SVALBARD, Norway — Google’s experimental attempt to operate artificial intelligence hardware in outer space achieved a major milestone this week.
On Friday, October 30, 2026, engineers successfully executed Google’s open-weights Gemma 4 AI model aboard a prototype satellite in low Earth orbit, receiving an orbit-generated haiku as the system passed over a ground station near Svalbard, Norway.
The short poem transmitted back to Earth began with the line “Blue marble below” and concluded with “Starlight feeds my core.”
Project Suncatcher and the Prototype Mission
The orbital test forms part of Google’s Project Suncatcher, a long-term initiative launched by Google Research to evaluate whether specialized machine learning hardware can run reliably in space using continuous solar power.
Project Suncatcher Mission Summary
+-------------------------------------------------------------------------+
| LAUNCH DATE: October 1, 2026 |
| LAUNCH VEHICLE: SpaceX Falcon 9 (Rideshare Mission) |
| BUS DEVELOPMENT PARTNER: Planet Labs |
| HARDWARE PAYLOAD: 4 Google Tensor Processing Units (TPUs) |
| ACTIVE HARDWARE CAPACITY: 1 TPU operating at a time (15–20 min windows) |
| OPERATIONAL AI MODEL: Gemma 4 |
| TRANSMISSION LOCATION: Svalbard Ground Station, Norway |
+-------------------------------------------------------------------------+
The prototype satellite was launched into orbit on October 1, 2026, aboard a SpaceX Falcon 9 rocket. Built in collaboration with satellite imaging firm Planet Labs, the bus carries four of Google’s high-performance Tensor Processing Units (TPUs).
Due to orbital power limits and thermal management constraints in a vacuum, power is restricted so that only a single TPU operates at a time during short active compute windows.
Technical Telemetry and Initial Prompt Testing
During the telemetry pass over Svalbard, engineers issued several test prompts to evaluate the processing chip. In addition to prompting the Gemma 4 model for creative space poetry, mission controllers tasked the onboard processor with processing diagnostic metrics and telemetry data to verify chip status and memory integrity.
Core Engineering Considerations in Space Compute
+-------------------------------------------------------------------------+
| THERMAL DISSIPATION: Absence of air requires specialized radiators and |
| heat pipes instead of traditional cooling fans. |
| RADIATION FLUX: Solar flares and cosmic rays risk bitflips or silicon |
| degradation over extended periods. |
| SOLAR HARVESTING: Continuous sun-synchronous orbits offer high power |
| generation potential for satellite compute clusters. |
+-------------------------------------------------------------------------+
Next Operational Phases and 2027 Roadmap
The success of the Gemma 4 execution kicks off a planned one-year testing phase for the prototype satellite.
Project Suncatcher Development Timeline
+-------------------------------------------------------------------------+
| OCT 1, 2026: Falcon 9 launch carries prototype satellite into LEO. |
| OCT 30, 2026: First successful Gemma 4 LLM inference & haiku returned. |
| LATE 2026 – 2027: Ongoing thermal vacuum and radiation endurance testing.|
| 2027 TARGET: Launch of twin interconnected satellites for multi-node |
| orbital compute tests. |
+-------------------------------------------------------------------------+
Over the coming months, Google engineers will observe how the hardware copes with high-energy cosmic radiation, solar particle events, and extreme orbital temperature shifts between direct sunlight and darkness.
Following the single-satellite prototype phase, Google plans to launch two interconnected satellites in 2027 to test high-speed optical links and evaluate whether multiple space-based nodes can process coordinated AI workloads.
Google-Headquarters-Picture-on-Flickr-by-Anthony-Quintano.
