The AI Infrastructure Boom Transforming Tech

When you drive past Loudoun County, Virginia at dusk, you can tell that something has changed. In fields where soy was once grown, long, windowless boxes squat. Cooling units hum at a low frequency that you can feel before you hear it. Near loading bays, trucks stand motionless, awaiting shipments of cables as thick as a man’s wrist. This is where the future of artificial intelligence is being poured, bolted, and wired together, even though none of it looks like what people had in mind.

When you finally see the numbers in print, they almost seem cartoonish. According to Goldman Sachs, annual spending on AI infrastructure will increase from approximately $765 billion this year to $1.6 trillion by 2031. In 2026, Amazon, Alphabet, Microsoft, and Meta alone intend to invest nearly $700 billion in buildout. In actuality, that is more than the entire dot-com fiber boom of the late 1990s, a comparison that investors frequently refer to, half as assurance and half as caution.

Observing this development gives the impression that the industry is no longer pretending AI is software. Anxieties about the supply chain are growing along with the industry’s weight. Taiwanese chips. German transformers. vital minerals from a few mines dispersed throughout unfriendly nations. You can tell that algorithms are no longer the bottleneck when a single PJM grid auction causes prices in a region to spike by fifteen billion dollars. Instead, it’s electricity and, more and more, the patience of the residents who live close to substations.

Of all companies, Dell is experiencing an odd second act. It used to be written off as the dull PC manufacturer that missed the cloud, but now it offers the server stacks and racks that hyperscalers physically require. In the most recent fiscal quarter, its AI revenue increased fourfold. The backlog of orders has exceeded $43 billion. Michael Dell most likely stopped anticipating this kind of comeback story around 2015. Whether it continues depends on unanswered questions, such as how many of these data centers will be used at anticipated levels in three years.

There are significant disparities in the geography. Approximately 75% of foreign direct investment entering developing economies now ends up in just ten nations, according to a recent UNCTAD warning. With little ability to connect, the others are watching the AI economy come together elsewhere. That is unsettling because a technology that is promoted as democratizing intelligence is actually concentrating capital more tightly at the infrastructure layer than nearly anything that came before it.

Local officials are beginning to resist, or at the very least engage in more intense negotiations, in places like rural Ohio and Phoenix. According to researchers at Brookings, these communities are being given warehouses disguised as economic miracles, along with temporary construction jobs, a small operational staff afterward, and unmanageable water bills. A few towns have started requesting workforce training, R&D collaborations, and co-investment. It’s unclear if the hyperscalers cooperate or simply relocate to the next county with less expensive electricity.

The rhyme with earlier booms is difficult to ignore. The internet eventually developed from the glut of dark cable left behind by the fiber-optic frenzy. Perhaps the same is true of AI infrastructure. Perhaps it doesn’t. In either direction, the buildings will stand humming in the dark while they wait to see if the demand was genuine or merely a very, very confident forecast.