Microsoft has reportedly begun canceling most of its Claude Code licenses, reversing course just six months after encouraging thousands of engineers, project managers, and designers to adopt the Anthropic-powered coding tool across the company. According to The Verge, Microsoft is now pushing employees toward GitHub Copilot CLI instead—a move that signals the tech giant's internal AI experiment has hit a financial wall. The cancellation won't affect Microsoft's broader Foundry deal with Anthropic, which includes up to $5 billion in investment and access to Claude models for Azure customers, nor Anthropic's $30 billion commitment to purchase Azure compute capacity.

Uber Burns Through 2026 Budget in Four Months

Microsoft isn't alone in hitting the brakes on AI adoption. Uber's CTO Praveepalli Naga revealed to The Information that the company had already exhausted its entire 2026 AI coding tools budget by April—less than a third of the way through the fiscal year. That spending spree came after Uber actively incentivized AI usage through internal leaderboards ranking teams by tool adoption rates, essentially gamifying the consumption of expensive compute resources. The pattern raises serious questions about whether corporate AI rollouts have been properly accounting for exponential cost growth as usage scales across organizations.

The Token Consumption Paradox

The economics driving these pullbacks reveal an uncomfortable truth about enterprise AI deployment. Goldman Sachs recently forecasted that agentic AI could drive a 24-fold increase in token consumption by 2030, reaching a staggering 120 quadrillion tokens per month as businesses deploy autonomous agents across operations. While individual token prices continue declining—research firm Gartner predicts inference costs on one-trillion-parameter models will drop roughly 90% by 2030—this deflation won't translate to cheaper enterprise AI bills. Agentic models require far more tokens per task than standard language models, and increased consumption can easily outpace falling unit costs.

Why Cheaper Tokens Don't Mean Cheaper Bills

Gartner senior director analyst Will Sommer warned that product leaders shouldn't conflate commodity token price drops with the democratization of frontier reasoning capabilities. The research firm pointed to three compounding factors: agentic workflows consume dramatically more tokens than traditional inference, aggregate usage growth can outrun per-unit cost reductions, and AI providers face structural incentives to maintain revenue margins rather than fully passing through infrastructure savings. "Chief Product Officers should not confuse the deflation of commodity tokens with the democratization of frontier reasoning," Sommer stated in a press release.

Nvidia's Compute Cost Reality

The financial calculus becomes even starker at the cutting edge of AI development. Bryan Catanzaro, vice president's applied deep learning at Nvidia, told Axios that compute costs for his team "are far beyond the costs of the employees." That admission from a company selling the hardware powering this transformation carries weight—training and running frontier models has become so expensive that human salaries are increasingly the smaller line item. Meanwhile, CEO Jensen Huang has publicly championed deploying 100 AI agents per Nvidia employee, a vision that's looking increasingly expensive to realize at scale.

Meta's 'Claudeonomics' Leaderboard Culture

The internal dynamics driving wasteful AI spending have become institutionalized at some firms. At Meta, an employee created an internal leaderboard called "Claudeonomics"—named after Anthropic's flagship model—to track which workers are consuming the most AI resources. Amazon has reportedly pushed employees toward "toxenmaxx," a strategy of maximizing token usage across operations. These cultural forces create perverse incentives where developers feel pressured to use AI tools even when human labor might be more cost-effective, particularly for straightforward tasks that don't require frontier model capabilities.

Key Takeaways

  • Microsoft is canceling Claude Code licenses after just six months, pivoting back to GitHub Copilot CLI amid runaway costs
  • Uber burned through its entire 2026 AI budget in the first four months of the year
  • Agentic AI could drive 24x growth in token consumption by 2030, reaching 120 quadrillion tokens monthly per Goldman Sachs
  • Gartner predicts inference costs will drop 90%, but enterprise bills may still rise due to increased consumption

The Bottom Line

The AI industry has been selling a story where scale equals savings, but the math doesn't work when usage grows faster than unit costs decline. Microsoft and Uber's course corrections aren't anomalies—they're early warning signs that the agentic AI future executives are promising shareholders will come with a price tag nobody wants to put on the slide deck.