An inside look at his analysis showing AI is a bubble
Michael Burry, the investor best known for predicting the housing meltdown before 2008, has turned his attention to one of the market’s most beloved themes: artificial intelligence.
Burry recently deregistered hedge fund firm Scion Asset Management, excluding it from routine regulatory disclosures. But he continues to actively invest and is doubling down on what he sees as the next big mispricing in the markets.
At the center of this view is Phil Clifton, Scion’s former associate portfolio manager, and his research provides the basis for skepticism. Clifton argues that while the adoption of generative AI is accelerating, the economics behind the industry’s massive infrastructure buildout have yet to offset the cost.
In his farewell letter to Scion investors in late October, Burry called Clifton “the most extraordinary thinker” he had ever encountered. CNBC obtained several of Clifton’s research notes written earlier this year, before he founded his own firm, Pomerium Capital, that help outline Scion’s bearish thesis on AI.
Clifton wrote that the investment community “expects much more economic significance from this technology than is likely to be achieved.” “Just because a technology is good for society or revolutionizes the world doesn’t mean it’s a good business proposition.”
low margins
On the surface, the use of AI seems ubiquitous. More than 60 percent of U.S. adults say they interact with artificial intelligence at least a few times a week, according to the Pew Research Center. But Clifton said the economics on the demand side were “surprisingly small.”
Market leader and cultural phenomenon OpenAI is expected to exceed $20 billion in annual revenue this year, but that figure is tiny compared to the size of its AI structure. Hyperscalers have quadrupled their capital spending in recent years to almost $400 billion annually, with expectations for $3 trillion over the next five years, according to Man Group.
“We assume that other productive AI services are insufficient in aggregate to justify the amount spent on infrastructure,” Clifton wrote.
Warnings of history
Scion sees a clear historical parallel with the telecommunications boom of the early 2000s, when heavy investments in fiber-optic networks far outstripped actual usage. Scion noted that U.S. capacity utilization has fallen to about 5 percent and wholesale telecommunications prices have fallen nearly 70 percent in a single year.
Clifton argues that cloud giants are now in a similar race as they expand their AI infrastructure with the assumption that future demand will eventually catch up. But if mass adoption of AI takes longer than expected, the economics of these massive data center deals could become untenable.
He noted that some Big Tech companies are already waffling on their commitments. Microsoft canceled data center projects that would use 2 gigawatts of electricity in the USA and Europe. citing excess supply. Alibaba’s president warned A bubble is forming in the artificial intelligence infrastructure.
Exposure to Nvidia
No company has benefited this much from AI spending Nvidia. The stock has surged along with unprecedented GPU orders from cloud providers. But Scion questions whether these customers will get an economic return on this investment.
Nvidia one year
An important element here is the depreciation policy. Tech giants have extended server life on the books by up to six years. But Scion claims that Nvidia’s product cycles now continue year after year, and older chips are becoming functionally obsolete and less energy efficient long before they’re even written.
Nvidia pushed back on that claim, saying its hardware remained productive much longer than critics said thanks to the efficiency provided by the company’s CUDA software system.
Yet Burry and other critics note a paradox. Nvidia says its newest chips are superior in performance, efficiency and capability, while also promising that older chips will remain economically viable. They say one of these defenses should be given.
Burry published a new Substack newsletter to lay out his bearish thesis on AI. We don’t know yet whether generative AI will eventually become a bubble, but for now Burry is on the cautious side of a fast-moving story.
