HYPErscalers: Is the AI boom dot-com redux?

During a recent high‑profile financial summit I attended in Hong Kong, the CEO of a leading asset management firm, who should have known better, said, “In the context of hyperscalers’ endowment, USD 2.5 trillion [on capex] over five years is not a lot.”

I baulked at the remark. In what parallel universe is USD 2.5 trillion a trivial amount? Answer: one in which Nvidia can put on USD 1 trillion market capitalization in the space of three months.¹

"In what parallel universe is USD 2.5 trillion a trivial amount?"

But how do you earn a return on USD 2.5 trillion of investment when the mid‑ to long‑range path to artificial intelligence (AI) monetization remains unclear? It points to two different AI capex (capital expenditure) booms: a relatively near‑term one that is underwritten by a discernible level of actual demand and a longer‑term speculative one tied to a quasi‑religious belief in exponential growth.

First, let’s consider some eye‑popping numbers. AI spending just hit 1% of U.S. gross domestic product (GDP) in a 1.8% growth economy—that’s over half of all new demand. Nividia alone reached USD 5 trillion in worth at the end of October, or 15% of the entire U.S. economy.² For context, when Cisco peaked in 2000 as the world’s largest company, it was just 5.5% of GDP. While much was made over JP Morgan’s USD 3 billion Manhattan headquarters, USD 40 billion data center projects are being quietly launched in Texas. Nobody seems to care.

Riding the hype wave

So are we in a bubble? There is certainly a bubble of people talking about bubbles—just have a glance at Google Trends if you don’t believe me (Figure 1). Meanwhile, there are those who believe we cannot be in a bubble precisely because we are talking about one. To my mind, there are three keywords in recent market developments that are worth attention: reflexivity, circularity, and debt.

Bubble mania

(Fig. 1) Google searches for “Is the stock market in a bubble?” have surged

As of November 12, 2025.
The chart is based on Google Trends search data, which reflect the relative popularity of search terms over time on a scale of 0–100, with 100
representing the peak popularity of the search term during the period. Google Trends data do not measure actual financial market conditions, investor
behavior, or sentiment directly. The results are influenced by search volume, geographic location, and Google’s proprietary algorithms.
Source: Google Trends.

x

Reflexivity asserts that prices do in fact influence the fundamentals and that these newly influenced sets of fundamentals then proceed to change expectations, thus influencing prices. The process continues in a self‑reinforcing pattern. The excitable market reaction to the recent AI megadeals is an example of just such a feedback loop.

Then there is the circular nature of these deals. Essentially, the providers of compute infrastructure are investing in native AI players that are in an investment phase. In the dot‑com era, this was known as “vendor financing,” and it became something of a profanity.

Finally, there is debt. So far, cash‑rich tech giants have bankrolled this AI frenzy with their own deep pockets and equity finance. But now we’re entering the credit phase—companies are turning to debt markets, often off balance sheet, and borrowing will likely accelerate.

We know that with generative AI and, further down the road, artificial superintelligence, we are looking at a new technological paradigm—and potentially a massive productivity leap. This is all great stuff, and it is easy to understand the temptation to keep riding the lucrative AI wave that began with the launch of ChatGPT in November 2022. Indeed, given current levels of index concentration, it would take a brave investor to get off.

Titans, hubris, and nemesis

An important lesson from history is that when we invent nice things, bubbles happen. But not all bubbles are the same. There are “bad” bubbles (tulips, gold, land), and when bad bubbles are financed by debt, it can pose a systemic risk to the economy. There are also “good” bubbles that create overcapacities in productive assets—think railroads in the gilded age of U.S. expansion between 1870 and 1900, electricity in the early 1900s, and the dot‑com boom of the late 1990s. Ultimately, wealth was created in each case, but the early investors lost a lot of money.

"...when we invent nice things, bubbles happen."

It is too early to predict what kind of bubble the AI capex boom will turn out to be. The long‑term impact will depend on whether today’s massive investments ultimately lay the groundwork for lasting productivity gains, or whether they create excess capacity that fails to deliver sustainable returns. When capacity is growing this fast, the supply and demand mismatch is always likely to flip from shortage to surplus. In the dot-com era, investors lost a lot of money on fiber and switches dug into the ground, and something similar could happen with AI if overbuild ties up vast amounts of capital in facilities that may not operate near full efficiency—or even worse, if the rapidity of the chip cycle renders the compute obsolete.

Complicating matters, the incentives of infrastructure providers (“picks and shovels” players) and AI application developers are not aligned. While infrastructure companies benefit from continued expansion and investment, application developers are focused on improving efficiency and lowering costs. To reverse the Dodo’s comment after the caucus race in Lewis Carroll’s Alice’s Adventures in Wonderland: “Everybody cannot win, and all can’t have prizes.” The optimistic projections for AI infrastructure highlight just how much hype is built into today’s hyperscaler valuations.

There is an irony that data center projects are being named after Greek titans Prometheus and Hyperion. In Greek mythology, nemesis always follows hubris. To throw a bit of Latin into the mix, caveat emptor.