Companies question cost of AI as tokenmaxxing spending adds up

Tech companies that went all in on using AI internally are now clamping down in the wake of soaring costs for intensive AI use. 

Last month, Uber acknowledged it had spent its entire 2026 AI budget in the first four months of the year, with the company’s COO saying it was becoming “harder to justify” internal AI costs. OpenAI CEO Sam Altman said earlier this month that AI costs have become a “huge issue” for its customers.

And it’s not just the big players. At a recent conference, leaders from Canadian startups said they’re feeling the costs of growing internal AI expenses, Betakit reports. 

The new focus is on finding ways to track costs and use AI more strategically. If tech companies rein in spending, though, what does that mean for the massive valuations of AI companies?

The ballooning expense comes down to the cost of using “tokens,” the units of data it takes to input prompts to AI and receive an output from AI. In practical terms, the quantity of tokens that companies are using — and it’s been quite a lot, thanks to the rise of “tokenmaxxing” — amounts to the cost of users interacting with AI. 

Rising cost of increasingly complex AI tasks

Overall, the cost of real-world applications of AI — called inferences — has been falling. However, tech companies are using AI for complex, “agentic” tasks like coding and chain-of-thought reasoning. It’s a very different world from individuals asking ChatGPT what to make for dinner.

“Those agents internally pose many, many queries in the process of getting you to your answer,” said Gary Marcus, a cognitive scientist and AI researcher. “I don’t think there’s any precise statistic here, but sometimes it takes 500 times as much, a thousand times as many [tokens].”

Until recently, many tech companies encouraged employees to go all-in on AI experimentation. The push to incorporate heavy AI use at work even spawned the term “tokenmaxxing”: burning through as many tokens as possible. Employees at Meta and Amazon, for example, for a time used internal leaderboards to compete for who was using the most tokens. Heavy token use served as a kind of shorthand for productivity. 

As recently as March, Nvidia CEO Jensen Huang said he’d be “deeply alarmed” if a software engineer making $500,000 wasn’t spending $250,000 a year on tokens.

But faced with high costs for token use, some tech businesses are rethinking their spending. Uber, for example, recently imposed a $1,500 US cap per month per employee per coding tool.

Companies have “moved away from very kind of naive experimentation,” according to Nestor Maslej, the CEO of a consulting firm that advises companies on the use of AI and former editor-in-chief of Stanford University’s AI Index Report. Instead, he says they are dealing with the realities of integrating AI in a broad-based way into their organizations.

Shifting to ‘tokenomics’

So, what’s the strategic way forward for businesses that both feel the pressure to innovate and adopt AI, but also want to control costs and see tangible benefits? 

Enter AI “tokenomics”: better understanding token cost, and using AI strategically and in a financially predictable way.

Maslej advises focusing on “micro-sized experiments to figure out where AI is useful as a tool.” Businesses need to evaluate: “Does it do things faster than a human and also [at] what cost?” he said.

There aren’t one-size-fits-all solutions, either, but will vary “not only on an organization by organization level, but also on a function level,” Maslej said. “The way HR uses AI is going to be different than the way legal or engineering uses AI.”

WATCH | Will Canada’s AI strategy create jobs?:

Will Canada’s push for AI make or break jobs?

The federal government’s new AI strategy lays out a plan for 250,000 new AI-relevant jobs in five years. But is that realistic? CBC Radio’s The House breaks it down.

Who pays the real cost of AI?

The AI sector is facing a reckoning, as companies assess whether costs associated with complex uses of the technology deliver a return on their investment. 

“If tokenmaxing is not sustained … then these [AI] companies are probably not going to make the revenue that they thought,” Marcus said.

It may leave AI companies between a rock and a hard place: on one hand needing to recoup token cost, but on the other, needing to keep market share in an intensely competitive environment. 

Anthropic’s current Enterprise plan, for larger businesses, includes both a flat fee, and a fee based on token usage. At the beginning of June, Microsoft-owned GitHub Co-Pilot changed its fees to tie price to token use. 

At the same time, the Wall Street Journal reports OpenAI is considering lowering the cost of using its tokens, likely in a bid to attract users from Anthropic. Chinese startup DeepSeek recently announced a 75 per cent discount on its primary model.

“To me, this is all evidence of a technology that’s still in its very early days in terms of not only its capability, but also how it’s priced,” Maslej said. “I do still think there is some level of cost that businesses will be willing to pay.”