The race to build artificial intelligence infrastructure is attracting unprecedented investment, but economists warn that future productivity gains and revenues may struggle to justify the enormous costs.
The global race to dominate artificial intelligence is entering an expensive new phase, with trillions of dollars being committed to data centres, advanced computer chips and computing infrastructure, raising questions about whether the technology can generate enough economic value to repay the investment.
A projection by PwC suggests cumulative global spending on data centres alone could exceed $30 trillion by 2050, a figure approaching the value of outstanding US government Treasury securities.
The scale of the projected investment highlights the extraordinary financial expectations surrounding artificial intelligence, which technology companies increasingly present as a force capable of transforming economies, businesses and labour markets.
However, beneath the optimism lies a growing economic question: will the productivity gains and commercial opportunities created by AI be large enough to justify the money being spent?
The concerns are becoming more pronounced as technology companies commit increasingly large sums to infrastructure whose financial returns may take years to materialise.
Anthropic, one of the leading companies developing advanced AI models, plans to spend $518 billion in the coming years, according to figures contained in an initial public offering prospectus reviewed by Reuters. The projected expenditure is more than 100 times the company’s 2025 revenue.
The commitments reflect expectations that artificial intelligence will eventually create entirely new industries and significantly increase economic productivity.
But economists say those expectations remain difficult to verify.
JPMorgan said in August that broad-based productivity improvements in the United States, the world’s leading AI investment market, remained elusive, raising questions about whether current company valuations can be sustained.
The concern is that investment is moving much faster than the economic benefits needed to support it.
The revenue challenge
The financial challenge extends beyond individual AI developers to the large technology companies building the infrastructure that supports the industry.
Google, Amazon and Microsoft are among the major companies investing heavily in data centres, computing capacity and related technologies to support the growing demand for AI services.
A Bain & Company study found that existing markets were unlikely to generate sufficient productivity gains and revenue to justify the industry’s current spending commitments.
It estimated that US hyperscalers and other companies involved in the AI race would need to generate more than $4.2 trillion in additional revenue over five years to finance the infrastructure expansion.
That would require artificial intelligence to create new commercial markets on a scale that has yet to be demonstrated.
Potential opportunities include AI-powered robotics, advanced manufacturing and the development of new materials for batteries and semiconductors.
The study warned that entirely new markets would have to emerge to close the gap between expected investment and potential returns.
For investors, the timing is critical. Infrastructure spending creates immediate financial obligations, while the commercial benefits of new technologies may take considerably longer to develop.
A productivity revolution or an investment bubble?
The United States accounts for an estimated three-quarters of global AI investment, making its economic performance particularly important to the industry’s future.
Stijn Van Nieuwerburgh, an economist at Columbia Business School, estimates that US investment in AI could reach approximately $9 trillion between 2025 and 2032.
That would represent an average annual investment equivalent to about 3.2% of US gross domestic product.
For the spending to generate a 10% return on investment, the US AI sector would need to produce approximately $3.55 trillion in annual revenue by 2032, far above its current earnings.
The financial exposure is further complicated by the use of debt to finance much of the infrastructure expansion.
Van Nieuwerburgh has warned that even relatively modest declines in demand, construction delays or falling asset values could produce disproportionately large financial losses.
JPMorgan has also questioned whether the productivity improvements expected from AI can support current valuations.
Using Nvidia, whose processors are central to the AI infrastructure boom, the bank estimated that US productivity growth would need to reach between 3% and 5% annually over the next decade to justify the company’s valuation.
That compares with the Congressional Budget Office’s baseline projection of 1.75% annual productivity growth over the same period.
The difference illustrates how much future economic success investors may already be pricing into AI companies.
The promise of machines that improve themselves
Despite the financial concerns, technology executives remain confident that artificial intelligence could produce economic changes far beyond those associated with previous technological revolutions.
Anthropic chief executive Dario Amodei has described the potential AI future as a “thing of transcendent beauty”.
OpenAI chief executive Sam Altman has similarly predicted an extraordinary acceleration in technological breakthroughs as AI systems become increasingly capable.
One of the most closely watched possibilities is recursive self-improvement, a process through which AI systems could help develop and improve subsequent generations of AI technology.
Jasjeet Sekhon, chief strategy officer at Google DeepMind, described the concept as a key part of the industry’s investment case during a summit at the University of California, Berkeley, in August.
If achieved at scale, the process could accelerate scientific research, product development and economic productivity.
However, it also raises concerns about the risks associated with increasingly autonomous and powerful AI systems.
For now, the economic consequences remain uncertain.
Anthropic’s economic researchers have modelled several possible scenarios for AI’s contribution to annual economic growth by 2030.
Starting from an assumed baseline growth rate of 2% without AI, their projections range from 2.4% under a modest impact scenario to 5.4% under a substantial impact scenario and 15.4% in an extreme scenario.
The company also acknowledged that faster AI-driven growth could result in significant job displacement.
Workers face an uncertain future
The debate over AI investment is increasingly extending into the labour market, particularly as businesses adopt systems capable of performing tasks traditionally handled by office workers.
Amodei predicted last year that artificial intelligence could eliminate half of all entry-level white-collar jobs within five years.
Although widespread job losses have not materialised at that scale, early signs of pressure are emerging among younger workers seeking employment in professions exposed to automation.
Research from Stanford University published in August found that employment among workers aged 22 to 25 in AI-exposed occupations, including accounting and paralegal work, was 19% lower than in occupations considered less vulnerable to AI, such as construction and cleaning.
Separate studies in the United States and Britain have also identified slower recruitment of early-career workers into some office-based positions involving tasks that AI can increasingly perform.
The findings suggest that the effects of artificial intelligence may become visible in hiring patterns before they translate into widespread unemployment.
Nevertheless, economists caution against assuming that every technological advance will immediately produce higher productivity or eliminate existing jobs.
Diane Coyle, an economist at the University of Cambridge, said the economic benefits of major technological breakthroughs have historically taken between 10 and 50 years to become fully visible.
Why an AI downturn may not end the technology
The enormous sums being invested in AI have revived comparisons with earlier infrastructure and technology booms, particularly the expansion of railways in the 19th century and the dotcom bubble of the late 1990s.
Both periods attracted substantial investment based on expectations of long-term economic transformation.
Both also experienced financial corrections when investors’ expectations exceeded the immediate commercial returns.
Yet the underlying technologies survived.
Railways continued operating after the financial crisis that devastated railway investors in the 19th century, while the internet remained central to the global economy after the collapse of many dotcom companies.
The same distinction could prove important for artificial intelligence.
Even if some AI companies fail, valuations decline or infrastructure investment slows, the data centres, computing systems and technological advances already developed could continue supporting future economic activity.
Coyle argues that the survival of essential infrastructure can make the longer-term economic consequences of a technology boom more important than the fortunes of individual investors.
The challenge for the AI industry, therefore, is not simply to demonstrate that artificial intelligence works or that it can transform society.
It must also prove that the economic value it creates can justify the extraordinary cost of building it.
