Consumer AI's Comeback Runs Into Brutal Unit Economics

Only 2.2% of consumers were paying for AI as of May, at an average spend of $31 a month — the sobering backdrop to this week's wave of consumer AI launches.

The products themselves look like winners. Meta's personal AI assistant, Muse, with its plush-like mascot Jolly, has become a surprise hit. OpenAI's Dots, released yesterday, appears to be chasing the same cartoony personal assistant concept. And the up-and-coming Instinct assistant just reached a $10 billion valuation on the strength of its agentic errand-running — booking travel, making restaurant reservations, cancelling subscriptions.

The bull case writes itself. Agentic AI has finally gotten reliable enough to handle everyday tasks, and companies are increasingly pitching that service to ordinary people who are getting genuine value from it. For investors, the setup looks a lot like the ChatGPT launch in 2022, when raw AI power opened a product category that never existed before.

There is a reason frontier labs have turned gun-shy about consumer AI. It is not because the technology falls short. Even staggeringly popular tech products are hitting a ceiling on how much money consumers will pay, and better models do not appear to be producing a more profitable consumer business. The result has been an industry-wide shift toward what the industry calls the Anthropic model: enterprise contracts and vertical-by-vertical expansion.

If Muse and Instinct are bucking that trend, it is largely because they are not focused on monetization yet. The underlying economics of consumer AI are not improving.

Andreessen Horowitz's semiannual State of Markets report, drawing on a PNC research report from this summer, tracks the slowly growing share of consumers paying for AI alongside the slowly growing amount they pay. Andreessen puts a positive spin on the data, saying, "it's still so early when it comes to mature AI adoption and utilization." But both charts look awfully linear. Even as models make huge improvements, the number of customers willing to pay for AI — and the amount they will pay — barely moves. The enormous performance jump from GPT-5.2 to Astra is barely visible on the chart.

The per-consumer math is worse. Take Netflix as the standard for a market-saturated online service, with 325 million subscribers. At $34 per customer, that yields only $11 billion in annual revenue — less than a third of OpenAI's operating costs.

Bank of America offers similar numbers. In March, the firm found roughly 3% of U.S. consumers paid for AI, up 40% from the previous year. A Menlo survey from September paints a slightly sunnier picture: a quarter of adults use AI daily, and half of those users pay for it.

The core problem is cost, not revenue. AI is an unusually expensive technology to operate, particularly next to lightweight predecessors like social networking or cloud computing. Even hundreds of millions of paying customers does not guarantee break-even.

OpenAI appears to have adapted. Its widely reported pivot to enterprise has been largely successful, with enterprise bookings reportedly doubling since July. Even the Dots launch carried a strong enterprise angle, demonstrating how the new personal agent could serve software engineers and agency creatives. The long-standing playbook — sell popular-but-cheap consumer services to businesses at a markup — is exactly what OpenAI seems to be following.

The outlook for Muse and Instinct is less clear. Muse has the juggernaut of Meta's personalized ad targeting behind it, which buys more monetization options and more time. Meta is already exploring the enterprise angle. Instinct plans to take a cut of purchases made through its agent, which might raise the ceiling, and it will presumably avoid the cost of training a frontier model.

Still, the economics of consumer AI put a hard cap on how large any player can grow without tapping enterprise revenue. The major labs have already learned that lesson. It is one of the few things about this industry that does not seem to be changing.