Image: Google

Google's first Project Suncatcher satellite is now in orbit. The prototype, built with satellite maker Planet, rode SpaceX's Transporter-18 rideshare out of California alongside 129 other payloads. Google says its team has made contact and the satellite is "operating as expected," and Planet confirmed it has reached the spacecraft, too.

This isn't a data center yet. It's a test of whether Google's TPUs, the AI chips it uses on Earth, can survive launch, radiation and heat in orbit. According to Ars Technica, the satellite (called MVP) carries just four TPUs, gets about a kilowatt of solar power, and can run Gemini workloads for only about 15 minutes before the chips need to cool down.

Project Suncatcher has liftoff! ☀️🚀

Announced last year, this moonshot project explores if we can one day host machine learning infrastructure in space. After years of research, we’ve just launched a prototype satellite, built in partnership with @planet, into orbit on… pic.twitter.com/GaxyvUoSbA

— Google AI (@GoogleAI) October 2, 2026

Google first announced Suncatcher in November 2025, pitching tight clusters of solar-powered satellites packed with TPUs and linked by lasers. The original plan was two prototypes with Planet by early 2027, and Google says two satellites are still coming in 2027 to test the laser links. Ars reports this extra pathfinder was added to get data sooner, using a satellite Planet had already built.

What Google claims, and what it still has to prove

Many of the headline numbers are Google's own. It says solar panels in the right orbit can be up to eight times more productive than on Earth. In proton-beam tests, its Trillium TPUs' memory showed glitches only after about 2 krad(Si), nearly three times the expected shielded dose for a five-year mission, with no hard failures up to 15 krad. Its preprint also projects launch prices could drop below $200 per kilogram by the mid-2030s. At that point, Google argues, costs could roughly match a ground data center's energy bill. For comparison, BCG puts today's launch prices at about $1,500 per kilogram.

Then there are the hard problems. Space has no air to carry heat away, so chips must dump it through radiators. BCG estimates a 100 kW satellite needs about 400 square meters of them. Google wants links of tens of terabits per second between satellites, which in one design means flying them just 100 to 200 meters apart. Big, tightly packed constellations raise collision and debris worries, and astronomers have already asked the FCC to deny SpaceX's much bigger plan. Latency matters, too: BCG argues chatbots and other real-time AI belong on Earth, while batch jobs suit orbit better.

Google's model of an 81-satellite cluster about 2 km across. Image: Google

Google isn't first, either. Nvidia-backed Starcloud put an H100 GPU in orbit in November 2025, trained a tiny model up there and even ran Google's Gemma. SpaceX has applied for up to a million orbital data center satellites, Blue Origin has filed for up to 51,600, and China's Three-Body Computing Constellation launched 12 satellites in May 2025.

Our take: Google's slow, step-by-step approach looks more credible than filing for tens of thousands of satellites, and publishing a peer-reviewed paper in Joule helps. But there's an awkward irony. Google's chips just flew on a rocket from SpaceX, a would-be rival, and its cost case depends on cheap, plentiful launches. Starcloud's CEO already says launch is "pretty constrained right now."

So, some open questions. Will Google publish real in-orbit error rates and cooling data, or just a success story? Does the $200/kg forecast hold if fully reusable rockets like Starship arrive late? BCG thinks orbital compute will still cost about 1.5 times as much as ground data centers in a decade, even with realistic improvements. Until those answers arrive, Suncatcher is a science experiment, not a quick fix for the backlash against data centers down here.