Microsoft and Uber are pulling back from AI coding tools after costs ran far ahead of budgets, a pattern that industry surveys suggest is becoming the norm. Microsoft began winding down most internal licenses for Anthropic's Claude Code in mid-May, with access in its Experiences and Devices division set to end June 30. Uber burned through its entire 2026 AI budget by April, just four months into the year, according to Forbes.
Why the costs spiraled
The root cause is token-based billing, which charges per usage rather than a flat fee. Uber deployed Claude Code to about 5,000 engineers, with per-engineer costs ranging from $500 to $2,000 monthly, Forbes reported. That added up fast, especially given that roughly 70% of committed code now comes from AI tools, indicating heavy reliance across major engineering teams. For Microsoft, the costs in its Experiences and Devices division became unsustainable, prompting the license wind-down.
Anthropic's Q2 2026 forecast of $10.9 billion in revenue would deliver its first profitable quarter, benefiting from increased Claude usage despite customer complaints about pricing. The company's own growth depends on the very billing model that's now straining its biggest customers.
A wider industry problem
A 2025 survey by Mavvrik found that 85% of companies miss AI cost forecasts by more than 10%, and 84% report AI spending cutting gross margins by over six percentage points. The numbers underscore how quickly AI tooling can eat into budgets. Across Big Tech, AI capital expenditure hit $650 billion in Q1 2026 alone.
The financial strain is pushing organizations to rethink how they manage AI spend. The number of FinOps teams overseeing AI costs doubled from 31% to 63% within a year, as companies scramble to get a handle on the bills.
Companies retrofit controls
After letting AI deployments run open in late 2025, companies are now retrofitting financial controls. Quotas, internal leaderboards, and cheaper model routing are becoming standard tools to cap spending. Microsoft's own AI workplace report claimed 80% productivity gains, but that didn't prevent the cost overruns from Claude Code. The productivity promise, it turns out, doesn't automatically make the price tag manageable.
For Uber, the budget exhaustion means a hard stop on new AI coding tool spending until the next fiscal year. Engineers who relied on Claude Code for the majority of their code contributions now face a sudden shift back to more traditional methods or to less expensive alternatives.
The question now is whether these new controls will be enough. With Anthropic expecting its first profitable quarter, the pricing pressure isn't going away. Companies that rushed to adopt AI coding tools are now scrambling to pay the bill, and the next test will come when budgets renew in 2027.
