Industry overview

Most of us know the Silicon Valley story. In 1939, Bill Hewlett and Dave Packard started their electronics company in a garage in Palo Alto, and that small operation helped give rise to one of the greatest concentrations of technological innovation the world has ever seen.

But there is another localized American technology story that never attracted as much attention, about 3,000 miles east, in the suburbs of Washington, D.C.

Back in the 1990s, an early internet exchange called Metropolitan Area Exchange-East, or MAE-East, helped turn Northern Virginia into one of the busiest crossroads of the young internet. Internet companies such as UUNET and America Online established major operations in the area, drawn by its communications infrastructure, proximity to the federal government, and growing concentration of technology businesses. As more companies arrived, the region became an increasingly attractive place for still more to build: internet connections were plentiful, skilled workers were nearby, and local officials welcomed the development.

Over the following decades, the area became known as Data Center Alley — one of the world’s largest concentrations of data centers, centered on Loudoun and Prince William counties.

But today, Data Center Alley is looking more like a cul-de-sac. Growth is coming to a halt. Just west of its heart sits Fauquier County, which seemed a natural place for the next wave of development, given AI’s enormous appetite for computing power. Developers introduced nine business-park data center proposals to Fauquier. Only one gained approval. The other eight did not. Then, on October 8, the Fauquier County Board of Supervisors voted to adopt new zoning rules that make data center construction even more restrictive.

The obstacles facing developers fall into three categories, and each one creates a financial problem for the companies trying to build the infrastructure behind AI.

First, localities are cancelling projects before they ever get off the ground. One of the eight withdrawn business-park proposals in Fauquier featured a 2.2-million-square-foot campus near Remington. Another developer abandoned its data center plans and will build homes instead.

Second, projects face mounting delays. Amazon.com Inc. (AMZN) won approval for a $550 million data center in Warrenton in February 2023. Residents sued to block the project, and more than three years later, it remains tied up in court.

Third, projects are incurring rising costs. Remington Technology Park, approved back in 2018, still can’t get enough electricity from the local grid. In July, county supervisors approved a plan for hundreds of natural gas fuel cells and backup diesel generators. Without that workaround, the development could face delays until at least 2030.

In the region that helped build the modern internet, developers are struggling to secure the land, legal approvals, and electricity needed to build its next generation of infrastructure. In today’s Smart Money, Eric Fry walks through what’s happening in this Virginia county, why it matters beyond the county itself, and how to position a portfolio for the next phase of the AI trade.

In his recent work, Fry has described how a two-sided vise squeezes the life out of every major tech boom, eventually. One side is the Revenue Jaw: technological advances that make services cheaper to provide, pushing down what companies can charge. The other side is the Expense Jaw: the rising cost of building and operating the infrastructure behind the boom. And Fauquier County has just shown three examples of that Expense Jaw tightening.

Every year spent in court can mean another year before a project starts generating revenue. Every permitting hurdle also adds time and expense. Every dollar spent generating electricity on-site adds to the cost of construction. And every project that developers must redesign, or abandon completely, impairs the prospective return on investment.

Fauquier isn’t an outlier. More than 300 towns, cities, and counties across America have adopted data center moratoriums or bans, according to research firm SemiAnalysis. And according to Gallup, 7 in 10 Americans oppose building a data center in their area. A big reason is electricity bills: people understandably worry they’ll pay more for power to support a boom whose benefits they may never see.

Residents aren’t the only ones facing higher bills. The tech giants first funded the AI buildout from their considerable cash reserves. Increasingly, they’re borrowing to keep it going, and every delay makes that debt more expensive to carry. And debt doesn’t wait. As the late Charlie Munger, longtime associate of Warren Buffett, bluntly said: “The liabilities are always 100% good; it’s the assets you have to worry about.” In other words, the money you owe is real and unavoidable, while the money you hope to earn from your investment is another matter entirely — an uncomfortable distinction when you’re spending billions on infrastructure that might not begin generating revenue for years.

To be fair, the buildout continues. Remington Technology Park got its power plan approved. The hyperscalers keep raising their spending budgets. And as AI becomes more efficient, demand for computing power will likely increase. The demand for AI isn’t imaginary. Neither are the obstacles to supplying it.

But the Expense Jaw alone doesn’t end a boom. There’s more danger ahead if the Revenue Jaw closes, too: as AI becomes cheaper to provide, competition could push down what companies can charge, even as construction costs keep climbing and debt payments come due.

Again, the AI buildout isn’t stopping. But it’s encountering delays and growing more expensive in places like Fauquier County. And those costs can fall hardest on the companies financing the infrastructure at the center of the boom: the “AI Builders.” History suggests that even excellent companies can become disappointing investments when they spend too much building infrastructure that eventually becomes cheaper and more plentiful. The internet was a tremendous success, yet that didn’t prevent investors in some of its most important infrastructure companies from losing fortunes.

The companies that come out ahead are often the “Appliers,” businesses that put the new technology to work once it becomes cheap and abundant. They don’t need to own the infrastructure. They just need to find profitable ways to use it. In his recent presentation, Fry explains this in more detail, reveals which companies sit within AI’s “vise,” and names the No. 1 AI Applier to buy right now.

Industry overview based on InvestorPlace

What it means for investors

For investors, the Fauquier story is a warning about where the risk sits in the AI trade today. The companies financing data centers — the AI Builders — face longer timelines, higher costs, and growing debt, all while the services those data centers will provide are getting cheaper. That squeeze is exactly what Fry’s two-sided vise describes, and it is why he argues the better place to invest may be the Appliers: companies that use AI once it becomes cheap and abundant, rather than paying to build it. Before buying into the buildout, investors should look at how much a company is borrowing, how long its projects take to start earning revenue, and whether its spending budgets are still rising — the hyperscalers’ budgets are one signal worth watching. The buildout is not stopping, but the easy money in financing it may already be behind us.

What it means for employees

This story has no direct employee angle: it does not report hiring, layoffs, or pay changes at any company. Indirectly, though, delays matter for workers in the places where data centers are planned. When a project is cancelled, tied up in court for years, or waiting on power until 2030, the construction jobs and the longer-term operations jobs — technicians, electricians, and other skilled workers — arrive later, or not at all. That is a real cost for communities like Fauquier County, even as it is the same skilled workforce that helped make Data Center Alley attractive to builders in the first place.

What it means for customers

For customers, the effect is indirect but real. The businesses and people who use AI services depend on the data centers being built, and when those projects cost more and take longer, the expense can eventually feed into what customers pay for cloud and AI services. There is also a more immediate customer angle at home: residents near proposed data centers worry their electricity bills will rise to support a boom whose benefits they may never see — a concern shared by 7 in 10 Americans, according to Gallup. The AI tools themselves keep getting cheaper and more capable, but the infrastructure behind them is getting more expensive to build, and someone has to pay for that gap.

Eric Fry

Note: Investment calculations and projections are for illustrative purposes only. Market conditions change frequently. Consult a qualified financial advisor before making investment decisions.