Uber cuts AI costs by quadrupling adoption, not restricting use

Uber has reversed its AI spending crisis not by cutting back on usage, but by dramatically expanding it—a counterintuitive strategy that its CTO says demonstrates how enterprises should approach AI cost management at scale.

The rideshare company exhausted its AI budget within months earlier this year after encouraging widespread employee adoption of tools like Anthropic's Claude Code, even creating leaderboards to gamify usage among software engineers. That aggressive push exemplified "tokenmaxxing," a trend where companies incentivized maximum AI tool usage without clear returns on investment.

The efficiency-through-scale approach

Rather than impose restrictions, Uber Chief Technology Officer Praveen Neppalli Naga told The Information the company "went back to the drawing board" and treated the problem as an engineering challenge. The result: Uber quadrupled the number of employees using frontier AI tools while simultaneously reducing the cost per token.

The company achieved this through several technical improvements, including better prompt caching, adjusting default model settings, evaluating new models for efficiency, and giving engineers visibility into their hourly AI usage and costs. "We've treated efficiency as an engineering problem rather than a budget problem," Naga wrote on X this week.

Why it matters

Uber's experience illustrates a critical inflection point for enterprise AI adoption. As companies face mounting pressure to justify massive AI investments—Deutsche Bank Research warned last month that productivity gains remain years away—the focus is shifting from raw usage metrics to cost efficiency. The Magnificent Seven tech companies saw profit margins grow from 15% to 25% between Q1 2023 and Q1 2026, while the rest of the S&P 500 achieved only 10% margin growth, underscoring how few organizations outside Big Tech are extracting real value from AI spending.

The Jevons paradox risk

Even as Uber celebrates lower per-token costs, economists warn the company may face Jevons paradox—a 19th-century observation that efficiency improvements often increase total consumption rather than reduce it. Named for economist William Stanley Jevons, who noted coal usage surged despite more efficient steam engines, the phenomenon appears to be playing out in AI markets today.

According to the Silicon Data Token Expenditure Index, token prices have dropped over 90% since 2023, yet large language model spending has doubled since late 2024. A Bain and Co. brief found that while token costs halved from December 2024 to 2025, consumption grew 450% as companies upgraded their AI tools.

"As tokens get cheaper, companies don't spend less but instead run more AI agents, automate more workflows and generate more code," wrote Apollo Chief Economist Torsten Slok.

Quality over quantity

Uber's leadership acknowledges the innovation promised by AI hasn't fully materialized. As of May, President and COO Andrew Macdonald said on the Rapid Response podcast that the company struggled to draw a direct line between AI usage statistics and measurable increases in useful consumer features.

Naga suggested Uber is now prioritizing quality over quantity in token spending, though he didn't disclose whether absolute computing usage has increased or decreased. "The next phase will not be characterized by who spends the most tokens, but about how people use them as efficiently as possible," he concluded.

These details were first reported by The Information and Fortune.