Three consumer AI assistants landed within days of each other, and TechCrunch’s Russell Brandom thinks the money behind them does not add up. In an analysis published Wednesday, he argues that people paying for AI is still a small slice of the public, and that serving them is expensive enough to cap how big a consumer-only company can get.
The week’s launches make the case for a comeback look strong. Brandom calls Meta’s Muse assistant a surprise hit, with a soft, toy-like character named Jolly acting as its face. OpenAI released Dots on 29 September, a cartoonish personal agent aimed at the same audience. Instinct, an assistant that books travel, makes restaurant reservations and cancels subscriptions, has reached a $10 billion valuation. Brandom grants the bull argument: agents now work reliably enough for everyday errands, and investors see a repeat of the 2022 ChatGPT launch.
His counterweight is payment data. Andreessen Horowitz’s semiannual State of Markets report, which took its numbers from PNC research published over the summer, found that 2.2 percent of consumers were paying for AI as of May, at an average of $31 a month. Brandom notes that both lines on the report’s charts rise slowly and almost in a straight line. The jump from GPT-5.2 to Astra, a big leap in capability, barely shows up. Better models, in other words, have not visibly changed how many people pay or how much.
Other surveys land nearby. Bank of America found roughly 3 percent of U.S. consumers paying in March, up 40 percent on the year before. A September Menlo survey is friendlier: 25 percent of adults use AI every day, and half of that group pays. Andreessen’s own read is that adoption is “still so early,” a fair point that the charts do not contradict.
The sharper problem, Brandom writes, is cost. AI is far more expensive to run than earlier consumer products such as social networks or cloud storage. His back-of-envelope test uses Netflix as the picture of a saturated subscription service. Take Netflix’s 325 million subscribers, assume about $34 from each, and the total is $11 billion a year. By his figures, that covers less than a third of what it costs OpenAI to run its business. Hundreds of millions of paying users, he argues, would still not guarantee a profit.
That explains the industry’s tilt toward business. Brandom describes OpenAI’s pivot to enterprise as largely successful, citing reports that enterprise bookings have doubled since July. Even the Dots launch, he notes, leaned on pitches to software engineers and agency creatives. He compares it with an old play: sell a popular, cheap service to companies at a markup.
Muse and Instinct are harder to judge. Meta has its personalized ad targeting to monetize Muse, which buys time before profit becomes urgent, and the company is already testing an enterprise angle for the assistant. Instinct has its own route: a commission on whatever users buy by way of its agent, which could lift its ceiling, and it presumably avoids the expense of training a frontier model. Brandom still concludes that, without business revenue, there is a hard limit on how large Instinct can grow.
This is an opinion piece built on third-party survey numbers, so the 2.2 percent figure is a snapshot from one report and not a measured law. The Netflix comparison is Brandom’s illustration, not a forecast. Still, it frames the choice for any startup pitching a consumer assistant this autumn: either show a revenue source beyond subscriptions, such as commissions or ads, or expect investors to ask why a business tier is not on the roadmap.
Reported by TechCrunch (Russell Brandom) on 30 September 2026.
