Google has successfully brought its first prototype space-based data center online. Launched earlier this month aboard a SpaceX Falcon 9 rocket alongside more than 100 other spacecraft, the experimental satellite has completed its three-week commissioning phase and begun executing AI inference tasks in low Earth orbit.
The milestone represents the first live operational test of Google custom hardware and AI software in space. To demonstrate that the platform was functioning, ground engineers sent a prompt to the spacecraft, which executed Google DeepMind's pre-loaded Gemma 4 model. In response, the satellite generated a short, space-themed haiku:
Blue marble below
Data streams in silent grace
Starlight feeds my core. ✨
While the demonstration highlights basic operational status, the underlying mission is focused on evaluating whether high-performance AI chips can withstand thermal stress and radiation outside Earth's atmosphere.
How the orbital hardware works
The prototype satellite carries four Google Tensor Processing Units (TPUs), the specialized application-specific integrated circuits that power AI computations in Google's ground data centers. Power generation on the satellite is modest, capped at approximately 1 kilowatt.
Due to the heavy electrical demands of the TPU silicon, the spacecraft cannot run all four processors simultaneously. Instead, flight operations can run only one TPU at a time for about 20-minute processing windows. This restriction creates unique operational challenges, requiring engineers to coordinate power management, processing runs, and space-to-ground telemetry within narrow time frames.
Google previously evaluated its TPUs in terrestrial particle accelerators to simulate orbital conditions. However, testing the physical chips on an active spacecraft provides real-world data that simulations cannot replicate.
Testing AI resilience against space radiation
A major goal of the test campaign is monitoring how cosmic radiation affects chip behavior, specifically regarding soft memory errors known as bit flips. These occur when high-energy radiation particles alter a stored binary value in computer memory from a zero to a one, or vice versa.
Bit flips can silently corrupt analytical data or model calculations without triggering standard system monitoring tools. Prior ground testing suggested that radiation-induced errors would have a minimal impact on inference tasks while posing a far greater risk to model training. The in-orbit tests will allow engineers to measure these effects under actual flight conditions.
Inter-satellite networks and future plans
Google's orbital initiative, known as Project Suncatcher, takes a distinct architectural approach compared to competing orbital computing concepts:
- Coherent distributed processing: Google intends to link multiple satellites using high-speed optical connections, enabling spacecraft flying in close formation to share compute workloads across a distributed cluster.
- Alternative approaches: Competitors such as Starcloud and Cowboy Space focus on running inference tasks on isolated individual satellites. Meanwhile, SpaceX's planned data center concepts utilize laser links but place satellites too far apart to support real-time shared computing.
If the prototype succeeds throughout its testing phase, Google plans to launch two additional satellites in 2027. Those second-generation spacecraft will feature expanded power generation capabilities and demonstrate inter-satellite optical communication links in space.
Long-term vision and economic hurdles
Project Suncatcher is designed as a long-term research effort aimed at establishing viable infrastructure for the 2030s. Parent company Alphabet maintains a significant financial stake in SpaceX, aligning the technical roadmaps of both organizations.
Scaling orbital data centers into an economically viable network will depend heavily on low-cost space transport. In a peer-reviewed paper published earlier this month, Suncatcher researchers estimated that SpaceX would need to launch its Starship vehicle roughly 1,800 times to achieve the operational cost efficiencies necessary for space-based data centers to become cost-effective.
While Google explores long-term orbital infrastructure, the company faces immediate commercial shifts on Earth. Alphabet stock experienced a downturn following recent financial reports indicating that integrating AI features into Search has slowed advertising revenue growth. Project Suncatcher highlights Google's commitment to pursuing high-risk, long-term technical concepts even as core commercial markets adjust to new AI integration demands.
Filed under: AI, TechNews, ProductLaunches, Hardware, Google, SpaceTech