On 17 August it was reported that Anthropic's annualised revenue run rate had passed $65 billion. Three months earlier the company announced a raise at a $965 billion valuation. This piece takes both numbers apart.
Errors or inaccuracies are possible. Spotted something that looks wrong? Write to me and I will correct it.
A Preliminary Note That Has to Come First
Neither Anthropic nor OpenAI has audited, public financial statements. Both are private companies.
The figure quoted for both is an "annualised revenue run rate" - the last month's revenue multiplied by twelve. It is not annual turnover, and it is not examined by an auditor. It is also unusually sensitive in a fast-growing business: one strong month lifts it, and a weak one pulls it down.
And different companies calculate it differently. Any comparison between them is an estimate, not a measurement.
The Path of the Valuation
The two most recent rounds were announced by Anthropic itself:
| Series G | Series H | |
|---|---|---|
| Announcement date | 12 February 2026 | 28 May 2026 |
| Raised | $30 billion | $65 billion |
| Post-money valuation | $380 billion | $965 billion |
| Run rate at announcement | $14 billion | $47 billion |
That is, within three and a half months, the valuation grew 2.5-fold.
Series H leads: Altimeter, Dragoneer, Greenoaks and Sequoia. Alongside them as co-leads: Capital Group, Coatue, D1, GIC, ICONIQ and XN.
And more interesting than the identity of the funds are the infrastructure partners who joined the round: Micron, Samsung and SK hynix. Three memory makers.
And this is a point worth pausing on, because it explains something about the nature of the raise.
The Series H announcement also set out the compute agreements: up to 5 gigawatts from Amazon, 5 gigawatts of next-generation TPU capacity from Google and Broadcom, and access to GPU capacity from SpaceX. Plus about $15 billion previously committed by hyperscalers, including $5 billion from Amazon.
When memory makers and cloud operators come in as investors in a company that buys capacity from them, the line between a capital raise and a commercial agreement blurs. This is neither improper nor unusual in this industry - but it does mean a valuation set in such a round also reflects commercial relationships, and not a clean valuation alone.
And What Was Reported This Week
Per a report of 17 August, the annualised revenue run rate passed $65 billion at the end of July.
| Date | Annualised run rate |
|---|---|
| End of 2025 | about $9 billion |
| February 2026 | $14 billion |
| May 2026 | $47 billion |
| End of July 2026 | over $65 billion |
From 9 to 65 in seven months. And per the same report, investors expect the year to close at a run rate of $100 billion to $120 billion.
As for the listing: the report notes an expectation of an IPO as soon as the autumn, at a valuation of $2 trillion or more. That is an investor expectation, not a company announcement - and Anthropic has published no announcement of an IPO date or valuation.
The Comparison with OpenAI
And here care is needed, because the figures come from different dates.
| Anthropic | OpenAI | |
|---|---|---|
| Latest raise | $65 billion, May | $122 billion, March |
| Valuation | $965 billion | $852 billion |
| Run rate, end of 2025 | about $9 billion | about $20 billion |
| Run rate, August 2026 | over $65 billion | about $40 billion |
The OpenAI figures in the last row come from that same 17 August report, which set the two companies side by side.
And the interesting comparison is not in the levels but in the rate.
OpenAI doubled its revenue - from about $20 billion to about $40 billion.
Anthropic grew its revenue sevenfold - from about $9 billion to over $65 billion.
At the start of the period OpenAI was roughly twice the size. At the end, Anthropic is roughly 1.6 times the size. The reversal happened in under eight months.
And on valuation: OpenAI raised in March at $852 billion; Anthropic raised in May at $965 billion. The gap in valuation is modest relative to the gap in rate - partly because the raises were two months apart, in a period when the numbers were moving fast.
The Revenue Multiples, and What They Do Not Say
Taking the latest valuation and dividing by the current run rate:
| Calculation | Multiple | |
|---|---|---|
| Anthropic | 965 divided by 65 | about 14.8 |
| OpenAI | 852 divided by 40 | about 21.3 |
And two caveats have to travel with that table.
The first: the valuation was set in a raise made when the run rate was lower. The multiple computed here divides a past valuation by present revenue - so it is lower than the multiple actually struck in the deal.
And the second, which is the important one: there is no profit line here. A revenue multiple on a company whose profitability is unknown, and which is committing to compute agreements measured in gigawatts, does not say whether the price is reasonable. It says only how much is being paid per dollar of revenue.
What Is Missing from These Numbers
Three things are absent, and each of them is material:
Profitability. No profit line has been published, so there is no telling whether this growth comes with a margin or against one.
The cost structure. The compute agreements are reported in gigawatts, not dollars. A commitment to capacity is a long-dated financial commitment, and its scale is not known.
And the revenue mix. Anthropic noted in February that the number of customers spending over $100,000 a year had grown sevenfold in a year, and that over 500 customers spend more than $1 million. That points to an enterprise business - but the full split between enterprise, developers and end consumers has not been published.
הזווית שלי
דעה אישית של אילן אברמוב - לא ייעוץ ולא המלצה
The number I read here is not the valuation but the rate at which it changed.
$965 billion is a figure hard to hold in the head, which makes it easy to argue about without saying anything. What is hard to argue with is that the run rate went from $9 billion to $65 billion in seven months. That is not a valuation estimate - it is actual selling.
And what interests me in the comparison with OpenAI is not which is larger, but how fast the order reversed. At the end of 2025 OpenAI was twice the size. Today the ratio is inverted. In an industry where everyone talks about a competitive moat, a reversal like that inside eight months says the moat, if it exists, is far shallower than is usually assumed.
And my reservation is substantive rather than polite. There are no audited statements, no profit line, and no expense breakdown. A run rate is the easiest number to grow and the hardest to verify - it is derived from a single month, and multiplied by twelve.
And what I would look for first in a prospectus, if and when: not the revenue, but the operating commitments to the compute suppliers. A company that has committed to gigawatts of capacity has also bought a large fixed cost, and that is the side of the equation a run rate does not show.
And when an investor in the company is also its supplier, as with the memory makers and the cloud providers - a valuation set in such a round deserves to be read with additional care.
(An important note: this is my personal opinion only, and nothing here is a recommendation to take any action.)
