Google Tests Project Suncatcher Orbital AI Data Center Satellite in Low Earth Orbit

Orbital AI Computing Experiment

Google confirmed operational testing for its Project Suncatcher prototype satellite in early October 2026. The refrigerator-sized spacecraft entered low Earth orbit following its launch aboard a SpaceX Falcon 9 rocket during the Transporter-18 rideshare mission.

Developed in collaboration with Planet Labs, the project evaluates the feasibility of deploying machine learning infrastructure directly into space. The primary payload consists of four custom Google Tensor Processing Units (TPUs) based on the Trillium architecture.

Mission objectives focus on measuring how commercial accelerator silicon withstands launch vibration, cosmic radiation, and thermal swings between direct sunlight and planetary eclipse.

Gemma Model Workloads and Hardware Telemetry

While in sun-synchronous orbit, the onboard TPU cluster executes inference tasks using Google's open-weight Gemma model:

• Inference cycles: Workloads run in 15-minute bursts to evaluate power distribution and heat dissipation without conventional water cooling.

• Solar power utilization: The spacecraft relies on near-continuous solar radiation to power compute silicon without drawing on terrestrial electrical grids.

• Hardware telemetry: Engineers are monitoring memory bit-flip rates caused by high-energy galactic cosmic rays outside the thick protection of Earth's atmosphere.

Initial telemetry indicates the Trillium silicon maintained stable execution parameters without unexpected thermal throttling.

Inter-Satellite Laser Roadmap for 2027

The current mission represents the opening phase of a multi-year technical roadmap:

• Follow-up deployment: Google plans to launch two additional prototype satellites in 2027.

• Optical transceivers: The second phase will test high-bandwidth laser communication links connecting spacecraft into a distributed orbital mesh.

• Distributed processing: This architecture aims to validate parameter synchronization and parallel processing directly in orbit.

Addressing Terrestrial Power Constraints

Surging electrical demand from terrestrial data centers has created major supply bottlenecks for frontier AI training. Orbital facilities offer an alternative path to scale compute capacity without consuming municipal power, land, or water reserves.

While launch costs and radiation shielding remain significant hurdles, the initial data from Project Suncatcher demonstrates that standard accelerator silicon can function reliably outside Earth's atmosphere.