New Ruggedized Edge AI Computers Emerge

The push for Modular Open Systems Approach (MOSA) is a key driver behind this new hardware. Mandated by the U.S. Department of Defense, MOSA requires modular designs with open standards to reduce costs, accelerate upgrades, and increase competition among vendors. This prevents vendor lock-in and allows for rapid insertion of new technologies from a wider range of specialized companies. Standards like the Sensor Open Systems Architecture (SOSA) and others under the MOSA umbrella define the specific hardware and software interfaces. These standards, often based on OpenVPX, specify details from card pinouts to data protocols, ensuring interoperability between modules from different manufacturers. This allows for quicker, more affordable integration of capabilities for C5ISR (Command, Control, Communications, Computers, Cyber, Intelligence, Surveillance, and Reconnaissance) systems. The Kite-Strike II mission computer exemplifies this trend, integrating an NVIDIA Jetson AGX Orin system-on-module (SOM) for up to 275 TOPS of AI performance. This GPGPU-based system, with its Ampere architecture featuring up to 2048 CUDA cores and 64 Tensor cores, is designed for compute-intensive tasks like sensor fusion and computer vision on platforms where space, weight, and power (SWaP) are limited. It is ruggedized to MIL-STD-810H, MIL-STD-461G, and MIL-STD-1275E standards. In contrast, Klepsydra's geolocation solution leverages a Field-Programmable Gate Array (FPGA) based hardware platform, specifically an AMD Zynq UltraScale+ MPSoC. This approach offloads key parts of the AI inference pipeline to programmable logic, minimizing CPU load and ensuring deterministic, real-time performance. The solution runs the KamNet convolutional neural network to determine a satellite's position from live images, a critical capability in GPS-denied environments. This hardware evolution introduces new challenges for software certification. Integrating AI/ML components into systems requiring DO-178C compliance is not straightforward, as the standard was not designed for non-deterministic algorithms. Industry and regulatory bodies like the FAA are actively developing guidance and adapting standards to ensure the safety and reliability of AI in safety-critical aerospace applications. The core business and technical strategy is to create a more competitive and agile defense acquisition lifecycle. By breaking down large, monolithic systems into severable modules, the DoD can acquire components from independent vendors, fostering innovation and allowing for incremental capability upgrades without replacing the entire system. This approach is critical for keeping pace with rapidly evolving threats and technologies.