Google launched its first orbital compute satellite today aboard a SpaceX rocket from California, sending one of its Tensor Processing Units - the chip it hopes will eventually make Nvidia sweat in zero gravity - into the great beyond. Built by Planet Labs, the satellite will test whether a TPU can supply a kilowatt of continuous power, stay cool, and run a series of models without throwing a tantrum in orbit.

"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 around the Earth. Once commissioned, the satellite will fire up its TPU in 15-minute bursts to avoid straining its power and thermal systems. A more purpose-built demo with two satellites is expected next year, and those versions will attempt to collaborate via a laser communications link.

Suncatcher isn't the only space AI payload on this rocket, which is carrying more than 100 different payloads, including missions from Satlyt and Cowboy Space Company. What sets Google apart - from those startups and from SpaceX itself - is that this is a long-term project. The "long-term moonshot," as Beals calls it, envisions a network of 81 satellites flying in close formation, processing in parallel.

"The bandwidth and the latency between TPUs really, really matters when you're trying to run a multi-rack workload…we're trying to look ahead to not just what workloads exist today, but where they will be in five years," Beals said. The rockets needed to scale up orbital data centers cost-effectively, of course, don't yet exist.

On Thursday, Google also released a peer-reviewed version of its white paper on orbital data centers, to be published in Joule. The paper offers one of the most rigorous analyses available of how compute gets to orbit. While the researchers stress it isn't an economic feasibility study, it paints a picture of rockets getting cheaper over time.

Like all data center companies, Google is looking to SpaceX to get its spacecraft off the ground. (Google is also a major investor in SpaceX.) Arguing that Elon Musk's rocket builders have achieved a price-reducing "learning curve" of about 20% a year since the Falcon 1, the authors believe launch prices could approach $200 per kilogram by 2035.

What will it take? Based on Falcon 9's payload history, they think that trajectory requires Starship to fly 370,000 tons of payload into orbit - roughly 1,800 launches over the next 10 years, or 180 a year, assuming it can fly 200 metric tons per mission. That's a big ask for a vehicle that has never flown more than five times in a year. SpaceX predicts far more; Elon Musk has suggested Starship could achieve an hourly flight rate in 2029, but Musk says a lot of things.

The good news in Google's updated research is that its chips will likely survive space radiation. The company had to redo tests blasting the chips in a particle accelerator after realizing the configuration provided more shielding than they would actually experience. This produced slightly more errors in the chip's logic circuitry, but Google is still confident its chips can handle large inference workloads in orbit for a satellite's five-year lifespan.

"The error rate is very low if you're thinking about typical inference operations, right? Like one in a million," Beals said. "On the other hand, it was already problematic for doing, say, some mega-scale training run where you're going to have many thousands of chips running for months."