Google’s first data center satellite is online and writing poetry

Google's Suncatcher prototype space data center is now online and communicating with the ground after launching earlier this month on a SpaceX Falcon 9 rocket alongside more than 100 other spacecraft. The satellite carries four Google Tensor Processing Units (TPUs) and generates up to 1 kW of electricity, though its power demands mean only one TPU can operate for about twenty minutes at a time. After a roughly three-week commissioning process, engineers sent their first prompt this week, asking the pre-loaded Google DeepMind Gemma 4 model to write a haiku about space; the model responded with a poem about the "blue marble below" and "starlight" feeding its core. Google also had the spacecraft report on orbital conditions and said it is operating nominally. The ongoing test campaign will examine how the chips perform under thermal stress and space radiation, which can cause bit flips that corrupt data without triggering typical monitoring tools. Ground testing suggested radiation's impact would be minimal for inference workloads but more problematic for training, and the team will now assess these dynamics in reality. If the onboard TPU proves capable, Google plans to launch two more satellites with the advanced chips in 2027, supporting higher power generation and demonstrating inter-satellite optical links. Google's project is largely alone in planning high-speed links between satellites to support coherent distributed workloads; SpaceX's data centers are expected to use laser links but orbit too far apart to share processing effectively. Both companies' visions are closely linked, as Alphabet holds a significant investment stake in SpaceX, and both will require Starship to be fully and rapidly reusable. A peer-reviewed analysis estimated SpaceX would need to fly Starship 1,800 times to achieve the same learning-driven cost reductions seen with earlier vehicles. Suncatcher is a classic Google moonshot focused on infrastructure that will be economical in the 2030s, though getting there may depend on how the company responds to near-term AI pressures and opportunities.

(Source：TechCrunch)