Uber is rethinking its AI spending. The company burned through its entire 2026 AI budget by mid-April. That’s four months into the year. Now, Uber’s leadership openly questions whether the money buys meaningful results.

Andrew Macdonald, Uber’s COO, spoke plainly. The company can’t link soaring AI token use to better features for users. Adoption stats look impressive. But the actual improvements? Not so much.

Uber rolled out Anthropic’s Claude Code to 5,000 engineers last December. Usage surged from 32% to 84% in just a month. By March, 95% of engineers used AI tools monthly. Nearly 70% of code commits involved AI assistance.

Yet the cost per engineer ballooned from $500 up to $2,000 monthly. The budget blew out early. CTO Praveen Neppalli Naga confirmed the annual AI token budget ran dry by mid-April. Uber’s R&D spend hit $3.4 billion in 2025, a 9% increase year-over-year. AI costs are now a big chunk of that.

Macdonald admitted it’s tough to draw a direct line between AI usage and “25% more useful consumer features.” The company is asking a simple question: Is AI spending paying off? So far, the answer is fuzzy.

This isn’t just Uber’s problem. Other tech firms face similar dilemmas. Some have pushed employees to maximize AI use, an approach dubbed “tokenmaxxing.” The idea was more AI equals more productivity. Reality is different. More tokens don’t guarantee better output.

Duolingo, for example, rolled back plans to factor AI use into employee reviews after internal resistance. Their CEO said AI didn’t always fit the workflow. This signals a broader industry reckoning.

AI tools rely on massive cloud infrastructure that costs real money. That cost often gets buried under buzz about innovation. But CFOs don’t pay in tokens—they pay in dollars. When bills rise faster than measurable gains, scrutiny follows.

Uber’s CEO Dara Khosrowshahi acknowledged the tradeoff. The company is slowing hiring partly due to AI investments. The promise of “employees with superpowers” collides with tight budgets. Hiring fewer humans to pay for AI raises tough questions about long-term strategy.

Uber’s experience highlights a growing tension in enterprise AI. The technology clearly works. It generates code, automates tasks, and speeds some processes. But more AI output can mean more bugs, more integration work, and more costly maintenance.

Spending money faster isn’t progress. The true test is whether AI enables better products, faster innovation, or cost savings that justify the expense. Uber’s blunt admission cracks open a conversation many companies avoid. If the AI bill is real, the return must be real too.

Expect more companies to demand clearer metrics. Usage dashboards won’t cut it. Leaders want data that ties AI spending directly to customer benefit, revenue growth, or efficiency gains. Otherwise, tokenmaxxing becomes just a fancy excuse for big bills.

For now, Uber remains committed to AI as a frontier. Autonomous vehicles, logistics, and marketplace dynamics depend on smart software. But the era of blind AI spending is ending. Companies will tighten budgets, impose usage caps, and demand accountability.

Uber’s candid stance is a warning shot. AI’s buzz won’t pay the bills. The next phase of AI in business will be about proof, not hype.