NVIDIA GTC 2026: Everything You Need to Know

2026-03-25 · #ai #nvidia #hardware #agents

NVIDIA GTC 2026 took place March 16–19 in San Jose, California. Over 30,000 attendees from more than 190 countries, over 1,000 sessions, and a two-hour keynote from Jensen Huang — the CEO of the company whose chips power virtually the entire AI industry.

This post covers every major announcement from the conference, from new hardware to AI agent platforms to Uber’s autonomous taxi partnership.

$1 Trillion in AI Orders Through 2027

Huang opened the keynote with a revised demand forecast. At GTC DC last year, NVIDIA projected $500 billion in orders for Blackwell and Vera Rubin systems. This year, Huang raised that figure to at least $1 trillion through 2027.

In a follow-up meeting with analysts, he clarified that the projection covers only Blackwell and Vera Rubin systems — it does not include newer products like the Groq LPU, standalone Vera CPUs, or the next-generation Feynman platform.

For context, NVIDIA is currently the world’s most valuable public company at roughly $4.5 trillion. The company has reported eleven consecutive quarters of revenue growth above 55%, and current-quarter revenue is expected to rise approximately 77% year-over-year to around $78 billion.

Vera Rubin — NVIDIA’s Next-Gen AI Platform

The flagship hardware announcement was Vera Rubin, NVIDIA’s successor to the Grace Blackwell system. According to NVIDIA, the platform comprises 1.3 million components and delivers 10x better performance per watt than its predecessor.

The NVL72 GPU Rack integrates 72 Rubin GPUs and 36 Vera CPUs, providing 20.7 TB of HBM4 memory and 1.6 PB/s of memory bandwidth. Vera Rubin is scheduled to ship to customers later in 2026, with the more powerful Vera Rubin Ultra following in 2027.

Groq 3 LPU — A New Kind of AI Chip

One of the conference’s biggest surprises was the Groq 3 LPU — a Language Processing Unit built on technology from the startup Groq, which NVIDIA acquired in December 2025 for approximately $20 billion.

The Groq 3 LPX rack houses 256 LPUs with 128 GB of SRAM and 40 PB/s of bandwidth. The key metric: according to CNBC, the LPU rack delivers 35x better token-per-watt performance compared to Vera Rubin GPU systems.

An important distinction — this is not 35x faster response generation. It is 35x more AI processing per unit of energy consumed. For data center operators, where power costs represent a significant portion of operating expenses, this could be transformative.

Huang described the pairing of GPUs and LPUs as “unifying two processors of extreme differences — one for high throughput, one for low latency.”

AI Agents — NemoClaw and OpenClaw

The central theme of GTC 2026 was AI agents — systems that don’t just respond to queries but autonomously execute multi-step goals.

NVIDIA announced full support for OpenClaw, the open-source agent framework created by Peter Steinberger. Since launching in January 2026, OpenClaw has become one of the fastest-growing open-source projects in history. Huang compared it to Linux and Android in terms of its significance for the agent ecosystem.

Built on top of OpenClaw, NVIDIA introduced NemoClaw — a production-grade version with three security layers: OpenShell runtime sandboxing, a privacy router, and network guardrails. NemoClaw enables enterprises to build secure, always-on AI agents with persistent memory, real-time planning, and built-in safety rules.

Huang put it bluntly:

“Every company in the world will now need to know what your OpenClaw strategy is. Just as we all had our Linux strategy, just as we all had to have an Internet strategy.”

Physical AI — Robots and Autonomous Taxis

Huang declared on stage that “Physical AI has arrived” — artificial intelligence is moving beyond software into the physical world.

Over 110 robots were showcased across the GTC expo floor. NVIDIA’s robotics partners include ABB, Fanuc, Hexagon Robotics, Disney, and PTC. The company unveiled Cosmos 3 models for physical world simulation and the Isaac GR00T system for humanoid robotics.

Uber announced a partnership with NVIDIA to deploy autonomous taxis powered by NVIDIA Drive AV software in 28 cities across four continents by 2028. The first cities will be Los Angeles and San Francisco, with rides starting in 2027. BYD, Hyundai, Nissan, and Geely also joined NVIDIA’s RoboTaxi Ready platform alongside existing partners Mercedes, Toyota, and GM.

DLSS 5 — AI-Powered Neural Rendering

NVIDIA also introduced DLSS 5, the fifth generation of its AI-powered graphics rendering technology. Unlike previous versions, which enhanced existing frames, DLSS 5 generates entire frames from scratch using neural networks.

Huang noted that the same neural rendering pipeline powering consumer gaming GPUs also drives industrial simulations in Omniverse and digital factory twins — connecting the consumer GPU ecosystem directly to enterprise AI applications.

What GTC 2026 Means Going Forward

GTC 2026 showcased NVIDIA’s transformation from a chipmaker into a full-stack AI infrastructure provider — spanning hardware, software, and a growing ecosystem of partners.

Three key trends from the conference:

AI agents are moving from experiment to production. According to NVIDIA’s “State of AI” survey, 44% of companies have either deployed AI agents or are actively testing them. In telecommunications, adoption reaches 48%.

Inference is overtaking training. The majority of AI compute is now spent generating responses and executing tasks, not training models. This is the driving force behind the LPU investment — chips purpose-built for inference workloads.

AI is entering the physical world. Autonomous vehicles, industrial robotics, and digital factory twins are no longer roadmap items — they are partnerships with announced launch dates and named cities.