Two presentations on the AI infrastructure boom
I'm continuing my series on the most interesting things I heard at the Value Investing Seminar in Italy earlier this month...
Today, I'd like to share two presentations on the AI infrastructure boom. Given that it has been a major driver of both the U.S. economy and stock market over the past few years, investors are rightly wondering if the boom has become a bubble that will burst.
The first presentation was by Gabriel Grego of Quintessential Capital Management, who's among the best-known and most successful activist short sellers in the world. In addition, he manages a traditional low-turnover, long-focused fund concentrated in high-quality companies.
He gave me permission to share his presentation, which you can access here: "AI Infrastructure: Boom, Bubble, or Both?"
Grego starts by underscoring that the AI infrastructure boom is "the largest capital cycle in the history of technology," citing several figures:
He then acknowledges the bear case and the many smart people behind it, including my friend Michael Burry of The Big Short fame:
He also outlines four major pillars of the bear case:
Grego is famous for his forensic investigations, which have led to him uncovering numerous frauds. Before shorting any company or sector, he fills out his "six-check scorecard," which assigns signals to six different financial factors.
He applied it to the AI infrastructure boom and came up with mixed results:
Grego doesn't think the comparison with the 1999 dot-com bubble is quite accurate, noting that it was driven mostly by multiple expansion, resulting in a more overvalued market and stocks.
In contrast, in the past five years, earnings growth has been the primary driver of the market and certain stocks like Nvidia (NVDA). So today, the price-to-earnings (P/E) multiples are more reasonable:
He also addresses one of Burry's main critiques: The value of the graphics processing unit ("GPU") chips that companies are spending hundreds of billions of dollars on is declining very rapidly.
Grego's analysis confirms these chips are indeed declining in value faster than the 5.5-year average that the hyperscalers are assuming – but not by very much. So he concludes that Burry is "half right":
After some additional analysis, Grego reaches the same conclusion I've come to: You want to "own the layer that keeps the value" – namely the "integrated franchises" of Alphabet (GOOGL), Microsoft (MSFT), Amazon (AMZN), and Meta Platforms (META) – and avoid the rest:
His favorite among these integrated franchises is Alphabet for several reasons: It has its own models, distribution, and demand... It's the cheapest of the four on conservative 8% growth... And Berkshire Hathaway (BRK-B) bought $10 billion worth of GOOGL stock.
The second presentation on the AI infrastructure boom was by Nicolas Chantier, the founder of N72, a Paris-based independent research firm.
He gave me permission to share his presentation, which you can access here: "The Return on a Trillion Dollars."
Like Grego, Chantier also addresses the debate over the reported book life versus the actual economic life of these AI chips. He concludes that their blended economic life is only 2.84 years:
This is only roughly half what the buyers are using to calculate the depreciation they report in their earnings. As a result, many companies are reporting "phantom profit – reported earnings that never convert to cash, surfacing later as impairment":
Chantier then looks at the layers of the chip value chain:
Lastly, he looks at which companies earn back the capital they're investing and reaches the same conclusion as Grego (and me): It's only the four hyperscalers – Alphabet, Microsoft, Amazon, and Meta:
Here you can also read Chantier's more detailed, free research paper on AI infrastructure, "DeepSeek, the Chinese Stack and the Ten-Year Cloud-to-Edge Trajectory."
Thank you both for sharing your excellent analyses! These are strong additions to the work my team and I have done in this sector – and it further reinforces our conclusion that the four hyperscalers are the most likely winners of the AI boom.
Best regards,
Whitney
P.S. I welcome your feedback – send me an e-mail by clicking here.
P.P.S. Susan flew home from Glasgow yesterday, while I flew to Budapest, Hungary to meet up with my parents, sister, aunt, and friend for another nine days in Europe.
My flight was six hours later than Susan's, so I bought a ticket for a red double-decker bus and toured Scotland's largest city all afternoon. I visited the Glasgow Cathedral, the nearby cemetery/necropolis, the largest terra cotta fountain in the world, the shipbuilding museum, the transportation museum, and a historic ship:
