Dwarkesh Podcast Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat
3154 snips
Apr 15, 2026 
Jensen Huang, Nvidia co-founder and CEO who helped pioneer GPU computing, dives into TPU rivalry, supply chain bottlenecks, and why Nvidia stays out of becoming a hyperscaler. He also gets into selling AI chips to China, CUDA’s ecosystem edge, and why the company bets on a few strong chip designs instead of many wild alternatives.
AI Snips
Chapters
Transcript
Episode notes
CUDA's Moat Is Install Base Trust And Optimization
- Jensen Huang says CUDA's moat is not just APIs but a trusted stack, giant install base, and presence in every cloud and device tier.
- He claims NVIDIA engineers still unlock 2x to 3x speedups for frontier labs because GPUs are F1 cars, not easy cruising CPUs.
Anthropic Won Alternatives Because NVIDIA Missed The Financing Moment
- Jensen Huang says Anthropic is an exception, not proof that ASICs broadly threaten NVIDIA.
- He argues Google and AWS won Anthropic compute partly by making multibillion-dollar investments NVIDIA was not yet prepared to offer, which he now regrets.
NVIDIA Avoids Becoming A Cloud To Preserve Focus
- Jensen Huang says NVIDIA should do as much as needed and as little as possible, so it supports clouds rather than becoming one.
- He applies the same logic to investing: help ecosystems like CoreWeave exist, support all major labs, and avoid picking winners.