Most AV companies are solving the wrong problem. They're obsessing over model architectures when the data is the real differentiating factor. On the latest NVIDIA AI Podcast episode, I sat down with host Noah Kravitz and Porsche's Tech Lead Tin Sohn to discuss what it takes to build safe, intelligent autonomous vehicles — and why data quality, simulation, and curation are the real bottlenecks.

📊 Data, not models, is the real differentiator: As foundation models become commoditized, competitive advantage lies in how teams curate training sets, evaluate performance, and systematically address failure modes.

🛠️ Simulation and synthetic data are now essential: You can’t capture every rare edge case in the real world — generating diverse, high-fidelity scenes is how AV systems learn to handle the unexpected. Platforms like @NVIDIA Drive are powering end-to-end solutions for AVs, from data collection to model testing and validation in simulation environments.

🔍 Trust requires explainability, not just accuracy: AVs must reason like humans, understanding spatial relationships, object dynamics, and physical constraints — then communicate their decisions clearly to drivers. Black-box accuracy isn't enough when lives are at stake.

🚗 Top automakers are becoming software companies: Leading teams are moving away from off-the-shelf solutions and building internal data loops to continuously refine models with their unique domain expertise and fleet data.

At Voxel51, we're proud to help companies like Porsche turn raw visual data into trustworthy, high-performance AI systems. If you’re looking to leverage your data to build industry-leading AI, this one's worth a listen.