Google's prototype space data center satellite Suncatcher has been launched and began communicating with ground control earlier this month after being deployed together with over 100 other spacecraft on a Falcon 9 rocket SpaceX.
This satellite is equipped with a Google Tensor Processing Unit (TPU), which can generate up to 1 kilowatt of power. Its goal is to pave the way for future networks composed of satellites flying in close formation, allowing these satellites to perform large-scale computational tasks in orbit. However, the high power requirements of TPU mean that it can only be activated for about 20 minutes at a time, which makes the already complex space-ground communication even more challenging.
During a debugging process that lasted about three weeks, the team tested each subsystem of the satellite, after which the spacecraft entered a working state. This week, when ground engineers issued their first command, they asked the model pre-installed on the satellite—Google’s DeepMind’s Gemma 4—to write a haiku about being in space. And so it wrote:
Blue planet below
The data flow is silent and elegant.Starlight nourishes my core ✨Google stated that the company tested the model with such space-themed puns (should they have sent a poet instead?), and more importantly, it required this spacecraft to report on the orbital environment around the satellites.
A spokesperson told TechCrunch that the satellite is "operating normally."
The ongoing tests will assess the performance of these chips under conditions of thermal stress and space radiation. Google has previously conducted a large number of tests on them in particle accelerators on Earth, but now it is necessary to observe the actual effects of radiation in space on the hardware, as well as how these effects will affect the processors. Radiation can cause "bit flips," which means changing the stored values in computer memory, thereby damaging data without triggering conventional monitoring tools.
Ground tests indicate that the impact of radiation on reasoning workloads should be minimal, while its effect on model training is likely to be more significant; however, now the Google team will be able to assess these dynamics in a real environment rather than in simulation.
If the TPU on this spacecraft is proven to be capable, the next step will be to launch two additional satellites equipped with advanced chips in 2027. These spacecraft will support higher power generation capabilities and demonstrate inter-satellite optical links.
Among the various concepts for space data centers, Google's Starcloud project is one of the few plans that rely on high-speed links between satellites to support coordinated distributed workloads; the data center of SpaceX is also expected to use laser links, but the orbital spacing between its satellites is anticipated to be too large to effectively share processing capabilities. Companies like Starcloud and Cowboy Space are expected to perform computational tasks on a single satellite.
Google is closely aligned with the vision of SpaceX. Google's parent company, Alphabet, holds a significant equity stake in SpaceX. To launch their orbital data centers, both companies need Starship – the heavy-lift rocket at the core of Elon Musk's space vision, which is still under development – to achieve complete and rapidly reusable capabilities.
In a peer-reviewed analysis published earlier this month, researchers from Suncatcher estimated that SpaceX would need to conduct 1,800 Starship flights in order to achieve the same cost reduction driven by the learning curve as its earlier launch vehicles.
Suncatcher is a typical example of Google's "moonshot" project – the company's focus is on designing infrastructure that will be economically viable by the 2030s. However, whether this goal can be achieved may depend on how the rest of the company's business responds to the pressures and opportunities brought about in the short term by AI. Google's parent company, Alphabet, saw its stock price plummet this week following the release of financial reports, as the company stated that efforts to AI into its search business are dragging down ad revenue growth.
