Elon Musk says space will soon hold nearly all compute. Google is still finding out if its chips can work there.
The New Stack Steven J. Vaughan-Nichols ● Covered by 2 sources
Google is launching a test satellite with its AI chips to see if space computing can work. Musk wants full AI data centers in orbit; Google is still checking the basics.
Based on reporting by The New Stack, Steven J. Vaughan-Nichols — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Elon Musk is talking as if the future is already decided: compute belongs in space, and soon. SpaceX and Nvidia are aiming to put a space-optimized Vera Rubin AI system in orbit by late 2027, with Starmind AI 1 satellites part of that push. Google is taking a much smaller step first.
Its Project Suncatcher is not a promise of orbital data centers. It is a test. This October, Google plans to launch a refrigerator-sized prototype satellite carrying four Tensor Processing Units, or TPUs, to find out whether its chips can survive the environment at all. The hardware can only run in 15-minute bursts before it has to shut down and cool.
That caution makes sense. Space hardware has to deal with launch vibration, sustained acceleration of up to 10 times Earth gravity, and components that can see 50 to 100 g. Once in orbit, the chips face radiation and thermal conditions that Earth labs can’t fully mimic. Brandon Lucia, a Carnegie Mellon University professor, described the radiation problem as random high-energy particle strikes hitting the insides of a chip, and said cooling is hard because the machines will be basking in the sun all day.
Google’s bigger idea is a constellation of solar-powered satellites linked by high-speed optical connections. The company says it wants those links to run at tens of terabits per second. It has already shown 800 Gbps in each direction, or 1.6 Tbps total, in a bench-scale optical-transceiver setup. But turning that into a real orbital compute network means flying satellites unusually close together, perhaps only hundreds of meters apart.
For now, though, the target is simpler: learn what fails and what survives. Google says this first launch is about seeing what works and applying those findings to later missions. A two-satellite test of the laser communications side is planned for 2027, which suggests the company is still in the proof-of-concept stage while others are already talking about orbiting AI infrastructure like it’s a done deal.
My take — AI-written commentary, not fact-checked reporting
This is the sane way to do space computing: test the chips before declaring victory from a keynote slide. Musk’s “100% of all compute” line sounds grand, but the universe has a nasty habit of ignoring hype. Google may be late to the sermon, but at least it is checking whether the congregation can survive launch.
Read more about this at: The New Stack