Many conversations about markets seem to land on the same question these days.

Is all of this spending on AI a bubble?

While we cannot answer it with certainty, we can frame the question in a way that helps us learn something about market history and how other periods with big spending on new technology turned out.

The idea, which we’ll call the “Rule of 25,” is simple to state.

Historically, investment booms tied to transformational technologies have not turned genuinely dangerous until cumulative spending reaches roughly a quarter of gross domestic product (GDP).¹

Below that threshold, the story has tended to be “too much, too fast” rather than “the end.”

Three Historical Booms, One Pattern

It’s possible to look at three past instances:

• The railroad boom of the 1860s, which financed a genuinely transformational build-out of American infrastructure and preceded the Panic of 1873.
• The electrification boom of the 1920s, when utilities and industrial firms poured capital into a new source of power before the crash of 1929.
• The telecom and internet boom of the late 1990s, when fiber-optic buildout and dot-com enthusiasm both climbed toward the same rough ceiling before the 2000–2002 unwind.

In each case, cumulative investment as a share of the economy climbed toward that same neighborhood before the cycle turned. Importantly, this is a historical observation, not a law of economics. We believe, however, that the recurrence of a similar ceiling across three very different technologies, a century and a half apart, is the kind of pattern-matching that deserves attention.

Where AI Actually Stands Today

This is the part of the conversation that usually gets skipped, because it requires an actual denominator. Barron’s own reporting puts total AI-related capital spending at roughly $6 trillion across the leading players over just a handful of years, with global hyperscaler capex projected at $3.7 trillion through 2029. Set against a $32 trillion U.S. economy,² even the more generous cumulative figures put AI investment well short of the 25% threshold that preceded prior turns. That doesn’t mean nothing can go wrong between now and then; it simply means that if this pattern holds, the spending itself has room to run before it becomes the historical trigger point.

The Part That Matters More Than the Threshold

Here is the detail that is arguably more useful than the 25% figure itself.

None of these three prior cycles ended on their own.

Each needed an outside shock to turn, something like a Fed rate hike, a credit event and in one case something as unrelated as the 1906 San Francisco earthquake shaking loose the financing behind the railroad-adjacent boom of that era. Valuations alone did not do the work of ending the cycle.

That possibly reframes the useful question for investors.

• It may be less “how close are we to some spending ceiling”
• It may be more “what would have to happen—a rate shock, a credit-market seizure, a financing partner stumbling—to turn sentiment here.”

It is true that increasingly complex arrangements, such as Nvidia’s financing commitments to OpenAI and special-purpose vehicles like Meta’s Beignet Investor structure carrying meaningful debt alongside equity, are exactly the kind of financing complexity that has made prior cycles potentially more fragile when a shock did arrive.

What Survives the Bust

The other point worth carrying into any dinner-table AI conversation could be this.

When these cycles have turned, the infrastructure itself did not disappear.

• Track was still track after 1873.
• The electrical grid still worked after 1929.
• The fiber laid during the telecom bust became the backbone that carried the next two decades of internet growth.

What gets built tends to remain, but it’s the financial claims on top of it that reset. The lag between “the market decides this was overbuilt” and “the country stops using what was built” has historically been measured in years, not forever.

A Framework, Not a Forecast

None of this tells you where AI-related stocks go tomorrow, next week or next month, and it shouldn’t be read as an argument that nothing can go wrong. What it offers is a more grounded way to frame the question. Cumulative AI investment remains a fraction of the threshold that has historically preceded genuine breaks in past technology cycles. The more productive thing to watch may not be a spending ceiling but the financing structure underneath the spending, and whether a shock, rather than a valuation reckoning, is what eventually tests it.

Sitting at the start of September 2026:

• The WisdomTree Artificial Intelligence & Innovation Fund (WTAI) is currently positioned with different physical ‘AI Bottlenecks’ in mind. A ‘bottleneck’ in this context could be something like memory or optical networking components. There is a whole complex facility being built and, without this much smaller number of very specific components, currently in shortage or backlog, the rest of it cannot function.
• The WisdomTree U.S. Quality Growth Fund (QGRW) is currently positioned in many of the largest, most high-quality companies driving a significant portion of the capital spending to increase the compute resources for AI.

We think the historical perspective on past periods of big spending on new technologies is useful for helping investors gauge their forward-looking expectations on strategies such as these.

¹ Source for the historical perspective of this piece, unless otherwise noted. This article also draws on thinking from Raymond James Strategist Tavis McCourt: Root, A. (2026, August 27). Is the AI capex bubble about to burst? What 250 years of market history tell us. Barron’s.

² Source: U.S. Bureau of Economic Analysis. (2026, August 26). Gross domestic product (second estimate), 2nd quarter 2026 [News release]. U.S. Department of Commerce.