When Anthropic filed to go public, the surprise was not the size of its losses but the identity of its best customers. Amazon and Google, the two cloud giants competing with Anthropic’s Claude for corporate AI budgets, together generated about $2.16 billion of the company’s revenue in 2025, or 47 cents of every dollar it took in, according to the IPO prospectus.
The company took in about $4.6 billion last year, roughly twelve times the prior year’s total. Most of that, about $3.8 billion, came from consumption-based pricing, the payments customers make as their applications call Claude. Subscription revenue added $789 million. The split matters because consumption revenue depends on customers keeping their workloads running, and the two biggest of those customers are the same cloud providers selling the underlying computing.
The arrangement is a study in awkward alliances. Amazon sells its own models through Bedrock and Google sells Gemini, yet both host Claude and direct some of their largest enterprise accounts toward it. Anthropic needs their chips and their sales channels; they need a frontier lab whose model keeps corporate customers from defecting to the other’s stack. The result is a revenue base in which the competitors are also the counterparties.
Claude’s rise from research curiosity to commercial workhorse is what the revenue figures describe. Enterprises adopted the model for coding, customer service and internal analysis, and the consumption line grew as those workloads scaled. The surge arrived in a year when every major lab was racing to convert fast-improving models into recurring revenue before the next round of training costs came due.
The operating loss widened to more than $8 billion, more than double the prior year’s figure. The number reflects the cost of the enormous infrastructure buildout Anthropic is financing while its revenue, however fast it is growing, is still a fraction of its spending.
A second disclosure in the filing concerns a customer of a different sort. Anthropic has agreed to pay as much as $84.5 billion through 2029 for Nvidia-based computing capacity supplied by xAI, the artificial-intelligence company owned by Elon Musk’s SpaceX, according to reports by Reuters and The Information. The figure is nearly double the roughly $45 billion that earlier disclosures implied, and it covers access to about 325,000 Nvidia GPUs spread across the Colossus data centers.
That a company training Claude would pay one of its sharpest rivals, xAI’s Grok, for compute is unusual even by the standards of this industry, where almost every lab is simultaneously a supplier, a customer and a competitor. The deal lets Anthropic rent capacity quickly rather than wait for its own data centers to come online.
Unlike most of Anthropic’s other commitments, the SpaceX arrangement can largely be canceled with 90 days’ written notice. Earlier filings from SpaceX’s own listing described payments of $1.25 billion a month through May 2029. The new top-line figure of $84.5 billion marks the ceiling if the deal runs its full course, not money already spent.
The SpaceX contract sits inside a much larger pledge. Anthropic expects to spend at least $518 billion on computing and infrastructure over the next decade, and roughly four-fifths of that is non-cancelable or locked into minimum commitments that must be paid whether or not the capacity is used. The partners named in the filing include Google, owed $111.1 billion; Amazon, owed $110 billion; Microsoft, owed $31.4 billion through May 2033; and Broadcom, tied to equipment leases.
The company was founded in 2021 by Dario Amodei and his sister Daniela, former OpenAI researchers who left to build a lab they argued would put safety ahead of speed. It ended 2025 with about $20.28 billion in cash and short-term investments, a cushion that sits beside hundreds of billions in future obligations. People familiar with the planning expect the IPO to value the company above $2 trillion.
The filing lands while OpenAI is raising at least $30 billion at a valuation near $1.4 trillion, a measure of how crowded the top of the market has become. Anthropic’s prospectus positions the company as the safety-minded alternative in that field, and the risk section, which runs to roughly a third of the document, is where that stance shows up most plainly.
The concentration in its customer base is the part of the filing investors will read most carefully. Nearly half of revenue rests on two companies that also compete directly with Anthropic, and many of its other largest customers have not signed long-term contracts. That leaves a fast-growing revenue base standing on relationships that could shift quickly.
What the filing does not settle is whether the spending is a moat or a millstone. Every dollar of the $518 billion commitment is a bet that demand for computing will keep compounding and that Anthropic can capture enough of it before the money runs out. The IPO is the next tranche of that bet, and the prospectus is the pitch.
