Trending
The US artificial intelligence buildout could require more economic resources than previous major infrastructure and technology expansions, according to a new Brookings study.
The study estimates that investment in AI data centres, power systems, networking infrastructure, specialised chips and other equipment could reach $10.3 trillion between 2025 and 2032.
That would equal an average of about 3.63% of US gross domestic product each year, according to economist Stijn Van Nieuwerburgh of Columbia University.
The projected investment would exceed the economic share absorbed by major US infrastructure booms involving canals, railroads, electrification, highways and telecommunications.
Van Nieuwerburgh presented the analysis in a paper prepared for the Brookings Papers on Economic Activity conference.
Major technology companies initially financed much of the AI expansion through their own cash reserves.
However, the scale of investment has increasingly pushed companies toward outside financing.
The Brookings study says AI infrastructure funding is moving into more complex arrangements involving joint ventures, private credit, securitisation, special-purpose vehicles, leases and loan guarantees.
Paolini Leads Italy Into Billie Jean King Cup Semi-Finals
These structures can spread financial exposure across technology companies, banks, private lenders, property developers and other investors.
Van Nieuwerburgh warned that such arrangements could make financial risks harder to identify before a potential downturn.
The study does not conclude that the AI sector currently poses systemic financial risks comparable to earlier credit booms. Instead, it calls for greater transparency as the industry’s financing structure develops.
The investment boom reflects the rapid expansion of data-centre infrastructure required to support AI services.
Van Nieuwerburgh estimates that the US could add about 183 gigawatts of new data-centre capacity over the next seven years.
That compares with roughly 57 gigawatts of existing capacity cited in the study.
The expansion will require large amounts of electricity, specialised equipment and supporting infrastructure.
Consequently, data-centre construction has become an increasingly important issue for utilities, policymakers and local communities.
The scale of the investment also depends on whether AI companies can generate enough revenue to support the infrastructure spending.
Van Nieuwerburgh estimates that the industry would need about $3.7 trillion in annual revenue by 2032 to achieve the expected return on investment.
He notes that current combined annual revenue estimates for OpenAI and Anthropic stand at roughly $100 billion.
That would require revenue growth of about 80% a year to reach the projected level, according to the study.
However, the analysis also identifies conditions that could support the investment.
Strong demand for AI applications, high data-centre utilisation and continued improvements in AI models could generate the cash flows needed to support the infrastructure.
The study compares the current AI expansion with earlier technology investment booms that experienced periods of rapid growth followed by financial stress.
Railroad investment in the late 19th century absorbed about 2.2% of US GDP annually, according to the analysis.
By comparison, the construction of the US interstate highway system and the telecommunications expansion that began in the 1990s each accounted for slightly more than 1% of annual GDP.
The AI buildout is projected to require a significantly larger share of the economy.
Van Nieuwerburgh said the combination of uncertain demand, rapid technological change, construction bottlenecks and higher leverage creates potential downside risks.
He also highlighted the difficulty of tracking exposures created through private financing structures.
The financial concerns come as the US technology sector races to build data centres and secure additional electricity supplies.
The Brookings study notes that AI infrastructure depends on continued access to power, hardware and reliable tenants.
Meanwhile, the rapid pace of construction has created concerns over electricity demand and the ability of utilities to expand capacity quickly enough.
Brookings has previously highlighted the challenge posed by rising data-centre electricity demand and the financial risks that could emerge if expected AI demand fails to materialise.
For now, the study does not predict an imminent financial crisis.
Instead, it argues that policymakers and investors need better information about the increasingly complex financial arrangements supporting the AI buildout.
As AI investment accelerates, the ability of companies to convert infrastructure spending into sustainable revenue will remain a key test for the industry.
[…] AI Buildout Could Consume $10.3 Trillion in US Investment by 2032 […]