Opinion: America's Growth Numbers Are Being Carried by a Few Data Centers

The Commerce Department's Bureau of Economic Analysis reported on October 29 that the U.S. economy grew at an annualized 3.4% in the third quarter, up sharply from 2.2% in the second. Reuters, the Associated Press, Barron's and The New York Times all led with versions of the same headline that day: a brisk, broad-looking acceleration, carried by consumer spending and investment. Taken at face value, it is good news, and Americans are entitled to feel some relief after two years of argument about whether growth was real or borrowed. But the number deserves a second look before anyone treats it as evidence that the economy is healthy in the way a 3.4% print used to imply. Increasingly, this is not economic growth in the old, diversified sense. It is a bet — a very large, very concentrated bet by a small number of technology companies on artificial intelligence infrastructure — and the national accounts are simply recording the size of that bet.

A Number That Flatters More Than It Informs

Private-sector economists who track the GDP components in real time reached a strikingly similar conclusion before the official figure even landed. Brian Wesbury and Robert Stein of First Trust Advisors, forecasting a 3.3% print days ahead of the release, wrote that the headline number "masks extremely rapid growth in AI/data centers and slow growth elsewhere." Their underlying math is startling: they estimated nominal spending on data centers was running at an annualized rate of 139% for the quarter, and that U.S. production of semiconductors and related electronic components grew at a 47% rate — the fastest pace for any quarter in a series going back to the early 1970s.

"Add it all up, and we get a 3.3% annual real GDP growth rate for the third quarter, which masks extremely rapid growth in AI/data centers and slow growth elsewhere." — Brian S. Wesbury and Robert Stein, First Trust Advisors, October 26, 2026

That is not a partisan complaint; it is a description of composition. GDP is an aggregate, and aggregates can rise even when the median experience does not improve, if a few categories grow fast enough. The question an opinion page is entitled to ask is whether that is a trivial accounting curiosity or something policymakers, investors, and voters should actually worry about.

The Soaring Share of One Sector

It is not a new phenomenon, but it is getting more extreme. Economists at the Federal Reserve Bank of St. Louis calculated that AI-related investment categories — information-processing equipment, software, R&D, and data centers — accounted for 39% of real GDP growth in the first three quarters of 2025, compared with 28% for the analogous categories at the peak of the dot-com buildout in 2000. In one quarter of 2025 that the researchers examined, those categories alone generated 30% of the quarter's entire growth. The AI boom, in other words, is now a larger share of what counts as American economic expansion than the internet boom ever was, by this measure, even before adjusting for how much larger the economy has since become.

The physical evidence backs this up. The Census Bureau's monthly survey of construction spending showed private data-center construction running at a record annualized pace of roughly $84.95 billion in August, up more than 73% from a year earlier — now exceeding general office construction by the widest margin in a series that dates to 2014. Over the same period, manufacturing construction, the sector tariff policy was supposed to be reviving, fell nearly 20% year over year. One category is booming; the rest of the industrial base is shrinking. The GDP report nets these out into a single comforting number.

This Could Still Be the Railroad, Not the Tulip

The fair counterargument deserves real weight. Every transformative infrastructure build — railroads, electrification, the long-haul fiber networks laid in the 1990s — looked wildly overbuilt and dangerously concentrated at the time, and most of it eventually proved its worth, even when individual investors or companies were wiped out along the way. ING's chief international economist, James Knightley, has estimated that AI-related investment accounts for roughly a third of U.S. GDP growth this year — a large share, but one he frames as a genuine productivity story rather than a bubble by definition. It is entirely possible that today's data-center buildout is laying down capacity that will look, in a decade, like the prescient infrastructure bet rather than the reckless one. Dismissing it as pure mania would be its own kind of overconfidence.

But the historical parallel cuts both ways. The dot-com-era buildout also produced a recession and a multi-year stock-market collapse once the capital spending slowed, even though the fiber it left behind was eventually used. Concentration risk and long-run value creation are not mutually exclusive; a bet can be correct in the long run and still be destabilizing in the short run for the people whose jobs, pensions, and 401(k)s are tied to it in the meantime.

Why the Concentration Itself Is the Risk

This is where the argument moves from description to judgment. When a large share of headline growth depends on the capital-spending decisions of a handful of companies, the economy's health becomes hostage to those companies' confidence — and that confidence has already proven jumpy. In late August, Alphabet's stock shed roughly $700 billion in market value, more than 15% from its spring peak, as investors grew anxious about "mounting AI bills" and the absence of a clear return on them. That is a single data point, not a trend line, but it illustrates the mechanism: if two or three hyperscalers decided next year that their AI returns did not justify this year's spending, the same GDP arithmetic that flattered the third quarter would work in reverse, and it would do so with unusual force, because so little of the recent growth has come from anywhere else.

That asymmetry is the real policy problem. A broad-based 3.4% built on wages, small-business formation, and diversified capital spending across dozens of industries is resilient almost by construction — no single firm's miscalculation can derail it. A 3.4% built substantially on the data-center capital expenditure of a handful of companies is not resilient in the same way, whatever its long-run merits. Treating the two as economically or politically equivalent, which the quarterly headline implicitly does, is the mistake worth resisting.

What Follows From This

None of this means the number was fabricated, or that the AI buildout is worthless, or that officials should talk the economy down. It means the growth figure should be read the way a careful investor reads a single stock's earnings — line by line, not just the headline — before anyone treats it as proof that the broader economy, and the households who do not work in AI or own its stocks, are sharing in the gain. The Federal Reserve, Congress, and financial commentators owe the public that distinction, especially when the next data-center earnings call, rather than the next jobs report, may be what actually decides whether the recovery holds.