The Prospectus is here: My (but not the) Anthropic IPO Valuation!

Summary

Aswath Damodaran reviews Anthropic’s IPO prospectus and develops his own valuation. He emphasizes that the prospectus is a snapshot of a very young business, not a substitute for forecasting its future. The same disclosures, he says, can support optimistic or pessimistic interpretations, so investors should develop and test their own investment story.

What the prospectus suggests

• Growth: Anthropic reported year-over-year revenue growth of roughly 1,300–1,400%. Damodaran says revenue had risen from almost zero to more than $16 billion in the latest quarter and could reach at least $100 billion by year-end. The precise revenue measure is not clarified in the subtitles.
• Customers: The company reported 11,900 customers, including 1,900 spending more than $1 million. Its five largest customers still represented 25% of revenue, creating customer-concentration risk.
• Growth risks: Preliminary third-quarter figures reportedly discussed with investors suggested growth had stalled. Damodaran cites compute constraints, the natural slowdown that can occur as a business scales, and intensifying competition—particularly OpenAI’s pricing—as possible causes.
• Revenue reporting: Anthropic reports gross revenue, while OpenAI reportedly shifted to net revenue. Damodaran says 15–20% of some partnership revenue is paid back to technology partners and recorded as marketing expense. He describes this as a reporting choice, not fraud, but says it reduces comparability and could make industry-wide revenue less transparent.
• Profitability: Computing costs had fallen from 250% of revenue six quarters earlier to under 50%. Damodaran estimates gross margin at about 80%, partly because affiliate payments are treated as marketing expense and computing costs have shifted from variable to fixed costs. Anthropic reported an operating profit in one quarter, but he cautions that sustained profitability depends on continued growth. Stock-based compensation could also be a burden.
• Reinvestment and obligations: Reported capital expenditure appears low, but the company has substantial contractual compute commitments of over $450 billion and total liabilities of over $600 billion, according to the presentation. Broadcom is identified as the largest single beneficiary of those commitments. Damodaran says the commitments secure capacity but create large fixed costs that could become dangerous if growth weakens.
• Risks and potential buffers: Repeat customer use could make revenue more durable. Damodaran cites a current cash balance of about $114 billion, potentially rising to roughly $200 billion after the IPO; his valuation calculation uses about $220 billion, including expected IPO proceeds. Regulation might protect established companies from some competitors, but could also increase costs and reduce margins.

Damodaran’s valuation framework

Damodaran frames valuation around four inputs: revenue, profitability, reinvestment, and risk. He invites investors to substitute their own assumptions.

1. Estimate the market and Anthropic’s eventual revenue. He identifies four questions that affect the addressable market: whether AI is a tool or replaces workers; whether it serves only high-wage workers or all workers; whether adoption is limited to some industries or broad across businesses; and whether the market is regional or global. He estimates the AI market at $5–10 trillion and sees Anthropic as a possible premium-market leader.
2. Choose a 2036 revenue target. His base case is $1 trillion in 10 years. Alternatives include $200 billion, $500 billion, $800 billion, $1 trillion, or $1.5 trillion, with the option to enter $2 trillion or $2.5 trillion. He says a tool-only outcome could imply much lower revenue.
3. Estimate steady-state operating margins. He assumes 10% for mass-market AI services and 40% for premium products. With a mix of 75% premium and 25% mass-market, that yields an unadjusted margin of 32.5%.
4. Allow for regulatory costs. His base case reduces the margin by 5 percentage points, to 27.5%. He also considers a scenario with no regulatory impact and says the effect could be larger.
5. Model reinvestment in computing. His base assumption is $3 of additional revenue for each $1 of invested capital, or a sales-to-capital ratio of 3. Scarcer or more expensive compute would require more reinvestment; cheaper and more available compute would require less.
6. Set the discount rate and account for failure risk. He uses a starting cost of capital of 9.74%, compared with a 9.16% median and 9.88% third quartile for U.S. companies in September 2026. He also cites 7.79% as the technology-sector median. His long-term assumption moves toward a 9.13% U.S. company median. He assigns a 10% bankruptcy probability, with assets recovered at half their value in that scenario. Anthropic’s BB+ bond rating is below investment grade.
7. Adjust for financing and contractual commitments. He treats the present value of compute commitments as debt, estimating it at $473 billion, even though it does not appear as debt on the balance sheet. He also accounts for cash, expected IPO proceeds, shares offered, restricted shares, and separately valued options.

Valuation outcome and investment implications

• In his base case, Damodaran estimates operating assets at about $1.7 trillion. After including cash and IPO proceeds, deducting the $473 billion present value of contractual obligations, and adjusting for failure risk, he estimates equity value at about $1.48 trillion, also described as roughly $1.47 trillion.
• Based on his assumed share count, he estimates about $447 per share. He calls the share count provisional and says he would revise the per-share estimate when more specific share data are available.
• His worst-case scenario produces equity value of negative $373 billion, effectively implying failure or bankruptcy. His best-case scenario—with revenue of $1.5 trillion, a 40% margin, cheaper compute, and a lower cost of capital—produces equity value of about $4 trillion.
• He says the anticipated IPO pricing discussion is around $1.5–2 trillion. At prices significantly above his estimate, he would not buy; he would require a substantial discount, partly because of governance concerns. He allows that he might buy if the price fell sufficiently below his value estimate.
• He distinguishes investing from short-term trading. An IPO trade, he says, would depend on market sentiment and momentum. He does not consider himself a trader and cautions that the stock could fluctuate as news emerges about AI adoption, regulation, and the business.
• He advises investors to test their own story by asking whether it is possible, plausible, and likely.

Governance and share structure

Damodaran highlights what he sees as unusually weak shareholder control:

• Anthropic is described as having five share classes:
  • Public Class A shares have one vote each.
  • Class N shares held by Alphabet and Amazon have economic rights but no votes.
  • Class T shares are held by a trust that can appoint a majority of directors.
  • A Class F share held by the seven co-founders carries 50.1% of voting rights.
  • The subtitles also mention employee shares with voting rights limited to certain matters.
• The founders select the members of the public-benefit trust. Damodaran argues that the public-benefit-corporation structure can give management legal room to make decisions that disadvantage shareholders and make it difficult for shareholders to force change.
• He says this governance structure is a reason to demand a larger valuation discount, not an automatic reason never to buy the stock.
• He also discusses staged insider share releases. Some insiders could sell up to 10% of their shares on the first day; other shares would unlock in tranches through 180 days, while founders and certain key people could trade after 365 days. He argues that anticipated unlocks are usually reflected in market prices in advance, so he does not expect their effect to be as large as often claimed.

Other cautions and context

• The prospectus is 416 pages. Damodaran contrasts this with Apple’s reported 80-page 1980 IPO prospectus and Microsoft’s 66-page prospectus, criticizing modern prospectuses as lengthy and partly shaped by disclosure and legal-protection concerns.
• He cautions against conspiracy theories about AI companies, hyperscalers, Nvidia, and alleged industry “double counting.” He interprets industry interdependence as potentially reflecting self-interest rather than a coordinated conspiracy.
• He says a cited $72.6 billion convertible-debt-related cost is attributable to the conversion option’s accounting mechanics, not evidence of accounting fraud by itself.

Companies, sectors, and instruments mentioned

• Companies and organizations: Anthropic, OpenAI, Alphabet/Google, Amazon, Broadcom, Nvidia, Apple, Microsoft, Airbnb, Uber, Facebook, Tesla, SpaceX, and The New York Times.
• Sectors and business areas: AI/LLMs and AI agents; enterprise software; mass-market and premium AI services; technology and hyperscalers; pharmaceuticals; social media; chips, data centers, and computing infrastructure.
• Financial instruments and assets: IPO shares, equity, restricted shares, employee stock compensation, options, convertible debt, bonds, cash, and contractual compute commitments.
• Tickers: No ticker symbols are mentioned.

Disclosures

Damodaran says the valuation reflects his own story and assumptions, warns that he could be wrong, and tells investors to form their own view. The subtitles contain no explicit “not financial advice” disclaimer.

Presenter and source: Aswath Damodaran discusses the Anthropic prospectus and his prior analyses of AI and Anthropic’s prospective valuation.