A corporate buzzword called “tokenmaxxing” is losing its shine as businesses across America find that throwing AI at every problem comes with a steep price tag — and not enough results to show for it.
The craze began earlier this year, fueled by enthusiasm from the tech industry around getting maximum output from AI products like OpenAI’s ChatGPT and Anthropic’s Claude. But that excitement has given way to a growing backlash as the bills have started arriving.
“It’s very easy to create something you don’t need with AI,” said Vincent Gusdorf, head of AI analytics at Moody’s Ratings and author of a new report urging companies to take a more careful approach to AI use.
So what exactly is “tokenmaxxing”? The term refers to pushing AI usage to its maximum by consuming as many tokens as possible. Tokens are the fundamental units of generative AI — small chunks of text that an AI system processes or produces, with each token equaling roughly three-quarters of a word. AI products typically set limits on how many tokens can be used, with higher-priced versions offering larger allowances.
“As bills started to pile in, people realized that those new tools are quite expensive and you need to use them wisely,” Gusdorf said.
Just months ago, Silicon Valley was celebrating heavy token use as a badge of honor for high-achieving employees. The image of the ideal tokenmaxxer was someone burning the midnight oil, possibly at the expense of their personal life, while directing multiple AI systems to work around the clock on their behalf.
OpenAI’s CEO said in May he was “excited to see what will happen with tokenmaxxing startups, both for how they work internally and the products they can build.” The CEO of Nvidia declared that if a highly paid engineer isn’t spending heavily on tokens, “something is wrong.” Facebook’s parent company even ran an internal contest that rewarded employees for token consumption.
The frenzy was a boon for major AI developers, but it began to unravel as it became clear that maxing out AI usage wasn’t the right move for most organizations.
Microsoft’s CEO acknowledged in a recent blog post that tokenmaxxing can become habit-forming, but he raised a warning flag: companies are effectively paying twice — once for the tokens themselves and again by handing over sensitive internal data to AI providers. His comments stood out for openly questioning the data security promises made by top AI companies, even as he promoted his own company’s alternative approach.
Palantir’s CEO was even more blunt, telling a financial news network earlier this month that something had gone “completely wrong.” He said he was voicing the frustration of American businesses quietly furious about spending large sums on tokens that produce no real value.
“The basic view among enterprises in this country is, ‘I’m going to chillax and waste my time with tokens. I’m going to get no value and they’re going to get my IP,’” he said.
A management consultant at Bain & Company said many of the large corporations her firm works with are now scrutinizing what they’re actually getting back from their AI investments.
“The token cost for them has been doubling, almost every other month,” she said. “Let’s say $200 per developer per month. Multiply that by 20,000 developers, which is often what we’re dealing with at these companies, and that quickly gets you to a number that is not a line item that any general manager has planned for.”
Part of the solution, she said, is simply matching the right tool to the right task — not reaching for the most powerful AI model to handle something as routine as writing an email.
“Not everything needs a Claude Opus 4.6,” she said, referring to one of Anthropic’s more advanced models designed for complex software work or in-depth research. “And yet you see so many companies, so many users, default to using Opus for everything, including generating emails.”
That realization has sparked interest in so-called AI “model routing” — technology that automatically directs simple tasks to cheaper, more efficient AI systems while reserving the heavy-duty models for genuinely complex work.
A software developer who leads developer experience at the startup Together AI said companies shocked by the “ridiculous amount of money” they’ve spent on AI subscriptions are moving away from rewarding heavy usage.
“It’s better to kind of just empower employees on how to use this stuff and let them use AI when and however much they need to,” said Hassan El Mghari, whose company provides developers access to a range of open-source AI models.
Meanwhile, those who still want to run up token counts are finding new options in low-cost open-source AI models from Chinese startups, including offerings from Moonshot’s Kimi and Zhipu’s GLM, which reportedly come close to matching the performance of top American models at a much lower price.
“There is some validity to the theory that this could push tokenmaxxing a little bit further,” said Raffi Krikorian, chief technology officer at Mozilla. “But if we look at the industry overall, I think it’s realizing that tokenmaxxing is a dumb thing.”
Krikorian compared the trend to an old software industry habit of measuring a programmer’s productivity by how many lines of code they wrote — a metric that eventually fell out of favor.
“I think tokenmaxxing is moving through the exact same pattern,” he said. “I think this is going to be an interesting blip that we’re all going to look back to laugh at in a year.”
