Anthropic's S-1 Reveals $11.6 Billion Quarter and $810 Billion in Compute Debt

Anthropic posted $11.6 billion in Q2 2026 revenue and its first GAAP operating profit of $974 million — and simultaneously disclosed long-term compute commitments of roughly $780 to $810 billion. Those two figures, disclosed in its S-1 filing with the SEC, frame the bet the company is asking public investors to underwrite. Anthropic plans to list on the Nasdaq under the ticker "ANT," with Goldman Sachs and Morgan Stanley leading the offering. Wall Street expects a valuation of roughly $2 trillion, according to Yahoo Finance, up from its most recent private valuation of about $965 billion — already ahead of OpenAI's roughly $852 billion.

Anthropic is the first major AI lab of its kind to file an S-1 publicly. That alone gives the filing market significance: it is the first detailed look at the unit economics of a frontier-model company, and the numbers will anchor how investors price every AI firm that follows. Chip stocks rallied on the news — Nvidia, Broadcom, and AMD all gained after the filing went public — because Anthropic's disclosed compute spending effectively functions as a forward order book for silicon.

How fast is the revenue actually growing?

Revenue jumped from $376 million in Q1 2025 to $11.6 billion in Q2 2026, a 1,368 percent year-over-year increase. For the first half of 2026, Anthropic brought in $16.3 billion total — more than triple its full-year 2025 revenue of $4.6 billion. The company turned its first positive GAAP operating profit of $974 million in Q2, even as compute costs rose 300 percent over the same period. The saving grace is arithmetic: costs grew slower than revenue.

Revenue per employee sits at $6.6 million, higher than any Magnificent 7 company except Nvidia, with about 5,300 people on staff. That figure flatters efficiency at a company still scaling its headcount; it says nothing about durability.

The customer base is broad but top-heavy in a specific way. More than 11,900 customers pay over $100,000 annually, and about 250 spend more than $10 million. No single customer accounts for more than ten percent of revenue, but five customers together make up 25 percent. The S-1 does not name them. If those five include companies like Meta, Google, or Microsoft — firms that could switch to their own or cheaper models — that concentration becomes a live risk the filing declines to quantify.

What does 900% net revenue retention mean?

Anthropic's net revenue retention rate sits at roughly 900 percent, meaning existing customers spend nine times what they did a year ago. That is the single most quoted number in the filing, and it deserves scrutiny. The source itself notes that most deployments are likely still shallow, depend on human oversight, and do not run AI at full speed. Retention of this magnitude reflects expansion into usage that may still be experimental, not yet contracted, recurring demand.

The billing mix tells its own story. Eighty percent of revenue comes from per-token billing (a token is the smallest billing unit for AI model usage, priced by actual consumption), while the remaining 20 percent comes from fixed subscriptions. That split suggests subscriptions may still be cross-subsidized by usage revenue — but it also means Anthropic has room to raise prices. The company's forecast for next year will be primarily usage-based, according to the FT.

Claude Enterprise users who also use Claude Code are the biggest spenders, paying about 20 times more than other customers, according to the S-1. Developer-tooling customers, in other words, carry the business.

Not all revenue is clean. Roughly 5 to 7 percent comes from customers who violate usage policies, primarily through model distillation (using a model's outputs to train another, cheaper model). Anthropic says it has already been cracking down. Government contracts, by contrast, account for only about one percent of total revenue — worth noting for a company often discussed in policy circles as a public-sector partner.

Why does $780 billion in commitments matter?

The filing's most consequential disclosure is structural: compute capacity contracts with Broadcom, Google, Amazon, Microsoft, and other providers add up to roughly $780 to $810 billion. Those costs are more or less fixed and locked in long-term. Revenue is variable and usage-dependent. Anthropic has, in effect, swapped a variable-cost business for a fixed-cost one at unprecedented scale, and it must keep growing fast simply to cover its future bills. If it cannot, the AI-fueled capital market could start to wobble — the filing's obligations are large enough that Anthropic's growth rate is now a systemic variable, not just a company-specific one.

To gain flexibility and lower costs over time, Anthropic is building its own data centers under the name "Anthropic Direct." Owning infrastructure converts some rental obligations into capital assets, but it also deepens the total capital committed to the AI buildout.

What is not in the filing?

Anthropic left its Q3 numbers out of the S-1, even though the company already has them. One possible reason, the source notes, is that growth slowed significantly in Q3. Companies filing IPO paperwork choose their reporting cutoffs, and an omitted quarter of deceleration is exactly the kind of omission a prospectus is not required to explain.

Momentum may have picked back up. OpenRouter data (a marketplace that captures only a tiny slice of total token demand and skews toward open-weight models) shows demand for Anthropic models climbing again after the October release of the 5.5 series, especially Opus 5.5. If Anthropic is dominating on a platform biased against frontier closed models, it is likely doing even better everywhere else. That inference is directional, not audited.

The takeaway in plain words: Anthropic has proven it can turn usage into profit at extraordinary efficiency, while committing to fixed compute obligations nearly seventy times its annualized revenue — and the public market will now decide whether that trade is worth $2 trillion.