Anthropic's $1.41 Trillion IPO Turns Cloud Giants Into Dual Winners

By Nathan Brooks

Anthropic's public offering targets a fully diluted valuation of roughly $1.41 trillion, and two of its largest suppliers will walk away with a combined equity position worth about $411 billion. Amazon and Alphabet, the two hyperscalers that host Anthropic's models, invested roughly $18 billion and $33.3 billion respectively. At the top of the price range, those stakes carry estimated values of $271.8 billion and $139 billion, including Amazon's convertible notes.

The offering itself is substantial. Anthropic is selling 825 million Class A shares at $70 to $78 apiece, which could raise base proceeds of about $64.35 billion. A concurrent $26 billion private placement brings the basic equity valuation slightly above $1 trillion. Market reports indicate institutional orders have already covered the full offering ahead of pricing.

What do the underlying numbers show?

The financials reveal a company scaling quickly. Anthropic posted third-quarter revenue of about $17 billion and an October annualized run rate of approximately $81 billion. Its enterprise base grew to more than 11,900 accounts generating at least $100,000 in revenue over the trailing twelve months. Compute expenses fell to roughly 55% of revenue in the first half, moving the company closer to adjusted operating profitability once non-cash convertible note markups are excluded.

The circular economics deserve attention. Filings show about $900.9 billion in total infrastructure commitments over the coming years, with roughly 80% categorized as non-cancellable. That includes at least $121.3 billion committed to Google Cloud and $110 billion to Amazon Web Services. About 47% of Anthropic's revenue flows through cloud marketplaces such as AWS Bedrock and Google Cloud Vertex AI, where Amazon and Alphabet collect an estimated 16% distribution fee. Anthropic then pays the same two companies to rent custom silicon, including Amazon Trainium2 and Google Tensor Processing Units.

The arrangement pays dividends already. Alphabet reported Google Cloud revenue of $30.71 billion, up roughly 103% year over year, at a 36.2% operating margin. Alphabet is also using its corporate credit rating to backstop project debt for a dedicated data center lease in Kentucky that will support Anthropic's Google hardware usage. AWS grew nearly 39% to $46 billion and generated more than $17 billion in operating profit. The contracted backlog reached roughly $578 billion at AWS and $561 billion at Google Cloud.

How do the new models fit in?

Anthropic just released Claude Fable 5.2 and Mythos 5.2, which generate tokens up to 30% faster than previous versions. Standard API pricing now sits at $8 per million input tokens and $40 per million output tokens. Fable 5.2 ships with automated safety classifiers that cut unnecessary safety blocks by roughly 55%. Mythos 5.2 targets vetted enterprise, cybersecurity and biomedical research institutions with lifted default restrictions.

Anthropic says Claude now writes more than 90% of the code changes merged into its core codebase, with human guidance. More than 50% of the Fortune 500 use the company's products. Cheaper, faster inference acts as a catalyst for compute consumption, which in turn feeds the hyperscalers' backlogs. It is a toll road, and the toll collectors own shares in the traffic.

What should investors weigh?

The pure-play carries caveats. At $1.41 trillion diluted, the stock trades at roughly 12 to 17 times forward run-rate sales while carrying heavy fixed compute liabilities. A multi-class structure gives a single Class F share 50.1% voting power, with an independent trust electing a board majority; public Class A shareholders get economics but little say, which may keep the stock out of certain major indexes.

The cloud giants offer a different profile. Alphabet trades at a trailing price-to-earnings ratio of 15 with a 0.31% dividend yield; Amazon trades at nearly 22. Together they generate over $210 billion in annual net income, though each faces roughly $200 billion in annual capital expenditures. Both also compete with their own customer: Google actively markets its Gemini models, so hyperscalers capture value whichever model an enterprise chooses.

With OpenAI also expected to go public later this year, the market will soon test how it prices model developers whose largest landlords are also their shareholders and rivals.