Google’s parent company Alphabet has already put the robots on your streets with Waymo, but that clearly wasn’t ambitious enough. Google has now announced Project Suncatcher, a self-described “moonshot” that wants to start putting the robots in space. In fact, the specific idea is to put the actual AI data centers themselves into orbit onboard a tightly clustered satellite formation. This is the actual ‘brain’ of the AI, which would then beam its thoughts down to Earth. Yes really.
That may seem like a pretty tall order, and Google’s own preprint paper admits that there are a lot of technological hurdles to clear before this becomes a reality. Still, the craziest part might be that each individual step in this plan seems plausible. Put a bunch of Google’s tensor processors on a small satellite? Simple enough. Launch a bunch of them into space? Low-Earth orbit satellite constellations are happening already, and everybody wants a piece of that action. Have those satellites optically communicate with one another? Doable. Then just send whatever the processors processed back down to Google on Earth. I see what they’re aiming at here!
If you’re wondering why Google would want to launch data centers into orbit, the answer is because AI requires immense amounts of electricity; the technology is expected to be nearly 12 percent of national energy consumption by 2030. That has huge consequences for greenhouse gas emissions, electrical bills (even for ordinary residents who live nearby), and grid stability. Beyond that, cooling all those processors down requires huge amounts of water. In other words, building the robots on Earth is bad. But in space, you avoid all these problems at once!
How to build a space brain
The exact plan is pretty wild. Since putting an entire data center onto one satellite would make for something too heavy (and expensive) to launch, Project Suncatcher would instead use a constellation of smaller satellites. But this isn’t like SpaceX’s Starlink: the constellation would fly in formation, a tight grouping of 81 satellites that would have to maneuver constantly to maintain optimum communication. The key here is that all 81 of them, combined, make for one single data center. They therefore have to talk to one another very rapidly to actually perform their calculations, or else the whole thing would be too slow to compete with a terrestrial center.
They would fly in a dawn-dusk orbit, meaning they would live in a perennial golden hour. That maximizes the amount of solar power they get, fully eight times more than what panels get on Earth. Electrical needs solved. As for cooling, well, they’re in space! Some engineering details still need to be worked out, but that’s the universal heat sink right there. No groundwater required.
Making science-fiction into fact
And Google’s not alone: NVIDIA, the kind of AI chipmakers, is also experimenting with space robots. The company intends to launch a test satellite in the near future, though its plans (for now) seem less complex or ambitious than Google’s. As for the search giant, CEO Sundar Pichai says that he hopes to have some test satellites in orbit sometime in 2027.
Will any of this lead to something tangible? It is an enormous challenge, and again, Google hasn’t worked out all the details. They do describe it as a “moonshot.” Still, it’s crazy how plausible it actually sounds when you break it out. And it really would be nice to solve AI’s ravenous hunger without consuming any of Earth’s resources. The hidden threat, of course, is that the more satellites go into orbit, the higher likelihood of crashes, which have serious consequences for us all.
