Google Tests Generative AI Inference on Orbiting Suncatcher Satellite
The Suncatcher satellite successfully executed Gemma 4 model prompts from low Earth orbit.
1Orbital AI Milestone Achieved
Google successfully received a haiku from its experimental Suncatcher satellite, marking a significant milestone for orbital machine learning. The onboard Gemma 4 model generated the poem while orbiting Earth, responding to prompts sent from a ground station in Svalbard, Norway. This test confirms the viability of running inference tasks in space environments.
2High Cost of Space Inference
Generating the twenty-token haiku was exceptionally expensive compared to terrestrial cloud costs. While standard Google Cloud API pricing for Gemma 4 is negligible, the satellite launch costs, estimated at seven thousand dollars per kilogram, inflate the price significantly. Google is currently analyzing these economics to justify future deployments of orbital computing infrastructure.
3Future Cost Parity Projections
Google anticipates that orbital datacenter costs could eventually rival terrestrial energy spending if launch prices drop to two hundred dollars per kilogram. By 2035, SpaceX launch capabilities might enable this economic shift. Prototyping now allows Google to prepare for a future where space-based computing becomes a strategic component of their global infrastructure.
4Thermal and Radiation Testing
Beyond creative tasks, Google is evaluating how TPUs perform under extreme thermal stress and radiation. Laboratory tests using a 67 MeV proton beam assessed the hardware's durability. The system's high-bandwidth memory showed sensitivity after a cumulative dose of 2 krad, exceeding the minimum survival requirement of 750 rad by three times.
5Silent Data Corruption Analysis
Researchers analyzed silent data corruption rates to ensure inference reliability in space. At an estimated orbital dose of 150 rad per year, the failure probability is approximately one per three million inferences. While this error rate is acceptable for inference, training jobs remain vulnerable, necessitating further study into system-level mitigation strategies.
6Expanding Orbital Testing Scope
The Suncatcher mission has moved beyond simple haikus to testing space-themed limericks and complex telemetry prompts. These technical reports provide real-time data on payload system status. This iterative testing approach ensures that the hardware remains functional while exposed to the harsh conditions of low Earth orbit throughout the mission's operational lifespan.
7Scaling Through Satellite Constellations
Google is shifting its strategy from deploying large, singular satellites to utilizing constellations of smaller units. The plan for next year involves testing a pair of interconnected satellites to evaluate communication protocols. This distributed architecture aims to improve resilience and scalability, moving away from the limitations of single, massive orbital hardware platforms.
8Strategic Implications for Infrastructure
This experiment represents a critical step toward decentralized, space-based computing architectures. By validating the Gemma 4 model's performance in orbit, Google is establishing the technical foundation for future orbital datacenters. Continued research into radiation hardening and system-level mitigations will determine the long-term feasibility of running complex transformer workloads in space.