Washington – A recent paper presented at a Brookings Institution conference highlights the massive scale of the United States’ artificial intelligence (AI) infrastructure expansion. Professor Stijn Van Nieuwerburgh of Columbia Business School estimates the build‑out will require roughly 3.6% of gross domestic product each year through 2032 – more than $10 trillion – surpassing historic projects such as the rollout of electricity, railroads, interstate highways and the internet.
Financing becomes increasingly intricate
The study notes that early AI spending was largely funded from the cash reserves of tech giants like Amazon, Meta and Alphabet’s Google. Today, the effort relies heavily on external financing, involving banks, private‑credit lenders, real‑estate firms and a host of special‑purpose vehicles. Van Nieuwerburgh describes the web of arrangements as “freaking complicated,” drawing a parallel to the opaque structures that contributed to the subprime mortgage crisis.
Potential systemic risk
Because the financing is highly leveraged and the revenue streams from AI applications remain untested, the author warns of “meaningful downside risk” should demand fall short of expectations. He points out that to justify the projected $10 trillion investment, the AI industry would need to generate about $3.7 trillion in annual revenue by 2032 – a figure that would require roughly 80% yearly growth from the combined revenues of OpenAI and Anthropic today.
Local concerns and broader economic impact
Municipalities across the country are already voicing worries about the strain on local resources, such as water, power and land use, as data‑center construction accelerates. Federal Reserve officials are also monitoring the boom for possible inflationary pressure. While some AI executives suggest a slower rollout could mitigate risk, the study emphasizes that the sector’s rapid growth, high leverage, and uncertain demand create a fragile financial environment.
What the data shows
Compared with historic technology rollouts, the AI build‑out’s projected share of GDP eclipses the 2.2% annual share railroads occupied in the late 1800s, and is more than three times the roughly 1% annual share of the interstate highway system in the 1950s and the telecommunications expansion of the 1990s. The paper estimates the United States will need about 183 gigawatts of new data‑center capacity over the next seven years, up from the current 57 gigawatts.
Van Nieuwerburgh concludes that while strong growth in AI applications could generate stable cash flows, the combination of uncertain demand, rapid technological change, execution bottlenecks and high leverage makes the sector vulnerable to a significant correction.
Original reporting: Appleton, WI News Feed (HLL/CB) — read the source article.
