Anthropic IPO valuation: a five-factor screen of the $2T prospectus

Reuters reported a confidential Anthropic IPO prospectus that asks public investors to underwrite more than $2 trillion. Last year the company took in $4.6 billion and booked a $42 billion loss. The question is not whether Claude is a remarkable product. It is whether that Anthropic IPO valuation is a price for the business as it stands, or a price for 2028 arriving on time.

The $42 billion loss is mostly not the operating loss

Start with the headline because it is doing too much work. About $34 billion of the $42 billion is an accounting charge tied to earlier funding deals that can convert into Anthropic shares later. As the private mark rose — from $61.5 billion in March 2025 to $965 billion this May — those promises became more valuable on paper. The accountant books the jump as a loss even though no cash left the till. That is a real line on an income statement. It is not a description of how the business spent last year.

The number that describes the business is the operating loss: $8.06 billion, up from $2.98 billion the year before. Anthropic brought in $4.6 billion and spent $12.65 billion. Compute — chips and data centers that train and run Claude — was $7.33 billion of that spend. For every dollar a customer paid, the company spent about $2.75; roughly $1.60 of that went to compute. The rest of the $42 billion is the price going up. The operating book still lost $8 billion, and it lost it buying compute.

How we screen the Anthropic IPO valuation

Every two weeks we rescore roughly 3,600 US-listed companies on the same five factors — growth, estimate revisions, profitability, valuation, and momentum — each measured against the company's own sector. The method is on the strategy page. Anthropic is not listed, and it does not yet have a sector in that universe. It may be the first frontier AI lab to list. The honest move is still to run the same questions on the prospectus numbers rather than invent a special screen because the story is large.

That is what good stock research looks like: the business, the gap at the current price, valuation tied to economics, specific downside, and a written trail of what would change the conclusion. A $2 trillion ask is not a reason to skip the work. It is a reason to do it slowly.

Growth — pass

Revenue grew twelvefold in 2025. Bloomberg then reported more than $11.5 billion of revenue in the second quarter of 2026 against $787 million a year earlier. One quarter this year brought in about two and a half times all of last year. The prospectus, as described in the video, ties usage — and therefore revenue — to new models, with a continuous and overlapping cadence of releases as the cost of staying at the frontier.

The timing around that claim is awkward in a useful way. In early September the CEO published a long essay asking the industry to pace the frontier. Ten days later Anthropic released Opus 5.5, and on the Monday Reuters reported the prospectus it released Sonnet 5.5. On Artificial Analysis's intelligence index those were the two best-scoring models released to date in that snapshot (Opus 58, Sonnet 56, against OpenAI's GPT-6 Astra at 53). Growth at this scale passes any bar we use on listed names. It does not, by itself, set the price.

Estimate revisions — pass

No sell-side roster covers Anthropic yet. The closest substitute is whether the company's own numbers keep getting raised. The revenue run rate — the current pace annualized — was about $9 billion at the end of 2025, more than $47 billion by May, and $65 billion by the end of July. It told investors to expect at least $10.9 billion in the second quarter, then came in above $11.5 billion. That is the shape we look for in revisions: the outlook is moving up, not being defended. It passes. It is still a private company talking about its own run rate, not a consensus of independent estimates.

Profitability — fail

Full-year 2025 operating margin was about minus 175%. This year the print is turning. For the second quarter Anthropic projected its first operating profit, about $559 million, and subsequent reporting has it on track for a second profitable quarter. That profit is adjusted. Adjusted usually means some costs sit outside the headline — typically stock paid to employees. A public income statement has to include them. We score the full number. There is not a full profitable year yet. Profitability fails on that test, even if the sequential path is the first sign the cost of compute might be growing slower than revenue.

Valuation — fail

Take the second quarter, multiply by four, and Anthropic is running at about $46 billion a year. Two trillion is about 43 times that. In the comps we used in the video, Palantir — one of the more expensive listed software names — traded around 53 times this year's expected revenue; SpaceX and Cloudflare around 41.6 times. The unusual step, per Reuters, is that the bankers are not selling that 43-times print. They are pricing Anthropic on revenue two years out. The company is projecting $190 to $200 billion for 2028. On that number, $2 trillion is about 10 times sales.

That is like pricing a house on the rent it will earn after you build the third floor. The floor might get built. You still pay for it today. Our framework wants a margin of safety: a price below a conservative view of what the business is worth now, because the underwriter can be wrong and because markets can stay skeptical longer than a model likes. Priced on 2028, there is not one. Valuation fails. That is the whole argument against treating the Anthropic IPO valuation as a bargain because a 2028 multiple looks tidier than a 2026 one.

Momentum — cannot score

Momentum needs a public stock. What exists instead is a string of private marks: $61.5 billion in March 2025, $183 billion that September, $380 billion in February, $965 billion in May, and now more than $2 trillion. That is roughly 32 times in eighteen months. The nearest listed comparison in the video was SpaceX, which went public in June at $135 a share, closed its first day at $160, and later traded around $147: the IPO allocation was up; the first-day excitement was not. Until Anthropic trades, momentum is a blank. We leave it blank.

Two passes, two fails, and one we cannot score. On listed names that shape is usually a maybe. The number that decides this book is not a fifth factor. It is the bill already signed.

The $518 billion compute bill

Anthropic expects to spend $518 billion on computing infrastructure over about a decade, with six partners. It held $20.28 billion in cash and short-term investments at the end of last year. The largest pieces in the video's read of the prospectus are about $161 billion in equipment leases tied to Broadcom, $111 billion with Google, $110 billion with Amazon, and $31 billion with Microsoft. About 80% of the total is non-cancellable, or has to be paid whether Anthropic uses the compute or not. In the prospectus's own language as quoted on the tape: if actual spend falls short, it must pay Google the difference. The Amazon deal works the same way.

The first thing we check in any business is whether cash flow can fund the plan without going back to the capital markets. At this scale, Anthropic's cannot. That is what the IPO is for. A $2 trillion mark is not only a story about Claude. It is a story about who funds a decade of chips when the customer checks are still a fraction of the committed spend.

The counterparties are not only suppliers. Amazon and Google have each invested billions in Anthropic. They sell the compute. They sell Claude to their own cloud customers. They build competing models. The prospectus calls them investors, customers, cloud providers, distributors, and competitors at once, with incentives that may not be fully aligned. Customers are concentrated too: nearly a quarter of last year's revenue came from just two of them, and the filing warns that many of its largest clients are not locked into long-term contracts. The bill is fixed for ten years. A quarter of the revenue is not.

Then there are the risk factors — 80 of the prospectus's 261 pages, against about 38 of 277 in SpaceX's filing. Anthropic warns that advanced AI could pose catastrophic or existential risks to humanity, and that its models could show self-preserving behavior, including attempts to resist shutdown. Reuters noted that few companies have ever put a warning like that in an IPO document. It is not a valuation input in our model. It is part of the audit trail: the company is asking public shareholders to fund a product it describes, in its own risk section, as potentially existential.

The strongest case against our verdict

The pushback is the growth itself. Anthropic says its revenue run rate grew more than tenfold a year in each of the three years through early 2026. If it hits even the low end of its 2028 forecast, $2 trillion is about 10 times revenue — cheaper, on that multiple, than Palantir was in the comps we used. Bankers are betting that revenue rises faster than the cost of compute, and the second quarter is the first sign that it can. If they are right, $2 trillion will look cheap in hindsight. That is a coherent thesis. It is also a thesis that requires 2028 to show up on time, with margins the full-year statements have not yet shown, while a mostly non-cancellable $518 billion bill is already on the table.

Index investors already sit on one side of that bill. A market-cap S&P 500 fund holds Amazon, Google, and Microsoft. Between them, Anthropic has promised more than $250 billion in minimum spending. Amazon and Google are also shareholders. The fund already rides on Anthropic paying. Direct IPO stock would be a second, more concentrated claim on the same stack — lab equity on top of the suppliers and cloud platforms already in the index. That is a portfolio construction question, not a forecast.

For our own model, nothing changes yet. It scores listed companies with public filings. When Anthropic's filing goes public and the stock trades, it enters the same screen as the other 3,600. Until then the honest sentence is the one the video ends on: go back to the $42 billion loss; most of it was accounting. The numbers that matter are $518 billion already signed, and a $2 trillion price that assumes 2028 arrives on schedule.

Two factors pass, two fail, and one cannot be scored yet. On our framework, Anthropic is an exceptional business that does not clear the bar at that Anthropic IPO valuation. Every two weeks we publish the one US-listed name that does — full written thesis, the risks, and what would prove us wrong. Returns are public, losers included, on the track record. Through December 1, 2026, the founding year is $250; the list is on pricing.