Google’s Project Suncatcher Begins Testing AI Computing in Orbit

Google’s Project Suncatcher is a long-term research “moonshot” exploring whether low Earth orbit can eventually host scalable, solar-powered machine learning data centers. As AI models demand increasingly massive amounts of power, Earth-bound data centers are struggling with grid connection delays and community pushback. Google’s solution is to look to space.

The Inaugural Test Mission

On October 1, 2026, Google launched its first Suncatcher prototype—a refrigerator-sized satellite named MVP, developed in partnership with Planet. It reached orbit aboard SpaceX’s Transporter-18 rideshare mission from Vandenberg Space Force Base.

This mission isn’t attempting to run a commercial AI service; it is an extreme hardware stress test for Google’s Trillium Tensor Processing Units (TPUs). The team is monitoring how the chips handle three main hazards:

• Launch Forces: A rocket launch subjects the payload to intense vibrations and up to 10g of sustained acceleration, with individual chips experiencing up to 50-100g.
• Space Radiation: Cosmic rays and solar flares can cause bit-flips and corrupt data. Before launch, Google blasted the TPUs with a 67 MeV proton beam at UC Davis to simulate orbital radiation, where they survived well beyond a typical five-year mission’s exposure.
• Vacuum Cooling:Earth-based data centers use fans and air conditioning, but space is a vacuum with no airflow.The TPUs are testing a combination of specialized heat pipes and radiators to diffuse their massive thermal output safely.

Why Put AI in Space?

The primary driver is clean, uninterrupted energy. Satellites positioned in a dawn-dusk sun-synchronous low Earth orbit experience near-constant sunlight.This allows them to harvest up to eight times more solar power than equivalent panels on Earth, without having to wait for daylight or deal with cloud cover.

The 2027 Vision and Beyond

If the current hardware survives, the next major hurdle is interconnectivity. AI workloads require massive data bandwidth to function. Google’s plan for 2027 and beyond is to fly tight clusters of satellites—sometimes just hundreds of meters apart—acting as a floating neural network. They aim to use Dense Wavelength Division Multiplexing (DWDM) lasers to beam data between satellites at 10 terabits per second, effectively matching the speed of terrestrial fiber-optic cables without any physical wires.