AEC AI and Tech Strategy Podcast AI and Digital Twins in AEC Infrastructure – Ep 103
23 snips
Mar 4, 2026 
Sean Young, NVIDIA director specializing in GPU-accelerated computing and digital twins, discusses AI-driven infrastructure tools. He explains physics-first digital twins and using synthetic data to train vision AI. He talks about agentic AI that generates geometry, runs physics checks, and coordinates workflows. He also covers NVIDIA-optimized data center design and practical steps for firms to start learning AI.
AI Snips
Chapters
Transcript
Episode notes
Digital Twin Means Physics First Simulation
- NVIDIA defines a digital twin as a physics-first complete simulation of reality rather than just a rendering, BIM, or point cloud.
- Physics includes Newtonian behavior and visual photonics so virtual sensors and AI trained on synthetic data behave correctly in the real world.
How Autonomous Driving Shaped NVIDIA's Synthetic Data Work
- NVIDIA's autonomous driving R&D creates synthetic data by capturing real-world driving and generating physics-accurate virtual scenarios for training.
- They mimic sensor outputs (camera, radar, lidar) with photoreal rendering so models generalize to real roads and conditions.
Start With A Contextual Minimum Viable Twin
- Build the minimum viable twin to the use case; more context usually improves AI accuracy but may increase compute and training time.
- For simple defect detection you can train on images of rust without full geometry, but include non-rust examples to avoid false positives.