listen to this article
Estimated 5 minutes
The audio version of this article is generated by AI-based technology. Mispronunciations may occur. We work with our partners to continually review and improve our results.
High-tech companies that were committed to using AI internally are now tightening their belts as the costs of intensive AI use soar.
Last month, Uber admitted that it had used up its entire 2026 AI budget in the first four months of this year, with the company’s COO saying:difficult to justifyOpenAI CEO Sam Altman said earlier this month that AI costs have become a “huge issue” for customers.
And it’s not just the big players. At a recent conference, Canadian startup leaders said they are feeling the costs of increased internal AI spending. Bettakit report.
The new focus is on tracking costs and finding ways to use AI more strategically. But what happens to AI companies’ massive valuations if technology companies rein in their spending?
The ballooning cost comes down to the cost of using “tokens,” the units of data needed to prompt and receive output from the AI. In fact, the amount of tokens being used by companies is significantly higher thanks to the rise of “tokenmaxxing,” which equates to the cost of a user interacting with an AI.
Increasing costs of increasingly complex AI tasks
Overall, the cost of real-world applications of AI called inference is decreasing. However, technology companies are using AI for complex “agent-like” tasks such as coding and thought-chain reasoning. It’s a completely different world than when an individual asks ChatGPT what to make for dinner.
“These agents ask a lot of questions internally as they try to come up with an answer for you,” said Gary Marcus, a cognitive scientist and AI researcher. “I don’t think there are exact statistics, but in some cases it can take 500 or 1000 times longer.” [tokens]. ”
Until recently, many technology companies encouraged their employees to go all-in on experimenting with AI. The push to incorporate heavy use of AI in the workplace has also given rise to the term “tokenmaxxing,” which refers to burning as many tokens as possible. For example, Meta and Amazon employees at one point used internal leaderboards to compete to see who was spending the most tokens. The use of large amounts of tokens served as a kind of shorthand for productivity.
As recently as March, NVIDIA CEO Jensen Huang said he would be “very alarmed” if a software engineer making $500,000 didn’t spend $250,000 a year on tokens.
But faced with the high cost of using tokens, some tech companies are rethinking their spending. For example, Uber recently imposed a cap of $1,500 per employee per month for each coding tool.
Companies are “moving away from very naive experiments,” he said. Nestor Masley He is 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 reality of integrating AI into organizations in a broad way.
Transition to “Tokenomics”
So what is the strategic path forward for companies feeling the pressure to innovate and implement AI, but still want to control costs and reap tangible benefits?
Introducing AI “tokenomics”: Gain a deeper understanding of token costs and use AI in strategic and financially predictable ways.
Masrezi advises focusing on “micro-sized experiments to figure out where AI can be useful as a tool.” Companies should assess: “Can we do things faster than humans? [at] How much does it cost? ” he said.
There is no one-size-fits-all solution either, and it will vary “not just at the organizational level, but at the departmental level,” Masley said. “The way HR uses AI is going to be different than the way legal or engineering uses AI.”
The federal government’s new AI strategy sets out plans to create 250,000 new AI jobs over five years. But is it realistic? CBC Radio’s The House explains it.
Who pays the real cost of AI?
The AI sector is facing liquidation as companies assess whether the costs associated with the complex use of the technology provide a return on investment.
“If token maximization is not maintained, these [AI] Companies probably won’t be able to make as much profit as they thought they would,” Marcus said.
AI companies may find themselves in a bind, needing to recover token costs while maintaining market share in a highly competitive environment.
current state of humanity enterprise planfor large enterprises, includes both flat fees and fees based on token usage. In early June, Microsoft-owned GitHub Co-Pilot changed its fees to tie prices to token usage.
The Wall Street Journal also reported at the same time. OpenAI is considering lower costs We’re thinking about using that token, perhaps to attract users from Anthropic. Chinese startup DeepSeek recently announced a 75% discount on key models.
“To me, all of this is evidence that this technology is still in its infancy, not only in terms of functionality but also in terms of price,” Masrej said. “I still think there are some costs that companies are willing to pay.”
