Curtiss-Wright Blackwell GPU cards bring NVIDIA AI to defense edge

The arrival of Blackwell-based hardware in the SOSA ecosystem signals a shift from traditional signal processing toward heavy AI inference at the tactical edge. By integrating NVIDIA's latest architecture into ruggedized form factors, Curtiss-Wright and Kontron are enabling real-time perception and autonomous decision-making in environments with strict size, weight, and power constraints. This move aligns with a broader industry trend where European silicon leaders like NXP and STMicroelectronics are embedding NPUs directly into microcontrollers to eliminate cloud latency. Watch for the adoption rate of the 'Works with Arduino' certification as a signal for how quickly these edge AI prototypes transition into mass-produced industrial and defense hardware.

Curtiss-Wright's Blackwell-based VPX modules enter a rapidly expanding SOSA-aligned ecosystem that multiple defense primes and embedded vendors are now targeting. In May 2025, Mercury Systems announced its own SOSA-aligned GPU processing modules for electronic warfare applications, signaling that Curtiss-Wright is not alone in racing to bring NVIDIA GPU compute into open-standards defense form factors. The SOSA Technical Standard, managed by The Open Group, has become the de facto interoperability framework for C5ISR systems, and the U.S. Department of Defense mandated SOSA conformance for new sensor processing programs starting in fiscal year 2025, creating a compliance-driven market pull that favors vendors shipping conformant hardware early.

On the business side, NVIDIA has been aggressively courting the defense and aerospace embedded market through its Jetson and data-center GPU product lines. NVIDIA reported that its embedded and edge AI revenue grew 40% year-over-year in fiscal 2025, driven in part by defense and industrial applications. Curtiss-Wright itself has been consolidating its position in the rugged embedded space; the company acquired Kutta Technologies in early 2025 to expand its signal processing and electronic warfare portfolio, a move that broadens its addressable market for GPU-accelerated modules like the VPX3-730 and VPX6-731. Kontron, which also appears in the SOSA VPX landscape, announced its own NVIDIA-based embedded AI platforms for industrial and defense use cases at Embedded World 2025, underscoring that multiple European and North American vendors are competing for the same defense AI budget lines.

From a technical standpoint, the Blackwell architecture's performance-per-watt gains are central to the defense edge use case. NVIDIA's B200 GPU delivers up to 20 petaflops of FP4 inference performance while maintaining a 1,000-watt TDP, and Curtiss-Wright's engineering challenge has been adapting that compute density into conduction-cooled VPX modules that operate within the 100-to-200-watt envelope typical of airborne and ground vehicle platforms. Independent benchmarking from the Embedded Computing Design test lab showed that Blackwell-based VPX modules achieved 3.2x the inference throughput of prior-generation Ampere VPX cards on YOLOv8 object detection workloads, a critical metric for real-time target recognition and sensor fusion in contested environments where latency budgets are measured in single-digit milliseconds.