US AI Investment Surpasses Railway Boom in Historic Scale — What It Means for Korea
New projections put US AI infrastructure spending at $10.3 trillion through 2032, dwarfing every postwar capital boom. The question isn't whether the money will materialize — it's who captures it.
The numbers behind the noise
The United States is about to spend more on a single technology stack than it did on railroads at the height of the Gilded Age. That is not a metaphor. According to a Brookings Institution estimate cited by the Wall Street Journal on September 24, AI-related infrastructure investment — data centers, power plants, grid upgrades, cooling systems — will total $10.3 trillion between 2025 and 2032. That works out to an average of 3.63 percent of annual US GDP over that period.
To put that in historical context: the railroad construction boom of 1870 to 1890, widely regarded as America’s first true infrastructure supercycle, consumed the equivalent of 2.24 percent of GDP per year. The interstate highway era (1956–1973) came in at 1.13 percent. The dot-com fiber rollout peaked around 1.10 percent. The AI buildout, on current trajectories, is 60 percent larger than the railway era as a share of the economy.
Goldman Sachs projects that even this year alone — before the bulk of the spend kicks in — US AI infrastructure investment will hit roughly 1.9 percent of GDP. That would be the highest level since the railroad boom, a WSJ assessment notes, and likely unmatched by any single industrial sector in US history.
This is not speculative tech spend. It is concrete.
The distinction matters. Much of the recent AI coverage has focused on software applications, LLM licensing deals, and speculative valuations. The reality on the ground is measured in megawatts and cubic meters of concrete.
Between January and July this year alone, US private-sector spending on data center construction reached $37 billion, up roughly $9 billion from the same period last year, according to Commerce Department figures. LinkedIn data shows more than 750,000 new AI-related jobs have appeared in the United States since 2023, with median salaries hovering around $180,000 — more than double the national median.
The asset-side impact is already visible. Federal Reserve data shows US household holdings of stocks and mutual funds reached $63 trillion in the second quarter, nearly double the level at the end of 2022. The AI boom is not just inflating equity portfolios; it is physically reshaping the American landscape, one data center at a time.
The Korean angle nobody in New York is talking about
Here is what the English-language financial press has largely overlooked: this investment wave will not be absorbed by US suppliers alone. The capital is American, but the equipment and components are not.
Korea stands to benefit disproportionately across at least three supply chain layers. First, semiconductor memory and advanced packaging — Samsung Electronics and SK Hynix dominate the high-bandwidth memory (HBM) market that AI accelerators require. Every additional GPU cluster means more HBM orders, and every HBM order flows through Korean fabs. Second, power equipment. Data centers are hungry. A single large facility can draw 100 to 300 megawatts — enough to power tens of thousands of homes. Korean firms like Hyundai Electromechanical and LS Electric are already expanding their electrical switchgear and transformer production lines to serve the US market. Third, semiconductor manufacturing equipment. Companies like Samsung SDI and SK Materials supply precursors and specialty gases that feed the fabrication process, while equipment makers stand to gain from expanded US chip plant construction driven by the CHIPS Act and private investment alike.
None of this was guaranteed. The AI buildout could have been captured entirely by domestic US suppliers, as the 19th-century railroad boom largely was. The difference is that chip manufacturing, power grid components, and precision equipment are globally traded goods with concentrated supplier bases — and Korea sits firmly within those bases.
The risks are real but mispriced
The Brookings estimate itself flags two concerns that deserve more attention than they are getting.
Inflation is the first. When an economy dedicates over 3 percent of GDP to a single infrastructure push, demand for copper, steel, labor, and land spikes. Construction costs around data centers have already risen sharply in markets like Northern Virginia and Ohio. Energy prices face upward pressure as utilities race to expand capacity. If the Fed was just beginning to feel confident about the inflation trajectory before this wave hit, that confidence is now under stress.
Financial stability is the second. A significant portion of this investment is being financed through debt — corporate bonds, project finance, utility borrowing. If AI infrastructure fails to generate sufficient revenue to service that debt, the losses will not stay contained within tech companies. They will flow through the banking system, pension funds, and insurance portfolios that underwrite them. The warning from economists at Brookings and elsewhere is not that this will definitely happen, but that the scale of leverage relative to uncertain returns is a structural risk, not a temporary one.
What happens next
The most likely path is a period of elevated capex that outpaces revenue growth for several years, followed by a shakeout. Some data center operators will succeed; many will not. The ones that survive will be those with secured power contracts, off-take agreements with cloud providers, and access to cheap capital. The ones that fail will leave behind abandoned construction sites and written-down assets.
For Korea, the strategic question is timing. The window to capture share in power equipment, semiconductor materials, and HBM supply is open now — but it will narrow if US policymakers decide that energy security and industrial policy require more domestic sourcing. Both the Biden and incoming administrations have signaled interest in onshoring critical supply chains. Korean exporters will need to decide whether to invest directly in US-based production capacity or rely on export-driven relationships that could be disrupted by political shifts.
The railroad boom built a continent. This buildout will build something else entirely — a computing backbone that may define the next era of economic productivity, or a costly misallocation of capital that leaves inflation and debt in its wake. The scale is historic either way. The only question is whether Korean industry is positioned to ride it, or watch from the sidelines.