Anthropic may be sitting on one of the most expensive calendars in tech. Its planned data center in Louisiana is estimated to generate about $1.3 billion in revenue every month once fully running.
That turns the construction schedule into a revenue question. Under the same estimates, bringing the site online just one month early could pull forward at least $1.3 billion for the maker of the Claude AI models.
The details: 611 acres, 330 MW and a lot of TPUs
The project is called River Bend. It sits on a 611-acre campus in West Feliciana Parish, Louisiana, and Hut 8 is developing it.
The first phase will deliver 245 MW of IT compute capacity, backed by 330 MW of total power capacity. The hardware inside will be Google TPUs, the custom chips Google designed specifically for AI workloads.
A quick translation of those two megawatt figures: 245 MW is the power that actually reaches the computers. The extra headroom up to 330 MW covers everything else a data center needs, such as cooling and electrical overhead.
The financing structure is a three-way arrangement. On December 17, 2025, Hut 8 signed a 15-year lease worth $7 billion with Fluidstack, and Google guarantees the deal. Anthropic will use the capacity from that initial phase.
Operations are expected to begin in the second quarter of 2027. A groundbreaking and first-look event took place in early October 2026, attended by Anthropic’s head of compute, Louisiana officials and Donald Trump Jr.
Why the margin math matters
The estimated $1.3 billion monthly figure assumes the site runs mainly on inference workloads. Inference is the part of AI that earns money: it covers answering user prompts and running models in production, as opposed to training them.
The economics look notably lopsided in Anthropic’s favor. Gross margins are estimated at around 85% after chip-leasing costs, which are pegged at roughly $200 million per month.
The local impact is also significant. The first phase of construction is expected to create up to 1,500 jobs. Once the facility is operating, it is projected to support roughly 75 to 100 permanent positions and generate up to $90 million a year in tax revenue for the parish.
Background: the race for compute
Louisiana is one piece of a much larger build-out. Anthropic is leasing more than 15 GW of compute capacity in total across AWS, Google and other providers.
River Bend is one of several purpose-built campuses planned for delivery between 2027 and 2028. The timing lines up with Anthropic’s plans for an initial public offering.
The arrangement also shows how AI infrastructure deals are now structured. The developer builds the site, a specialist operator holds the lease, a hyperscaler guarantees it, and the AI lab consumes the output.
What this means
For anyone tracking Anthropic ahead of a possible IPO, the key takeaway is that compute capacity and revenue are now discussed in the same breath. If the estimates hold, each month of delay at River Bend carries a cost measured in billions of dollars of revenue that gets pushed back.
That gives every party involved a strong incentive to move quickly. Hut 8 has a 15-year anchor tenant structure backed by Google’s guarantee. Fluidstack sits in the middle as the leaseholder. Google gains another large customer running on its TPUs rather than a rival’s chips.
There are clear caveats. The $1.3 billion monthly revenue figure is an estimate, and it assumes the site is fully built and running inference at scale. The Q2 2027 start date is a target, and large construction projects are known for slipping rather than finishing early.
Demand is the other variable. The revenue projections depend on customers continuing to send enough AI workloads to fill 245 MW of compute. Anthropic is betting on that across more than 15 GW of leased capacity, so the Louisiana site is less a single wager than one chip on a very large table.
For West Feliciana Parish, the stakes are more concrete. Up to $90 million a year in tax revenue and up to 1,500 construction jobs would make a meaningful difference for a rural Louisiana community.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.