AI on Satellites: A Startup's Plan to Make Orbit Think

Satlyt, a two-year-old startup with headquarters in Sunnyvale, California and Nairobi, Kenya, has raised an eight-million-dollar seed round to build software for AI on Satellites — models running on hardware already circling the planet. The financing was first reported by TechCrunch. The round was led by Houston-based Non Sibi Ventures, according to SmallSat News reporting on the announcement.

The investor list behind the raise is a mix of American and Africa-focused venture firms: TLcom Capital, Antler, Slauson & Co., Launch Africa Ventures, Enza Capital, Askya Investment Partners, Axian Investment, Gaingels, Demos and BAG Collective joined the round alongside existing backers, WeeTracker reported. With the fresh capital, the company plans to grow its engineering team around its Nairobi hub while pushing product development and wider deployment of its software across more spacecraft.

Satlyt's pitch starts with a physical problem. Modern satellites generate enormous volumes of imagery, sensor readings and telemetry, but the pipe that carries data back to Earth — the downlink — is narrow, expensive and slow. Most satellites today beam raw data home and wait for ground computers to make sense of it. Satlyt flips that sequence: its hardware-agnostic software runs AI workloads on the computers already aboard third-party satellites, so onboard models decide what matters and transmit only that. It is edge computing, moved into orbit.

The idea came from co-founder Rama Afullo, a Kenyan-American engineer who previously worked on cloud computing at Google and on the Starlink team at SpaceX. Afullo told TechCrunch he pitched orbital computing internally at both companies and was turned down at each, then left to co-found Satlyt in 2024. He describes the strategy with a smartphone analogy: if SpaceX's vertically integrated approach is the iPhone of orbital computing, Satlyt wants to be the Android — an open software layer that runs across satellites built by different companies.

"Running applications on existing spacecraft gives operators a way to do more with the computing hardware they already have in orbit... Our job is to make those applications practical to deploy and operate," Afullo said in comments carried by the announcement coverage.

A Google AI model already flew in orbit

Satlyt has already shown the concept works. Earlier this year the company ran Google DeepMind's open-weights model Gemma aboard a Momentus spacecraft, using it to analyze system logs and stack traces — the debugging records engineers use to diagnose faults. The onboard model shrank those error transmissions by more than 60 percent before anything was sent to Earth, the AI Insider reported. Afullo says that kind of reduction could save hundreds of thousands of dollars per satellite each year, a powerful argument for operators who pay dearly for every byte beamed home.

The company just put its software into orbit on a bigger test. Its latest deployment flew aboard a SpaceX rideshare mission on a TakeMe2Space satellite, carrying workloads for several clients at once: research software from a NASA-funded project testing cloud-computing protocols in orbit, image-processing software for space-surveillance startup Stellerian, and application hosting for TakeMe2Space itself, TechCrier reported. Prototype hardware for Google's Project Suncatcher, the tech giant's own orbital data-center experiment, shared the same ride.

What comes next is a scale-up of that multi-satellite idea. The company plans to demonstrate distributed computing tasks synchronized across two satellites next year, TradersUnion reported, with a stated long-term target of running its software on 20 percent of satellites by 2030. The seed funding also lands as the market for AI on Satellites heats up, with investor appetite for AI infrastructure beyond the data center running high, a trend tracked in our AI coverage and business coverage.

The rivals building the hardware instead

Satlyt's approach is deliberately asset-light. Rivals including Starcloud and Cowboy Space Company are taking the heavier, more expensive bet: building dedicated data-center hardware and launching it into orbit. Satlyt argues that the hardware is already up there — most satellites carry onboard computers that sit underused — and that a software layer can turn that idle capacity into shared cloud infrastructure, much as virtualization software once did for server rooms on Earth.

The asset-light route is cheaper but comes with harder questions. Satlyt has to prove its software can run reliably on hardware it did not build, stay secure against attacks, and work with the systems of many different satellite operators — all in an environment where a software failure is hundreds of kilometers from the nearest reboot button. The 2030 target, which would make its platform the default operating layer for a large share of the space industry, is the kind of goal that reads as either vision or hubris depending on how the next two years of missions go.

Why orbiting computers matter to people on the ground

For readers on Earth, the promise is faster answers from space. Earth-observation satellites that analyze imagery in orbit could flag wildfires, floods or oil spills in minutes instead of after a slow downlink and a ground-station queue. Connectivity and navigation systems could diagnose their own faults without waiting for instructions from mission control. And the Nairobi engineering hub, backed by Africa-focused investors in the round, puts the company at the center of Africa's growing space-tech scene — a sign that the next space-race supply chain will not be built entirely in California.