To pop or not to pop: the smart way to invest in AI without worrying about the bubble

You either love or hate it, but you can’t ignore it: AI has helped drive the stock market to higher highs amid elevated interest rates and recessionary fears. That’s pretty impressive.

So far this year, the Mag Seven is up by over 34%. But after such a strong run, it’s normal to wonder whether the sector is too rich.

It’s also common to draw parallels between the dot-com bubble and today’s hype, leading investors to wonder if there’s an AI bubble that’s about to pop, too. Goldman Sachs’ big AI headline of the month is “To buy, or not to buy, that is the question.” The note from September 5, led by Peter Oppenheimer, suggests the answer is “to buy” but also to diversify.

Strong fundamentals have supported Big Tech’s rising prices, with global earnings per share for the sector up by 400% relative to the broad market at 25%, the note read. But Goldman’s team also acknowledged that concentration risk is real and new entrants could create greater competition, suggesting it’s time to broaden your exposure and consider the smaller technology players and dividend payers. Such rebalancing would improve risk-adjusted returns.

The bear case around GenAI says hundreds of billions of dollars in capital expenditure are being spent on infrastructure but very little in the way of explicit revenue, says John Belton, a portfolio manager for growth equities at Gabelli Funds. Still, he doesn’t believe the negative outlook will affect the sector’s trajectory long-term.

The current case against the bubble is that GPUs are being purchased and deployed immediately, says Brian Colello, an equity strategist at Morningstar Equity Research. This is unlike the internet bubble, when a lot of infrastructure was built for demand still to come.

However, fear and uncertainty will cause a lot of volatility, which investors need to accept. So, if you plan to ride it out, you should buckle up with some stern conviction.

A smart way to AI

Nancy Tengler, the CIO at Laffer Tengler Investments, managed money during the internet boom of the 1990s and sees many of the same parallels, but that’s not necessarily bad.

Today’s tech darlings aren’t at peak earnings or as expensive as internet companies were then. Nvidia may seem hyped up, but at 35 times estimated earnings, it’s still a catch relative to its projected growth, Tengler said. She noted that on a forward-looking basis, it is expected to grow at 288% this year, 120% next year, and 41% in 2026.

Companies in a fast growth phase like this can often be at 60, 70, or 80 times their estimated earnings, Tengler noted. The aggregate 24-month forward P/E of the Mag Seven is at about 24 with a net income margin over the next 12 months estimated at 28%. In comparison, in 2000, the top seven, which included some big names still around today like Microsoft, Intel, Oracle, and IBM had a 24 month forward P/E of 52 with a net income margin of only 16% during the last 12 months of the tech bubble, according to Goldman’s note.

She acknowledged that the GPU maker’s revenue is decelerating, but that doesn’t mean it’ll hit the stock’s price. However, that’s the mantra that’s being used by Wall Street’s less experienced, she said. On the up side, Nvidia could push through by improving margins, she noted.

The big players also have big customers.

“Management teams at these companies aren’t going to just build willy-nilly for fun,” Tengler said while emphasizing that one company’s capital expenditure budget is another company’s revenue source. She pointed to Oracle as an example: the cloud company’s co-founder Larry Ellison projected growth from operating 162 data centers to 2,000. The outcome means continued demand for chips from Nvidia, Broadcom, and AMD.

But here’s the catch: Nvidia’s projected decelerating sales means it’s not part of Tengler’s 12 best ideas list, which rests on a five to seven-year low turnover strategy. Instead, she substituted it with the steady-dividend payer Broadcom (AVGO), or what Tengler calls “the poor man’s Nvidia”.

She advises investors who want to continue buying Nvidia to view volatility as their friend and add to their position during pullbacks.

Additionally, since a big part of Nvidia’s revenue comes from a handful of buyers, investors can use its customers’ projected or slowed spending as a leading indicator of the GPU maker’s revenues, Colello said.

It’s difficult to paint the entire sector with one Brush, says Belton. The AI value chain is the framework he likes to use as a differentiator. It begins with the base, or what he calls the microlayer, where computing infrastructure such as chips and processors are made. The second layer would be data center infrastructure providers and suppliers, including power management companies. And the third is the application providers, which are the companies creating services for end users to harness AI.

“It’s pretty clear to me that at the base layer, particularly the chip level, the results for those companies are very real,” Belton said. “And I think you have to take it case by case, and there’s going to be winners and losers like everything else. But for the winners, I don’t think we are in a bubble.”

For a company like Nvidia, most of its business isn’t coming from generative AI. It comes from machine learning or big data workloads that various companies and governments use, Belton noted.

The base layer also has an oligopoly made of a few semiconductor capital equipment companies such as ASML and Applied Materials, which chipmakers must turn to, Belton said. Cadence Design Systems is another key player in this part of the value chain because it’s part of a duopoly as one of the two major design automation software providers for the semiconductor space, he added.

Belton also has a positive conviction for the second layer, where data centers are built and supplied because there’s already rising demand. His firm’s holdings include big public cloud platform providers like Microsoft, Google, and Amazon’s AWS.

His growth funds also have exposure to the industrial companies servicing the data centers, including the power suppliers and electrical components companies such as Eaton Corporation, a preferred vendor for large hyperscale data centers.

But not all energy stocks that got a lift from the AI theme are worthy. Their services could be commoditized, which creates competition, Belton noted.

Sometimes a company’s growing AI-based revenue means a decline in its core business model, Colello said. He pointed to AMD as an example. It has sold traditional CPUs to cloud companies for over two decades. Recently, many of their large customers have shifted to AI servers. So, an AMD GPU in an AI server is a win. But that shift also means its CPUs aren’t being bought, he said.

Finally, as you get further out of Belton’s AI-value chain, where the application providers are, the risk increases, here, there has been less direct business and revenue to point to.

“I think we’re still waiting for some generative AI-specific killer use cases to commercialize,” Belton said. “And I think that’s where I have the least conviction.”