Satlyt raises $8 million to run AI models on other companies' satellites
Satlyt says it has already run Google DeepMind's Gemma model on a Momentus spacecraft, cutting the size of error reports sent to the ground by more than 60%.
Satlyt, a startup that writes software for running AI models on satellites, has raised an $8 million seed round led by Houston-based Non Sibi Ventures, TechCrunch reported on Thursday. The company does not build satellites. Chief executive Rama Afullo told TechCrunch it aims to be a software layer that works across many operators' spacecraft.
Afullo previously worked in Google's cloud division and briefly at SpaceX's Starlink, and pitched orbital computing inside both before founding Satlyt, with offices in Sunnyvale and Nairobi, according to TechCrunch. He contrasted the company's open approach with what the outlet called SpaceX's closed, all-in-one, iPhone-style model.
The company says it has completed two demonstration missions. In one it deployed Google DeepMind's Gemma model on a Momentus spacecraft, which TechCrunch says cut the size of software error reports sent to the ground by more than 60%. A third mission is due to launch this week on a TakeMe2Space satellite, with NASA and Stellerian among its customers.
Afullo told TechCrunch he wants Satlyt on 20% of satellites by 2030, and expects builders to treat a processor as standard kit. "You're putting up a satellite without putting up a GPU on it, at the very least, you're doing yourself a disservice," he said. The 20% figure is the founder's goal, not a forecast we can check.
Big companies are testing the same idea. International Cyber Digest reported on Thursday that Google launched one of its AI chips into orbit for the first time, on a Planet Labs-built satellite carried by a SpaceX rocket, as the first hardware test for Project Suncatcher. It said Google estimates space data centres need about 1,800 Starship flights over ten years, a figure we have not verified.