On October 1, Google launched a satellite carrying one of its Tensor Processing Units (TPUs) to space. On October 30, the company announced that it had received the first AI response from the onboard chip — a haiku about space. The achievement marks the "Hello World" milestone for Project Suncatcher, a research initiative exploring whether data centers could ever be viable in orbit by the 2030s.
The satellite was built in partnership with Planet Labs PBC and launched aboard a SpaceX Falcon 9 rocket on the Transporter-18 rideshare mission, which carried more than a hundred other spacecraft. Aboard are four Trillium TPUs, Google's own custom chips designed for AI training and inference, drawing about 1 kW of power from solar panels.
As the satellite passed over Svalbard, Norway, a ground station sent a prompt to the open-source Gemma 4 model running on the TPU, asking it to compose a poem about being in space. The response came back to Earth, confirmed by Google in an official blog post: "Blue marble below / Data streams in silent grace / Starlight feeds my core."
Beyond the haiku, the team tested limericks and telemetry prompts asking the satellite to report system health. The goal is to verify whether TPUs can reliably execute workloads under extreme space conditions — radiation, thermal stress, and vacuum, which complicates cooling since there's no air for convection.
The Project Suncatcher team had already tested the chips in a thermal vacuum chamber simulating the space environment. Early results showed that the Trillium TPUs tolerate a total ionizing radiation dose higher than what they'd receive during a five-year mission. But the real validation requires being in orbit — that's where the gap between simulation and reality shows.
Google published a peer-reviewed paper in the journal Joule detailing the project. The paper sketches a long-term vision where dozens of satellites orbit in swarms, communicating via laser links to maintain the bandwidth needed for AI processing. Each future satellite would carry dozens of TPU chips, operating continuously and forming actual orbital data centers.
Google estimates that if launch costs fall to $200/kg, the cost of launching a kilowatt of solar power to orbit would be about $810 per year — comparable to what US data centers spend on electricity. The learning-curve analysis suggests this price becomes viable by the mid-2030s, assuming Starship flies about 180 times per year.
The timing of the announcement is curious. Google itself is grappling with AI-related internal challenges. Alphabet reported this week a stock price decline after acknowledging that efforts to insert AI into search products are hurting advertising revenue growth. The company faces a split between maintaining existing infrastructure investments and funding long-term research projects like Suncatcher.
The same launch that carried Suncatcher also transported other orbital data center experiments. Cowboy Space tested the Reason-1 satellite, which attempts to beam power via laser, and Star Catcher tested Protostar, beaming power between two spacecraft. All these experiments happened on the same rideshare — confirming that the orbital data center industry is moving beyond theory.
If the first objective is simply to prove chips can survive in space, how long can the company sustain this kind of research before market pressure demands tangible near-term results?
Sources: Google Blog, The Verge, TechCrunch
✓ Independent sources cross-checked and verified before publishing
