A startup wants satellites to think for themselves instead of waiting for instructions from Earth

Satlyt develops software that lets AI models run directly on satellites while they're in orbit, rather than requiring every decision or anomaly to be transmitted down to ground control for a human response. The company doesn't build its own spacecraft at all.

Why it matters: that's a meaningful distinction in the space industry right now. Instead of competing to launch hardware, Satlyt is betting the real opportunity is in the software layer that makes existing and future satellites smarter, regardless of who built them.

Satlyt closed an $8 million seed round led by Houston-based Non Sibi Ventures, with partner Bernard Harris, a former NASA astronaut, backing the investment.

Why it matters: having a former astronaut's firm lead the round adds a layer of credibility that goes beyond typical venture capital interest, someone who has direct, firsthand experience with exactly the operational gap this company is trying to close.

Background: sending data from a satellite down to mission control, known as downlink, is slow and expensive. Every piece of information a satellite needs a human to review has to travel through limited communication windows and bandwidth.

In real life it's like a factory worker who has to mail a letter to headquarters and wait for a reply before fixing even a minor glitch on the line, instead of just handling it on the spot.

Why it matters: letting AI make certain decisions onboard reduces how much data needs transmission in the first place, which Satlyt says can save hundreds of thousands of dollars per satellite annually while also speeding up how fast problems actually get addressed.

Satlyt has already run a real demonstration. In an earlier mission, the company deployed Google DeepMind's Gemma AI model on a Momentus spacecraft, cutting the data needed for software error reporting by more than 60%.

Why it matters: that's a concrete, measurable result from an actual spacecraft, not just a projected estimate. For a seed-stage company, having working flight heritage behind the pitch is a meaningfully stronger position than a purely theoretical product.

Satlyt was founded by Rama Afullo, a Kenyan American engineer who previously worked at Google's cloud computing division and briefly at SpaceX's Starlink division in 2024. He reportedly pitched orbital data center concepts internally at both companies before deciding to build the idea independently.

Why it matters: having direct insider experience at both a major cloud computing company and a leading satellite operator gives Afullo a specific vantage point most space startup founders don't have, understanding the problem from both the computing side and the orbital hardware side at once.

Afullo describes the long-term goal as becoming "the Android of orbital computing," an open platform usable across many different satellite operators, in contrast to a closed, single-company approach like SpaceX's own tightly integrated satellite systems.

Why it matters: that framing positions Satlyt directly against the industry's dominant player without competing on hardware at all. If the comparison holds, Satlyt is betting its growth on becoming the common software layer underneath a fragmented market of satellite operators, rather than trying to out-build SpaceX directly.

Satlyt is headquartered in Sunnyvale, California, with an additional base in Nairobi, Kenya.

Why it matters: that dual-location structure reflects both Silicon Valley's space-tech investment ecosystem and the founder's own background, and it's a reminder that space startups increasingly draw talent and operations from well beyond the traditional US aerospace hubs.

Background: several major players have been pushing in a similar direction recently. Starcloud launched an Nvidia H100 GPU to orbit in late 2025 and is already running AI inference on satellite imagery. Google and Planet announced a joint program called Suncatcher to fly Google's own AI chips on Planet's satellites. Loft Orbital has been operating shared satellite infrastructure where multiple customers run AI models in orbit at once.

Why it matters: that broader momentum suggests Satlyt isn't chasing a niche idea nobody else sees value in. Major cloud and satellite companies investing in the same general direction, onboard computing and AI inference in orbit, signals genuine industry conviction that this shift is coming, not just one startup's speculative bet.

Background: industry analysts have pointed to specific technical hurdles before orbital computing scales up significantly: custom silicon built for the radiation and power constraints of space, modular hardware that can be swapped or upgraded after launch, and launch costs low enough to make the economics work at a larger scale.

Why it matters: those constraints explain why Satlyt's current pitch focuses on smaller, sharper use cases, like reducing downlink data for error reporting, rather than claiming it can run full data centers in orbit today. A defined, provable return on investment for a narrow task is a more achievable near-term goal than a wholesale reinvention of how satellites compute.

Background: the push toward orbital computing isn't limited to US companies. A Chinese initiative called the Three-Body Computing Constellation launched its first twelve satellites in May 2025 and has since demonstrated distributed computing and inter-satellite networking, including deploying an AI model with 8 billion parameters directly in orbit.

Why it matters: that international competition adds urgency to the broader race. If multiple countries and companies are independently converging on the idea that satellites should compute and reason onboard rather than relying purely on ground control, it strengthens the case that this is a genuine technological shift rather than a passing trend.

• This won't change anything you interact with directly today, but it's part of a broader shift toward satellites that react faster to the problems they're built to monitor, from weather to communications.
• Watch for more "orbital software" startups positioning themselves as infrastructure, not hardware. Satlyt's VMware and Snowflake comparisons signal where investors increasingly see opportunity in the space industry.
• A working flight demonstration, not just a pitch deck, is becoming the bar for credibility in space-tech funding rounds, and Satlyt clearing that bar at the seed stage is notable.
• If this approach succeeds, expect satellites broadly to become more autonomous and less dependent on constant ground communication, a shift relevant to anyone relying on satellite-based services, from weather forecasting to connectivity.