By Brett Arends
Lone Nvidia bear Jay Goldberg shares why he’s still not a buyer of the AI superstock
Nvidia founder and CEO Jensen Huang, the man of the moment.
How does it feel to be the only analyst on Wall Street who is bearish about AI superstock Nvidia Corp. (NVDA)?
“It feels fantastic,” Jay Goldberg tells me with a laugh. “Everybody asks me this question.”
Goldberg, a research analyst at Seaport Research Partners, is the only analyst with a “sell” or “underperform” recommendation on Nvidia’s stock. Of the other 65 (yes, really), 60 give stock a “buy” or “outperform” rating and five give it a neutral “hold,” according to FactSet.
“I have never told my clients to ‘short’ Nvidia,” he adds, referring to the technique for trying to make money if a stock falls. “But I’ve always positioned my thesis as, ‘Nvidia is going to underperform the sector.’ And that has actually played out. If you look at the AI sector, Nvidia has underperformed since April 1 when I launched coverage.”
Goldberg is a former research analyst at Deutsche Bank. He also worked in the tech sector. He spent a decade in China and remains in touch with former colleagues there, while staying on top of developments. One of the (many) reasons he’s skeptical about Nvidia’s stock at current levels is that Taiwan Semiconductor, the company that actually produces the physical chips, is already running at full capacity. “They’re sold out,” he says. “And once they’ve sold out, where does the upside come from?”
But Goldberg’s analysis isn’t just about Nvidia stock. His thesis is important for everyone who invests in the stock market, even if they just have their 401(k) invested in broad-based index funds that track the S&P 500 SPX. When the last two bubbles burst, in 2000 and 2008, it wasn’t just the investments at the center of the mania – technology stocks and housing, respectively – that tanked. The entire market went down about 50%.
A market is supposed to match buyers and sellers. If you run out of sellers, that’s when you get in trouble. One of the oldest saws on the street of shame is that a bubble doesn’t peak “until the last bear turns bullish.” (Based on the events of 1999-2000 and 2006-07, it’s probably more accurate to say it doesn’t peak until the last bear capitulates or gets fired.)
Goldberg is standing his ground.
“It’s complicated,” he says.” I’m getting increasingly bearish about the AI cycle, the AI bubble. I fully subscribe to this as a bubble. Semiconductors are cyclical. Eventually, gravity will reassert itself. This could end in six weeks. It could end in three years.”
We’re already at the stage where big companies, particularly Nvidia, ChatGPT-owner OpenAI and others, are providing the capital to their own customers. This so-called “vendor financing” was a notorious feature of the dot-com and tech bubble of the late 1990s, and dragged everything down when the bubble burst.
Goldberg says things could get “even crazier” as companies issue massive amounts of debt to fuel their expansion. And that is starting to happen.
Meanwhile, the investment euphoria for the emerging technology of artificial intelligence has outstripped the proven demand from end customers.
“I’m especially nervous about the demand side of the AI trade,” he says. “I don’t think we fully understand the use case of AI. Enterprise adoption is tepid,” referring to usage by businesses.
A recent study by MIT found that 95% of the companies that have invested in AI have so far earned “zero return.” Meanwhile, up the road from MIT at Harvard Business School, the latest business review asks – with a straight face – “AI Companies Don’t Have a Profitable Business Model. Does That Matter?”
As MarketWatch has recently written, hedge-fund manager Harris “Kuppy” Kupperman, among others, has already run a slide rule over the math of the AI mania and found it comes up short.
Goldberg argues that many investors underestimate how expensive it is to run AI cloud-computing server farms, which need an enormous amount of electricity. He also says AI chips will be rendered obsolete far quicker than many in the industry say.
“The more important question is not the physical life of the server, it’s the economic life of the server,” Goldberg says. “In 2022, if you had a GPU, you could pay for it in six months. Now the payback is still somewhere between 11/2 and two years.” That’s “still pretty good,” he says. But the direction is down. “What matters is the obsolescence factor.”
This is an argument already made by Michael Burry, the hedge-fund manager made famous by Michael Lewis’s “The Big Short,” which successfully predicted the global financial crisis in 2008. Burry recently started betting against AI superstocks such as Nvidia. “He’s onto something,” Goldberg says.
In a surprising and mysterious development, Burry just deregistered his hedge fund, Scion Asset Management, from the Securities and Exchange Commission. “On to much better things Nov. 25th,” he wrote. It’s not clear yet what Burry means. But if he’s a bear who’s capitulating, that’s another ominous sign.
-Brett Arends
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
11-15-25 1238ET
Copyright (c) 2025 Dow Jones & Company, Inc.
