Satlyt has raised an $8 million seed funding round to develop software that runs artificial intelligence models and distributes computing workloads among satellites. The company differs from SpaceX, Google, and other companies working to build orbital data centers because it does not develop its own spacecraft. Instead, it aims to provide a software layer that can run on satellites operated by multiple parties.
Satlyt was founded by Rama Afullo, a Kenyan-American engineer who previously worked on cloud computing at Google and then briefly worked on SpaceX’s Starlink network in 2024. The company has offices in Sunnyvale, California, and Nairobi. Houston-based Non Sibi Ventures led the investment round, with participation from Bernard Harris, a former NASA astronaut for more than 20 years.
Computing Closer to the Data Source
Many satellites rely on ground control centers to process problems and analyze sensor readings. But sending data to Earth, known as a downlink, can be expensive and slow. Satlyt is therefore initially focusing on running artificial intelligence onboard the spacecraft to make some decisions and process data before transmission.
The company says its software ran Google DeepMind’s Gemma model aboard a spacecraft operated by Momentus earlier this year. The model helped reduce the volume of transmissions related to software errors onboard the spacecraft by more than 60%, a result Satlyt says could save hundreds of thousands of dollars annually for each satellite.
Multi-Party Test in Orbit
Satlyt’s software is scheduled to launch aboard a spacecraft built by TakeMe2Space, an Indian startup developing computing hardware for satellites. The test includes three customers: NASA, which is funding a test of cloud-computing protocols in space; Stellerian, which wants to test image-processing workloads related to space monitoring; and TakeMe2Space, to demonstrate the spacecraft’s ability to host software from other companies.
Satlyt says its next step is to link a shared computing system between two different satellites, an effort it expects to carry out next year. The success of this test would bring the company closer to offering a third-party orbital computing cloud to spacecraft manufacturers.
What Changes in Practice?
The core value of this model is not building a complete orbital data center, but turning the processing capabilities already present on satellites into a service that can be managed and shared. This gives spacecraft manufacturers and operators a software option that is not tied to a single infrastructure, unlike closed models that combine the spacecraft, software, and platform into one ecosystem.
However, the project is still at an early stage. The number of high-powered GPU units in orbit is limited, and the economics of space data centers and launch capacity remain unresolved factors. Satlyt’s near-term viability depends on increasing the number of satellites carrying advanced processors and demonstrating that its platform can operate across different spacecraft, rather than creating huge orbital data centers from the outset. Afullo is targeting deployment of the software on 20% of satellites by the end of the decade, a goal announced by the company and not yet a proven result.