The volume of funds directed towards new technologies has reached an unprecedented level, surpassing the amounts spent during the technological revolutions associated with railroads or the internet. According to a PwC forecast, global spending on data centers alone could exceed $30 trillion by 2050, which is almost comparable to the nominal value of outstanding US Treasury bonds. PwC notes that these expenditures 'surpass' the spending of the railroad boom or dotcom era, even after accounting for inflation.
Anthropic, one of the leading companies in the AI race, plans to invest $518 billion in the coming years—an amount more than a hundred times its 2025 revenue, according to an IPO prospectus shared with Reuters. Proponents of the company argue that AI technology will be more transformative than the advent of steam engines and the industrialization they spurred.
Nevertheless, economists point out that behind the impressive forecasts and massive spending by AI companies, as well as high valuations, lie assumptions about massive productivity growth and future profits that are not yet supported by sufficient evidence or historical precedent.
In August, JPMorgan noted that widespread productivity growth in the US, which leads the AI race, 'remains elusive,' casting doubt on the sustainability of current AI valuations. A Bain & Company study showed that productivity gains from existing markets would be insufficient to justify current costs, and 'entirely new markets must emerge to close the funding gap.' These new markets could include the use of AI-driven robots or the development of new materials for batteries and semiconductors.
Bain also reported that American hyperscalers such as Google, Amazon, and Microsoft, along with other participants in the AI race, must find over $4.2 trillion in new revenue over the next five years to fund infrastructure expansion.
When do tech booms end?
According to a study published last month, the key question is whether 'applications will arrive in time to recoup these costs.' Few doubt AI's potential to transform all aspects—from office work to research labs, much like past revolutions reduced travel time from days to hours or connected the world via keyboard.
The mathematical calculation of investment returns or loan maturities remains more constant, forcing economists to analyze the consequences for the global economy beyond the ups and downs of investment cycles. JPMorgan writes that historical experience shows that tech booms often conclude when infrastructure development ceases to yield sufficient profit.
Citing Nvidia, an American company whose chips are the foundation of the AI revolution, JPMorgan calculated that to justify its valuation, productivity growth in the US must be 3–5% annually over the next 10 years. This is a significant increase compared to the baseline expectation of the U.S. Congressional Budget Office of 1.75% annual productivity growth over this period.
According to Columbia University economist Stijn Van Nieuwerburgh, for the US alone—which accounts for about three-quarters of global AI investment, according to some estimates—investments will reach approximately $9 trillion from 2025 to 2032, equivalent to spending 3.2% of US GDP annually. He predicts that by 2032, the US AI sector should generate around $3.55 trillion in annual revenue to ensure a 10% return on investment, whereas it currently contributes only a small fraction of that amount.
In his October review conference, Van Nieuwerburgh noted that the leverage structure of most debt financing AI infrastructure means that 'a relatively modest deterioration in demand, delays, or asset costs could lead to much larger losses.'
The speed of miracles
These staggering figures have not deterred US AI leaders, who speak of the changes occurring with otherworldly enthusiasm. Anthropic CEO Dario Amodei stated that the future of AI could become a 'thing of transcendent beauty,' while OpenAI CEO Sam Altman said that 'the speed of achieving new wonders will be enormous,' as models learn to improve themselves and accelerate breakthroughs.
Jasjeet Sekhon, Director of Strategy at Google DeepMind, announced at the University of California, Berkeley summit in August that this self-learning, known as recursive self-improvement, is a 'key part of the investment hypothesis' and could deliver unprecedented productivity growth if achieved.
Although recursive self-improvement potentially could lead to exponential AI development, it has also raised concerns about existential risks to humanity. Nevertheless, productivity growth rates may prove slower than corporate finance departments assume.
Diane Cole, an economist from the University of Cambridge, noted that the impact of past revolutionary technologies usually took 10 to 50 years to fully materialize. The Anthropic economics team modeled several scenarios for the additional annual growth AI might provide in 2030. Based on a 2% baseline in a non-AI environment, they projected growth of 2.4% in a moderate AI impact scenario, 5.4% in a substantial scenario, and 15.4% in an extreme scenario. They also indicated that higher growth would lead to the loss of more jobs, without assigning probabilities to any outcome.
Amodei predicted last year that AI could eliminate half of all low-level office jobs within five years. However, some researchers argue that its impact has currently been limited to complicating job searches for those seeking office employment. Studies in the US and UK have shown a slowdown in hiring entry-level specialists in office roles where AI excels, even if overall employment remains high.
Researchers from Stanford University reported in August that employment for workers aged 22–25 in AI-affected roles, such as accountants and paralegals, was 19% lower than in jobs that AI found difficult to replicate, such as cleaners and construction workers. Nevertheless, even if the promised transformation takes longer than the figures surrounding AI companies suggest, the real economic benefit should persist, just as trains kept running after the 1873 panic bankrupted railroad magnates, and the internet did not shut down after the dotcom crash in the 1990s.
Cole concludes: 'History is our friend in trying to understand this. As long as the infrastructure needed to support all future productivity effects remains, it is fine.' - Steven Eisenhower.