On November 19, Anthropic rang the bell at Nasdaq under the ticker ANT. That day it set two records: the largest IPO by fundraising globally, and the first time an AI foundation model company laid out its complete financial statements in the public market.
But what really made me stop and do the math was how, the next day, different media outlets described the same figure differently—some said the fundraising was $81.3 billion, while the official announcement put it at $93.5 billion. Both numbers are correct; the $12.2 billion difference happens to be exactly one 'over-allotment.' This fact alone is worth a read for enterprise digitalization leaders: even the size of an IPO has different bases of measurement, so when it comes to the 'supplier ledger' we use with AI, we need even more to learn to do the calculations ourselves.
I. 81.3 billion or 93.5 billion: an IPO, accounts that can be reconciled in four lines
First, let’s lay out the basis. The issue price is $83 per share; the only difference between the two books is "whether to count the extra shares for the underwriters":
| Caliber | Number of shares | Unit price | Amount raised | Source and Date |
|---|---|---|---|---|
| Media Day-One Talking Points | 980 million Class A shares | $83 | Approximately $81.34 billion | Xinhua News Agency/Securities Times 11-19 |
| Official delivery standards | 1.127 billion Class A shares | $83 | Approx. $93.54 billion | Anthropic Official Announcement 11-19 |
| Difference = Over-allotment | 147 million shares | $83 | Approximately $12.2 billion | The underwriter fully exercises the option |
| Concurrent private placement | 313.25 million Class N shares | $83 | Approximately $26.0 billion | AMD/Broadcom/Google/Micron/Microsoft/NVIDIA/TSMC |
| Official total | — | — | Approximately $119.54 billion | IPO + Private Placement, Official Wording |
Let's check the math: 980 million × 83 = 81.34 billion; adding the fully exercised over-allotment of 147 million shares (147 million × 83 = 12.20 billion), it comes to exactly 93.54 billion; the private placement of 313.25 million shares × 83 = 26.00 billion, and adding the two gives 119.54 billion—which matches the official “about $119.5 billion” exactly. The so-called “largest in history,” calculated by the media at 81.3 billion, has already surpassed SpaceX’s 75 billion in June this year; calculated officially at 93.5 billion, it raises that record even further.
Combined IPO and private placement (official figures, 2026-11-19)
Its first-day performance was also quite dramatic: it opened at $111.50, with its fully diluted market capitalization briefly surging to about $2 trillion, before closing back at $97.17, up 17.07% from the IPO price. Its fully diluted valuation based on the IPO price was about $1.5 trillion, accounting for roughly 2% of the total market capitalization of the U.S. public equity market. Subscription demand reached nearly seven times the offering size—this enthusiasm and worries about an "AI bubble" are almost two sides of the same coin.
II. 33.2 Billion in Revenue and 104 Billion in Losses: What Do the 90 Pages of Risk Factors in the Prospectus Say?
The most tangible benefit of going public is that the books are open. The prospectus discloses: for the nine months ended September 30, revenue was $33.24 billion, up about 1,200% year over year; as of October 31, the annualized revenue run rate had risen to about $81 billion, a net increase of about $10 billion in one month. This is astonishing growth.
+1200%
Year-over-year revenue growth for the first nine months (revenue: US$33.24 billion)
$104 billion
Net loss for the first three quarters (US$25.1 billion in the same period of 2025)
The losses are equally staggering. A clarification of definition is necessary here, otherwise it is very easy to misread: this US$104 billion net loss mainly comes from the changes in fair value of convertible notes and derivative liabilities, which is an accounting measurement result and does not mean that US$104 billion in operating cash was burned. But even so, it means that after going public, this company's financial statements will remain continuously exposed under the magnifying glass of the public market, and any slowdown in growth will be priced in immediately.
What deserves the CIO's attention even more is that over 90-page "Risk Factors" section. It is essentially a ready-made supplier risk checklist—except that ordinarily we simply cannot get our hands on it.
There is another layer of industrial relationships worth unpacking: the seven participants in this private placement—AMD, Broadcom, Google, Micron, Microsoft, NVIDIA, and TSMC—are both its shareholders and mostly its compute suppliers, while Google and Amazon are also its investors and cloud partners. This deep binding of "shareholder as supplier as customer" will significantly increase the supply chain concentration of the enterprise AI stack.
Market sentiment has already fractured. Former SEC Chair Gensler warned that the AI investment boom could burst before 2028, Bridgewater's Dalio said "we are close to the bursting point," and UBS's research unit pointed out that volatility in the tech sector has risen to its highest since the dot-com bubble burst. These are opinions, not conclusions, but when a CIO makes a three-year technology plan, they cannot assume that the price and supply of AI services will always improve linearly.
III. How CIOs Should Read This Prospectus: Three Procurement Recommendations
Read an IPO as financial news, and you move on once you've finished it; read it as a supplier's manual, and you can glean three things you should do.
First, create a "counterparty health check" for core AI suppliers. In the past, when choosing ERP, we looked at counterparty continuity risk; now, when choosing foundation model suppliers, we must likewise examine: cash reserves, revenue structure, customer concentration, and post-IPO profitability pressure. Write "what is our contingency plan if this company raises prices by 50% or restricts quotas in the next 12 months" into the assessment form, rather than waiting until the renewal notice arrives to think about it.
Second, turn "model portability" from a bonus point into a hard acceptance metric. The deeper the cross-shareholding in the industry chain, the higher the lock-in risk. At acceptance, clarify three things: whether prompts and knowledge bases can be exported, whether evaluation sets can be rerun on different models, and whether the work hours and costs of switching models have been actually measured. If these three questions cannot be answered, it is equivalent to depositing the enterprise's organizational memory in a supplier's account.
Third, infer procurement cadence from public financial reports. A model vendor that has just gone public and is under profit pressure will most likely next shift from "subsidies for growth" to "usage-based charging to improve gross margin." Right now, the AI budget should be split away from the single "pay-per-call" model, with a portion allocated to self-hostable open-weight models as a hedge—this is the same logic as our previously discussed point many times that "step-based billing contracts must specify routing rights."
In the final analysis, the significance of this IPO for enterprise digitalization does not lie in how many billionaires it created, but in turning a long-vague supplier into a company that must account clearly on a quarterly basis. For the first time, CIOs have publicly available, verifiable counterparty information with an auditable basis—whether they use it well depends on whether we know how to read these books.
All data in this article are annotated with sources and dates: Anthropic's official IPO closing announcement (2026-11-19), reports by Xinhua News Agency and Securities Times (2026-11-19), reports by People's Daily English-language app and China News Service (2026-11-20), and Sina Finance's analysis of the AI bubble (2026-11-20). The amount raised is presented separately under two calibers—official and media—with the calculation process shown; the net loss caliber has been explained; judgments by market participants are opinions, not facts. Data as of 2026-11-22.
Zhongdi Information · Specializing in Odoo and digital solutions for enterprise management
