Tokens or humans? the new corporate trade-off

AI is turning out to be much more expensive than anyone expected, and CFOs at major US companies now face a brutal new choice: tokens or people?

It was a picture of the two corporate AI CEOs at the center of the structure described to CNBC this week. Disclosures of what’s happening at the Fortune 500 paint a stark picture of the threat rising costs pose to the AI ​​business. This is a risk the market has yet to realize, as it has reached record levels and new trillion-dollar companies have emerged. Micron.

Arvind Jain, CEO of enterprise AI company Glean, told CNBC that the number one issue for every business right now is sky-high AI budgets.

“Companies tell us their AI budgets will run out in a month or two, and these are annual budgets,” he said.

This is because the cost of artificial intelligence has not fallen as buyers expected. More precisely, it rose. Each new model coming out of frontier labs is roughly twice as expensive per token as the model it replaces, putting enterprise AI in what Jain calls “an unsustainable path right now.”

“This is the first time I remember technology costing the same as people, and you make that comparison: choose technology or people,” he said. “We’ve never had this conversation in history because technology is such a small fraction of the overall cost of any business.”

Growing AI budgets are increasingly replacing future headcount increases, he says.

Matan Grinberg, CEO of Factory AI, which leads engineering efforts across all advanced AI models, described this shift as an identified resource allocation issue currently emerging within leadership teams.

“Companies are saying, if we can optimize something, is it the number of employees we have or is it the AI ​​spend per employee?” Grinberg said.

Companies go through three different phases in about a year, Grinberg said. The first boards to join demanded their CEOs do something about AI. It was later called tokenmaxxing, or the use of artificial intelligence by any means necessary, regardless of cost. In the third phase, leadership teams re-evaluate their needs for premium models.

“Do we need to use Opus-level intelligence for every mission?” Grinberg said. “There’s just no need for it.”

The basis of the squeeze is that the technology works but has not yet paid off.

“The way AI works today is very powerful but very inefficient,” Jain said. “At this point, the value provided by artificial intelligence lags behind the costs businesses incur.”

A big part of the problem is inefficiency in model selection. About 95 percent of enterprise AI use still runs on the most expensive frontier models, even for tasks that could be handled by cheaper alternatives, Jain said.

There’s a simple fix: redirecting the easy job to the cheaper level. Jain said this is the lowest hanging fruit.

“You really have 10 times the savings you can achieve with the right model routing on the front end,” he said.

This is also the step behind Factory AI, which automatically sends each task to the model best suited to it. The trick, Grinberg said, is to realize how rarely a business actually needs the top level. He compared the gap between the newest leading models to two experienced academics.

“Opus 4.7 versus Opus 4.8 is like the difference between a professor who’s been a professor for 13 years and a professor who’s been a professor for 15 years,” Grinberg said. “It’s really hard for the average person to tell the difference.”

The entire AI trade is based on the bet that historic demand will continue, with buyers largely indifferent to the cost. But Fortune 500 insiders suggest demand may be much more sensitive to price than trading assumes.

To read Learn more about what AI price calculation means for valuations at OpenAI and Anthropic, which build their business models on premium pricing.