Big Tech's AI spending spree is hitting a wall — and the bills are enormous

The enterprise AI boom is running into an uncomfortable reality: the tools are expensive, the billing is unpredictable, and the productivity gains are harder to prove than the pitch decks suggested. Companies like Microsoft and Uber have already started pulling back, and the numbers behind their decisions are striking.

The costs that nobody budgeted for

Uber rolled out Claude Code — Anthropic's AI coding assistant — to its engineering teams starting in late 2025. Adoption shot up from 32% to 84% of engineers by March 2026. The problem: per-engineer monthly costs ran between $500 and $2,000, and the company burned through its entire 2026 AI budget of $3.4 billion in just four months, per Fortune (May 26, 2026). Uber's COO Andrew Macdonald publicly acknowledged the spiral — and admitted there's no clear evidence that higher token consumption translates to proportionally better products for users.

Microsoft went the other direction. After a pilot that saw developers prefer Claude Code over its own GitHub Copilot, the company canceled most of its internal Claude Code licenses with a June 30, 2026 deadline — the same day its fiscal year ends. The timing suggests a deliberate budget reset rather than a technical complaint.

The most extreme case is anonymous. An AI consultant told Axios (May 28, 2026) that one client racked up roughly $500 million in Claude Code charges in a single month — because nobody had set a usage cap for employees.

The structural problem with token billing

All three major coding AI providers — Anthropic, OpenAI, and GitHub — charge by token consumption rather than a flat fee. That model works fine at small scale. At enterprise scale with agentic workflows (where AI runs multi-step tasks autonomously), token usage can multiply 5–30 times faster than anyone anticipated. The cost per token has actually fallen around 98% since 2024, but volume has exploded far faster than prices have dropped.

The result is budget volatility that finance teams have no established playbook for. Four friction points keep surfacing across corporate AI rollouts, according to Axios reporting: use-case mismatch, token cost overruns, slow human adoption despite the spend, and leadership overcorrection once the bills arrive.

What comes next

A pullback doesn't mean AI tools disappear from enterprise software stacks. It means buyers are demanding fixed-price options, monthly spending caps, and — most critically — proof that the output is worth the cost. The question Uber's COO asked out loud is now being asked in boardrooms across the US: does 70% of code coming from AI actually mean better software for customers, or just more code?