Satlyt's Orbital Computing Test Ran One Real Computer, and Hit a Timing Bug

Satlyt, a Sunnyvale and Nairobi startup that wants to turn other companies' satellites into a shared computing network, ran a NASA-funded test earlier in October in which a single physical computer aboard a spacecraft carried the experiment, according to TechCrunch's October 30 report. The other computers in the scenario were simulated.

What the test was

The work was done with the University of Houston after 13 months of development. The software divides tasks among computers according to the communication links between them and the compute each has free. The simulated network had four computers, four users and eight services. The real hardware planned the work in orbit and sent results to Earth.

CEO Rama Afullo, who previously worked at Google and SpaceX and founded the company in 2024, said: "The software development, onboard execution, and return of outputs succeeded." He also said "The finding is a trade-off": one planning approach performed best but needed more computing power than the others.

Element	Real or simulated
Computers in the scenario	4, simulated
Physical computers in orbit	1
Users and services	4 users, 8 services, simulated
Links between separate spacecraft	Not yet tested

The bug and the gaps

TechCrunch reports a timing bug: the software stamped the arrival time of data rather than the elapsed delay. The software still needs testing under different workloads, hardware and connectivity conditions. Untested pieces include how several physical computers pass information to one another, secure transfers, resource scheduling and recovery when links drop.

NASA has not selected Satlyt for any mission or chosen it to deploy the technology, TechCrunch notes. The agency runs a separate data-transfer network for delayed links, which it began using earlier in 2026.

The company behind it

Payload reported on October 1 that Satlyt raised an $8 million seed round led by non sibi ventures, with TLCOM, Antler, Slauson & Co., Launch Africa Ventures and others participating. Afullo said total funding is "well past $10M". The company has a NASA STTR contract worth $4 million for delay-tolerant edge protocols.

Its model is a virtual orbital data center. It does not build compute hardware; it adds AI inference to partner satellites, either before launch or by uploading software in flight, so that only essential outputs are sent to the ground. According to TechCrunch, the software is already on three satellites in orbit. Payload reports a May demonstration of inference in orbit using Google's Gemma model.

The claim to hold it against

Afullo's goal, per Payload, is to onboard 30% of all non-Starlink satellites by the end of the decade. Three satellites carry the software today. Closing that gap depends on operators granting permission and on the untested links between spacecraft, which is the same step this experiment skipped.

The pitch also sits against a crowded field of terrestrial compute. Runware's modular pods and Intel's Classiq purchase show where other infrastructure money is going on the ground and in new silicon.

What would change this read

The next checkable milestone, in TechCrunch's account, is demonstrations of computing work shared across separate spacecraft, after broader validation. If that run shows two real machines exchanging tasks over a real link with the timing bug fixed, the network claim has evidence. Until then the result is one computer, one orbit and a trade-off between planning quality and power.