Aswath Damodaran says artificial intelligence product and service companies would need about $5 trillion in annual revenue within a decade to support an assumed aggregate valuation of $5 trillion.

That is about 20 times his estimate of the industry’s current annualized revenue, which he puts at no more than $250 billion.

Damodaran’s calculation is not a prediction. He starts with the price investors have placed on AI businesses and works backward to ask what those companies must eventually deliver.

Damodaran is the Kerschner Family Chair in Finance Education and a professor of finance at New York University’s Stern School of Business. Known as the “Dean of Valuation,” he is closely followed for his work on the numbers needed to support company and stock-market valuations.

He assigned the industry a blended operating margin of 20%. Under that assumption, it would need about $5 trillion in annual sales if the AI market matured within 10 years.

If it took 15 years, the requirement would exceed $8 trillion. The longer investors must wait, the larger the eventual business needs to be for today’s valuation to make sense.

Damodaran published the analysis on Thursday, 20 August.

“Big markets don’t always become big businesses,” he wrote.

AI sales remain far behind spending

Damodaran estimates that AI products and services currently generate no more than $250 billion in annualized revenue. That figure includes a reported revenue run rate of more than $65 billion at Anthropic and about $40 billion at OpenAI. It also includes AI revenue attributed to Microsoft, xAI and other companies.

A run rate takes sales from a recent month or quarter and projects them across a full year. It is useful for tracking fast-growing companies but does not represent revenue already earned over 12 months.

Anthropic’s figure reflected its annualized revenue run rate at the end of July, Reuters reported.

The biggest revenue and operating-profit beneficiaries so far have been infrastructure suppliers, Damodaran said. Chipmakers, electrical-equipment companies and power providers are being paid as data centers are built. AI model and application developers still have to demonstrate that their economics can support durable profits.

Damodaran calculated that Alphabet, Amazon, Meta, Microsoft, Oracle and CoreWeave had invested a combined $1.7 trillion in recent years. Not all that spending can be classified strictly as AI investment. Data centers and cloud systems often support several services. The figure nevertheless shows the scale of the infrastructure being built around AI.

Hyperscaler capital expenditure exceeded $400 billion in 2025 and could approach $800 billion in 2026, according to JPMorgan Asset Management. Damodaran called this infrastructure the “most expensive factory in history.” That buildout raises the amount of revenue and profit the AI ecosystem must eventually generate.

Anthropic needs far more growth

Damodaran also tested what Anthropic would need to achieve at a rumored IPO valuation of $2 trillion. He assumed the company would eventually retain 30 cents in after-tax operating profit from every dollar of revenue. Under his model, Anthropic would need close to $1.2 trillion in annual revenue within 10 years.

If it took 15 years to mature, the revenue requirement would approach $2 trillion. The $2 trillion valuation is a scenario, not Anthropic’s current price. The Claude developer raised $65 billion in May at a reported valuation of $965 billion.

Damodaran did not say the calculation proved Anthropic was overpriced. The answer depends largely on how widely AI is eventually used.

Investors who believe AI agents will replace employees across many industries can argue for a much larger market. Those who expect AI to remain mainly a tool that helps people work faster will arrive at a lower figure.

How large can AI become

Damodaran placed the theoretical upper limit for the global AI market at about $26 trillion, based on total employee compensation. This is a ceiling rather than a realistic revenue estimate because reaching it would imply replacing virtually all paid labor.

The realistic market will depend on which jobs AI can perform, how much companies will pay and whether the technology replaces workers or helps them become more productive.

Software and financial services are among the most exposed industries because much of their work is digital and can be checked relatively easily. Jobs involving physical work, personal contact or difficult judgment are harder to automate.

AI replacement also makes more financial sense in countries where wages are high.

AI remains expensive to run

AI companies may struggle to achieve the profit margins investors associate with traditional software. Once software has been developed, selling another copy costs almost nothing. AI services, however, incur computing costs every time a customer submits a request.

The cost of producing individual AI tokens has fallen as chips and models improve. More capable systems, however, use more tokens, process more data and require more expensive infrastructure.

This is pushing AI companies towards charging customers according to how much computing power they use. Subscription plans will probably remain available, but heavy use is likely to come with limits or additional charges.

Damodaran expects basic AI products to compete on price and scale. More advanced systems will compete on performance, customization and access to proprietary data.

Power shortages, water constraints, opposition to data centers, privacy requirements and measures to protect displaced workers could raise costs further.

Damodaran stopped short of calling AI a bubble but said the industry had reached its “bar mitzvah moment,” a coming-of-age point at which companies must begin converting technological promise into revenue, profits and cash flow.