Jensen Huang on Nvidia’s Supply Chain Moat, TPU Competition, and Selling to China
Jensen Huang in conversation with Dwarkesh Patel. Published April 2026. Jensen argues that Nvidia’s durable moat is not CUDA or architecture but its grip on the increasingly scarce and complex supply chain for advanced AI hardware. Covers whether TPUs can break Nvidia’s dominance, why Nvidia chooses not to become a hyperscaler, and Jensen’s view on selling AI chips to China.
Key ideas
- Nvidia’s moat is supply chain, not CUDA. As AI compute scales toward multi-trillion-dollar buildouts, the binding constraint is not GPU architecture or software — it is the manufacturing and logistics infrastructure for the components that make up a data centre. Nvidia has spent decades building these supply chain relationships; competitors cannot replicate them quickly.
- “If our next several years are a trillion dollars in scale, we have the supply chain to do it.” This is the core strategic claim: at the scale the industry is targeting, supply chain execution is what separates winners from those who cannot deliver.
- TPUs are real but not decisive. Google’s TPUs and other custom silicon are competitive for specific workloads but do not break Nvidia’s overall position. The supply chain moat matters at the system level, not just the chip level.
- Nvidia should not become a hyperscaler. Jensen argues the hyperscaler model requires a fundamentally different business — customer relationships, data centres, enterprise sales, operational infrastructure — that would distract from Nvidia’s core strength and create channel conflict with its best customers.
- Should sell chips to China. Restricting sales primarily harms US companies while doing little to prevent China from developing AI capabilities through domestic production or alternative sourcing. Jensen’s view is that engagement is strategically preferable to restriction.
Supply Chain as Moat
Jensen’s argument is structural: as AI data centre buildout scales to trillion-dollar annual investment, the scarce resource is not chips per se but the integrated supply chain — advanced packaging, HBM memory, cooling systems, power delivery, and the systems integration expertise that ties them together. These are multi-decade relationships with TSMC, SK Hynix, memory suppliers, and custom component manufacturers. No new entrant can replicate them in the 2-3 year timelines that matter for the current buildout.