Descriptions:
In a sit-down interview at Nvidia’s GTC conference, CEO Jensen Huang joined the All-In Podcast to explain the company’s strategic transformation from a GPU vendor into what he calls an “AI factory” platform. The centerpiece is Dynamo, Nvidia’s inference operating system, which enables disaggregated inference—splitting the AI processing pipeline across heterogeneous hardware including GPUs, CPUs, networking processors (BlueField), and now Groq LPU chips following Nvidia’s acquisition.
Huang detailed why Groq processors should occupy roughly 25% of next-generation Vera Rubin data center racks, particularly to handle prefill-decode disaggregation in agentic workloads. He described modern AI agents—which access working memory, long-term memory, use tools, and coordinate with other agents running diverse model types—as fundamentally different from earlier LLM inference, justifying Nvidia’s expanded product surface and an estimated 33–50% increase in total addressable market. The conversation covered Vera Rubin’s design for heterogeneous workloads, the role of networking processors, and how scale-up versus scale-out switching fits into the full stack.
Beyond hardware, the interview touched on physical AI and robotics, sovereign AI funding dynamics, and striking real-world examples from GTC attendees—including a CEO who described replacing an entire enterprise software stack in 90 minutes with an agentic system on a Sunday night, and a genomics team that reproduced what would have been a seven-year PhD thesis in 30 minutes using AutoResearch. The interview offers a direct view into Nvidia’s long-term roadmap framed in Jensen Huang’s own words.
📺 Source: All-In Podcast · Published March 19, 2026
🏷️ Format: Interview
