Consumer AI is back in the spotlight. Meta’s personal AI assistant Muse, together with the cartoon character Jolly, unexpectedly gained popularity very quickly, OpenAI released its own agentic assistant Dots, and the startup Instinct, which specializes in handling everyday errands, has already reached a $10 billion valuation.

At first glance, this looks like the beginning of a new wave of consumer AI. Agentic systems have finally become reliable enough to book trips, make restaurant reservations, cancel subscriptions, and handle other everyday tasks on their own. For users, this is real value, and for investors, it is a potentially new mass market.

But this story has a serious problem: the popularity of consumer AI is still not translating into equally compelling economics. Running AI models is significantly more expensive than operating most previous internet services, while users’ willingness to pay for AI is growing much more slowly than the capabilities of the models themselves, TechCrunch reports.

The number of users is growing, but few are willing to pay

Data from the semiannual Andreessen Horowitz State of Markets report, based on research by PNC Research, reflects the current situation.

In May 2026, about 2.2% of consumers were using paid AI services, with average spending of $31 per month. The figures are gradually increasing, but the pace remains fairly linear.

At the same time, model capabilities changed much faster over the same period. The significant performance jump between generations of AI models is barely reflected in the chart of consumers’ willingness to pay more or sign up for paid subscriptions.

Andreessen Horowitz views the low penetration rate as a sign that the market is still at an early stage. But there is another interpretation: it is possible that further technical improvements in AI alone will not lead to proportional growth in consumer spending.

Other studies present a somewhat more optimistic picture. In March, Bank of America estimated the share of American consumers who pay for AI at about 3% — 40% more than a year earlier. According to a September study by Menlo, about a quarter of adults use AI daily, and half of them pay for such services.

Even these estimates, however, do not change the main problem: most users are still not willing to pay regularly for AI as much as would be required to cover its cost.

Even a huge audience may not make AI profitable

Consumer services have another distinguishing feature: the problem lies not only in the size of revenue, but also in the cost of generating it.

AI is noticeably more expensive to operate than many previous mass-market internet technologies. Social networks and other digital services can serve huge audiences with relatively low marginal costs. Generating responses with powerful AI models requires significantly more computing resources.

That is why even hundreds of millions of paying users do not necessarily guarantee profitability.

Netflix can be used as an example of a mature mass online service for comparison. With an audience of 325 million subscribers, even an average of $34 per user per month would generate about $11 billion in annual revenue.

For OpenAI, that is less than one-third of the operating expenses the company incurs today.

In other words, the problem of consumer AI is not only about finding enough people willing to pay. It is also necessary to ensure that the cost of serving each user allows the company to make money from that audience.

That is why AI companies are moving into the enterprise sector

It is precisely this economic reality that explains why the largest AI developers are increasingly focusing on business customers.

OpenAI, for example, has significantly strengthened its enterprise direction in recent months. According to Axios, the company’s annual recurring revenue has approached $70 billion, while enterprise bookings have doubled since July.

Even the launch of Dots demonstrates this strategy. Although the product is positioned as a personal agent, OpenAI is simultaneously showcasing its use for developers and creative agencies.

The logic here is quite simple: if a popular consumer service cannot generate enough money from individual subscribers, its capabilities can be sold to companies at a higher price.

This is especially important for AI. A business client can pay significantly more if the system saves employees time, automates operations, or replaces part of manual work. For an ordinary user, the value of a chatbot response or a completed errand is often limited to a much lower amount.

Muse and Instinct have their own ways to make money

Muse and Instinct are trying to build a consumer AI business differently.

Meta has a huge advantage in the form of advertising infrastructure and personalized targeting. This gives the company more monetization options and allows it not to turn a Muse subscription into its only source of income.

Meta is already also considering enterprise applications for Muse, expanding the AI agent to small businesses.

Instinct has a different model. The startup expects to earn commissions from purchases that users make through the agent. If AI really becomes an intermediary between people and stores, travel services, restaurants, and other services, such a model could prove significantly more scalable than a standard subscription.

In addition, Instinct is not required to independently finance the development of its own cutting-edge foundation model on the same scale as the largest AI labs. This potentially reduces one of the most expensive cost items.

The main question is whether consumer AI can move beyond subscriptions

The success of Muse, Dots, and Instinct shows that demand for personal AI agents really does exist. But a popular product and a profitable business are not the same thing.

Consumer AI faces an unusual combination of two constraints: users are not willing to pay endlessly high amounts, while fulfilling AI requests remains expensive. The more capable the agent becomes and the more tasks it takes on, the higher the product’s potential value — but at the same time, computing costs also rise.

That is why large AI companies are gradually arriving at the same model: a consumer product may be an important entry point, but serious business scale has to be found in enterprise contracts, transaction commissions, or other sources of revenue.

This is where the main limitation of today’s consumer AI lies. The technology is becoming noticeably better, agents are learning to handle more and more everyday tasks, and demand is growing. But the underlying math is still changing much more slowly than the technology itself.