MW The massive run-up in AI stocks this year may be built on a 'token mirage'
By Barbara Kollmeyer
Trading veteran Kevin Muir, author of the Macro Tourist on Substack, raises questions about true demand for artificial intelligence
Many heavyweight companies are all-in on AI, but is that really going to keep the stock market rally going?
The breathless rally seen this year among semiconductor and memory stocks may be based on a flawed understanding of company AI usage.
That's according to an institutional-trader-turned-blogger who posted the warning on his Macro Tourist publication on Substack. Kevin Muir readily admits, and has done so recently, that he was "caught flat-footed" over the AI run-up over the past few months.
"I believe I understand what drove the rally. I see where I went wrong and what I missed. However, I now face a quandary: the market is likely misreading a bullish signal it has received," he wrote on Monday.
Muir said evidence of surging rental prices for the graphic processing units, or GPUs, used to train large language models and their plummeting availability - correctly flagged by 3-Fourteen Research founder Warren Pies and others - have been helping to drive that rally. The Van Eck Semiconductor ETF SMH is up 45% this year, and the Roundhill Memory Index ETF DRAM has surged nearly 65%, he noted.
But the stock market has "taken this demand and extrapolated it into the future," with chip and memory stocks adding $7.8 trillion in market cap this year. Muir believes insatiable GPU demand is based on a hard corporate push to get employees and engineers to spend tokens - units of data processed by artificial-intelligence models - even if at high costs, he said.
"Tech CEOs face a strange environment where they are forced into a massive AI arms race," according to Muir. "They are petrified that one of their competitors will discover AGI (artificial general intelligence) and render their own companies redundant. Therefore, tech CEOs have little choice but to compete in AI spending."
He referred to a deep dive by the House of El report that looked at the human cost of using tokens - consumed any time AI is responding to a question, writing code or running an agent. El, identified as a computer scientist and polymath on the House of El channel on YouTube, noted recent reports about Meta and Amazon engineers letting agents run idle or complete meaningless tasks just to achieve - and be rewarded for - higher token-use counts.
Muir, additionally, pointed to a recent Axios report that one company had spent $500 million in a single month because no limits had been placed on employee Claude usage.
"Right now, the market is assuming that the past three months uptick in AI usage represents true demand. Yet, what if the circumstances developed in a way that had too many tech companies using AI compute with little regard for cost? What if that monster surge in AI usage that had guys like Warren Pies correctly rushing in and buying the AI stocks was nothing more than a demand mirage?" Muir asked.
Those seeking to gauge the rally in semiconductor and memory stocks ought to be aware that it's been driven at least in part by "token maxing and other questionable AI use by large tech firms," he said. And that may have exaggerated the pace of adoption over the last two months.
The dot-com bubble," Muir observed, "didn't end when the internet stopped growing. No, it ended when it stopped growing as fast as was expected."
The PHLX Semiconductor Index SOX, off nearly 1.5% early Monday, has surged 78% in 2026 and nearly 162% over the past 12 months.
Top components of the "SOX" index include Nvidia (NVDA), Taiwan Semi $(TSM)$, Broadcom $(AVGO)$ and Micron $(MU)$.
Read on: A 'volatility spasm' is set to provide the toughest test yet to the 9-week-old stock-market rally
-Barbara Kollmeyer
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
June 01, 2026 10:07 ET (14:07 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments