Just months after Silicon Valley urged everyone to “use AI for everything,” the bill is arriving – and it’s bigger than expected. What started as a subsidized gold rush of cheap chatbots has morphed into an expensive reality check as businesses confront runaway costs driven by a new generation of AI tools.
When ChatGPT exploded onto the scene in late 2022, AI companies followed the classic tech playbook: offer rock-bottom (or even free) access to hook users and build market share. Investors happily footed the bill in the name of “subsidized intelligence.” But the era of giveaways is ending. With OpenAI and Anthropic preparing for potential IPOs later in 2026, the pressure is on to turn usage into actual revenue. Prices are rising across the board.
The biggest culprit? AI agents – autonomous systems that don’t just answer questions but do things: book meetings, write and review code, manage files, or even run multi-step workflows. Unlike simple chat queries, a single agent task can spin up dozens of sub-agents, each burning through tokens (the basic billing unit for AI usage). One task can consume many times more tokens than a standard conversation.
Inside big tech, an internal arms race took hold. Companies like Meta set up leaderboards ranking teams by token consumption as a proxy for productivity. Nvidia’s CEO Jensen Huang told engineers they should burn tokens worth roughly half their salary each year to stay competitive. The result? “Tokenmaxxing” – employees firing off AI prompts for everything, sometimes just to game internal metrics.
The consequences have been swift and painful. In some cases, AI costs have exceeded the salary of the employee using the tool within one or two months. Uber burned through its entire 2026 AI budget in just four months (by April 2026) after pushing Claude Code usage. Microsoft has canceled high-cost licenses for certain third-party tools, and multiple firms are now auditing runaway bills.
Even Meta is hitting the brakes. In an April 2026 internal memo, Chief Technology Officer Andrew Bosworth warned staff: “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.”
Uber’s Chief Operating Officer Andrew Macdonald went further in late May 2026, telling interviewers there’s still “no clear link” between higher token spend and more useful product features shipped to customers. “That link is not there yet,” he said.
Compounding the problem is a persistent shortage of the GPUs and data-center capacity needed to power today’s models. Demand continues to outstrip supply, keeping inference costs elevated even as per-token prices have fallen in some cases. The combination of higher usage volume and still-expensive compute has created a perfect storm.
Analyst Jack Gold of J.Gold Associates sums it up bluntly: “In some cases people are seeing the cost of tokens exceed the cost of the employee within a month or two of use, just because they’re using it too much.”
Faced with sticker shock, companies are getting pragmatic. Many are shifting from premium proprietary models to free or low-cost open-source alternatives that, while less powerful, handle the majority of routine tasks just fine. Others are implementing strict guardrails, real-time token monitoring dashboards, and ROI tracking to ensure every prompt delivers measurable value.
Consultants and CIOs are now asking tougher questions: Is this agent actually replacing hours of human work, or just generating more (sometimes bloated) output? Is the productivity gain real, or are we measuring motion instead of progress?
The AI industry is still in its adolescence. The trillions of dollars poured into infrastructure must eventually deliver returns, and the era of “move fast and subsidize everything” is giving way to disciplined spending. As Kevin Simback of Delphi Labs put it, “the tides are beginning to turn” – and an era where the big AI companies actually need to make money has begun.
For enterprises, the message is clear: AI remains one of the most powerful tools of the decade, but blind enthusiasm is giving way to objective accounting. The binge is over. Now comes the hangover – and the smarter, more sustainable phase of adoption that will determine who actually wins in the long run.
