Major Companies Rein In Employee AI Use After Costs Explode

The era of unlimited AI experimentation at major corporations has come to an abrupt end. Companies including Uber, Meta, Microsoft, and Salesforce are now rationing employee access to artificial intelligence tools after costs spiraled far beyond expectations.

The Rise and Fall of Tokenmaxxing

The problem has a name inside the industry: tokenmaxxing. The term describes employees consuming as many AI tokens as possible, sometimes to signal innovation rather than to produce real results. Tokens are the basic data units that measure how much a large language model, or LLM, processes during each interaction — roughly one token equals about 1.5 words of text.

At some companies, the practice turned into an informal competition. Meta employees created a leaderboard tracking who consumed the most tokens, which the company has since taken down. Amazon workers reportedly spun up AI agents to complete unnecessary tasks just to keep their usage numbers high.

The financial impact has been severe. Uber burned through its entire 2026 token budget within the first four months of the year, partly due to heavy use of coding assistants. Salesforce CEO Marc Benioff revealed the company’s annual bill from AI provider Anthropic is projected to reach approximately 300 million dollars.

Companies Fight Back With New Controls

After more than a year of urging employees to use AI for everything from writing code to conducting research, executives are now reversing course. Microsoft cancelled access to certain AI coding tools for employees in several product divisions, directing them toward cheaper internal alternatives instead.

The shift reflects growing pressure to demonstrate return on investment. Uber’s Chief Operating Officer Andrew Macdonald noted that without a direct link between token spending and useful features shipped to users, the costs become difficult to justify.

Some experts argue the growing pains are normal. Technology analyst Azeem Azhar compared the current moment to the early days of factory electrification, when companies needed time to redesign entire workflows before seeing productivity gains from a new technology.

Still, the message from corporate leadership is clear: the days of unchecked AI spending are over, and companies now want proof that every token spent delivers real value. The tokenmaxxing era has officially ended.