Satlyt raises $8M for orbital AI software

Rama Afullo left SpaceX in 2024 convinced that computers in orbit could do more than relay data. His new company, Satlyt, has raised an $8 million seed round to build software that runs AI models on satellites, the company told TechCrunch exclusively.

Afullo, a Kenyan American engineer who previously worked in Google's cloud computing business and had a brief stint at SpaceX's Starlink network in 2024, co-founded Satlyt with headquarters in Sunnyvale, California, and Nairobi. The company's software is designed to let many satellites share the work of large computing jobs in space, rather than tying each spacecraft to a single operator or vendor.

Pitching orbital compute internally

Afullo says he tried to sell the idea inside both of the companies where he worked before striking out on his own.

"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 tells TechCrunch.

— Rama Afullo, co-founder of Satlyt

So he left SpaceX and started Satlyt with the goal of building software for satellites. The company's pitch is that orbital data centers are coming, and that the software layer to operate them will matter as much as the hardware itself.

A software layer, not a spacecraft builder

Unlike SpaceX, Google, or startups such as Starcloud and Cowboy Space Company, Satlyt does not plan to build its own spacecraft. Afullo's team is working on software to operate AI models on satellites, an approach he compares to VMware and Snowflake — platforms that let users run complex software without worrying about the underlying infrastructure.

Afullo frames the strategy as an alternative to vertically integrated orbital data centers. "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 other words, the company wants to offer software that works across many companies' satellites.

Two demonstration missions already flown

Satlyt has already flown its software on two demonstration missions, according to the company. Its initial focus is finding operational efficiencies for spacecraft, which typically depend on flight controllers on the ground to solve problems that arise.

Sending data back to Earth, known as downlink, is expensive and often slow. Using AI to make decisions on the spacecraft can mean real savings, either when resolving anomalies or by processing sensor readings onboard instead of on the ground.

Gemma model cut transmission size by more than 60%

Earlier this year, Satlyt deployed Google DeepMind's Gemma AI model onboard a spacecraft operated by Momentus. According to Afullo, the model cut the size of a transmission about onboard software errors by more than 60% — the kind of efficiency he says can save hundreds of thousands of dollars per satellite each year.

That result is the clearest data point the company has offered so far for its core claim: that processing data in orbit, rather than shipping it to the ground, can reduce both cost and delay for satellite operators.

Thursday's launch and its three customers

The upcoming launch will put Satlyt's software to work onboard a spacecraft built by TakeMe2Space, an Indian startup that builds computing hardware for satellites. The mission involves three customers, each testing a different workload.

• NASA, which is paying the company to test protocols for cloud computing in space.
• Stellerian, a startup focused on space surveillance — tracking objects in orbit — that wants to test image processing workloads.
• TakeMe2Space itself, to show the satellite can host other companies' software.

Satlyt's software is set to launch Thursday on a SpaceX rocket alongside the first prototype for Google's space data center effort, Project Suncatcher. SpaceX has since gone all-in on orbital data centers, a shift from the period when Afullo says his internal pitches were turned down.

Few GPUs in orbit today

Right now, few high-powered GPUs are in orbit. For the moment, the most likely opportunity lies in the kinds of workloads Satlyt is already working on — anomaly resolution, sensor processing, and image analysis — rather than in general-purpose cloud computing at scale.

Satlyt's next project is to show that it can create a shared computing system — a cloud — spanning two different satellites. The company expects to attempt that next year. If successful, it would allow Satlyt to build a true compute cloud in orbit as a third-party provider to spacecraft builders.

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

Non Sibi Ventures led the seed round

Satlyt's seed round was led by Houston-based Non Sibi Ventures. Partner Bernard Harris, who served as a NASA astronaut for more than 20 years, helped ground the firm's conviction in the novel, space-based business.

One reason the firm invested was that Satlyt does not depend on the highest-octane vision of space data centers, which faces 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."

A target of 20% of satellites by 2030

Afullo's hope is to be live on 20% of satellites by the end of the decade, when he expects virtually every spacecraft builder to be putting GPUs and similar advanced processors in 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 says.

Satlyt is among dozens of other startups vetted by TechCrunch and appearing at Disrupt as part of the broader Battlefield competition happening October 13-15 in downtown San Francisco.

What the bet hinges on

The company's case rests on a specific reading of where satellite hardware is headed: that advanced processors become standard equipment, and that operators will want a software layer to make those processors useful across mixed fleets. That is a different bet from building orbital data centers outright, and it is the one Non Sibi says it is comfortable with.

For now, the evidence Satlyt can point to is limited to two demonstration missions and a single published efficiency figure from the Gemma deployment. The company's plan to link two satellites into a shared computing system, expected next year, would be the first test of whether its software approach works beyond a single spacecraft.

For businesses and government agencies that buy satellite data or operate spacecraft, the near-term question is narrower than the orbital-data-center debate: whether onboard AI can cut downlink costs and speed up anomaly response in practice. Satlyt's flying software — including the mission alongside Google's prototype — gives it a chance to demonstrate that on real hardware rather than in a pitch deck. This suggests the coming years could show whether the software layer, not just the hardware, becomes a meaningful part of how orbital computing gets sold.