Google's Suncatcher Satellite Is Online, Carrying a Single AI Chip in Orbit

Google's Suncatcher prototype satellite is online and communicating with the ground as of October 30, 2026, according to TechCrunch. The spacecraft carries a single Google Tensor Processing Unit (TPU), a custom accelerator chip built for AI workloads, and is pre-loaded with Google DeepMind's Gemma 4 model.

The satellite launched earlier in October 2026 on a SpaceX Falcon 9 rocket, alongside more than 100 other spacecraft. A September 24 Ars Technica report had said the Suncatcher test was scheduled to launch on October 1. Google describes Project Suncatcher as a moonshot to put AI in space, and its stated long-term vision is a constellation of AI satellites (Google).

The prototype is designed to operate in low Earth orbit (LEO), the relatively close band of orbit where most satellites fly, so that Google can assess its AI chips in the space environment (Reuters). The project also covers testing hardware survival in space and designing cooling systems for AI chips (Google).

Power sets the operating envelope. The satellite generates up to 1 kW of electricity, and its TPU can only be switched on for about 20 minutes at a time because of power demands. Before the payload was ready for work, the satellite went through a commissioning phase of about three weeks, during which its subsystems were tested.

Once commissioned, the engineers' first prompt asked Gemma 4 to write a haiku about being in space. The team also tested the model with space-themed limericks and asked the spacecraft to report on orbital conditions. A Google spokesperson said the Suncatcher prototype is operating nominally.

The ongoing test campaign will examine how the chips perform under thermal stress and in the space radiation environment. Ground testing suggested that radiation would have minimal impact on inference workloads, meaning the task of running a trained model on new inputs, while training a model would be more of a problem.

Google's next step depends on the result. If the onboard TPU proves capable, the company plans to launch two more satellites equipped with advanced chips in 2027. Those two spacecraft are intended to support higher power generation and to demonstrate inter-satellite optical links, which are laser connections between satellites. Both are planned, not flown, and neither has a confirmed schedule beyond the 2027 target.

In my view, the power budget is the detail that most shapes what this prototype can tell Google. A 20-minute duty cycle, the share of time the chip is actually active, is enough to characterize how a chip behaves under radiation and thermal load, but it is a long way from a continuously available compute node. The 2027 plan, with higher power generation and optical crosslinks, is the point at which the architecture would have to be shown to work as a system rather than as a single component.

The broader context here is the engineering problem Suncatcher is built around. Terrestrial data centers are constrained by power, cooling, and land. Moving the accelerator into orbit changes the cooling problem to radiative heat rejection, since there is no air or water to carry heat away, and changes the power problem to what a solar array can deliver to a single small spacecraft. Whether those trade-offs eventually work in Google's favor will depend on data from the thermal and radiation campaign now underway. For an industry that has spent the last several years expanding AI capacity on the ground, the prototype is a practical measurement of whether some of that capacity could ever be placed elsewhere.