Microsoft, one of the world’s largest technology companies and a major backer of AI infrastructure, recently canceled the majority of its internal Claude Code licenses due to cost overruns. The company is redirecting its engineering teams toward GitHub Copilot, its own more cost-effective coding assistant.

The $500 million wake-up call

One enterprise client reportedly racked up $500 million in costs on Anthropic’s Claude in a single month.

Uber’s experience tells a similar story, just at a different magnitude. The ride-hailing giant’s COO acknowledged that the company depleted its annual AI coding budget in roughly four months. AI spending, the executive noted, is “harder to justify” given the pace at which costs are accumulating relative to measurable returns.

When AI costs more than the humans it replaced

AI compute costs have surpassed human labor expenses in some organizations. Nvidia VP Bryan Catanzaro flagged this trend, which flips the entire economic thesis for AI adoption on its head.

Some organizations have framed layoffs around automation, cutting headcount based on the assumption that AI would pick up the slack at lower cost. When the AI bills come in higher than expected, those companies find themselves in a particularly awkward position: fewer humans, bigger technology bills, and no clear path to the efficiency gains that justified the cuts in the first place.

The consulting world has noticed the shift. Industry advisors describe a move away from what some have called “tokenmaxxing,” the practice of maximizing AI usage across an organization without much regard for cost or return. The new playbook is more disciplined: identify the specific use cases where AI genuinely delivers value, typically coding-related tasks, and cut back everywhere else.