AI on satellites cuts Earth downlinks, Satlyt says

Satlyt, a new space software company co-founded by Rama Afullo, has raised an $8 million seed round to bring AI computing directly to satellites. The plan is to let satellites run AI models onboard rather than sending every decision or data packet back to Earth for processing. The funding, led by Non Sibi Ventures, aims to turn satellites into revenue-generating managed services through software that can operate across different spacecraft and hardware platforms.

Afullo, a former Google cloud manager and SpaceX Space Data Center veteran, says the vision is to create a shared, in-space compute cloud. The idea is to let satellites handle workloads locally, such as anomaly resolution or sensor data processing, which can dramatically cut down on expensive and slow downlinks to Earth. Satlyt has already demonstrated its software on two missions, including a model onboard a spacecraft operated by Momentus that used Google DeepMind’s Gemma AI model. In tests tied to that demonstration, the amount of data sent back to Earth for processing shrank by more than 60% when the AI handled the issue reporting on the spacecraft itself.

Satlyt’s seed round comes at a time when SpaceX is pushing orbital data centers, and Google has its own space data initiatives. Satlyt isn’t building its own spacecraft; instead, it provides the operating software so other satellite builders can run AI workloads on their hardware. The company plans a broader demonstration this week on a SpaceX rocket alongside TakeMe2Space’s first satellite computing prototype. The mission will involve NASA as a customer to test cloud-like protocols for space computing, plus Stellerian, a surveillance-focused startup, and TakeMe2Space itself to validate hosting third-party software in orbit.

In practical terms, Satlyt is pitching a space-based alternative to cloud computing: a way to process data where it’s collected, reducing the need to downlink massive datasets and potentially enabling faster decisions for tasks like observation, navigation, or anomaly detection. Afullo compares Satlyt’s goal to VMware and Snowflake in the space context, framing the product as a platform that lets many satellites share compute resources much like modern cloud providers on Earth.

Satlyt’s investors see opportunity in the growing number of satellites and the push to add GPUs and other powerful processors to spacecraft. Non Sibi partner Kent Lucas says the business case doesn’t require every satellite to carry the most expensive hardware; it’s about enabling the right workloads to run onboard and on a scalable software layer. The plan is to have a foothold on roughly one-fifth of satellites by the end of the decade, a target that underlines how quickly in-space AI ideas could scale if the economics line up.

What this could mean for everyday readers is easier access to timely, space-based data processing without long waits for ground-based analysis. It could also influence how satellite services are priced and how quickly new space applications reach users on Earth. For now, Satlyt is early in its journey, but the company is actively marketing to aerospace companies and agencies looking to reduce latency and downlink costs while expanding the kinds of operations that can run in space.