Anthropic Starts Leasing Proprietary Data Centers While Still Using Amazon and Google Cloud Power

Anthropic launches “Anthropic Direct” data‑center strategy

Anthropic’s S‑1 filing, released on October 26, states the company now leases AI‑optimized data centers and the chips inside them, a program it calls Anthropic Direct.

The move adds to Anthropic’s existing practice of renting compute capacity from Amazon Web Services (AWS) and Alphabet’s Google Cloud.

Fluidstack, the firm behind a $50 billion Texas and New York data‑center deal announced in November 2025, is identified in the filing as the partner that designs, builds, and operates these leased facilities for Anthropic.

Anthropic plans to sign additional leases, expanding the footprint of its privately‑managed infrastructure.

Financial commitments underline the stakes for Amazon and Google

The S‑1 reveals Anthropic has pledged $110 billion to AWS through 2036, tied to Amazon’s Trainium AI chips.

A parallel commitment of $114.5 billion to Google’s cloud services runs through 2033, according to CNBC’s reading of the prospectus.

Amazon’s capital expenditures surged from $52.7 billion in 2023 to $132 billion in 2025, and the company reaffirmed a target of about $220 billion for 2026.

Amazon reported AWS revenue growth accelerating to 39 % in its October 29 results, with a backlog of $578 billion, and CEO Andy Jassy projected growth above 40 % for both 2027 and 2028.

Free cash flow fell sharply from $32.9 billion in 2024 to $7.7 billion in 2025, and a 12‑month period ending September 30 showed a $42.4 billion cash outflow.

These figures illustrate how critical large AI customers like Anthropic are to recouping Amazon’s massive capex outlays.

Analyst view: Anthropic is “reduplicating” AWS

On the October 27 episode of Prof G Markets, investor Paul Kedrosky summed up the development: “They’re essentially reduplicating AWS, Amazon Web Services, in both directions.”

Kedrosky argued Anthropic’s dual investment in product development and underlying cloud infrastructure reduces reliance on its biggest rivals and avoids the marketplace fees they charge.

He expects Anthropic’s spending on its own data‑center leases to accelerate as the company scales.

Because Anthropic’s long‑term compute agreements require payment for capacity regardless of actual usage, the new leases sit on top of existing obligations rather than replacing them.

The S‑1 notes that the equipment and building costs are recorded as lease commitments, keeping the spend off Anthropic’s balance‑sheet capital‑expenditure line.

Consequently, AWS retains the $110 billion revenue floor even as Anthropic builds partial independence.

Anthropic’s strategy creates a hybrid model where the firm controls critical hardware while still tapping the elasticity of public clouds for peak demand.

This arrangement may influence how other AI firms negotiate cloud contracts and consider private‑infrastructure investments.

Overall, the development signals a nuanced shift in the relationship between a leading AI startup and its two biggest cloud providers.

Why This Matters

Anthropic’s partial move to privately leased data centers introduces a new competitive dynamic for AWS, potentially limiting the cloud giant’s upside from one of its largest AI customers.