Uber CEO, COO says AI spending not yet tied to measurable improvements

Ninety-five percent of Uber's engineers use AI tools every month. Seventy percent of committed code is now AI-generated. By almost any adoption metric, the rollout has been a success.

So why is the COO saying the costs are getting harder to justify?

Andrew Macdonald said impact of AI spending is hard to measure

Uber (UBER) COO Andrew Macdonald made a candid admission at a conference on May 25 about the company's AI spending, according to Business Insider.

Despite near-total adoption of AI coding tools across Uber's engineering workforce, he said he cannot draw a clear line between that usage and measurable improvements in consumer-facing products.

More AI:

• Micron sits at the center of a red-hot chip rally
• IBM CEO sends blunt message on AI and quantum computing
• Anthropic CEO makes shocking admission about AI

"That link is not there yet, right?" Macdonald said. He added: "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% more useful consumer features.'"

The comments are significant because they come from the number two executive at a company that has been publicly enthusiastic about AI.

They are also candid in a way that is unusual for C-suite communications around AI, where the dominant narrative has been one of unqualified optimism, Business Insider confirmed.

The tokenmaxxing problem Uber is now confronting

The backdrop to Macdonald's comments is a phenomenon that has been spreading quietly across enterprise AI deployments: tokenmaxxing.

The term describes a dynamic where employees maximize their AI token consumption, either because they believe it signals productivity or because the tools have become embedded in their daily workflows to a degree that drives up costs regardless of intent, according to AI Weekly.

Token pricing means every query, code generation request, debugging session, and documentation task consumes compute that the company pays for. When 95% of engineers are interacting with these tools daily, the bill accumulates, regardless of whether the output translates into faster product delivery.

The catalyst for Macdonald's public acknowledgment was Uber CTO Praveen Neppalli Naga's earlier disclosure. In an April interview with The Information, Naga revealed that Uber had burned through its entire 2026 budget for Claude Code and Cursor in just four months, AI Weekly confirmed.

That comment went viral inside and outside the company and triggered what Macdonald described as a "head-exploding moment" that forced leadership to examine the relationship between token consumption and actual business outcomes.

Why high AI adoption rates no longer tell the full story

The uncomfortable truth Macdonald surfaced is one that applies well beyond Uber. High AI adoption rates and high token usage have become the metrics companies reach for to demonstrate AI progress. But those metrics measure inputs, not outputs.

A company can have near-total engineer adoption and still be unable to prove that the AI investment is producing proportional gains in the things that matter: features shipped, bugs fixed, customer problems solved.

That measurement gap is now urgent enough for Uber's COO to raise publicly. It is also creating pressure on AI tooling vendors. If procurement teams at large enterprises start demanding outcome-linked justification rather than seat-count renewals, the commercial model for tools like Claude Code and Cursor faces scrutiny, according to AI Weekly.

Uber is not alone. Microsoft questioned the cost of Claude Code licenses before canceling them across its Experiences and Devices division. Duolingo has also been working through similar questions about AI tool ROI.

The pattern suggests that enterprise AI adoption is entering a second phase, one where the question is no longer whether to deploy AI but whether the deployment can be justified on financial terms.

The pressure is also showing up in hiring decisions. Earlier in May, CEO Dara Khosrowshahi said on an earnings call that Uber was slowing hiring to counter its investments in AI, Business Insider reported.

Together, the CEO's hiring restraint and the COO's ROI concerns tell a consistent story: Uber is spending heavily on AI, absorbing the cost through headcount discipline, and still cannot fully demonstrate that the output justifies the bill.

Key figures on Uber's AI spending and Macdonald's comments:

• AI adoption at Uber: 95% of engineers use AI tools monthly; 70% of committed code is AI-generated, Business Insider noted.
• Budget overrun: Uber CTO Praveen Neppalli Naga disclosed in April that Uber burned through its entire 2026 Claude Code and Cursor budget in four months, according to AI Weekly.
• COO's key quote: "That link is not there yet, right?" referring to the connection between token usage and consumer product improvements, Business Insider confirmed.
• The problem named: Macdonald used the term "tokenmaxxing" to describe the dynamic of rising AI token consumption without proportional output gains, AI Weekly reported.
• Broader pattern: Uber joins Microsoft and Duolingo in publicly questioning whether token volume translates to business outcomes, according to AI Weekly.

What this means for investors watching enterprise AI spending

The enterprise AI investment thesis has rested heavily on a productivity argument: AI tools make workers faster, output rises, and margins improve. Uber's COO is now publicly questioning whether that chain holds in practice, at least at the current stage of deployment.

That does not mean AI tools do not work. Macdonald was not calling for a rollback. His point was narrower and arguably more important. The company cannot yet measure whether the productivity gains are proportional to the cost, and that uncertainty makes the spending harder to defend to finance leadership.

For investors tracking AI infrastructure spending, the signal is specific. The companies selling AI tokens, models, and compute infrastructure are benefiting from the current adoption wave, regardless of whether enterprise buyers can demonstrate ROI.

But the longer that ROI remains unmeasurable, the more pressure will build on pricing models, contract renewals, and the sustainability of current spending levels. Uber's COO just made that pressure visible in a way that will be difficult for other enterprise AI buyers to ignore.

Related: Microsoft CEO sends shocking message to employees

The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.

This story was originally published May 26, 2026 at 10:23 AM.