Google Tests Project Suncatcher with Orbital Satellite

Built with Planet Labs, the refrigerator-sized spacecraft carries four TPU accelerators to test cosmic radiation resilience and thermal limits.

Google launched its first experimental Project Suncatcher satellite, kicking off a research effort to evaluate whether running artificial intelligence processing in space is viable. The spacecraft lifted off on October 1 aboard a SpaceX Falcon 9 rocket during the Transporter-18 rideshare mission from Vandenberg Space Force Base in California. Within hours of deployment, Google Research engineers established two-way communications and confirmed that onboard systems were functioning within normal parameters.

The prospect of moving compute off Earth has drawn interest amid soaring power demands from terrestrial data centers. However, operating advanced silicon in space introduces immediate physical constraints that ground-based facilities never face.

What Project Suncatcher Is Testing

Aerospace imaging specialist Planet Labs designed, built, and operates the satellite platform. Roughly the size of a standard kitchen refrigerator, the payload bay houses four Google Tensor Processing Units (TPUs). These are Trillium-class accelerators, identical to the hardware Google deploys across its server racks on Earth.

This early flight is not an active commercial data center, nor is it designed to train foundation models from orbit. Instead, engineers want to observe how off-the-shelf silicon handles launch vibrations, cosmic radiation bombardment in low Earth orbit, and thermal cycling across repeated orbits.

The Physics of Cooling in a Vacuum

Ground facilities draw power from local grids and cool racks using forced-air circulation or liquid cooling loops, dissipating heat primarily through convection. In orbit, solar panels provide abundant energy along the flight path, but removing excess heat runs straight into the vacuum of space.

Without air molecules to carry thermal energy away, the satellite must rely entirely on radiative heat transfer. Every watt of heat generated by active chips must conduct to exterior radiator panels and disperse as infrared radiation. Radiative cooling is inherently slower than terrestrial convective cooling.

Because of this thermal bottleneck, the satellite cannot run continuous compute cycles. Test protocols limit runs of Google’s open Gemma AI model to bursts lasting no more than 15 minutes. Once a compute run finishes, the platform pauses processing to allow the radiator panels sufficient time to shed the heat before initiating another task.

Economic Viability and Next Steps

Alongside the orbital test, Google researchers published a peer-reviewed paper in the academic journal Joule examining the structural physics and financial models of space compute. The findings remain grounded: Project Suncatcher represents basic exploratory research with a five to ten year horizon.

The underlying math works only if orbital transport costs decline sharply. The team notes that making orbital compute economically sensible requires high-capacity, fully reusable launch systems, such as SpaceX’s Starship, operating at routine scale.

Google plans to run the current satellite for up to twelve months to log component wear and radiation degradation. A follow-up mission is already scheduled for 2027, when Google and Planet Labs aim to launch two synchronized satellites to test formation flying and inter-satellite optical laser links.

Sources

• Google Blog — Official post by Travis Beals, Google Research
• Planet Labs — Orbital platform specifications and announcement
• NPR — Coverage of orbital computing and hardware testing
• TechCrunch — Analysis of launch economics and Starship cadence requirements
• Fortune — Perspective on data center power limits and space research