Satlyt is betting that the emerging market for computing in orbit will need a software layer that is not tied to any single spacecraft builder. TechCrunch reports that the startup has raised an $8 million seed round to develop software for running AI models on satellites and, eventually, sharing large computing jobs across machines owned by different companies.

Founder Rama Afullo previously worked in Google’s cloud business and briefly at SpaceX’s Starlink network. He later co-founded Satlyt, which has headquarters in Sunnyvale, California, and Nairobi. Rather than manufacturing satellites, the company wants to provide an operational layer that hides differences in the underlying hardware, borrowing the logic of cloud-software companies that let customers use infrastructure without managing every component themselves.

A satellite processes and compresses operational data before sending it to Earth.
Onboard AI could reduce costly downlinks by analyzing errors and sensor readings in orbit.

That hardware-neutral approach separates Satlyt from companies pursuing complete orbital data centers. Afullo compared the strategy to building an open ecosystem across spacecraft fleets instead of a closed stack controlled by one operator. The analogy remains an ambition: TechCrunch reports that Satlyt has flown its software on two demonstration missions, but it has not yet shown a production cloud spanning many satellites.

The immediate use case is more modest and practical than a large data center in space. Satellites still rely heavily on ground controllers, while transmitting information back to Earth can be slow and expensive. Software that analyzes sensor readings or diagnoses anomalies onboard could reduce the amount of data that must be downlinked and let a spacecraft respond more quickly.

Earlier in 2026, Satlyt deployed Google DeepMind’s Gemma model on a spacecraft operated by Momentus. According to TechCrunch, the model reduced the size of a transmission about onboard software errors by more than 60 percent. Afullo said efficiencies of that kind could save hundreds of thousands of dollars per satellite each year, though the report does not provide independent cost verification.

Two satellites share computing workloads through a protected orbital connection.
Satlyt plans to test a cloud that spans two separate satellites next year.

Satlyt’s next launch is scheduled to put its software on hardware built by Indian startup TakeMe2Space. The mission includes three customers with different tests: NASA is paying to evaluate cloud-computing protocols in orbit, space-surveillance startup Stellerian plans to run image-processing workloads, and TakeMe2Space wants to demonstrate that its satellite can host outside software.

The more difficult milestone is planned for next year, when Satlyt intends to connect computing across two separate satellites. Success would support its claim that spacecraft can operate as a shared cloud rather than isolated machines. Significant limits remain: few powerful graphics processors are currently in orbit, launch capacity is constrained, and TechCrunch notes that the economics of large orbital data centers are still uncertain.

The seed round was led by Houston-based Non Sibi Ventures. Satlyt’s long-term target is to operate on 20 percent of satellites by the end of the decade, based on Afullo’s expectation that advanced processors will become standard equipment. That goal is aspirational. For now, the company’s evidence is narrower but concrete: two demonstrations, a new mission with paying users, and a plan to test whether satellite computing can cross the boundaries between spacecraft manufacturers.