Google Just Put Its AI Chip in Space — Here's Why That Matters

Google sent a satellite into orbit today on a SpaceX rocket launched from California, and inside it is one of Google's own AI chips — a Tensor Processing Unit, the company's answer to Nvidia's famous chips. It's the first time Google has put its advanced hardware in space. The satellite, built by Planet Labs, will switch the chip on in short 15-minute bursts to test whether it can run AI without overheating or running out of power. As one Google executive put it, ground tests are good, but nothing beats the real thing.

The bigger idea is called Project Suncatcher: eventually building data centers in orbit. Data centers are the giant warehouses of computers that power everything from search to chatbots, and they eat enormous amounts of electricity, land, and water for cooling. In space, sunlight is available around the clock and nobody minds the heat. Google's vision is a tight formation of 81 satellites working together, passing data between each other with lasers. The company is clear that this is a long-term "moonshot," aimed at the AI workloads of five years from now, not today.

Here's the catch, and it's a big one: rockets. Google's newly peer-reviewed research paper estimates that launch prices could fall to roughly $200 per kilogram by 2035 — but only if SpaceX's Starship hauls about 370,000 tons into orbit. That means roughly 1,800 Starship launches over ten years, or 180 a year. Starship has never flown more than five times in a single year. Elon Musk has promised far more, but he promises a lot of things.

The good news is radiation. Google re-ran its chip tests in a particle accelerator after realizing its earlier setup gave the chips more shielding than they'd actually get in space. The updated results showed about one error in a million operations for typical AI tasks — easily survivable for a satellite's five-year life. Giant, months-long training runs, however, remain a problem.