Uber Burns Its Entire 2026 AI Budget by April — Claude Code Costs $500-$2,000 Per Engineer Monthly
Uber’s CTO went on record with The Information this week: the company’s entire 2026 AI tooling budget is gone before May. Four months into the fiscal year, the ride-hailing giant is rebuilding its spend plan from scratch.
“I’m back to the drawing board, because the budget I thought I would need is blown away already,” said Praveen Neppalli Naga.
The culprit is not a failed contract or an infrastructure overrun. It is a coding assistant.
How Claude Code Took Over Uber Engineering
Uber gave 5,000 engineers Claude Code access in December 2025, with internal leaderboards ranking teams by usage to accelerate adoption. It worked faster than anyone modelled.
By February 2026, Claude Code usage had nearly doubled from the December baseline. By March, 84% of Uber developers were classified as agentic coding users in internal telemetry. Cursor, the competing tool, plateaued over the same window. Claude Code did not.
The current state of Uber’s engineering org by the numbers:
| Metric | Figure |
|---|---|
| Engineers using AI tools monthly | 95% |
| Committed code from AI | Up to 70% |
| AI-authored backend updates | 11% |
| Claude Code adoption (Dec → Feb) | 32% → 63% |
| Monthly API cost per engineer | $500 – $2,000 |
| AI-generated PRs per week | ~1,800 |
The $20/Month Myth
Enterprise procurement teams priced Claude Code as a per-seat SaaS product. It is not one.
Claude Code bills on tokens consumed. An engineer doing occasional tab-completion is a different order of magnitude from one running parallel agentic workflows: full codebase reads, multi-file refactors, automated test generation, PR drafting. The latter consumes tokens at rates that bear no resemblance to a subscription model.
Uber’s R&D budget for 2026 sits at $3.4 billion — up 9% from the prior year. That is not a company without resources. It is a company that discovered the gap between per-seat SaaS assumptions and per-token agentic reality at full engineering org scale. The result: one fiscal quarter burned the annual AI allocation.
Industry guidance is now converging on 2–3x initial AI tooling estimates as the planning baseline for organizations above a few hundred engineers.
Who Comes Next
Naga said Uber is now testing OpenAI’s Codex alongside Claude Code as it expands its agentic stack. The company describes its direction as “agentic software engineering” — AI systems handling coding, testing, and deployment with other AI systems supervising the process, not just assisting human developers.
The budget problem Uber surfaced is not unique to Uber. It is the first large-scale confirmation of what token-based pricing looks like when adoption genuinely takes off at an engineering organization. Other companies are running the same experiment. Most have not announced the result yet.
The productivity numbers were real — 11% of live backend code changes written by AI is a meaningful shift. The question enterprise CTOs now face is whether the productivity gain justifies a budget line that does not behave like software and does not have a ceiling.