Speed Is Not the Thesis: How to Read Anthropic's Numbers Before the IPO
- jcarvallo4
- Aug 24
- 7 min read
Anthropic built the fastest revenue ramp in the history of enterprise software. That is a verifiable fact. Whether it justifies a two-trillion-dollar valuation is a different question — and the one worth answering before October.
There is an article circulating that captures the market's current mood on Anthropic rather well: the company has left OpenAI behind, has no comparable competitor, and should be aiming at a three-trillion-dollar valuation before year-end. It is an enthusiastic piece and, on several of its data points, an accurate one.
The problem is not the data. It is what the enthusiasm does with it.
Anthropic is, almost certainly, the most extraordinary corporate growth story of the past decade. It is also a company that will shortly ask public investors — including, very probably, pension funds — to accept a sales multiple no one has ever paid for a business of this size. Both things are true at once, and the discipline of holding them apart is exactly what separates analysis from cheerleading.
1. What is genuinely extraordinary
Start by giving the fact its due, because it is remarkable.
In January 2024, Anthropic was running at an annualized revenue run-rate of roughly $87 million. By the end of July 2026, according to figures the company shared with investors and that Bloomberg reported first, that number had passed $65 billion.
The path between those endpoints matters as much as the endpoints themselves: roughly $9 billion at the close of 2025, $14 billion in February 2026, $19 billion in March, $30 billion in April, $47 billion in May, and $65 billion in July. That is $18 billion of run-rate added in two months.
For scale: Salesforce took roughly two decades to reach $30 billion in annual revenue. Anthropic crossed that run-rate mark in a little over three years from its first booked dollar.
The engine is identifiable rather than diffuse. Around 80% of revenue comes from enterprise customers — more than 300,000 organizations — and within that block Claude Code, the coding tool launched in mid-2025, stands out, reaching $1 billion of run-rate in about six months. This is not a mass-consumer phenomenon. It is corporate adoption with operating budget behind it.
2. The crossover with OpenAI
In April 2026 something happened that looked improbable two years ago: Anthropic passed OpenAI on revenue run-rate.
At the close of 2025 the ratio was $9 billion to $20 billion. By August 2026, per Bloomberg and CNBC reporting, it is $65 billion to $40 billion.
A methodological warning is warranted here. The two companies do not calculate the metric the same way. Anthropic reports on a gross basis and includes marketplace partner commissions; OpenAI reports net of Microsoft's share. The real gap is probably narrower than the apparent one. The crossover happened, but the scale of the lead is inflated by accounting, not by operating reality.
3. The distinction that decides everything: run-rate is not revenue
This is where most of the coverage goes off the rails, and where an investor needs to listen carefully.
Annualized run-rate (ARR, in the jargon) is a startup convention: take the most recent month and multiply by twelve. It is an extrapolated snapshot, not a result. It is useful for measuring momentum. It is not useful for calculating multiples.
Anthropic's booked calendar-2026 revenue, according to third-party estimates, will most likely land between $20 billion and $26 billion. The gap with the $65 billion run-rate is not an accounting trick — it is the arithmetic of a company growing very fast — but it completely changes the implied valuation.
The company closed its Series H in May 2026, raising $65 billion at a post-money valuation of $965 billion. On run-rate, that is roughly 15 times sales: expensive, but arguable for an asset growing this way. On estimated booked revenue, it is roughly 42 times.
And if one takes seriously the $2 trillion figure the press has reported as investors' target for the October listing — which is not a number the company has confirmed — we are talking about roughly 31 times run-rate, or 87 times booked revenue.
Eighty-seven times sales. That is the number to keep in mind, not the 15x.
4. The serious bull case: unit economics are improving
Having said that, it would be dishonest to present only the critique. There is one data point that supports the bull case far better than any growth chart, and it is the one that should matter most in the S-1 when it becomes public.
According to internal projections reported by The Wall Street Journal, first-quarter 2026 revenue was $4.8 billion and second-quarter revenue was projected at around $10.9 billion. Over that same stretch, compute cost per dollar of revenue fell from $0.71 to $0.56.
That is what matters. Doubling revenue while lowering the unit cost of serving it is the definition of operating leverage, and it is precisely what most AI skeptics argued would not happen. On that basis, Anthropic projected its first quarter of positive operating income — roughly $559 million — while OpenAI projects losses of around $14 billion for the same year.
If that cost curve holds, the multiple debate changes in nature. If it does not hold, no multiple survives.
5. What the enthusiasm leaves out
Four risks no promotional note mentions and that a prospectus will have to address.
a) The financing is partly circular. Anthropic's compute cost is low in part because Google and Amazon absorb a share of the bill through equity-backed cloud commitments. In other words: the suppliers are also shareholders, and the shareholders are also indirect customers of the ecosystem they finance. The company also carries on the order of $80 billion in committed infrastructure spend through 2029. If that structure is renegotiated or the chip mix changes, the $0.56 re-inflates. This is not an accusation; it is the variable to watch.
b) Profitability is fragile and unaudited. Anthropic itself warned investors that planned infrastructure spending for late 2026 and 2027 is likely to push operating results back into negative territory. One positive quarter is not a profitable company, and there are critiques — the most widely read being Ed Zitron's — arguing that the figure excludes items, among them stock-based compensation and certain equity-funded commitments, that public-company accounting would have to reflect. When the S-1 goes public we will have audited figures for the first time. Until then, all of the above is information the company supplied to private investors.
c) Concentration. Eighty percent of revenue comes from enterprises, and a substantial share of recent growth comes from a single product in a single category: coding assistance. It is an enormous and expanding category, but it is also the most competitive and the most exposed to price pressure from low-cost open-weight models. A company valued as a diversified platform that in practice depends on one product line is a thesis risk, not a quarterly one.
d) Political risk. Anthropic was excluded from Department of Defense contracting and is litigating against the US administration over that exclusion; the case remains open. In June 2026, Commerce Department export controls forced a temporary suspension of access to its highest-capability models. For a company whose advantage depends on holding the technological frontier, regulatory exposure is not noise — it is a first-order variable.
6. Structural versus cyclical
As always, it is worth separating which part of this story is a regime change and which part is where we sit in the cycle.
Structural: the shift of corporate software spending toward AI tooling, the improving unit cost of inference, and the fact that enterprise adoption — unlike consumer adoption — comes with recurring budget and high switching costs.
Cyclical: the market's appetite for long-duration, cash-negative assets, the abundance of capital willing to fund infrastructure without demanding near-term return, and the listing window that opened in 2026.
The bull case on Anthropic depends on the first. The two-trillion-dollar multiple depends on the second.
7. For the investor: what can actually be done today
Here is the practical part, which is where the enthusiasm is doing the most damage.
You cannot buy Anthropic today. There is no listed share. And in May 2026 the company warned eight secondary-market platforms — among them Forge, Hiive and Sydecar — that unauthorized transfers of its shares are void and will not be recognized on its books. Any product offering "pre-IPO exposure to Anthropic" via special purpose vehicles, tokenization or syndicated interests should be treated with extreme suspicion. The risk is not only price: it is that the instrument may not represent what it claims to represent.
And for anyone tempted to buy on day one, the closest precedent is instructive. SpaceX listed on 12 June 2026 at $135 per share, at a valuation of roughly $1.77 trillion. The stock touched $225.64 intraday four days later and by late June was trading around $153. Whoever bought at the peak lost close to a third in two weeks, with the long-term thesis fully intact. Asset quality and entry price are two separate decisions.
Conclusion
Anthropic did something without precedent, and it deserves analytical recognition rather than reflexive skepticism. Going from $87 million to $65 billion of run-rate in thirty-one months, improving unit costs along the way, is a first-order industrial achievement.
But the conclusion that follows is not "you should buy." It is more demanding than that. With an implied multiple ranging from 15 to 87 times sales depending on the denominator, with $80 billion committed to infrastructure, with profitability the company itself warns may not hold, and without a single audited financial statement available to the public, the useful question is not whether Anthropic is a great company. It almost certainly is.
The question is at what price it stops being a great investment. That question gets answered when the prospectus becomes public — not before — and anyone answering it today with conviction is opining, not analyzing.
It is a matter of weeks. The document is worth waiting for.
This content is for informational and educational purposes only. It does not constitute investment advice, an offer, or a recommendation to buy or sell any security, and it does not take into account the objectives, financial situation or particular needs of any investor. Investments in pre-IPO companies and in initial public offerings involve significant risks, including the possible loss of the entire principal. The figures cited come from press reports and from information the company supplied to private investors; they have not been audited or independently verified. Past performance does not guarantee future results.
Sources
Bloomberg — Anthropic's Annualized Revenue Tops $65 Billion Before IPO (17 August 2026)
CNBC — Anthropic tells investors annualized revenue run rate climbed to $65 billion in July (17 August 2026)
CNBC — Anthropic confidentially files IPO prospectus with SEC (1 June 2026)
Axios — Anthropic's revenue run rate reportedly surpasses $65 billion pre-IPO (17 August 2026)
The Wall Street Journal — Q2 2026 revenue and operating income projections (May 2026)
Reuters — Anthropic annualized revenue projections
Financial Times — investor expectations for year-end 2026
Anthropic — Series G (February 2026) and Series H (May 2026) announcements
Anthropic — statement on Fable and Mythos model access (July 2026)
CNBC — SpaceX IPO takeaways (12 June 2026) and subsequent trading coverage
Ed Zitron, Where's Your Ed At — Anthropic's "Profitability" Swindle (21 May 2026)
Michael Spencer, AI Supremacy — We've never seen an Anthropic before (20 August 2026)
Comments