Google’s first Suncatcher prototype satellite has begun communicating with the ground and running an AI model in orbit. The spacecraft is an early test of whether satellites could eventually work together as data centers for large inference workloads.
A first prompt from orbit
The satellite launched earlier this month with more than 100 other spacecraft on a SpaceX Falcon 9 rocket. After about three weeks of commissioning to check its subsystems, it was ready for its first task from engineers on the ground.
They asked Google DeepMind’s Gemma 4, pre-loaded on the spacecraft, to write a haiku about being in space:
Blue marble below
Data streams in silent grace
Starlight feeds my core. ✨
Google also tested the model with space-themed limericks and asked the satellite to report on orbital conditions around it. A spokesperson told TechCrunch the satellite was operating nominally.
Power and radiation shape the test
The prototype carries four Google Tensor Processing Units (TPUs) and can generate up to 1 kW of electricity. Its power supply is limited, however: it can operate only one TPU at a time, and that chip can run for about twenty minutes before power demands become a constraint. This also makes communication between the spacecraft and the ground more difficult.
The test campaign will examine how the chips perform under thermal stress and in the space radiation environment. Google has tested its TPUs in particle accelerators on Earth, but the satellite gives the team a chance to observe radiation’s effects on hardware in space.
Radiation can cause “bit flips,” changing stored values in computer memory. That can corrupt data without setting off typical monitoring tools. Ground tests suggested radiation might have a minimal impact on inference workloads and pose more of a problem for training a model. The orbital tests will show how those expectations hold up in reality.
Google’s next step depends on the results
If the onboard TPUs prove capable, Google’s next step is to launch two more satellites equipped with the advanced chips in 2027. Those spacecraft would support higher power generation and demonstrate inter-satellite optical links.
Google’s plan centers on high-speed links between satellites, which could let them support coherent distributed workloads. Other companies, including Starcloud and Cowboy Space, expect to run inference loads on individual satellites. SpaceX is also expected to use laser links, though its data centers may orbit too far apart to share processing effectively.
A long-term project with big dependencies
Suncatcher is a Google moonshot focused on infrastructure that could be economical in the 2030s. Reaching that point depends on technical progress, including whether the chips can operate reliably in orbit and whether satellites can generate and share enough power to do useful work together.
Both Google and SpaceX would also need Starship, the still-in-development heavy-lift rocket at the center of Elon Musk’s space vision, to become fully and rapidly reusable to launch their orbital data centers. In a peer-reviewed analysis published earlier this month, Suncatcher researchers estimated that SpaceX would need to fly Starship 1,800 times to achieve the same learning-driven cost reductions it saw with its earlier vehicles.
The project’s long horizon sits alongside immediate pressures around AI at Google. Alphabet, Google’s parent company, saw its stock plunge this week after reporting that efforts to insert AI into its search businesses are hurting ad revenue growth. Suncatcher is aimed at infrastructure for the 2030s; its future will unfold while the company faces near-term questions about AI’s effects on its existing business.