Google is sending two satellites to orbit in early 2027 to test whether artificial intelligence can run in space. It sounds like science fiction, but the engineering problem is surprisingly concrete: can the chips that power today's AI models survive the radiation, extreme temperatures, and isolation of orbit?
The project, called Suncatcher, would use Google's TPU chips — the same processors running Gemini 3 — mounted on solar-powered satellites 400 miles above Earth. Data would travel between satellites and ground stations via laser beams. On paper, the appeal is obvious. Space-based data centers could sidestep two of the biggest headaches for Earth-based servers: the massive power draw needed to keep them cool, and the physical space required to house them.
Google isn't alone in this thinking. SpaceX is exploring similar ideas, as are several smaller startups planning to launch AI-equipped satellites within the next few years. The logic is sound enough that multiple teams are willing to bet significant resources on it.
We're a new kind of news feed.
Regular news is designed to drain you. We're a non-profit built to restore you. Every story we publish is scored for impact, progress, and hope.
Start Your News DetoxBut the gap between "sounds good" and "actually works in space" is enormous. Thermal management without air or water cooling is a brutal constraint. Solar storms and space debris pose constant threats. Laser communication systems need to maintain the speed and reliability of fiber optic cables — a technical challenge that hasn't been fully solved at scale. And if something breaks 400 miles up, you can't send a technician. Repairs would require expensive robotic missions or launching replacement satellites.
Then there's the economics question, which may matter more than the engineering. Space-based computing only makes financial sense if you can deploy it across dozens or hundreds of satellites, and if launch costs keep falling. Right now, neither condition is guaranteed.
A successful 2027 demonstration would be real progress — proof that AI chips can function in orbit. But it would also be just a beginning. Building an actual network of space data centers would take years of solving problems we haven't even encountered yet. If this ever becomes mainstream, it's a decades-long story, not a 2027 story.
For now, Google has a genuinely hard engineering problem and the resources to tackle it. Whether the solution turns out to be worth the cost is a question the market will answer much later.
