Data centers. In space.
It’s quickly becoming one of the most buzzed-about ideas among tech CEOs with skin in either the AI or rocket-launching game. Elon Musk, via Tesla and SpaceX, and Jeff Bezos, through Blue Origin, are both reportedly circling the concept. Sam Altman wants in too. Google $GOOGL ( ▲ 0.72% ) CEO Sundar Pichai recently said it’s a matter of when, not if, arguing that within a decade orbital data centers could become “a more normal way” to build AI infrastructure.
The logic isn’t totally wild. Earth is great for humans, but it’s increasingly hostile territory for AI data centers. Between grid bottlenecks, permitting delays, and political resistance, connecting massive compute hubs on land is often harder than the engineering challenge of operating them in orbit. Even Google’s own energy team has suggested the institutional friction on Earth may be tougher than the physics of space.
The Launch Cost Problem
Deutsche Bank analyst Edison Yu lays out the first and biggest hurdle: rockets are still too expensive.
Citing Google estimates, Yu says launch costs would need to fall below $200 per kilogram to make orbital data centers viable. Today, they sit closer to $1,500 per kilogram. That implies an 87% cost reduction, which Yu says would require SpaceX’s Starship to become fully operational and launch at a regular cadence. In other words, this entire idea hinges on Starship actually delivering on its promise.
Cooling Isn’t as Easy as Space Is Cold
Space may be cold, but it turns out it’s not great at cooling.
According to Yu, large AI clusters in orbit would need massive passive radiator panels to dissipate heat. Without a major breakthrough in radiator design, cooling alone could make space-based data centers impractical. Simply put, the vacuum of space doesn’t solve thermal management the way people assume it does.
Depreciation Gets Even Uglier in Orbit
Remember Michael Burry’s argument that companies are understating GPU depreciation? Space might prove him even more right.
Yu notes that radiation exposure shortens chip lifespans, especially for high-bandwidth memory. Cosmic rays and high-energy protons constantly bombard satellites, accelerating degradation. While shielding with heavy materials like lead or aluminum could help, that adds mass, which in turn raises launch costs. It’s a vicious cycle.
Good Luck Fixing It
Finally, there’s maintenance.
Fixing broken equipment on Earth is hard enough. In space, it’s exponentially worse. Yu argues that orbital transfer vehicles capable of advanced maneuvers would be prohibitively expensive, meaning satellites would need far more robust hardware upfront to reduce the odds of failure. That pushes manufacturing costs even higher.
Still, Deutsche Bank doesn’t think the idea belongs in the sci-fi bin just yet.
“There are clearly technical challenges to making this a viable endeavor,” Yu writes, “but these seem to be engineering constraints as opposed to physics.”
In other words: not impossible: just brutally expensive, complicated, and very far from ready.