Satya Nadella told Microsoft employees to stop using the most powerful AI models for every task, as runaway token spending threatens the economics of artificial intelligence.
Back
Microsoft's Nadella warns tokenmaxxers as AI costs surge 320%
Satya Nadella told Microsoft employees to stop using the most powerful AI models for every task, as runaway token spending threatens the economics of artificial intelligence.
Microsoft Chief Executive Officer Satya Nadella said "a lot" of tokenmaxxing is happening inside the company and urged employees to match AI models to task complexity, as enterprise AI bills climb 320% even as per-token prices have fallen 98%.
"Don't use frontier models for non-frontier problems," Nadella said at a live taping of The New York Times' "Hard Fork" podcast. "Let's kind of match these things such that you get the outputs, you get the economics — it can't be a race to doing things that just don't add value."
Nadella pointed to Microsoft Copilot's auto mode, which routes tasks to the most cost-effective model, as a template for disciplined AI use. The comments come as Uber Technologies Inc. disclosed it had blown through its entire 2026 coding budget by April, and one unnamed company burned through $500 million in tokens in a single month after failing to cap Claude licenses, according to Axios. Amazon.com Inc. and Meta Platforms Inc. have quietly killed internal AI leaderboards, while Walmart Inc. and Starbucks Corp. pulled back AI agent plans. An Amazon senior vice president told staff in a leaked email to "stop using AI just for the sake of using AI."
The cost-control push threatens the business models of OpenAI and Anthropic, which sell tokens as their primary product and are pursuing trillion-dollar IPOs. OpenAI alone has been losing more than $1 billion a month, partly from giving ChatGPT away to hundreds of millions of users. Anthropic co-founder Daniela Amodei suggested both firms had spent so heavily they sat "months from bankruptcy" and had "run out of options" beyond filing for trillion-dollar listings.
The Tokenmaxxing Hangover
Nadella acknowledged his own addiction to the practice. "I'm a tokenmaxxer too, it's addictive," he said. "But you have to step back when the novelty wears off to say, 'What is it that I'm trying to create?'" He added that he recently built a tool using vibe coding that keeps software projects updated by following workplace conversations — a targeted use case rather than blanket AI deployment.
The shift marks a reversal from 2025, when Silicon Valley executives pushed workers to use AI as much as possible, often through internal leaderboards tracking token consumption. Token prices for GPT-4-level responses have dropped to about $0.40 per million tokens from roughly $20 in late 2022, a 98% decline, yet enterprise AI bills have surged an estimated 320% as autonomous agents drive consumption higher, according to The Next Web.
The Industry Reckons With Waste
Cognizant Technology Solutions Corp. Chief Executive Officer Ravi Kumar S. called tokenmaxxing "a vanity metric" at a Fortune conference, accusing OpenAI's Sam Altman and Anthropic's Dario Amodei of "fearmongering" about AI's job impact. OpenAI's Altman acknowledged the strain in a CNBC interview, calling it a "fair issue" that clients keep asking "how long do I have to wait for AI benefits to show up in revenue?"
The backlash extends beyond corporate budgets. A March Pew poll found just 10% of Americans feel more excited than worried about AI's future, while 70% said they do not want a data center built near them. Data Center Watch counted at least 48 projects blocked or delayed in 2025. In Utah, local opposition forced Shark Tank investor Kevin O'Leary to shrink the land footprint of his planned Stratos data center by 75%.
Microsoft shares trade at roughly 33 times forward earnings. The company's ability to contain AI infrastructure costs while monetizing Copilot across its 220,000-person workforce will be a key driver of margin expansion. For Nvidia Corp., which has been the primary beneficiary of the AI buildout, the cooling of tokenmaxxing could signal slower GPU demand growth. Michael Burry of Big Short fame has maintained a short position against Nvidia, according to public filings. The S&P 500 has declined to add OpenAI, Anthropic, or SpaceX to the index, further pressuring their IPO valuations.
This article is for informational purposes only and does not constitute investment advice.
