Anthropic’s equity could be worth as much as $4.04 trillion, or nothing at all, according to valuation expert Aswath Damodaran, who also calls the company’s share structure “so deviously self-serving that it would put Kim Jong Un to shame.”
In his worst case, revenue limps to $200 billion by 2036, margins settle at 10%, compute stays scarce and costly, and regulation hits hard. The equity is then worth minus $373 billion, and the company does not survive as a going concern. In his best case, revenue surges to $1.5 trillion, margins reach 40%, compute becomes cheap and regulation barely bites. The equity is then worth $4.04 trillion.
That gap of over $4 trillion between the two outcomes captures the problem facing anyone trying to put a price on the Claude maker ahead of its listing. Damodaran’s own base case lands at about $1.475 trillion, at the lower end of the $1.5 trillion to $2 trillion range being discussed for the IPO. He calls it an upbeat story, which makes the fact that it still falls short of the IPO band all the more telling.
What the prospectus shows
Anthropic filed its 416-page prospectus on October 26. Damodaran, who has long argued that IPO filings have grown longer while becoming less informative, says this one has the usual mix of useful data, legalese and distractions. He sorts the disclosures into what optimists and pessimists will each seize on.
On revenue, the growth has been exponential. Revenue in the first half of 2026 was about 14 times the year-ago level. But growth slowed in the third quarter as OpenAI gained market share, compute capacity became a constraint, and some enterprises moved to cheaper open-source models. Enterprises spending over $1 million a year on Claude rose to 1,900 from under 100 a year ago, yet the top 5 customers still account for 25% of revenue.
On profitability, the headline loss for the first half was $80 billion, but $72 billion of that was a non-cash charge from expensing the conversion option on a convertible bond. Adjusted for that, the operating loss was over $8 billion.
Also read: Jio Platforms joins Meta, Alibaba, SpaceX in $100 billion founder-led IPO valuation club
The trend is moving in Anthropic’s favour. The company claims an operating profit in the latest quarter, every expense line is shrinking as a share of revenue, and compute costs, its biggest expense, have fallen from over 250% of revenue to 48% quarterly.
The catch, Damodaran notes, is that the company still needs strong revenue growth to stay ahead of the costs coming its way.
On reinvestment, the financial statements barely show Anthropic’s spending on compute. Neither capital expenditure nor debt reflects it in any meaningful way. The company is open, though, about commitments that run to hundreds of billions of dollars in future years, and over $600 billion in total. These commitments have helped Anthropic catch up with OpenAI on compute, which until recently held the lead.
They have also turned its highest cost from a variable expense into a fixed one. That flatters gross margins today but raises operating leverage, and chips age fast and will need replacing. Data centres, facing community opposition, may also get costlier.
On risk, Anthropic’s pitch is that once enterprises embed Claude in their operations, they won’t walk away. Damodaran says that claim is yet to be tested, and AI spending may well turn out to be cyclical, with companies pulling back in bad times.
The compute commitments expose the company to default risk, reflected in a BB+ bond rating. Its defence is a cash balance of $114 billion, likely to rise by another $100 billion from the offering, and easy access to capital as a listed company worth over a trillion dollars.
He also flags the circularity critics have pointed to. Anthropic is closely tied, in both operations and financing, to other big players in the AI business, which creates room for double counting and opacity.
The past versus the future
Still, Damodaran’s central message is that investors are reading too much into the filing. The prospectus is rooted in the past, he argues, while the value of a business lies in what investors expect it to deliver. For a company as young as Anthropic, that gap is large. He likens the exercise to studying a precocious kindergartner’s report card to decide how much to put into the child’s college fund.
It is a useful corrective at a time when the S-1 is being combed line by line for proof that AI either works or doesn’t. Damodaran expects both bulls and bears to spin the same numbers and walk away with their views reinforced.
He also warns investors against falling for conspiracy theories, whether of hyperscalers, chipmakers and AI labs colluding to create a mirage of success, or of accounting fraud. Self-interest and creative accounting exist, he says, but they are common to all companies, and chasing such theories shuts out other explanations.
To value Anthropic, he says, investors need to make their own calls on 4 questions: how big the market gets, where margins settle, how much reinvestment growth requires, and how risky the business is. He has shared his spreadsheet so readers can plug in their own assumptions and set up a shared Google Sheet where they can compare their valuations with others.
Damodaran’s base case, step by step
Market size: Damodaran sees AI’s most powerful models displacing well-paid white-collar workers across some industries, creating an end market of $5 trillion or more. Anthropic, in his reading, is aiming for the premium end of that market and looks like a leader there.
He also takes the ambition seriously, spelt out in the prospectus, to compete in some of the businesses its agents disrupt, with pharmaceuticals a possible play. On that basis, he believes revenue approaching $1 trillion within a decade is plausible. For perspective, no company in the world earns that much today, with Amazon and Walmart at around $700 billion to $750 billion.
That is the boldest assumption in his model, and the one investors will argue over most.
Margins: He expects the AI market to split in two. A mass market of cheaper, often open-source models will compete on price and earn margins similar to business services, at around 10%. A premium market of closed frontier models, powered by client data, will earn margins closer to top-tier enterprise software, at around 40%.
With almost 80% of Anthropic’s revenue coming from enterprises paying by usage, he assumes a 75:25 split in favour of premium. That gives a pre-tax margin of 32.5%, which would place it in the ninth decile of US companies.
Reinvestment: He expects the compute crunch to persist in the near term and ease over time. He assumes Anthropic will need to invest $1 for every $3 of additional revenue, in line with tech companies today, which still means very large reinvestment each year given the scale of growth.
Risk: He puts the cost of capital at 9.74%, higher than the 9.16% median for US firms, to reflect the burden of the compute commitments. He assigns a 10% chance that the company fails, with half its value lost if that happens. On regulation, he notes that both Anthropic and OpenAI have invited government oversight, which cynics read as a way to keep out competition, especially from China. Either way, he expects regulation to raise costs rather than shut the business down, much like social media. He trims his margin estimate to 27.5% to account for it.
A governance warning
Damodaran saves his sharpest criticism for governance, and this is where his piece turns from cautious to scathing.
“I think Anthropic has managed to put those companies to shame with its combination of a benefit corporation structure (which gives the company the legal license to do pretty much whatever it wants with your money) and a voting share design that is so deviously self-serving that it would put Kim Jong Un to shame.”NYU PROFESSOR
He traces the problem to Google’s IPO, which broke precedent in US markets by issuing shares with different voting rights while institutional investors stayed silent. Tech companies have since used the same structure to protect founders. Anthropic, he says, takes it much further. It has 5 classes of shares. A benefit trust picks the board, and the founders control 50.1% of the vote no matter how many new shares are issued in future.
It is also a public benefit corporation, and its prospectus states plainly that it may take decisions that hurt its business performance or stock price. Damodaran notes that the company’s leadership has actively cultivated concern about AI as a destructive force, which makes such protection look reassuring. But he argues the structure does not shield society from AI’s harms. It puts power in the hands of the founders and the trustees they pick, and asks investors to hope they use it well.
Investors who have watched companies use ESG as cover for self-serving behaviour, he says, will find this the same structure on steroids. Damodaran holds dual-class stocks such as Alphabet and Meta and won’t rule out such companies. But he says Anthropic has outdone them all with a voting design that hands near-total control to its founders.
He reads the structure as a calculated dare to institutional investors: object by staying out of the IPO, or stay passive. Anthropic, he says, has bet they will stay passive.
On lock-ups, insiders other than founders can sell up to 10% of their shares on day 1, with the rest unlocked in tranches over 180 days. Founders and key figures can trade only after 365 days. Damodaran thinks too much is made of unlock dates. With large institutions and plenty of liquidity, he expects the impact to be muted, and adds that the smart money selling at unlocks is often less smart than people assume.
Damodaran has already said the IPO will be priced rather than valued, with buyers on listing day reacting to mood and momentum rather than fundamentals. Markets owe nothing to investors who care about fundamentals, he says. So he will wait for the pricing, and if it comes in well above his value, he will sit out the frenzy.
He won’t short the stock either. He has faith in his valuation, but no reason to believe the market will agree with him within his time horizon.
In a post on X, he said the story will have twists and turns, and that he is convinced it will become a buy at some point. He points to Facebook. He found it overvalued at its 2012 IPO at $36 and bought it 3 months later at $18. With Anthropic, he expects a roller coaster, and says there will almost certainly come a day when the stock is undervalued. Given its governance, though, he will want a much larger discount before he buys.
Anthropic plans to list on the Nasdaq under the ticker ANT. The filing does not yet carry a price range or share count, and the listing is expected in Thanksgiving week. Rival OpenAI’s IPO is on hold, with Sam Altman ruling out a listing this year.