Anthropic Just Made $559 Million in a Quarter. Nobody Asked the Right Question.
Anthropic is about to post $559 million in operating profit for the quarter ending June. Eighteen months ago, it told investors it would not break even until 2028.
That is the number nobody is leading with.
The headlines have correctly captured the scale: Anthropic is in talks to raise at a $900 billion valuation, its revenue is on track to hit $10.9 billion this quarter — up from $4.8 billion in the first quarter — and its annualized revenue crossed $43 billion in April, up from $9 billion at the end of 2025. Eight of the Fortune 10 are now customers. More than 500 companies spend over $1 million a year with Anthropic, up from twelve two years ago. These are extraordinary numbers by any measure.
But the question worth asking is simpler than the valuation math: is this growth real?
There are two ways to double your revenue quickly. You can sell more, or you can charge more. One means the enterprise AI adoption curve is even steeper than the most optimistic models predicted. The other means the growth is partially a repricing story, and the next renewal negotiation will look very different from the last one.
The available pricing data suggests repricing is not the primary driver — at least not through the model lineup currently on offer. According to Price Per Token, which tracks AI API pricing across providers, Claude Sonnet 4 launched in May 2025 at $3 per million input tokens and $15 per million output tokens. When Sonnet 4.6 arrived in February 2026, the price was identical: $3 and $15. The flagship Opus tier launched at $15/$75 in May 2025 and has not changed. No price increases appear in the current model lineup during the period Anthropic grew from $9 billion to $43 billion in annualized revenue. This does not prove volume drove the growth — enterprise contracts, volume discounts, and custom arrangements are not reflected in list prices — but the list prices offer no evidence of upward repricing.
Whether it is volume, repricing, or some combination is not publicly confirmed. Anthropic declined to specify what portion of revenue growth came from expansion versus price changes. The company did not respond to a request for comment on token pricing history.
Anthropic books revenue from cloud resellers — AWS, Google, Microsoft — on a gross basis, counting total end-customer spend as top-line revenue and paying out partner fees as expenses. This is a legitimate accounting treatment, but it makes cross-company comparisons difficult. Sacra, which tracks private market company metrics, flagged that the gross-basis treatment inflates Anthropic's reported revenue relative to peers that report net. Neither CNBC nor the Wall Street Journal addressed this distinction in their coverage of the profit milestone.
The $559 million operating profit figure also excludes stock-based compensation. Anthropic has not disclosed its SBC figure, making a fully loaded profit calculation impossible from public sources. That is not unusual for a high-growth private company, but it means the profit margin story is more complicated than a single quarterly number suggests.
Then there is the compute bill.
Anthropic is paying SpaceX $1.25 billion per month through May 2029 to rent capacity at the Colossus data center in Memphis, according to terms revealed in SpaceX's IPO prospectus. That is $15 billion annually — more than a third of its annualized revenue — flowing to a company controlled by Elon Musk, who has publicly attacked Anthropic's safety commitments, declined to sign the AI safety commitments Anthropic co-founded, and whose own AI company xAI is a direct competitor. The arrangement has no obvious strategic alternative, but it is a significant dependency that does not appear in the revenue celebration coverage.
Anthropic CFO Krishna Rao said in the company's February Series G announcement that Anthropic had earned its first dollar of revenue less than three years prior. The company went from zero to $14 billion in annualized revenue by February 2026, and to $43 billion annualized by April. The growth is not in dispute. The composition of that growth is.
Eighteen months is a long time in AI. Whatever the answer is, it will matter for every company that assumed it had more runway to catch up.