NVIDIA Bets Big on AI Agents and Teases Its Most Advanced Chip Yet at GTC 2026

NVIDIA Charts a New Course at GTC 2026

NVIDIA's annual GPU Technology Conference kicked off in San Jose on Monday, drawing over thirty thousand attendees from one hundred and ninety countries for what CEO Jensen Huang has called a pivotal moment in the company's evolution. The four-day event, running March sixteenth through nineteenth, features more than seven hundred sessions spanning AI infrastructure, robotics, and autonomous systems.

Feynman: The Next Frontier in AI Silicon

Among the headline announcements was a first official look at the Feynman GPU architecture, NVIDIA's successor to the Vera Rubin platform. Named after Nobel laureate Richard Feynman, the chip is slated for a twenty twenty-eight commercial launch and will be manufactured using TSMC's A16 process at one point six nanometres, making it the first major chip at that scale. The architecture introduces silicon photonics, replacing traditional electrical signals with light to move data between chips, and pairs with an eighth-generation NVLink interconnect running at seven point two terabits per second. Crucially, Feynman is being designed as an inference-first processor, built specifically for the long-context, multi-step reasoning that AI agents demand.

NemoClaw: NVIDIA's Play for the Agent Economy

On the software side, NVIDIA formally launched NemoClaw, an enterprise-grade open-source AI agent platform. Inspired by OpenClaw, the autonomous agent framework acquired by OpenAI earlier this year, NemoClaw is designed to let companies deploy AI agents that carry out complex, multi-step tasks with minimal human oversight. The platform is hardware-agnostic and includes built-in security features. NVIDIA has already begun pitching it to major enterprise partners including Salesforce, Cisco, Google, Adobe, and CrowdStrike, offering free access in exchange for development contributions.

A Trillion-Dollar Bet

With daily Build-a-Claw workshops teaching developers to customise agents and memory partners Samsung and SK Hynix demonstrating next-generation HBM4 memory on the show floor, NVIDIA is positioning itself at the centre of what it believes will be a trillion-dollar shift in computing infrastructure, from passive AI models to autonomous, reasoning agents.