Uber and Microsoft Cut Claude Licenses as AI Costs Soar 2026

Uber and Microsoft are cutting back on Claude licenses as AI costs skyrocket, signaling a major shift in enterprise AI adoption. The era of subsidized AI is ending, and companies are now facing the real price of integrating advanced AI tools into their workflows.

Uber's Claude Code Experiment: From Enthusiasm to Budget Crisis

Uber's CTO, Praveen Neppalli Naga, recently shared how the company deployed Claude Code to its 5,000 engineers. Adoption surged from 32% to 84% in just one month, with engineers using the tool extensively. However, this enthusiasm came at a steep price: the actual cost per programmer jumped from $500 to $2,000 per month, blowing through Uber's annual AI budget in just four months.

This case highlights the hidden costs of AI tools that charge by usage. While the initial productivity gains were impressive, the financial implications were unsustainable without careful management.

Microsoft's Strategic Pullback

Microsoft, despite having virtually unlimited computing resources through Azure, has also decided to rein in spending. The company is removing internal Claude Code licenses from its Experiences + Devices division. The reasons are twofold: first, to curb operational expenses before the end of its fiscal year; second, to push developers toward its own GitHub Copilot.

This move underscores a broader trend: companies are re-evaluating third-party AI tools in favor of in-house solutions that offer better cost control and integration.

GitHub Copilot's Shift to Usage-Based Pricing

GitHub, owned by Microsoft, is also preparing for the future. Starting June 1, 2026, all GitHub Copilot plans will abandon the flat-rate option and move to usage-based billing. The base subscription price remains, but the change reflects the need to align costs with actual usage.

This shift is likely to affect how developers use Copilot, as they become more mindful of their consumption to avoid unexpected charges.

Key Takeaways for Enterprises

• AI costs can escalate quickly when usage-based pricing is not properly managed.
• Vendor lock-in is a real concern as companies like Microsoft prioritize their own tools.
• Budget forecasting becomes challenging with variable AI expenses.
• Productivity gains must be weighed against total cost of ownership.
• Usage-based models require new cost-control strategies to prevent overspending.

Comparing AI Tool Costs: Claude vs. GitHub Copilot

The table illustrates the financial unpredictability of usage-based AI tools. While flat-rate pricing offers stability, it may not be sustainable for providers as usage grows. The background to this story is set out in DeepSeek Gains Ground Among US Companies as AI Costs Bite. A related thread runs through Anthropic Builds Custom AI Chips to Power Claude and Cut Costs. Earlier coverage in Lloyds Bank AI Strategy Cuts £2bn Costs, Boosts Growth shows how the argument developed.

What This Means for the AI Industry

The pullback from Uber and Microsoft is a wake-up call for the AI industry. Companies must now carefully evaluate the total cost of ownership for AI tools, not just the initial subscription fees. Vendors will need to offer more predictable pricing models or risk losing enterprise customers. A closer look at the same subject appears in Revolut Cuts AI Coding Costs 10x by Switching 2,000 Engineers to GLM 5.3. Anyone following the wider picture can start with Australia Gig Worker Minimum Wage: Uber Eats Cost Impact. The same theme surfaces in Bill Gates' Speeding Tickets: How a Porsche Almost Cost Him His License, which takes a different angle.

Moreover, the move toward in-house solutions like GitHub Copilot suggests that companies value integration and control over third-party offerings. This could lead to further consolidation in the AI tool market. Context on how this played out before sits in Steam Driving Sim Demo Costs Devs $1,000 Daily in AI Tokens. A companion piece, AI Advice Fail: Andrew Hastie's Copilot Reply to Assisted Dying, approaches the same subject from another angle.

FAQ