According to Ars Technica, Google has confirmed it’s working on Project Suncatcher, a plan to put AI data centers in space using orbiting TPUs. The company aims to launch its first pair of prototype satellites with TPUs by early 2027, with full-scale deployment targeted for the mid-2030s when launch costs could drop to $200 per kilogram. Google’s research shows solar panels in orbit are up to eight times more efficient than Earth-based systems, and their TPUs can handle almost 2 krad of radiation before data corruption occurs. The satellites would maintain proximity within a kilometer using free-space optical links that have demonstrated 1.6 Tbps speeds in early testing. This represents Google’s latest “moonshot” project following its 15-year journey from early self-driving cars to Waymo’s near-autonomous vehicles.
Sponsored content — provided for informational and promotional purposes.
<h2 id="why-space-makes-sense”>Why space makes sense
Here’s the thing about Earth-based data centers: they’re becoming increasingly problematic. They guzzle enormous amounts of electricity and water, they generate significant heat and noise, and communities are increasingly pushing back against having them built nearby. Space solves a lot of these issues in one go. You get nearly constant sunlight for solar power without atmospheric interference, no neighbors to complain, and unlimited real estate. Basically, it’s the ultimate NIMBY solution.
The real challenges
But let’s be real – this isn’t just about strapping some servers to a rocket. The engineering hurdles are massive. Keeping satellites within a kilometer of each other while orbiting at thousands of miles per hour? That’s precision flying on an unprecedented scale. Then there’s the radiation problem – space is brutal on electronics. Google‘s testing their TPUs with proton beams is smart, but space radiation is unpredictable. And what about maintenance? When something breaks on Earth, you send a technician. When your orbital TPU fails, you’re looking at a multi-million dollar replacement mission.
The bigger picture
What’s fascinating is how this fits into the broader space infrastructure trend. We’ve already seen satellite constellations like Starlink prove that orbital networks are viable. Now we’re talking about moving actual compute power up there. If this works, it could fundamentally change how we think about cloud computing. Instead of building massive data centers in specific locations, we could have distributed computing power literally orbiting the planet. The implications for global connectivity and low-latency services are pretty mind-blowing.
Is this realistic?
I’ll be honest – when I first read about this, it sounded like pure science fiction. But then you look at the details in Google’s pre-print study and realize they’re actually doing the hard engineering work. They’re not just dreaming – they’re running radiation tests, developing formation flying models, and planning concrete prototype launches. The 2027 timeline for first prototypes feels aggressive but plausible. And if launch costs really do drop to $200/kg by the 2030s, the economics start making sense. We’re probably a decade away from this being commercially viable, but the fact that Google is seriously pursuing it tells you something about where they see the future of AI compute heading.
