Uber burned its entire 2026 AI budget by April after rolling out Claude Code in December 2025, with per-engineer API costs hitting $500-$2,000/month.
Key Takeaways
Claude Code launched at Uber in December 2025; usage doubled by February 2026 and consumed the full annual AI budget by April.
95% of Uber engineers use AI tools monthly; 70% of committed code originates from AI.
Cursor plateaued in adoption while Claude Code dominated engineering workflows.
Per-engineer monthly API costs ranged $500-$2,000; Uber CTO says the company is “back to the drawing board” on AI budgeting.
With R&D at $3.4B annually, the overage signals that enterprise AI coding tool budgets set before widespread adoption are structurally underbuilt.
Hacker News Comment Review
Commenters broadly questioned whether Uber measured productivity gains against the spend – the budget overrun data exists but ROI evidence was absent from the announcement, drawing skepticism about whether velocity improvements justified the cost.
Several engineers challenged the $500-$2,000/month figure as not intrinsically large compared to hiring costs, arguing the real problem is that the budget was set before adoption data existed, not that the tool is too expensive.
A recurring thread flagged that the 70% AI-originated code stat is partly a performance-review artifact: when AI tool usage is tied to evaluations, adoption numbers get gamed, making the metric unreliable as a productivity signal.
Notable Comments
@phillipcarter: Points out $500-$2,000/month per engineer is cheap versus hiring; frames this as a forecasting failure, not a cost crisis.
@woah: Notes API pricing vs. subscription pricing discrepancy for Claude – large context sessions can blow through hundreds of dollars without careful context hygiene, suggesting Uber may be paying API rates where subscriptions would be cheaper.
