The test, completed earlier this month after 13 months of work, was conducted with the University of Houston as part of a NASA-funded research project. Satlyt’s software assessed how computing tasks could be allocated according to available processing capacity and communications links. The result advances the startup’s effort to turn existing spacecraft into a distributed data-processing network-but it was a software-planning demonstration, not a fully functioning orbital data centre.
The experiment used a simulated network of four computing hosts, four users and eight services. Only one physical computer aboard a spacecraft ran the software; it evaluated the simulated network and sent its output to Earth. The test showed that Satlyt could generate computing plans on hardware in orbit. It did not demonstrate several satellites exchanging data and carrying out shared workloads across real spacecraft.
That distinction defines the next stage of the work. Satlyt founder and chief executive Rama Afullo says that the company needs broader testing across workloads, hardware and connectivity conditions. It also needs to demonstrate computing tasks shared among separate spacecraft, rather than simulated machines running on a single onboard computer. The company has identified a measurement issue in how its software recorded communication delays, but Afullo said the software ran onboard and returned its outputs successfully.
Satlyt’s software is already installed on three satellites in orbit, according to Afullo. The company, founded in 2024, aims to let operators set permissions, priorities and operating limits, then have software assign tasks across available computing resources. The startup’s broader ambition is a virtual computing layer that could let developers run applications in space without building a separate infrastructure for every mission. NASA has not selected Satlyt for a specific mission or chosen its technology for operational deployment.
The potential appeal is less about replacing data centres on Earth than processing selected information closer to where it is collected. Satellites often face limited onboard computing resources and intermittent communications, while contact windows for transmitting data to ground stations can be constrained. If a spacecraft detects a wildfire, for example, onboard analysis could help identify and send relevant information sooner, instead of downlinking large volumes of raw imagery for later processing on Earth.
The concept must contend with real constraints. Spacecraft have limited power, processing capacity and communications bandwidth; links can be intermittent, and systems need to coordinate securely despite changing connectivity. NASA’s own Delay/Disruption Tolerant Networking approach is designed for space communications that face delays or interruptions, using store-and-forward mechanisms so data can be held until a connection becomes available. That is a communications framework, not evidence that Satlyt’s system is already operating as a multi-satellite computing network.
Afullo described the test as revealing trade-offs: one workload-planning approach performed best but required more computing resources than alternatives. That balance matters in orbit, where a more sophisticated scheduling method must justify the processing power it consumes. Satlyt says it still needs to validate its software under varied operating conditions and address practical requirements including usable communications links, operator permissions, secure transfers, resource scheduling and recovery when a network connection becomes unavailable.
The work is part of a broader trend towards moving some analysis closer to sensors and spacecraft. NASA has developed networking technologies to improve data movement across disrupted links, while companies are exploring onboard processing and satellite-to-satellite communication. These efforts address related but distinct challenges: networking moves information reliably, while Satlyt is working on software to decide where computing tasks should run.
Satlyt’s immediate goal is broader validation and integration, followed by demonstrations of computing work shared between separate spacecraft. Its test marks a building block, not a commercial service or a proven orbital data-centre network. Whether the model can scale will depend on reliable links, available onboard resources and convincing operational demonstrations.
