Deutsche Bank highlights four roadblocks for tech CEOs’ obsession: Data centers in space

Data centers. In space.

It’s the hottest topic among CEOs with skin in the data center or rocket-launching business, with Tesla (and SpaceX) CEO Elon Musk and Jeff Bezos, former Amazon CEO and founder of Blue Origin, purportedly engaged in this race to make one small step for man and a giant leap for AI-kind. OpenAI CEO Sam Altman reportedly wants in, too.

Google CEO Sundar Pichai thinks it’s a matter of when, not if, recently telling Fox News, “There is no doubt to me that, a decade or so away, we will be viewing it as a more normal way to build data centers.”

Earth may be an extremely hospitable environment to live for those of us who spent millions of years evolving to do so, but that’s less true for AI data centers.

Tyler Norris, Google’s head of market innovation on the advanced energy team, summarized the problem as such:

In other words, the institutional and political difficulties associated with getting data centers on Earth built and connected to the grid are seemingly more difficult than the engineering challenges associated with having these projects operating in outer space.

But Deutsche Bank analysts led by Edison Yu flagged four obstacles that would need to be tackled for this to move from sci-fi to the nonfiction section.

• Rocket launches cost too much. Yu cited estimates from Google indicating that launch costs would need to be below $200 per kilogram (or about $441 per pound, for heathens) to be viable, and currently pins costs at about $1,500 per kilogram. It doesn’t take a rocket scientist to tell you that means costs need to go down by about 87%, though it will likely take a legion of rocket scientists to figure out how to make that happen. This “requires SpaceX Starship to be operational and launching on a regular cadence,” he writes.

• Space is cold but bad at cooling. “To properly cool a large AI cluster, a data center satellite would require massive passive radiator panels,” per Yu. “Therefore, we think some type of breakthrough is required in the radiator design to make the data center truly viable.”

• Space makes Burry more right about depreciation. Remember Michael Burry’s argument that companies were understating GPU depreciation costs? Well, this critique about chips having a shorter useful lifespan becomes a lot more salient in space (especially for high-bandwidth memory chips, per a white paper from Google). “Radiation can cause faster degradation of chips. Cosmic rays and high-energy protons constantly bombard the satellite,” Yu wrote. “There are some simple solutions for this such as wrapping the servers in heavy lead or aluminum. However, this would naturally add mass to the satellite.”

• Fixing stuff in space is hard. Hell, fixing stuff on Earth is hard enough! It took me seven weeks to get my landlord to get a technician to repair my air-conditioning unit this year. Yu believes “the cost of building [an orbital transfer vehicle] capable of carrying advanced maneuvers is too expensive,” so satellites would need better hardware to increase the likelihood that nothing goes wrong, and that would drive up the cost of production.

Now, tech CEOs don’t just wax eloquent about stuff that’s straight out of science fiction for the mystique (though that’s probably a part of it). Pursuits that seem so futuristic that they border on the absurd on first hearing are massive potential moneymaking opportunities. (If there’s a flying autonomous car available in my lifetime, shut up and take my money!)

The good news? This flight beyond Earth’s atmosphere is not a flight of fancy, according to Deutsche Bank.

“There are clearly technical challenges to making this a viable endeavor but these seem to be engineering constraints as opposed to physics,” Yu wrote.