Naturally, Rama Afullo’s initial response was, “Let’s GOOOO,” he remarked. His enterprise, Satlyt, is developing software aimed at transforming existing satellites into data processing networks, akin to the concept of orbital data centers. Recently, he made significant progress towards this goal. After a year of dedicated effort, the company successfully conducted a test earlier this month. Collaborating with the University of Houston on a NASA-funded research initiative, the test demonstrated how the startup’s software can distribute tasks among computers in orbit, based on communication capabilities and available computational resources.
Afullo explained that the objective is to develop software that connects multiple satellites to facilitate the sharing of computing tasks. Users of the system “will only need to define permissions, priorities, and operating limits,” while the software autonomously makes decisions within those parameters, Afullo stated, according to hinews. The software is currently operational on three satellites in orbit, and this test represents a crucial step towards realizing that vision. Although the test utilized NASA-funded research, the agency has yet to select it for any specific mission or to deploy its technology.
Afullo noted that NASA missions encounter similar challenges that Satlyt aims to resolve — “limited computing resources, intermittent communication, and the necessity to relay useful information back to users.” In this test, the company established a simulated network consisting of four computers, four users, and eight services. The software evaluated how to allocate and execute tasks on these computers, taking into account the availability of computational power and communication. Only one physical computer aboard a spacecraft in orbit executed the software, while the other three were simulations, ultimately transmitting the results back to Earth.
The test confirmed that Satlyt’s software can formulate plans on hardware in orbit, but further testing is required to understand how these computers exchange information in practice when multiple physical computers are involved. “The finding is a trade-off,” he remarked. One method proved most effective, although it necessitated greater computing power than the alternatives. Nonetheless, Satlyt, which ranked among the top five in the TechCrunch Disrupt Battle this year, considers this a significant achievement. “It’s a building block toward satellites sharing computing work,” he asserted.
The Satlyt team includes engineer Leina Moli, founder and CEO Rama Afullo, engineer Junn Wangari, and CTO Nelson Psenjen. Image Credits: Satlyt under a Satlyt license. NASA has already established a technology network designed to facilitate data transmission through space, even amidst communication delays. The agency began utilizing this technology earlier this year and has already conducted tests on aircraft. Afullo, who previously held positions at Google and SpaceX, founded Satlyt in 2024 with the vision of transforming spacecraft into virtual data centers. Its software aids operators in running AI and other applications onboard, processing data closer to its source and minimizing reliance on downlinks, where satellites transmit data back to Earth, and ground-based analysis primarily conducted by humans.
For instance, if a spacecraft detects a wildfire on Earth, such software can expedite the transmission of pertinent information. “The goal is to provide useful answers to users more quickly and optimize the use of limited communication windows,” he explained. “This experiment tested one planning component of that broader capability.” Satlyt indicated that it still needs to evaluate its software under various workloads, hardware configurations, and connectivity conditions. They also identified an issue regarding how their software was referencing time delays, Afullo noted — it was recording it as arrival time (the time it arrived) instead of elapsed delay (the duration it took to arrive), which represents a different metric.
However, this does not imply that the experiment was unsuccessful. “The software development, onboard execution, and return of outputs were successful,” he stated. Yet, Afullo acknowledged that considerable work remains. “We also require functional communication links, permissions from participating operators, secure data transfers, resource scheduling, and recovery protocols when a connection between computers becomes unavailable.” They aspire to expand their operations beyond low Earth orbit in the near future.
“Our forthcoming technical steps involve broader validation and integration, followed by demonstrations of computing tasks shared across different spacecraft,” he stated. “In the long run, we aim for developers to create applications for space without needing to reconstruct the infrastructure for each mission. That’s what virtual AI data centers in space signify for us.”
