Nvidia GTC 2026 kicks off in San Jose with Jensen Huang keynote on AI infrastructure

San Jose is hosting over 30,000 attendees this week as Nvidia's annual GTC conference gets underway, drawing engineers, researchers, and enterprise buyers from 190 countries. The scale alone tells you something about where the industry's attention is right now. Everyone wants to know what comes next for AI hardware, and Nvidia is the company most people are watching.

CEO Jensen Huang is scheduled to deliver the opening keynote on Monday, March 16, at SAP Center. His presentations at GTC have become something of an annual ritual for the AI industry. Last year's announcements sent Nvidia's stock surging within days. This year, the company has been tight-lipped about specifics, but the agenda points are already circulating, and they cover a lot of ground.

Vera Rubin GPU architecture takes center stage

The most anticipated disclosure is Nvidia's Vera Rubin GPU architecture. Named after the astronomer who helped confirm the existence of dark matter, Vera Rubin is expected to succeed the current Blackwell lineup. Huang is expected to go into detail on performance benchmarks and memory bandwidth numbers. Whether those figures arrive with firm ship dates or just roadmap slides remains to be seen, but given how Blackwell ramped in 2025, buyers will be pushing hard for specifics on availability.

Inference optimization is also on the agenda. As companies move from training large models to actually running them at scale, the cost and speed of inference have become the bottleneck most enterprises care about. Nvidia has been working on software-level improvements alongside hardware, and GTC is the natural stage for those announcements.

NemoClaw and the push into agentic AI

Nvidia teased its NemoClaw platform in the days before the conference opened. NemoClaw is described as an enterprise AI agent system, designed to let businesses build and deploy autonomous agents that can handle multi-step tasks without constant human intervention. The name is new, but the concept has been building across several Nvidia software releases over the past year.

Agentic AI has become the phrase everyone in enterprise software is using right now. The difference with NemoClaw, at least based on what Nvidia has hinted at, is that it sits closer to the hardware layer. That matters for latency. A system that can make decisions at the chip level rather than routing everything through a cloud API is going to behave differently in production. Enterprises running real-time workloads will care about that distinction a lot.

A new laptop CPU targeting gaming and local AI workloads

Less expected was the tease of a new laptop CPU. Nvidia has historically stayed out of the general-purpose CPU market on the consumer side, leaving that to AMD and Intel. The new chip appears to target two distinct use cases: gaming performance and local agentic AI workloads. Running AI agents locally, without sending data to a cloud server, is a growing concern for both privacy-conscious consumers and enterprises with strict data residency requirements.

If Nvidia can ship a laptop chip that handles both high-frame-rate gaming and local model inference at acceptable power draw, it changes what a gaming laptop actually is. That is a big if, and the technical details will determine whether this is a serious product or a concept announcement. The keynote on Monday should clarify which it is.

What attendees and analysts are watching for

Beyond the hardware, GTC this year has a heavy software and developer tools track. Nvidia's CUDA ecosystem remains one of the biggest competitive advantages the company holds. Training and deploying AI on non-Nvidia hardware is technically possible but still painful in practice, largely because so much AI software has been written with CUDA in mind first. Announcements around developer tooling, inference runtimes, and enterprise software integrations are expected throughout the week, not just in Monday's keynote.

The conference runs through Thursday, March 19. Sessions span autonomous vehicles, healthcare AI, physical simulation, and robotics alongside the core AI infrastructure content. For companies that have been waiting to make procurement decisions about next-generation AI clusters, the next four days will give them a clearer picture of what Nvidia's roadmap actually looks like for the rest of 2026.