04/15/2026
Everyone’s talking about AI like it’s a software race.
It’s starting to look a lot more like an infrastructure race to us.
Nearly half of the AI data centers expected to come online in the U.S. by 2026 are now facing delays or cancellations, including major projects tied to OpenAI’s Stargate initiative.
That’s a loud signal.
Because while everyone’s busy obsessing over bigger models, better chips, and what AI can do next, the real issue is a lot less glamorous: The physical backbone needed to power all of this isn’t ready.
And the bottlenecks aren’t exotic. In fact, they’re painfully basic:
👉 Not enough power
👉 Not enough electrical equipment
👉 Hardware supply constraints
👉 Trade tensions making key materials harder to get
In other words:
AI may be moving at warp speed… but the grid is still tying its shoes.
And that’s where this gets interesting.
Out of roughly 12 gigawatts of expected new AI data center capacity, only about one-third is actually under construction right now.
So while AI demand is accelerating, the real world is basically saying: “Great. Now where exactly are you plugging all this in?”
That means this isn’t just a tech story anymore.
It’s a power story. A land story. A logistics story. A who-got-there-first story.
The winners in the next phase of AI may not just be the companies with the smartest models.
They may be the ones that secured power, land, fiber, equipment, and permitting before everyone else showed up.
And that’s where real estate comes in.
Because as power gets tighter, the most valuable land may not just be in the fastest-growing markets.
It may be the land closest to substations, transmission access, fiber, and jurisdictions that won’t take forever to approve a project.
So yes, AI is changing software.
But it’s also starting to reprice industrial land, powered sites, and secondary markets that can support large-scale compute.
Turns out the next AI gold rush may be won less by code… and more by whoever controls the dirt under the power lines.