The era of AI maximalism is grinding to a halt. It was only months ago that CEOs were forcing employees to use AI as much as possible for tasks like coding. But at some point during their AI binge, the big wigs stopped to check their tab, and are now having second thoughts. Their employees are hooked on AI coding tools, but the costs of using them are spiraling out of control.

How businesses go about reconciling these costs with their AI evangelism is “going to be an absolute nightmare,” an unnamed big tech executive told The Economist.

It’s an ironic reversal of fortunes for companies in the tech sphere, which have become one of the main adopters of AI. Bosses have been happy to slash their workforces and replace them with AI coding agents that can churn out mountains of code, encouraging their surviving employees to make use of AI help as much as possible. Some, like Amazon, instituted a leaderboard ranking employees by the number of AI tokens they used, as if they were competing in a video game. Meta also even factored AI usage into performance reviews.

This wasn’t an entirely top-down phenomenon. AI bros have embraced this maximalism in tongue-in-cheek fashion, giving it the aptly meme-y named ethos of “tokenmaxxing.”

All of this has backfired in predictable fashion. At one business, a single employee spent over $150,000 a month on AI tokens. An Nvidia executive admitted that he was spending more on AI costs for his research team than what he pays the actual employees. One unfortunate company reportedly blew through $500 million in a month on Claude usage fees. On average, new research from the Ramp AI Index found that the most “AI-pilled” businesses are spending around $7,500 per employee every month on AI.

Now the noise coming out of the industry is cautionary rather than exuberant. AI isn’t the problem, of course, but how you use it. Experts have advised imposing token limits on employees, being more selective about where AI is deployed, and using cheaper models. Signaling the vibe shift, Amazon and Meta have ditched those AI leaderboards, and a top Uber executive said AI wasn’t yielding clear productivity gains compared to their expensive costs; soon after those comments, Uber imposed a $1,500 monthly token cap per employee.

That AI customers are already reeling from AI costs and tapering off usage doesn’t bode well for the model makers. Token costs right now could be the cheapest they’ll ever be, as they’re effectively subsidized by the companies providing the models to get customers on board. 

But can AI companies afford to keep their prices low, when their road to profitability remains elusive? It’s a question that OpenAI might be speeding headlong into answering. While many raise their rates and switch over to usage based billing, the Sam Altman-led firm is reportedly considering slashing its rates to launch a price war with arch rival Anthropic, in anticipation of it doing the same. The gamble could secure it more customers in the short term and assure the business world that their models will still be worth the investment, but all eyes will be on both companies to see if cheap AI access is sustainable in the long run, or just a ditch measure before they start charging more again.

More on AI: Meta’s Super Expensive New AI Team Is Already a Complete Catastrophe