Ex-Google and SpaceX PM raises $8M for Satlyt's on-satellite AI software

Satlyt, a startup building software that runs AI workloads on satellites, has raised an $8 million seed round led by Houston-based Non Sibi Ventures, TechCrunch reports. The company was founded by Rama Afullo, a Kenyan American engineer who worked in Google's cloud business and spent part of 2024 at SpaceX's Starlink. According to TechCrunch, he pitched the idea of computing in space internally at both companies, was turned down each time, and left to build it himself.

The funding arrives just as the field heats up. Satlyt's software is scheduled to fly on a SpaceX rocket this week, in the same launch as the first prototype for Google's orbital data center effort, Project Suncanter — and SpaceX itself has since gone all-in on space-based data centers.

Selling the operating layer, not the hardware

Unlike SpaceX, Google, or startups such as Starcloud and Cowboy Space Company, Satlyt has no intention of building its own spacecraft. Operating from Sunnyvale, California and Nairobi, the team is writing the software layer that would let AI models run across satellites owned by many different operators. Afullo compares the approach to VMware and Snowflake, platforms that let users run demanding software without managing the infrastructure underneath.

He frames the competitive position in mobile terms: if the companies building orbital data centers are the iPhone, Satlyt wants to be the Android — a horizontally integrated, open ecosystem spanning many vendors' satellites.

Where the savings come from

The near-term business case rests on the cost of moving data. Downlink — transmitting information from a spacecraft back to the ground — is slow and expensive, and satellites today generally depend on flight controllers on Earth to resolve problems. AI running onboard can triage anomalies and process sensor readings before anything is sent home.

Satlyt has already flown its software on two demonstration missions. Earlier this year it deployed Google DeepMind's Gemma model on a spacecraft operated by Momentus, and TechCrunch reports the model cut the size of a transmission about onboard software errors by more than 60%. Afullo says that kind of efficiency can save hundreds of thousands of dollars per satellite each year.

This week's launch puts the software on a spacecraft built by TakeMe2Space, an Indian startup that makes computing hardware for satellites. The mission serves three customers: NASA, which is paying Satlyt to test protocols for cloud computing in space; Stellerian, a space-surveillance startup testing image-processing workloads; and TakeMe2Space itself, demonstrating that the satellite can host other companies' software.

Next on the roadmap is a bigger technical milestone: creating a shared computing environment spanning two different satellites, which the company expects to attempt next year. Success would be a first step toward a genuine compute cloud in orbit, offered as a third-party service to spacecraft builders. In Afullo's telling, the model turns a satellite into a revenue-generating managed service.

A bet that doesn't need orbital data centers

Few high-powered GPUs are in orbit today, and the grand vision of massive space data centers still faces uncertain economics and a launch bottleneck. That uncertainty is precisely what attracted Satlyt's lead investor. Non Sibi Ventures partner Kent Lucas told TechCrunch the company doesn't need data centers in space to succeed — the business can grow with the rising number of satellite launches alone. The firm's partner Bernard Harris, who served as a NASA astronaut for more than two decades, helped underwrite its conviction in the space-based business.

Afullo's stated goal is to be live on 20% of satellites by the end of the decade, when he expects virtually every spacecraft builder to be flying GPUs or comparable processors. His argument is blunt: launching a satellite without a GPU is, at minimum, leaving value on the table.

Why it matters

An $8 million seed is small by AI infrastructure standards, but the direction it signals is notable. As SpaceX, Google and well-funded startups race to build orbital data centers, someone will have to write the software that turns scattered satellite hardware into a usable, shared compute platform. Satlyt is betting that the operating layer — not the spacecraft — is the durable business, much as cloud software outlasted any single server vendor on the ground. Its measurable downlink savings also give it a business case that works today, with the anomaly-handling and sensor-processing workloads satellites already have, rather than depending on a far-off future of orbital mega-data centers. If onboard GPUs become standard equipment, the software that manages them becomes infrastructure every satellite builder needs — and Satlyt is one of the first companies trying to own it. The company is also among the startups presenting at TechCrunch's Battlefield competition at Disrupt in San Francisco on October 13–15.