Uber's chief operating officer says the company can't draw a clean line between its rising AI bill and anything riders actually notice. Andrew Macdonald made the comments on the Rapid Response podcast over the weekend, weeks after Uber's engineers burned through their entire 2026 AI coding budget by April.
The link that isn't there
Macdonald didn't dress it up. After talking with senior engineering leaders, he said, more tokens weren't producing proportionally more useful features. "That link is not there yet, right?" Maybe more is shipping quietly, he allowed, but he couldn't tie the usage stats to what he called "25% more useful consumer features," and that hedge, captured in Fortune's report, is a rare bit of doubt from a number-two exec at a company that has been loudly pro-AI.
The cost side is concrete enough. Uber's monthly API charges reportedly ran $500 to $2,000 per engineer across roughly 5,000 developers, which is how you exhaust a yearly budget by spring. Uber has also said agents now write about one in ten lines of its code. That's an input number. It tells you the tools get used, not that what they produce is worth the meter.
So the next argument inside Uber, Macdonald said, is token cost versus headcount. If you can't point to the features, paying for the tokens starts to look like a gym membership nobody visits.
Microsoft hit the same wall
The timing is awkward. Microsoft is canceling thousands of Claude Code licenses across the group behind Windows, Office, Teams and Surface, steering engineers toward its own GitHub Copilot CLI by June 30. The stated reason is toolchain unification. The reason that lines up with the close of Microsoft's fiscal year is money.
Both stories run on the same mechanism. Flat seat licenses hide the meter; usage-based pricing switches it on, and a tool everyone loves quietly drains a budget nobody capped. Microsoft handed engineers Claude Code in December, watched adoption climb, and is now pulling it roughly half a year later.
Even the chipmaker is feeling it
Here's the part that should bother the optimists. Bryan Catanzaro, an Nvidia executive running applied deep learning, told Axios last month that for his team the cost of compute already runs far beyond what the company pays its people. Nvidia sells the compute. When the firm minting money on GPUs says the GPUs outcost the humans, the story about AI quietly replacing expensive labor gets harder to tell.
None of this stops the spending. It just moves the hard questions up the calendar. Microsoft's cutoff lands June 30, the same day its fiscal year ends, and that's the next real test of whether this is a budget trim or a change of heart.
