Satlyt Lands $8M Seed Round to Bring AI to Satellites

Satlyt, a startup co-founded by Rama Afullo—formerly a product manager at Google and SpaceX—is putting in motion a plan to run artificial intelligence directly on satellites. Fresh off an $8 million seed round, the company aims to shift more computation from Earth to space, enabling satellites to handle complex AI workloads rather than relying entirely on ground stations for data processing.

Headquartered in Sunnyvale, California, and Nairobi, Kenya, Satlyt was born from Afullo’s conviction that orbital data centers are more than just theoretical. His stints in Google’s cloud division and at Starlink under SpaceX convinced him that the time was right to build software that lets multiple satellites collaborate on compute-intensive tasks. Satlyt doesn’t plan to launch its own hardware; instead, it builds the software infrastructure that runs AI models on existing spacecraft.

What Satlyt’s Software Does—And Has Done

Satlyt enables satellites to make onboard decisions typically handled on Earth—such as detecting anomalies or processing sensor data—cutting down on the slow and costly process of downlinking raw data. Earlier this year, Satlyt partnered with DeepMind to deploy the Gemma AI model on a Momentus-operated satellite. That model reduced data transmission tied to onboard software errors by more than 60%, potentially saving satellites hundreds of thousands of dollars per year in operational costs.

This week, Satlyt’s software is being launched aboard a satellite built by TakeMe2Space, an Indian hardware provider. The mission includes clients such as NASA, which is testing space-based cloud computing protocols; Stellerian, which handles space surveillance and imagery processing; and TakeMe2Space itself. The aim is to demonstrate that third-party software can be hosted in space alongside partner hardware.

Funding and Ambitions

The $8 million seed funding was led by Non Sibi Ventures, a firm with roots in space exploration. One of its partners is Bernard Harris, a former NASA astronaut, who sees value in Satlyt’s software-first approach that doesn’t depend on each satellite builder owning physical infrastructure. Another investor noted that Satlyt’s growth prospects rest more on increasing satellite launches than on high-risk, high-capital data center projects in orbit.

Satlyt has loftier goals in its roadmap. Over the next year, the company plans to build a distributed cloud computing layer spanning multiple satellites—across companies—to form a unified compute network in space. By decade’s end, Afullo hopes Satlyt’s software will be running on roughly 20% of new satellites, assuming the trend of integrating GPUs or equivalent compute on spacecraft continues.

Today, there are few high-powered GPUs operating in orbit. But Satlyt believes there’s a substantial opportunity in the more immediate “mid-tier” AI workloads it’s targeting now—sensor analysis, anomaly detection, and other inference tasks that don’t require full-scale data centers but still confer major savings in bandwidth, latency, and operational cost.

ANALYSIS: The vision Satlyt presents taps into a broader push toward decentralizing cloud compute and pushing it closer to the edge—or in this case, the sky. As satellites proliferate, latency and downlink costs become increasingly painful, especially for Earth observation, surveillance, and Internet-of-Things services. If Satlyt can pull off a multi-satellite cloud and demonstrate consistent savings, its software model could become foundational for a new layer of infrastructure. What to watch: partnerships with satellite OEMs, regulatory constraints for onboard AI, and whether adoption holds in orbit at scale.