Google Suncatcher’s prototype space data center is online and communicating with the ground after launching along with more than 100 other spacecraft on a SpaceX Falcon 9 rocket earlier this month.
The satellite is equipped with four Google Tensor Processing Units (TPUs), generates up to 1 kW of power, and is intended to pave the way for a future network of close-flying satellites that will run large-scale inference workloads in orbit. However, the power demands of the TPUs mean that they can only be powered on one for about 20 minutes at a time, complicating the already difficult business of space-to-ground communications.
After a commissioning process that tested the satellite’s subsystems over about three weeks, the spacecraft was ready for operation. When engineers on the ground sent out their first prompt this week, they asked Google DeepMind’s Gemma 4, a model preloaded on the satellite, to write a haiku about being in space. So it would look like this:
blue marble below
Data stream within silent grace period
Starlight nourishes my core. ✨
Google said it used these space-themed limericks to test the model (should it have sent a poet?), and perhaps more importantly, the company asked the spacecraft to report orbital conditions around the satellite.
A spokesperson told TechCrunch that the satellite is nominally operating normally.
An ongoing test campaign will investigate how the chip performs under thermal stress and space radiation environments. Google has extensively tested its TPUs in particle accelerators on Earth, but now it needs to see how the actual effects of radiation on hardware in space affect its processors. Radiation can cause “bit flips,” which cause values to be stored in computer memory, potentially corrupting data without triggering common monitoring tools.
Ground-based testing suggested that the effects of radiation are minimal for inference workloads and more problematic for model training, but now the Google team will be able to evaluate these dynamics in practice rather than in simulation.
If the spacecraft’s TPU proves capable, the next step will be to launch two more satellites with advanced chips in 2027. These spacecraft will support higher power generation and demonstrate optical links between satellites.
Google’s Suncatcher project is almost alone in its vision for space-based data centers, with plans for high-speed links between satellites to support coherent distributed workloads. SpaceX’s data centers are also expected to use laser links, but they will be orbiting too far away to efficiently share processing. Companies like Starcloud and Cowboy Space expect to run inference loads on individual satellites.
Google and SpaceX’s visions are closely related. Google’s parent company, Alphabet, has a large stake in SpaceX. And the companies will need Starship, the heavy-lift rocket being developed at the center of Elon Musk’s space vision, to be fully and rapidly reusable to launch an orbital data center.
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 cost savings seen with earlier vehicles.
Suncatcher is the quintessential Google moonshot. The company is focused on designing infrastructure that will be economical in the 2030s. Getting there, however, may depend on how other parts of the company respond to the pressures and opportunities of AI in the short term. Google’s parent company Alphabet reported this week that its efforts to incorporate AI into its search business were hurting advertising revenue growth, sending its stock price plummeting.
Correction: The Suncatcher prototype satellite has four TPUs, but only one can operate at a time.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
