Google takes the AI data centre race to outer space

[SAN FRANCISCO] At a lab in San Francisco last month, a team of technicians in protective suits and hairnets poked, prodded and inspected a refrigerator-sized satellite commissioned by Google.

First, they examined the satellite’s solar panels, which would unfurl in space and face the sun.

The satellite was then placed on a table and shaken rapidly in a vibration test, to see whether the chips inside would be damaged or the machine would come apart during a journey into space. The technicians painstakingly painted a small line over each screw to help indicate any loosening during the shaking.

The satellite passed the vibration test, with its screws staying put and the chips remaining intact. James Manyika, Google’s senior vice-president for research, called the results “great”, but he said he still wondered what would happen when the satellite went to space.

The work was part of an ambitious effort from Google called Project Suncatcher, which aims to put artificial intelligence data centres into space and to harness the sun’s energy to power them.

On Oct 1, the satellite from the San Francisco lab – which Google has named MVP – is set to become the first step towards that goal by a major tech company. It will be loaded onto a SpaceX Falcon 9 rocket at the Vandenberg Space Force Base near Santa Barbara, California, and then launched into orbit.

Google provided The New York Times with the first inside look at the project, which seemed like a science-fiction dream just a year ago.

Elon Musk, Jeff Bezos, Sam Altman and others have pledged support for orbital data centres, but the idea faces many challenges, including withstanding radiation in space and the costs of getting machines beyond the atmosphere. Yet as data centres grapple with terrestrial opposition and physical limits, having computing facilities float above Earth has become increasingly attractive.

To be clear, Google is not sending up a full data centre – only an experimental precursor to one. MVP contains four specialised computer chips called “tensor processing units”, which have the computing power of one server in a data centre. The satellite’s solar panels will supply only about 1 kilowatt of power – roughly the amount needed to run a hair dryer – to the chips.

But that will be enough for Google to test its hardware against the harsh conditions of space. The satellite will answer simple AI queries and operate for a year, though it is set to circle Earth for up to six years. Eventually, it is expected to succumb to Earth’s gravitational pull and burn up as it descends through the atmosphere.

Manyika said Google’s expectations for the satellite were tempered.

“We don’t expect, to be perfectly frank, that we’ll have anything usefully operational in the next few years,” he said, comparing the mission to the company’s early efforts to build driverless cars. “Remember how Google was researching for like 15 years before anything showed up? I think this is going to look like that.”

Expanding from one satellite to a vast network of them that operate like a giant data centre will take years and enormous funds, said Brandon Lucia, a professor of electrical and computer engineering at Carnegie Mellon University.

“If you do this on a large scale, there are additional engineering problems,” he said. “That is uncharted waters.”

Project Suncatcher was the brainchild of Blaise Aguera y Arcas, 51, a Google vice-president and AI researcher who leads a team studying intelligence. About three years ago, he attended a gathering of entrepreneurs and AI researchers where the main topic was the escalating energy needs of AI systems. Aguera y Arcas left the event convinced that data centres needed to move to space to harness solar power.

The idea “has been on my mind since I was kid”, he said. “There are long-standing ideas in science fiction about using stars for computation.”

Aguera y Arcas soon took the notion to Manyika, who was convinced it would not work but agreed to tests that might determine whether AI chips could be cooled in space and survive the radiation there.

So in February 2025, Google began taking AI chips to Crocker Nuclear Laboratory in Davis, California. There, an enormous particle accelerator known as a cyclotron, which contains magnets once used in a similar machine during the Manhattan Project, blasted the chips with a radiation dose equivalent to the amount of exposure they would get from five years in space.

Radiation can cause “bit flips”, errors that alter the physics of a circuit to change its binary code from a zero to a one or a one to a zero. These can cause big problems. But the results of Google’s cyclotron tests were promising, with the company finding that restarting the chips could usually reset the bit flips.

In May 2025, Aguera y Arcas joined a meeting that Manyika had arranged to pitch Sundar Pichai, Google’s CEO, on going further with the project. To his surprise, there was another attendee: Sergey Brin, Google’s co-founder.

Brin and Pichai quickly greenlit the project. “OK, so this is a good idea,” Brin had said, according to Aguera y Arcas. “Let’s talk about how we’re doing it.”

With their approval, Project Suncatcher accelerated. Google declined to say how much it is spending but estimates that the cost of orbital data centres will become roughly equivalent to terrestrial ones in the mid-2030s, as the price of sending satellites into space decreases.

The biggest challenge was how to cool the AI chips, which heat up when they perform calculations and process information. Fans, which typically help dissipate the heat, do not work in space. So the Silicon Valley company instead developed a cooling system that uses layers of conductive material to expel the heat into space.

The bottom layer is made up of Google’s AI chips, which sit on a green motherboard. The next layer consists of “thermal interface material”, a pale green putty that comes in sheets like Fruit Roll-Ups and connects the chips to layers of aluminium and copper, radiating heat away from the motherboard. Finally, there is a radiator panel, which projects heat into space.

The chips can operate for about 15 minutes in space before needing to be shut down so they can cool off, said Travis Beals, Google’s senior director of product management for Project Suncatcher. During that time, the chips can process short queries so Google’s AI, known as Gemini, can respond.

Google is already looking ahead. Next year, it plans to launch a pair of satellites. It also has designs for fleets of more than 80 satellites that will fly in close formation and communicate with one another as they process AI queries. And Google is meeting with designers to potentially build a custom satellite the length of a soccer field.

“If five years from now, everything we’ve done has worked perfectly, it probably means we’ve not taken enough risk and we’ve not learned as much as we could,” Beals said.

“If we’re really successful with this in the long run, this will ultimately be boring, and people won’t think anything of the fact that their Gemini query might be getting served in space.” NYTIMES