Advanced nuclear technologies, recognized as reliable energy sources for AI-driven data centers, are gaining momentum in the United States, particularly in Texas. Deploying these systems requires a streamlined licensing process and robust safeguards to ensure operational safety and security. To address this, we propose developing a Nuclear Foundation Model (NFM), utilizing large language model (LLM) to integrate diverse data streams for effective anomaly detection, material accountancy, and cyber-threat identification.

Rapid materials discovery requires accurate, efficient prediction of key properties like energy, atomic forces, and stress tensors from atomic structures. Existing methods either offer computational efficiency with limited precision (invariant models) or precise symmetry capture but high computational cost (equivariant models). Our proposed foundation model integrates these strengths, significantly enhancing accuracy and computational efficiency for diverse materials science applications.