Will data centers be useful infrastructure — or tomorrow’s stranded boxes?

In Abilene, Texas what was once a 10-minute trip has somehow stretched to two hours. The traffic stems from the thousands of drivers who are now commuting each morning to the sprawling construction site that figures to be a $500 billion data center complex, one of the world’s largest. 

The widely publicized boom is concentrated: Just a few big American tech giants will spend half as much on AI infrastructure this year alone as all of Germany will over the next five. Yet it’s also far-flung: nearly 5,500 data centers across all 50 states.

That makes a national craze with very local implications. We’ve had those before: 19th century railroads, 20th century highways and the telecom buildout near the 21st. 

Still, at an estimated 1.2% of GDP, this year’s data center investment total would mark the third highest infrastructure investment in American history — trailing only the peak of rail mania and the height of the interstate highway system.

All have one thing in common: Each era’s overreach eventually became tomorrow’s backbone. But not all bubbles burst in the same way. What local leaders want to know is how the current AI infrastructure buildout maps to that past.

Most major innovations since 1820 came with bubbles — 37 of 51, according to a 2018 analysis, including electricity and telephones. Firms collapsed and investors were ruined, but big spending transformed the economy. When was the pain worth it? A 2020 book, “Boom and Bust,” argued that bubbles sparked by policy changes are most economically damaging for too little gain — like the American housing bubble of the 2000s. In contrast, bubbles driven by new technologies more often created something lasting, like the automobile and gasoline buildout of the 1920s, even through the chaos.

The question for place-based economic developers today is what happens when the construction traffic is gone and the mostly empty data centers remain. 

“Leaders need to make sure the benefits are real, not just another ribbon-cutting,” said Rae’Mah Henderson, a Techstars investment associate in Birmingham, Alabama. She’s also a co-host of Builders Live, Technical.ly’s monthly podcast on entrepreneurial ecosystem building trends.

Politicians do love celebrating data centers, which presently come with big headline investment figures, significant property tax contributions, upfront construction jobs and are often associated with exciting tech giants — like Apple or Alphabet. In pursuit of empty space and energy, many of the biggest are being built near small towns and in rural counties, rather than existing hubs like Virginia’s data center corridor.

“There’s a difference in economic development between building utility systems and building an entrepreneurial economy,” said Victor Hwang, founder/CEO of advocacy group Right to Start and another Builders Live co-host. “Both matter, but too often officials talk as if bringing Google or Amazon to town is the same as empowering local founders. It’s not.”

As industry analysts have asked: Are data centers truly infrastructure, or merely real estate? 

Already, after nearly a decade of exponential growth, new data center construction has slowed, according to new research from investment giant Apollo. The expansion is still torrid by otherwise constrained American infrastructure standards, but it may mark the end of the “lay-track-anywhere” phase of AI’s physical buildout.

What data centers are (and aren’t)

Data centers are the most visible part of what gets called AI infrastructure — but they’re only one (very big) piece of a sprawling network of machines that do lots of math super fast.

Broadly, AI infrastructure includes the entire physical and digital supply chain that makes large-scale computation possible: the chips and accelerators that perform the math, the high-speed fiber networks that move data, the power and cooling systems that keep machines running and the software frameworks that orchestrate all of it. 

If the internet was built on wires and websites, AI is built on compute and cooling. 

At the moment, data centers are where this all becomes concrete. They are industrial-scale facilities packed with servers that store and process data, drawing massive energy loads and requiring extensive water for cooling. Local governments offer incentives to lure them, often promising new jobs. But as Technical.ly has reported, those jobs are largely front-loaded: thousands of construction workers today, a few dozen maintenance staff tomorrow.

Is the buildout worth it? MIT economists estimate 95% of firms are getting zero return on AI investments. But usage keeps growing: AI is now by default part of nearly every Google search, Amazon listing, Facebook feed, banking chatbot and more. Plus, 40% of American workers used generative AI tools last month. Boosters bet that with enough use cases, the gains will come, and nobody would want to undershoot demand.

“There are only too many gas stations until you run out of gas,” said Lightship Capital founding partner Brian Brackeen, and our final co-host. “You can literally run out of compute, and we’ve already seen what that looks like when cloud outages take a third of the internet offline.”

That logic explains why nearly every state is chasing data-center projects — and why governors are rolling out energy incentives, like PA Gov. Josh Shapiro’s $39 million program to prepare industrial sites for hyperscalers such as Amazon or Meta.But chasing ribbon cuttings is not the same as building an ecosystem. As regions compete to host the “engine rooms” of AI, and residents express concerns over the coming tide, the smarter question isn’t how many boxes get built. It’s how well those boxes connect back to people, entrepreneurs and long-term local capacity.

The boom, the brake and the bubble talk

The AI infrastructure buildout has been led by many of the most profitable businesses in human history — the so-called “Magnificent Seven,” including Alphabet, Amazon, Meta and Nvidia. Though newer, more speculative entrants have joined, the risk of today’s infrastructure bonanza is concentrated among firms with gargantuan cash reserves. 

Worry more about the people and the communities coexisting with the steady hum of an unexplainable future. 

The 1990s telecom infrastructure left investors holding fiber that no one yet needed — until the next decade, when the internet caught up. But it was a much less intrusive stranded asset. The question now is whether today’s rural data centers will follow that pattern: essential but premature, or wasteful and abandoned.

Related news

Speak now: Pennsylvania commission is reviewing how data centers impact energy costs

Small modular reactors to power Delaware? Nuclear task force begins talks on feasibility

Eos Energy makes $353M headquarters move to Pittsburgh after nearby data center deal

By 2030, McKinsey predicts companies will invest almost $7 trillion in capital expenditures on data center infrastructure globally. Despite the risks, economic development leaders will pursue their slice of the buildout. Advanced research and emerging industries will benefit. 

Local communities are starting to ask whether the return is worth the load.

If (slightly) slowing rates of construction signal a more mature phase, now’s the time to shift from quantity to quality: smarter siting, clearer contracts and stronger local linkages. Regions that do this will turn a speculative surge into lasting advantage.

The next infrastructure cycle always feels like a gamble while it’s happening. But the winners aren’t those who bet biggest — they’re the ones who make sure the tracks, pipes, wires or servers connect back to people. Residents stuck in unfamiliar traffic and congestion may be impatient partners.

“Railroads moved people. Data centers move bits,” said Hwang. “The more we can move people — their ideas, their networks — the more dynamism we create.”