Data centers present sprawling engineering and political problems, with ravenous appetites for land and resources. Building them on Earth has proven problematic enough — so why is everyone suddenly talking about launching them into space?
Data centers are giant warehouses for computer chips that run continuously, with up to hundreds of thousands of processors packed closely together taking up a mammoth footprint: An Indiana data center complex run by Amazon, for example, takes up more real estate than seven football stadiums. To operate nonstop, they consume immense amounts of electricity, which in turn is converted to intense heat, requiring constant cooling with fans and pumped-in water.
Fueled by the ongoing boom in artificial intelligence, Big Tech is so desperate to power its data centers that Microsoft successfully convinced the Trump administration to restart operations at the benighted Three Mile Island nuclear plant in Pennsylvania.
The data center surge has spawned a backlash, as communities grow skeptical about their environmental toll and ultimate utility of the machine learning systems they serve.
It’s in this climate that technologists, investors, and the world’s richest humans are now talking about bypassing Earth and its logistical hurdles by putting data centers in space. And if you take at face value the words of tech barons whose wealth in no small part relies on overstating what their companies may someday achieve, they’re not just novel but inevitable. The Wall Street Journal reported last month that Jeff Bezos’s space launch firm Blue Origin has been working on an orbital data center project for over a year. Elon Musk, not known for accurate predictions, has publicly committed SpaceX to putting AI data centers in orbit. “There’s no doubt to me that a decade or so away we’ll be viewing it as a more normal way to build data centers,” Google CEO Sundar Pichai recently told Fox News.
The prospect of taking a trillion-dollar industry that is already experiencing a historic boom and literally shooting it toward the moon has understandably created a frenzy within a frenzy.
But large questions remain: Is it even possible? And if it is, why bother?
Orbital computing boosters claim the reason is simple: Data centers are very hot. Space, as sci-fi teaches us, is very cold. Data centers need a lot of energy, and the sun produces an effectively infinite supply of it. The thinking goes that with free ambient cooling and constant access to solar power (unlike terrestrial solar panels, these wouldn’t have to contend with Earth’s rotation or atmosphere), an orbital data center could beam its information back to our planet with few earthly downsides.
Experts who spoke to The Intercept say it’s nowhere near this simple. Despite the fact that putting small objects like satellites into orbit has become significantly cheaper than decades past, doing anything in space remains an extremely expensive and difficult enterprise compared to doing it on the ground. And even if the engineering problems are surmountable, some question the point.
There are varying visions of space data centers. Musk’s idea seems to be based on constellations of smaller satellites carrying computing hardware; others envision massive spacecraft the size of skyscrapers filled with graphics-processing units.
“If you wanted to spend enough money, you could absolutely put GPUs in space and have them do the things that data centers are supposed to do,” Matthew Buckley, a theoretical physicist at Rutgers University, told The Intercept. “The reason that I would say it is an incredibly stupid idea is that in order to make them work, you’re going to have to spend incredible amounts of money to keep them from melting. And you could solve that problem much easier by not launching them into space. And it is unclear why on earth you would want to do that.”
“You’re going to have to spend incredible amounts of money to keep them from melting. And you could solve that problem much easier by not launching them into space.”
Outer space is largely a cold vacuum, but objects in Earth’s orbit are subjected to temperature extremes. Ali Hajimiri, an electrical engineering professor at Caltech, pushed back on the “general notion of a cold vacuum of space. Actually space can become very cold or very hot.” The International Space Station, carrying a computer payload producing a mere fraction of the heat of a large-scale data center, has to carefully contend with temperatures of between 250 and -250 degrees Fahrenheit depending on whether it’s exposed to direct sunlight. But even when an object in orbit is subjected to extreme cold temperatures, the nature of space’s vacuum behaves drastically differently than hot and cold within our atmosphere.
On Earth, you can remove a boiling kettle from the stove and the energy within will gradually transfer to the surrounding air, cooling the vessel and its contents back to room temperature. In space, there is no air, water, or other medium to which one can transfer heat, thus the coldness of space would do nothing to cool a scorching hot piece of silicon. “If you put a GPU in space and powered it, it would melt,” said Buckley.
“Heavy is not good for space.”
Without ambient air or any other medium to ferry away heat through convection, a hypothetical space data center would need to rely on thermal radiation. Washington-based Starcloud is among the most prominent startups pitching orbital data centers as a concept, and says it’s working to build a 5 gigawatt space facility, a staggering figure that represents about 10 percent of all electricity currently consumed by data centers on Earth, according to a recent Goldman Sachs estimate. Starcloud says it would get rid of the astounding amount of heat generated in such a facility through the use of enormous radiators — essentially large pieces of metal that absorb the heat directly from the onboard chips and then radiates it out into space. Physics dictates that this would require radiators unlike anything that’s ever been constructed: Starcloud says it would use 16 square kilometers of radiators, taller and wider than four Burj Khalifa skyscrapers stacked end to end. How such a thing would be launched into or constructed in space, a project without any precedent, is unclear.
“If you want to create this heat transfer system, either heat pipes and all those things, those things are heavy,” Hajimiri said. “And heavy is not good for space.”
Then there’s the sun. Proponents of space data centers also point to the fact that a solar panel in space can receive uninterrupted solar energy without diminishment from weather or Earth’s atmosphere. But all of this sunlight generates extreme heat of its own, requiring further cooling. And any efficiency gained by putting the panels closer to the sun, argued Buckley, is largely negated by the extreme inefficiency of having to put them into space in the first place.
Other unsolved problems abound. While space is thought of as empty, it’s filled with radiation that can damage computer hardware or corrupt the data stored within. Earth’s orbit is also filled with debris. This orbiting space trash presents the biggest hurdle, according to John Crassidis, a mechanical and aerospace engineering professor at the University of Buffalo. Near-misses and space junk collisions are a real danger for satellites — objects a small fraction of the size of mammoth orbiting data centers. Last month, Starlink executive Michael Nicolls announced one of the company’s satellites — infinitesimal compared to Starcloud’s plan — nearly collided with a Chinese satellite. “This stuff’s going 17,500 miles per hour,” Crassidis said of space debris, and even contact with a tiny fragment could be catastrophic. “It doesn’t take too big of a hole. I think it’s half an inch radius to explode the whole [International] Space Station.”
“I think it’s half an inch radius to explode the whole Space Station.”
Though Crassidis doesn’t object to companies pursuing these projects, he cautions that flooding Earth’s orbit with chip-ferrying satellites could make a dangerous situation worse. He pointed to Kessler syndrome, a theoretical scenario in which low Earth orbit becomes so crowded with objects and trash that it becomes unusable by humans.
Any floating data center would also have to contend with the difficulties of communicating between space and Earth; even Starlink’s broadband satellites are extremely slow compared to the fiber optic connections plugged into terrestrial data centers. University of Pittsburgh electrical and computer engineering department chair Alan George told The Intercept that sending data between Earth and space is just one of “many extreme challenges to overcome.” And if it can’t be solved, the whole endeavor is for naught. “Bold claims are being made based upon technologies that don’t yet exist,” he said.
“If you have hundreds of billions of dollars, you can launch enough infrastructure to keep it cool. Why would you do that when you can just put it an ugly building at the end of the block?” he said. “I’m not saying that you could never do this if you just decided to set money on fire. I’m just saying I don’t understand the motivation to do this.”
The motivation may be as financial as it is scientific. SpaceX is rumored to be approaching an initial public offering that could potentially be bolstered by plans for orbiting data centers, and any Big Tech entity knows it can reap publicity and share price benefits by mentioning “AI” at any available opportunity. Space is trendy, “AI” is booming (or bubbling), and the combination of the two could spur further investment.
Starcloud co-founder and CEO Philip Johnston was unfazed by these challenges in an interview with The Intercept. He said his company’s vision of a 5-gigawatt facility is 10 to 15 years away, by which point he believes SpaceX launches will be so frequent and carry such huge payloads that bringing the raw materials to orbit shouldn’t be difficult. Johnston dismissed as “annoying” criticism of his company’s plan to cool hot chips in space. “Nothing we’re doing is against the laws of physics and nothing requires new physics to make it work. It’s not like we’re building a fusion reactor.”
In his view, it’s simply a matter of scaling up existing technology. Johnston said he doesn’t believe his company will compete with Earth-based facilities for several years, at which point he thinks Starcloud will begin launching large constellations of smaller satellites carrying computing hardware that will mesh together, rather than one giant object. This modular approach, Johnston said, will also take care of the obsolescence issue: Older hardware can simply be left to burn up upon reentering the Earth’s atmosphere. For the time being, he said the company will cater to the specialized needs like processing satellite imagery, with potential customers including the U.S. Department of War. The company counts In-Q-Tel, the venture capital arm of the U.S. intelligence community, among its backers. Johnston told The Intercept that the “CIA is interested in what we’re doing,” but declined to comment further.
Experts who spoke with The Intercept didn’t wholly oppose these projects because the sheer enormity of the challenge could yield engineering breakthroughs. But many also suggested that the mammoth investment in resources and ingenuity required would be better spent on the surface.
Hajimiri says he believes the engineering problems could be solved eventually, and that crazy ideas can yield scientific and societal benefits. A decade ago, he pursued a similar project on a far smaller scale. He and his team dropped it for simple reason: Chips need to be replaced. The processors used to train state-of-the-art large language models are rendered obsolete in a matter of years. It’s this need for newer and better chips that has taken the value of chipmakers like Nvidia into the stratosphere. But it’s not just buying the latest and greatest. Things go wrong: Processors sometimes fail, power supplies burn out, wiring needs to be fixed. In earthly data centers, the solution is easy. Technicians use their hands to pop in a replacement processor, for example.
“Data centers need full-time humans to deal with the occasional hardware emergencies,” said Dimitrios Nikolopoulos, an engineering professor at Virginia Tech who works on high-performance computing. “And I don’t know how this is gonna be dealt with in space.” Johnston predicted that robot repairmen would eventually solve this problem.
We’re independent of corporate interests — and powered by members. Join us.
Join Our Newsletter
Thank You For Joining!
Original reporting. Fearless journalism. Delivered to you.
Will you take the next step to support our independent journalism by becoming a member of The Intercept?
By signing up, I agree to receive emails from The Intercept and to the Privacy Policy and Terms of Use.
When an orbital data center’s hardware grows obsolete, companies would need to figure out how to upgrade them. Otherwise it becomes a piece of space trash two-and-a-half miles across.
Jesse Jenkins, an engineering professor at Princeton who works on energy technologies, said the tech world is simply looking in the wrong place. “The fact that we are considering building data centers in space because it’s too hard to build and power them on land should be an indictment of our ability to deploy new energy and data infrastructure at scale in the United States.”
The biggest problem is the simplest, said veteran aerospace engineer Andrew McCalip. Though the cost of putting things in space has decreased dramatically, it’s still vastly greater than building a data center on land. “Can we host a GPU in space cheaper than hosting it in a building in Oregon?” he asked. The answer remains an emphatic no.
McCalip is also skeptical of Johnston’s claim that Starcloud represents a green alternative to terrestrial data centers. Launching craft large enough and frequently enough to make orbital data centers feasible would require infeasibly vast volumes of liquid oxygen fuel, McCalip said, and manufacturing enough to match the ambitions of SpaceX (and other companies hoping to hitch a ride to orbit) would likely entail burning a lot of fossil fuels.
It’s enough to make you ask once more: Why do all of this in space?
“The benefit,” McCalip said, “would be this sort of vague ‘Humanity gets better at doing things in space.’”