Dwarkesh Patel interviews Jensen Huang about Nvidia’s moat, supply-chain scaling, CUDA, competition from TPUs/ASICs, cloud customers, and China export policy. Huang argues Nvidia’s advantage is not just chips but the full accelerated-computing stack, ecosystem, install base, and co-design with customers and suppliers, while pushing back hard on the idea that selling chips to China is a strategic mistake.
Watch on YouTube ›Get the market thesis, key claims, assets, contradictions, and follow-up questions from any financial video — then unlock a version personalized to your portfolio, watchlist, and favorite speakers.
This is a long-form interview centered on whether Nvidia’s moat can persist as AI commoditization narratives intensify. Jensen Huang frames Nvidia as the middle layer in a transformation from electrons to tokens, arguing that the hard part is not commoditized because it requires deep engineering, software, systems design, and ecosystem orchestration. He repeatedly emphasizes that Nvidia tries to do ‘as little as possible’ itself while partnering broadly upstream and downstream, building a wide AI ecosystem across supply chain, clouds, model builders, and application developers. A major theme is supply-chain scale. Huang argues Nvidia’s growth is not constrained by insurmountable bottlenecks: if demand is clear, the industry can ‘swarm’ around shortages like CoWoS, memory, packaging, or even EUV capacity within a few years. …
Near term, Nvidia still looks tactically supported as long as demand stays ahead of supply and major platforms keep ordering into the queue. The immediate risk is policy noise around China and any signs that alternative accelerators are winning specific workloads faster than expected.
Over the next few quarters, the base case is continued scale-up rather than a supply wall, with Nvidia trying to convert ecosystem breadth into sustained share. The key validation point is whether CUDA, install base, and customer co-design keep outweighing TPU/ASIC fragmentation in real deployments.
Structurally, Huang is arguing that accelerated computing is the long-lived regime and Nvidia is the most likely default platform for it. If that regime persists, the moat will come from software, developer gravity, and system-level co-design more than from any single generation of silicon.
Software commoditization does not automatically commoditize Nvidia because Nvidia’s core job is the hard transformation from electrons to tokens.
He argues the real value lies in making tokens valuable through difficult engineering and system design, not merely assembling commodity parts.
Nvidia’s moat comes from its broad ecosystem across supply chain, clouds, model makers, and application developers.
He repeatedly says Nvidia has the largest partner ecosystem upstream and downstream and spans all five layers of the AI stack.
AI tool usage will expand exponentially, causing software tools like Synopsys and Cadence to see far more instances and usage.
He argues agents will increasingly use tools, increasing rather than reducing demand for software toolmakers.
What does Nvidia's competition from TPUs mean for the company going forward?
He says Nvidia builds a much broader thing than a TPU: accelerated computing for AI plus many other workloads like data processing, physics, and drug discovery. Because Nvidia systems are flexible, operator-friendly, and backed by a large ecosystem, he argues the company can serve far more applications than TPU or ASIC competitors.
Why is Nvidia able to achieve performance and efficiency gains beyond Moore's Law?
He says the gains come from new models and kernels, especially MoEs, plus CUDA-based programmability and co-design across the processor, fabric, libraries, and algorithms. He points to Blackwell versus Hopper as an example of a leap that could not be explained by Moore's Law alone.
Why didn't Nvidia make the big investments in OpenAI and Anthropic sooner, when it had the cash?
He says Nvidia invested as soon as it could and would have done so earlier if it had been able. At the time, Nvidia had never invested outside the company, underestimated the scale required, and assumed the labs could be funded by venture capital.
Unlock the full claims, asset map, scores, related transcripts, follow-up questions, and AI chat — shaped around your portfolio, watchlist, favorite speakers, and risks.
