Microsoft, Uber Hit by Rising Agentic Coding Costs

Microsoft canceled most internal licenses for Anthropic’s Claude Code after token-based billing pushed agentic coding costs above company projections. The company began winding down broad access in mid-May 2026, and the Experiences and Devices division will end access on June 30.

Engineers in multiple Microsoft groups had rapidly adopted the agentic coding tool. The company reported large internal productivity gains from AI tools, but teams running the tool at scale generated usage that outpaced planned spending and made token-based billing unsustainable.

Uber reported it exhausted its full 2026 AI budget by April after deploying Claude Code to about 5,000 engineers roughly four months earlier. Internal figures put average monthly per-engineer costs between $500 and $2,000. Company estimates indicate about 70% of committed code now originates from AI tools, driving vendor bills higher than expected.

A 2025 corporate survey from Mavvrik found 85% of companies miss AI cost forecasts by more than 10% and 84% reported AI spending reduced gross margins by more than six percentage points. Big Tech capital expenditure on AI reached about $650 billion in the first quarter of 2026. The share of companies with FinOps teams managing AI budgets rose from 31% to 63% within a year.

Anthropic, the maker of Claude Code, projected $10.9 billion in revenue for the second quarter, a level the company said would move it toward profitability. At customer sites, finance and engineering leaders are adding controls to deployments that were largely open in late 2025. Common measures include hard quotas on model use, routing requests to lower-cost models when possible, and internal leaderboards or alerts that flag high spend.

Uber’s chief technology officer, Praveen Neppalli Naga, described the budget strain directly: “I’m back to the drawing board, because the budget I thought I would need is blown away already.” Other industry participants have described the situation as a growing cost challenge.

Companies will report next-quarter results that include AI spending and may reflect the early effects of tighter governance, model routing and cost controls. Observers say similar budget pressures could appear in other sectors building large-scale AI infrastructure, including crypto-related projects that replicate high-consumption architectures.