Scaling Physical AI and Physics-Informed Surrogates with NVIDIA and Google Cloud

About the Session

AI-physics is rapidly emerging as a new computing paradigm for industrial engineering, complementing traditional simulation with learned models that can accelerate design exploration, optimization, and decision-making across automotive, aerospace, energy, and semiconductor verticals. Advances in geometry-aware architectures, graph neural networks, transformers, and hybrid physics-ML models are improving accuracy, generalization, and scalability. 

These workloads behave very differently from traditional AI/LLM pipelines. Rather than operating on linear token streams, AI-physics models train and infer over large 3D meshes, point clouds, continuous fields, and complex engineering geometries, often tightly coupled to high-fidelity solvers, optimization loops, and design systems. This creates distinct requirements around memory bandwidth, data movement, distributed training, model parallelism, and low-latency inference.

Join Todd McDevitt (NVIDIA) as we explore the role of AI-physics in industrial engineering, recent progress in model architectures and training at scale, and real-world examples across CFD, thermal, structural, and multiphysics modeling. We will also examine why this represents a distinct cloud growth opportunity for Google Cloud: AI-physics combines large-scale training, bursty simulation-driven data generation, and repeated production inference creating demand for elastic GPU infrastructure, high-performance storage and networking, and tightly integrated cloud workflows that can scale from model development through deployment.