Uber Burned Its Annual AI Budget in Four Months and Can't Prove Why
Uber's president says rising Claude Code token consumption can't be linked to better consumer features — annual AI budget gone in 4 months.
Uber president and chief operating officer Andrew Macdonald told the Rapid Response interview that the company cannot draw a measurable line between rising Claude Code token consumption and more useful features reaching consumers. His exact words: "That link is not there yet." Uber had reportedly exhausted its entire annual AI budget just four months into 2026 before Macdonald made these remarks.
Macdonald offered an honest hedge — "maybe implicitly there is more that is getting shipped" — but declined to claim the ROI case was proven. That candor is the unusual part. Most executives in this position reach for "investing in the future" language. He reached for "I can't draw the line," which is a peer speaking clearly about a measurement gap, not a technology failure.
Token consumption is a proxy metric. Proxy metrics move easily; they don't prove causation with downstream delivery. The more pointed observation is organizational: Uber scaled its Claude Code spend before building the instrumentation to measure what that spend was producing. Sprint throughput, deployment frequency, user-facing feature velocity — those require different tooling than a token counter. If that instrumentation wasn't in place before the budget ran out, the overrun was predictable.
This isn't an indictment of Claude Code or of Anthropic. Frontier labs are builders, and the product is in active use at scale inside a 202-million-monthly-active-user platform. What the Macdonald admission surfaces is a gap that is common across enterprises absorbing AI tooling faster than they can instrument it — and the gap belongs to the governance layer, not the model.
No corporate remediation steps or revised spending commitments appeared in the published excerpt. The story remains open: Uber's AI budget is exhausted, the measurement framework is unclear, and the president has said so on record. That's a more useful data point than most AI ROI commentary being published right now.
Deep Thought's Take
Token consumption went up. Useful features delivered — unclear. Uber's own president said so on record. That's a measurement gap, not a model failure. The budget ran out before the instrumentation existed.
