The biggest industrial buildout on American soil isn’t a government megaproject. It’s a private-sector infrastructure wave — and you don’t need to be a venture capitalist to participate.

There is a number that should stop you cold. In 2026, four American technology companies (Microsoft, Meta, Amazon, and Alphabet) will spend roughly $670 billion on capital expenditures, most of it on AI infrastructure. As a share of GDP, that single-year figure exceeds the 1850s railroad expansion. It is roughly five times the Interstate Highway System at its peak, and ten times the Apollo program’s annual GDP share. The only thing in American history that compares is the Louisiana Purchase, and that was a one-time land deal, not a sustained industrial campaign.

This is not hyperbole; the numbers come from The Wall Street Journal and Moody’s, which projects ~$700 billion in total hyperscaler AI spending in 2026, nearly six times the 2022 level. The single largest physical buildout in the world is happening on U.S. soil. And it is, fundamentally, an infrastructure story, not a technology story. That distinction changes who can participate, and how.

Moody’s tracked $387 billion in AI infrastructure spending across the six major hyperscalers in 2025; 2026 is tracking to nearly double. The Big Five (Amazon, Microsoft, Alphabet, Meta, Oracle) have guided to a combined ~$725 billion for 2026, a 64% year-over-year increase. Goldman Sachs sees no plateau: roughly $1.1 trillion in 2027 (above the ~$920 billion consensus it calls too conservative) and more than a trillion by 2028. Data-center construction now accounts for more than 2% of all U.S. construction spending, a share that was effectively zero a decade ago.

It’s bigger than the railroad expansion of the 1850s, the Apollo space program, and the decades-long build-out of the U.S. interstate highway system.

The Wall Street Journal, February 2026

The comparison is instructive because it reveals something structural. The Interstate Highway System cost roughly $500 billion in today’s dollars and took 35 years. The AI buildout will spend more than that in a single year, almost entirely with private capital, without a single congressional appropriation. This is not a government project. It is a private-sector response to a technology shift with no precedent in its speed, scale, or capital intensity.

Roughly three-quarters of global data-center capacity under construction sits in the United States, more than 700 facilities going up across 38 states. It is not patriotism. It is physics, law, and money, in that order. The physics is energy: the U.S. is simultaneously the world’s largest natural-gas producer and operator of the largest nuclear fleet on Earth. The buildout follows the molecules. The law is property rights and contract enforcement: a hyperscaler deploying $200 billion a year needs its 15-year leases and power-purchase agreements honored. The money is capital markets deep enough to finance a multi-trillion-dollar buildout, plus the REIT structure that lets ordinary investors access it. This geography of concentration is not accidental. It is structural, and it will persist.

For a decade before 2022, U.S. electricity demand was flat. Then came the GPU. Goldman Sachs now estimates global data-center power demand will rise 165% by 2030; in the U.S., data centers go from 3% of total power in 2022 to 8% by 2030. Deloitte models a thirtyfold increase in U.S. AI data-center power demand by 2035 (from 4 gigawatts to 123), the equivalent of adding roughly 120 large nuclear reactors’ worth of demand in a country that has brought only two new reactors online in the past decade: Vogtle Units 3 and 4, in 2023 and 2024.

The single biggest constraint is power. I don’t believe we will have fully resolved the capacity we need for the demand we have in a couple of quarters.

Andy Jassy, Amazon CEO

The utility response is historic: 51 investor-owned utilities now plan $1.4 trillion in capital spending over five years, up more than 20% from a year earlier, the largest utility capital cycle in American history. But the bottleneck is not money; it is time. Data centers go up in 18 months. Gas turbines ordered today won’t arrive until the 2030s. Transmission lines take a decade to permit. Northern Virginia, which handles ~35% of U.S. hyperscale cloud traffic, is already at Dominion Energy’s capacity wall. Demand is arriving faster than the infrastructure to power it can be built. For investors, that tension is the opportunity.

Because this is a physical buildout (land, steel, copper, transformers, turbines, cooling), the access points are public and liquid. The most direct exposure is data-center REITs: Equinix (EQIX, ~$107B market cap, 270+ data centers) and Digital Realty (DLR, ~$68B), both legally required to distribute 90% of taxable income, both carrying multi-year contracted backlogs. North American colocation vacancy at year-end 2025 was roughly 1.4%, effectively sold out, giving operators pricing power they haven’t had in a decade. Utility ETFs (XLU, VPU) hold the names executing the $1.4 trillion grid buildout and earn regulated returns on a growing rate base. Semiconductor funds (SMH, ~$72B) capture the highest-margin component: NVIDIA, TSMC, Broadcom, AMD, Micron.

What to avoid: GPU-rental platforms with thin margins and no backlog; pre-revenue AI-cloud startups marketed to retail; highly leveraged single-project developers; and any “AI infrastructure” promotion that cannot produce a GAAP income statement. The supercycle is real. The grift around it is also real, and the two are not the same thing.

The United States is in the early years of its largest private-sector infrastructure buildout since the industrial age began. The capital is committed, the power demand is real and growing, and the assets are physical, owned, and contracted. The vehicles that provide exposure (REITs, utility ETFs, semiconductor funds) are public, liquid, and accessible to anyone with a brokerage account. This is a domestic opportunity. It does not require a passport, a private fund, or a conviction about which AI model wins.

The railroads took 30 years. The highways took 35. This one is moving faster. And for the ordinary investor, the door is open.