Why AI’s trillion-dollar data centre boom could be the next big bubble

(Representative Image)

The artificial intelligence boom has triggered one of the most expensive infrastructure expansions in modern history. From North Dakota to Texas, vast data centres are rising, packed with billions of dollars’ worth of chips. Top four technology companies have committed more capital to AI infrastructure in three years than it cost over four decades to construct the U.S. interstate highway system, at constant dollars. But the payoff is uncertain since today's AI revenues are still small compared with money being spent, the Wall Street Journal reported.

A risky business model

Firms such as CoreWeave rent data centres, put in Nvidia chips, and sublease computing capacity to AI companies, a model driven by high-interest finance and long-term leases. The financing is risky since customer contracts are usually for just two to five years, while lease commitments take a decade. This creates a mismatch that can end up being expensive if AI adoption flags, reminding us of the overbuilding of fibre optic cables in the 1990s dot-com bubble.

Why investors worry

Experts estimate the current wave of spending at trillions of future AI profits. Bain & Co. recently predicted that AI would need to earn some $2 trillion annually by the year 2030 in order for the spending to be worthwhile, many times more than the total revenues of the tech titans of the present day. To put this in perspective, AI-related revenues in 2024 totalled just about $45 billion, pointing to the gap between current returns and hoped-for possibilities.

Signs of a bubble

Analysts see echoes of past technology bubbles. In the late 1990s, telcos spent over $100 billion laying fibre, only for much of it to remain idle for years. Similarly, AI chips devalue rapidly as new models are designed, with it difficult for companies to make back investment. In the meantime, the roll-out of ChatGPT-5 was regarded as incremental not revolutionary, adding to fears that breakthroughs in generative AI are slowing while costs spiral out of control.

The bull case for AI

Supporters argue the spending spree is different this time. Unlike the 1990s internet boom, today’s tech giants are cash-rich, and AI applications are instantly accessible worldwide. Hundreds of millions already use tools like ChatGPT weekly, and if AI reaches the point of reliably automating white-collar work, the resulting productivity gains could more than cover the costs. Executives like Oracle’s Larry Ellison predict AI could eventually add double-digit growth to global GDP.

The Ellendale, North Dakota view

For towns like Ellendale, where a $15 billion AI data centre is under construction, the boom promises advantage and peril. The town has taken loans to include housing and infrastructure for the workers. If the AI market delivers as expected, Ellendale could be a lively hub. But if demand sags, it will be left with debt and unused buildings—a miniaturized version of the bet the world tech industry is making.

The bottom line

Investment in AI has been at historic highs, with hopes that revenues will finally overtake debt overhangs and overbuilding to close before they bury investors. Whether this is the start of a new industrial revolution or the first seeds of another dot-com bubble collapse will depend on how quickly companies and consumers start paying for AI in bulk. The coming few years will reveal if today's boom gets converted into long-term prosperity or short-term-to-be-painful correction.