According to DCD, Google is partnering with Planet Labs on Project Suncatcher to launch TPU AI chips into space, with the first two satellites scheduled for early 2027. The company’s research paper outlines a vision for 81-satellite clusters forming 1km-radius arrays in low-earth orbit. Google CEO Sundar Pichai acknowledged this “moonshot” faces complex engineering challenges, though early radiation testing shows their Trillium TPUs survived simulated space conditions. The project comes as multiple companies including SpaceX, Blue Origin, and startups like Starcloud-1 are pursuing space data centers, creating a new orbital computing race.
Sponsored content — provided for informational and promotional purposes.
Why even bother with space data centers?
Here’s the thing – this isn’t just about being cool (though it’s definitely that). Google’s paper suggests that if launch costs drop to around $200 per kilogram – which they theorize could happen by 2035 if SpaceX’s Starship launches frequently – then the economics start making sense. Basically, the cost of launching hardware could become comparable to terrestrial energy costs. And in space, you’ve got unlimited solar power and natural cooling through radiation. But we’re talking about needing Starship to launch 180 times per year to hit those numbers. That’s… ambitious to say the least.
The technical nightmares they’re facing
Look, putting data centers in space isn’t just about launching servers and calling it a day. Google identified some massive hurdles. First, networking – their terrestrial data centers use custom optical interconnects that handle hundreds of gigabits per second. Current satellite links max out around 100Gbps. Their solution? Fly satellites incredibly close together – we’re talking hundreds of kilometers or less – to enable multiple high-speed optical beams between them. But maintaining that tight formation at 650km altitude has never been done before.
Then there’s radiation. They tested their TPUs in a particle accelerator simulating five years of space radiation, and while the chips survived, the high-bandwidth memory had uncorrectable errors. Google says it’s “likely acceptable for inference,” but training AI models? That requires more study. And when things break? No human is going up there to replace a TPU. Their solution is simple: just pack extra chips as backups.
Everyone wants in on space computing
This isn’t just Google’s wild idea. Elon Musk says SpaceX “will be doing” space data centers. Jeff Bezos predicts gigawatt data centers in space within 10+ years. Even former Google CEO Eric Schmidt bought a rocket company specifically for orbital data centers. And we’ve already seen Starcloud-1 launch with an Nvidia H100 chip, proposing a massive 5GW data center across a 4km solar array.
Google’s approach is different though – they’re betting on modular clusters of smaller satellites rather than massive single structures. Their research paper argues that building huge monolithic data centers in space introduces its own problems: assembly challenges, clumsy collision avoidance, and added structural mass. But the modular approach creates its own networking headaches. It’s basically choosing your poison.
Let’s be real about the timeline
Two satellites by 2027 sounds manageable. But 81-satellite clusters forming kilometer-wide arrays? That feels like science fiction with today’s technology. The launch cost assumptions alone require Starship to become the workhorse SpaceX promises, and we haven’t even talked about the thermal management challenges or the fact that space is, you know, trying to kill electronics constantly.
Still, the fact that Google is publishing serious research about this tells you something. They’ve clearly done the math and decided this is worth exploring. Whether Project Suncatcher becomes the future of computing or just another ambitious Google moonshot that fizzles out… well, we’ll find out starting in 2027.
