Companies Are Pulling Back on AI Use as Token Costs Start to Bite
Corporate enthusiasm for generative AI is meeting a hard financial reality: AI usage can become extremely expensive, extremely fast.
In the early days of the current AI boom, many executives pushed employees to use AI tools as much as possible. The idea was simple: if workers leaned heavily on AI, companies could boost productivity, speed up development, and potentially reduce labor costs. Some business leaders even treated heavy AI usage as a sign of innovation, encouraging teams to “max out” their use of AI tokens.
But that attitude is now changing across parts of Big Tech and Corporate America.
AI tokens are the units used to measure inputs and outputs in many AI systems. A prompt, a document summary, a coding request, or a long back-and-forth conversation with a chatbot all consume tokens. The more complex the task, and the more often employees use these tools, the higher the token bill becomes.
For companies with thousands of employees, that can add up to a massive expense.
Not long ago, some executives framed heavy AI usage as essential for staying competitive. Nvidia CEO Jensen Huang famously suggested he would be concerned if highly paid engineers were not spending heavily on AI tools to improve their work. He compared avoiding AI-assisted design tools to a chip designer refusing modern software and working with pencil and paper instead.
Now, however, businesses are discovering that unrestricted AI use is not always cheaper than human labor. In some cases, it may be far more expensive than expected.
Several major companies are reportedly tightening control over AI spending. Microsoft has scaled back many Claude Code licenses, while Uber leadership has indicated that rising AI costs are becoming harder to justify. Meanwhile, internal AI usage leaderboards at companies such as Meta and Amazon have reportedly disappeared after attention around how much employees were using these systems. While those companies have not publicly framed the move as a cost-cutting measure, removing leaderboards suggests that encouraging maximum token usage may no longer be the goal.
The issue is not limited to Big Tech. One reported case involved a company that accidentally spent around half a billion dollars in a single month because employee access to Claude was not properly limited. That kind of runaway spending highlights a growing concern for businesses adopting AI at scale: without strict controls, budgets can spiral quickly.
At the same time, the productivity gains from AI remain uneven. Some reports suggest that AI tools may save workers roughly an hour per day in certain roles. However, broader research into public AI deployments has found that many projects fail to deliver meaningful profit or meet performance targets. In other words, companies may be paying large sums for AI systems that do not always produce a strong return on investment.
This marks an important shift in the enterprise AI conversation. For the past few years, many companies treated generative AI as an inevitable upgrade to the workplace. It was often promoted as a tool that could replace repetitive tasks, accelerate coding, improve customer service, and reduce headcount.
But the economics are proving more complicated.
AI is not free. The cost of running large language models, processing prompts, generating responses, and supporting enterprise-wide usage can be substantial. When employees use AI casually or excessively, the expense can become difficult to defend, especially if the results are inconsistent.
The new focus for companies is likely to be controlled AI adoption rather than unlimited experimentation. Businesses may still use AI for coding, automation, research, writing assistance, customer support, and internal workflows, but with tighter rules around access, token consumption, and measurable results.
The message is becoming clear: AI can be powerful, but it must prove its value. For corporations, the question is no longer just whether employees should use AI. It is whether the productivity gains are worth the bill.
