Here’s the thing nobody’s talking about: AI isn’t running out of chips or money. It’s running out of dirt.
Microsoft, Google, and Amazon have the capital and the processors to build data centers everywhere. What they don’t have? A power outlet that won’t take three to five years to connect. Water rights that won’t trigger a community revolt. Land that isn’t already spoken for.
This is the actual bottleneck crushing the AI infrastructure boom—and it’s about to force the entire industry to look up.
**The Problem That Broke the Grid**
Data centers need three things: power, cooling, and space. Earth is running short on all three. Interconnection queues in the U.S. now stretch years. Water rights in the West are getting tighter by the month. Prime real estate for data centers is disappearing faster than available GPU inventory.
Bloomberg estimates half of all AI data center projects in the U.S. will get delayed this year because of power constraints. That’s not a supply chain hiccup—that’s a structural problem.
**The Solution: Steal Power From the Sun, Cooling From Space**
Enter orbital data centers. It sounds like sci-fi, but it’s already happening.
Solar panels in low Earth orbit get 1,400 watts per square meter of raw energy. On Earth, even the best solar farms average 20-60 watts per square meter after accounting for clouds, night, and atmospheric loss. Space also offers something Earth can’t: a near-perfect vacuum sitting a few degrees above absolute zero. That means you can dump heat straight off GPUs into the void without fans, water, or any cooling infrastructure.
The math is wild. Today, running an H100 GPU for an hour in a terrestrial data center costs about $1. In orbit? $142. But here’s the kicker: $85 of that is pure launch cost. The energy itself is basically free once you’re up there.
**The Cost Curve Is Moving in One Direction**
SpaceX’s Starship is designed to drop launch costs from $3,000 per kilogram to $50-100 per kilogram at scale. Google’s own engineers published a study showing that at $200/kg, orbital compute becomes cheaper than Earth-based infrastructure.
The crossover happens around 2038. After that, orbital compute doesn’t just become competitive—it becomes the obvious choice. And the cost curve keeps falling while terrestrial costs keep rising due to resource scarcity.
**Why This Matters Right Now**
SpaceX just confidentially filed for an IPO targeting a $1.75 trillion valuation—potentially the largest in market history. That’s not hype. That’s capital committing to this thesis at scale.
When that IPO hits, every hyperscaler will have to respond. Microsoft can’t let Elon Musk own the infrastructure layer. Google can’t cede the compute substrate. Amazon can’t let a rival define the next generation of cloud. They’ll all pile in, which accelerates launch demand, which funds launch competitors, which drives costs down faster.
**The Play**
The real money in infrastructure booms comes from the picks-and-shovels layer, not the hyperscalers themselves. Look at Nvidia during the cloud boom. Look at Vertiv on power infrastructure.
For orbital compute: Rocket Lab (RKLB) is the embedded infrastructure play. Microchip Technology (MCHP) dominates radiation-hardened chips. Broadcom (AVGO) and Marvell (MRVL) handle networking. Alphabet (GOOGL) is the highest-quality hyperscaler exposure.
The thesis is simple: AI demand is exponential. Earth’s resources aren’t. The gap gets filled by something that exists outside Earth’s constraints.
The grid is coming. The only question is whether you’re positioned before it arrives.
