AI vs. Math: Will Massive AI Investments Pay Off in Time?
Global data center spending could hit $30 trillion by 2050. But economists warn productivity gains aren't showing up yet - and the math doesn't add up.
👁 123 reads
Global spending on data centers alone could exceed $30 trillion by 2050, according to a PwC analysis cited by Reuters. That’s comparable to the value of all outstanding US Treasuries and, as PwC emphasizes, “exceeds what the railway boom and dot-com bubble consumed combined, even after adjusting for inflation.” But the history of those previous booms isn’t just about technological triumph. It’s also a graveyard of investors who bet on infrastructure before revenues emerged to justify it.
Numbers That Are Hard to Grasp
To understand the scale, it’s worth looking at some concrete figures. Anthropic, one of the companies in the AI race though far from the revenue leader, disclosed plans in its IPO prospectus to spend $518 billion in the coming years. That’s over 100 times its 2025 revenues. The company’s backers compare AI’s potential to the steam revolution and the industrialization it triggered. The enthusiasm is understandable: Anthropic, OpenAI, Google DeepMind, and Meta AI are working on technologies that could genuinely transform everything from office work to research laboratories.
The problem is that between vision and balance sheet lies a chasm that’s very hard to fill with optimism alone. The United States accounts for roughly three-quarters of global AI investment, so macroeconomic data from there provides the most important test. And that test is currently failing.
Productivity That Doesn’t Show Up in Statistics
JP Morgan wrote bluntly in August: broad productivity gains in the US “remain elusive,” undermining the ability to sustain current AI industry valuations. This isn’t the opinion of marginal skeptics. It’s the conclusion of one of the world’s largest banks, which has analyzed technology investment cycles for years.
“Historical precedents suggest that technology-driven booms often end when infrastructure buildout stops delivering sufficient returns,” JP Morgan wrote.
Bain & Company went even further. The firm calculated that productivity gains generated by existing markets won’t be enough to justify current spending. “Entirely new markets” must emerge - Bain lists AI-driven robots and development of new materials for batteries and semiconductors among them. More importantly, Bain estimated that hyperscalers - Google, Amazon, Microsoft, and others building global infrastructure - must collectively generate over $4.2 trillion in new revenue in just five years to finance continued expansion. As the report published last month summarized: “The question is whether applications will emerge early enough to pay for it.”
Nvidia as a Litmus Test
JP Morgan used Nvidia as an example - the company whose chips form the backbone of the entire AI revolution - to show what would need to happen for current valuations to make sense. The answer: annual US productivity growth would need to reach 3 to 5 percent for the next 10 years. For comparison, baseline forecasts from the Congressional Budget Office (CBO) assume productivity growth of 1.75 percent annually over the same period. The difference between these numbers isn’t just a statistical discrepancy; it’s a question of whether Nvidia’s valuation is rooted in reality or wishful thinking.
It’s worth noting that Nvidia’s stock has dropped and rebounded multiple times in the past two years, reacting to every market signal - from Chinese export regulations to quarterly results. The very fact that one company can be treated as a proxy for an entire technological revolution says something important about the structure of this boom: capital has concentrated in infrastructure before anyone knows who will be using it and for what.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →Is This a Bubble or Still Construction?
The answer to that question depends on which historical analogy you use in your analysis. The 19th-century railway boom consumed enormous capital, brought numerous bankruptcies - and changed the world. The dot-com bubble wiped out hundreds of companies - and gave us Google, Amazon, and the modern internet. Both revolutions involved a phase where infrastructure was built faster than revenues could justify it.
The difference lies in pace and scale. $30 trillion spent on data centers is something that can’t be compared to any previous technology cycle. At that scale, even partial disappointment - say, productivity growth of 2 percent instead of 4 - could mean valuation corrections felt across the entire financial market, not just the tech sector.
The race continues. Money flows. Mathematics patiently waits for its turn.
More on this from QUANTSLANT
guides·What Is a Webhook? Connect AI to Forms, Email & SpreadsheetsA webhook is a URL where one app automatically sends data to another when something happens.