The economics of frontier AI are starting to crack, and nowhere is that more clear than at Anthropic. As the company heads towards what could become the biggest IPO ever, it faces an ominous signal: Many of its U.S. customers are choosing cheaper artificial intelligence models over its most advanced option.
That matters because it is testing a core commercial assumption underpinning frontier AI: that the most advanced models will continue to command a price premium. It’s an assumption that investors have already backed with hundreds of billions of dollars.
Those sums are being deployed into an extraordinarily capital-intensive race to develop top models. But that makes economic sense only if those models can continue to earn superior margins over “good enough” alternatives.
For now, the numbers still look solid. Both Anthropic and OpenAI have reported sharply accelerating revenues in recent months, and the former has even recorded its first adjusted operating profit. Yet Anthropic faces growing difficulty in persuading customers to pay for its leading model, Fable 5.
Companies have rushed to deploy AI. But as the bills mount, they are becoming more cost conscious. OpenAI and Anthropic have responded by introducing cheaper models below the frontier, while continuing to defend premium pricing for their flagship systems.
But if “good enough” beats “best” in delivering business value, the economics of that strategy no longer add up. To be clear, this is not simply a pricing problem, but the first sign that the business model underpinning frontier AI is starting to wobble.
The AI business model is breaking
The technology is advancing faster than the business case for it.
The latest models may be more capable, but for many companies they are not delivering proportionately better outcomes or productivity. That’s why many customers are deciding they simply aren’t worth the premium.
That’s where the strategic mistake begins. Anthropic and OpenAI are betting on the wrong source of competitive advantage. They believe spending ever more money pushing the frontier will be enough to dominate the enterprise market.
It probably won’t.
They are confusing technological leadership with commercial leadership. Those two things are no longer the same.
Uber shows us where the enterprise market is heading. Earlier this year, the ride-hailing company gobbled up a year’s worth of AI tokens—the units of usage used to bill customers—in just four months.
Uber then stopped treating every task the same; simple jobs went to cheaper models, while the most expensive systems were reserved for the hardest work. The payoff was huge. AI usage across the business increased more than ninefold without a corresponding rise in spending.
The lesson is quite obvious: Companies won’t standardize on one frontier model any longer. Increasingly they’ll buy outcomes instead, and when that happens, the frontier model stops being the product and becomes just another input. The business model therefore has to change.
Stop selling models. Start selling outcomes
The first step is changing what they sell. Too many companies still think they need just one AI model. They don’t. Like Uber, they’ll increasingly assemble portfolios instead, selecting whichever delivers the best result for the task at hand.
Frontier labs need to adapt to that reality by offering portfolios of models at different price points and competing on the business outcomes they deliver instead of the performance of any single system. In that world, today’s frontier AI becomes tomorrow’s “freemium” offering: free or cheap enough to get customers in the door.
The next step is to turn those models into a platform that automatically chooses the optimal AI for every task. At that point, the competitive battle shifts from models to the products and services wrapped around them.