Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast

This episode features a conversation with Jensen Huang, CEO of NVIDIA, focusing on the company's pivotal role in the AI revolution and his leadership philosophy. Huang explains NVIDIA's evolution from chip-scale design to "extreme co-design" at the rack and data center level, encompassing GPUs, CPUs, memory, networking, storage, power, and cooling. This holistic approach is necessary because complex problems now require distributed computing, aiming for speedups far beyond merely adding more computers.

Huang details the challenges and strategies behind this co-design, highlighting his unique management style where a large direct staff of 60+ experts from diverse disciplines collaborate on problems collectively rather than through one-on-one meetings. He traces NVIDIA's path from a specialized accelerator company to an accelerated computing company, mentioning key innovations like the programmable pixel shader, FP32 integration, Cg, and ultimately CUDA. A critical strategic decision was putting CUDA on GeForce GPUs, which, despite consuming significant profits and temporarily reducing NVIDIA's market cap from approximately $8 billion to $1.5 billion, was essential for building a developer "install base" – a foundation Huang argues is paramount for any computing architecture's success.

The discussion delves into Huang's leadership approach for making bold, future-defining bets. He emphasizes the importance of intense curiosity, a strong belief in a reasoned future, and a continuous process of shaping the belief systems of his board, management, employees, and even industry partners over time. This ensures widespread buy-in when major announcements are made. Finally, Huang discusses his continued belief in AI scaling laws (pre-training, post-training, test time, and agentic scaling) and addresses perceived blockers, arguing that synthetic data will overcome pre-training data scarcity and that inference ("thinking") is intensely compute-intensive, not simple, requiring significant resources for reasoning and planning.