Michael Burley, the investor known for predicting the housing meltdown in 2008, turned his attention to one of the market’s most beloved themes: artificial intelligence.
Mr. Barry recently deregistered his hedge fund company, Scion Asset Management, and removed it from regular regulatory disclosures. But he remains an active investor, doubling down on what he thinks will be the next big mispricing in the market.
The central figure in that view is Phil Clifton, a former associate portfolio manager at Scion, whose research supports the skepticism. Clifton argues that while the adoption of generative AI is accelerating, the economics behind the industry’s large-scale infrastructure build-out do not yet justify its cost.
In a farewell letter to Scion investors in late October, Mr. Barry called Mr. Clifton “the most extraordinary thinker I have ever met.” CNBC obtained some of Clifton’s research notes, written before he launched his own firm, Pomerium Capital, earlier this year, outlining Scion’s bearish thesis on AI.
The investment community “expects this technology to have economic significance that far exceeds its potential to deliver,” Clifton wrote. “Just because technology is good for society or revolutionizes the world doesn’t mean it’s a good business proposition.”
low profit margin
On the surface, the use of AI appears to be ubiquitous. According to the Pew Research Center, more than 60% of U.S. adults say they interact with AI at least several times a week. But Clifton said demand-side economics were “surprisingly small.”
OpenAI, a market leader and cultural phenomenon, is on track to generate more than $20 billion in annual revenue this year, but that number pales in comparison to the scale of its AI buildout. Hyperscalers have quadrupled their capital investment spending in recent years to about $400 billion a year, which is expected to rise to $3 trillion over the next five years, according to Man Group.
“We assume that other generative AI services, taken together, will not be sufficient to justify the amount spent on infrastructure,” Clifton wrote.
history warning
Scion believes there are clear historical parallels with the communications boom of the early 2000s, when heavy investment in fiber-optic networks far exceeded actual usage. Capacity utilization in the U.S. has fallen to about 5%, and wholesale communications prices have collapsed by about 70% in one year, Scion noted.
Clifton argues that cloud giants are now competing equally, expanding their AI infrastructure with the assumption that future demand will eventually catch up. However, if mass adoption of AI lasts longer than expected, the economics of these large-scale data center deals may become unsustainable.
He noted that some Big Tech companies are already starting to wobble on their commitments. Microsoft has canceled data center projects in the United States and Europe that would have used 2 gigawatts of electricity, citing oversupply. Alibaba’s chairman has warned that a bubble is forming in AI infrastructure.
Nvidia revelations
No company is benefiting more from AI investments than Nvidia. The stock price also soared due to unprecedented GPU orders from cloud providers. But Scion questions whether those customers will see any economic return on their investments.
Stock chart iconStock chart icon
Nvidia 1 year
The key factor here is depreciation policy. The tech giant has extended the lifespan of its servers on its books to six years. But NVIDIA’s product cycles currently run every year, and Scion argues that older chips become functionally obsolete and less energy efficient long before they become worthless.
Nvidia disputed this claim, saying that its hardware can remain productive much longer than critics say, thanks to the efficiencies made possible by its CUDA software system.
Still, Barry and other critics are conflicted. Nvidia says its latest chips are better in terms of performance, efficiency, and features, while also promising that older chips remain economically viable. They say one of those defenses needs to be given.
Burry has launched a new Substack newsletter laying out his bearish theories on AI. Whether generative AI ultimately turns out to be a bubble remains to be seen, but for now Barry is once again on the cautious side of a fast-moving story.
