Microsoft Reduces Access to Claude Code, Uber Burns Through 2026 AI Coding Budget in Four Months
Microsoft has reportedly begun to cancel most direct licenses for Claude Code, redirecting engineers to GitHub Copilot CLI to control high token costs, after the tool rapidly gained popularity within the company.
Uber's CTO stated that the company has exhausted its entire 2026 AI coding tool budget within four months, with adoption rates soaring among approximately 5,000 engineers, leading to monthly costs per engineer ranging from $500 to $2,000.
These two incidents expose the cost pressures of AI coding agents under large-scale enterprise use; despite the tools being well-received, the token billing model has exceeded expectations.
Source: Public Information
ABAB AI Insight
Microsoft previously encouraged employees to experiment with Claude Code; this rapid reduction in access continues its strict cost control over AI tools while relying on its own GitHub Copilot ecosystem to reduce external dependencies.
In terms of capital strategy, companies are shifting AI budgets from external high-performance models to self-controlled infrastructure and hybrid deployments. Uber's internal leaderboard incentivized adoption but led to overspending, motivated by a desire for a more controllable ROI and to avoid runaway token consumption.
Similar to early cloud computing adoptions where companies faced unexpected bills, and the transition of SaaS tools from free to high subscription costs, AI coding tools are currently in a phase of transitioning from pilot enthusiasm to cost optimization.
Essentially, this reflects capital concentration: high usage of AI agents drives resources towards a few efficient models and self-built computing power. The mechanism lies in the token billing and exponential adoption forming compounding costs, forcing large enterprises to shift from external procurement to vertically integrated infrastructure to gain long-term pricing power.
ABAB News · Cognitive Law
The better the tool, the more explosive the bill.
Adoption speed outpaces budget models, a common trap in the AI era.
The true winners are not those who use the most, but those who can control costs.