Anthropic's Confidential IPO Filing Shows $518 Billion in AI Infrastructure Commitments, Most Locked In

Anthropic has told prospective investors in a confidential IPO prospectus that it expects to spend at least $518 billion over the next decade building AI infrastructure with six partners, according to a filing seen by Reuters and reported by wtvbam.com and Konsulteer. The company filed confidentially with the U.S. Securities and Exchange Commission in June, and the paperwork has not been made public. A prospectus was shared with a small group of partners, people familiar with the matter told the Financial Times, according to Euronews and Yahoo News Malaysia.

The $518 Billion Breakdown Across Six Partners

Per the filing details reported by Reuters, Anthropic expects to spend at least $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft under long-term infrastructure service agreements. Separately, the company carries about $161.2 billion in Broadcom-related equipment lease obligations, described in the filing as largely non-cancelable and the single largest item disclosed. Anthropic also disclosed agreements with Elon Musk's xAI that could result in up to $84.5 billion of spending on Nvidia-based computing capacity through 2029, and AMD has committed to buying up to $5 billion of Anthropic stock while supplying AI computing capacity expected to exceed $20 billion.

Summing the six disclosed figures ($161.2bn Broadcom + $111.1bn Google + $110bn Amazon + $84.5bn xAI + $31.4bn Microsoft + $20bn AMD) produces about $518.2 billion, our calculation, which lines up with the filing's headline figure of "at least $518 billion" cited by Reuters.

Non-Cancelable Terms and Contract Structure

The filing draws a clear line between commitments Anthropic can walk away from and those it cannot. The Google agreement runs from April 2026 through July 2033, and the Amazon agreement from May 2026 through April 2036, according to the prospectus language reported by Reuters. Both carry make-whole provisions: filing language states, "If our actual spend falls short, we must pay Google the difference," with similar terms applying to the Amazon agreement. The Broadcom equipment leases are described in the reported excerpts as largely non-cancelable. Treating those commitments plus the Microsoft cloud agreement as fixed — the reported excerpts do not spell out Microsoft's cancellation terms — implies roughly $413.7 billion ($161.2bn + $111.1bn + $110bn + $31.4bn) of the total is effectively locked in, our calculation, or about 80% of the $518.2 billion sum.

By contrast, the xAI agreements covering up to $84.5 billion of Nvidia-based capacity through 2029 are largely cancelable on 90 days' notice, a materially different risk profile from the cloud and equipment-lease commitments. Anthropic told investors, per the Reuters-sourced reporting, that it is shifting from a cloud-only model toward building its own AI infrastructure, combining dedicated data centers with directly leased chips and other computing equipment.

Revenue, Losses and Liquidity Against the Commitments

The scale of the buildout stands against a company still deep in losses. Revenue jumped twelvefold to almost $4.6 billion in 2025, but operating expenses reached $12.65 billion, more than half of it on computing and infrastructure, leaving an operating loss of more than $8 billion, according to the filing details reported by Euronews and Yahoo News Malaysia, both citing the Financial Times. Anthropic's net loss for 2025 was about $42 billion, which included a roughly $34 billion accounting charge tied to financing that could eventually convert into shares.

The company had $20.28 billion in cash and short-term investments at the end of 2025, a fraction of the long-term infrastructure obligations disclosed elsewhere in the filing. More recently, revenue reached $11.5 billion in the second quarter of 2026, and Anthropic is on course for a second straight quarter of operating profit on an adjusted basis, per the Financial Times reporting relayed by Euronews. The filing also flags customer concentration as a risk: close to a quarter of last year's revenue came from just two customers, and many of Anthropic's largest clients are not locked into long-term contracts.

Dependence on Partners Who Are Also Competitors

Per the prospectus language reported by Reuters, Anthropic told prospective investors that access to computing power is becoming a central constraint on AI development, with future demand for advanced systems expected to exceed available supply, and that growth will be "limited principally by the availability of compute." The filing also warns of a structural tension: Amazon, Google and Microsoft simultaneously act as investors, customers, cloud providers, distributors and competitors developing their own AI models, creating incentives that "may not be fully aligned" with Anthropic's interests. Risk-factor language cited by Reuters states plainly: "If the compute we have access to from third parties is curtailed, repriced, or terminated ... our business, financial condition, and results of operations could be adversely affected."

Separately, reporting on the same filing, including from CNBC-TV18, has highlighted that Anthropic devotes a substantial portion of its risk factors to warnings that advanced AI models could behave unpredictably, resist shutdown, or pose what the company calls catastrophic or existential risks. Euronews and Yahoo News Malaysia, citing the Financial Times, report that almost a third of the S-1 is devoted to risk factors.

IPO Timing and Valuation

No offering size has been disclosed in the reported filing excerpts. The listing is expected on the Nasdaq this autumn, according to the Financial Times reporting cited by Euronews, and backers reportedly expect Anthropic to list at a valuation above $2 trillion, more than double the $965 billion set in its funding round in May. Anthropic has not publicly commented on the reports and did not immediately respond to a request for comment on the filing, according to the wtvbam.com report.

Bottom Line

The confidential prospectus paints a picture of a company betting heavily on locked-in compute capacity, with roughly 80% of its disclosed $518 billion decade-long infrastructure spend, our calculation, structured so it cannot be easily unwound, even as current cash and revenue remain small relative to those obligations. The gap between $20.28 billion in cash at the end of 2025 and hundreds of billions in multi-year commitments underscores why the company is pursuing what could be one of the largest IPOs on record, though the timing, structure and ultimate size of any offering remain unconfirmed pending public disclosure of the S-1.