Google's Project Suncatcher satellite runs Gemma 4 AI model in orbit

Google says its Project Suncatcher prototype, a satellite built to test whether artificial intelligence computing can work in space, has run the company's Gemma 4 AI model in orbit and sent a complete response back to Earth. The milestone was reported on Friday, October 30, 2026. Google's team sent a command to the satellite, switched on one of its Tensor Processing Units (TPUs, Google's own AI chips), ran a prompt asking for a haiku about space, and received the answer as the satellite passed over Svalbard, Norway. Google chief executive Sundar Pichai called it a "hello world" moment for the project.

Training and running AI models takes enormous amounts of electricity, and data centres on the ground are running into limits on power, land and water for cooling. The idea behind Project Suncatcher is that satellites in orbit could draw on near-constant sunlight through solar panels and run AI chips there instead. A data centre in space is still a distant prospect, but before anything larger can be built, engineers need to show that ordinary AI chips can switch on, run a real model and return results in the harsh conditions of orbit. That is what this test demonstrates.

It is important to be clear about scale. The prototype can power only one chip at a time, for about 20 minutes, and it answered short poetry prompts. Some coverage has described it as an orbital data centre, but others treat it as a measured technical step rather than a working facility. The company's own description, a "hello world" moment, is the kind of first test programmers run to check that a system works at all.

There are also serious open questions. Getting rid of heat is difficult in a vacuum, because there is no air to carry it away, and AI chips run hot. Radiation in orbit can damage electronics. Critics also point out that large numbers of satellites add to the risk of space debris and to light pollution that affects astronomers. The next round of heat and radiation testing will show how far the idea can go.

For India, where data-centre operators also need large and reliable power supplies, the experiment is worth following as one of several attempts to find new ways to power AI. If computing in orbit ever becomes practical, it would also matter to satellite builders and launch providers, including India's own space industry.

Short-term: Google will run further tests on the prototype, including stress tests for heat and radiation. Results will show whether the chips stay reliable over time in orbit.

Long-term: If the approach works at larger scale, part of the world's AI computing could one day move off the ground, easing pressure on power grids. If heat, radiation or cost prove too difficult, the project may remain an experiment.

Who is affected: AI companies and data-centre operators looking for new power sources; satellite and launch companies; astronomers concerned about crowded orbits; and, eventually, users of AI services, whose costs depend on the price of computing.

Questions and Answers