The next big technology step is to put data centers in place. Can it work?

Standing in front of a friendly crowd in March, Elon Musk laid out his plan for the future of his companies, and it was really out of this world.

Musk announced that his space development company, SpaceX, which had recently merged with his artificial intelligence company, xAI, would install data centers around the Earth.

It all depends on electricity, he explained. He said: “You have little power on Earth.” “The place has the advantage that it is always sunny.”

Musk envisions clusters of data-gathering satellites orbiting the planet, powering the AI ​​revolution from above. It’s a great place to advertise SpaceX. This week, Bloomberg reported that the company has filed confidential documents with the Securities and Exchange Commission with the goal of going public this summer.

Musk also says it makes financial sense. “I actually think the cost of putting AI into space will drop below the cost of global AI sooner than most people expect,” he said. “I think it will only be two or three years.”

Others have doubts. Musk’s timing is an “optimistic interpretation,” according to Brandon Lucia, a professor of electrical and computer engineering at Carnegie Mellon University who specializes in putting computers on satellites. The napkin figures look impressive, and the energy is free after that – but it seems that there are many obstacles to building a data center among the stars.

The problem of the world of energy

Here on Earth, the problem is obvious: AI is powering the world. Global energy consumption is expected to nearly double to about 1,000 terawatt hours by the end of the decade, according to estimates by the International Energy Agency.

To bridge this gap, some companies are building dedicated gas turbines, while others are investing in nuclear technology. It’s not enough, according to Philip Johnston, CEO and co-founder of Starcloud, which wants to build orbital data centers.

“We’re still facing the challenges of how you can build new energy jobs in the world,” Johnston said. “In six months, they’ll be leaving the chips in the warehouse because they don’t have the power to open them.”

Starcloud announced its first spacecraft this fall with an Nvidia H100 chip on board. The company has demonstrated the ability to launch the Google Gemini AI model from space, and plans to launch a second spacecraft in October. “That’s 100 times more powerful than the original,” Johnston said, although it’s still only expected to produce 8 kilowatts of power.

Google is also pursuing the idea of ​​building data centers in space with a project known as Suncatcher. It is considering a group of 81 satellites that it plans to build in partnership with the satellite imaging company Planet. Two prototype satellites will launch in early 2027, according to the companies.

“Orbital data centers are an idea whose time has come,” Will Marshall, CEO of Planet, wrote to NPR in an email. “Of course it will cost more than the world can negotiate, but now is the time to deal with this.”

Everything has to be big

Going from prototype satellites to something useful is not that easy. Among other things, the required power of microchips used for artificial intelligence is very large.

To understand how much power is needed, consider the largest power generating facility in space today: the International Space Station (ISS).

The ISS’s solar panels are about half the size of a football field and produce about 100 kilowatts of power, according to Olivier de Weck, a professor of astronomy at the Massachusetts Institute of Technology. “It’s basically the amount of power that one big car engine produces.”

Replicating a 100-megawatt data center in space would require a facility 500 to 1,000 times larger, depending on the orbit.

“Is that possible? Yes, I think it is possible, but not next year and certainly not for three years,” he said.

And power is not the only thing needed; satellites must also provide recovery to microchips. While it is true that space is cold, it is also empty. This means that when a satellite gets hot, there is no easy way to get rid of that heat – it just keeps on increasing.

Rebekah Reed, a former NASA official now at Harvard University’s Belfer Center for Science and International Affairs, said:

The best solution is radiators, which direct water to large panels where the heat can be dissipated. So in addition to solar panels, the AI ​​satellite will need another set of large radiators.

“When you put these large radiators together with the large solar arrays needed to power and cool them, you’re talking about very large satellites, or very large constellations,” Reed said.

Another way is to build small satellites and fly them in fixed patterns called constellations. Such constellations allow heat and energy problems to be distributed, but to function, these satellites would have to send a lot of information back and forth. That probably means using lasers to beam data between satellites. But even traveling at the speed of light, the time it takes to get data from one satellite to another is long enough to slow down a computer.

Google’s Project Suncatcher moves flying groups of satellites in very tight groups to reduce that latency. Meanwhile, Musk plans to produce more than a million satellites and place them in orbit around the poles of the Earth. He recently unveiled the first generation of the AI ​​Sat Mini drone – which has solar arrays up to 180 meters (about 600 feet) – during his presentation.

Putting all that stuff in place would cost money – a lot of money. Currently, it can cost $1,000 per kilogram to launch a satellite into orbit. Google believes that costs must drop by at least a fifth to $200 per kilogram before data centers in space start to make sense.

Musk thinks he can do it with his new Starship rocket, which is still in development. Starcloud’s Johnston says Starship is central to more than just SpaceX’s vision. He told investors: “If you don’t think Starship is going to work, don’t invest in us – that’s totally fine.”

To update the server

Even if a company can get a data center in space, running it will involve a lot more than moving microchips into orbit.

Data centers on Earth are not stable buildings full of depreciating shifts, says Raul Martynek, CEO of DataBank, a company that manages 75 data centers, mostly located in the United States. They require constant maintenance and improvement, all of which are done by the staff.

Take DataBank’s IAD1 data center in Ashburn, Virginia. The space is 144,000 square feet filled with rows and rows of black computer cabinets, filled with microprocessors. It’s still efficient, as these devices go, but it still uses about 13 megawatts of power at any given time (130 times more than the International Space Station).

“We have vendors here every day,” says James Mathes, who runs IAD1.

Workers are constantly in and out of these data centers, installing new equipment, updating microchips and making repairs. And to stay competitive, local data centers will need to do the same.

Some of that can be done in software, and Musk points out that the chips can be rigorously tested on the ground before being sent up. But the fact remains that companies that lease data centers often want to physically access them for another reason.

Martynek, who has spent decades in telecom, says he’s not worried about local data centers taking away business from his company.

“It seems like there are a lot of ifs and developments that will have to happen, and I find it hard to believe that everything can happen in two or three years,” he said. “Nobody in data center country is losing sleep.”