Google tests its first AI chip in space aboard a SpaceX launch

Google’s prototype orbital compute satellite launched Thursday aboard a SpaceX rocket from California — the first time the company has sent one of its advanced chips into space. Built by Planet Labs, the satellite will test whether a Google Tensor Processing Unit, the company’s rival to Nvidia’s GPUs, can function in orbit: supplying a kilowatt of continuous power, cooling the chip, and running models to check if anything goes wrong.

“We’ve done testing on the ground, but you know, there’s no test that’s completely as good as the real thing,” said Travis Beals, the Google executive managing Project Suncatcher, the company’s plan to build large-scale compute clusters in orbit. Once commissioned, the satellite will fire its TPU in 15-minute bursts to avoid straining its power and thermal systems. It uses a standard Planet Labs platform, but the two companies are already preparing a demo for next year using two satellites built specifically for advanced compute, which will attempt to communicate over a laser link.

Suncatcher isn’t the only space AI payload on this SpaceX rocket, which is carrying more than 100 different payloads, including missions from Satlyt and Cowboy Space Company. What sets Google’s project apart, Beals said, is that it’s a long-term effort: the company envisions a network of 81 satellites flying in close formation and processing in parallel, since bandwidth and latency between TPUs matter enormously for a multi-rack workload — and the rockets needed to scale up orbital data centers cost-effectively don’t yet exist.

On Thursday, Google also released a peer-reviewed version of its white paper on orbital data centers, due to be published in the journal Joule. The paper isn’t an economic feasibility study, the researchers stress, but it lays out how the company sees rocket launch costs falling over time. Pointing to the roughly 20%-a-year price-reducing “learning curve” SpaceX has achieved since the Falcon 1, the authors argue it’s reasonable to expect launch prices near $200 per kilogram by 2035 — a target that would require Starship to fly about 370,000 tons of cargo into orbit, or roughly 1,800 launches over the next decade at 200 tons per mission, for a vehicle that has never flown more than five times in a year.

Google’s research also suggests its chips will survive the radiation of space: the company had to redo particle-accelerator tests after realizing its chip configuration gave the chips more shielding than they would actually get in orbit, producing slightly more errors in the chip’s logic circuitry — though Google remains confident the chips can handle large inference workloads in orbit for a satellite’s five-year lifespan. Beals said the error rate is very low for typical inference operations, about one in a million, though he noted it would already be a problem for a mega-scale training run involving thousands of chips running for months.

Why it matters

If Google’s cost math holds, orbital compute could eventually offer a new way to scale AI infrastructure beyond what Earth-bound power and land constraints allow — but the plan hinges on Starship ramping up to roughly 180 launches a year, more than thirtyfold its current annual record, something SpaceX has yet to demonstrate.