Google's Data Center Satellite Phones Home, and Its First Words Are a Haiku
Google's first prototype satellite for Project Suncatcher is online and talking to the ground. The refrigerator-sized spacecraft launched on October 1 aboard a SpaceX Falcon 9 rocket, sharing the ride with more than 100 other spacecraft. Its cargo is small but unusual: four Google Tensor Processing Units, meant to test whether data-center-class AI chips can actually compute in space.
After launch, engineers spent about three weeks working through a commissioning checklist: powering up the host and TPU trays and running early machine learning tests. This week they sent the first prompt to orbit, asking the onboard Google DeepMind Gemma 4 model to write a haiku about being in space. The satellite generated the answer overhead and beamed it back: "Blue marble below / Data streams in silent grace / Starlight feeds my core."
In its official blog post, Google called the reply its "Hello World" milestone: the first AI response ever produced by a data-center-class TPU in orbit. Engineers also tested space-themed limericks and telemetry-style prompts, checking that the chip could handle both creative text and spacecraft housekeeping queries.
Project Suncatcher lead Travis Beals put the logic simply: the sun produces nearly all the usable power in the solar system, so running AI on sunlight is the most direct source available. Google's accompanying research estimates that solar panels in a dawn-dusk orbit, in near-constant sunlight, can generate up to eight times more power than the same panels on Earth, with no land purchases, no cooling water, and no grid connection queue.
The prototype carries four TPUs and generates up to about 1 kilowatt of power. Because the chips draw heavily and run hot, only one TPU runs at a time, in bursts of roughly 15 to 20 minutes, after which computation must stop so radiators can shed the accumulated heat. Cooling, not radiation, is now the mission's defining constraint.
The near-term winners are satellite builders and launch providers; the prototype was built with Planet. The long-term stakes fall on ground-based data centers, power grids, and cooling supply chains if orbital inference ever scales. Google itself cautions that the economics are nowhere near closed: this prototype flies in a standard orbit and still relies on batteries, while the always-sunlit orbit and close-flying constellation remain future plans.
Google plans to launch two more satellites next year to test links between spacecraft. The long-range vision is clusters of satellites, each carrying dozens of TPUs, exchanging data by laser and working together as an orbital data center for large inference workloads. A three-line haiku is a modest start; the harder tests are heat rejection and networking.
The satellite proves chips can think in orbit, but only for minutes at a time. If your town faced power shortages, would you rather send its data centers to space, or fix the grid and cooling on the ground first? Have you ever lost work to a blackout or a soaring power bill?