Amazon’s AI business has crossed a $25 billion annualized revenue run rate. It got there in roughly three years.
CEO Andy Jassy told investors on the July 30 earnings call that the unit is growing “several hundred times faster” than AWS did in its early days.
The numbers behind the milestone
In its Q2 2026 report, Amazon disclosed that its AWS AI business and its custom chip business each passed the $25 billion run-rate mark. Both posted triple-digit year-over-year growth.
A quick translation for anyone allergic to finance terms: a run rate takes current revenue and projects it across a full year. It is a snapshot of momentum, not a booked annual total.
The broader cloud unit also had a strong quarter. AWS revenue reached $42.2 billion, up 37% from a year earlier.
That works out to a $169 billion annualized run rate. It also marks AWS’s fastest growth rate in the last 18 quarters.
AWS posted an operating margin of 39.4%, alongside substantial growth in operating income.
Amazon reported a backlog of $496 billion, which also grew at a triple-digit rate year over year. Demand is exceeding its capacity for 2026 and 2027.
The $220 billion question
Amazon raised its 2026 capital expenditure guidance to approximately $220 billion, up from $200 billion. The company tied most of that spending to AI infrastructure and custom chips. Rising memory costs added further pressure on the budget.
Amazon’s trailing twelve-month free cash flow fell to an outflow of $7.6 billion, largely because of the heavier investment.
Anthropic, OpenAI and the chip bet
A key part of the story is Trainium, Amazon’s in-house AI chip. The company has major agreements with both Anthropic and OpenAI to use Trainium.
Amazon’s managed AI service, Bedrock, also had a standout quarter. Bedrock usage in Q2 surpassed all previous quarters combined.
What this means
For Amazon shareholders, the trade-off is now explicit. The company is pairing its fastest AWS growth in 18 quarters and a 39.4% operating margin with negative free cash flow and a capex budget of approximately $220 billion.
Capacity constraints mean Amazon cannot recognize revenue it cannot serve, so the timing of data center and chip buildouts becomes critical to the story.
Rising memory costs add another variable. If component prices keep climbing, each dollar of capex buys less capacity, which could stretch the timeline for working through the backlog.
Customer concentration is also worth watching. Anthropic and OpenAI are anchor tenants for Trainium, and much of the custom chip momentum depends on those relationships holding up.
Key metrics to track include AWS growth rates, backlog conversion, operating margin, and whether free cash flow begins to recover from its $7.6 billion trailing outflow.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.