AI Costs Are Forcing Big Companies to Put Limits on Workers: Report

Corporate America is discovering that artificial intelligence may be revolutionary, but it is also noting that it is not cheap and as a result is beginning to ration access to its employees in an effort to cut costs.

After more than a year of urging workers to use AI for everything from coding to research to routine office tasks, some of the country’s biggest companies are beginning to ration access, track usage more aggressively, and steer employees toward cheaper tools as costs rise faster than expected.

The pressure point is the token, the basic unit used to measure AI computing. Every prompt consumes tokens. As companies rushed to prove they were AI-ready, usage exploded. So did the bills. According to a Wall Street Journal report, some companies have burned through annual AI budgets in just a few months, while others have seen costs double or triple.

Executives at companies including Uber, Meta, Microsoft, Salesforce, and DoorDash have either discussed or implemented new controls aimed at making sure AI spending produces measurable gains, not just activity.

An Uber executive told the Journal that the company had already exhausted its annual budget for “agentic” AI use by March. Microsoft limited access to Anthropic’s Claude Code program for some employees, directing them instead toward an internal coding assistant. Salesforce has introduced a system to connect token use to business outcomes.

Meta Chief Technology Officer Andrew Bosworth was blunt in an April memo cited by the report. “It has been great to let people experiment, but now we have too many overlapping tools,” Bosworth wrote. “Nobody should be using AI tools just for the sake of using them. All motion is not progress, and token usage alone is not a measure of impact of any kind.”

The problem has a nickname inside the industry: “tokenmaxxing.” It refers to employees using as much AI computing as possible, sometimes because they believe heavy usage signals innovation. But expensive premium models are often used for simple tasks that cheaper systems could handle.

Matan Grinberg, chief executive of coding automator Factory, put it simply: “If your daughter needs tutoring in algebra, you can probably find someone cheaper than Albert Einstein.” That realization is pushing companies to triage AI use.

Others are building homegrown tools to reduce dependence on outside providers. Anthropic, OpenAI, and Google all offer lower-cost versions of their flagship models, but some corporate buyers remain cautious about using the cheapest available systems, particularly when they come from Chinese developers.

Still, the broader AI market is not slowing. Google said at its recent I/O event that it now processes more than 3.2 quadrillion tokens a month across its surfaces, about seven times its volume from a year earlier. Anthropic announced Thursday that it raised $65 billion at a $965 billion post-money valuation, underscoring how much investor money is still chasing the AI wave.

There is also a productivity problem. EntelligenceAI, a startup that analyzed data from more than 2,000 companies using advanced AI coding tools, found that only 18% of spending on tokens translated into shipped coding products that reached real users.