Even some of the most ardent boosters of Artificial Intelligence (A.I.) concede that the current amount of money rushing into the sector could constitute an investment bubble—meaning that firms developing A.I. are “being valued far beyond their revenues or proven business models.” Open AI CEO Sam Altman has admitted that investors in A.I. may be “overexcited,” with some investors likely to get “very burnt.” Amazon founder Jeff Bezos stated the AI market was an “industrial bubble” where stock prices were “disconnected from the fundamentals” of their businesses. More recently, the Bank of England has warned of a “sharp correction” in the value of major tech companies on growing fears of an A.I. bubble.
The massive data centre construction boom fueled by A.I. investment has been characterized as building the material ‘backbone’ of the digital economy. So what would happen to this backbone should the A.I. Bubble burst?
When the financial media does contemplate the consequences of a major market correction, they usually focus on the impacts to the stock market and the major tech firms funding A.I. But the massive investment in data centre construction entangles far more stakeholders than companies like Google or Meta alone. Regions have staked their economic development on it, politicians have risked their re-election, utilities have bet their growth on it, and some communities have bet their very future. Given the stakes, and the very real possibility that current levels of investment are unsustainable, we should have at least some idea of what is at risk, and what—if anything—we can do to protect ourselves and our communities.
By any measure, the amount of capital pouring into data centre projects is “staggering.” As Wired magazine describes it, “rarely, if ever has a single technology absorbed this much money this quickly.” And while some of the most profitable firms on the planet like Amazon, Google and Meta are driving this investment, the amounts are so large that even major tech firms are financing these investments via debt.
Morgan Stanley predicts $250 billion to $300 billion of debt issuance in 2026 for the hyperscale tech giants alone. And yet, despite these massive levels of investment, the A.I. industry has yet to demonstrate any real, sustained profitability. Instead, firms are investing billions in building out A.I infrastructure without the revenue streams required to offset their costs, gambling that greater AI mass adoption (and, finally, profitability) will arrive sooner rather than later.
It is this mismatch between investments and returns that make the popping of the AI investment bubble a virtual certainty. The only question is when and how far-reaching its impacts will be. An obvious historical analogue is the 2000’s dot com crisis, when exuberant investors so overbuilt the infrastructure of the early internet that it vastly outstripped demand. The result was miles of unused fibreoptic cables and abandoned data centres as the expectations of future internet use far outstripped the reality.
While a financial crisis in its own right, the damage from the dot com bust primarily affected the tech sector itself. The fear this time is that due to the major tech firms’ use of ‘circular’ accounting practices, coupled with the financialization and securitization of vast amounts of data centre debt, the end of the AI bubble would be more akin to the 2008 financial crisis, with effects much more far-ranging and potentially devastating to the wider economy.
No matter how it plays out, any significant market correction would directly impact existing and proposed data centre projects. Without sufficient revenues and the end of investment in the sector, the ability of tenants large and small to make continued lease payments to data centre providers would cease, quickly driving many of these facilities into insolvency.
Due to the special-purpose design of data centres, it is not an asset that can be easily used for other purposes, making them all the more difficult to lease or sell for any other use. The residents of Vaudreuil-Dorion, Quebec witnessed the abandonment of a $1.3 billion Ericsson-owned data centre a mere ten months after its construction in 2016. As Patrick Brodie and Julia Velkova write of the abandoned site;
As of 2020, the building remains mostly vacant, sold to a US real estate company and locally administered by a Montréal firm. After active local political mobilisation, government tax breaks, and years of anticipation of local jobs and prosperity, the inert site now employs only a skeleton crew of security and maintenance professionals tasked with not letting the impressive structure fall to ruin while waiting for new buyers.
One would imagine that a more contemporary hyperscale data centre, given its size, on-site energy generation and cooling systems would be even more difficult to sell or lease in a post-bubble market. Governments would be on the hook for any sunk costs into developing and servicing these sites in addition to the loss in future tax revenues from a now non-operational business. It is this question of who will pay for these stranded assets post-bubble that is motivating citizens groups across the U.S. to demand greater protection from their government.
Of equal concern would be the costs borne by ratepayers for energy generation, transmission and other electrical upgrades to service a data centre that can no longer pay its bills. Traditionally, these costs would be universalized across all ratepayers over the course of many years. But with utilities spending record amounts to specifically service data centre energy needs, state governments in the U.S. are looking to pre-emptively protect ratepayers from absorbing those costs in the event of a significant market correction.
Minnesota has enacted a package of laws designed to protect consumers from the adverse economic effects of data centers. This includes a statute that requires that data centres pay their full cost of service, with no costs shifted to other ratepayers. In addition, the statute specifically requires that other customers of the public utility are not placed at risk for paying stranded costs associated with the utility serving a data centre. Wisconsin is also entertaining similar regulations that would require data centres to provide a financial security pledge that the utility can access should the data centre not be able to pay for the infrastructure built to serve it.
Another strategy that has been proposed would be the public appropriation of stranded assets, like data centers. As data centres entered various states of insolvency, governments could acquire these assets and their computing power at a discount. Such a strategy could be used to construct a fully public AI model.
As Nathan Sanders and Bruce Schneier argue, an “AI model built and funded by Canada for Canadians, as public infrastructure,”could give Canadians “access to the myriad of benefits from AI without having to depend on the U.S. or other countries. It would mean Canadian universities and public agencies building and operating AI models optimized not for global scale and corporate profit, but for practical use by Canadians.”
Such a vision is not fanciful, Switzerland has already embarked on a ‘pro-social’ large language model built on public infrastructure that uses the country’s computational resources to solve real-world problems and maximize social benefit.
Yet, while other countries urgently plan to mitigate the impacts of a volatile AI economy on its citizens, Canada’s leaders seem oddly detached from these concerns. Both federal and provincial governments seem largely indifferent to the obvious harms this investment boom has caused in the U.S, with astoundingly little discussion of how we might ensure we do not emulate the American experience.
Recognizing that a majority of Canadians are skeptical of the benefits of AI, Canada’s minister of artificial intelligence and digital innovation, Evan Solomon, regularly intones how important building trust is to building Canadian’s acceptance of AI. Failing to protect Canadians through proper regulation and allowing us to replicate the U.S. example that has ignited grassroots opposition to AI data centres across that country will only further alienate Canadians from the supposed promises of AI.
