AI cost crisis hits tech giants as employees burn $1.3M monthly tokens

Many tech companies, including Microsoft and Amazon, are pushing employees to use AI tools to boost productivity, but this strategy is backfiring due to soaring token costs. According to The Verge, Microsoft is urging staff to switch from Claude Code to its own Copilot CLI, ostensibly for internal tool preference, but sources say the real reason is the escalating cost of Claude Code tokens as usage surges. This isn't isolated—Fortune reports similar pullbacks across the industry. The core issue is that agentic AI (autonomous, multi-step tasks) consumes up to 1,000 times more tokens than simple LLM queries. Peter Steinberger, creator of OpenClaw, revealed his team spent over $1.3 million on token costs in a single month.

The phenomenon, dubbed 'tokenmaxxing,' sees employees using AI for unnecessary tasks to hit internal usage targets, inflating scores at Amazon, Microsoft, and Meta. Nvidia CEO Jensen Huang famously said engineers should use AI tokens worth at least half their annual salary to be productive. However, decreasing token costs (due to model training efficiencies) are paradoxically driving total spending up—a textbook Jevons Paradox. As AI becomes cheaper per token, companies are deploying it more aggressively, offsetting any cost savings. The result: AI usage can now be more expensive than hiring humans, especially with limited productivity gains. It remains unclear if firms will adjust policies, but the trend suggests that replacing labor with AI may not yield the expected savings if token costs outpace efficiency gains.