A new report from Exponential View estimates that the generative AI economy produced $110 billion in revenue over the past 12 months and is now operating at an annualized run rate of about $175 billion.
The analysis, published as part of the outlet's first State of the AI Economy report, says the figures were built from the bottom up and adjusted to avoid double-counting spending that flows through multiple layers of the AI supply chain. The authors said the work is based on a proprietary model that combines public disclosures, supplier data, customer information, leaks and self-reported figures where available.
The report argues that the supply side of the AI market is relatively easy to observe because many of the major beneficiaries are public companies. Chipmakers, memory suppliers, power equipment vendors and data center operators often disclose sales, order books and investment plans. By contrast, the demand side is harder to track because much of the revenue comes from privately held firms such as OpenAI, Anthropic, Cursor and ElevenLabs, which are not required to publish financial results.
To estimate total spending, the researchers said they counted only the amount paid by the end customer. That means if one company pays another AI provider, and that provider then pays a cloud vendor to deliver the service, the model aims to capture the customer dollar once rather than multiple times. The authors said this approach gives a more accurate picture of actual market size.
The report does not include all forms of AI-related value. It excludes internal productivity gains from AI tools, such as improved ad targeting at Meta or Google, as well as savings from internal use by major technology companies. It also leaves out professional services and systems integration costs, which can make up a large share of corporate AI projects. Chinese data is also not included in this version of the report.
Exponential View said the AI revenue base is expanding faster than prior technology waves, at roughly three times the pace of the mobile and internet eras. The report adds that many companies have moved past small pilots but remain early in the process of scaling AI across their operations.
The authors also looked at whether AI revenue can support the heavy infrastructure spending required to build and run the systems. Their model separates AI-specific capital spending from broader spending by cloud providers and spreads the cost of computer equipment and other infrastructure over several years. Under those assumptions, the report says revenues tied to hyperscalers are just enough to cover depreciation costs.
The analysis further argues that falling token prices do not necessarily reduce total market spending. It estimates that a 10 percent price cut tends to increase token usage by 12 percent to 18 percent, which can still push total spending higher. The report suggests that tokens are a useful billing unit, but not necessarily the best measure of the economic value created by AI.
Beyond the revenue estimate, the report says it examines how AI demand is affecting electricity use, how token pricing is changing and how consumption-based billing could expand the market. It also outlines several scenarios for future growth based on different assumptions about prices and model capability.
The authors said the first version of the report is intended as a starting point and invited feedback as they refine their approach. Even with the uncertainties around measurement, the headline conclusion is clear. AI has become a substantial business already, and the market appears to be growing quickly.