Belgian company EDGX’s two Sterna computing payloads have been in orbit since March. In a 28 October update, it said one returned its first operational data within two weeks of launch; the other was still awaiting satellite commissioning. Two further missions are booked for 2027.[1]
ESA’s original launch report describes a hosted experiment: an off-the-shelf AI processor travelling aboard a satellite alongside other payloads. One planned test was locating sources of interference in radio communications. The purpose was to measure computing performance and power consumption in space.[3]
The computer has a separate supervisor
The architecture documented in ESA’s ASCEND project record divides the machine into two domains. The NVIDIA Jetson processor runs Linux and containerised applications. A separate, radiation-tolerant microcontroller supervises it, managing power, monitoring health and detecting, isolating and recovering from faults.[4]
That supervisor also supports recovery through alternative boot partitions and a known-good fallback software image. The design gives the computer a route back to a usable state when its normal software environment needs recovery. The AI workload and the machinery watching over it have different jobs.[4]
Sterna’s product specification adds software-controlled power modes from 10 to 45 watts. That gives mission designers a way to fit computing demand to the spacecraft’s available power, while retaining the familiar NVIDIA software tools used to develop applications on the ground.[5]
A software boost still needs a power budget
The next Sterna mission is scheduled for February 2027 on Gracious Gloria. EDGX’s roadmap calls for twelve hours of continuous operation with customer software, using the Jetson Orin NX in Super Mode.[2]
NVIDIA’s January 2025 technical explanation shows what that mode changes. JetPack 6.2 unlocks higher processor frequencies and adds a 40-watt reference mode for Orin NX. The uncapped MAXN SUPER setting can throttle back when it exceeds the thermal design power budget. NVIDIA advises matching the power profile and thermal design to the application.[6]
A software upgrade can therefore expose more performance without making the physical requirements disappear. How fast a customer’s model can run still depends on the power and heat the complete payload can accommodate.[6][5]
EDGX’s October 2027 mission is planned to combine a hyperspectral camera, Sterna, the newer Thor-based Morus computer and a relay link. Alongside selling computers, the company is developing a service for customers to upload software and rent computing time on its missions. The next flights move that proposition from supplying a processor toward operating somebody else’s application in orbit.[2][1]
Sources & further reading
- EDGX launches AI computing platform in orbit on SpaceX Transporter-16 using NVIDIA Accelerated Computing EDGX · 28 Oct 2026 · Accessed 28 Oct 2026
- EDGX: missions and roadmap EDGX · Accessed 28 Oct 2026
- Off-the-shelf AI-powered data processing unit successfully launches aboard 16U satellite, with ESA support European Space Agency · 31 Mar 2026 · Accessed 28 Oct 2026
- Neuromorphic AI Onboard: ASCEND project record European Space Agency · Accessed 28 Oct 2026
- Sterna: space-qualified AI edge computer for satellites EDGX · Accessed 28 Oct 2026
- NVIDIA JetPack 6.2 Brings Super Mode to NVIDIA Jetson Orin Nano and Jetson Orin NX Modules NVIDIA · 16 Jan 2025 · Accessed 28 Oct 2026