HBS research puts paid AI adoption at 2.9% of US consumers

Only 2.9% of US consumers paid for at least one major generative AI service, according to research highlighted November 13 by HBS Working Knowledge. The study estimates about 6 million paying individuals as of April 2026—and finds that winning a subscription is different from keeping it, with low retention across the four providers examined.

The findings come from the October working paper “Competition in Generative AI: Evidence from Consumer Demand.” Its authors are Harvard Business School professor Mark Egan, HBS doctoral students Dominic Russel and Claire Shi, and Georgia State University professor Vincent Yao. They examine consumer demand for services from OpenAI, Anthropic, Google and xAI.

Paying customers, not the whole AI audience

The team analyzed 114 billion transactions collected by financial technology company Fiserv from early 2021 through April 2026. That cutoff matters: these are estimates of the paying market earlier this year, not a November snapshot. The analysis focuses on consumer payments and excludes enterprise clients, so it should not be read as a measure of the entire AI business.

The distinction also affects comparisons between providers. The paper describes paying consumers as accounting for about 60% of OpenAI’s revenue at the beginning of 2026. Individual subscribers historically accounted for 20% of Anthropic’s revenue, with enterprise clients making up a larger share. Consumer subscription behavior therefore captures different portions of those two businesses.

Paying subscribers also skew toward a particular audience. They tend to earn more than $200,000 a year, according to the research. People younger than 55 and residents of coastal or major metropolitan areas are more likely to pay. The study’s consumer market is not evenly distributed across income, age or geography.

AI enthusiasm does not guarantee platform loyalty

The researchers distinguish interest in AI from attachment to a particular provider. Their model characterizes subscribers as AI enthusiasts, but not platform loyal. It also finds that Anthropic’s and OpenAI’s services have become more substitutable over time—meaning consumers increasingly treat one as an alternative to the other.

That differs from the habit of paying for several media subscriptions, such as Netflix and Disney+. The team created a “multihoming index” to track consumers’ willingness to subscribe to multiple AI services. It was high but declining across all companies. The researchers argue that tools combining functions such as coding and writing give people less reason to maintain several subscriptions.

A premium for some, price pressure overall

Some customers value a specific service much more than others do. The model estimates that consumers in the top quarter of the ChatGPT preference distribution would pay $37 more per month than those in the bottom quarter. That is a difference between groups’ willingness to pay, not a proposed subscription price.

Yet the broader pricing result points in the opposite direction: a 10% price increase produces a 46% decline in modeled demand. This is a model estimate, not an observed response to an announced price rise. The researchers interpret the results as strong competitive pressure, with high prices pushing consumers toward rival providers.

The study discloses ties to its data supplier: Egan and Yao are unpaid members of Fiserv’s Small Business Index advisory council, and Yao also works as a paid consultant.

For paying subscribers, the researchers estimate about $35 per quarter in additional value from AI subscriptions. Their conclusion is conditional: unless providers substantially differentiate model quality, increasingly interchangeable services will constrain pricing power and leave consumers capturing a large share of the value.