Uber CEO Dara Khosrowshahi says AI spend “blew” the company’s annual budget in one quarter

Uber Technologies CEO Dara Khosrowshahi said the company “blew through” its planned artificial intelligence budget for the year in a single quarter, and that the overspend has pushed Uber to slow hiring. Speaking in an interview with investor Patrick O’Shaughnessy, Khosrowshahi said AI spending was also the reason Uber planned to adjust its “hiring goals,” adding that the company would “meter headcount increases.”

Khosrowshahi attributed the productivity shift behind the higher AI usage to engineers becoming “much more efficient,” and said productivity gains were a factor as Uber looked at how to manage the tradeoff between output and cost. The comments also positioned AI as a lever for internal speed, even as budget overruns made it harder for leadership to treat AI as a steady, predictable expense.

The remarks come as Uber is trying to bring structure to how it uses AI in day-to-day engineering work. TechCrunch reported that Bloomberg said Uber instituted internal caps on employee spending for agentic coding tools, including Anthropic’s Claude Code and Cursor. Under the reported policy, employees face a monthly $1,500 cap per employee and per agentic coding tool, and usage is tracked through an internal dashboard that each employee can access.

TechCrunch also reported that the caps can be exceeded in certain cases with permission, suggesting Uber is aiming to control the default cost trajectory without fully eliminating flexibility for teams with urgent needs. The “agentic coding” label refers to AI systems that do more than answer questions, by generating code changes, running development tasks, and iterating toward software work products.

Uber’s cost controls have been framed as a response to an earlier pattern of faster-than-anticipated AI adoption. TechCrunch said Uber’s CTO had previously indicated the company exhausted its annual AI budget within months after encouraging staff to use AI “as much as possible,” and even ranking internal usage competitively through leader boards.

In parallel, Uber’s technology leadership has pointed to tangible engineering activity coming from AI. Benzinga, citing the discussion, said Uber has reported that 11% of its live backend code is now written by AI agents, and that Uber’s research and development spending rose to $3.4 billion in 2025, with expectations that it will keep rising.

Even with evidence that AI is changing how software gets built, Uber executives have also acknowledged the difficulty of proving a tight link between individual productivity metrics and customer-facing results. TheStreet reported that Uber’s COO, Andrew Macdonald, said it is “very hard to draw a line” between stats and the scale of improvements in useful consumer features, highlighting a persistent measurement gap that many companies face as AI becomes embedded in workflows.

Uber has not disclosed in the reporting publicly available here the exact size of the AI budget it originally set for the year, the specific cost centers affected, or how much hiring that “headcount metering” ultimately changed in practice. It also remains unclear how the company is accounting for potential offsetting benefits, such as faster release cycles or reduced defect rates, versus the incremental token, compute, and tool costs that contributed to the early budget exhaustion. Investors will likely look for further clarity in future earnings materials on AI-related expenses and whether the company can sustain productivity gains without recurring budget shocks.