Anthropic Files S-1: $11.6 Billion in Quarterly Revenue and $780 Billion in Commitments

Anthropic is the first major AI lab to publicly file a Form S-1 with the SEC. The company plans to list on Nasdaq under the ticker ANT, with Goldman Sachs and Morgan Stanley leading the offering. According to Yahoo Finance, Wall Street expects a valuation of about $2 trillion. The last private valuation was roughly $965 billion, which already put Anthropic ahead of OpenAI and its $852 billion. Revenue in the second quarter of 2026 reached $11.6 billion, and for the first time in its history Anthropic reported a positive GAAP operating profit: $974 million.

Growth that is hard to digest

In the first quarter of 2025, Anthropic earned $376 million. Five quarters later, in the second quarter of 2026, revenue was $11.6 billion. That is up 1368% year over year, or roughly 31x. In the first half of 2026, the company booked $16.3 billion, more than triple all of 2025, which totaled $4.6 billion.

Compute costs over the same period grew 300%. Revenue grew faster, and that gap produced the $974 million operating profit. Revenue per employee is $6.6 million, with a headcount of about 5300. Among the Magnificent 7 companies, only Nvidia is higher.

Two details from the prospectus. Government contracts account for about 1% of revenue. Another 5–7% comes from customers who violate usage policies, primarily through model distillation. Anthropic is already fighting this.

Who pays and for what

Net revenue retention is around 900%: existing customers spend nine times more on average than they did a year ago. Most deployments are probably still shallow, depend on human oversight and do not run at full speed, so the headroom for growth looks substantial.

The customer base breaks down as follows, according to the S-1:

• more than 11,900 customers pay over $100,000 a year;
• about 250 customers spend more than $10 million;
• no single customer accounts for more than 10% of revenue, but the five largest together account for 25%;
• Claude Enterprise users who also use Claude Code spend roughly 20 times more than everyone else;
• 80% of revenue is based on actual consumption, where the minimum billing unit is a processed token, and the remaining 20% comes from fixed subscriptions.

Revenue concentration could become a risk. If the five largest buyers include Meta, Google or Microsoft, they could switch to their own models or cheaper ones. The 80-to-20 split also hints that subscriptions are still partly subsidized, which leaves Anthropic room to raise prices. According to the FT, next year's forecast will be built mainly on usage-based payments.

Fixed bills against variable revenue

Compute commitments to Broadcom, Google, Amazon, Microsoft and other providers total roughly $780–810 billion. These costs are barely flexible and are scheduled years ahead, while revenue stays variable and depends on how many tokens customers push through. To cover future bills, Anthropic has to keep growing fast. If growth stalls, the whole AI economy running on borrowed compute would wobble too. To gain flexibility and cut costs, the company is building its own data centers under the name Anthropic Direct.

The prospectus leaves out third-quarter data, even though the company already has it. One possible explanation is a noticeable slowdown in growth. But there is also a signal pointing the other way: according to OpenRouter, demand for Anthropic models rose again after the October release of the 5.5 series, especially Opus 5.5. The service covers only a small share of total token demand and skews toward open models. If Anthropic leads even there, it is probably doing no worse in enterprise channels.

Shares of Nvidia, Broadcom and AMD rose after the S-1 was published. The market read the prospectus as confirmation that compute demand is backed by real revenue.

For investors, the main question now is Q3: will its numbers confirm the trajectory? For developers and teams building products on Claude, the shift to per-token payment matters. AI spending is increasingly a variable cost line, and it makes sense to plan it with a cushion for price increases. Anyone building a product on a single API should work out a fallback on another model in advance: the prospectus names exactly this customer mobility as the main risk to Anthropic's revenue.