While the capital flow towards the artificial intelligence sector surpasses the expenditures seen in the railways or internet periods, economists point to the mismatch between the return schedule of investments and productivity increases in the field.
In the face of companies' multibillion-dollar spending plans and high company valuations, it seems that the expected productivity jump has not yet spread to the general public.
According to the projections of PwC, one of the world's leading accounting firms, global expenditures on data centers are expected to exceed 30 trillion dollars by 2050.
This amount represents a size close to the current total value of U.S. Treasury securities. It is noted that even when adjusted for inflation, these estimates exceed expenditures during the railroad and dotcom eras.
Anthropic, one of the major actors in the industry, aims to spend $518 billion in the coming years, according to its IPO prospectus. This amount corresponds to more than 100 times the company's revenue in 2025.
While some argue that the technology will create a greater transformation than steam engines, economists emphasize that there is not enough evidence to support future profit assumptions.
According to the news of Reuters agency, US-based investment bank JP Morgan finds the sustainability of valuations doubtful, stating that broad-based productivity increases "have not yet been achieved" in the USA, which is the leader of the race.
A study prepared by Bain & Company shows that the efficiency gains from existing markets will not be enough to cover these expenses.
According to the report, to close the financing gap, entirely new markets must emerge, ranging from artificial intelligence-driven robots to the development of new batteries and semiconductor materials.
Bain & Company calculates that technology giants such as Google, Amazon and Microsoft, and other industry players building global infrastructure, will need to generate more than $4.2 trillion in new revenue over the next five years to finance infrastructure construction.
Reminding that historical examples have shown that technology-focused rises ended with insufficient returns on infrastructure investments, JP Morgan made a calculation based on chip manufacturer Nvidia.
The bank noted that annual productivity growth in the U.S. would need to be between 3 percent and 5 percent over the next 10 years to justify the company's current market value.
The US Congressional Budget Office's basic annual productivity increase expectation for the same period is 1.75 percent.
Columbia Business School economist Stijn Van Nieuwerburgh calculated that expenditures between 2025 and 2032 in the USA, which undertakes approximately three-quarters of global artificial intelligence investments, could reach 9 trillion dollars. This amount corresponds to 3.2 percent of the annual US gross domestic product.
Speaking to Reuters, Van Nieuwerburgh stated that for a 10 percent return on investment, the sector must produce annual revenue of $3.55 trillion by 2032, and that current revenues constitute a very small part of this.
The economist also noted that a moderate deterioration in demand could lead to large losses due to the leverage structure in infrastructure debt.
Industry leaders continue to emphasize the potential of technology.
While Anthropic manager Dario Amodei argued that the future will bring an extraordinary transformation, OpenAI CEO Sam Altman noted that new inventions will accelerate with self-developing models.
Google DeepMind Strategy Director Jasjeet Sekhon also stated that self-training of models can provide unprecedented efficiency.
On the other hand, economist Diane Coyle from Cambridge University reminded that it usually took 10 to 50 years for the productivity effects of past revolutions to be reflected in the economy.
Anthropic, on the other hand, modeled different scenarios regarding the impact of artificial intelligence on growth in 2030. Compared to the base growth assumption of 2 percent, it was calculated that the growth could be 2.4 percent in the moderate scenario, 5.4 percent in the comprehensive scenario and 15.4 percent in the extreme scenario. The study noted that high growth would mean more employment losses.
In an assessment last year, Amodei predicted that artificial intelligence could eliminate half of entry-level white-collar jobs within five years. Research shared by Stanford University shows that the employment of people aged 22-25 in professions open to artificial intelligence, such as accountants and legal assistants, remains 19 percent lower than in fields such as construction work or cleaning.
Economists estimate that a similar economic legacy will remain, reminding us that despite the collapse of railroad companies in the Panic of 1873 or the bursting of the dotcom bubble in the 1990s, the infrastructure established is permanent.
