Satlyt Tests Task-Sharing Software Across Orbiting Computers, With Only One Machine in Space

Satlyt has completed a successful test of software that distributes computing tasks among computers on orbiting spacecraft, according to TechCrunch. The test ran in early October 2026 with the University of Houston as part of a NASA-funded research project. It followed 13 months of work toward turning satellites into data processing networks.

The software assigns tasks based on communication between nodes (the individual computers in a network) and the compute capacity available on each machine. In practice, it works like a dispatcher deciding which worker takes which job, based on who is free and who has the right tools.

For the test, Satlyt built a simulated network of four computers, four users and eight services. It used that network to examine how tasks were assigned across the nodes.

The test was only partly hardware-based. One physical computer on a spacecraft in orbit ran Satlyt's software, while the other three computers in the network were simulations. Results were sent back to Earth. As of October 30, 2026, the company's software is on three satellites in orbit.

NASA has not selected Satlyt for any particular mission, and as of the same date it has not chosen to deploy the company's technology.

Rama Afullo founded the company in 2024 with the aim of turning spacecraft into virtual data centers. Afullo previously worked as a product manager at Google and SpaceX. In early October, Satlyt announced an $8 million seed round, an early-stage funding round, led by Non Sibi Ventures. The company said the money would go toward building virtual AI data centers in space, according to its announcement.

Satlyt is not the only company pursuing computing in orbit. In September 2026, Google said it would launch a prototype satellite the following week for the first in-orbit test of Project Suncatcher, an effort to test AI chips in space, as Reuters reported. Google's effort focuses on accelerator hardware (chips built to speed up AI workloads) in orbit, while Satlyt's announced work centers on the scheduling and coordination layer that would sit across multiple machines.

In my view, the gap between this result and an operating orbital data center is still wide. A four-node simulation with one live processor shows that the task allocation logic behaves as designed under the conditions tested. It does not yet show how the scheduler handles link outages, radiation-induced faults (errors caused by charged particles striking electronics), thermal throttling (slowing a chip to shed heat), or inter-satellite link latency under real orbital geometry. Those are the questions that will determine whether the approach survives contact with operations.

The broader context here is cost. Satellites that process data before downlink (sending it to Earth) can, in principle, reduce the volume of raw data sent to ground stations. A distributed scheduler could also let a constellation share work when one node is busy or out of contact. Whether that advantage outweighs the power, cooling and hardware-reliability penalties of running compute in orbit is an open engineering and economic question, and the test does not answer it.

Still, in my view, the direction is a constructive one. Moving from one machine running a workload to a set of machines that coordinate who does what is a step distributed systems engineers have spent decades refining on the ground. If that logic can work on satellites with intermittent links and tight power budgets, a constellation could do more of its own processing before data reaches Earth. The coming test campaigns, and whether NASA or another customer commits to a mission, will show how far the idea can go.