Veteran Stockpicker Sees Risks for AI, Big Tech. Why It Could Be Worse Than the Dot-Com Bust.

Rajiv Jain. (Photograph by Guerin Blask)

Rajiv Jain, chairman and chief investment officer of GQG Partners , who rode the technology boom with stocks like Nvidia in recent years, is worried.

He sees behavior reminiscent of the complicated vendor-financing setups at Cisco Systems, Lucent, and Global Crossing in the late 1990s, which ultimately became emblematic of the dot-com bust. And he says an unraveling of the artificial-intelligence boom could be even worse than the collapse of the internet-stock bubble.

Jain, a panelist on the Barron’s Roundtable, admits he may have been early, selling Nvidia, Alphabet , and Amazon.com earlier in the year and paring his stake in Microsoft. But he says the “all-in on one-way bet on AI,” combined with decelerating growth, narrowing free cash flow margins, and increasing competition for many of the sector’s heavyweights is enough for him to move on to more defensive stocks.

Too much money is going into a technology still in its infancy, without the network effects and other positive traits that made other technologies longer-term winners, says Jain. He is one of the 4% of active large-cap managers who were investing during the dot-com era. And to him, the AI frenzy and corporate behavior within the sector is giving off dot-com vibes.

We spoke with Jain this week to learn why he is so concerned and skeptical of AI’s promise. An edited version of our discussion follows.

Barron’s: What are the parallels to the dot-com era?
Jain: This is dot-com all over again, with no sustainable profits. OpenAI’s model is flawed because it doesn’t scale well, which is the fundamental issue. Less than 3% of their customers are paying. That is unlikely to change much because they have a high percentage of customers located in price-sensitive emerging-markets countries.

And unlike in software, where the marginal customer’s revenue flows to the bottom line, in [large language models] currently, every query is compute intensive and the customers who use AI the most will most likely end up being low margin revenue because of the compute intensive nature. That is why we think their cash losses are increasing.

I am all for investing ahead of the curve, but that makes sense where there are moats, network effects, and switching costs. Why was Uber such a good business? Because the more users, the better the experience for both the driver and the consumer. Here, you don’t have this dynamic.

And there isn’t much differentiation currently, so it’s easy to switch, which makes these models—OpenAI or Anthropic—ultimately a commodity that should at best earn marginal profits! Plus, our research shows most of the world going toward open-source, small-language models.

What behavior are you seeing in the sector that worries you?
Cloud took decades to become meaningful. It’s silly to assume AI has matured and plan to spend trillions of dollars. Every other day we hear of disruption—small models, domain-specific models outperforming LLMs, or something else.

A major part of VC funding has gone into AI start-ups. Plus, large tech companies are investing in these start-ups. Look at Nvidia’s $100 billion investment announcement in OpenAI: It was just a letter of intent and came days before the closing of OpenAI’s latest fundraise. Was it to help them gain traction?

If OpenAI blows up, this AI story is over. Even Nvidia doesn’t have $100 billion in free cash flow. Last year, it was $60 billion.

There is much brazenness in this circularity: Nvidia invests in CoreWeave, which in turn buys [graphics processing units] from Nvidia, and Nvidia subsequently guaranteed offtake, [or future purchases,] from CoreWeave. Similarly, OpenAI would get money from Nvdia, and subsequently OpenAI buys compute from Oracle, which then buys GPUs from Nvidia. This is turning financing cash flows into revenue and artificially creating revenue and profits—identical to Cisco, Lucent, or Global Crossing of 1999. Just much bigger in size and scale.

When I invest in a start-up [or company], it doesn’t go into the profit-and-loss statement because it isn’t an expense, and it doesn’t impact my free cash flow. But when the start-up buys compute from Google or AWS, that is revenue for those hyperscalers. The numbers aren’t being looked at with enough skepticism.

In this example, the “clean” free cash flow is a lot less. The impact is to inflate margins, which is boosting earnings for the significant part of the S&P 500 . Capex gets amortized over 6-10 years but revenue mostly outside of land gets booked within 12 months.

The unwind of this will likely be very painful: Markets need capex growth for this bullishness to continue. If Google or AWS say we love AI but don’t see the need for massive growth in capex, the AI theme would be in trouble.

Beyond stocks, there’s the societal impact of spending over a trillion dollars within just two years on a nascent technology and chips that are pretty much useless within three years, taking that investment away from other promising technologies.

Beyond the circular investing, what else is of concern?
Look at restricted stock unit or stock-based compensation as a percent of revenue, which also contributes to the over-earning aspect. Why hasn’t that gone down given their increased size?

Companies adjust the stock-based compensation from their P&L but add it to the free cash flow. That makes those cash flows look better. And if stock prices don’t continue to rise, companies will be forced to pay their employees additional compensation in cash.

But these accounting practices aren’t new. Why does it matter now and impact the trajectory of AI spending?
In the past, free cash flow was so strong it didn’t matter, especially when capex was growing moderately. Now it matters because [free cash flow] is funding the AI capex. To compound this problem, cloud is a less attractive business than before because it is mature.

About 80% of large enterprises in the U.S. have two or more cloud providers. Enterprises can shift incremental load from one provider to another for pricing—and there’s a massive price war brewing in cloud computing. GPU rental pricing in the third quarter alone was down by 20%.

Sell-side analysts have mistakenly assumed that pricing is stable and this is an oligopolistic market. The chief executive of CoreWeave said their job is to cut prices and gain market share. Oracle is undercutting other players by 40% + and gaining share because they claim they have a more cost-effective cloud product.

Oracle earnings barely grew when reported last month. Revenue growth was only 12%, hardly blistering, and this is the best of times. AWS had 700 basis-point margin deterioration last quarter. That is huge!

But now they also must compete against dozens of neoclouds, [or specialized cloud providers]. Margins will only go lower as supply increases to all players. Anyone who has covered telecoms should appreciate what is going on here.

Google CEO Sundar Pichai said AI is more profound than fire or electricity. Why are you skeptical about AI’s utility?
Look, I am hardly a permabear! We were investors in these companies not that long ago. But the more you look at the numbers, the more you see a problem. Look at the internet: We knew it was useful even in 1997-1998. Cisco estimated the internet economy was generating $300 plus billion in revenue in 1999. We have $25 billion in total revenue right now from AI.

AI isn’t usable for mission-critical purposes. We talk to dozens of large companies every month and almost none have suggested they are planning to spend meaningful amounts on AI. Everyone sees it as a glorified tool.

Sure, it will be disruptive in some areas—like assisting in code development in technology or assisting in underwriting in insurance. But if AI goes away, it isn’t going to change my life.

Thanks, Rajiv.

Corrections & Amplifications: In 1999, Cisco Systems estimated that the internet economy was generating more than $300 billion of revenue. An earlier version of this article incorrectly said the company reported that much revenue for that year.

Write to Reshma Kapadia at reshma.kapadia@barrons.com