NVIDIA GTC 2026 — Every Major Announcement You Need to Know

NVIDIA just wrapped GTC 2026 in San Jose, and Jensen Huang didn’t hold back. Over four days, the company unveiled new chips, a completely new architecture roadmap, open-source AI agent tools, and partnerships that stretch from hospitals to highways. If you use AI tools in any capacity, this event shapes what you’ll be working with over the next two years.

Here’s what actually matters — no fluff.

Vera Rubin: The Next Generation of AI Chips

The headliner hardware announcement is the Vera Rubin platform — NVIDIA’s successor to Blackwell. It’s a full-stack AI computing platform built for agentic AI, the kind of AI that doesn’t just answer questions but takes actions across software and the real world.

The platform includes seven chips and five rack-scale systems. The DGX Station GB300 is the most accessible piece: 748 gigabytes of coherent memory, up to 20 petaflops of AI compute, and the ability to run models with up to 1 trillion parameters. You can order one now from ASUS, Dell, GIGABYTE, MSI, and Supermicro.

For context, that’s a desktop-sized machine running models that previously required a data centre. If you’ve been following our coverage of AI agents and how they’re replacing traditional SaaS, this is the hardware backbone that makes those agents viable at scale.

Feynman and Kyber: What Comes After Vera Rubin

NVIDIA also previewed its next-next generation architecture, codenamed Feynman. It pairs a new Rosa CPU (named after Rosalind Franklin) with LP40 next-gen language processing units, all connected through Kyber — a new rack architecture fitting 144 GPUs in a vertical layout for higher density and lower latency.

This is a 2027+ roadmap item, but it signals that NVIDIA is planning for AI workloads to get significantly more demanding — think persistent AI agents running 24/7 across entire business operations.

DLSS 5: AI-Powered Gaming Gets a Major Upgrade

For gamers and creators, DLSS 5 is the headline. It introduces 3D-guided neural rendering, which NVIDIA says delivers real-time photoreal 4K performance. In plain language: your GPU uses AI to reconstruct frames so games look dramatically better without needing brute-force rendering power.

If you’ve got an RTX card, this is the kind of free performance boost that makes your existing hardware feel like an upgrade.

NemoClaw: Open-Source AI Agent Deployment

This one’s big for anyone building with AI. NemoClaw is an open-source framework developed in partnership with the OpenClaw project (which pulled over 100,000 GitHub stars in its first week). It gives enterprises a way to deploy autonomous AI agents with proper guardrails — runtime policy enforcement, network security, and privacy routing baked in.

Think of it as the plumbing that lets companies actually ship AI agents into production without the security nightmares. If you’ve been experimenting with building AI agents from scratch, NemoClaw provides the enterprise-grade infrastructure layer you’d eventually need.

Six Frontier Model Families

NVIDIA announced a Nemotron Coalition of six specialised model families:

This is NVIDIA positioning itself not just as a chip company but as a full-stack AI platform — from the silicon to the models to the deployment tools.

Autonomous Vehicles Are Accelerating

The autonomous vehicle announcements were stacked. Uber is launching a fleet powered by NVIDIA Drive AV software across 28 cities on four continents by 2028, starting with Los Angeles and San Francisco. New automaker partners include BYD, Hyundai, Nissan, and Geely, all building Level 4 autonomous vehicles on NVIDIA’s Drive Hyperion platform.

Isuzu and China’s Tier IV are even building autonomous buses. This isn’t a research demo — these are production timelines with named companies and specific cities.

Healthcare Robotics Gets Its Own AI Platform

NVIDIA launched the first domain-specific physical AI platform for healthcare. The toolkit includes Open-H (776 hours of surgical video data), Cosmos-H (a model for generating surgical video), and GR00T-H (a vision-language-action model for clinical tasks). Early adopters include CMR Surgical, Johnson & Johnson MedTech, and Moon Surgical.

The $1 Trillion Revenue Forecast

Jensen Huang forecasted that purchase orders between Blackwell and Vera Rubin would exceed $1 trillion through 2027. AWS alone is deploying more than 1 million NVIDIA GPUs starting this year, spanning Blackwell and Rubin architectures.

That number puts the scale of AI infrastructure investment into perspective. The compute demand, according to Huang, has increased “1 million times over the last few years.”

What This Means for You

If you’re learning and using AI tools, GTC 2026 signals a few things. First, AI agents are moving from experimental to production-ready — NemoClaw and the Vera Rubin platform are built specifically for this. Second, the hardware is getting more accessible: a DGX Station on your desk can now run trillion-parameter models. Third, AI is embedding into everything from gaming (DLSS 5) to surgery to autonomous buses.

The practical takeaway: the tools you learn today are going to run on significantly better infrastructure within 12-18 months. If you’re just getting started, our discover page is a good place to find the right AI tool for your workflow.

The full keynote replay and session recordings are available on NVIDIA’s GTC page.