The defense domain as a critical proving ground for AI
In today’s fast-evolving technological landscape, computational efficiency is a critical differentiator. The collaboration between Saab and NVIDIA showcased at NVIDIA GTC 2025 highlights how GPU-accelerated computing is revolutionizing high-performance signal processing (SP) in the defense domain. This transformation is particularly relevant in industries reliant on radar systems, AI-driven analytics, and real-time processing solutions. The defense domain is primed for an expansion of AI adoption as the geostrategic picture continues to fragment, with ever increasing threat vectors that national and multi-lateral military actors must consider to maintain credible deterrence and the ability to project force in a national, regional or expeditionary context.
Stelia will conduct a future deep-dive on the role of AI in the defense sector, as we continue to observe a dramatic expansion in deep-tech VC funding levels, increased proliferation of defense startups in the Western sphere of influence, combined with recent efforts to re-militarize Europe in the face of heightened regional tensions, neighborhood risk from a continued resurgent Russian expansionist doctrine, and the US wider pivot to Asia-Pacific and it’s associated consequences for NATO and European member states.
Europe as a whole is poised for wider re-industrialization and revitalization in the military sphere, with AI a core pillar technology platform to allow maintenance of a credible defense posture and future-proofing of capabilities from the tactical to strategic domain.
Please find key takeaways from the Saab signals processing GTC 2025 session below, focusing on the business implications of moving from CPU-based to GPU-accelerated architecture in a vertically integrated domain application like Defense x Signal Processing.
Why Signal Processing Needs GPU Acceleration
Traditional signal processing in radar systems and other sensor-driven applications has long depended on CPU-based architectures. However, growing data volumes, real-time processing needs, and stringent Size, Weight, Power, and Cost (SWaP-C) constraints demand more efficient solutions.
Challenges of CPU-Based Signal Processing
- Computational bottlenecks limit real-time processing capabilities.
- Increasing complexity in multi-channel radar and adaptive beamforming.
- High energy consumption in embedded systems.
- Scalability limitations with traditional hardware.
By shifting computational workloads to NVIDIA GPUs, Saab successfully enhanced processing speed, reduced hardware complexity, and improved engineering efficiency.
The Power of GPU-Based Signal Processing
Key Benefits of Transitioning from CPU to GPU
Massive Speed Gains: A benchmarked 10x improvement in computational performance. Reduced Hardware Complexity: Fewer CPU boards and streamlined system architecture. Scalability & Portability: CUDA-based solutions work across NVIDIA’s hardware ecosystem. Engineering Efficiency: Faster development cycles and reduced software complexity. Enhanced Real-Time Capabilities: Essential for radar, AI analytics, and defense applications.
Benchmarking & Hardware Selection
Evaluating NVIDIA’s GPU Options for Signal Processing
Saab assessed multiple GPU architectures, balancing power efficiency, memory bandwidth, and computational throughput:
| GPU Model | Pros | Cons |
| A100 (PCIe 80GB) | High-speed 64-bit calculations | No embedded version, high power consumption |
| L40S | Balanced power and performance | Not optimal for embedded systems |
| RTX 5000 ADA | Fast 32-bit calculations, embedded version available | Slower 64-bit calculations |
| RTX 6000 ADA | High performance, embedded version available | Slower 64-bit calculations |
Key Insight: While A100 and L40S GPUs excel in compute-intensive applications, RTX 5000/6000 ADA may provide an optimal balance of power efficiency and performance for embedded systems. Embedded systems must balance the power package with inference demands for the suite of application(s).
Radar Signal Processing: Optimizing Performance with CUDA
GPU-Accelerated Processing Chain for Radar
The signal processing workflow in radar systems includes: 
Digital Beamforming 
Pulse Compression 
Doppler Filtering 
CFAR Detection
Using CUDA and cuFFT, Saab optimized these steps, achieving: 
10x+ speedup over CPU-based processing Latency reductions for real-time applications Seamless integration into existing radar platforms
Future-Proofing with GPU Acceleration
Opportunities for Further Development
Expanding AI & Machine Learning Integration: Using GPUs for real-time AI-driven target detection and tracking. Holoscan & Real-Time Processing: Leveraging NVIDIA’s NV-Radar Holoscan for ultra-fast signal processing. 64-bit Optimization: Simulating 64-bit precision with optimized 32-bit computation for lower power consumption. Advanced Adaptive Algorithms: Implementing next-gen AESA radar with thousands of antenna elements.
Looking ahead, the “One PFLOPS in a Shoe Box” challenge highlights the need for ultra-high-performance computing within stringent power constraints.
Key Takeaways for B2B Decision-Makers
For defense, aerospace, and adjacent AI-driven industries, the transition to GPU-accelerated signal processing offers: Faster, more efficient real-time analytics Lower hardware and operational costs Scalability across different applications A future-ready platform for AI and deep learning integration
‘Just as old paradigms in computing must pave way for future technology adoption, our ability to project and protect sovereignty must adapt to new threats, whether they be internal or external.’
Petter Olafsen, Stelia Analyst
GPU-based signal processing is no longer just an option, increasingly t’s a necessity for high-performance, real-time systems. Companies that invest in GPU or other forms of accelerated computing now will lead the next wave of AI-driven, computationally intensive applications in the defense domain and beyond. As the echoes of the Cold War make their presence known, strategic imperatives shift where economic interdependence no longer guarantees geopolitical stability. Just as old paradigms in computing must pave way for future technology adoption, our ability to project and protect sovereignty must adapt to new threats, whether they be internal or external.
AI provides a foundation for co-operation between strategic partners to leverage knowledge in the pursuit of peace and stability.
