Anthropic says preliminary Q2 results beat its own $559 million profit forecast — the same two months a $45 billion compute deal was running at a discount

Preliminary Q2 2026 figures reported this week put Anthropic's revenue above $11.5 billion — ahead of the $10.9 billion the company itself projected to investors in May — with a positive adjusted operating profit, Anthropic's first, according to the same reporting.

The claim matters beyond one company's bottom line. Frontier AI labs have spent the past two years telling investors that today's losses are the price of a business that becomes highly profitable at scale — and Anthropic clearing that bar in 2026, if the preliminary figure holds, would be the first real evidence for that pitch from any of the field's top three labs by valuation. It also lands five months into the aftermath of Anthropic's $965 billion Series H, the same May round whose investor materials produced the $10.9 billion forecast this week's figures now claim to have beaten — and the same valuation the company confidentially filed to go public against in July.

The comparison point is Anthropic's own. The Wall Street Journal reported in May, citing internal documents shared with investors ahead of a funding round, that the company projected $10.9 billion in Q2 revenue — up from $4.8 billion in the first quarter — and a first-ever adjusted operating profit of $559 million, a margin of roughly 5%. The same documents credited part of the swing to computing efficiency: Anthropic's compute cost per revenue dollar fell from 71 cents to 56 cents in a single quarter. Anthropic itself cautioned, in the same disclosure, that continued profitability was not guaranteed — planned infrastructure spending could push Q3 or Q4 back into a loss.

What actually changed this week

What's new isn't the shape of the claim, it's the number. This week's preliminary figures landed above the May forecast, and Anthropic is again reported as having posted a positive operating profit. Neither the May projection nor this week's update is audited — both are Anthropic's own account, the kind of preliminary figures a private company circulates while marketing a funding round, not the kind a public company later certifies in a quarterly filing.

The timing question

The scrutiny this week has focused less on the revenue number than on what sits underneath it. SpaceX's own IPO filing disclosed in May that Anthropic will pay Musk's company nearly $45 billion over roughly three years — $1.25 billion a month — for dedicated access to SpaceX's Colossus I and II GPU clusters, more than 200,000 Nvidia chips drawing hundreds of megawatts. The filing states that capacity ramped up during May and June 2026 at an initial discounted rate. Those are the same two months behind Anthropic's profit claim.

None of that makes the profit figure fabricated. A discounted onboarding period is standard in large compute contracts, and Anthropic's efficiency gain — 71 cents to 56 cents per revenue dollar — is a real operating metric disclosed in May, before the SpaceX ramp even began, and isn't explained by one vendor discount alone. What the filing does establish is that a meaningful share of the cost side of Anthropic's most-cited financial claim was priced below its long-term rate for exactly the window that claim covers — a coincidence neither company has quantified publicly.

A second discrepancy in the public record is harder to resolve. Anthropic CFO Krishna Rao told a federal court, in a March 9 declaration filed as part of the company's lawsuit against a Pentagon blacklist, that revenue was "exceeding $5 billion to date." (This newsroom located the court docket confirming the declaration exists and its date, but is citing the exact $5 billion figure from wheresyoured.at's reporting on the filing rather than a verbatim read of the primary document text.) Separately, Anthropic has told reporters its annualized run-rate reached roughly $19 billion around the same period. The two aren't necessarily contradictory — "to date" and "annualized run-rate" can describe different things measured different ways — but neither Anthropic nor the reporting on either figure spells out exactly what each one includes, which is the same gap that runs through most of this quarter's numbers: precise on the top line, vague on the definition underneath it.

The two months Anthropic used to prove it turned a profit are the same two months its biggest new compute bill was running below its long-term rate.

Where this sits against the rest of the field

Anthropic's timeline still compares well against its two closest rivals on its own terms. OpenAI has told investors it doesn't expect to be profitable until at least 2030, against a compute-spending target of roughly $600 billion through that year. Anthropic, on its own preliminary and unaudited account, says it got there in 2026 — two years ahead of its original internal forecast, and years ahead of OpenAI's. "First" and "first, on preliminary figures Anthropic reported about itself during an active fundraising process" are different claims, though, and this quarter is still the second one.