How Anthropic Clinched $517 Billion in Compute Deals in 11 Months

Anthropic in the last year has scrambled to line up cloud computing deals with SpaceX, Google and others to meet the skyrocketing demand for Claude Code and Cowork that took the company by surprise this year. As the AI powerhouse gears up for a possible record-setting initial public offering, potential investors are wondering: Will that compute be enough?

Anthropic has never publicly disclosed a figure for its total compute capacity, and it’s unclear exactly how much the company is paying right now or has pledged to pay in the future. To get an estimate, The Information analyzed public statements and our own prior reporting. Since October, Anthropic has entered into agreements for at least 14.8 gigawatts of compute capacity it could tap over the next several years. That’s in addition to between 1 GW and 2 GW it had secured before October, according to a person familiar with the matter.

The recent agreements are likely to cost it far more than the $180 billion it told investors in December 2025 it would spend on renting servers through 2029. It could spend as much as $517 billion mostly over the next decade, based on commitments to rent cloud capacity, chips and leases on data centers, according to the analysis of the current announced deals.

Anthropic could be working on more deals than it has made public, but it also may not utilize all the agreed-upon capacity. And it’s hard to compare its spending projections with those of rival OpenAI, since Anthropic’s deals frequently extend beyond 2030. Anthropic signed on as a direct tenant for a 20-year lease with former crypto miner Riot Platform in Texas, for instance. (Most of the known ones are from three to six years.)

Still, Anthropic’s bevy of announcements suggests its planned total capacity may be far less than what OpenAI said it has has lined up. That’s even though Anthropic’s annualized revenue, which passed $65 billion, according to Bloomberg, has eclipsed OpenAI’s revenue of more than $40 billion as of July.

OpenAI in April told investors it’s planning to have 30 GW of compute by 2030. Last month, it projected spending around $750 billion on compute through 2030, according to a person familiar with the forecast, up from a forecast of $665 billion earlier this year.

Executives at OpenAI, which has inked compute deals at a blistering pace with Oracle, Microsoft, CoreWeave and others, have said those commitments have given the company an edge over its rival. Such forecasts also underscore the energy demand of AI datacenters: 1 GW is enough to run a city the size of San Francisco.

This competition helps explain why potential Anthropic investors are eager to get an update on its compute capacity—and spending—in its IPO prospectus, which it could make public in the coming weeks. Its ability to secure servers will determine whether it has enough compute to keep up with growth without undermining its margins or turning away customers.

Anthropic and its chief financial officer, Krishna Rao, have made compute deals a priority in the last year. Early partners and investors Amazon and Google remain the biggest suppliers, responsible for 11 GW that will cost more than $300 billion over the next 10 years or so, the analysis shows.

But in the last year Anthropic entered agreements with companies that had been closely aligned with OpenAI or were competing directly with Anthropic.

In November, it announced a deal with Microsoft, a longtime OpenAI backer, committing at least $30 billion to rent roughly 1 GW of Microsoft Azure servers running Nvidia chips.

Then in May, Anthropic announced it had entered an agreement with SpaceX to pay Elon Musk’s rocket-ship company $1.25 billion a month through May 2029 to rent AI servers, for a total potential cost of up to $45 billion.

In recent months, Anthropic has also partnered with a range of smaller companies that buy Nvidia graphics processing units. Just last month, Anthropic signed $80 billion worth of cloud contracts with startups Lambda and U.K.-based Nscale for at least 460 megawatts of capacity over six years.

Anthropic is also increasingly announcing agreements to directly lease data centers that could run off chips from partners such as Broadcom, Google and Advanced Micro Devices.

In November, Anthropic said it was planning to invest $50 billion on an initiative to build its own data centers with cloud startup Fluidstack. The data centers will operate in Texas and New York and should start coming online this year.

More recently, it entered into a lease agreement with former bitcoin miner TeraWulf, which is developing an AI data center in Kentucky that will eventually supply 401 MW of compute.

Anthropic also has ambitions to build its own data centers that could eventually house chips it plans to buy. In April, it signed a deal for multiple gigawatts of Google’s tensor processing units, designed by Broadcom, and later in July, 2 GW of AMD chips.

Some of the announced capacity may not come online in the planned time frame or indeed ever. Data center developers need to secure land parcels near substations and electric and gas transmission lines as well as permits and approvals from local governments. They face mounting pushback from municipalities wary of noise and energy strains from data centers. And they are confronting a shortage of components and links to the power grid.

To deal with some of these constraints, SpaceX is laying the groundwork for a foundry to build turbine vanes and blades for natural gas turbines, The Information reported.

What’s more, cloud computing contracts typically contain some language allowing participants to tear up an announced deal. For instance, in SpaceX’s Anthropic partnership, either party can cancel the deal with 90 days’ notice.

The Wall Street Journal first reported on OpenAI’s recent compute projections and some Anthropic compute deals.