An AI consultant told Axios that one of their enterprise clients spent $500 million in a single month on Anthropic’s Claude after no one set usage limits on employee licenses. The figure, disclosed as part of a broader investigation into corporate AI spending, represents what may be the single largest documented case of uncontrolled AI cost overrun.
The incident has become the flashpoint for a wider reckoning across corporate America, where companies that spent the past year encouraging employees to experiment freely with AI tools are now confronting bills that dwarf any productivity gains.
The Tokenmaxxing Reversal
The culture that produced the $500 million bill has a name: “tokenmaxxing.” The term describes employees deliberately maximizing AI token consumption to appear tech-forward or hit management-mandated usage targets. According to Axios, the shift from subsidized flat-rate AI pricing to usage-based billing made previously invisible consumption suddenly catastrophic.
Ali Ansari, CEO of model training firm Micro1, told Axios that the enterprise is undergoing a “healthy swing” away from tokenmaxxing. The data suggests the swing is anything but gentle.
Who Is Pulling Back
The backlash is hitting major tech companies simultaneously:
Uber burned through its entire annual budget for autonomous AI tools in three months. CTO Praveen Neppalli Naga disclosed the company had exhausted its Claude Code allocation by April. COO Andrew Macdonald told Rapid Response that it’s “very hard to draw a line” between token spending and “actually producing 25% more useful consumer features.”
Meta CTO Andrew Bosworth issued an internal memo: “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,” according to Times of India.
Microsoft stripped employee access to Anthropic’s Claude programs entirely, pushing staff toward internal coding assistants.
Salesforce built an internal tracking system requiring that every AI token used by a worker be tied directly to a profitable business outcome.
Amazon shut down its internal leaderboard that tracked employee AI usage, per the Financial Times.
The Bifurcation Thesis
Not everyone sees a uniform pullback. SaaStr founder Jason Lemkin told the 20VC podcast that companies “with well north of a million in revenue per employee” will continue to tokenmaxx indefinitely. “If you’re already hyper-efficient, you will find more ways to use AI,” he said. “The folks that are less efficient, that are larger organizations and more traditional, I think, as the year goes on, will become more skeptical, especially if prices go up.”
OpenAI CEO Sam Altman acknowledged the dynamic directly: “There’s a lot of great things I hear from companies. The negative one I hear is: our spending is going up and up, people feel like they’re being very productive, but where is the revenue, where are the actual productivity gains?” He told a Forbes Australia event that it will take “a little bit longer” for companies to figure out efficient implementation.
The Agent Cost Problem
The tokenmaxxing backlash has direct implications for autonomous agent deployments. When human employees burn through tokens on casual queries, cost overruns are linear. When agents spawn sub-processes, call APIs in loops, and run 24/7 without human-in-the-loop oversight, the same governance gap scales exponentially. Factory, a tech automation firm, told Axios that employees at a top financial institution were burning hundreds of thousands of dollars monthly using expensive AI subscriptions for casual conversation and basic questions.
Google disclosed that its token processing has reached 3.2 quadrillion tokens per month, seven times higher than last year, according to Times of India. As agent frameworks proliferate and autonomous systems claim increasing shares of that volume, the companies that survive the cost reckoning will be those that built spending controls before giving agents the keys.