Anthropic told investors over the weekend that its annualized revenue run rate reached roughly $65 billion at the end of July, about a sevenfold increase from a year earlier. The update landed two days after the company disclosed preliminary second-quarter revenue of more than $11.5 billion — up from $787 million in the same quarter last year — alongside its first quarter of positive adjusted operating income.
The number is a coincidence worth pausing on. In its Series H round, Anthropic raised $65 billion at a $965 billion post-money valuation. Twelve weeks later it is generating that same $65 billion figure on an annualized basis. The company confidentially filed listing paperwork with the SEC in June and is working with Morgan Stanley, Goldman Sachs and JPMorgan on the offering.
Markets did not spend Monday celebrating. The S&P 500 closed down 0.5% and the Dow lost 272 points as oil climbed and the 30-year Treasury yield touched its highest level since 2007. But inside a red tape, storage and memory names ran: SanDisk gained about 8%, Western Digital about 6%, Micron about 5%. Investors are increasingly reading AI-lab revenue — not just Nvidia’s order book — as the signal for whether AI infrastructure spending is durable.
The arithmetic
Three numbers explain what changed. Anthropic booked roughly $10 billion of revenue across all of 2025. It did $4.73 billion in the first quarter of 2026 alone. Then it did more than $11.5 billion in the second — a jump of about 143% in a single quarter.
Now annualize that second quarter: $11.5 billion times four is roughly $46 billion. The exit run rate one month later was $65 billion. That gap is the whole story. The quarter that produced the headline was already stale by the time it was reported, because most of the growth happened inside it and kept going after it closed. A company moving that fast makes a trailing multiple almost useless — and makes a forward one an act of faith.
At $965 billion, the private mark values the business at roughly 15 times its current run rate. That is not an outrageous multiple for software growing sevenfold. It is an outrageous multiple if the growth rate halves twice.
Our take: The revenue figures are corroborated across the company’s own investor update and multiple independent reports. The profit figure deserves a harder look. “Positive adjusted operating income” is not net income, and adjusted is doing real work in that sentence — adjusted measures typically exclude stock compensation, and they say nothing about multi-year compute obligations signed for future periods. A frontier lab crossing into profitability would be a genuine landmark. A frontier lab crossing into adjusted profitability one quarter before a filing is a number built for a roadshow. The S-1 settles it, because the S-1 has to use GAAP.
Why this is a markets story, not an AI story
Because the read-through is priced into things you can actually buy. Anthropic is private; memory, storage, networking and power are not. Every incremental dollar of AI-lab revenue is the argument that this cycle’s capital spending is being funded by demand rather than by balance sheets — and the balance-sheet question is live, given the roughly $3 trillion of AI-related commitments sitting in Big Tech’s footnotes rather than in its debt line.
A $65 billion run rate at one lab does not resolve that. It does make the demand side harder to dismiss. Anthropic’s chief rival, OpenAI, is running near $40 billion annualized — a gap that did not exist eighteen months ago, and one that inverts the conventional ranking of the two.
What to watch
- The S-1 going public. Confidential filings are made public roughly 15 days before a roadshow. That document converts “adjusted” into GAAP and shows gross margin — the single most important unknown for any AI lab.
- Compute commitments in the footnotes. Multi-year capacity deals are the industry’s largest off-income-statement obligation. Their size and duration will matter more than the revenue headline.
- Whether Q3 holds the exit rate. A $65 billion annualized pace implies a September quarter above $16 billion. Anything materially under that means July was a peak, not a pace.
- Customer concentration. Management credits enterprise adoption. Enterprise revenue is stickier than consumer — and more concentrated. Customer counts matter here.
Preliminary figures can move before a filing, and every number above is exactly that: preliminary and self-reported ahead of a listing. What is not preliminary is the direction. A company that made $787 million in a quarter last year made more than $11.5 billion in the same quarter this year. Whatever Wall Street argues about between now and the opening bell, it will not be arguing about whether the demand showed up.