A venture capitalist just spent $10 million on a hole in the ground that could once end the world, and now he's betting it will power the next generation of AI.
The Summary
- Nik Halik bought a decommissioned Cold War nuclear missile silo in Colorado for over $10 million and is converting the 200,000-square-foot, 165-foot-deep facility into an AI data center
- Tech companies are racing to turn nuclear silos and abandoned mines into ultra-secure data centers as geopolitical tensions rise
- The shift signals that AI infrastructure is moving from cloud abstraction to physical fortress, and whoever controls the bunkers may control the compute
The Signal
Nik Halik, an Australian entrepreneur who's skydived over Everest and dove to the Titanic wreck, has spent five years converting a 1959 nuclear missile facility into what he envisions as a modern AI data center. The site once housed three 4.5-megaton nuclear missiles. Now it's becoming compute infrastructure, 165 feet underground with 200,000 square feet of space that was literally designed to survive nuclear war.
This isn't a one-off eccentric project. Multiple tech companies are converting nuclear silos and abandoned mines into hardened data centers as "fears of global conflict grow." That phrase, buried in the second source, is doing a lot of work. When Big Tech starts prepping for kinetic war, they're not being paranoid. They're reading insurance actuarial tables.
"The AI infrastructure race is moving from the cloud to the bunker."
The economics here are interesting. Halik calls himself a "value-facturer" who targets assets he can upgrade. He paid $10 million for a facility the U.S. government spent roughly $350 million to build in today's dollars. That's 97% depreciation on hardened infrastructure with power systems, ventilation, and blast doors already installed. For context, a purpose-built data center runs $6-12 million per megawatt of capacity. A refurbished nuclear silo gives you the structure, the security, and the marketing story for a fraction of greenfield cost.
The timing matters. AI training runs are energy-intensive, geographically concentrated, and increasingly geopolitically sensitive. If you're running frontier models, you need:
- Stable power (nuclear silos were designed for independent grid operation)
- Physical security (literally bombproof)
- Cooling (underground temps are naturally stable)
- Redundancy (these facilities have backup systems for backup systems)
Halik's silo delivers all four. The fact that other companies are making similar moves suggests the market agrees. When infrastructure that was designed to survive the end of the world becomes the preferred home for AI compute, that's not nostalgia. That's threat modeling.
"When tech companies start buying doomsday architecture, they're not making a statement. They're pricing in scenarios most people aren't ready to think about."
The parallel trend is also worth noting: thousands of Americans are joining survival camps like Fortitude Ranch designed to survive catastrophes or World War III. While individuals are prepping bug-out bags, corporations are prepping compute bunkers. Same fear, different scale. One group is worried about food and water. The other is worried about losing access to the models that run half the economy.
The Implication
If you're building in AI, watch where the hyperscalers put their backup infrastructure. When frontier labs start announcing partnerships with hardened facilities, that's a leading indicator about which geographies and threat models they're taking seriously. For governments, this is a reminder that critical AI infrastructure is increasingly private and portable to whoever owns the concrete.
For everyone else, the fact that Cold War relics are becoming hot real estate for the hottest technology tells you something about how fragile the current stack really is. The cloud was never really in the cloud. It was always in warehouses. Now some of those warehouses are 165 feet underground with three-foot-thick blast doors.