Uber says it has identified a more sustainable approach to AI deployment after exhausting its entire 2026 artificial intelligence budget within the first few months of the year.

In an interview with The Information earlier this year, Uber chief technology officer Praveen Neppalli Naga acknowledged that he went "back to the drawing board" on spending allocations after the company had encouraged employees to use AI tools — particularly Anthropic's Claude Code, an AI-powered coding assistant that helps engineers write, debug, and review software — as extensively as possible. Uber even created internal "leaderboards" to rank software engineers by their AI tool usage.

The push reflected a broader corporate trend known as "tokenmaxxing," in which companies incentivized workplace AI adoption, only for many to retreat when returns on investment failed to justify the accelerated spending. The term derives from tokens — the basic units of text that large language models process and that providers bill on — making raw token consumption a rough proxy for AI spend. Uber was no exception, but Naga said the company has since developed a more efficient deployment model.

"We're seeing some very interesting trends on AI costs," Naga wrote in an X post on Wednesday. "I think it's another signal that we're coming to the end of the so-called tokenmaxxing era."

Naga explained that Uber quadrupled the number of employees using frontier AI tools, which brought down the cost per token. The company achieved this through improved prompt caching, adjustments to default model settings, efficiency-focused evaluation of new models, and dashboards that enabled engineers to monitor their AI usage and costs on an hourly basis.

"You might expect costs to rise as adoption accelerates," Naga continued. "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."

Rising Stakes for AI Returns

Pressure is mounting on companies to demonstrate returns on substantial AI investments. In July, Jim Reid, global head of macro and thematic research at the Deutsche Bank Research Institute, cautioned that AI productivity gains remain years away.

According to Apollo, profit margins for the Magnificent Seven — Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, and Tesla — expanded from 15% to 25% between the first quarters of 2023 and 2026, while the rest of the S&P 500 saw only 10% margin growth over the same period — suggesting limited returns on AI investment outside the immediate technology sector.

As of May, Uber was still working to translate AI investment into measurable product improvements.

"That link is not there yet," Uber president and chief operating officer Andrew Macdonald said in an appearance on the Rapid Response podcast. "Maybe implicitly there's more that is getting shipped, but 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.'"

Jevons Paradox Looms Over AI Spending

Even as Uber develops strategies to reduce per-token costs, the company faces a risk that economists have long highlighted: the Jevons paradox, whereby total spending on a resource increases even as its unit cost falls.

Named after 19th-century economist William Stanley Jevons, who observed in 1865 that coal consumption surged despite the Watt steam engine making coal use more efficient, the concept is now manifesting in AI.

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.

"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," Apollo chief economist Torsten Slok wrote in a recent blog post.

A Bain & Co. brief published in June reinforced this finding. It reported that token costs halved from December 2024 to 2025, while tokens consumed grew by 450% over the same period as companies upgraded their AI tools.

Naga noted a shift in Uber's philosophy toward prioritizing quality over quantity in token spending, though he did not disclose whether the company is consuming more or less computing power than at the start of the year.

"This is the future of applied AI at enterprise scale," he concluded. "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."

This story was originally featured on Fortune.com.