The ugly economics of consumer AI

This week’s launches — Meta’s personal AI assistant Muse (with its plush mascot, Jolly), OpenAI’s Dots (released yesterday), and the $10 billion-valued agentic assistant Instinct — have revived the idea that consumer AI could be a hit business again. Agentic tools now handle everyday tasks like booking travel or canceling subscriptions reliably enough that, to some investors, it looks like the 2022 ChatGPT moment all over again.

But frontier labs have grown cautious about consumer AI for a different reason: the economics. Even wildly popular AI products are hitting a ceiling on what consumers will pay, and better models haven’t clearly made the consumer business more profitable — pushing much of the industry toward an enterprise-first approach instead.

Andreessen Horowitz’s semiannual State of Markets report, citing a PNC research report from this summer, shows just how linear that adoption curve has been: as of May, only 2.2% of consumers were paying for AI services, spending an average of $31 a month. The firm called this “still so early when it comes to mature AI adoption and utilization,” but the jump between GPT-5.2 and Astra — a huge leap in model capability — is barely visible in the growth trend.

The per-consumer math doesn’t help much either. Using Netflix’s 325 million subscribers as a benchmark for a mature, market-saturated service, $34 per customer would generate only about $11 billion a year — less than a third of OpenAI’s operating costs. Other surveys tell a similar story: Bank of America found roughly 3% of U.S. consumers paid for AI in March, up 40% year-over-year, while a Menlo survey from September found a quarter of adults use AI daily, with half of those paying.

The deeper issue isn’t revenue, but cost: AI is far more expensive to run than earlier consumer technologies like social networks or cloud computing, so even hundreds of millions of paying users wouldn’t guarantee profitability. OpenAI has adapted by leaning into enterprise, with bookings reportedly doubling since July and even the consumer-facing Dots pitched partly at software engineers and agency creatives. Muse benefits from Meta’s ad-targeting business as a cushion, and Meta is already exploring its own enterprise angle; Instinct plans to take a cut of purchases made through the agent instead.

Why it matters

The underlying economics put a hard ceiling on how big a consumer AI product can get without an enterprise revenue stream to back it up — a lesson the major labs have already absorbed, and one any new entrant will eventually have to confront.