The Tough Economics of Consumer AI

 

Meta’s personal AI assistant, Muse, and its plush-like mascot Jolly have become a surprise hit. OpenAI’s Dots, launched just a day ago, appears to be pursuing a similar cartoon-style personal assistant concept. Meanwhile, up-and-coming Instinct has reached a $10 billion valuation, driven by its ability to handle everyday tasks such as booking travel, making restaurant reservations, and cancelling subscriptions.

 

The bullish case for consumer AI is easy to understand. Agentic AI is becoming reliable enough to handle real-world tasks, and companies are increasingly bringing those capabilities directly to consumers. For investors, it can look a lot like the launch of ChatGPT in 2022: a major leap in AI capabilities creating an entirely new product category.

 

But there’s another side to the story. Frontier AI labs have become increasingly cautious about consumer products—not because the technology isn’t capable, but because consumer willingness to pay remains limited. Even highly popular AI products are struggling to translate massive improvements in model performance into proportionally larger consumer revenues.

 

That has pushed much of the industry toward an enterprise-focused model, with companies targeting business contracts and expanding into specialized verticals.

 

The success of products like Muse and Instinct could challenge that trend, particularly if they can find alternative ways to monetize users. But the underlying economics of consumer AI remain difficult.

 

Andreessen Horowitz’s latest State of Markets report offers a useful snapshot. Drawing on data from PNC Research, it found that as of May, just 2.2% of consumers were paying for AI services, with an average monthly spend of $31.

 

Andreessen Horowitz describes this as evidence that AI adoption is still in its early stages. There is certainly significant room for growth. But the data also suggests that consumer adoption and spending are increasing relatively slowly, even as AI models continue to improve dramatically.

 

The gap becomes even clearer when looking at potential revenue. Using Netflix’s roughly 325 million subscribers as a benchmark for a mature consumer subscription business, an average of $34 per customer would generate approximately $11 billion in annual revenue—still less than one-third of OpenAI’s reported operating costs.

 

Other surveys paint a somewhat more optimistic picture. Bank of America found in March that around 3% of U.S. consumers were paying for AI, up 40% from the previous year. A September Menlo survey offered an even brighter outlook, finding that about one-quarter of adults use AI daily and roughly half of those users pay for AI services.

 

But the bigger challenge for consumer AI may not be revenue—it’s cost.

 

AI is extraordinarily expensive to operate compared with earlier internet businesses such as social networks and many cloud-based services. Even hundreds of millions of paying users may not guarantee profitability if inference and infrastructure costs remain high.

 

OpenAI appears to have recognized this reality. Its shift toward enterprise customers has reportedly gained momentum, with enterprise bookings said to have doubled since July. Even the Dots launch included an enterprise angle, highlighting potential applications for software engineers and creative agencies.

 

That follows a familiar technology-business playbook: build a popular consumer product, then monetize the underlying technology through higher-value enterprise customers.

 

For Muse and Instinct, the path is less clear.

 

Meta has the advantage of a massive advertising ecosystem and sophisticated personalization infrastructure, giving Muse more potential monetization options and more time to develop its business model. Meta is also exploring enterprise opportunities.

 

Instinct is taking a different approach by potentially earning a percentage of transactions completed through its agent. That could increase its revenue potential, while avoiding the enormous cost of developing and training a frontier model could improve its economics.

 

Still, the fundamental challenge remains: consumer AI has a difficult cost structure and limited consumer willingness to pay.

 

Products may become dramatically more capable, but that does not automatically mean consumers will pay dramatically more. For companies building consumer AI businesses, eventually finding a path into enterprise revenue—or another high-margin revenue stream—may be essential to achieving sustainable scale.

 

The technology is changing rapidly. The economics are proving much harder to change.