Columbia professor says AI data center boom will exceed $10 trillion investment

The ongoing $10.3 trillion AI infrastructure and data center buildout represents the largest financial and economic gamble in U.S. history, far outpacing past industrial booms.

Driven by a massive wave of artificial intelligence development, this multi-year spending surge is absorbing vast amounts of capital, labor, and energy, reshaping state policies and the broader macroeconomic landscape.

The artificial intelligence infrastructure boom could dwarf America’s late-19th-century railroad expansion, driven by an estimated $10.3 trillion in capital spending through 2032, according to a new report from a Columbia Business School professor.

The research paper, published by the Brookings Institution and written by economist Stijn Van Nieuwerburgh, projects that annual U.S. investment in data centers, power systems, and specialized chips will average 3.63% of gross domestic product (GDP) between 2025 and 2032.

That annual economic commitment exceeds peak historical spending on major national infrastructure drives, including the railroad boom of 1870–1890 (2.2% of GDP), the interstate highway system in the 1950s (1.1%), the telecommunications and internet boom (1.1%), and early 20th-century electrification (0.5%).

Off-Balance-Sheet Leverage

While AI is often described as a software breakthrough, Van Nieuwerburgh highlights a massive physical buildout that requires novel and increasingly complex financial engineering. Major technology hyperscalers, such as Meta and Microsoft, are rapidly outspending their operating cash flows to secure computing capacity.

To fund this expansion without weakening their credit ratings, companies are shifting capital expenditures off their balance sheets into joint ventures, special-purpose vehicles (SPVs), private credit, and long-term lease agreements.

Examining a case study of Meta’s Hyperion Data Center project in Louisiana, Van Nieuwerburgh detailed how Meta partnered with private credit firm Blue Owl to create an SPV called Beignet. The entity issued $27 billion in debt against $30 billion in assets—a 90% leverage ratio—while keeping the project’s debt off Meta’s corporate balance sheet.

Accounting Gaps and Systemic Risk

The paper notes that accounting standard loopholes create a “probability vacuum” regarding future liabilities. Under current Generally Accepted Accounting Principles (GAAP), commitments like future lease renewals or residual value guarantees don’t require balance-sheet disclosure if individual outcomes fall below a 70% probability threshold, even when one obligation is virtually guaranteed.

While Van Nieuwerburgh stopped short of calling the current financing structure a systemic risk or a speculative bubble, he cautioned that off-balance-sheet structures make it difficult for regulators and investors to measure overall leverage across the sector accurately.

“It’s premature to call this systemic risk, but better measurement and disclosure should come while the industry’s capital structure is still evolving,” Van Nieuwerburgh said of the staggering $10.3 trillion capital layout.

Unlike the 2008 subprime mortgage crisis, the current AI buildout is largely financed with long-term debt backed by investment-grade corporate tenants, reducing immediate maturity-mismatch risks.

However, Van Nieuwerburgh warned that rapid technological obsolescence of graphics processing units (GPUs) and potential drops in long-term AI demand could redistribute risks across private credit funds, institutional investors, and tech balance sheets.