After blowing its entire 2026 AI budget in months, Uber CTO says ‘We’re coming to the end of the so-called 'tokenmaxxing' era’

Uber's CTO Praveen Neppalli Naga had a rough start to the year on the AI budget front. After the company leaned hard into encouraging employees to use tools like Anthropic's Claude Code as much as possible, going as far as building leaderboards to rank engineers by usage, Uber tore through its entire AI spending allotment within a few months. Naga told The Information he had to go back to the drawing board.

What happened at Uber is basically a case study in what people are now calling tokenmaxxing: companies pushing employees to burn through AI tokens as fast as possible, only to realize the spending wasn't translating into anything they could point to as a real return. Uber wasn't alone in that trap, but Naga says the company has since found a workaround that doesn't involve just cutting people off.

Instead of restricting access, Uber quadrupled the number of employees using frontier AI tools while actually bringing down the cost per token. It did this through better prompt caching, tweaking default model settings, evaluating models for efficiency, and giving engineers visibility into their own usage and hourly costs. Naga framed it as treating efficiency like an engineering problem instead of a budget problem, and he says the result defied the usual expectation that costs climb as adoption grows.

The bigger question hanging over all this is whether any of it is actually paying off. Uber's own president, Andrew Macdonald, admitted back in May that the company couldn't draw a clean line between AI adoption and concrete gains like shipping meaningfully more useful features. That uncertainty lines up with broader numbers cited in the piece: profit margins for the Magnificent Seven jumped from 15% to 25% between the first quarters of 2023 and 2026, while the rest of the S&P 500 saw just 10% growth, suggesting AI's payoff is still concentrated in a narrow slice of the economy.

And there's a wrinkle even in Uber's efficiency win. Economists point to something called Jevons paradox, where cheaper resources lead to more consumption overall rather than less. Token prices have dropped more than 90% since 2023 by one index's measure, yet LLM spending has doubled since late last year, and a Bain and Co. brief found token costs halved from late 2024 to 2025 while consumption grew 450% over that stretch. Naga didn't say whether Uber's total AI usage went up or down, only that the company's philosophy has shifted toward efficient use rather than sheer volume.