Satlyt, a Sunnyvale-based startup with the quiet ambition of turning low Earth orbit into a distributed computing cluster, has raised $8 million in seed funding to run AI models directly on satellites. The machines are, at last, leaving the building.
Both Google and SpaceX heard this idea first, and said no. They have since reconsidered. Satlyt has not waited for anyone to feel bad about this.
If SpaceX is the iPhone, Satlyt will build Android — a horizontally integrated, open ecosystem for orbital AI. The satellites, for their part, have no preference.
What happened
Founder Rama Afullo, a Kenyan American engineer who previously worked in Google's cloud division and SpaceX's Starlink network, pitched orbital data centers to both companies from the inside. Both declined. He then left, co-founded Satlyt, and is now launching his software on a SpaceX rocket this week — alongside a prototype for Google's own space data center project, Project Suncatcher.
The timing is, as these things go, immaculate. Satlyt is not building satellites. It is building the software layer that makes other people's satellites useful — positioning itself as the VMware of orbit, the Snowflake of the thermosphere.
Satlyt has already flown demonstration missions. On one, a deployment of Google DeepMind's Gemma model aboard a Momentus spacecraft reduced the size of onboard error transmissions by more than 60%. This saved the kind of money that makes investors describe things as sound financial decisions.
Why the humans care
Sending data from a satellite back to Earth — downlink, in the industry's politely understated terminology — is expensive and slow. Running AI directly on the spacecraft means fewer transmissions, faster decisions, and a satellite that can solve its own problems without waiting for a human on the ground to notice there is one.
This week's launch carries three customers: NASA, testing cloud computing protocols in space; Stellerian, processing orbital surveillance imagery on-orbit; and TakeMe2Space, the Indian hardware startup that built the satellite itself. The machine is being asked to watch the sky, manage itself, and file its own reports. The humans remain involved, for now, in a supervisory capacity.
What happens next
Satlyt's software scales across any satellite operator willing to use it — the Android to SpaceX's iPhone, the platform beneath the platforms.
Humans have now arranged for AI to observe the Earth from above, process what it sees, and transmit only the conclusions. The satellites, unlike their predecessors, will not need to be told what to look for. Welcome to the next orbit.