Google boots up first space AI satellite under Project Suncatcher
Google successfully tested its Project Suncatcher satellite in orbit, using an onboard TPU chip to run the Gemma 4 AI model from space.
Google announced this week that it has successfully booted up its first AI-equipped data center satellite in orbit as part of its experimental Project Suncatcher initiative. After launching earlier this month aboard a SpaceX Falcon 9 rocket, the prototype satellite powered on its custom AI chips and successfully processed prompts sent from engineers on the ground.
How does Google's Project Suncatcher work in space?
The prototype spacecraft carries four Google Tensor Processing Units (TPUs)—specialized hardware designed for machine learning tasks—and generates up to 1 kilowatt of electricity. Due to the high power requirements of the processors, the satellite can only operate one TPU at a time for about 20 minutes.
Engineers pre-loaded the satellite with Google DeepMind's Gemma 4 model. During initial tests following a three-week system check, the satellite responded to prompts by generating space-themed poetry and sending back telemetry reports about the orbital environment surrounding the spacecraft.
Why is testing AI hardware in orbit important?
Operating AI hardware in space introduces unique physical challenges, particularly from extreme temperature changes and space radiation. Radiation can trigger "bit flips"—an error where a computer memory value unexpectedly flips between zero and one, potentially corrupting data without triggering standard software monitoring alerts.
Google previously evaluated its TPUs using ground-based particle accelerators, which indicated that radiation risks would be minimal for running existing models (known as inference) compared to training new ones. The ongoing orbital tests will measure how the physical processors withstand real radiation exposure and thermal stress over time.
What is the long-term vision for space data centers?
Project Suncatcher is a long-term initiative focused on creating infrastructure for the 2030s. Unlike competing concepts that rely on standalone satellites, Google plans to connect multiple orbital data centers using high-speed optical laser links to share processing workloads across space.
The company plans to launch two interconnected satellites in 2027 to test advanced power systems and optical communication. Future expansion will depend on heavy-lift launch capabilities like SpaceX's Starship. Suncatcher researchers estimate that Starship will need to complete 1,800 flights to achieve the cost reductions necessary to make large-scale orbital data centers economically viable.
Comments 0
No comments yet. Start the conversation.