Satlyt will fly its satellite-computing software on a SpaceX rocket

A software layer for orbit

Afullo, a Kenyan-American engineer, worked in Google’s cloud-computing division before briefly joining SpaceX’s Starlink in 2024. He pitched ideas to leaders at both companies and was turned down. He later left SpaceX and founded Satlyt with partners. The company has offices in Sunnyvale, California, and Nairobi.

Its model is closer to infrastructure software than satellite manufacturing. Afullo compares Satlyt to VMware and Snowflake: platforms that let customers run complex software without having to manage the underlying infrastructure. His analogy for the broader market is an iPhone and Android: if SpaceX and other orbital data-center builders are making the iPhone, Satlyt wants to be the open system that works across satellites from different manufacturers.

Satlyt has already sent its software to space on two test missions. Its early focus has been making spacecraft more efficient. Today, when a satellite encounters a fault, it typically relies on ground control; sending data back to Earth can be slow and expensive. Onboard AI could instead analyse sensor readings or respond to problems locally.

Earlier this year, Satlyt installed Google DeepMind’s Gemma model on a spacecraft built by Momentus. The model cut the volume of software-error data sent down from the spacecraft by more than 60%. Afullo estimates that the reduction could save hundreds of thousands of dollars per satellite each year.

What this launch will test

On Thursday’s launch, Satlyt’s software will run on a satellite built by Indian startup TakeMe2Space, which develops computing hardware for spacecraft. The mission has three customers, each testing a different part of the proposition:

The near-term opportunity is constrained by the small number of powerful GPUs in orbit, so Satlyt is working on tasks it can run with the hardware available. Its next planned project is to test whether two different satellites can form a shared computing system, or cloud. The company expects to run that test next year. If it works, Satlyt could offer spacecraft manufacturers an independent orbital-computing service.

The bet that does not require orbital data centers

Satlyt’s seed round was led by Houston-based Non Sibi Ventures. The firm’s partner Bernard Harris, who spent more than 20 years as a NASA astronaut, helped build investor confidence in the business. One reason to back Satlyt, Non Sibi partner Kent Lucas said, is that the startup does not need the most ambitious orbital data-center plans to succeed. Those projects still face uncertain economics and limited launch capacity; Satlyt could benefit simply from more satellites.

Afullo hopes Satlyt’s software will be installed on 20% of satellites by the end of the decade. He expects nearly every spacecraft manufacturer to equip its satellites with GPUs and other advanced processors by then. He describes the goal as turning a customer’s satellite into a managed service that generates revenue.

I think the appeal is clear: Satlyt can sell into a growing satellite fleet without owning the spacecraft. But the company’s larger promise depends on that fleet becoming capable enough to run useful workloads. Its next test—linking two different satellites into one cloud—will show whether “open system” is more than a software pitch.