On March 26, 2026, Syntiant Corp. and Novi Space announced the successful in-orbit demonstration of real-time AI object detection on a commercial satellite. The collaboration validated that ultra-low-power, quantized neural networks can perform high-accuracy computer vision tasks in the extreme resource constraints of Low Earth Orbit (LEO), where both onboard compute power and communications bandwidth are severely limited.
By processing imagery directly on the satellite, the system identifies targets such as ground vehicles and ships in real time. This “edge AI” approach eliminates the need to downlink massive volumes of raw data to Earth, significantly reducing latency and operational costs while providing actionable intelligence faster than traditional ground-based processing methods.
The in-orbit demonstration represents a critical shift in how space agencies and commercial operators manage data. By moving the “brain” of the satellite from the ground to the spacecraft itself, the partnership addressed two of the most significant barriers in modern space operations: limited communication bandwidth and the “data graveyard” problem, where 90% of captured satellite imagery is never analyzed due to high downlink costs.
Technical Breakthrough: The SP240 Space-Edge Computer
The demonstration was executed on Novi Space’s SP240 onboard computer, which features an AMD Versal adaptive SoC powered by a dual-core ARM Cortex-A72 processor. This flight-qualified hardware utilized Syntiant’s proprietary development tools to deploy a suite of quantized vision models optimized for minimal memory and power consumption.
A key differentiator of the test was the ability to retrain and redeploy AI models in less than 24 hours. This rapid-cycle capability allows satellite operators to adapt to evolving mission needs—such as switching from maritime ship detection to terrestrial vehicle tracking—without ending the mission or requiring hardware modifications.
Details of the In-Orbit Demo
The demonstration utilized a commercial satellite equipped with the Novi Space SP240 space-edge computer. This hardware is specifically designed for the harsh environment of Low Earth Orbit (LEO), featuring radiation-hardened components that prevent the “bit flips” and hardware failures typically caused by cosmic rays.
- Real-Time Inference: Syntiant’s quantized neural network (QNN) models performed autonomous object detection, identifying ground vehicles and maritime ships directly from the raw sensor feed.
- Edge Processing vs. Traditional Downlink: Instead of sending a high-resolution image (which could be several hundred megabytes), the satellite sent only the “metadata”—essentially a small text packet confirming the object’s coordinates and type.
- Model Agility: A core achievement was the ability to retrain and “hot-swap” AI models in under 24 hours. This allows a satellite to switch its mission—for example, shifting from wildfire detection to urban traffic monitoring—without needing to be decommissioned or physically altered.
Strategic Context and ‘GENIE’ Platform
Novi Space is positioning this capability as part of its open-access GENIE platform, which aims to transform satellites from simple “cameras in the sky” into intelligent, autonomous systems. By reducing the “needle in the haystack” problem of satellite data—where only a small fraction of captured imagery contains useful information—Novi and Syntiant are enabling a shift from simple Earth Observation to true Geo-Intelligence.
Syntiant, which has deployed over 100 million physical AI solutions globally, is leveraging this milestone to extend its “Edge AI” footprint into the burgeoning orbital data center market. The partnership highlights a trend toward decentralized space architectures where intelligence lives exactly where the data is generated.
Executive Perspective
“Our collaboration with Novi shows that advanced AI can operate in the highly constrained environment of space,” said Ethan Wais, GM of Federal at Syntiant. “Using our development tools, we trained and optimized multiple computer vision models and deployed them to run efficiently onboard a commercial satellite, despite the limited bandwidth in LEO.”
“By running Syntiant’s optimized AI models onboard our satellite systems, we significantly reduce bandwidth requirements while enabling faster, more informed decision-making for space-based missions,” said Michael Bartholomeusz, CEO at Novi Space. “Just as importantly, the ability to retrain and redeploy these models in less than 24 hours allows satellites to quickly adapt to evolving mission needs.”
Proliferated Edge Compute
The success of this demonstration paves the way for the fractional leasing of space-based infrastructure. Novi Space plans to expand its GENIE constellation to enable daily revisit rates anywhere on Earth by the end of 2028, offering developers and government agencies the ability to deploy their own AI-driven applications directly onto existing orbital compute layers.
