Uber burned through its entire 2026 AI budget by the start of the second quarter. One of the most technologically sophisticated companies on the planet ran out of AI money before spring was over.

The mix-and-match era

Rather than lock themselves into expensive contracts with a single AI provider, companies are blending models from multiple vendors, cherry-picking cheaper or more specialized options depending on the task at hand.

The Wall Street Journal reported on May 28 that this strategy of mixing and matching has become a defining trend among both startups and established tech players. The motivation is straightforward: soaring compute and inference costs have forced enterprises to ration their AI resources.

OpenAI eyes price cuts as Anthropic circles

OpenAI is reportedly considering drastic reductions to its token pricing, the primary billing mechanism for AI usage. Token pricing is essentially how AI companies charge per unit of text processed. Lower token prices mean cheaper AI for everyone downstream.

The catalyst isn’t altruism. It’s competition. Anthropic, maker of the Claude family of models, has been steadily gaining ground in the enterprise market. OpenAI’s potential price cuts are a direct response to that pressure.

When Altman publicly acknowledged that AI costs represent “a huge issue,” he was laying the groundwork for a strategic pivot.

Both OpenAI and Anthropic have confidentially filed for IPOs ahead of these pricing discussions.

What this means for investors and the broader market

If OpenAI and Anthropic start aggressively undercutting each other, the short-term winners are enterprises that have been struggling to afford AI at scale. But for investors eyeing those upcoming IPOs, price cuts compress margins for companies that already spend enormous sums on compute infrastructure and talent.

The Uber anecdote is particularly telling. If a company with Uber’s resources and technical sophistication is blowing through its AI budget in a single quarter, it suggests that current pricing levels are simply unsustainable for widespread enterprise adoption.

When companies treat AI models as interchangeable commodities, model differentiation starts to matter less than cost efficiency and integration flexibility.