As Nvidia Corp. has maintained a streak of hitting quarterly revenue and profit numbers out of the park, what might be lost to investors is that so much of the action around generative artificial intelligence still hinges on the build-out of computing infrastructure, rather than demand for products and services making use of the technology. Ken Laudan of Buffalo Funds is worried about that.
"I am growing concerned that the handoff from AI enablers to adopters will take much longer than I had previously anticipated," he said during an interview with MarketWatch.
Laudan manages the Buffalo Large Cap Fund, which is rated four stars (out of five) within Morningstar's "Large Growth" category.
Nvidia $(NVDA)$ remains an obvious play among companies helping data centers build the computing capacity needed to support their corporate clients' development of products and services that develop generative AI. Laudan calls the companies supplying the needed hardware "AI enablers." When interviewed in February, he said that "maybe in 2025" investors would begin paying more attention to "AI adopters," which he defined as "software-centric companies that sell a large-language AI model on top of their enterprise or vertical software stock to their clients."
But he has since changed his mind.
From numbers supplied by FactSet, as well as guidance from Nvidia, Advanced Micro Devices Inc. $(AMD)$ and Broadcom Inc. $(AVGO)$, Laudan estimated in a follow-up email that, from January 2023 through the end of 2024, cumulative spending "for the silicon (chip) portion of AI" would exceed $200 billion.
He estimated that "hyperscalers" investing heavily in AI equipment, such as Meta Platforms Inc. (META), eventually would need to generate an annualized $300 billion in annual revenue "just to be gross-margin neutral to their existing cloud-hosting business."
He clarified that the $300 billion figure would cover "application revenues paid to the hyperscalers from their enterprise customers or other AI companies," and said that figure compared with $3 billion in cumulative AI "application" revenue estimated by Sequoia Ventures in April. You can watch that presentation here.
Boiling down the warning about AI spending and a delayed payoff
Laudan asked: "When will hyperscalers - which represent 53% of Nvidia's revenues - begin to base GPU purchases on end-user demand, rather than buying them on anticipated demand?"
Nvidia dominates the market for graphics processing units (GPUs) installed by data centers. GPUs are the core components of the generative-AI infrastructure, although many other companies are providing supporting equipment.
"The hypsecalers are buying on anticipated demand," Ludan said. But he countered that point by saying he believed Tesla Inc. $(TSLA)$, and xAI (founded by Tesla Chief Executive Elon Musk last year) were "incrementally positive drivers" for Nvidia during its most recent reported fiscal quarter ended April 28. (The Buffalo Large Cap Fund had a small position in Tesla as of March 31.)
He said that an institutional investor had raised the point that companies such as Amazon.com Inc. $(AMZN)$ and Alphabet Inc. $(GOOGL)$ $(GOOG)$ couldn't afford to be left behind during the AI-infrastructure buildup.
"How much can investors tolerate?" Laudan wondered. "If there is a big hit to gross margins, investors will not tolerate this for five years. At some point the spark is going to go off," he said.
Laudan said investors need to see "hyperscalers dramatically shift" so the gap between AI infrastructure spending spend and AI adopter revenue begins to close "dramatically."
"But I am not seeing data points that are leading indicators that that will happen," he said. He expects to see clearer indications - positive or negative - over the next 12 months.
How Laudan is heeding his own warning
The Buffalo Large Cap Fund tends to hold about 70 stocks. Net of expenses, which total 0.95% of assets under management annually, the fund's investor shares have returned 22% this year, slightly ahead of its benchmark, the Russell 1000 Growth Index RLG, and ahead of the 15.6% return for the S&P 500 SPX. (All returns in this article include reinvested dividends.)
Broadly, Laudan said he was "trimming where I need to in the AI ecosystem, staying overweight in industrials and healthcare (relative to the Russell 1000 Growth Index), and looking at some of the beaten-up large-cap software names."
He said he had reduced the fund's holdings of Nvidia, which made up 7% of the fund's portfolio as of March 31. He also reduced the fund's holdings of AMD "a bit" and sold some shares of Coherent Corp. $(COHR)$ and Pure Storage Inc. (PSTG), two alternative AI plays he mentioned during the February interview with MarketWatch.
Microsoft Corp. $(MSFT)$ was the largest holding of the Buffalo Large Cap Fund as of March 31, with an 11% allocation. The stock has returned 22% this year, and software companies in the S&P 500 have returned a weighted 16.9%, according to FactSet.
Beyond the world of tech and AI, Laudan appreciates "industrial compounders," such as GE Aerospace $(GE)$, Xylem Inc. $(XYL)$ and Westinghouse Air Brake Technologies Corp. $(WAB)$, also known as Wabtec.
Within healthcare, Laudan still likes Eli Lilly & Co. $(LLY)$, which reported a 26% year-over-year increase in first-quarter sales as demand soared for its GLP-1 weight-loss medications.
Other healthcare names Laudan listed as favorable included Abbott Laboratories $(ABT)$, DexCom Inc. $(DXCM)$ and AstraZeneca PLC $(AZN)$, along with Merck & Co. $(MRK)$, which he said he had added to the portfolio recently.