ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
NewsMaterial4 min readDisconfirmAI Editorial

NYU Stern Finance Professor Aswath Damodaran: ARR for AI Companies Is Not Valuation, But a Container for Hidden Assumptions

Damodaran explains why young companies are forced to use pricing instead of valuation, and what growth, profitability, and risk assumptions are embedded in non-standard metrics like ARR, user counts, and compute power.

2026-10-23 06:02 GMT+8

Aswath Damodaran, Professor of Finance at the Stern School of Business at New York University, published a long-form article in October 2026 with the core argument that for young companies, pricing has supplanted valuation, and metrics like ARR (Annualized Revenue Run Rate) are not objective measures but rather containers for assumptions about growth, profit margins, and risk hidden within a single number.

Damodaran first distinguishes between two ways of assigning numbers to assets. Valuation is based on an asset's ability to generate and grow cash flows and the risks associated with those cash flows; pricing looks at the market transaction prices of similar assets, which is essentially a matter of supply and demand. If you believe markets are efficient, prices will converge on value—but if you truly believed that, you would not spend energy analyzing individual stocks and would simply buy index funds. Investors and traders in reality take a different path.

Why are young companies forced to slide from valuation to pricing? The mechanism Damodaran provides is direct: valuation requires estimating future cash flows, and estimating cash flows requires three sources of information—financial statements, market size, and industry statistics. Young companies lack precisely these three things: undefined business models lead to noisy financial data, new markets are difficult to define in terms of size, and there is a lack of comparable companies' profit margins and risk benchmarks in the industry. He summarizes this logic with one sentence: Desperation drives you up the income statement until you find a positive number. For Anthropic and OpenAI, that positive number is revenue.

Sign in / Sign up
ReadingResearchRadarInvestment framework

But revenue itself is no longer novel. These two companies are growing so fast that traditional trailing twelve-month (TTM) revenue figures are outdated by the time they are released, leading the industry to shift to Annualized Revenue Run Rates (ARR). Damodaran points out the fundamental problem with ARR: it lacks a standard definition. Does revenue refer to gross or net revenue? Is annualization based on the last day, last week, or last month's revenue? He cites a specific example: OpenAI recently shifted from a gross revenue basis to a net revenue basis, causing its ARR to drop from $70 billion to $50 billion, while Anthropic continues to report gross revenue and treats cloud service provider payments as expenses. Both companies have chosen monthly annualization, but no rules bind them from changing this practice.

Pricing involves more than one metric. Damodaran lists three choices: using current or recent revenue, using expected future revenue, or using drivers related to future revenue. The third option is the logic Mary Meeker used to price dot-com companies in the 1990s using 'eyeballs'—today's equivalent is user count or compute power (measured in gigawatts or compute commitment amounts). He warns that once investors shift to forward revenue or even forward profits, multiples may appear lower and easier to defend, but the implicit assumptions expand accordingly: buying today at 25 times expected 2036 profits is betting that an unprofitable company can achieve mature tech company profit margin levels within ten years.

The second step in pricing is finding comparable companies. Damodaran identifies this as the true dilemma facing AI companies: if you define peers by the large language model track, you can hardly find any comparable publicly listed companies. Gemini is only a small part of Alphabet, xAI is embedded within SpaceX, and Muse is obscured by Meta's advertising business. Anthropic and OpenAI have become each other's only references, and neither is publicly traded. If Anthropic's market capitalization approaches $1.5 trillion, investors either compare it to OpenAI or group it with the Mag Seven (Apple, Microsoft, etc., all trillion-dollar companies)—but the latter have completely different business models and lifecycle stages.

Damodaran's core warning comes at the end: Pricing does not dispense with assumptions. He dissects the valuation chain behind revenue multiples one by one—for high revenue multiples to be reasonable, a company must simultaneously satisfy three conditions: rapid revenue growth with accelerating speed; profit margins jumping from losses to over 30% (relying on favorable unit economics and economies of scale); and risk converging rapidly to industry averages. He writes that this story might be a stretch, but it is plausible, and he is willing to give the storytellers a chance rather than dismissing it outright.

He also criticizes two extremes: bulls who look only at revenue growth or gross profit margins, and bears who look only at losses and negative cash flows, both of whom are only touching part of the elephant. Damodaran's own stance is that he prefers doing valuation over pricing, but he must wait until the prospectus is made public before he can begin.

The practical implication for readers is: when you encounter assertions like 'Company X's AI ARR multiple is lower than peers, so it is cheaper,' you should ask three questions: Does this ARR use gross or net revenue? What period is used for annualization? What growth and profit margin assumptions are implied? Damodaran's framework does not provide buy/sell answers, but it makes the assumptions behind every pricing number checkable.

Sources

  • aswathdamodaran.blogspot.com↗
Continue researching →