Anthropic IPO Valuation: Damodaran's $1.475 Trillion Model and What Matt Levine, Ben Thompson, and Other Top Analysts Say

Aswath Damodaran, the NYU Stern professor who teaches valuation, values Anthropic's equity at $1.475 trillion. That is $447 per share on his share count, and just below the $1.5 to $2 trillion range bankers have floated for the IPO. He published the model, every input, and a spreadsheet in his Musings on Markets post three days after the S-1 went public.

This post rebuilds his valuation step by step and turns his 13 charts and tables into data you can read on a phone. Then it sets his view next to the other analysts worth reading on this filing. They are Matt Levine at Bloomberg, Ben Thompson at Stratechery, Tomasz Tunguz, Azeem Azhar, MBI Deep Dives, and the bears. For the raw numbers in the filing, see our Anthropic S-1 guide.

Who Said What, in One Table

Damodaran Reads the Prospectus: Glass Half Full, Glass Half Empty

Damodaran sorts the 416-page filing into four inputs: revenue growth, profitability, reinvestment, and risk. For each one he gives the bull reading and the bear reading side by side.

The margin trend is the strongest bull fact in his tables. The chart uses his quarterly numbers.

He calls the prospectus "a snapshot of a young company that is evolving" in his summary. To him, reading it is like reading the report card of "a kindergartner" and then sizing its college fund. He also warns readers away from two traps: theories of an AI company cabal, and claims of accounting fraud.

The Compute Commitments Are Debt

Damodaran treats the compute contracts as borrowing. The S-1 balance sheet shows little debt, because the contracts are future purchase commitments, not loans. He discounts the contracted payments at 6.28%, the borrowing rate for a BB+ company.

In his valuation spreadsheet, total debt is $472.3 billion, which includes this $363 billion.

Other analysts count the total differently, because some deals are minimums and some are "up to" amounts paid on delivery.

Tunguz states the bear case in one line: the revenue is consumption-based with no lock-in, so "the payments arrive whether customers do."

Damodaran's Five Choices, and Yours

Damodaran does not ask you to accept his numbers. For each input he lists five or six choices and marks his own. Use the table to build your own valuation.

Revenue is the input that matters most. He sets a trillion dollars as the target. Today no company in the world has more than $1 trillion in revenue. Amazon and Walmart are at about $700 to $750 billion.

How the $1.475 Trillion Is Built

The model forecasts cash flow for 10 years, adds a terminal value, adjusts for failure risk, then moves from the business to the shares. Each step uses numbers from his spreadsheet, in US$ billions.

1. Discount the next 10 years of free cash flow at his cost of capital: $424.7B. Cash flow is negative in year 1 (-$9.3B) and zero in year 2, because reinvestment eats all profit.
2. Add the present value of the business after year 10. Terminal growth is 4.9%, the cost of capital falls to 9.13%, and return on capital is 15%: $1,398.4B.
3. Sum the two: $1,823.0B.
4. Apply failure risk. With a 10% chance of failure and 50% recovery, the value is 0.9 × 1,823.0 + 0.1 × 0.5 × 1,823.0 = $1,731.9B.
5. Subtract debt, including the compute commitments: -$472.3B.
6. Add cash, including about $100B of IPO proceeds, and non-operating assets: +$212.4B and +$3.9B.
7. Subtract the value of employee options: -$0.5B.
8. The result is equity value of $1,475.4B. Divided by 3,298.14 million shares, that is $447.34 per share.

About 77% of the operating value comes after year 10. That is normal for a young growth company, and it is why the revenue target matters more than the discount rate.

On his share count, the IPO range implies about $455 per share at $1.5 trillion and about $606 per share at $2 trillion. Those are our recomputations, and the final share count in the offering may differ.

The Best and Worst Cases

Damodaran runs the extremes of his own choices.

He calls both extremes "possible, but unlikely" outcomes. He asks readers to put their own story through his "3P test": is it possible, is it plausible, is it probable? He also set up a shared spreadsheet where readers enter their own valuations next to his.

Matt Levine: What to Promise, What to Threaten

Levine's Money Stuff column, "Anthropic Raises Concerns," starts from a framing that fits every AI IPO. When you write a prospectus for a frontier AI company, you choose what to promise in your "good vision of AI" and what to threaten in your "bad vision."

He compares three filings. SpaceX promised "the largest actionable total addressable market ('TAM') in human history," which Levine puts at $28.5 trillion. That is a lot of money, but a finite number. OpenAI's pitch for artificial general intelligence is more abstract. Anthropic's filing leans hard on the bad vision.

Fortune reports that Levine objects to the passive voice in Anthropic's risk language, such as "Concerns have been raised both within our organization and by other researchers, regulators, policymakers." Raised by whom, about what, and what did the company do? Much of the column is about governance, the same subject that angers Damodaran.

Ben Thompson: Profitable Core, Endless Dilution

Thompson's Stratechery piece is titled "Anthropic S-1, The Cost of Compute, Who Is Paying?" His summary has three parts. "The core business is very profitable." But "competition may eat all of the profits." And "investors are set up for endless dilution."

The dilution point links to Damodaran's reinvestment input. A company that must buy ever more compute to stay at the frontier pays for it with cash, debt, or new shares. Damodaran's model funds it from cash flow and IPO proceeds. If cash flow falls short, new shares fill the gap.

Tunguz, Azhar, and MBI: The Operator Lens

Three analysts who study software businesses read the same filing as an operating company.

Tomasz Tunguz (his blog) compares the metrics to the best software IPOs:

• Revenue grew 11.9 times in 2025 to $4.6 billion, more than four times OpenAI's 253% growth.
• Net revenue retention is about 900% as of September 30, 2026. Existing customers spend nine times what they did a year earlier.
• No customer is above 10% of revenue. The top five are about 25%.
• Nearly half of revenue flows through cloud marketplaces.

Azeem Azhar (Exponential View) concludes that "the lab has become a business, and it looks like it can become reliably profitable." He adds two warnings. More than 80% of revenue comes from models six months old or less, so the product must keep moving. And Anthropic might slip into a small loss in Q3 because of payments to its new public benefit foundation.

MBI Deep Dives (analysis) works backward from the price. Assume Anthropic trades at about 25 times operating income in 2030, like the hyperscalers. Then a buyer at about $1.5 trillion who wants a 10% annual return needs about $92 billion of GAAP operating income that year.

MBI's most useful number is how much compute serves paying customers. The rest trains the next model.

Revenue grew about 14 times while serving compute grew about 2.3 times. MBI frames the open question as three choices. If frontier training is a choice, Anthropic can cut it and profits jump. If it is a race, about $420 billion of research compute through 2030 is the cost of staying in the game. If it is a flywheel, better models earn more per unit of compute, and 25 times earnings is too low. MBI ends: "I don't yet have the conviction I would need to put my own capital into this IPO."

The Bears

The skeptics focus on whether customers outside tech get paid back.

• Torsten Sløk, chief economist at Apollo, quoted by Fortune: AI has "delivered zero financial payoff to corporate America outside of tech." He adds: "Revenue per worker is rising for the Mag 7 but is flat for the S&P 493."
• New Constructs, quoted by CNBC before the public filing, called it "the most ridiculous IPO of 2026" and said, "We don't think Anthropic has a viable business."

Sløk's point is the demand-side version of Damodaran's operating risk. If customers do not earn a return, AI spending becomes the first budget line cut in a downturn. The committed compute bills do not shrink with it.

Governance: Where the Critics Agree

Damodaran, Levine, and outside governance experts land in the same place here. Anthropic has five share classes.

Anthropic is also a public benefit corporation. The S-1 says the company may make decisions that "negatively impact the short-, medium-, or long-term performance of our business or the market price." Damodaran's verdict is harsh. In his words, the structure "does not protect society from AI's evils, but puts power in the hands of the few" at the top. He calls it ESG cover "on steroids" for investors. He says the voting design is "so deviously self-serving that it would put Kim Jong Un to shame" as well.

He still owns dual-class stocks such as Alphabet and Meta, so he does not rule the stock out on governance alone. He says it needs a larger discount. A Harvard Law School governance analysis points to a limit on the trust's power: a supermajority of stockholders can end the trust and remove the directors it picked. New York City Comptroller Mark Levine is no relation to Matt. He wrote to Anthropic on October 14 asking it to pause the IPO over safety disclosures and shareholder accountability.

On the lock-up, insiders other than founders may sell up to 10% of their shares on the first trading day. The rest unlocks in tranches over 180 days, and founders can sell after 365 days. Damodaran thinks unlock dates get too much attention, because "the smart money" is "not that smart."

What Damodaran Will Do on IPO Day

He will wait for the price. If Anthropic prices well above $1.48 trillion, he will "sit out the frenzy" and not buy. He will not sell short either. He has no reason to expect the market to agree with him within his time horizon.

He expects a buying chance later. He valued Facebook at its 2012 IPO and found it overvalued at $36. Three months later it traded at $18, and he bought. His view is that there will almost certainly be a day when Anthropic is undervalued, and it may come sooner than people think.

What This Means for Companies That Buy AI

All of these analysts end up on the same question from different sides: do customers get a return on what they spend? Anthropic's 900% net revenue retention says many customers do. Sløk's data says most companies outside tech do not yet. That gap is the whole valuation debate in one sentence.

For a buyer, the lesson is simple. AI vendors will keep cutting prices to fill committed compute, so model cost will fall. The return depends on what you build with it: the workflows, the data, and the evaluations that show the work is correct.

Get a Return on Your AI Spend

OpenNash puts forward-deployed engineers on your team and works with Fortune 2000 companies to turn model access into measured business results. A typical engagement follows these steps:

1. Rank your AI use cases by business value and data readiness.
2. Build the first workflow on your own data, with evaluations that measure accuracy and cost per task.
3. Route each task to the cheapest model that passes the evaluation, so falling model prices go to your margin.
4. Put the workflow into production with logs, access controls, and human sign-off where the risk needs it.
5. Report the return each quarter in the terms your CFO uses.

At the end, you own the workflows, the evaluation sets, and the documentation.

Book a call with an OpenNash forward-deployed engineer or email us at hello@opennash.com. In the first session, we review your top AI use cases and your data, and show which ones give the highest business value and the fastest ROI.