A statistical line was crossed this spring that captures the sheer scale of the artificial intelligence buildout better than almost any earnings call or chip launch. According to Commerce Department construction figures analyzed by MarketWatch, private-sector spending on building data centers has now surpassed public spending on transportation-related structures, a category that includes airports, marine terminals, and mass transit systems. For the first time in the history of the data series, the computers that train and run AI models are commanding more construction capital than the roads, rails, and runways that move people and goods.
The milestone is not a rounding error or a one-month anomaly. Data center construction reached a record annualized rate that recent reporting has placed in the neighborhood of $45 billion to $50 billion, and the segment now accounts for roughly 2.3 percent of all US construction spending. Projections for the full year suggest that figure could climb substantially higher if the current pace holds, with some estimates running well past $100 billion on an annualized basis. The category has gone from a footnote in construction statistics to one of the dominant forces in American commercial building in the span of about two years.
A Fivefold Surge in Two Years
The speed of the shift is what makes it remarkable. Data center construction spending has surged roughly fivefold over the past two years, a rate of expansion that few sectors of the economy have ever matched outside of wartime mobilization. The driver is no mystery. The largest cloud and AI companies, Microsoft, Alphabet, Amazon, Meta, and Oracle, have collectively committed to capital expenditure budgets that now run between $660 billion and $690 billion for 2026, nearly double the prior year’s level. A large and growing share of that money flows directly into physical construction: the concrete shells, the power infrastructure, the cooling systems, and the high-density electrical work required to house tens of thousands of accelerator chips per building.
The construction figures capture only part of the total spending, since the chips themselves, the networking gear, and the servers are counted as equipment rather than structures. But the building component alone has become large enough to reshape the construction industry’s order book. While traditional commercial categories such as office buildings have stagnated or declined, data center projects have been outbidding conventional developments for labor, materials, and prime sites near electrical substations and fiber routes.
Why the Transportation Comparison Matters
The comparison to transportation spending is instructive because public infrastructure outlays are normally the benchmark against which large private investments are measured. Government spending on transportation structures reflects deliberate, budgeted national priorities: keeping airports functional, maintaining transit systems, and building the physical backbone of commerce. That a handful of private technology companies, pursuing a commercial bet on artificial intelligence, are now collectively pouring more into building computing facilities than the public sector spends on this slice of transportation infrastructure says something profound about where the economy believes its future growth will come from.
It also raises questions that economists are actively debating. AI data center spending has grown at least tenfold since 2022 and now approaches an estimated 2 percent of total US gross domestic product on its own. By some calculations, hyperscaler capital expenditure budgets are on track to represent more than 2 percent of GDP, a share large enough to materially affect headline economic growth figures. When a single category of private investment becomes large enough to move national accounts, its trajectory becomes a macroeconomic story rather than merely a corporate one.
The Bull Case and the Bear Case
Supporters of the buildout argue that the spending is the rational response to a genuine technological transition, and that the companies making these bets are among the most cash-rich and disciplined enterprises in history. They point out that previous waves of infrastructure investment, from the railroads of the nineteenth century to the fiber-optic buildout of the late 1990s, ultimately created enormous productive capacity even when individual investors were burned along the way. In this telling, the data centers being built today are the foundation for a decade or more of AI-driven productivity gains, and the construction surge is simply the visible tip of a structural shift.
Skeptics counter that the pace of spending has outrun the pace of revenue. The direct revenue generated by AI services across the largest providers remains a small fraction of what is being spent to build the infrastructure, and no prior technology cycle has required this much capital this quickly. The bears worry that if AI monetization disappoints, the result could be a glut of expensive, rapidly depreciating computing capacity and a sharp pullback in construction that would ripple through the broader economy. The same concentration that makes the buildout powerful on the way up makes it dangerous on the way down. Our analysis of whether the AI capital spending boom is a bubble or a smart bet lays out how the hyperscalers are now devoting the overwhelming majority of their operating cash flow to this infrastructure rather than returning it to shareholders.
The Power Problem Underneath
What the construction figures do not fully capture is the energy challenge that accompanies every new facility. Data centers are extraordinarily power-hungry, and the buildout has triggered a parallel scramble to secure electricity. Utilities are planning enormous capital programs of their own to add generation and transmission capacity, and the competition for power has become one of the binding constraints on how fast the AI infrastructure can actually grow. This dynamic has driven a wave of consolidation and investment in the power sector, including deals such as the multibillion-dollar utility acquisition aimed at locking up AI power capacity that we covered earlier this year.
The geographic implications are significant. Data centers are increasingly sited not where people live but where power is cheap and abundant, reshaping regional economies and electricity markets. The buildout is also going global, with sovereign wealth and national governments racing to establish their own AI infrastructure footprints, a trend visible in the Gulf states’ ambitions to become regional AI hubs even amid regional instability.
What Investors Should Take From the Milestone
For investors, the construction milestone is a useful reality check on just how committed the largest technology companies are to the AI thesis. Spending of this magnitude is not easily reversed; the capital is being sunk into long-lived physical assets, and the companies have tied their competitive futures to the bet. That commitment cuts both ways. It provides a powerful tailwind for the entire supply chain, from chipmakers to electrical contractors to power producers, and it raises the stakes on whether the eventual revenue materializes to justify the outlay.
The practical takeaway is that the AI buildout has moved beyond the realm of software and silicon into the physical economy, where it is now competing with and surpassing traditional infrastructure for capital. Whether that proves to be the foundation of a productivity supercycle or the setup for a painful correction will depend on monetization that has not yet arrived at the scale the spending implies. Either way, the data centers being poured today are a concrete, literally, expression of the largest corporate bet of the decade.
Frequently Asked Questions
What milestone did data center construction just reach?
For the first time, private-sector spending on building data centers has surpassed public spending on transportation-related structures, a category that includes airports, marine terminals, and mass transit. Data centers now account for roughly 2.3 percent of all US construction spending.
How fast has data center construction spending grown?
It has surged roughly fivefold over the past two years and reached a record annualized rate in the range of $45 billion to $50 billion, with full-year projections running substantially higher if the current pace holds.
Who is driving the spending?
The largest cloud and AI providers, Microsoft, Alphabet, Amazon, Meta, and Oracle, have committed to combined capital expenditure budgets of roughly $660 billion to $690 billion for 2026, nearly double the prior year, with a large share flowing into physical construction.
How big is AI data center spending relative to the economy?
AI data center spending has grown at least tenfold since 2022 and now approaches an estimated 2 percent of US gross domestic product on its own, large enough to materially affect national economic growth figures.
What is the main constraint on further growth?
Electricity. Data centers are extremely power-hungry, and securing generation and transmission capacity has become a binding constraint, triggering massive utility investment and competition for power.
What is the risk if AI revenue disappoints?
Because spending has outrun revenue, a shortfall in AI monetization could leave a glut of expensive, rapidly depreciating computing capacity and trigger a sharp construction pullback that would ripple through the broader economy.
