Uber poured billions into artificial intelligence tools last year. Its research and development expenses hit $3.4 billion in 2025, a 9% jump from the year before. The Verge reported that figure alongside plans for even heavier outlays. Yet the payoff remains elusive. Andrew Macdonald, Uber’s president and chief operating officer, said as much in a candid interview released this weekend.
“That link is not there yet, right?” Macdonald told the Rapid Response podcast. “I think maybe implicitly there is 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 25 percent more useful consumer features.'” He added that the trade becomes harder to justify without a direct connection to features shipped to users. Short sentences. Clear frustration.
Macdonald’s remarks land at a pivotal moment. Uber’s chief technology officer, Praveen Neppalli Naga, revealed earlier that the company burned through its entire 2026 AI budget in just four months, largely on Anthropic’s Claude coding tools. The news sparked internal shock. Executives responded by slowing hiring to offset the rising costs. CEO Dara Khosrowshahi framed the move as strategic. He pointed to AI agents producing roughly 10% of code changes and urged employees to boost throughput by 20%, 30%, 50%, or even 100%.
“If every person at this company can increase their throughput by 20%, 30%, 50%, 100%, then I think metering headcount growth and leaning in on AI investment is going to be well worth it,” Khosrowshahi said during the company’s first-quarter earnings call, as covered by Yahoo Finance UK. Optimism from the top. Questions from the operating chief.
But Macdonald pressed on the gap between activity and outcome. Senior engineering leaders reported higher token consumption and more code commits via Claude. Twenty-five percent of commits in one quarter came that way. Still, no corresponding surge in projects rescued from the cutting room floor or meaningful new functionality delivered to riders and drivers. The numbers climb. The business impact stays fuzzy.
Executives across tech now confront the same uncomfortable math.
Macdonald didn’t stop at code. He addressed the hype around agentic AI reshaping commerce. A year earlier, Uber’s board worried that chatbots would disintermediate the entire ride-hailing and delivery model. “We’re working with pretty much all of the large model companies as they roll out, try to roll out commerce, and there hasn’t really been anything that’s taken off yet,” he said. “It doesn’t mean it won’t happen.” Measured. Cautious. A far cry from predictions that apps like Uber would become obsolete.
His comments echo broader concerns. Gizmodo highlighted how AI can feel free to individual users dreaming up use cases. “Somebody’s paying the bill,” Macdonald noted. That bill now runs high enough to force trade-offs in headcount. And the productivity data to support those choices hasn’t arrived.
Recent coverage sharpens the picture. A Business Insider report from two days ago detailed the “tokenmaxxing” phenomenon, where internal leaderboards encourage maximum AI usage. Uber isn’t alone. Klarna has quietly rehired humans after chatbot experiments fell short. Duolingo pulled back on AI-generated reviews. Patterns emerge. Spending accelerates. Measurable returns lag.
Stanford’s 2026 AI Index, released recently, shows U.S. private AI investment reached $285.9 billion in 2025. Massive sums. Yet economy-wide productivity statistics show no clear breakout tied to these outlays. A Bruegel working paper from last year already flagged the tension between exploding costs and slow productivity growth. Newer analyses from Deloitte and BCG, referenced in recent discussions, find that only a small fraction of organizations see quick payback. Most wait two to four years. Some see none at all.
So what explains the disconnect? Part of it lies in measurement. Code commits rise. Token counts explode. But useful features? Those require human judgment, integration, testing, and alignment with user needs. Engineers still review every AI-generated change. The “super engineers” Khosrowshahi envisions exist. Their output advantage, however, proves slippery to quantify at the company level.
And. There’s the infrastructure bill coming due. Hyperscalers commit hundreds of billions to data centers and chips. Investors once accepted the spend as table stakes for future dominance. Now they ask tougher questions. Chamath Palihapitiya argued on the All-In podcast earlier this year that AI agents must deliver at least twice a human’s productivity once oversight and energy costs enter the equation. Uber’s experience suggests many firms still fall short of that bar.
Macdonald acknowledged potential. “I think over the coming quarters and years, maybe that will become clearer,” he said. The underlying metrics trend astronomical, after all. Yet today the link stays missing. Boards want evidence. Finance teams watch budgets evaporate. Employees sense the shift in hiring priorities.
Uber built its early lead on data and algorithms for matching drivers and pricing rides. AI was supposed to supercharge that edge. Instead, it introduced new variables. Higher compute spend. Harder justification conversations. A reminder that adoption metrics differ from business outcomes.
Other companies watch closely. The $700 billion in planned AI infrastructure investment this year, noted in multiple recent reports, assumes eventual returns. If firms like Uber cannot close the loop between token spend and shipped value, that assumption faces pressure. Productivity gains have historically taken time to materialize after major technology shifts. This cycle may prove no different.
But time carries cost. Every quarter of unclear returns makes the next round of investment tougher to defend. Macdonald’s frank assessment doesn’t signal retreat. Uber continues to work with major model providers and expand AI use. It does signal a demand for better metrics, clearer causation, and honest accounting of what the technology actually moves forward.
The bill keeps arriving. Someone pays it. The question now is whether the features, the throughput, and the new projects arrive in proportion. For Uber and many peers, that answer remains not yet.
WebProNews is an iEntry Publication