The roof of a data centre in Frankfurt, Germany, on Aug. 22.Michael Probst/The Associated Press
I’ve been writing a lot about what’s abnormal in the investing landscape. After 40-plus years following markets, there are things that jump out at me as being extreme, irrational, or just plain silly.
I’m referring to themes and products that are far off trend or disconnected from basic financial fundamentals.
Some irregularities play out at the fringes (meme stocks; SPACs) but are good indicators of investor sentiment and risk taking, while others have a meaningful impact on the economy and capital markets. My topic here, the amount of capital being spent building data centres, is in the latter category.
It’s a phenomenon that can’t be ignored by any investor. It’s unprecedented and will be an important factor for future returns.
Economic impact
The need for data centres has been growing with the shift to cloud computing, but the demands of artificial intelligence are taking it to a new level.
The numbers are staggering. McKinsey & Co. says, “by 2030, companies will invest almost US$7-trillion in capital expenditures on data-centre infrastructure globally.” In 2025, spending on data centres contributed more to economic growth than consumer spending, the main engine of the U.S. economy.
I’ve seen capex booms before in mining, energy, real estate and automobiles, but nothing like this.
Many unknowns
There’s a long list of questions to answer in determining whether this spending boom will prove to be a new order for capital allocation or just the mother of all cyclical surges.
Demand: Will demand be as high as anticipated? Will the shift of emphasis from training AI models to inference, the process by which AI systems respond to user requests, has an impact on demand and the type of chips required? As I understand it, training is more computationally demanding but inference brings with it huge volumes.
Bottom line: Will companies be able to monetize their models? The efficacy of AI isn’t in question but the pathway to profitability is less clear. It will require that users generate enough incremental revenues and savings to justify the cost.
The big valley: Assuming an affirmative on the previous question, how big will the gap be between spending and profits? There will likely be a valley to cross, and history suggests that innovation leaders don’t always make it to the other side. Think back to the fibre-optic developers (Worldcom and 360 Networks), and the railroads a century before them. The mega-tech companies will navigate it, but there are other players in the ecosystem that aren’t as sturdy. They’ve become too dependent on the trend and/or are using debt liberally to increase capacity.
The losers: We don’t know who the winners will be. Several companies may be on the podium, or it will be winner take all. Either way, there will be losers, which prompts further questions. What happens to the capacity being used to train their models? How will it be absorbed, and at what price?
There’s more.
Who will benefit: The economy benefited hugely from the fibre-optic buildout and the railroad boom. The rising stock market tells us that AI will have a similar effect. What we don’t know is who will benefit the most. Will it be the providers who are putting the infrastructure in place or the end users?
Power: AI is moving at lightning speed. Power generation not so much (especially as the development of renewables is being discouraged in the U.S.). McKinsey points out that US$3-trillion of the total spend will go toward peripheral investments such as power infrastructure and real estate. Can power also grow at an unprecedented rate, and if not, how will it affect the spending cycle?
Obsolescence: There’s a chance that the power-hungry chips of today will be usurped by more efficient versions tomorrow. We can’t expect the innovation cycle for semiconductors to stop. Will the current crop of data centres become the equivalent of high-cost mines and oil fields?
The hangover
After every binge, there’s a morning after. If the hangover is severe, it will reach far beyond the tech darlings of today. The skewing of capital spending at the macro level have an impact on allocation decisions at companies, decisions that hinge on answers to the questions above. Many companies’ success, including Canada’s Celestica, is tightly linked to the data-centre boom.
There are always unanswered questions in investing. As Bob Hager, my former partner, used to say, “If you wait for certainty, you’ll miss the market.”
What’s so interesting with this uncertainty is the size of the ante. It’s a big bet and one that investors must keep an eye on.
Tom Bradley is a portfolio manager with Purpose Investments, co-founder of Steadyhand Investment Management, a member of the Investment Industry Hall of Fame and a champion of timeless investment principles.