Major tech companies are hitting a financial wall with artificial intelligence. Microsoft told its engineers to stop using a popular AI-powered coding tool, Anthropic's Claude Code, because bills were getting out of hand. At the same time, Uber blew through its entire yearly AI budget by April. Companies love what AI can do, but they can no longer afford the price tag.
Microsoft Drops Claude, Migrates to GitHub Copilot
Microsoft rolled out Anthropic's Claude Code to internal teams in late 2025. Engineers in the Windows and Teams departments loved using it to write code faster. However, high usage created massive token bills that shocked executives. Microsoft is now canceling most Claude Code licenses by June 30, 2026, forcing workers onto its own product — GitHub Copilot. This shift shows that even the biggest software companies cannot handle uncontrolled AI token costs.
Uber Engineers Burn $500–$2,000 Monthly, Budget Gone in Four Months
Uber gave Claude Code to roughly 5,000 engineers in December 2025. By March 2026, more than 84% of those workers used automated agents daily, generating about 70% of all code committed. But heavy users racked up bills between $500 and $2,000 every single month. As a result, Uber exhausted its full 2026 AI budget in just four months. CTO Praveen Neppalli Naga admitted the company must rethink its financial assumptions.
The Hidden Trap: Unit Costs Drop, Total Spending Soars
The cost per million tokens has fallen hundreds of times over the past few years. Yet enterprise AI spending is skyrocketing because employees use tools far more often. Simple queries cost little, but new AI agents run background loops and are much more expensive. A 2025 tracker showed 78% of businesses overran their AI budgets by 47%. Finance teams are now stepping in to cap usage. Companies are looking at cheaper open-source models or smaller local models for basic tasks, saving premium models only for the hardest problems.
The next two years will be about cutting costs and finding efficiency. The era of unchecked AI adoption is over. Companies will build strict usage caps and better tracking tools. AI definitely helps engineers work faster, but it does not save money automatically. This crunch is a healthy reality check for the industry: powerful tools require careful management. The winners will be those who figure out how to run automation without breaking the bank.