Aerospace and Defense Leaders Discuss AI Surrogate Modeling in FedRAMP-Compliant Cloud HPC Environments

Aerospace and Defense Leaders Discuss AI Surrogate Modeling in FedRAMP-Compliant Cloud HPC Environments

At SC24, the premier supercomputing conference, Rescale brought together top Aerospace and Defense experts for a panel discussion to share their advice and experiences in high performance computing (HPC).

Four major Aerospace and Defense (A&D) institutions were represented, including:

• Lawrence Livermore National Laboratory: Dr. Aaron Fisher, Director of HPC for the Energy Innovation Program
• Department of Defense (DoD):  Dr. Kevin Newmeyer, Acting Director for the Department of Defense HPC Modernization Program
• Northrop Grumman: Cole Turner, Sr. Principal Software Engineer, HPC
• BAE Systems: Scott Grabow, HPC Technical Expert and Systems Engineer

Sharing insight and camaraderie, these distinguished A&D professionals explored emerging numerical methods, such as differentiable computing and AI surrogate modeling. They also emphasized the role of high-precision GPUs in accelerating complex computations. Throughout the panel, the discussion consistently highlighted the criticality of cybersecurity, emphasizing its importance in HPC.

Now, let’s dive into the recap, or you can head here to watch the full recording.

AI Surrogate Modeling 

The panelists reflected on their AI adoption journeys, sharing how they navigated both excitement and skepticism. While each faced unique challenges, they all shared the struggle of balancing time and resources between adopting new AI-based methods and meeting traditional HPC demands.

Dr. Aaron Fisher, Director of HPC at Lawrence Livermore, shared how he made AI tools widely available to encourage experimentation while Dr. Newmeyer of the DoD focused on breaking down silos of expertise to enable AI adoption. Cole Turner of Northrop Grumman and Scott Grabow of BAE Systems focused on merging traditional CAE with AI-driven surrogate modeling, shifting teams from CPU-based simulations to GPU-accelerated learning. 

Fisher noted, “So really, we’re deep in the game of surrogate modeling… Our modeling and simulation products provide a massive data amplification for AI training, enabling AI to generate synthetic, yet highly reliable, data.”

Turner added, “We are still in the early stages of adopting AI in defense applications. Our focus is on building a flexible computing infrastructure that can integrate future AI models while supporting existing CPU-driven workloads.”

Interestingly, all the panelists seemed to be focused on identifying the right use cases for AI surrogate modeling. Their experiences highlighted the importance of flexible strategies, collaboration, and the rethinking of legacy workflows.

FP64 GPUs and Differentiable Computing

AI/ML often operates effectively on widely available 32-bit precision GPUs, but demanding fields like computational fluid dynamics (CFD), structural analysis, and weather prediction require the higher accuracy 64-bit FP64 GPUs. 

Grabow  observed, “AI/ML has an advantage right now. What it’s forcing people to do is really look at their models and consider how much precision they actually need.”

High-precision GPUs enable differentiable computing by providing the necessary accuracy for gradient calculations, allowing neural networks to train effectively and converge faster. This mathematical technique allows AI models to be more efficiently integrated into physics-based solvers, ultimately allowing AI to solve real-world problems more effectively.

Expanding on Grabow’s point, Fisher highlighted how these recent mathematical advances are reshaping engineering. “Differentiable computing is revolutionizing topology and shape optimization by accelerating iterative design processes, leading to faster and more efficient engineering solutions.” 

This was helpful advice for those developing modeling and simulation applications. Simply put, differentiable computing, combined with HPC-optimized compiler frameworks like LLVM, will dramatically improve optimization algorithms and drive progress in aerospace and defense applications.

However, not every company was rewriting their numerical solvers just yet. Several were balancing investments between traditional approaches and emerging techniques.

Northrop Grumman, for example, is prioritizing building HPC resources that can flex. As Turner noted, “Our goal is to create a system that adapts to evolving engineering needs as AI-driven models become more prevalent.” He added, “We’re not seeing strong demand for that yet, but using Rescale provides us with a flexible platform.”

Harnessing AI for National Security Challenges

Not surprisingly, even the Department of Defense’s most sensitive programs are facing disruption from AI. 

Dr. Kevin Newmeyer of the DoD outlined the unique challenges of integrating AI to support strategic military operations. He explained, “We rely on physics-based models for aircraft and weapons development, but AI-driven surrogate models allow us to run real-time combat simulations at unprecedented scales.” 

With AI adoption on the rise, pinpointing the right use cases will be essential. Newmeyer provided a strategic outlook, emphasizing that testing force-on-force scenarios and adapting strategies in real-time will shape the future of military preparedness.

Cybersecurity

Security has become a critical concern as collaborative workflows, data sharing, and reliance on external cloud resources introduce new vulnerabilities. The panel made it clear that sophisticated attackers are not just increasing in number but also refining their methods to target both sensitive research data and critical infrastructure.

How many attacks are we talking about? The panelists described an onslaught that shows no sign of slowing. “Tens of thousands of hits a day,” said Newmeyer. Fisher added, “We’ve got cybersecurity teams that are fielding thousands of attacks a day.”

While BAE may appear less vulnerable than the DoD, security is still a top concern. Grabow described the struggle to move beyond a “checkbox approach” to common vulnerabilities and exposures. With over 250,000 vulnerabilities reported last year alone, keeping pace has become nearly impossible. His organization is now shifting toward prioritizing known exploited vulnerabilities to better manage this overwhelming challenge.

The security discussion concluded with Fisher stressing segmentation as a strategy, such as air-gapped systems, to isolate and protect critical workloads. He also highlighted the need for FedRAMP compliance to secure cloud-based HPC environments. 

Recruiting and Mentoring Tomorrow’s HPC Leaders 

After the cybersecurity discussion, where Grabow pointed out the need for skilled professionals to advance AI-driven threat detection, the panel shifted focus to cultivating future talent.

The panel unanimously confirmed what the audience likely already knew — a growing talent gap. As AI, cybersecurity, and cloud technologies converge, traditional HPC roles are undergoing significant transformation and becoming increasingly difficult to fill.

“What keeps me up at night is who’s going to replace me and my senior staff here in the next couple of years. How do you develop those people?” Newmeyer  asked.

He then reinforced the importance of cross-training and interdisciplinary education. “We need problem solvers, not just domain experts. The future of HPC depends on professionals who can bridge the gap between physics-based simulation, AI, and cybersecurity.”

While the panelists had different management approaches, they all shared a strong focus on talent development as leaders.

Conclusion

We hope this panel recap highlights the Aerospace and Defense Industry’s collective interest in advancing AI, HPC, and cybersecurity to shape the future of national security.

A key takeaway from the discussion is that the next generation of HPC is already taking shape. For example, Turner shared how operations are already transforming at Northrop Grumman: “Cloud computing allows us to move away from monolithic, one-size-fits-all HPC clusters. Instead, we can dynamically provision compute resources tailored to specific workloads, maximizing efficiency and cost-effectiveness.”

The discussion wrapped with a look back at the evolution of computing, starting all the way back with Apollo 11. To gain a deeper understanding of the panelists’ historical perspectives, shaped by years of experience, and to hear all the details, watch the full SC24 video.

Please see our events page for future events like SC24. If you have any AI, HPC, security, or compliance questions, we would love to hear from you. Reach out to us here.