Microsoft and Uber Cancel Claude Licenses Over Surging Costs

AI Token Trap: Microsoft and Uber Cancel Claude Licenses Over Surging Costs

The Bill for “Vibe Coding” Arrives: Microsoft and Uber Burn Through AI Budgets

The reality of the artificial intelligence boom is hitting enterprise finance teams hard. While agentic AI tools have been widely praised for accelerating software development, reports indicate they are becoming exceptionally expensive to run at scale.

According to reports from The Verge and tech insiders, Microsoft is winding down internal access to Anthropic’s Claude Code, transitioning developers back toward its own in-house alternatives ahead of the new financial year. The move highlights a growing, industry-wide challenge: token-based billing models are systematically blowing up corporate software budgets.

📉 Microsoft’s Experiences & Devices Division Pulls Back

Microsoft originally launched an internal pilot for Anthropic’s Claude Code in December 2025, granting thousands of its developers access to the specialized command-line assistant.

However, the pilot became a victim of its own success:

The Budget Burn: Under Anthropic’s usage-based, token-metered pricing structure, the division managed to burn through its entire annual AI budget in just a few months.

The June 30 Deadline: To rein in operating expenses before its new fiscal year kicks off in July, Microsoft has set a firm cancellation deadline of June 30, 2026, to remove the licenses.

Affected Teams: The rollback heavily impacts the Experiences & Devices division—the massive umbrella engineering unit responsible for Windows, Microsoft 365, Outlook, Teams, and Surface hardware.

Internal developers are now being aggressively redirected toward the GitHub Copilot CLI. Because Microsoft owns GitHub outright, steering engineers toward Copilot allows the company to absorb the underlying computing costs internally rather than writing massive checks to a direct competitor.

🏎️ Not An Isolated Incident: Uber’s Four-Month Burnout

Microsoft is far from the only tech giant hitting a structural wall with metered AI tooling. Internal memos reveal that ride-hailing giant Uber recently ran into an identical budgetary crisis.

Uber’s CTO reportedly issued a warning to staff after the company exhausted its entire 2026 artificial intelligence budget in just four months. The budget collapse occurred after Claude Code spread organically to roughly 5,000 internal Uber engineers, racking up massive token fees faster than the company’s financial forecasting models could anticipate.

🔄 The Enterprise Cost Trap: Flat Seats vs. Token Meters

The financial friction stems from a massive disconnect between traditional enterprise software procurement and the modern AI API economy.
[ Legacy Enterprise Model ] ──► Flat Monthly Fee per Seat ──► 100% Predictable Spend
[ Agentic AI Model ]        ──► Usage-Based Token Fees    ──► Exponential Budget Risk
In standard software development, autocomplete tools or fixed SaaS extensions charge a predictable monthly flat fee per user. In contrast, agentic coding tools like Claude Code act as active agents. If a developer triggers an agent to debug a codebase, the tool may autonomously read thousands of lines of context, execute a command, read the stack trace, and rewrite massive files—repeating this loop thousands of times a day.

Because corporations pay for every single input, output, and cached token passing through the API, a highly active developer can quietly rack up thousands of dollars in usage fees in a single month. Most corporate procurement and finance teams simply don’t have the FinOps guardrails or forecasting frameworks built to monitor or cap this type of real-time consumption.

🤝 The Irony: Hardware Talks Continue Behind the Scenes

What makes Microsoft’s internal software cutback particularly fascinating is that the overarching corporate relationship between Microsoft and Anthropic remains tightly intertwined.

While Microsoft is dropping Claude Code internally to protect its balance sheet, it is simultaneously in deep talks to sell its own custom silicon to Anthropic. Microsoft is pitching its second-generation Maia 200 AI chip to Anthropic, with CEO Satya Nadella publicly boasting that the hardware offers “over 30 percent improved tokens per dollar” compared to existing silicon.