WisdomTree Research Chief: AI Boom Isn’t a Bubble – But the Stakes Are Rising

TLDR

• AI infrastructure earnings surged 54% in Q2, while demand and order backlogs remain strong.
• Chris Gannatti says today’s AI boom does not resemble the dot-com bubble on valuations.
• Rising debt and a delayed productivity payoff could increase risks as AI spending accelerates.

The AI infrastructure boom is delivering extraordinary earnings growth and showing little sign of running out of steam. But the enormous scale of spending – and increasing use of debt to finance it – could leave markets vulnerable if demand and productivity gains take longer than expected to materialize.

That’s according to Chris Gannatti, global head of research at asset manager WisdomTree. He said the current AI buildout does not resemble the dot-com bubble on traditional valuation measures. But history suggests that major technology investment cycles rarely proceed without significant market corrections.

“It does not feel like a bubble from a perspective of valuation,” he said in an interview with The AI Innovator. “But it is 100% true that every time the world and human society has had one of these new technologies, be it the railroad, electricity, the internet, mobile computing – all these things relate to people getting overly excited.”

Historically, that excitement eventually encounters a catalyst that brings a “rather severe pullback in market indices,” Gannatti said. Such corrections can take years to recover.

Today’s potential vulnerability is unusually important because AI infrastructure companies have become a major engine of U.S. corporate profit growth. WisdomTree research found that AI infrastructure companies – including hyperscalers and companies benefiting from their capital spending – contributed approximately half of the S&P 500’s second-quarter earnings growth year-over-year.

While the broader market itself was robust – the median S&P 500 company posted 14% earnings growth year-over-year – AI infrastructure Q2 earnings grew at 54%, according to WisdomTree. That gives a relatively narrow group of companies an outsized influence on overall profit growth.

Is there enough demand?

For now, Gannatti sees strong evidence that demand is real.

Data-center suppliers across chips, memory, networking and other components have accumulated order backlogs, he said. Unlike some of the speculative infrastructure spending during the dot-com era, much of today’s AI capacity is being built amid demand that companies say already exists.

“There are just all these different backlogs where we know the order is there, the demand is there, the commitment is made, and now the company has to physically make the thing and actually deliver it,” Gannatti said.

That distinction matters when comparing the AI boom with the internet bubble.

During the dot-com era, companies poured money into fiber-optic networks based partly on expectations that demand would eventually arrive. When adoption lagged expectations and debt payments came due, many companies failed.

Today, Gannatti said, AI infrastructure companies argue that they could use substantially more computing capacity if it were available. Supply constraints rather than an absence of customers remain a major limitation.

Valuations also look different. Gannatti said an expensive stock today might trade at roughly 30 to 40 times earnings, compared with triple-digit price-to-earnings ratios for some companies during the technology bubble.

WisdomTree senior economist Jeremy Siegel, Wharton finance professor emeritus who famously called the dot-com market top in March 2000, said  that markets today are “not anywhere near” the internet bubble of the late 1990s. Nvidia and the tech giants now spending a fortune on AI infrastructure are ‘real companies. They’re making profits” instead of internet companies with no earnings but sky high valuations.

Waiting for the productivity payoff

Today, however, the central risk is timing.

Companies are investing hundreds of billions of dollars in data centers and related infrastructure on the expectation that AI will ultimately make businesses and workers significantly more productive. But those gains are not yet clearly visible across economic productivity statistics.

“It’s not yet true that you can go across the board for entire economies and say, look at the AI-driven productivity stats, they’re just through the roof,” Gannatti said.

That does not mean the productivity gains will not materialize, he said. Previous general-purpose technologies, including personal computers and smartphones, took time to reshape businesses and the broader economy.

He quoted Nobel Laureate and MIT professor Robert Solow’s famous 1987 line about the productivity paradox: “You can see the computer age everywhere but in the productivity statistics.”

WisdomTree’s core belief is that AI eventually delivers meaningful productivity gains. “That’s our base case,” Gannatti said. “That ultimately, the productivity gains show up.”

The risk is what happens if the payoff arrives more slowly than the financial obligations created by the infrastructure boom.

Large technology companies are spending an increasing proportion of their free cash flow on AI infrastructure and beginning to use more debt and other financing vehicles, Gannatti said. (For example, Google posted $5.9 billion in negative free cash flow in the second quarter, reportedly for the first time in company history.) That raises risks if expected growth fails to materialize.

“As long as you can make the payments on schedule, everybody’s happy,” he said. “If you miss a payment, then that is another source of risk.”

Nvidia sits at the center

Nvidia’s central role in the AI economy makes its results particularly important to watch.

Gannatti said Nvidia has become something resembling a financial hub for parts of the emerging “neocloud” industry, helping these newer AI compute providers get financing so they can buy Nvidia chips and build infrastructure.

That creates a potential concentration of risk if data-center construction slows. Local and political resistance to new data centers due to issues over power use, noise and other disturbances could become constraints.

“Data center construction is a key source of demand for Nvidia chips. So if we’re slowing down data center construction, if that becomes the reality, the demand for Nvidia chips should slow down,” Gannatti said, unless it is able to further diversify its revenue stream to China and elsewhere.

For now, Nvidia and other infrastructure suppliers continue to report backlogs, providing Gannatti with another indicator of current demand.

But Gannatti said Nvidia’s importance means an earnings disappointment could carry implications well beyond one company.

“The longer that Nvidia functions as sort of the center of the whole story, the more important it becomes,” he said.

Memory and other constraints

The infrastructure bottleneck is also shifting.

Gannatti identified memory as the most serious current constraint because rising AI computing demand is colliding with increasing memory requirements across smartphones, computers, vehicles and other electronics.

“Every Nvidia system is using more memory, every new laptop and cell phone is using more memory, and we see the price effect of that,” he said.

Higher memory costs could eventually affect products outside the AI industry. If memory shortages push up prices for cars, televisions and other electronics, Gannatti said the increases could eventually appear in inflation numbers.

That would add another economic dimension to the AI infrastructure boom: The same investment expected eventually to raise productivity could create shorter-term pricing pressure as AI competes with consumer electronics and other industries for components.

What could trigger a correction?

Gannatti said history suggests that large infrastructure booms can make markets more sensitive to shocks that might otherwise be manageable.

Unexpected interest-rate increases, geopolitical conflicts, financial-system stress or another unforeseen event could spark a correction at a time when companies are investing at a historically large scale. According to PwC, data center capital spending is forecast to rise to $800 billion per year in 2026, $1.1 trillion in 2030 and $1.8 trillion annually in 2050.

But a correction would not necessarily mean the underlying technology thesis had failed.

The dot-com crash destroyed companies and investor wealth, but the internet ultimately transformed the economy. Gannatti sees a similar distinction between the short-term market cycle surrounding AI and the technology’s potential long-term economic impact.

For now, he said, the evidence remains stronger on the healthy side of that equation.

“What we see today is on the healthier side, as opposed to a side where you know it’s time to get really concerned and get really worried,” Gannatti said.

The indicators he is watching include hyperscaler spending, corporate revenue growth, infrastructure order backlogs and eventually greater financial disclosure from major AI model companies as they go public.