The data center boom is here: Experts explain how to build AI infrastructure correctly

To fast-track the buildout of data centers that power artificial intelligence (AI), tech giants like Amazon, Google, Meta, Microsoft, Oracle, and OpenAI are pouring money into construction projects. Globally, companies are predicted to spend $375 billion in 2025 on AI infrastructure—a 67% surge from last year, according to an estimate by bank UBS.

While this new era of rapid data center construction aims to power innovation, from scientific breakthroughs to AI-enhanced public services, it comes with distinct challenges and more isn’t always better, warn Virginia Tech artificial intelligence experts Walid Saad and Dimitri Nikolopoulos. Here’s why.

Data centers carry high environmental costs

“Training ever-larger models on massive data sets requires enormous computing power, which in turn drives up energy demand and environmental costs,” Saad said. “It is not the most sustainable path for better AI.”

“What we’ve learned is that true AI progress requires more than brute-force computing. The next leap forward will come from algorithms that build an internal understanding of how the world works, often called ‘world models.’

“These models let AI learn efficiently, generalize to new situations, and make decisions using far less data and energy. Leading research groups, such as Meta’s FAIR lab and our own Virginia Tech teams, are pioneering this direction because it holds the key to both sustainable AI and smarter AI.”

Data centers need to be sustainable, distributed, and accessible

“Without strong public-interest guardrails, more data centers may just deepen existing divides and environmental costs,” Nikolopoulos said. “It’s not just about how many data centers we build—but what kinds, where, and for whom.”

“To truly lead, the country needs AI infrastructure that is powered by clean energy and designed for efficiency, that is widely distributed in locations beyond just Silicon Valley and Northern Virginia, and that is accessible to researchers, startups, and public-interest institutions—not just Big Tech.”

Data center construction requires smart guidelines

“More data centers means increased environmental impact for the grid, water resources, and emissions,” Saad said. “It is important to at least have some guidelines so that our critical infrastructure can sustain this growth. AI innovation and environmental responsibility can advance hand in hand, not by scaling up endlessly, but by scaling smartly, designing algorithms that learn more like humans and waste far less energy.”

Data centers should benefit everyone

“If AI is to shape education, medicine, and governance, we must ensure the infrastructure is not gatekept behind closed platforms or unaffordable services controlled by a few firms,” Nikolopoulos said.

“U.S. leadership in AI should be measured not just in processing speed, but in how broadly the benefits are shared and how wisely the resources are used. We need to shift our mindset from ‘scale at all costs’ to ‘impact per watt, per dollar, per community.'”