In Brief …
AI’s booming. Data centers are scaling. But the energy to run them? Lagging. From grid delays to transformer bottlenecks, infrastructure—not algorithms—is the real threat to progress. This is a construction problem disguised as a tech story.
AI Is Hungry. The Grid Isn’t Ready.
Everyone’s obsessed with AI’s potential. But those breakthroughs don’t run on inspiration.
They run on electricity.
The International Energy Agency projects global data center electricity use could double by 2030, hitting nearly 1,000 terawatt-hours annually. That’s more than Japan uses in a year. And a huge chunk of that growth is coming from generative AI.
In the U.S., McKinsey estimates we’ll need an additional 80+ gigawatts of new capacity by 2030—more than triple today’s data center load.
Spoiler: We’re not building fast enough.
Why Can’t Energy Infrastructure Keep Up with AI?
AI isn’t some distant threat to the grid—it’s here, now, scaling fast. In Northern Virginia—the world’s data center capital—utilities are already strained. Ireland has paused new data center connections in Dublin until at least 2028. London, Frankfurt and Singapore are all facing the same crunch.
Here’s the rub: data centers can go up in 18 to 24 months. But power infrastructure takes five to 10 years, if you’re lucky. In the U.S., just getting through NEPA permitting can eat up four to six of those years.
Even if you get the green light, good luck finding the gear. Transformers—the industrial backbone of grid upgrades—are in critically short supply. IEEE Spectrum reports lead times stretching to four years, with costs up as much as 80% since 2020. And it’s not just AI that’s feeling the heat. Housing developments, EV charging stations and solar projects are all waiting on the same parts.
Are SMRs the Answer?
Small modular reactors (SMRs) offer clean, reliable baseload power in a compact package. It’s no wonder the tech giants are paying attention. Google inked a deal with Kairos Power to eventually offtake 500 MW of SMR-generated electricity. Microsoft brought on a nuclear integration lead. In Virginia, a 19-data-center campus is being designed around SMR capacity.
But these are long bets. Between permitting, public skepticism and costs, SMRs aren’t expected to deliver meaningful power until the 2030s. They’re a future fix. We need answers now.
How Smart Builders Are Staying Ahead
Site for Power, Not Just Proximity: Hyperscalers are choosing Iowa, Indiana and Texas over the usual suspects. Why? Fewer constraints, faster permits, more available capacity.
Build Their Own Supply: If the grid can’t keep up, some are going off-grid. Think on-site gas turbines, solar+battery microgrids, even plans to sell excess power back.
Order Early, Standardize Fast: Some developers are stockpiling transformers years in advance or deploying prefab substation kits to avoid delays.
Use AI to Build for AI: National Grid UK is using AI to optimize grid performance and accelerate renewables integration. U.S. data centers are also testing demand-shifting models to balance loads and reduce peak stress.
This Isn’t a Computer Problem. It’s a Construction One.
We keep hearing AI will change everything. But here’s what might slow it down:
- Copper
- Concrete
- Permits
- Process—and by that, we mean the messy, human stuff: outdated project management workflows, a shortage of skilled workers, slow adoption of digital tools and bureaucratic slog that extends far beyond the permitting office.
The race isn’t just about GPUs. It’s about whether we can get enough electrons to where they’re needed—on time. The next big breakthrough won’t arrive because someone built a smarter model.
It’ll arrive because someone finally built the substation.