Unlike SpaceX, Google or startups such as Starcloud, Satlyt has no plans to build its own spacecraft. Afullo compares his company to VMware and Snowflake, whose platforms let users run software without managing the underlying infrastructure, and says Satlyt wants to work across many companies' satellites rather than compete with the orbital data centre builders.

The company has already flown its software on two demonstration missions. Earlier this year it deployed Google DeepMind's Gemma AI model on a spacecraft operated by Momentus, cutting the size of a transmission about onboard software errors by more than 60%. Because sending data back to Earth is expensive and slow, processing readings on the spacecraft itself can save real money; Afullo estimates hundreds of thousands of dollars per satellite each year.

A coming launch will put Satlyt's software on a spacecraft built by TakeMe2Space, an Indian startup, with three customers: NASA, which is paying to test protocols for cloud computing in space; space surveillance startup Stellerian, testing image processing; and TakeMe2Space itself. Next year Satlyt aims to demonstrate a shared computing system spanning two different satellites, a step toward an orbital compute cloud run as a third party. Afullo hopes to be live on 20% of satellites by the end of the decade.