Currently, the volume of capital directed towards artificial intelligence (AI) has never been so large, surpassing the investments made in past technological revolutions, such as railways and the internet. This scale of investment raises questions about the time required for such applications to generate the expected return.
According to a PwC projection cited by Reuters, global spending on data centers could exceed US$ 30 trillion (approximately R$ 159 trillion) by 2050. This amount approaches the stock of United States Treasury bonds and, according to the consultancy, surpasses investments made during the expansion periods of railways and dot-com companies, even after inflation correction.
Anthropic, one of the central companies in this competitive scenario, plans to invest US$ 518 billion (about R$ 2.75 trillion) in the coming years, as detailed in its public offering prospectus analyzed by Reuters. This amount represents more than one hundred times the revenue the company recorded in 2025.
Proponents of these large investments argue that AI will bring an economic impact superior to the transformations caused by the steam engine and industrialization. However, much of this expectation is based on future productivity gains and profits, for which concrete proof and historical precedents guaranteeing the arrival of results at the necessary pace are still lacking.
JPMorgan signaled in August that broad productivity gains in the United States, the leader of the AI race, remain difficult to find, creating uncertainty about the sustainability of the sector's valuations. A Bain & Company study reached a similar conclusion, indicating that current market productivity gains would not be sufficient to justify existing investment levels; to cover this gap, entirely new markets would be needed.
Possibilities for these new markets include the use of AI-operated robots and the development of innovative materials for semiconductors and batteries. Bain points out that American companies responsible for AI infrastructure—such as Google, Amazon, and Microsoft—and other participants in the race would need to generate over US$ 4.2 trillion (about R$ 22.3 trillion) in new revenue over the next five years to finance the expansion. The central question is whether the applications capable of generating such revenue will arrive in time to repay the already built infrastructure.
JPMorgan also warned that historically, technological expansion cycles tend to end when infrastructure investments stop yielding adequate returns. To achieve a 10% return on this investment, the US AI sector would have to generate about US$ 3.55 trillion (approximately R$ 18.8 trillion) annually by 2032, a figure that is currently only a small fraction.
Additionally, the use of leveraged structures to finance part of the infrastructure increases risks. One economist pointed out that a modest decrease in demand, delays, or asset value drops could result in much larger losses.
Future Prospects Despite the Numbers
Despite the presented data, industry executives continue to describe a future full of profound changes. Dario Amodei, CEO of Anthropic, has stated that an AI-driven future could be something of 'transcendent beauty.' Sam Altman, CEO of OpenAI, has also stated that the speed of new discoveries provided by the technology will be immense as models learn to improve themselves.
Jasjeet Sekhon, Director of Strategy at Google DeepMind, considers what is called recursive self-improvement as a crucial element in the AI investment thesis. If achieved, this process has the potential to accelerate system advancements and generate unprecedented productivity gains. However, this same possibility is linked to concerns about existential risks to humanity.
Even if AI manages to restructure the economy, the factor of time may be the main obstacle. Diane Coyle, an economist at the University of Cambridge, observed that the productivity impact of previous revolutionary technologies generally took between ten and fifty years to disseminate throughout the economy.
Anthropic's economic team simulated several scenarios for additional AI growth in 2030. Starting from a projected growth of 2% in a no-AI scenario, growth would rise to 2.4% in a moderate impact hypothesis, 5.4% in a substantial impact scenario, and 15.4% in an extreme situation. The study also indicated that higher growth would result in greater job losses, although it does not assign probabilities to the scenarios.
Amodei had predicted the previous year that AI could eliminate half of entry-level administrative positions within five years. However, recent research suggests that the impact seems more related to the difficulty faced by newcomers to the labor market in securing office positions. Studies conducted in the United States and the United Kingdom showed a slowdown in hiring entry-level professionals for administrative roles where AI has greater capability.
Researchers at Stanford University reported in August that employment of young people aged 22 to 25 in areas affected by AI, such as accounting and legal assistance, was 19% below the record in occupations that are more difficult for technology to replicate, such as construction and cleaning.
Nevertheless, even if the promised transformation takes longer than the numbers from AI companies indicate, infrastructure investments can continue to generate long-term economic benefits. History provides examples of this: railways continued operating after the financial panic of 1873, which led many railway companies into bankruptcy, and the internet did not disappear after the dot-com bubble burst in the 1990s. The infrastructure established in these cycles remained available to support the productivity gains that came later.