techcrunch.com web signal

Google's Suncatcher Prototype Comes Online, Runs Gemma 4 in Orbit


<span>Google's Suncatcher Prototype Comes Online, Runs Gemma 4 in Orbit</span>
Google AI Infrastructure Chips ai-infrastructure

TL;DR

  • Google's Suncatcher prototype, with four TPUs on a 1 kW power budget, is running Gemma 4 in orbit after a Falcon 9 launch this month.
  • The spacecraft can only power one TPU for about 20 minutes at a time, complicating already-tricky space-to-ground communications.
  • Google plans two more satellites with advanced chips and inter-satellite optical links in 2027, targeting infrastructure 'economical in the 2030s.'

Google says its Suncatcher prototype, a four-TPU satellite the company calls a space data center, is online and talking to the ground after launching this month on a SpaceX Falcon 9 alongside more than 100 other spacecraft, TechCrunch reports. Engineers asked the onboard Google DeepMind Gemma 4 model to write a haiku about being in space; the opening line was "Blue marble below." A Google spokesperson told the outlet the satellite was "operating nominally."

The engineering story is the power envelope. The spacecraft generates up to 1 kW of electricity, and "the power demands of the TPU mean it can only turn one on for about 20 minutes at a time, complicating the already tricky business of space-to-ground communications." Four TPUs are flying; one is on at any given moment. The roughly three-week commissioning campaign is primarily a radiation and thermal-stress test, measuring real on-orbit effects on the chips rather than the Earth-side simulations the team had been running.

If the TPUs hold up, two more Suncatcher satellites "equipped with the advanced chips" are scheduled for 2027, with higher power generation and inter-satellite optical links. The economic case Google itself has sketched only reaches infrastructure "economical in the 2030s," and leans on SpaceX hitting a steep cost curve: Suncatcher researchers estimated Starship would need to fly 1,800 times to repeat the learning-driven cost declines seen with its earlier vehicles. It is a small hardware bet set against a very large balance sheet; Alphabet's latest 10-Q disclosed more than $1T in AI commitments this week.