On 1 October, a SpaceX rocket lifted off from California with dozens of satellites on board. Among them, a box the size of a fridge. Inside: four artificial intelligence chips designed by Google.
This is the first stage of a project named Suncatcher. Its ambition: one day to install data centres in space.
Why go so far
The answer comes down to one word: energy. The data centres that run AI consume colossal amounts of electricity. According to the International Energy Agency, they could use around 3% of the world's electricity by 2030, twice as much as today.
On Earth, this appetite is causing increasingly visible problems. We have covered them: data centres building their own power plants, local residents who no longer want them.
In space, Google's satellite orbits on a path chosen so that its solar panels are almost never in shadow. It therefore receives sunlight continuously, without needing heavy batteries. The project lead sums it up this way: the sun produces almost all the energy in our solar system, and all the other sources humanity exploits represent only a tiny fraction of it.
What this first satellite does
Google itself presents it as a very minimal test. The main aim is to check that the chips survive the journey and work in space. They will run Gemma, a Google AI model, to answer simple questions.
Before launch, the satellite was shaken to simulate lift-off, placed in a chamber that reproduces the vacuum of space, and its chips were bombarded with particles to test their resistance to radiation.
Next year, Google plans to send two more satellites to test laser communication between them. Eventually, the company envisages clusters of satellites, each equipped with dozens of chips, linked to one another and to Earth by beams of light.
It is a paradox few people know about. A chip that computes generates heat. On Earth, that heat is removed with air or water. In the vacuum of space, there is neither: heat can only escape by radiating, through large radiators. That is why the chips on this satellite will only work in fifteen-minute sessions. And these radiators are among the heaviest parts of the craft, which is expensive at launch time.
Google is not alone
Last year, a young company, Starcloud, already sent an Nvidia chip into orbit. SpaceX says it wants to deploy computing satellites for AI as early as 2028. Nvidia has presented a version of its chips designed for space. The idea, which was science fiction two years ago, has become a race.
The doubts
The obstacles remain immense. The rockets that launch these satellites burn fuel, which puts the environmental balance into perspective. Space debris threatens any installation in orbit. And no one knows when, or even whether, these data centres will become profitable: an analyst at the firm Gartner believes the idea makes no economic sense.
The project lead himself does not think space data centres will replace those on Earth any time soon.
What we take away
This satellite will not revolutionise AI tomorrow. But it says something about the era: AI's hunger for energy has become so great that the world's largest companies are now looking to the sky to feed it.
And there may be another reason, less openly acknowledged: in space, there are no neighbours to complain about the noise.