Satlyt raises $8 million to put AI software on satellites

A software layer for orbit

Afullo told TechCrunch he had pitched the idea at both Google and SpaceX and was turned down. He left SpaceX and founded Satlyt with partners to build software that lets satellites work together on larger computing tasks.

The distinction matters as SpaceX develops orbital data centers and Google prepares to fly the first prototype of Project Suncatcher. Satlyt’s pitch is not to own the hardware, but to make software work across satellites built by different companies. Afullo compares the company’s role to VMware and Snowflake: platforms that run complex applications without requiring customers to manage the underlying infrastructure.

He describes the ambition as an open ecosystem: if SpaceX and other orbital-data-center builders are the iPhone, Satlyt wants to be Android.

What the software has done

Satlyt’s near-term case is less about powerful GPUs in orbit than about making existing satellites more efficient. When a spacecraft encounters a problem, it often relies on ground operators. Sending data back to Earth—a downlink—can be expensive and slow. Onboard AI could help satellites make decisions themselves, including analyzing anomalies and processing sensor readings.

Earlier this year, Satlyt ran Google DeepMind’s Gemma model on a satellite operated by Momentus. The model cut the volume of data sent about onboard software errors by more than 60%. Afullo said that could save hundreds of thousands of dollars per satellite each year.

This week’s mission will run Satlyt’s software on a spacecraft built by TakeMe2Space. Three customers are involved:

There are few powerful GPUs in orbit, so Satlyt is concentrating on data processing and spacecraft efficiency for now. Its next project is a cloud-computing system distributed across two different satellites, which it plans to test next year. If successful, that could let Satlyt offer spacecraft makers an outsourced orbital cloud.

The bet beyond data centers

The $8 million seed round was led by Houston-based Non Sibi Ventures. Bernard Harris, a fund partner and former NASA astronaut who worked at the agency for more than 20 years, played a role in the investment decision.

Non Sibi’s rationale is that Satlyt does not need the largest orbital data centers to succeed. The economics of those systems remain uncertain, and launch capacity could be a constraint. Non Sibi partner Kent Lucas said Satlyt only needs more satellites to go into orbit.

I think that makes the software-first strategy more credible than a direct bet on giant space-based computing infrastructure—but it also leaves a harder question: how many satellite operators will want a shared platform, and what will persuade them to install it? Afullo hopes Satlyt’s software will be on 20% of satellites by the end of the decade. He expects nearly every spacecraft manufacturer to include GPUs and other advanced processors by then, and argues that launching a satellite without a GPU is already a missed opportunity.

That ambition depends on hardware adoption Satlyt does not control. Its opportunity is to make onboard computing useful before orbital data centers become routine; its risk is that satellite owners may not need a common software layer as quickly as the company expects.