“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
Amazon has started controlling Nvidia GB300 server racks before some AWS data centers are equipped with sufficient power to operate them. This is leaving the hardware worth billions of dollars idle as the company is hurrying to boost its AI capabilities.
According to Business Insider, internal emails from August revealed initial rack receipts of about $5 billion, as well as $3 billion on a monthly basis thereafter. These figures are forecasts rather than confirmed purchases.
Under the Early Rack Receipt policy of AWS, ownership is realized prior to the receipt, which at times means that the equipment is kept in warehouses.
An Amazon representative claimed that this practice was a:
“best business practice.”
In addition, Amazon denied allegations of not having a deployment plan.
When the chip arrives before the power
AWS used to assume ownership of racks when they reached their locations; now it is paying for them before installation or electricity has been made available.
These “unpowered” systems, sometimes referred to as “dark GPUs,” do not earn any money during the waiting time. In the meantime, they are losing value, increasing the burden on cash flow.
A cash-flow clock that keeps ticking
According to Business Insider, Amazon is projected to spend $220 billion on capital investment in 2026.
From the company’s report to the SEC, Amazon had cash capital investment of $96.3 billion and operating cash flow of $71.4 billion during the first half of the year. The company’s spending also included spending on its fulfillment infrastructure.
Amazon separately spent $13.7 billion on OpenAI and $10 billion on Anthropic in the second quarter. However, the investments to these two companies do not fall under any category of capital expenditure.
At a September staff meeting, AWS infrastructure executive Prasad Kalyanaraman described the AI business as:
“hugely free cash flow negative.”
He warned the strain could last several years, according to Business Insider.
CEO Andy Jassy, however, sees AI as a:
“once in a lifetime opportunity.”
There is a cause for optimism. AWS earnings rose by 37%, and contracted backlog climbed to $496 billion in the second quarter.
Amazon’s search for power expands
Cleanview estimates that Amazon is developing a total of 44 GW of capacity in its data centers in the US, including 27 GW from 13 newer markets.
It also points out that Amazon has planned gas generation of 17.5 GW, while it has contracted the supply of carbon-free electricity of 25.6 GW. All these figures represent different commitments rather than operational capacity that could simply be added up.
The problem is not limited to Amazon. Bain points to the lack of electricity, equipment, and worker skills, as well as permit delays. Moreover, the IEA has recently pointed out the increasing stress on the grid.
Local opposition makes it more difficult to expand. According to Cryptopolitan, Amazon has pledged $1 billion to support the communities hosting its data centers.
Why operational capacity could shape AI prices
BCG expects AI-compute pricing to become more usage-based and potentially dynamic.
Power shortages could influence where AI facilities are built, who gets computing capacity, and what customers pay worldwide.
For Amazon, buying GPUs is only part of the challenge. The bigger test is getting them running and turning that investment into revenue.
The smartest crypto minds already read our newsletter. Want in? Join them.