The companies that own the customer relationship will ultimately beat those that simply own the best AI

The economics of frontier AI are starting to show cracks, with one notable example being Anthropic. As this company nears what could be the largest IPO in history, it faces a troubling sign: many of its U.S. customers are opting for less sophisticated AI models over Anthropic's most advanced one. This is significant because it challenges a fundamental assumption about frontier AI—the belief that the most advanced models will always come with a price premium.

This belief has been a cornerstone of investor confidence, with hundreds of billions of dollars being poured into the development of top models. However, this economic model only makes sense if these advanced models can continue to generate superior margins compared to "good enough" alternatives. Currently, both Anthropic and OpenAI have reported substantial revenue growth, with Anthropic even achieving its first adjusted operating profit.

Yet, Anthropic is struggling to convince customers to pay for its leading model, Fable 5. Companies are racing to adopt AI, but with increasing operational costs, they are becoming more cost-conscious. In response, OpenAI and Anthropic have launched cheaper models below the frontier while maintaining premium pricing for their flagship systems.

However, if "good enough" proves to be more valuable in delivering business results, the economics of this strategy may no longer hold. This issue is not solely a pricing problem, but a sign that the business model behind frontier AI is beginning to falter. The technology is advancing at a pace faster than the economic rationale behind it.

While the latest models are more capable, many companies are finding that they do not yield correspondingly better outcomes or productivity. This is where the strategic error begins. Anthropic and OpenAI are relying on the wrong source of competitive advantage. They assume that continuously pushing the frontier will be sufficient to dominate the enterprise market.

This belief is misguided, as it conflates technological leadership with commercial leadership. These are two distinct concepts that are no longer interchangeable. To illustrate this point, consider Uber's recent strategy. Earlier this year, the ride-hailing company acquired an entire year's worth of AI tokens—units of usage charged to customers—in just four months.

Uber then began to differentiate tasks, assigning simpler jobs to cheaper models while reserving the most expensive systems for the most challenging work. The result was remarkable: AI usage increased more than ninefold, without a proportional increase in spending. The key takeaway is clear: companies will no longer standardize on a single frontier model.

Instead, they will increasingly purchase outcomes. In this new paradigm, the frontier model will no longer be the product but merely another input. Consequently, the business model must adapt. Companies must stop selling AI models and start selling outcomes. The first step is to shift focus from single AI models to portfolios. Like Uber, companies will increasingly assemble portfolios of models at different price points, selecting the one that delivers the best result for each specific task.

Frontier AI labs must adapt to this reality by offering a diverse portfolio of models, competing based on the business outcomes they enable rather than the performance of any single system. In this new landscape, today's frontier AI could become tomorrow's "freemium" offering—cheap or free enough to attract customers. Finally, the next step is to transform these models into a platform that automatically selects the optimal AI for every task.

Once this is achieved, the competitive battle will shift from models to the products and services built around them.