Satlyt Raises $8M to Put AI Compute on Satellites

A satellite can collect more data than its operator can economically move to Earth. The expensive part is not always the sensor. It is deciding which bytes deserve the ride home.

Satlyt has raised an $8M Seed round led by non sibi ventures to move more of that decision onto the spacecraft. The Sunnyvale- and Nairobi-based company builds software that runs AI and other workloads on computers already aboard satellites, reducing dependence on constrained downlink windows and ground processing.

The financing includes TLCOM, Antler, Slauson & Co., Launch Africa Ventures, Enza Capital, Askya Investment Partners, Demos Capital, BAG Collective, Gaingels, AXIAN Investment, and other backers. Satlyt says the capital will expand its engineering and customer-delivery teams, support deployments across more spacecraft, and advance its plan for virtual AI data centers in space. The larger bet is an infrastructure bet: can a neutral software layer make many companies’ satellites behave like programmable nodes in one orbital computing network?

What Satlyt’s $8M Seed Funds

Founder and CEO Rama Afullo started Satlyt after roles spanning Tesla, Google’s cloud business, and SpaceX’s Starlink organization. TechCrunch reported that Afullo had tried to advance the idea inside both Google and SpaceX before leaving to build it independently. CTO Nelson Kigen Psenjen brings the distributed-computing expertise behind the software layer.

Satlyt is not building spacecraft. Its product is designed to run on hardware owned by satellite operators, including systems already in orbit that can accept software updates. That distinction matters to the economics. Building an orbital data center requires launch capacity, power, thermal management, specialized hardware, and ownership of the spacecraft. Satlyt is pursuing a software-first wedge that begins with computing capacity operators already paid to place in orbit.

The $8M round gives Satlyt more room to turn demonstration work into repeatable customer delivery. It also raises the standard of proof. A successful workload on one spacecraft shows that onboard processing is possible. A commercial platform must install reliably across different hardware, protect command authority, support customer applications, and produce results that are worth more than the bandwidth and engineering effort they save.

Why Onboard AI Matters

Satellite operators work under constraints that terrestrial cloud teams rarely have to negotiate. Power is limited. Thermal headroom is physical. Communication windows can be brief. Downlink capacity is expensive, and raw imagery, telemetry, logs, and sensor readings can exceed what an operator wants to transmit.

A Google DeepMind case study offers Satlyt’s clearest technical evidence so far. Satlyt deployed a quantized Gemma 3 1B model onboard a satellite to analyze system logs, software errors, and stack traces produced by image-processing workloads. In two representative fault-injection tests, the model reduced diagnostic payloads from 1,319 bytes to 469 bytes and from 1,318 bytes to 464 bytes, reductions of 64.4% and 64.8%. It also produced a root-cause diagnosis and recommended action.

Those are focused test results, not proof of an autonomous orbital cloud. They still show the practical value of local inference. A spacecraft that can turn raw diagnostic context into a compact, useful explanation can reduce downlink demand and shorten the path from an anomaly to a ground operator’s decision. Satlyt is also evaluating smaller and more efficient model configurations because every watt and gigabyte of memory used for AI competes with the spacecraft’s primary mission.

From One Spacecraft to Shared Orbital Compute

Satlyt’s ambition extends beyond local processing. The company wants spacecraft from different operators to share compute, storage, and routing through a common software layer. That would turn isolated machines into something closer to a distributed cloud, without requiring Satlyt to own the satellites.

The company’s latest deployment rode on a TakeMe2Space spacecraft carried by SpaceX’s Transporter-18 mission. The Transporter-18 launch succeeded on October 1, carrying 130 payloads to low Earth orbit. Satlyt said its deployment supports research software connected to NASA-funded work with the University of Houston and a commercial imagery-processing application. Launch success establishes that the hardware reached orbit. The applications still have to run successfully and generate evidence from the orbital environment.

That sequence is easy to blur in space technology. A launch is a milestone, not the end of verification. Satlyt’s next major technical step is to coordinate computing across two separate satellites. If that works across different operators and hardware, the company will have moved from onboard software toward the shared infrastructure described in its pitch.

The Commercial Wedge Is Idle Compute

The commercial proposition is simpler than the orbital language around it. Satellite operators have already paid to launch computers. Satlyt wants to help them process more data locally, improve spacecraft monitoring, and eventually sell unused computing capacity. Software companies could gain a route to orbital customers without becoming spacecraft manufacturers.

Investor TLCOM describes the market as fragmented across closed operator systems, with much of the available onboard compute sitting idle. The firm says Satlyt has signed more than 30 operator memoranda of understanding, secured NASA STTR work, and deployed on several third-party satellites. Those are investor-reported traction signals rather than audited operating metrics, but they explain why a software layer could attract capital before multi-satellite orchestration is fully demonstrated.

Satlyt also carries a geographic advantage that fits the product thesis. Its headquarters span Sunnyvale and Nairobi, its engineering team has a substantial African presence, and the company has relationships with the Kenya Space Agency and Angola’s GGPEN. A neutral layer that works across operators and jurisdictions could matter to governments and commercial customers that do not want one vertically integrated provider controlling hardware, compute, and data movement.

What the Round Still Has to Prove

Satlyt now has to turn technical possibility into dependable infrastructure. The company must support different spacecraft architectures, survive long operating cycles, secure remote software delivery, and show that local AI improves mission economics without compromising control or reliability. It must also persuade operators that sharing compute across organizational boundaries is useful enough to justify new standards and trust.

The timing is unusually visible. SpaceX, Google, and startups are all testing versions of orbital computing, while satellite operators continue to deal with the immediate cost and delay of moving data to Earth. Satlyt does not need the grandest version of the orbital data-center market to arrive first. It can create value by making existing spacecraft more productive today.

The $8M Seed finances that narrower path toward a much larger network. Satlyt’s opportunity lives in the handoff between machines that collect data and people who need answers. Each successful deployment gives the software more hardware to coordinate, more workloads to learn from, and another operator deciding whether the valuable result should come home instead of every raw byte.