Own analysis · Markets
Anthropic's fair value: $93 a share
ANT closed its first day at $97.17 and a diluted market value of about $1.8tn. The prospectus accounts and a three-scenario model put fair value near $93: the market is neither expensive nor cheap. It is demanding a perfect 2027.
Anthropic debuted on Nasdaq on Thursday 19 under the ticker ANT. It sold 980 million shares at $83, the largest IPO on record: $81.34bn, ahead of SpaceX's $75bn in June. The first trade crossed at $111.50, 34.3% above the offer price; the stock reached $114.99 and closed at $97.17, up 17.1%. The fully diluted market value is about $1.8tn, according to CNBC.
The question is no longer whether the IPO was a success. It was. The question is whether $97.17 is the right price. An analyst who only buys below his estimate needs three things: the actual prospectus accounts, a comparable multiple and a range of scenarios. With those three pieces, my fair value for ANT is $93 a share, with a range of $48 to $145.
What the prospectus says (and what it does not)
Revenue in the first nine months of 2026 was $33.24bn, up 1,206% on the same period of 2025. In the third quarter it booked $16.96bn, up 47% on the second quarter, when it had grown 144%. The annual series shows the scale: $41m in 2023, $386m in 2024 and $4.59bn in 2025. Growth is still extraordinary, but it is decelerating on a sequential basis.
Profit is another story. In the third quarter the company swung back to an operating loss of $567m after making $974m in the second. Over nine months the operating loss was $1.26bn and the adjusted operating margin fell to 15% from 19% in the previous quarter. Cost of revenue was $4.81bn, about 14 cents for every dollar of revenue; compute and infrastructure spending rose to $9.86bn, or 58% of revenue, from 48% before.
The headline on Wednesday was a net loss of $104.0bn over nine months. Almost all of it is accounting: $94.3bn comes from the revaluation of convertible notes, not cash leaving the business. Operations generated $6.6bn of cash. That is the difference between a real impairment and a book entry driven by the company's rising valuation.
The customer base explains the rest. At 30 September it had 11,900 customers spending more than $100,000 a year, contributing $26.1bn, or 74% of trailing-twelve-month revenue. About 40 million small customers provide the other 26%. More than 1,900 customers spend over $1m. Net revenue retention is 900%, according to the company, and the top five customers account for 25% of revenue. Cloud marketplaces carry 44%.
The liability that really matters is not in net income: it is the $900.9bn of purchase and lease commitments, almost all for compute and infrastructure. $70.7bn falls due in the 12 months after 30 September and $451.6bn before 2030. Cash and short-term investments were about $132bn. These are take-or-pay contracts: if demand cools, the spending does not disappear.
And there is a third piece: price. Contractual cuts and discounts took about $600m off revenue between the second and third quarters, against $70m a year earlier. AI is deflationary; to grow, volume has to offset falling prices.
The number that matters is not net income. It is whether gross margin and cash hold up while the next wave of compute comes online.
The map gathers 20 headquarters: nine in the Americas, four in Europe and seven in Asia. San Francisco is home to Anthropic, OpenAI and Perplexity; London to Google DeepMind and Stability AI; Paris to Mistral; Hangzhou to DeepSeek and Alibaba; Beijing to Baidu and Moonshot. The concentration is not accidental: talent, capital and compute cluster in a few hubs. For ANT, competition for researchers, customers and power is local and intense, not global and diffuse.
| Lab | Headquarters | Founded | Model |
|---|---|---|---|
| Anthropic | San Francisco (US) | 2021 | Claude |
| OpenAI | San Francisco (US) | 2015 | GPT |
| Google DeepMind | London (UK) | 2010 | Gemini |
| Meta AI | Menlo Park (US) | 2013 | Llama |
| Microsoft AI | Redmond (US) | 2023 | Copilot |
| xAI | Palo Alto (US) | 2023 | Grok |
| Mistral AI | Paris (France) | 2023 | Mistral |
| DeepSeek | Hangzhou (China) | 2023 | DeepSeek |
| Alibaba | Hangzhou (China) | 1999 | Qwen |
| Baidu | Beijing (China) | 2000 | ERNIE |
How to price a stock that does not make money
You cannot use the P/E ratio: there is no positive profit. The tool is the sales multiple. For 2026 I estimate about $53bn of revenue ($33.24bn in nine months plus a fourth quarter around $20bn, consistent with the $81bn annualised October run-rate). For 2027 I model three scenarios: $110bn, $150bn and $190bn. The first requires doubling revenue; the second, almost tripling it; the third, multiplying it by 3.6.
The base case applies 11 times sales, the peer median cited by William Blair. The bull case uses 14 times, the top of the high-growth peer range. The bear case uses 8 times, which fits a sharp slowdown and normalising prices. These are not arbitrary multiples: they are the range in which high-growth software companies trade and the range analysts themselves use for ANT.
| Scenario | 2027 revenue | EV/sales | Value per share | Probability |
|---|---|---|---|---|
| Bear | $110bn | 8x | $48 | 25% |
| Base | $150bn | 11x | $90 | 50% |
| Bull | $190bn | 14x | $145 | 25% |
| Weighted | — | — | $93 | 100% |
I use 18.3bn fully diluted shares, a slightly conservative figure against the 18.5bn implied by CNBC's market value. The prospectus lists 13.22bn Class A and Class N shares; options and restricted stock units sit on top. On 13.22bn, value per share would be 38% higher, but that would be misleading: the dilution is real.
The weighted result is $93. The probability is not a forecast: it makes explicit the weight I give each scenario. If the reader thinks the bull case deserves more weight, fair value rises; if the bear case is more likely, it falls. The virtue of the model is not the final number but the fact that it forces the assumptions into the open.
What the price already discounts
At $97.17 and 18.3bn shares, the market pays $1.78tn. At 11 times sales, that requires $162bn of 2027 revenue; at 8 times, $222bn; at 14 times, $127bn. The close sits above our base case and below the bull case.
The first day moved between the top of the range and the base case. The $111.50 open valued the company at $2.04tn and required $185bn of sales at 11 times; the $92.72 low left it at $1.70tn and required $154bn, almost exactly the base case. The market opened at the top and closed in the base case. That is not a directional signal; it is a sign that the price is still looking for equilibrium.
Analyst price targets are not the same as fair value. Morningstar published a $92 base value, with $188 in the bull case and $25 in the bear case. New Street has $105, KGI $100, Needham $150 and Melius $155. The consensus is around $120. These are 12-month targets: they embed the growth analysts expect, not the price they would pay today. That is why my $93 sits below consensus and in line with Morningstar.
| Firm | Target | Note |
|---|---|---|
| Morningstar | $92 | Base value; bull $188, bear $25 |
| KGI Securities | $100 | Outperform |
| Freedom Finance | $101.90 | Buy |
| New Street | $105 | Buy; “generational IPO” |
| Citizens | $125 | Market Outperform |
| Wolfe | $135 | Outperform |
| Needham | $150 | Buy |
| Melius | $155 | Buy |
The four risks that move the range
One: compute. The $900.9bn of commitments is more than 11 times the $81bn annualised revenue. $70.7bn falls due within 12 months. If growth stalls, those payments eat the cash and turn a 15% margin into an operating loss.
Two: concentration. The top five customers account for 25% of revenue and 44% arrives through cloud marketplaces. The 900% retention rate is a company-defined, unaudited metric; if it normalises, growth slows. The prospectus warns that many customers have no long-term contracts and can cut spending.
Three: price. Cuts took $600m off revenue between the second and third quarters. Competition from OpenAI, Google and open-weight models pushes prices down. If price falls faster than volume, margin does not hold.
Four: the stock. The IPO sold 980 million shares and about 315 million employee shares were eligible to be sold from day one. The general lock-up expires after 180 days, around May 2027, and the founders' after a year. Nasdaq-100 inclusion on 21 December may add passive demand; the lock-up expiry may add supply. Today's price has digested neither.
What would change my mind
Three numbers. First: fourth-quarter revenue above $20bn and a 2027 guide above $150bn. Second: an adjusted operating margin above 20% while compute commitments grow more slowly than revenue. Third: the top five customers below 20% and marketplaces below 40%. If all three land, a 14 times multiple is defensible and value rises to $145. If the first fails, the $48 bear case stops being a theoretical exercise.
I would also watch the price of compute. Anthropic bought $5bn of short-term compute in the third quarter because demand exceeded capacity; that compute costs far more than its multi-year contracts. If scarcity persists, the 15% margin is a ceiling, not a floor.
And there is one figure that is not in the prospectus: opportunity cost. At $97, ANT competes with companies that already make money and trade on similar multiples. The investor is not just buying growth; they are buying growth with a $900bn contractual obligation behind it. That is why fair value cannot be the price of the first trade.
Conclusion. Anthropic is neither an obvious bubble nor a bargain. At $97.17, the market pays for the base case and leaves no margin of safety. My fair value is $93, with a range of $48 to $145. Anyone buying at this price is betting that 2027 will beat my base case; anyone selling is betting it will not. The IPO was historic. The price, for now, is reasonable: no more, no less.
This article is for information only and is not investment advice. The valuations are scenarios with explicit assumptions, not forecasts. Before investing, read the prospectus and consult a financial adviser.
Sources
- Anthropic (18 November 2026): “Anthropic prices initial public offering”. $83 price, 980 million shares and the underwriters' option.
- CNBC (19 November 2026): “Anthropic IPO live updates: ANT stock price and Nasdaq debut”. Open, close and fully diluted market value.
- Reuters (18 November 2026): “Anthropic prices blockbuster IPO at $83 a share”. Offering size and the comparison with SpaceX.
- Anthropic, PBC: final prospectus (Form 424B4, 18 November 2026). Revenue, margins, customers, compute commitments and share counts.
- AI Business (17 November 2026): “Anthropic's IPO prospectus, read line by line”. Breakdown of the net loss and commitments.
- Invezz (18 November 2026): “Anthropic stock price targets”. Morningstar, New Street, Freedom and KGI; bull and bear cases.
- Yahoo Finance (18 November 2026): “Investors overwhelmingly bullish on Anthropic”. 7-8 times 2027 sales and an 11 times peer median, according to William Blair.
- MarketWatch: ANT analyst estimates. Consensus price target.
- Interactive map: AI lab headquarters. 20 dots with company, city, founding year and model; data from company pages and Wikipedia (November 2026).
- Wikipedia: Anthropic, OpenAI, Google DeepMind, Meta AI, Mistral AI, DeepSeek and the other labs on the map. Headquarters and founding years checked on 20 November 2026.