Uber's CTO said it plainly this spring. "I'm back to the drawing board, because the budget I thought I would need is blown away already."
The company had rolled out Claude Code to roughly 5,000 engineers in December 2025. By April 2026 — four months later — the entire annual AI budget was gone. Not because anything went wrong. Because everything went right. Engineers loved it. Adoption jumped from 32% to 84% of the engineering org in months. About 70% of committed code now comes from AI. And nobody had modeled what that would cost, because nobody had ever modeled it before.
A separate enterprise spent $500 million on AI services in a single month. The reason: nobody had switched on a usage cap for its Claude licenses. Unlimited access looked like flexibility right up until the invoice arrived.
This is the AI story nobody was talking about six months ago. The conversation has shifted from "how do I get my team to use AI" to "how do I make sure I'm not Uber."
The Numbers Behind This Week's Story
- 78% of IT leaders have encountered unexpected charges from consumption-based AI pricing. — Zylo 2026 SaaS Management Index
- 79% of enterprises experienced AI cost overruns in the past year.
- 40% of AI agent projects will be cancelled by 2027 due to cost overruns alone — not technical failure. — Gartner 2026 AI Hype Cycle
- Agentic AI uses 5–30x more tokens per task than a standard chatbot. — Gartner March 2026
Why This Happened
The mechanism that blew Uber's budget is the same one operating at every scale. AI tools have moved from flat-fee subscriptions to consumption pricing — every prompt, every agent action, every automated task now generates a cost that compounds. Sam Altman said it plainly in March: "We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter." That future arrived faster than most finance teams were ready for.
The problem wasn't the tools. It was the absence of four things that should have existed before any tool was deployed at scale: a defined scope, a cost model built before deployment, usage caps or spend alerts that fire before the invoice, and someone accountable for outcome — not just adoption rate.
Three Questions Your Firm Should Be Able to Answer Right Now
What is your firm's total monthly AI spend — across every tool, every account tier, every team member? Not the subscriptions you formally approved. All of it — including personal accounts used for work, free tiers that become paid when usage crosses a threshold, and any API costs from tools your team has connected together.
What specific outcome is each AI tool connected to — and what number tells you it's working? Not "it helps with productivity." A number. Client onboarding time. Follow-up response rate. Hours on document review. Revenue per person. Something that existed before the tool and should be different because of it.
Who is accountable for AI spend — and do they see it before the invoice, or after? Uber's answer to this was "nobody, apparently." Anthropic launched Claude Enterprise spend controls on July 2, 2026 specifically because enterprises kept hitting this wall.
Uber's budget crisis wasn't caused by bad tools or reckless engineers. It was caused by measuring adoption instead of outcomes — and having no system to see the cost until it was already spent. Gartner forecasts 40% of AI agent projects will be cancelled by 2027 due to cost overruns — not technical failure, not market fit. Just the absence of discipline. The firms that avoid that list aren't using less AI. They're using it against a defined scope, with a defined cost model, measured against a defined outcome. The meter is running. The question is whether you know what it's measuring.