Why the bursting of the AI bubble would be a much more worrisome event than many people think

Open this photo in gallery:

When it comes to the AI boom there are two questions that come to mind. The first is, will it burst? The second and more interesting question, however, is, will the sectoral event eventually turn out to be a systemic event?

If the sectoral event does not scare you, the systemic event will knock your socks off.

To determine the possibility that the bubble will burst, let’s examine what history is telling us when it comes to industries that went through boom-bust cycles, and draw similarities with the AI-related boom. It’s based on a newsletter by Highwood Value Partners, a Whistler-based independent portfolio manager, that provided some excellent historical perspective.

The current rush for AI-related infrastructure build resembles the railroad boom of the 1850s, as well as the internet/broadband boom of the late 1990s. Just like these previous periods, the AI build is financed by shareholders of public companies as well as bond investors and private credit.

The rationale for AI computer capacity (i.e. storage, memory and overall performance) is like the railroad expansion and the belief that if we build it, customers will come. Or that “supply creates its own demand,” which is the core principle of Say’s Law of Markets. It states that the aggregate production of goods and services in an economy generates the exact amount of income needed to purchase all of them. But a few years after all the capital expenditures for the railroad infrastructure had been spent, investors realized that the return on their investment would not meet what they had been promised. As a result, a quarter of all railroads in the United States and the United Kingdom ended up in receivership.

Prosperity's Path: AI data centres are the future. Canada must overcome the backlash

The story repeats in the case of the internet boom of the 1990s. The expectation was that internet traffic would double every 100 days. The companies that built the infrastructure and provided the picks and shovels of the boom were among the top 10 most valuable companies, such as Microsoft, General Electric, Cisco, Intel, NTT, Lucent and Nokia.

Like the railroad builders, they did not make a satisfactory return on their investment. While the internet has had a great impact on society, these companies were terrible investments over the medium haul. In aggregate, they delivered a 60 per cent negative return over the following 10 years.

In 1999, these seven companies plus BP and Exxon made up about 26 per cent of the S&P 500 index. Today, they account for about 36 per cent. The hyperscalers – Alphabet, Amazon, Microsoft, Meta and Oracle – have financed a lot of their capital expenditures with huge amounts of debt, namely about US$570-billion in 2026. A lot of capital is a bet on expectation of huge returns going forward.

What is worse is that hyperscalers and model developers like OpenAI and Anthropic are linked. The former are major owners of the latter two who are supposed to pay the hyperscalers for compute to rent (i.e. graphics processing unit (electronic circuit) rental services for AI and machine learning projects). But as my last article (co-authored with economist Christian Dippel) indicated, OpenAI and Anthropic are financially fragile: single-product companies selling a commoditizing product into a market with free open-source models, financed by recurring capital needs that assume a profitability they have not yet shown. These are the most likely companies to go to zero, and they are the ones looking to soon go public. This will circle back to the hyperscalers.

Thomson Reuters to roll out in-house AI model as quarterly revenue rises

Now, could this sectoral event turn into a systemic event?

In short, yes. The AI-related boom and its possible burst could cause widespread damage to the world economy.

Here are the key fault lines, with a hat tip to a recent paper by YuvB Strategic Value Partners, which influenced a lot of my thinking.

• Unlike previous crises, the U.S. government balance sheet is severely handicapped by the unprecedented scale of federal debt. Meanwhile China, whose balance sheet was stellar in 2008-09, is now in recession (i.e. China’s property sector has been cut in half since 2021, investment in housing has dropped from 12.3 per cent of GDP in 2020 to 6.1 per cent in 2025, and all this while debt-to-GDP rose to 287 per cent in 2025 from 163 per cent in 2008).
• The Fed cannot cut interest rates aggressively without accelerating ongoing dollar reserve erosion, which will make it more difficult for the U.S. to borrow in its own currency and may cause balance of payments issues in the face of persistent current account deficits.
• The U.S. government, unable to raise taxes (without suppressing economic growth) or cut spending (without eliminating core entitlement programs that are embedded in social contracts), may inflate the system. But this will lead to higher interest rates and worsen debt-related service costs (already second to social security).
• Foreign demand for U.S. Treasuries is already retreating, and the bursting of the bubble will make the retreat faster with a predictable effect on interest rates.

And so, in light of all these policy constraints, a sectoral collapse – like the bursting of the AI bubble followed by a collapse of capital expenditures and wealth destruction – can easily turn into a systemic event.

Bottom line: fasten your seat belt and brace for turbulence.

George Athanassakos is a professor of finance and holds the Ben Graham Chair in Value Investing at the Ivey Business School. His latest book is “Value Investing: From Theory to Practice.”