Google Needs SpaceX to Launch Starship 1,800 Times – Here’s Why
Google space data centers took a major step forward today, as the company’s prototype orbital compute satellite launched onboard a SpaceX rocket from California, marking the first time Google has sent one of its advanced chips into space.
Built by Planet Labs, the satellite will prove whether a Google Tensor Processing Unit, the company’s direct competitor to Nvidia’s GPUs, can actually function in orbit. That means supplying a continuous kilowatt of power, properly cooling the chip, and running a series of models through their paces to catch anything that goes wrong.
A Long-Term Vision for Google Space Data Centers
“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 broader plan to eventually build large-scale compute clusters in orbit around Earth.
Once commissioned, the satellite will fire up its TPU in 15-minute bursts specifically to avoid straining its power and thermal management systems. While this particular satellite uses a standard Planet Labs platform, the two companies are already working on a follow-up demo expected to launch next year, featuring two satellites built more specifically for advanced compute workloads, designed to collaborate directly via a laser communications link.
A Long-Term Moonshot, Not Just a Test Flight
Suncatcher wasn’t the only space AI payload riding aboard this particular SpaceX rocket, which carried more than 100 different payloads total, including missions from Satlyt and Cowboy Space Company. What sets Google’s initiative apart from those other projects, and from SpaceX itself, is its genuinely long-term scope.
Beals described the project’s focus as building specifically for the space infrastructure and AI workloads that will exist years from now, envisioning eventual Google space data centers made up of 81 satellites flying in tight formation, processing data in parallel. “The bandwidth and the latency between TPUs really, really matters when you’re trying to run a multi-rack workload,” Beals said. “We’re trying to look ahead to not just what workloads exist today, but where they will be in five years.”
Why Google Thinks Rockets Will Get Much Cheaper
On Thursday, Google also released a peer-reviewed white paper on orbital data centers, offering one of the most rigorous publicly available analyses of exactly how compute gets delivered to space. The paper, set to publish in the journal Joule, includes a particularly notable section on how Google envisions access to space evolving, though researchers stress it isn’t intended as a formal economic feasibility study.
Like every major data center company, Google is relying heavily on SpaceX to get its spacecraft off the ground, and Google happens to also be a major investor in SpaceX itself. Pointing to what they describe as roughly a 20% annual price-reducing “learning curve” SpaceX has achieved since launching its first Falcon 1 rocket, the paper’s authors believe it’s reasonable to expect launch prices to fall to around $200 per kilogram by 2035.
The Massive Scale Starship Would Need to Reach
Achieving that price point would require Starship to fly a staggering 370,000 tons of payload into orbit, based on comparable cost trajectories seen with the Falcon 9. That translates to roughly 1,800 total launches over the next decade, or about 180 per year, assuming each mission can carry 200 metric tons.
That’s an enormous ask for a vehicle that has never flown more than five times in a single year to date. SpaceX, for its part, predicts it will eventually fly far more frequently than that, with Elon Musk suggesting Starship could reach an hourly flight rate as soon as 2029, though Musk has made similarly ambitious predictions before.
Will Google’s Chips Actually Survive Space Radiation?
One genuinely encouraging finding from Google’s updated research involves chip durability. The company had to redo radiation testing in a particle accelerator after realizing its original chip configuration provided more shielding than it would actually experience in real orbital conditions. That retesting produced slightly more errors in the chip’s logic circuitry, though Google remains confident its chips can still handle large inference workloads in orbit across a satellite’s full five-year expected 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,” highlighting a key distinction between everyday AI tasks and the far more demanding workloads Google space data centers will eventually need to support.