Jensen Huang takes the stage at 11 a.m. PT today for what may be the most consequential GTC keynote in Nvidia’s history. The two-hour address from SAP Center in San Jose will lay out a fundamental shift in the company’s strategy: from training chips to inference, from cloud-first to edge computing, and from AI that responds to prompts to AI that acts autonomously.
30,000 attendees from 190 countries are watching. Here’s what to expect.
The Vera Rubin Architecture
At the center of GTC 2026 is Nvidia’s next-generation chip platform, Vera Rubin - the successor to Blackwell, named after the astrophysicist who proved the existence of dark matter.
The architecture pairs a Rubin GPU with a Vera CPU, both manufactured on TSMC’s 3nm process with HBM4 memory. The numbers are significant: each Rubin GPU delivers 50 petaflops of inference performance, a 5x improvement over Blackwell’s GB200. Early specifications suggest up to 288GB of HBM4 memory per unit.
More practically, Nvidia claims Vera Rubin will reduce the number of GPUs needed to train mixture-of-experts models by 4x compared to Blackwell systems. Inference token costs should drop by 10x.
The chips entered full production at the start of 2026, with volume shipments ramping through the second half of the year.
The Groq Integration
In December 2025, Nvidia paid $20 billion for Groq’s assets - by far its largest acquisition ever, roughly triple its 2019 Mellanox deal. The purchase brought Groq’s Language Processing Unit technology in-house, along with founder Jonathan Ross and key engineers, several of whom previously designed Google’s TPUs.
Groq’s architecture takes a fundamentally different approach to inference. Where GPUs move data constantly between memory and compute, Groq’s LPU uses deterministic execution with massive on-chip SRAM, eliminating bandwidth bottlenecks. Independent tests showed Groq running inference roughly 2x faster than any competitor.
Today’s keynote is expected to reveal how Nvidia is integrating Groq technology into a new inference-focused processor. This marks a departure from Nvidia’s “one GPU does everything” philosophy toward specialized silicon for different workloads.
The $20 billion price tag - roughly 2.9x Groq’s $6.9 billion valuation from three months earlier - suggests Nvidia viewed the startup as an existential threat worth eliminating. Now that threat is an asset.
NemoClaw: Agents for the Enterprise
When OpenClaw exploded onto the scene in January 2026, it gained 60,000 GitHub stars in 72 hours. By March it had accumulated over 247,000 stars, making it the fastest-growing open source project in history. Jensen Huang called it “the most important software release probably ever.”
OpenAI ultimately acquired OpenClaw and hired its creator, Peter Steinberger. But Nvidia had been watching - and building its response.
NemoClaw is Nvidia’s open-source platform for enterprise AI agents, set for full reveal at GTC. Where OpenClaw targeted individual users running local AI assistants, NemoClaw is built for companies deploying agents across their workforces.
The platform features multi-layer security with role-based access controls, signed skills, and activity logging. It’s explicitly hardware-agnostic, designed to run on Nvidia, AMD, Intel, or CPU-only systems. Companies can deploy agents without relying on proprietary commercial APIs.
Early partners reportedly get free access in exchange for contributing to the project, positioning NemoClaw as the enterprise foundation for the agentic AI era that GTC 2026 is built around.
Robotics: GR00T N1.6 and Physical AI Days
The humanoid robot moment is arriving, and Nvidia wants to be Android - the operating system that powers everything.
Isaac GR00T N1.6, unveiled at GTC, is a vision-language-action foundation model purpose-built for humanoid robots. It applies chain-of-thought reasoning to physical control, allowing robots to reason through novel situations step by step rather than pattern-match against training data.
Two dedicated “Physical AI Days” run alongside the main conference, with demonstrations from Boston Dynamics, Caterpillar, LG Electronics, NEURA Robotics, and others building on Nvidia’s Isaac platform.
Running on Nvidia’s Jetson edge computing hardware, these systems can process vision and decision-making locally without cloud connectivity - critical for industrial settings where latency and connectivity aren’t guaranteed.
CPUs Take Center Stage
Perhaps the most surprising shift: CPUs may matter more than GPUs for agentic AI.
According to CNBC, Nvidia is developing CPU-centric data center servers for the agentic era. The conference floor will reportedly feature a CPU-only rack, signaling that autonomous agents running continuously require different compute profiles than batch training workloads.
The 88-core Vera CPU (codenamed Olympus, built on Arm v9.2-A) replaces the Grace CPU from previous generations. It’s designed for always-on agent workflows where predictable, efficient processing matters more than peak throughput.
What This Means
Nvidia is positioning itself at every layer of the AI stack as the industry pivots from training to deployment:
Chips: Vera Rubin for training, Groq-derived silicon for inference, Jetson for edge robotics Software: NemoClaw for enterprise agents, Isaac for robotics, OpenClaw support for developers Models: GR00T for humanoids, integration with every major foundation model provider
The “Agentic AI Inflection Point” framing isn’t marketing - it reflects a genuine architectural shift. AI systems that run continuously, take actions autonomously, and operate at the edge require different infrastructure than systems that process prompts in the cloud.
Whether Nvidia can maintain its dominance through this transition is the $3 trillion question. Today’s keynote will show how seriously the company is taking the challenge.
The Bottom Line
GTC 2026 marks Nvidia’s pivot from selling picks and shovels to owning the mine. The combination of Vera Rubin, Groq integration, NemoClaw, and robotics foundations represents a bet that the next phase of AI isn’t about bigger models but about models that do things in the world.
Jensen Huang’s keynote streams live at nvidia.com starting at 11 a.m. PT. No registration required.