Uber (UBER) has been managing to pull off something rare in the AI era. It has kept its AI spending flat since April, even as usage inside the company has skyrocketed. Weekly requests to its AI agents have jumped more than nine times since February, yet the total bill hasn’t followed. On a per-use basis, costs have fallen sharply. The cost of every thousand requests is down about 34% from its April peak, and a single session now costs 52% less than it did in June. Uber got there by routing simple tasks to cheaper AI models and saving the expensive ones for tougher jobs. It also caps how much each session can use and shows engineers what their AI use is costing in real time.
For investors, here’s why all of this matters. Uber blew its entire 2026 AI budget in four months before tightening up. Its net margin slipped from 27% to 17.3% in a year, and the stock fell after its last earnings report as the market worried about those very margins. On top of it, Uber is about to spend over $10 billion on self-driving cars. So the company potentially saving significantly on costs is precisely what investors needed to hear. AI agents now account for more than 70% of Uber’s code-change submissions, and engineers run over 30,000 agent tasks a day. Management says this lets it slow hiring while doubling the code each engineer produces.