Google’s Suncatcher prototype space data center is operational and communicating with ground controllers after launching earlier this month on a SpaceX Falcon 9 alongside more than 100 other spacecraft. The satellite carries four Google Tensor Processing Units (TPUs) and produces up to 1 kW of electrical power.
Purpose of the prototype
The craft is intended to help prepare for future constellations of closely flying satellites that perform substantial inference workloads in orbit. Because of the satellite’s power budget, however, only one TPU can be powered at a time for roughly twenty minutes, which complicates already challenging space-to-ground communications.
First tests and the onboard model
Engineers spent about three weeks commissioning and testing the satellite’s subsystems. During initial operations this week, the team sent a prompt to the language model preloaded on the satellite — Google DeepMind’s Gemma 4 — and asked it to write a haiku about being in space. The model responded:
Blue marble below Data streams in silent grace Starlight feeds my core. ✨
Google said the team also tested other space-themed short poems and requested the spacecraft report on local orbital conditions.
Radiation, thermal stress and hardware reliability
The ongoing test campaign will evaluate how the TPUs perform under thermal stress and in the space radiation environment. Although Google has tested its TPUs in particle accelerators on Earth, the company now needs on-orbit data to see real radiation effects on the processors. Radiation can cause “bit flips” — changes in stored memory values — which may corrupt data without triggering standard monitoring tools.
Ground testing suggested radiation would have minimal impact on inference workloads but could present greater problems for training models. The on-orbit tests are designed to measure these dynamics in real operational conditions rather than simulations.
Next steps and strategic context
If the onboard TPUs prove capable, Google plans to launch two additional, more advanced-chip-equipped satellites in 2027. Those follow-on spacecraft are intended to support higher power generation and demonstrate inter-satellite optical (laser) links.
Google’s Starcloud project is among the few space-based data center visions that plan high-speed links between satellites to support coherent distributed workloads. SpaceX’s proposals are also expected to use laser links, but their planned orbital spacing would likely be too large for efficient shared processing. Companies such as Starcloud and Cowboy Space anticipate performing inference loads on individual satellites rather than tightly coupled distributed processing.
Google and SpaceX’s plans are interrelated: Alphabet, Google’s parent company, holds a significant investment stake in SpaceX. Both companies’ plans for orbital data centers assume availability of Starship, the heavy-lift rocket under development that needs to be fully and rapidly reusable to achieve the projected economics.
A peer-reviewed analysis published earlier this month by Suncatcher researchers estimated that SpaceX would need to fly Starship about 1,800 times to realize the same learning-driven cost reductions it achieved with earlier vehicles.
Business implications
Suncatcher is a typical Google moonshot: the effort focuses on designing infrastructure intended to be economical in the 2030s. Reaching that point depends in part on how the company handles near-term AI opportunities and pressures. Alphabet’s stock fell this week after the company reported that efforts to integrate AI into its search products are suppressing ad revenue growth.
Correction
Correction: the Suncatcher prototype satellite carries four TPUs but can operate only one at a time due to power constraints.
A Google spokesperson told TechCrunch the satellite is operating nominally and that ground teams will continue detailed on-orbit testing.
