The same week a scooter guy raised $5 million to put servers in orbit, one of the world's biggest data center investors called it a science project—and he's betting gigawatts on Earth instead.
The Summary
- DigitalBridge CEO Marc Ganzi says AI's next bottleneck is power, not chips or data centers, and he's dismissing space-based data centers as unrealistic infrastructure
- Orbital, founded by ex-Spin CEO Euwyn Poon, just raised $5 million to launch 10,000 space data centers—a plan that sounds ambitious even by Silicon Valley standards
- The split reveals the fundamental question: Do you solve AI's power crisis with new physics or bigger power lines?
The Signal
Marc Ganzi runs DigitalBridge, one of the planet's largest digital infrastructure investors. His company just acquired ArcLight and is building out a massive gigawatt pipeline to power AI data centers on the ground. When Bloomberg asked him about space-based data centers, he didn't hedge. He called them a science project, not a real infrastructure play.
Two days earlier, Euwyn Poon announced a $5 million seed round for Orbital, his startup aiming to launch 10,000 data centers into space. Poon previously built 250,000 e-scooters at Spin. Now he wants to skip terrestrial power grids entirely and run inference in low Earth orbit. The pitch: infinite solar, no cooling costs from ambient heat, and latency that's competitive with ground-based edge computing.
"AI's next bottleneck is power, not just chips or data centers."
Here's the real tension. Both Ganzi and Poon agree on the problem: AI training and inference demand orders of magnitude more electricity than current grids can supply at the speeds companies need. Ganzi's solution is ArcLight and gigawatt-scale terrestrial power infrastructure. Build it where the grid already exists. Upgrade transmission. Add capacity. It's expensive and slow, but it's proven physics.
Poon's solution is orbital. Launch the compute, skip the power lines, harvest solar 24/7. No permitting battles. No substation upgrades. No fighting with utilities over interconnection queues. The problem is you're putting servers in space, which introduces radiation, launch costs, maintenance nightmares, and the non-trivial challenge of beaming processed data back to Earth at scale.
Key contrasts:
- Ganzi is betting billions on terrestrial gigawatt pipelines. Poon raised $5 million to prototype orbital compute.
- DigitalBridge's model is B2B infrastructure at scale. Orbital is venture-backed moonshot territory.
- One sees power as a grid problem. The other sees it as a gravity problem.
The timing matters. Ganzi isn't dismissing space infrastructure because it's conceptually impossible. He's dismissing it because hyperscalers need power in 18 months, not 10 years. Microsoft, Google, and Amazon are signing gigawatt-scale power purchase agreements right now for ground-based data centers. They need inference running next quarter, not next decade after SpaceX has launched 10,000 server racks.
The Implication
If you're building agents or training models today, your infrastructure is Earth-bound. Ganzi is right that the immediate bottleneck is terrestrial power, and the money is flowing there. But Poon's bet isn't about today. It's about what happens when every power grid on the planet is maxed out and inference demand is still climbing. Space data centers sound like science fiction until they're the only option left with room to scale.
Watch where the hyperscalers put their next $10 billion. If it's all ArcLight-style terrestrial deals, Ganzi wins this round. If even one of them writes a check to Orbital or a competitor, the science project just became infrastructure.