Google’s first Project Suncatcher satellite reached orbit on Thursday aboard SpaceX’s Transporter-18 rideshare mission, carrying the company’s Tensor Processing Units to test whether AI chips can work in space.
The satellite, built with Planet, will spend the coming weeks gathering data on how the chips cope with radiation and extreme temperatures. Google said it had made contact and operations were normal.
Alongside the launch, Google published a peer-reviewed paper in the journal Joule on the idea behind it: clusters of solar-powered satellites linked by lasers, running machine learning in orbit. Travis Beals, who runs the project, described it to TechCrunch as a long-term moonshot aimed at future demand rather than today’s workloads.
The radiation results are encouraging for one kind of work. Errors were rare enough, around one in a million, for typical inference. Beals said that rate was already a problem for a training run with thousands of chips running for months.
The harder number is launch. The analysis assumes SpaceX gets prices close to $200 a kilogram by 2035 while carrying 200 tonnes per flight. By TechCrunch’s reading of the paper, that means Starship lifting about 370,000 tonnes of payload, roughly 1,800 launches over ten years, or 180 a year.
Starship has never flown more than five times in a year.
The researchers stress that their work is not an economic feasibility study, which leaves the central question open. Proving that a chip survives in orbit takes weeks. Proving that orbit is cheaper than the ground depends on a launch cadence nobody has come close to achieving.
