
Tech & AI
By TrendScope Daily Team · November 1, 2026
For years, the pitch for putting data centers in space sounded like science fiction with a PowerPoint attached: free solar power, free cooling from the vacuum of space, no zoning fights with neighbors. On October 30, a small Bay Area startup called Satlyt took a real, if modest, step toward making it work — running a NASA-funded test that let a constellation of satellites plan and share computing tasks among themselves, entirely in orbit.
Satlyt, founded in 2024 by Rama Afullo, a former Google and SpaceX product manager, doesn’t build its own satellites. Instead, it writes software that turns existing satellites — whatever hardware is already in orbit — into what the company calls “virtual AI data centers in space.” The idea is that a spacecraft spotting a wildfire, a flood, or a ship in distress could process that imagery on the spot and relay only the useful result, rather than beaming raw data back to Earth and waiting for a ground station to crunch it.
What Satlyt Actually Tested
The October test, run in partnership with the University of Houston under a NASA-funded research project, came after 13 months of development. Satlyt simulated a network of four computers, four users, and eight services, with one physical computer aboard a spacecraft carrying out the experiment and relaying the results back to Earth. The company’s software is already running on three satellites currently in orbit, though NASA has not selected Satlyt for any specific mission or committed to deploying the technology more broadly.
The goal was narrower than it might sound: not to prove satellites can do useful AI work yet, but to prove they can autonomously plan who does what. “The software development, onboard execution, and return of outputs succeeded,” Afullo said of the run.
“It’s a building block toward satellites sharing computing work.”
— Rama Afullo, founder and CEO, Satlyt
A Bug, a Trade-Off, and an Early Verdict
Satlyt’s own account of the test is refreshingly unglamorous for a company riding the orbital-computing hype wave. Engineers found that one planning approach performed best overall, but demanded more onboard computing power than the alternatives — what Afullo called “a trade-off” the team will have to weigh as it moves toward real missions. They also caught a bug: the software had been logging the arrival time of a task rather than the actual delay it experienced, a distinction that matters a lot when you’re trying to measure how fast a satellite network can really respond.
Those are the kinds of findings that would barely register in a splashier announcement, but they’re also exactly what a NASA-funded research test is supposed to surface. Afullo was careful not to oversell the result: the test showed the software can generate plans using orbital hardware, he said, but transferring information reliably between multiple separate computers in space still needs further testing before Satlyt can claim satellites are genuinely sharing workloads.
Everyone Wants a Data Center in Orbit
Satlyt is a small player in a field that is suddenly attracting very large checkbooks. The startup closed an $8 million seed round in early October, led by investors betting that AI’s power and cooling bottlenecks on Earth will eventually push some computing into orbit, where solar panels run nearly nonstop and the vacuum of space handles heat dissipation for free. It also placed among the top five companies at this year’s TechCrunch Disrupt Startup Battlefield.
But Satlyt is working at a very different scale than its best-funded rivals. Starcloud, which is building purpose-made satellites stacked with Nvidia GPUs, has raised roughly $170 million for the effort. Alphabet, Google’s parent company, has said it plans to launch prototype AI satellites by 2027 as part of its own orbital-computing research. Against that backdrop, Satlyt’s pitch is less about launching new hardware and more about squeezing coordination and efficiency out of satellites that are already flying — a software-first bet in a race that a lot of rivals are trying to win with hardware and capital instead.
Why It’s Not Just a Space Story
The timing isn’t an accident. Big tech companies are collectively planning to spend hundreds of billions of dollars on AI data centers on the ground this year alone, and the physical limits — land, power grids, water for cooling — are becoming part of the public conversation around AI’s cost. Orbital computing won’t replace a hyperscale data center anytime soon; today’s test used a single physical computer standing in for a whole simulated network. But Satlyt’s NASA-backed work points at a genuinely different use case: processing satellite data, like wildfire or flood imagery, where the advantage isn’t raw compute, it’s not having to wait for a round trip to Earth.
Afullo said Satlyt is already looking past low Earth orbit, using the same NASA-funded research track to explore streaming 4K video from the Moon. The next milestones, he said, are broader validation of the planning software and, eventually, demonstrations of computing work actually shared across multiple separate spacecraft — the step this month’s test was explicitly designed to set up, rather than complete.
For readers who’d rather watch the space race from the couch than fund it, the LEGO NASA Artemis Space Launch System set is a reasonable stand-in for the orbital ambitions startups like Satlyt are chasing, one brick at a time.
As an Amazon Associate, TrendScope Daily earns from qualifying purchases made through links in this article.
Quick questions
What did Satlyt test in space?
NASA-funded software that let a constellation of satellites plan and share computing tasks among themselves in orbit.
Does Satlyt build satellites?
No. It writes software that turns satellites already in orbit into what it calls virtual AI data centres in space.
Why put data centres in space?
Supporters point to constant solar power, natural cooling and no land disputes, and to processing images on board so only useful results are sent down.
Related on TrendScope Daily
- OpenAI’s DevDay 2026: Meet Dots, the “Always-On” Agent Built to Work While You Don’t
- Anthropic Says Claude Now Leads 26% of Its Own R&D — Up From Zero in February
- Answer Engine Optimization, Explained: How AI Chatbots Are Rewriting the Rules of Search
- More Tech & AI stories: browse the Tech & AI section
Bottom Line
Satlyt’s October test didn’t prove satellites can run AI workloads at scale — it proved a piece of planning software can operate on real orbital hardware and come back with usable, if imperfect, results. In a field where Alphabet and well-funded rivals like Starcloud are placing much bigger bets on new hardware, Satlyt is betting that software coordination between satellites already in the sky is the faster, cheaper path to the same destination.