GTC 2026

The official launch of the VR200 NVL72 GPU platform — the next architecture after Blackwell — is the most anticipated hardware announcement of GTC 2026. Named after pioneering astronomer Vera Rubin, the platform is expected to deliver a 3.3x inference performance boost over the previous Blackwell generation and slash per-token costs by up to 90%.

The 72-chip NVLink rack configuration (NVL72) builds on the NVL36 Blackwell cluster architecture, scaling the interconnect fabric to support larger model inference in a single rack footprint. The Groq LPU licensing deal is expected to play a role in how inference is handled within the Vera Rubin ecosystem.

Huang's promise to "surprise the world" has focused analyst attention on a rumored architecture codenamed "Feynman" — a chip that, if the supply chain signals are accurate, would represent the most fundamental architectural departure in Nvidia's history.

Feynman is speculated to leverage silicon photonics — transmitting data using light rather than electrons — built on a 1.6nm process node. If confirmed, this would place Nvidia beyond the 2nm frontier where both Intel's 18A and TSMC's N2 currently operate, and would dramatically reduce memory bandwidth bottlenecks that currently limit large model inference.

Alongside hardware, GTC 2026 is positioning itself as the defining event for Agentic AI — autonomous systems that don't simply answer questions but execute complex, multi-step workflows across tools, databases, and external systems independently.

A major focus will be OpenClaw, an open-source project for building and deploying persistent AI agents — systems that maintain state and memory across sessions rather than resetting with each query. Jensen Huang has been explicit that he sees agentic AI as the next $10 trillion layer of the software stack — above and alongside existing enterprise platforms.

The most surprising development heading into GTC 2026 is the confirmed presence of Intel as a strategic partner — once Nvidia's primary CPU rival, now a collaborator following their landmark September 2025 agreement in which Nvidia invested $5 billion in Intel stock.

Intel is showcasing new x86 CPUs with native NVLink support — a fundamental architectural integration that allows unprecedented data throughput between Intel processors and Nvidia GPUs, eliminating PCIe bandwidth as a bottleneck in AI training and inference systems. The partnership also includes the first glimpse at Intel x86 SOCs integrating Nvidia RTX GPU chiplets, targeting thin-and-light laptops with professional AI inference capability.

Analysts are also watching for any confirmation that Intel's 18A foundry process will manufacture future Nvidia chips — which would be the most significant Intel Foundry customer win since the company restructured its foundry business.