An Interview with Nvidia CEO Jensen Huang About Accelerated Computing

In this March 2026 interview, Jensen Huang explains Nvidia's comprehensive approach to AI infrastructure, positioning the company as more than just a chip manufacturer. He discusses how AI crossing the threshold of practical utility in the past year was driven by improved reasoning capabilities, reduced hallucinations, and better grounding - particularly evident in coding applications where AI can now verify and iterate on its output.

A major focus is Nvidia's recent Groq acquisition, which addresses the need for ultra-low latency inference to serve AI agents and coding assistants used by enterprise customers willing to pay premium prices for faster performance. This represents a disaggregated approach to inference, separating high-throughput from low-latency workloads. Huang also introduces Vera CPUs, designed specifically for AI agent workloads with emphasis on single-threaded performance and high I/O bandwidth, contrasting with traditional hyperscale CPUs optimized for core count.

Huang addresses geopolitical concerns, particularly criticizing 'doomers' for creating fear around AI that could handicap American technological leadership. He argues that restricting AI development domestically while China advances could repeat Europe's experience in previous industrial revolutions. He emphasizes that AI encompasses five layers (power, chips, infrastructure, models, applications) and American leadership requires success across all layers, not just bundling them together.

Throughout the interview, Huang maintains that Nvidia remains focused on being a platform and infrastructure company rather than competing directly in end-user applications, preferring to enable others while staying at the technology frontier.