NVIDIA GTC 2026: Vera Rubin, a Feynman preview and the trillion-dollar outlook
In one sentence At the GTC 2026 keynote (March 16, San Jose) Jensen Huang presents Vera Rubin as a complete seven-chip platform, previews the next Feynman architecture, adds Groq 3 LPUs to the lineup and sees at least $1 trillion in revenue across 2025-2027.
GTC is NVIDIA's annual conference where it shows what will land in AI data centers. Jensen Huang's keynote took place on Monday, March 16, 2026, at the SAP Center in San Jose.
The star is Vera Rubin, the platform that succeeds Blackwell. NVIDIA presents it not as a single GPU but as a whole system: seven different chips, five rack-scale systems and a supercomputer, including the new Vera CPU. Microsoft Azure was the first cloud to power up Vera Rubin NVL72 racks, which roll out to other customers in the following months.
Huang also showed what comes next: the Feynman architecture, with a new CPU called Rosa. The lineup also gains Groq 3 LPUs, chips specialized in answering with very low latency.
On the business side, Huang said he sees at least $1 trillion in revenue from 2025 through 2027. On the desktop there were updates to DGX Station and DGX Spark, and in robotics new automotive and industrial partners.
Companies
NVIDIA
Tools
Vera Rubin, Vera Rubin NVL72, Groq 3 LPU, DGX Station, DGX Spark, NemoClaw
Tags
Sources