AI’s race to transform world before the money runs out

A staggering amount of investment is flowing into artificial intelligence (AI), dwarfing previous technological revolutions such as railways or the internet. According to PwC, cumulative spending globally on data centers could reach $30tn by 2050, almost matching the value of outstanding US Treasuries. Anthropic, one of the leading AI firms, plans to spend $518bn over the coming years, more than 100 times its 2025 revenue.

Backers of AI technology assert it will bring about transformative changes comparable to the advent of steam engines and industrialization.

However, the projections and immense investments by AI companies come with assumptions about vast productivity gains and future profits, which economists question due to a lack of evidence and historical precedent. JP Morgan noted that broad-based productivity gains in the US, the leader in AI, have remained elusive, casting doubt on the sustainability of AI valuations.

To fund the infrastructure expansion, US hyperscalers and other players in the AI race need more than $4.2tn in new revenue over the next five years, according to Bain & Company. The challenge lies in whether the desired applications will materialize in time to make up for the investments. Historically, technology-driven booms often end when infrastructure buildouts no longer generate sufficient returns.

Using Nvidia as an example, JP Morgan estimated that U.S. productivity gains would need to be 3% to 5% annually over the next 10 years to justify its valuation. This is significantly higher than the baseline expectation of 1.75% annual productivity growth for the same period by the US Congressional Budget Office. In the United States alone, investment in AI could reach $9tn from 2025 to 2032, equivalent to 3.2% of US GDP annually, according to Columbia Business School economist Stijn Van Nieuwerburgh.

To earn a 10% return on investment by 2032, the US AI sector would need to generate about $3.55tn in annual revenue, which it currently earns a fraction of. The leveraged structure of much of the debt funding AI infrastructure implies that a modest deterioration in demand, delays, or asset values could lead to much larger losses.

Despite the projections, US AI executives speak with an otherworldly zeal about the potential changes ahead, such as an "AI future" that could be "a thing of transcendent beauty." However, the pace of productivity gains from past revolutionary technologies typically takes 10 to 50 years to materialize. Studies have even shown a slowdown in early career hiring for white-collar positions exposed to AI, despite overall employment remaining strong.