By Uber burned through its entire 2026 AI budget in four months on Claude Code. What started in December 2025 as a controlled rollout became a consumption crisis by April 2026, with CTO Praveen Neppalli Naga publicly admitting he’s “back to the drawing board.” For a company spending $3.4 billion annually on R&D, this isn’t a rounding error. Uber is the first major enterprise to publicly disclose an AI coding budget disaster.
The Numbers: 95% Adoption, 70% AI Code
By April 2026, 95% of Uber engineers were using AI tools monthly, with 70% of committed code originating from AI—the highest publicly reported percentage at any major tech company. Monthly API costs per engineer ranged from $500 to $2,000, averaging $150-$250 but spiking higher for power users. Usage doubled between December and February as developers discovered Claude Code’s multi-step agentic capabilities.
By March, 84% of Uber’s developers were classified as “agentic coding users,” delegating entire workflows to AI rather than just accepting autocomplete suggestions. The annual budget allocation evaporated in four months. Naga told The Information: “I’m back to the drawing board because the budget I thought I would need is blown away already.”
The Consumption Pricing Trap
Claude Code uses consumption-based pricing: seat fee plus all usage billed at API token rates with no caps. In pilots with 50 engineers at $200/month each, that’s $10,000 monthly—manageable. Roll it out to 5,000 engineers at $500-$2,000 each, and you’re looking at $2.5 million to $10 million monthly. Nobody budgeted for that.
The cost structure compounds exponentially. AI agents resend full conversation history with every API call. A coding session starting at $0.50 can end at $5.00 just from context accumulation. Multi-step agentic workflows multiply this further.
Compare this to competitors. GitHub Copilot charges $10-$39 per user monthly—flat and predictable. Cursor uses usage caps per tier. Claude Code has no caps. The better your engineers use it, the more it costs. Because Claude Code offers superior multi-step reasoning versus competitors, engineers chose the best tool, not the cheapest. Internal leaderboards Uber built to gamify adoption accelerated consumption by design.
Why Engineers Won’t Give It Up
The productivity gains are real. Developers save an average of 3.6 hours weekly, with enterprise productivity increases of 10-30% and ROI ranging from 2.5x to 6x. The catch: those calculations often exclude actual token costs, counting only seat licenses.
Uber’s CTO faces an impossible dilemma. He can’t afford current usage. He can’t take the tools away without destroying productivity and morale. And he can’t budget for it when usage can double in three months. The tools work too well to abandon and cost too much to sustain at full throttle.
The Governance Gap
Only 43% of organizations have formal AI governance policies, and just 21% have mature agentic AI governance models. Uber deployed Claude Code company-wide without the spending controls DevOps teams apply to AWS. No usage caps per engineer. No monitoring for token spikes. No budgetary guardrails.
The AI governance market is projected to grow from $309 million in 2025 to $5.9 billion by 2035, and Uber’s crisis just accelerated that timeline. CTOs should audit AI tool spending now and implement controls before costs spiral. Set usage caps per engineer. Monitor token consumption like cloud costs. Evaluate flat-rate alternatives if budget predictability matters.
What Happens Next
Uber is the first to admit this publicly, but every enterprise using Claude Code at scale faces identical math. The Hacker News thread discussing this has 339 comments split between “productivity is worth it” and “pricing is unsustainable.” Both are true.
Anthropic has every incentive to maintain consumption pricing—more usage equals more revenue. But enterprise backlash is building. Expect market pressure to force capped enterprise plans with usage allowances included. Competition from flat-rate providers will intensify.
If a $3.4 billion R&D budget couldn’t control this, your company probably can’t either—unless you implement governance before, not after, the budget explodes.
