Is the U.S. Economy Propped Up by Data Centers? What the Data, the Fed, and History Say

The short answer

Data center and AI infrastructure spending is one of the largest single drivers of U.S. growth right now, but whether it is "propping up" the economy depends on how you measure it. Published estimates of AI-related investment's share of recent GDP growth range from about 11% to 92%, mostly because of how imports and time periods are treated.

Our read: it is a major source of growth and a concentrated risk, not the only thing holding the economy up. Larry Fink, Austan Goolsbee, and Kevin Warsh each describe a different piece of the picture.

Hey everyone,

The claim has become common enough to deserve a real look: that the U.S. economy is being held up by data centers. In recent weeks, three of the most-watched voices in finance have weighed in from different angles. BlackRock CEO Larry Fink warned that halting data center construction would cause a recession. Fed Chair Kevin Warsh called the AI spending surge a "shock." And Chicago Fed President Austan Goolsbee said the expansion is "largely shoving other parts of the economy down."

So how much is the economy really leaning on this buildout, who benefits, who bears the risk, and what does history say? This post lays out the data from official and research sources, the strongest arguments on each side, and the debate over data center bans. Sourcing note: every number below comes from a named, linked source at the end of the post. Quotes are attributed to the outlet where we verified the wording, and where estimates conflict, we show the range instead of picking one.

Short answer: are data centers propping up the U.S. economy?

Data center and AI infrastructure investment is now one of the largest single drivers of U.S. growth, and measured growth is more dependent on it than at any time in decades. But "propped up" depends heavily on how you measure it: estimates of AI's share of recent growth range from about one-tenth to more than nine-tenths. The honest conclusion is that the buildout is a major source of growth and a concentrated risk, not the only thing holding the economy up, and that both the optimists and the pessimists are partly right.

How big is the data center buildout?

Data center construction (Aug 2026, annualized)

~$85B

Computing infrastructure, share of GDP (Q1 2026)

~1.5%

Hyperscaler capex, 2026 pace

~$800B

Data centers' share of U.S. power, 2023

4.4%

Start with construction. According to the U.S. Census Bureau's construction spending data, private data center construction hit a record annualized pace of about $85 billion in August 2026, up roughly 73% from about $49 billion a year earlier. Over the same period, general office construction fell to about $45.8 billion. Data centers have out-spent offices every month since September 2025. Census doesn't count the servers and racks inside, so the full cost is larger.

U.S. private construction spending, annualized ($B)

Source: U.S. Census Bureau construction spending (August 2026 release), as reported by Yahoo Finance. Census figures are preliminary and get revised. They exclude servers and racks, and power plants and transmission lines are counted separately.

Including equipment, Epoch AI estimates that computing infrastructure (computers, data center construction, and networking) reached about 1.5% of U.S. GDP in the first quarter of 2026, up from an average of about 0.7% in 2015 to 2022, with roughly 0.8% of GDP attributable to AI specifically. On the corporate side, J.P. Morgan Asset Management estimated that the big cloud companies ("hyperscalers") were on track to spend nearly $800 billion on capital spending in 2026, roughly ten times their 2019 level and far above the roughly $150 billion a year they spent before the AI boom.

How much of U.S. growth is it, really?

This is where the headlines and the research diverge. In late 2025, Harvard economist Jason Furman pointed out that investment in information-processing equipment and software was only about 4% of GDP but accounted for "fully 92% of GDP growth" in the first half of 2025, according to Fortune's reporting of his analysis. Without those categories, he calculated, growth would have been about 0.1% annualized.

Other careful estimates put the share much lower, for two main reasons. First, much of the equipment is imported (chips and servers, for example), and imports are subtracted in GDP accounting, so counting the gross investment overstates the net effect. Second, the studies measure different periods and different categories. J.P. Morgan Asset Management estimated that an AI-related investment proxy added 0.47 percentage points to a 2.1% pace of real GDP growth over the past year, about one-fifth of growth, after netting out imported hardware, and described the effect as "relatively modest" next to consumer spending. The St. Louis Fed's tracking of AI-related investment categories found they contributed roughly 30% of GDP growth in the second quarter of 2025 and about 11% in the third.

Estimated share of U.S. GDP growth from AI-related investment

Sources: Fortune (Furman, H1 2025; gross investment, not netting imports), Federal Reserve Bank of St. Louis (Q2 and Q3 2025 contributions of AI-related investment categories; the post does not discuss imports), J.P. Morgan Asset Management (0.47 of 2.1 percentage points, net of imported hardware; "about one-fifth"). These measure different periods and definitions and should be read as a range.

Furman himself added a caveat. Without the AI boom, he said, the economy "would probably have lower interest rates [and] electricity prices," which could "maybe make up about half of what we got from the AI boom." In other words, the counterfactual isn't zero growth. And consumer spending remains the biggest part of the economy by far.

The pattern over time also matters. The St. Louis Fed found that the contribution of information processing equipment spiked in early 2025 and then fell back, because the growth rate of spending slowed even though spending stayed high. That's a subtle point with a big implication: an economy that leans on rapid increases in investment feels a drag when those increases flatten, even if nothing collapses.

What are Larry Fink, Austan Goolsbee, and Kevin Warsh saying?

These three don't agree with each other, which is part of what makes the debate worth reading.

Larry Fink: the bull, and a warning about bans. In an October 14 interview with CNBC, the BlackRock chairman and CEO made the case that the buildout is a pillar of growth and jobs:

"If we have a moratorium of some sort in this country, we will be in a recession."
Larry Fink, BlackRock, CNBC interview, October 14, 2026

He also defended the sector's employment effects:

"Data centers are not the problem. They're creating more new jobs than ever before."
Larry Fink, as reported by Yahoo Finance from the CNBC interview

Austan Goolsbee: the buildout is crowding out other activity. The Chicago Fed president told Fortune in September that the data center boom is lifting one part of the economy at the expense of others:

"I would characterize the expansion of the data centers as very hot, but largely shoving other parts of the economy down."
Austan Goolsbee, Chicago Fed, interview with Fortune, September 2, 2026

He added that the sector's rise is "stepping on others" as they compete for the same resources, and that "we're not far from that turning into aggregate overheating." He described the situation as a rebalancing across sectors, which he distinguished from economy-wide overheating, so it's a warning, not a diagnosis.

Kevin Warsh: a demand shock now, a productivity payoff later. Speaking at the IMF and World Bank meetings in Bangkok on October 16, the Fed chair framed AI spending as a shock that policymakers must take seriously:

"It's obviously hitting the demand side of the economy before it's hitting the supply side of the economy."
Kevin Warsh, Federal Reserve Chair, IMF meetings, October 16, 2026, as reported by ABC News

He added that "you've got to recognize this as a shock." He still expects AI to pay off eventually, but cautioned that "the path from where we are to there is likely to be a very bumpy one." The context makes this more than an academic debate. ABC News reported that U.S. prices rose 3.7% in September, a four-month high, per the Bureau of Labor Statistics, and that the Fed raised rates by a quarter point in September, its first increase since 2023. The same report tied part of the inflation to an oil supply shock, so AI isn't the only factor.

The case for optimism

Real demand and real jobs. Construction, electrical work, manufacturing, and energy all benefit from the buildout, which is the heart of Fink's argument.
Productivity is picking up. The Bureau of Labor Statistics reports that nonfarm business productivity rose 2.2% over the year ending in the second quarter of 2026, and has grown at a 2.1% annualized rate since late 2019, versus 1.5% over the previous business cycle. That's consistent with a technology payoff, though BLS data don't attribute the gain to AI, and economists haven't settled how much is due to it.
It's a general-purpose technology. As we argued in our Industrial Revolution post, technologies like this tend to take years to show up in productivity, then matter a great deal. Warsh's view is that the long-run effect could be a positive supply shock.
The spenders are large, established companies. That's a different starting point from many of the 1990s telecom start-ups, though the IMF points to rising leverage and concentrated exposures as amplifiers of any correction (see below).

The case for caution

Concentration. If one category drives a large share of growth, a pullback hits the headline numbers hard, even if the rest of the economy is healthy.
Inflation and crowding out. Goolsbee's point: competition for electricians, land, and power can push prices up in unrelated sectors, and Warsh's point: spending arrives before the supply it's meant to create.
Power demand. The Department of Energy, citing a Lawrence Berkeley National Laboratory report, found data centers used about 176 terawatt-hours in 2023, up from 58 in 2014, or 4.4% of U.S. electricity. Projections for 2028 range from 6.7% to 12%, an increase of 325 to 580 terawatt-hours.
Financial fragility. The IMF's October 2026 Global Financial Stability Report lists an AI-related market correction among its main risks: weaker or less durable AI investment and profitability could lead to a correction in AI-related asset prices, and rising leverage and concentrated exposures could amplify it. The Register's coverage of the full report adds that it flags circular financing, in which AI firms act as "one another's investors, financiers, and customers," and warns that markets are "increasingly exposed to an AI-driven correction" (quotes as reported by The Register). The IMF's central message is balanced: markets have stayed resilient, the global banking system "remains broadly resilient," and risks remain elevated. It does not predict a collapse.

Data centers' share of U.S. electricity use

Source: U.S. Department of Energy, summarizing the Lawrence Berkeley National Laboratory 2024 United States Data Center Energy Usage Report (December 2024). Usage rose from 58 TWh in 2014 to 176 TWh in 2023. The 2028 figures are scenario projections.

What does the telecom boom and bust say about today?

Investors and economists keep returning to the late-1990s telecom and dot-com boom, and the data supports at least part of the comparison. A St. Louis Fed analysis found that AI-related categories contributed about 0.97 percentage points to GDP growth in the first nine months of 2025, or roughly 39% of growth, versus 0.81 points, or about 28%, in 2000. (The 2000 comparison excludes data centers, which the dataset only covers from 2014, so the gap would be larger if they were included.)

Share of GDP growth from tech investment categories

Source: Federal Reserve Bank of St. Louis, "Tracking AI's Contribution to GDP Growth" (January 2026). Contributions are approximations (growth rate times GDP share). Part of the 2025 data center figure is estimated.

What happened next in telecom is the cautionary tale. According to a Federal Reserve Bank of San Francisco study of the period, capital spending by telecom service providers rose from about $47 billion in 1995 to a peak of about $121 billion in 2000, and real investment in information technology grew about 24% a year between 1995 and 2000 before plunging in 2001. The study attributes part of the overshoot to expectations that were too high, noting, for example, that a Level 3 Communications CEO had said Internet traffic would double every three to four months, which didn't happen. Capital spending by the new competitive local carriers (CLECs) fell by nearly 80% from its 2000 peak by 2002.

Telecom capex, 1995

$47B

Telecom capex, 2000 peak

$121B

Real IT investment growth, 1995-2000

~24%/yr

New-carrier capex drop, 2000-2002

~80%

The lesson isn't that the technology failed. The Internet was real. It's that investment can overshoot a genuine technology and that the correction can be painful for the investors and sectors involved. The honest counter-argument is that today's spending is led by large, established companies, which was less true of many telecom start-ups. The IMF's warnings about leverage and concentrated exposures are a reminder of why the funding structure is worth watching.

Would data center bans cause a recession?

Opposition to data centers has grown, driven by concerns over electricity prices, power-grid strain, land, and water, and local governments have responded with moratoriums and restrictions. That's the backdrop for Fink's warning that a national moratorium would push the U.S. into a recession. Two things are worth separating.

A hypothetical national ban versus what's actually happening. Fink's comment was about a broad moratorium. The actual wave of restrictions is local and uneven. Industry research firm SemiAnalysis, in a September 2026 review of more than 400 local instruments in 17 states, counted more than 300 local moratoriums or bans nationwide, but estimated that of roughly 20 gigawatts of planned capacity inside restricted areas, only about 1.5 gigawatts was actually delayed. It concluded that moratoriums "aren't yet at scale" and aren't currently killing the buildout, though they could slow it if they become much more widespread and restrictive.
Fink's claim is a warning, not a consensus forecast. It rests on the premise that growth now depends heavily on capital spending, which the data above partly supports (and the range of estimates shows is contested). Furman's caveat is relevant here too: part of what a slowdown would take away could be offset by lower interest rates and electricity prices, though that's his estimate about a world without the boom, not a forecast of a ban's effects.

There's also a policy question beneath the economics. Communities raising concerns about electricity costs, water, and transparency aren't simply wrong, and Goolsbee's data on competing demand for workers and resources suggests there are real trade-offs. The debate is less "ban or build" than who pays for the grid and how fast the capacity can be added.

What it means for investors and households

Know your exposure. If a large share of index returns and earnings growth comes from companies tied to AI spending, a broad index fund holds more of this bet than it may appear.
Watch the financing. The IMF points to rising leverage and concentrated exposures as the things that could amplify a correction, and The Register's reading of the full report highlights circular financing. How the buildout is funded matters more than any single quarter's capex number.
Watch inflation and the Fed. If AI-driven demand keeps prices elevated, interest rates may stay higher for longer, which affects mortgages, valuations, and borrowing costs. The Fed's next meeting is October 28.
Keep perspective. History suggests the technology can be transformative and still produce losers among the companies financing it. Diversification is the main protection.

For more on how these themes connect, see our posts on Anthropic's leaked IPO prospectus, nuclear power and AI, and why P/E ratios struggle with pre-profit companies.

Key takeaways

The buildout is big and growing fast: about $85 billion a year in data center construction alone, and computing infrastructure at about 1.5% of GDP.
"Propped up" depends on the method. Estimates of its share of growth range from about 11% to 92%, mostly because of how imports and time periods are treated.
The experts disagree for good reasons. Fink stresses jobs and growth, Goolsbee stresses crowding out, and Warsh stresses a demand shock now and productivity later.
History is a warning, not a verdict. The telecom boom was a real technology and a real bust.
Bans are so far a small effect, but a live risk. Fink's recession warning concerns a broad moratorium, not today's patchwork.

More soon,
Learn to Love Money

Frequently asked questions

How much does data center spending contribute to U.S. GDP growth?

It depends on the method. Jason Furman estimated information-processing equipment and software accounted for 92% of first-half 2025 GDP growth, gross of imports. The St. Louis Fed found about 30% in Q2 2025 and 11% in Q3, and J.P. Morgan Asset Management estimated about one-fifth after netting out imported hardware.

How big is the data center buildout?

The Census Bureau reported private data center construction at a record annualized pace of about $85 billion in August 2026, up roughly 73% from a year earlier. Epoch AI estimates computing infrastructure reached about 1.5% of U.S. GDP in Q1 2026, and J.P. Morgan estimated hyperscaler capital spending near $800 billion for 2026.

What did Larry Fink say about a data center moratorium?

In an October 14, 2026 CNBC interview, BlackRock's Larry Fink said, "If we have a moratorium of some sort in this country, we will be in a recession." His comment concerned a broad national moratorium, not today's patchwork of local restrictions.

Does the AI boom look like the dot-com telecom bubble?

Partly. A St. Louis Fed analysis found AI-related categories contributed about 39% of GDP growth in the first nine months of 2025, versus about 28% in 2000. Telecom capital spending then fell about 80% for new carriers from the 2000 peak to 2002. The counterargument is that today's spenders are large, established companies, though the IMF warns about leverage and concentrated exposures.

Are data center bans slowing the buildout?

Not much so far. SemiAnalysis counted more than 300 local moratoriums or bans, but estimated that only about 1.5 gigawatts of roughly 20 gigawatts of planned capacity in restricted areas was actually delayed, and concluded moratoriums "aren't yet at scale."

NOT FINANCIAL ADVICE. This post is for education only and is not a recommendation to buy or sell any security or a prediction of an economic outcome. Quotes are reproduced as published by the cited outlets. Economic estimates are approximations that differ by method, and projections are scenarios, not forecasts. Data is as of mid-October 2026 and may be revised. Do your own research or talk to a licensed advisor before investing.