Article — "Anthropic Races to an October IPO at $2T+ as Revenue Hits $65B ARR": Article — "Anthropic Races to an October IPO at $2T+ as Revenue Hits $65B ARR": Two numbers in the same reporting package contradict each other, and almost nobody has noticed.
The first: Anthropic's annualized revenue run rate hit $65 billion at the end of July, up from $47 billion in May and just $9 billion at the close of 2025. That's $18 billion of annualized revenue added in two months.
The second: the $2 trillion IPO valuation that investors are pitching for an October float isn't built on the $65 billion figure at all. It's built on a number from 2028.
Hold those two facts next to each other and the entire story reframes. The run rate is the seduction. The valuation is the bet. They are not the same math, and conflating them is the mistake almost every headline is making this week.
The run rate is the seduction. The valuation is the bet. They are not the same math, and conflating them is the mistake almost every headline is making this week.
Last week, this newsletter flagged Anthropic's $965 billion Series H mark from May — a $65 billion round that made it the most valuable AI startup on paper — and asked whether the $1 trillion private-secondary level would hold or prove the top. This week that question got forced into daylight. The company is now targeting a public print at $2 trillion or more in October, per the Financial Times, which would make it the largest IPO in history, reportedly eclipsing SpaceX's $1.77 trillion June record.
Start with the run rate, because it's the number doing the emotional work.
"Annualized revenue run rate" is a projection of a full year's revenue based on a recent, shorter snapshot. It is not cash in the bank. Anthropic's $65 billion was extrapolated off roughly two months of activity — the same two months in which the figure jumped $18 billion.
TechCrunch's own reply guy said it plainly under the post: the $65 billion figure "was built on just two months of $18 billion growth, which itself followed a slower rise from $9 billion." The growth is real. But run rate is a lens that magnifies the most recent, fastest stretch and treats it as the steady state.
There's a harder number underneath, and it's genuinely impressive: preliminary Q2 revenue of $11.5 billion, reportedly a 14-fold jump from $787 million a year earlier, and — for the first time — positive adjusted operating profit reportedly of $559 million. Anthropic reportedly turned profitable roughly two years ahead of the forecast it gave investors.
That is the fact that should move the story. A frontier lab printing operating profit at all upends the assumption that these companies are cash bonfires. OpenAI, by contrast, sits at a $40 billion run rate and makes no such claim.
But note the word "adjusted." And note what the $559 million sits in front of: the training bill for the next frontier model. According to one report, while unit inference cost keeps falling with scale, the investment in training the next generation of cutting-edge models is still rising. One quarter of thin adjusted profit is not a cycle. It's a data point.
Now the turn — the part the headlines skip.
Per Reuters, cited across the reporting, banks and investors in the pre-listing discussions are reportedly using EV-to-revenue multiples and applying them directly to Anthropic's 2028 revenue forecast: $190 billion to $200 billion. Not 2026 revenue. Not the run rate. The number the company expects to hit two full years out.
Using revenue instead of profit to value a high-growth software company is ordinary. Reaching two years forward to do it is not. Wall Street is, in the words of one report, pricing "Anthropic two years from now" to decide what Anthropic is worth today.
This isn't unprecedented, and that's the point. Cerebras's backers cited 2028 revenue expectations before its IPO this year. SpaceX's financial forecasts reportedly ran out to 2029 before its record June float. Forward-pricing a private-to-public transition on a multi-year projection is now an established trick for companies growing faster than any trailing multiple can rationalize.
The mechanics are clean enough. If you value Anthropic on today's profits, the GPU spend, the data centers, the training runs and the talent make it look absurdly expensive. If you assume that infrastructure eventually supports a platform several times larger at improving margins, today's spend is just the cost of the land grab. The $2 trillion ask is a wager on the second scenario.
Which brings us to the asterisk that no one has disaggregated.
Buried in the bull case is a line worth reading twice: Anthropic's high revenue has "benefited to a certain extent from subsidies from upstream computing power manufacturers and procurement support from strategic investors," and part of it is "of related transaction nature."
Translate that. Amazon has reportedly committed up to $8 billion and is both a strategic shareholder and Anthropic's compute partner. Google has reportedly put in more than $1 billion and seeded the company at the start. When a compute provider subsidizes your inference costs and also buys your product — and also owns a slice of you — some fraction of that gleaming run rate is money moving in a circle.
No one has yet separated the clean, market-demand revenue from the strategically-motivated and compute-subsidized revenue. Until someone does, the $65 billion is a blended figure, and the 2028 forecast built on top of it inherits whatever impurity sits underneath.
There's a related-methodology wrinkle, too. Anthropic and OpenAI may calculate their revenue metrics differently — TechCrunch flags this directly — which means the tidy "Anthropic $65B beats OpenAI $40B" comparison animating investors may not be comparing like with like.
Then there's the demand side, which is not guaranteed to hold. Anthropic's main model reportedly costs more than 2.5 times OpenAI's flagship to call, and Chinese open-source models a fraction of that. Some enterprise customers have reportedly already moved to cheaper alternatives. A reported 32% API market share is a strong hand, but it's a hand held at a price premium in a market where the price of intelligence keeps falling.
Here's why this print matters beyond Anthropic.
This is the first time a frontier lab converts a private secondary mark into a public print. The $965 billion Series H was reportedly set by roughly six sophisticated backers with information advantages and every incentive to mark up. A public IPO forces that number in front of index funds, pensions and retail — buyers who have to underwrite the 2028 story on faith rather than access.
The question this newsletter left open on August 4 — does the private mark hold or prove the top — is now a $2 trillion public ask. The doubling from $965 billion to $2 trillion in roughly three months wasn't driven by a change in the business. It was driven by a change in which forecast the market decided to price.
So the real test in October is narrow and specific. It is not whether Anthropic is growing — it plainly is, faster than almost anything on record. It is whether public investors will underwrite a 2028 number on the same faith six insiders extended it, and whether that number, when the S-1 finally strips the confidentiality off, turns out to be built on clean market demand or partly on dollars that Anthropic's own backers recycled back into it.
The run rate captivated the market. The forecast will price the deal. The footnote — how much of both is real, unsubsidized, arm's-length revenue — is the thing that decides whether $2 trillion is a floor or the top.
We'll know more when the S-1 goes public. For now, the most important number in the Anthropic story is the one nobody has published: the share of that $65 billion that isn't coming, in some form, from the same people setting the price.