Google's AI Chip Test Satellite Is Up and Writing Poetry
Google's Suncatcher prototype satellite is online and talking to the ground, as of October 30, 2026, according to TechCrunch. The spacecraft carries one Google Tensor Processing Unit (TPU), a chip built for AI work. It is also loaded with Google DeepMind's Gemma 4 AI model.
The satellite launched earlier in October 2026 on a SpaceX Falcon 9 rocket, along with more than 100 other spacecraft. A September 24 Ars Technica report had said the test was scheduled to launch on October 1.
Google calls Project Suncatcher a "moonshot" to put AI in space. Its long-term plan is a group of AI satellites working together (Google).
The prototype is meant to fly in low Earth orbit, so Google can see how its AI chips hold up in space (Reuters). The project also checks whether hardware can survive in space, and it tests ways to cool AI chips (Google).
Power is the main limit. The satellite makes up to 1 kilowatt of electricity. Its TPU can only run for about 20 minutes at a time, because of how much power it needs.
Before the chip was ready for work, the satellite spent about three weeks in commissioning. During that period, engineers tested its systems.
Once that was done, the engineers' first request to Gemma 4 was for a haiku about being in space. They also tried space-themed limericks and asked the spacecraft to report on conditions in orbit. A Google spokesperson said the prototype is operating nominally, which means within normal parameters.
The testing will continue. Engineers will watch how the chips handle heat and the radiation found in space. Ground tests suggested that radiation would have little effect on inference, which is when a trained model answers questions. Training a model, a much heavier job, would be more of a problem.
If the chip performs well, Google plans to launch two more satellites in 2027, with more advanced chips. These are meant to generate more power and to test laser links between satellites. Both are plans only. Neither has flown, and neither has a confirmed date beyond 2027.
In my view, the power limit is the detail that matters most. Twenty minutes of operation is enough to see how a chip reacts to radiation and heat. It is a long way from a computer that is always available. The 2027 satellites are where the design would have to work as a complete system, not just as a single part.
The broader context here is the basic problem Suncatcher is trying to solve. Data centers on the ground need large amounts of power, cooling, and land. In space, heat cannot be carried away by air or water, so it has to be released as radiation. Power comes from solar panels, and the question is how much one small spacecraft can produce. Whether these trade-offs work in Google's favor will depend on the data from the tests now underway. For an industry that has spent years building AI capacity on the ground, the prototype is a practical test of whether some of that capacity could one day be placed elsewhere.
