Satlyt raises $8M to run AI software on satellites

Satlyt, a company building software that lets satellites run artificial intelligence models, has closed an $8 million seed round, its co-founder Rama Afullo told TechCrunch.

The Kenyan American engineer had earlier spent time in Google's cloud computing division and then had a short stay at SpaceX's Starlink communications network in 2024. Those experiences left him convinced that computers in orbit held unexploited potential.

He recalled that he attempted to sell the idea inside both organisations. "When I was at SpaceX, I tried to pitch this internally. They said no. When I was at Google, I tried to pitch this internally. They said no," he said.

He subsequently departed SpaceX and set up Satlyt, which has offices in Sunnyvale, California, and Nairobi, Kenya. The company's goal is to write software for satellites that would allow many spacecraft to divide up the work of heavy computing tasks performed in space.

Market conditions appear to favour the venture. SpaceX has in the meantime committed heavily to orbital data centres, and Satlyt's software is due to fly Thursday aboard a SpaceX rocket, on the same mission as the first prototype for Google's space data centre project, known as Project Suncatcher.

Satlyt, however, has no plans to construct its own satellites, in contrast with SpaceX, Google and startups such as Starcloud and Cowboy Space Company. Afullo's team concentrates on software that runs AI models on satellites, an approach he compares to VMWare and Snowflake, whose platforms allow customers to run sophisticated software without concerning themselves with the infrastructure underneath.

"The folks like SpaceX who are doing orbital data centers—if they are the iPhone, we'll build Android as a horizontally integrated, open ecosystem," he said. In practice, that means Satlyt intends to supply software capable of operating across satellites made by a range of manufacturers.

Two demonstration missions have already carried the company's software. Its first priority is operational efficiency for spacecraft, which normally rely on ground-based flight controllers to fix whatever goes wrong. Downlinking data to Earth is costly and frequently sluggish, so allowing AI to take decisions aboard the spacecraft can produce genuine savings, whether by handling anomalies or by processing sensor data on the satellite rather than on the ground.

Earlier this year, Satlyt ran Google DeepMind's Gemma AI model on a spacecraft operated by Momentus. That model shrank the size of a transmission concerning onboard software errors by more than 60%, an efficiency gain Afullo says can translate into hundreds of thousands of dollars saved per satellite annually.

This week's launch will install Satlyt's software aboard a spacecraft assembled by TakeMe2Space, an Indian startup that develops computing hardware for satellites. Three customers are involved in the mission: NASA, which is paying the company to test protocols for cloud computing in space; Stellerian, a startup working on space surveillance, or the tracking of objects in orbit, which wants to test image processing workloads; and TakeMe2Space itself, which aims to demonstrate that its satellite can host other companies' software.

High-powered GPUs remain scarce in orbit at present. The most probable near-term opening lies in the workloads Satlyt is already pursuing. Its next undertaking, which it expects to attempt next year, is to demonstrate a shared computing system—a cloud—that spans two separate satellites. Should that succeed, the company would be able to operate a genuine compute cloud in orbit, supplying spacecraft builders as a third-party provider.

"The way to think about it is that we turn your satellite into a revenue-generating managed service," Afullo said.

Houston-based Non Sibi Ventures led the seed round. Partner Bernard Harris, a former NASA astronaut with more than 20 years of service, helped anchor the firm's confidence in what it sees as a novel space-based business.

Part of the appeal for the investor was that Satlyt does not hinge on the most ambitious version of the space data centre idea, which confronts uncertain economics and a launch bottleneck. "We don't need data centers in space for Rama to be wildly successful, right?" Non Sibi partner Kent Lucas told TechCrunch. "It can just be driven by the number of satellites going up."

By the end of the decade, Afullo hopes Satlyt's software will be live on 20% of satellites. He expects that by then virtually every spacecraft builder will be fitting GPUs and comparable advanced processors to their satellites.

"If you're putting up a satellite without putting up a GPU on it, at the very least, you're doing yourself a disservice, right?" he said.

Satlyt is one of dozens of startups reviewed by TechCrunch that are appearing at Disrupt through its wider Battlefield competition, which runs October 13-15 in downtown San Francisco.

Comments

Join the conversation: Sign in

No comments yet. Be the first to share your thoughts.