The bill came due fast
Uber Technologies burned through its entire 2026 AI budget within the first few months of the year. The culprit: a company-wide push to maximize employee use of frontier AI tools, including Anthropic's Claude Code, complete with internal 'leaderboards' ranking software engineers by usage volume.
The strategy had a name — 'tokenmaxxing' — and a predictable ending. Uber chief technology officer Praveen Neppalli Naga told The Information he went 'back to the drawing board' on allotted spending after the blitz failed to deliver returns that justified the pace of consumption.
Engineering the fix
Naga announced a course correction Wednesday in an X post. Uber quadrupled the number of employees using frontier AI tools while simultaneously bringing down the cost per token. The levers: improved prompt caching, adjusted default model settings, evaluation of new models for efficiency, and giving engineers real-time visibility into their own AI usage and hourly costs.
'You might expect costs to rise as adoption accelerates,' Naga wrote. 'We've seen the opposite. Not because we've restricted access, but because we've treated efficiency as an engineering problem rather than a budget problem.'
His conclusion: 'The next phase, whatever we call it, will not be characterized by who spends the most tokens, but about how people use them as efficiently as possible.'
Returns remain elusive
The efficiency gains arrive against a backdrop of unresolved questions about AI's actual productivity payoff. As of May, Uber president and chief operating officer Andrew Macdonald acknowledged the link between AI investment and measurable output was not yet visible. 'That link is not there yet,' Macdonald said on the Rapid Response podcast. 'It's very hard to draw a line between one of those stats and 'Okay, now we're actually producing like 25% more useful consumer features.''
Deutsche Bank Research Institute's global head of macro and thematic research Jim Reid warned last month that AI productivity gains were still years away. Profit margins for the Magnificent Seven grew from 15% to 25% between Q1 2023 and Q1 2026, while the rest of the S&P 500 saw only 10% margin growth over the same period — suggesting AI returns remain concentrated in the immediate tech sector.
The Jevons trap
Even Uber's efficiency win carries a structural risk. According to the Silicon Data Token Expenditure Index, the price of a single token has dropped more than 90% since 2023 — yet large language model spending has doubled since late last year. A Bain & Co. brief published in June found token costs halved from December 2024 to 2025, while tokens consumed grew 450% over the same period.
Apollo chief economist Torsten Slok framed the dynamic clearly: 'As tokens get cheaper, companies don't spend less but instead run more AI agents, automate more workflows, and generate more code, pushing aggregate expenditure higher even as the unit cost of intelligence collapses.'
Naga did not specify whether Uber is consuming more or fewer total computing resources than earlier this year.
The market is watching the math
The numbers come first, and here they are unambiguous: a budget blown in months, a pivot to efficiency, and an industry-wide pattern where cheaper inputs produce higher total bills. For enterprise leaders, the Uber episode is a case study in what happens when adoption incentives outrun financial discipline. Free enterprise rewards efficiency — but only when management treats cost as a constraint, not an afterthought. The tokenmaxxing era may be ending; the accountability era is just beginning.
