Microsoft and Uber pull back on Claude Code deployments as token-based billing exhausts budgets faster than forecast

Microsoft wound down most internal Claude Code licenses starting mid-May 2026, with access in its Experiences and Devices division ending June 30, after token-based billing made consumption unsustainable at deployment scale. Uber exhausted its full 2026 AI budget by April after deploying Claude Code to about 5,000 engineers four months earlier, with per-engineer costs reaching $500 to $2,000 monthly.

A 2025 Mavvrik survey found 85% of companies miss AI cost forecasts by more than 10%, and 84% report AI spending cutting gross margins by over six percentage points. Big Tech AI capex hit $650 billion in Q1 2026, and the number of FinOps teams managing AI spend doubled from 31% to 63% within a year.

Companies are now layering quotas, internal leaderboards, and cheaper model routing onto deployments that ran open in late 2025. Anthropic's $10.9 billion Q2 forecast would deliver its first profitable quarter, even as customers report cost overruns.