Generated by Codex with GPT 6 Luna XHigh

Techmeme surfaced Kevin Roose’s September 24 New York Times report, “Google’s Space AI Data Center Project Suncatcher”. Google plans to launch a refrigerator-sized prototype called MVP on October 1, carrying four Tensor Processing Units (TPUs) into low Earth orbit. The point is to find out whether AI hardware can work there, not to deploy a useful orbital data center next week.

A test of the hard parts

The attraction is steady sunlight: in its test update, Google says solar panels in the right orbit could produce up to eight times more power than on Earth. But power is only one constraint. TPUs concentrate heat, and a vacuum has no air to carry it away. Heat must move through pipes to radiators and then escape as radiation. Google says the prototype will test that cooling system in orbit; Ars Technica reports that the chips will run for about 15 minutes at a time before needing a cooling break. The satellite is expected to remain in orbit for a year, which is very different from running its processors continuously for a year.

Launch loads and radiation are other unknowns. Google says vibration testing survived forces meant to simulate the rocket ride, and proton-beam tests suggest its Trillium TPUs tolerate more radiation than a five-year mission would deliver. Those are encouraging lab results, but only the flight can show how the hardware behaves in orbit. The company developed this first mission with Planet Labs for SpaceX’s Transporter-18 rideshare.

From one satellite to a cluster

The larger idea is a constellation of solar-powered satellites, each carrying many chips and linked by lasers. Training or serving large models across them would require precise formation control and fast, reliable connections. Google’s earlier design paper sketches the system; this mission tests neither a large cluster nor its economics. Google says two satellites are planned for a 2027 test of inter-satellite links.

That gap matters. Suncatcher’s near-term question is whether a TPU can survive and compute in space. A scalable service would also need sustained cooling, high-bandwidth links, repair-free reliability, affordable launches, and coordination among moving satellites. The next useful signal is operational data from the prototype—especially how much compute it can deliver between thermal pauses—not the more distant promise of space replacing ground data centers.