In the 1800s, America built railroads faster than it had passengers to fill them. Towns sprang up where the tracks went and withered where they didn’t. Fortunes were made laying the rails, and more fortunes were lost when the financing caught up with reality and the Panic of 1873 hit.
The railroads still changed the country. Both of those things were true at the same time.
AI data centers are this century’s railroad. The money is on a scale few people grasp, it’s landing in a handful of places, and the people who will profit most aren’t always the people who live next to the tracks. I covered what these buildings do to the environment in the environment article and what they mean for the race with China in the geopolitics article. This one follows the money.
How much money is going into AI data centers?
More than almost anything else in the economy.
The five biggest US cloud and AI companies, Amazon, Alphabet, Microsoft, Meta and Oracle, committed $660 billion to $690 billion in capital spending for 2026, according to the Futurum Group. That’s nearly double the roughly $380 billion they spent in 2025. Amazon alone planned about $200 billion.
Construction shows up in government data. The Census Bureau now tracks data centers separately, and in April 2026 spending on their construction ran at an annual rate of $50.7 billion, up about 27 percent in a year. That’s more than the country spent building ordinary office buildings. Data centers are now more than half of all private office construction in the United States.
Suppliers have grown accordingly. On October 29, 2025, Nvidia became the first public company worth $5 trillion. TechCrunch noted that put it above the entire stock markets of every country except the United States, China and Japan. In a gold rush, sell shovels.
Nvidia sells the shovels.
Do AI data centers help the national economy?
In early 2025, they were most of it.
Harvard economist Jason Furman calculated that investment in information processing equipment and software was about 4 percent of GDP but accounted for 92 percent of US GDP growth in the first half of 2025. Without it, he estimated, the economy would have grown at an annualized rate of just 0.1 percent. That category is largely the computers, chips and software going into data centers.
Furman added a fair caveat. Without the AI boom, interest rates and electricity prices would probably have been lower, and other parts of the economy would have grown more. He put that offset, very roughly, at about half the difference.
Even with the caveat, that’s a lot of weight on one industry. Supporters see it as the start of a productivity wave that will lift everything. Skeptics see an economy leaning on a single pillar.
Both camps are watching the same numbers.
Is the AI data center boom a bubble?
Nobody knows, and anybody who tells you for certain is selling something.
Revenue is the first worry. Bain & Company estimated in September 2025 that by 2030 the industry will need about $2 trillion a year in new revenue to pay for the computing it’s building. Even counting the savings AI produces, Bain projected a shortfall of about $800 billion.
Businesses may not be getting their money’s worth yet, either. An MIT report released in mid-2025, The GenAI Divide, found that 95 percent of organizations it studied were seeing no measurable business return on generative AI, despite $30 billion to $40 billion in enterprise spending.
The study was built on a review of more than 300 public AI initiatives, 52 interviews and 153 survey responses, and its authors blamed tools that don’t retain feedback or improve over time. They said infrastructure, regulation and talent weren’t the main problem.
It’s a snapshot, and a small one. It still got everyone’s attention.
Then there’s circular money. In September 2025 Nvidia announced it would invest up to $100 billion in OpenAI, largely to build data centers that would be filled with Nvidia chips. In February 2026 that plan gave way to a smaller investment of about $30 billion.
Jay Goldberg of Seaport Global Securities said on Bloomberg Television that it was “kind of like having your parents co-sign on your first mortgage,” and called it bubble-like. Axios pointed out that Cisco’s financing of its own telecom customers in the 1990s fed the overbuilding that ended in the dot-com crash.
Others shrug. Max Kettner of HSBC argued that spending corrects itself when it stops making sense: “in three years, once it’s gone too high, too far, cool. You stop.”
That’s the railroad story again. A boom can be a bubble and an investment in the future at the same time. The tracks laid before 1873 didn’t vanish when the companies that laid them went broke.
Somebody else ran trains on them.
AI data centers are this century’s railroad. Fortunes get made laying the rails, more get lost when the financing catches up, and the tracks stay either way. – Richard LoweShare on X
What do AI data centers do for a local economy?
They pay a lot of taxes, hire very few people for the long haul, and change the place around them.
Loudoun County, Virginia, has the largest concentration of data centers in the world. It collected about $1.1 billion in taxes on the computer equipment inside them last year, roughly 38 percent of its general fund revenue, according to Moneywise. Ten years earlier the figure was about $150 million. The county has cut its real property tax rate every year for a decade, from $1.145 per $100 of assessed value in 2016 to $0.805 in 2026.
If you own a home in Loudoun, that’s real money.
Not every resident likes the trade. One county supervisor told The New York Times, “We’ve become addicted to the data centers for their tax revenues, but at what cost?” Residents complain about generator and cooling noise and about new high-voltage lines, including a proposed 185-foot transmission tower in one family’s backyard.
Addiction is the right word for the risk. A county that gets nearly two of every five general fund dollars from one industry has bet its schools and roads on that industry’s future. If the boom slows, or the next generation of campuses goes somewhere cheaper, the hole in the budget lands on the homeowners whose taxes were cut.
How many jobs does an AI data center create?
Thousands to build it. A few hundred at most to run it.
When Vantage Data Centers planned a 1.1 million-square-foot campus near Reno, 2024 business records projected more than 4,000 temporary construction jobs and 73 permanent jobs over the following decade. Meta said its $10 billion Hyperion campus in Richland Parish, Louisiana, would employ more than 5,000 trade workers at peak construction and more than 500 people in permanent operations. Researchers at the University of Michigan concluded that data centers don’t bring high-paying tech jobs to their communities and compared them to infrastructure like bridges and highways.
That comparison is fair, and it cuts both ways. Nobody builds a highway for the toll booth jobs. You build it for what it makes possible. A data center’s local value is the tax base and the construction wave. The payroll is small.
The construction wave is real. The Associated Builders and Contractors estimated the industry needed to bring in about 439,000 new workers in 2025 and about 349,000 in 2026. Its chief economist, Anirban Basu, said most of the 2026 demand comes from retirements more than new projects, even with the AI boom. For electricians, pipefitters and heavy equipment operators near a big campus, though, the boom has meant steady work and overtime for years at a stretch.
Are states giving AI data centers too many tax breaks?
The bills are bigger than anyone expected.
Most states that court data centers exempt the servers and equipment inside them from sales tax. Because AI equipment is so expensive and gets replaced every few years, those exemptions have grown fast.
Good Jobs First, a nonprofit that tracks corporate subsidies and is critical of them, reported in June 2026 that Georgia’s exemption cost about $2.5 billion in fiscal 2026, Texas about $1.3 billion, and Ohio about $1.6 billion in 2025. Virginia’s combined state and local losses neared $2 billion in fiscal 2025. Fourteen states with these exemptions don’t publish full annual figures on what they cost.
One Louisiana deal shows the scale. Fortune reported in May 2026 that Meta’s Hyperion campus in Louisiana stands to receive about $3.3 billion in sales and use tax breaks over 20 years, an estimate Good Jobs First called conservative. Meta has committed more than $300 million to roads, wastewater and other local infrastructure as part of the deal.
Supporters say the exemptions are how states win projects that would otherwise go next door. In their view, the property taxes and construction spending outweigh what’s given up. Opponents say the companies would build anyway, since they need the power and the land. States, in that telling, are bidding against each other to hand money to some of the richest corporations on earth. “These are wasteful subsidies for an industry that is growing very quickly and doesn’t need any public investments or support,” Kasia Tarczynska of Good Jobs First told Fortune.
Both sides agree on one thing. Most states can’t tell their own taxpayers what the deals cost.
Do AI data centers raise local electricity bills?
In some places, and that’s a cost that falls on every household whether or not it gets a tax cut.
The market monitor for PJM, the grid that serves Virginia and 12 other states plus Washington, D.C., attributed $6.3 billion of the July 2026 capacity auction’s costs to data centers, about 38 percent of the total. An industry-funded review by the consultants E3 found no evidence that data centers have historically been subsidized by other customers, and pointed out that some of the fastest-growing states had some of the smallest rate increases.
I went through that argument in detail in the environment article. The economic point is simple. A county’s tax windfall and its residents’ power bills are two sides of the same ledger, and they don’t always land on the same people.
Which countries are winning the AI data center economy?
The United States builds the most and makes the most money from it. But the boom is spreading wherever there’s land, power and a friendly government.
Malaysia is the fastest example. Johor, the state just across the causeway from Singapore, went past 900 megawatts of data center capacity in about three years, a milestone that took Singapore 12 to 14 years, according to KrASIA. Singapore is short on land and energy. Johor has both, and it’s close enough to connect back. Malaysia attracted at least 210 billion ringgit, about $51 billion, in digital investment in 2023 and 2024.
Johor’s boom brought the same problems it brings in America. A Johor state investment committee began rejecting new data center applications that called for liquid cooling because of water shortages. Three water plants are under construction, the first phase due by June 2027. Malaysia also imposed its own export controls on high-performance AI chips to avoid becoming a backdoor around US restrictions, while trying to stay neutral between Washington and Beijing.
Gulf states are buying a seat with money and energy. Stargate UAE, announced in May 2025 by the Emirati company G42 with OpenAI, Oracle, Nvidia, Cisco and SoftBank, is a one-gigawatt cluster inside a planned five-gigawatt campus in Abu Dhabi. Ireland took the opposite path, freezing new data center grid connections around Dublin for about four years, until late 2025. By 2025, data centers used 23 percent of the country’s metered electricity.
And then there’s the supply chain. Taiwan makes about 90 percent of the world’s most advanced chips. Every data center built anywhere sends money back to the island, and to the Dutch, Japanese and American companies that make the chipmaking machines. The railroad needed steel mills. The AI buildout needs TSMC.
Who wins and who loses from the AI data center boom?
So far, the clearest winners are the suppliers: chipmakers, power equipment makers, construction firms and the skilled trades. Counties that win campuses on good terms win too, at least as long as the tenants stay.
The losers are harder to see. Ratepayers in grids where data center demand drives up capacity costs. Taxpayers in states whose exemptions cost billions with little disclosure. Neighbors who live with the noise and the power lines. And, if the revenue never catches up to the spending, investors and lenders who financed the overbuild.
Still unanswered is the biggest question of all: whether AI itself produces enough productivity to justify the spending. If it does, the spending will look like the railroads in hindsight: messy, overbuilt in places, and worth it. If it doesn’t, a lot of very expensive buildings will sit half empty, and some counties will discover what it means to depend on one industry.
Is the AI data center buildout going to stop?
I don’t think so. As I wrote in the environment article, I think AI data centers are inevitable. There may be a crash, or a long shakeout. The buildings will still get built, and the capacity will still get used, by whoever owns it when the dust settles.
That makes the local deals matter. A county or state negotiating with a data center developer is signing up for decades. The ones that come out ahead will be the ones that make the companies pay for their own grid upgrades, disclose what the tax breaks cost, tie incentives to real commitments, and avoid letting one industry carry the whole budget.
Follow the tracks
Railroads made some people rich, ruined others, and left a country crisscrossed with track that everyone used for a century. AI data centers are on the same path. Watch where the rails go, who pays to lay them, and who’s still holding the bonds when the financing catches up.
The AI and writing hub has more on AI’s bigger picture. If you’d like help figuring out where AI belongs in your own writing work, my AI services cover that. Buy a ticket if it gets you somewhere. Just know who owns the railroad.
