Anthropic Goes Public: A $1.4 Trillion Public Stress Test

11/19 2026 399

On November 19, 2026, Nasdaq will welcome a newcomer under the ticker "ANT." The company, founded just five years ago, has set an offering price range of $70 to $78, implying a fully diluted valuation of roughly $1.4 trillion at the top end, and public reports indicate that IPO orders are already close to seven times the shares on offer.

What makes Anthropic's listing worth pausing to examine is not that it has set another record, but that it has, for the first time, placed a question that previously existed only in private markets onto the public market's pricing table: when the tension among a frontier AI company's revenue growth, cost structure, and long-term commitments is broken down into quarterly reports, what kind of "AI business" is the market actually willing to pay for?

01. Behind the sevenfold oversubscription, the choice of denominator is more telling than the frenzy over the numerator

The market's enthusiasm for Anthropic hardly needs elaboration. Led by Goldman Sachs and Morgan Stanley, with more than twenty banks participating in underwriting, Nvidia subscribed about $10 billion in a concurrent private placement, and Amazon and Google will respectively hold about 14.7% and 13.7% of shares after the IPO, totaling about 28.4%.

The subscription multiple is close to seven times, and the offering price may even be set above the top of the range. All of this points to a familiar story: scarce top-tier AI assets, plus pent-up allocation demand.

However, the $1.4 trillion valuation itself already represents a cooldown: it is about 30% lower than the roughly $2 trillion market expectation this summer, and also below the valuation record set by SpaceX when it listed in June this year. But what truly deserves attention is not "being unable to get an allocation," but "at what price allocations are being chased."

If the $1.4 trillion is divided by the annualized revenue run rate of $81 billion at the end of October, the price-to-sales ratio is around 17 times, which does not look unreasonable; if replaced with the actual GAAP revenue of about $35.2 billion over the past four quarters, the multiple immediately jumps to around 40 times; and if, according to the company's roadshow materials, all compute and infrastructure costs are deducted from revenue, only about 43% remains, corresponding to a multiple of nearly 93 times.

Underwriters and most buy-side investors choose the most favorable denominator, which is understandable in itself, but it means that what the public market takes over is not a static "cheapness," but a dynamic assumption that places extremely high demands on growth sustainability.

The subtlety of IPO pricing lies precisely here: the market has set a valuation using the most optimistic measure, yet it demands that the company prove quarter by quarter after listing, using the strictest standards, that it deserves it.

02. Revenue is real, but so is the fact that "every dollar of revenue first pays for compute"

Anthropic's growth data would be astonishing in any industry. Revenue in the first half of 2026 was $16.286 billion, up about 13 times year-on-year; ARR rose from $1 billion at the beginning of 2025 all the way to $81 billion by the end of October.

More than 11,900 customers spent at least $100,000 on Claude over the past year, and net revenue retention reached 900%, indicating that customers who have already come in are not testing the waters, but deepening their use.

The enterprise market data is equally persuasive: Ramp statistics show that among enterprises purchasing AI for the first time in October this year, 63% chose Anthropic, compared with only 25% a year earlier; preference skew in new purchases says more about momentum than existing share.

However, the other side of the revenue structure is equally clear. Revenue billed by API usage accounts for 80%, meaning revenue is tightly bound to customers' call frequency and scale, rather than coming from locked-in long-term contracts.

There is another detail in the prospectus that is easy to overlook: model versions released in the six months ending September 30, 2026, contributed more than 80% of revenue in the three months before that date. Revenue growth is highly dependent on new models continuing to lead; once the iteration pace slows, customer usage and willingness to pay will come under pressure at the same time.

A more direct signal comes from unit economics: compute costs in the third quarter grew 77% quarter-on-quarter, while revenue grew 47% quarter-on-quarter, and revenue recovered per dollar of compute fell from $2.08 to $1.72.

Growth is still fast, but at the margin, each dollar of revenue is becoming more "compute-heavy," not less.

This change has already shown up in the income statement: in the third quarter, under GAAP, the company swung from profit to loss, with an operating loss of $567 million, compared with operating profit of $974 million in the second quarter; the book net loss of $104 billion in the first three quarters mainly came from fair value remeasurement of convertible bonds and derivative liabilities and does not correspond to cash outflow.

03. Nearly $900 billion in commitments, and revenue without locked-in duration

If the preceding financial discussion still remained at the level of "expensive or not," then compute commitments push the question to the level of "can it bear the weight."

As of 2045, Anthropic has signed long-term procurement and lease commitments of nearly $900 billion, about half of which must be paid by the end of 2030, with about $69.8 billion due within the next twelve months.

The rigidity of these commitments deserves closer scrutiny: many contracts are non-cancelable, or require payment regardless of actual usage.

On the counterparty side, the contract with Amazon is no less than $110 billion and runs through 2036, with Google no less than $114.5 billion, with Microsoft about $44.9 billion, with Broadcom for TPU equipment leasing about $161.2 billion, and with SpaceX's AI division a contract ceiling of about $120.5 billion.

In contrast, the elasticity on the revenue side is evident. A large number of enterprise customers have not signed long-term usage contracts; revenue is calculated based on actual AI usage, and customers can increase or decrease consumption as demand changes.

In its pre-IPO credit analysis, Egan-Jones directly listed this mismatch as a key focus: Anthropic must lock in GPU and data center capacity in advance, but customers are under no obligation to lock in spending at the same time. The company's own prospectus also acknowledges that multiple infrastructure transactions will bring "structural time mismatch and liquidity risk," with payments required before compute capacity comes online.

CEO Dario Amodei called compute "the most core operating challenge" during the roadshow and admitted that budgeting amid rapid iteration of technology and revenue involves enormous uncertainty. This is not a matter of one company's management capability, but the inherent time structure of the frontier model business: costs are front-loaded in years and decades, while revenue fluctuates in months and usage.

The public market will vote on this structure every quarter. The combined stake of nearly 30% held by Amazon and Google raises governance questions from another direction. These two companies are both Anthropic's most important cloud and chip partners and giants with their own competitive AI products.

Related-party transactions, strategic dependence, and potential conflicts of interest will not disappear because of listing; they will only become more visible due to heavier disclosure obligations. What public market investors receive is not only a ticket to AI growth, but also a map of related-party relationships that requires continuous scrutiny. Anthropic's IPO is destined to be compared with SpaceX's record and repeatedly cited in the debate over whether there is an "AI bubble."

But the judgment worth taking away may be more plain: the company has spent five years proving that large models can generate real and astonishing revenue, and what it must prove next, quarter by quarter in the public market, is whether, when compute commitments are measured in hundreds of billions of dollars, customer usage in months, and model leadership in versions, this business can maintain its rhythm amid time mismatch. What $1.4 trillion buys is not today's Anthropic, but the probability that it will continue not to make mistakes over the next five years.

Listing is not the finish line; it is turning that probability from a private discussion in the primary market into a public answer updated every second on a Nasdaq screen.

- End -

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.