AI was supposed to cut costs. Microsoft and Uber are finding it is more expensive than paying human employees
Microsoft is cancelling most of its direct Claude Code licenses six months after rollout, shifting to GitHub Copilot CLI due to unexpected cost spikes from employee adoption. Uber exhausted its 2026 AI coding tools budget in four months despite aggressive internal incentives, highlighting a broader trend where AI token-based pricing inflates expenses as usage grows, outweighing future cost reductions for enterprises.
Microsoft has reversed course on its AI coding tool strategy, cancelling most direct licenses for Anthropic’s Claude Code and redirecting employees to use GitHub Copilot CLI instead. The decision comes six months after the company provided broad access to Claude Code across its engineering, project management, and design teams, only to find that widespread adoption led to higher-than-anticipated costs. Uber’s experience mirrors Microsoft’s challenges. The ride-hailing company’s chief technology officer, Praveen Neppalli Naga, revealed in April that Uber had burned through its entire 2026 AI coding tools budget within four months. Despite internal leaderboards encouraging teams to maximize AI tool usage, the rapid spending underscores a systemic issue: AI compute costs rise with token consumption, regardless of efficiency gains. The problem stems from how AI models are priced. Large language models charge per token, meaning both increased efficiency and higher usage drive up total expenses. Companies like Amazon have even promoted ‘tokenmaxxing,’ pushing employees to maximize token consumption, while Meta developed an internal tool called ‘Claudeonomics’ to track AI usage intensity. Goldman Sachs predicts agentic AI—autonomous systems handling multi-step tasks—could surge token consumption to 120 quadrillion tokens per month by 2030, a 24-fold increase. Though Gartner forecasts a 90% drop in inference costs for large models by then, falling token prices won’t offset rising demand, as agentic models require far more tokens per task. AI providers are unlikely to pass cost savings fully to businesses. Nvidia’s vice president of applied deep learning, Bryan Catanzaro, acknowledged the cost dilemma in an interview with Axios. While compute costs may decline, the financial burden on enterprises could persist due to escalating token usage. The reversal at Microsoft and Uber signals a broader reckoning: AI adoption may not deliver the cost-cutting promises initially expected.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.