A corporate mod of “tokenmaxxing”. artificial intelligence technology is pushing its limits as workplaces that throw artificial intelligence into everything are seeing costs rise without a corresponding increase in productivity.
What began as the tech industry’s spring hype for squeezing as many AI-generated jobs as possible out of products like OpenAI’s ChatGPT and Anthropic’s Claude has shifted to a summer. opposite reaction.
“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 recommending a more disciplined approach.
“Tokenmaxxing” refers to maximizing the use of tokens—the building blocks of generative AI that correspond to small chunks of text that an AI system reads or writes. Each token is about three-quarters of a word. And there’s usually a limit to how much you can use, with more expensive versions of AI products offering higher caps.
“As the bills started piling up, people realized that those new tools are quite expensive and you have to use them wisely,” Gusdorf said.
Technology leaders consider high AI usage a badge of honor
Just a few months ago, Silicon Valley executives were promoting high consumption as a signal of high-performing employees. The stereotypical Tokenmaxxer was staying up late—perhaps ignoring their significant other—while orchestrating an army of 24-hour AI agents running errands on their behalf.
OpenAI CEO Sam Altman said in May that he was “excited to see what happens with tokenmaxxing startups, both in terms of how they work internally and the products they can build.”
Nvidia CEO Jensen Huang said, “If your $500K engineer isn’t burning $250K in tokens, something’s wrong.” Facebook’s parent Meta had an internal contest that rewarded token usage.
Trend increased income for the core developers of the big AI language models like Anthropic and OpenAI, but failed as it became clear that it wasn’t necessarily the best strategy for everyone else.
Microsoft CEO Satya Nadella has acknowledged that tokenmaxxing can be addictive, but warned in a recent blog post that customers of those models are paying twice for AI, first in spending on tokens and second by giving them all their proprietary data. While promoting Microsoft’s own approach, Nadella’s comments were unusual in how he raised doubts about the data protection assurances of major AI providers.
Palantir CEO Alex Karp went further, telling CNBC earlier this month that something had gone “absolutely wrong.” He said he was channeling the voice of US businesses privately “glad” to pay so much for tokens that create no value.
“The basic view among enterprises in this country is, ‘I’m going to sit back and waste my time with tokens. I’m going to have no value and they’re going to take my IP,'” Karp said.
Workplaces require more and better ‘direction’ of their AI work
Bain & Company management consultant Jue Wang said many of the large businesses her firm advises have taken a closer look at the returns on their AI investments.
“The token cost for them has doubled, 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 in these companies, and that quickly gets you to a number that’s not an item any general manager has planned for.”
Sometimes that just means not using the AI equivalent of a sledgehammer to crack a nut.
“Not everything needs a Claude Opus 4.6,” she said of one of Anthropic’s most capable models suitable for software engineering or deep research. “And yet you see so many companies, so many users, who use Opus for everything, including email generation.”
This has led to a search for tools that do “model driving” of AI – in which easier questions are automatically sent to cheaper and more efficient AI systems, and more complex tasks go to more powerful models.
Open source AI models built in China offer less expensive alternatives
Software developer Hassan El Mghari said companies’ shock at the “ridiculous amount of money” spent on subscriptions to AI products by major US companies has led many to shy away from high-reward usage.
“It’s better to empower employees on how to use these things and let them use AI when and as much as they need,” said El Mghari, who leads developer experience at startup Together AI, which supplies developers with a variety of “open source” AI models.
At the same time, those who favor hoarding as many tokens as possible are having a field day with new open source models from Chinese startups such as Moonshot Chemistry or Zhipu’s GLM, which almost match the capabilities of the best American models at a fraction of the price.
“There’s some validity to the theory that this could push tokenmaxxing a little further,” said Raffi Krikorian, chief technology officer at Mozilla. “But if you look at the industry as a whole, I think you’re realizing that tokenmaxxing is stupid.”
It’s similar, Krikorian said, to how software companies used to consider how many lines of code a programmer wrote as a good productivity metric. This later fell out of favor.
“I think tokenmaxxing is following the same pattern,” he said. “I think this will be an interesting shot that we’ll all be looking back at to laugh at in a year’s time.”
Subscribe to our free newsletters
Our weekly newsletter Closing arguments provides the latest on ongoing trials, major litigation and court decisions across the US and the world, while monthly Under the lights feeds legal dirt from Hollywood, sports, Big Tech and the arts.
