Tech Companies Scaling Back AI Spending After Budget Blowouts
Microsoft has canceled most of its direct Claude Code licenses and is redirecting engineers toward GitHub Copilot CLI instead. The move comes six months after the company opened access to the Anthropic tool and encouraged thousands of developers, project managers, and designers to experiment with it. The rapid adoption forced Microsoft to reverse course.
Uber faced a similar problem. The company burned through its entire 2026 AI coding tools budget in just four months, according to reporting from The Information. Uber had actively encouraged adoption by ranking teams on an internal leaderboard based on AI tool usage.
The pattern reveals a core economic problem: companies can't sustain the cost of widespread AI adoption at current pricing, despite aggressive internal campaigns to maximize tool use.
The Token Consumption Trap
Meta and Amazon are among firms still pushing employees to use more AI. Meta created an internal leaderboard called "Claudeonomics" to track which workers use the most AI. Amazon is encouraging employees to "tokenmaxx"-consume as many AI tokens as possible.
But this strategy creates a paradox. While the cost per token is falling, total consumption is rising faster. Goldman Sachs forecasts that agentic AI could drive a 24-fold increase in token consumption by 2030, reaching 120 quadrillion tokens per month.
Gartner research shows that by 2030, inference on sophisticated AI models will cost 90% less than in 2025. Yet that price decline won't translate to cheaper AI for businesses. Agentic models consume far more tokens per task than standard models, and AI providers won't pass all savings to customers.
"Chief Product Officers should not confuse the deflation of commodity tokens with the democratization of frontier reasoning," said Will Sommer, a senior director analyst at Gartner.
The Agent Problem
Nvidia CEO Jensen Huang has said he expects 100 AI agents to work alongside every employee at his company eventually. Other executives are making similar bets on an agentic future where digital workers operate across the enterprise.
That vision could carry a much steeper price tag than leaders expect. If token consumption rises faster than unit costs fall, the math breaks down quickly.
Bryan Catanzaro, vice president of applied deep learning at Nvidia, put it plainly: "For my team, the cost of compute is far beyond the costs of the employees."
The cancellations at Microsoft and Uber suggest that companies are starting to confront this reality. Early forecasts about AI replacing human labor may have underestimated the actual economics of adoption at scale.
Learn more about AI productivity strategies and AI coding tools to understand how organizations can use these technologies more effectively.
Your membership also unlocks:
