AI's Real Problem Isn't Chips—It's Power. And Space Data Centers? Still Science Fiction.

Marc Ganzi, CEO of DigitalBridge, just told Bloomberg something the industry has been quietly sweating over: AI's bottleneck isn't chips anymore—it's electricity. Data centers are pulling kilowatt-hours like there's no tomorrow, and the real problem isn't building faster processors but guaranteeing gigawatts of stable, affordable power to actually run them. It's unglamorous logistics, but far more urgent than what you'll read in the hype cycle.

That's why DigitalBridge acquired ArcLight, a company focused on energy infrastructure. It's not as flashy as launching satellites, but it's exactly the kind of move that actually moves the needle. The company is building a massive pipeline of power capacity because it understands that the real constraint right now isn't silicon—it's the electrons flowing through it.

Then there's Ganzi's well-deserved skepticism about the trendy idea of space-based data centers. Sounds futuristic, right? Servers orbiting Earth, cooled by the vacuum of space. But according to the CEO, at least for now, it's more of a captivating science project than a real infrastructure strategy. The challenges—launching and maintaining satellites, latency issues, communication overhead—still make the concept too expensive and complicated to scale practically.

So while SpaceX and others have fun with orbital data centers, DigitalBridge is putting real money where the actual problem lives: powering billions of GPTs, LLMs, and AI models that keep multiplying like rabbits. Less flashy, but the kind of unglamorous problem that, once solved, actually changes the game.

The subtext is clear: this isn't space versus earth—it's reality versus hype. In 2026, energy remains the most boring and simultaneously critical piece of the AI puzzle. Terrestrial data centers, surrounded by stable power sources (nuclear, hydroelectric, utility-scale solar), remain the practical option. Space, at least for this generation, stays where it belongs: in research papers and venture capitalist fantasies.