Uber Burned Its Entire AI Budget in Four Months. Microsoft Did Too.
In a nutshell
The AI era has a cost problem that no one in Silicon Valley wants to talk about clearly. This week, two companies did — and the implications reach every enterprise that has bet on AI productivity gains to justify AI spending.
The Uber Admission
Uber's Chief Technology Officer, Praveen Neppalli Naga, went viral in April for admitting Uber burned through its 2026 Claude Code budget in four months.
Uber COO Andrew Macdonald said this week that despite high usage of AI within the company, he could not draw a clear line between that usage and measurable improvements in consumer-facing products. "That link is not there yet," Macdonald said.
Four months. An entire annual budget. Gone. And the COO cannot demonstrate that the spending generated measurable consumer-facing improvement.
That is not a technology failure. It is a cost architecture failure. Uber, one of the world's most sophisticated technology companies, did not anticipate how quickly token-based AI billing would consume a fixed annual budget when deployed at enterprise scale with agentic coding tools.
Microsoft's Parallel Experience
We reported yesterday that Microsoft quietly cancelled its internal Claude Code pilot after token billing ate the entire annual AI budget, redirecting developers to Copilot.
Microsoft — the company that has invested over $100 billion in OpenAI and built its entire developer strategy around AI — ran an internal AI coding pilot and cancelled it because the consumption model was financially unsustainable.
Microsoft and Duolingo have also begun publicly questioning whether high AI adoption rates automatically translate to business outcomes.
Three major enterprises — Uber, Microsoft and Duolingo — publicly questioning AI ROI in the same week. This is not a coincidence. It is the first wave of a reckoning that the enterprise AI market has been building toward since agentic pricing replaced flat subscriptions.
The Pricing Architecture That Creates the Problem
Some AI firms are shifting pricing plans to capture increased AI usage. Anthropic changed its pricing model, moving away from a flat fee to a usage-based model, meaning autonomous agents are now charged per token of compute use. In March, OpenAI CEO Sam Altman articulated the industry's broader direction: "We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter."
The utility meter metaphor is clarifying. Electricity bills are predictable because consumption patterns are stable and metered in advance. AI agent token consumption is neither — it scales unpredictably with task complexity, conversation length and the number of autonomous steps an agent takes to complete a task.
A human software developer costs a predictable salary. A Claude Code agent costs an unpredictable number of tokens per task — and complex debugging or refactoring tasks can consume tokens at rates that make the human alternative significantly cheaper.
A separate Gartner study forecasts AI agent software spending will reach nearly $207 billion in 2026, up more than 139% from the $86.4 billion spent in 2025.
$207 billion in AI agent software spending — growing at 139%. Against a backdrop of enterprises discovering they cannot predict or control those costs. That combination is the definition of a market correction waiting to happen.
What This Means for GAFAM
The enterprise AI cost crisis is the story that GAFAM earnings calls have not yet been required to address directly. Google, Microsoft, Amazon and Meta report AI revenue growth — driven by cloud AI services that enterprises are consuming at unprecedented rates. They do not report on whether the enterprises consuming those services are generating measurable returns.
The Uber and Microsoft admissions suggest the answer, for at least some enterprise AI deployments, is not yet. When that answer becomes widespread enough to affect AI budget renewal decisions — when CFOs stop approving AI spending that cannot demonstrate ROI — the revenue growth that GAFAM's AI businesses are reporting will face structural pressure.
That reckoning is not this quarter. It may not be next year. But the Uber budget story and the Microsoft cancellation are the early signals. gafam.ai noted them.
The European Perspective
European enterprises considering large-scale AI deployments should treat the Uber and Microsoft admissions as the most valuable data points of the week — more valuable than any vendor presentation or analyst forecast. The lesson is specific and actionable: token-based AI billing at agentic scale generates costs that fixed annual budgets cannot accommodate without explicit consumption controls. European CIOs implementing AI coding tools, AI agents or any consumption-priced AI service should require real-time token spend dashboards, monthly budget alerts and automatic consumption limits as procurement conditions — not optional features. The enterprises that build these controls before deployment will avoid the Uber experience. The enterprises that discover the need for them after four months will have already learned the same lesson at greater cost. gafam.ai will be watching.
We are not first. We are right.