Jensen Huang sat with Lex Fridman for nearly 2.5 hours and covered the full stack: rack-scale engineering, AI scaling laws, supply chain bottlenecks, memory and power constraints, TSMC dependency, China export controls, and whether NVIDIA hits $10 trillion. NVIDIA is currently valued near $4 trillion. That number is the baseline, not the headline.
The most technically dense sections run from the 22-minute mark through the 52-minute mark. Huang breaks down the three biggest blockers to AI scaling, memory architecture chief among them, and explains how NVIDIA's extreme co-design philosophy, building compute, networking, and software as a unified system rather than discrete parts, is the actual moat. Not the chips alone. The moat section at 1:15 is worth reading the transcript directly.
The conversation closes on AGI timelines, the future of programming, consciousness, and mortality. Huang does not hedge on the programming question. His answer about what software development looks like in 10 years is specific and worth hearing in his words, not a summary.
[WATCH ON YOUTUBE →]