Nvidia’s GTC 2026 kicks off Sunday in San Jose, with CEO Jensen Huang delivering the keynote Monday at 11 a.m. PT. More than 30,000 attendees from 190 countries will gather for what the company promises is a full-stack showcase: chips, software, models, and applications. Here’s what to expect.
The Main Event: Vera Rubin Architecture
The spotlight falls on Rubin, Nvidia’s next-generation GPU architecture that represents a massive leap over the current Blackwell generation.
The numbers are significant. According to Nvidia’s technical blog, a single Rubin GPU packs 336 billion transistors - 1.6x more than Blackwell - and delivers 50 PFLOPS of NVFP4 inference performance, a 5x increase. Memory bandwidth nearly triples compared to Blackwell, hitting 22 TB/s thanks to HBM4 memory with doubled interface width.
Each GPU can address 288 GB of HBM4, and the NVL72 configuration pairs 72 GPUs with 36 CPUs for 260 TB/s of scale-up bandwidth.
Manufacturing uses TSMC’s 3nm process. Though originally slated for mass production in the second half of 2026, Nvidia says Rubin entered full production in Q1 2026.
NemoClaw: Nvidia’s Answer to the OpenClaw Acquisition
When OpenAI acquired OpenClaw in February, enterprises found themselves suddenly dependent on a cloud-only AI agent platform. Nvidia’s response is NemoClaw, an open-source AI agent platform built for enterprise security and privacy.
According to CNBC, Nvidia has been pitching NemoClaw to Salesforce, Cisco, Google, Adobe, and CrowdStrike. The platform is hardware-agnostic by design, running on Nvidia, AMD, Intel, and other processors - a notable departure from Nvidia’s usual chip-centric strategy.
NemoClaw incorporates multi-layer security safeguards and privacy controls directly into its core, addressing the unpredictable behavior and data leakage risks that plagued early AI agent deployments. Because it’s open source, partners get free usage with early access in exchange for contributing to the project.
The agentic AI market is projected to reach $28 billion by 2027, and Nvidia clearly wants to own the infrastructure layer.
Thinking Machines Lab Partnership
The most concrete hardware commitment announced so far: a multiyear strategic partnership with Mira Murati’s Thinking Machines Lab to deploy at least one gigawatt of Vera Rubin systems for frontier model training.
That’s an extraordinary scale. For context, one gigawatt could power roughly 750,000 homes. Nvidia made a significant investment in Thinking Machines Lab (exact amount undisclosed), and the partnership includes designing training and serving systems optimized for Nvidia architectures.
Deployment is targeted for early 2027. Murati, formerly of OpenAI, founded Thinking Machines Lab to make AI systems “more widely understood, customizable and generally capable” - a mission that requires serious compute.
N1X: Nvidia Returns to Laptop CPUs
For the consumer market, watch for the N1 and N1X chips - Nvidia’s return to laptop CPUs after years away from the space.
According to Tom’s Hardware, Windows on Arm laptops using N1X are set to debut this quarter, with Dell and Lenovo among the launch partners. The N1X reportedly packs 20 cores, supports up to 128GB of unified LPDDR5X memory, and may include CUDA cores equivalent to an RTX 5070 GPU.
These will be Copilot+ PCs running Windows 11 on Arm, targeting gamers with a unified CPU/GPU package. The N1 targets mainstream laptops; the N1X is for professionals and enthusiasts.
DGX Spark: AI Supercomputer on Your Desk
The DGX Spark, first announced at CES 2026, is now shipping through Acer, ASUS, Dell, GIGABYTE, HPI, Lenovo, and MSI.
At $4,699 (up from launch price), the Spark delivers a petaflop of AI performance in a compact desktop form factor. It can run inference on models up to 200 billion parameters and fine-tune models up to 70 billion parameters locally.
The GB10 Grace Blackwell Superchip at its heart features fifth-generation Tensor Cores with FP4 support. Software updates have delivered up to 2.5x performance improvements through TensorRT-LLM optimizations and speculative decoding.
GTC includes hands-on “Build-a-Claw” sessions where attendees can create custom AI agents using the OpenClaw framework and deploy them on DGX Spark systems.
AI Inference: The Next Battleground
Beyond hardware announcements, the conference theme centers on AI inference at scale. Nvidia has been pitching the concept of “AI Factories” - data centers transformed into intelligence engines with high-performance training and low-latency inference.
Rumors suggest a dedicated inference processor targeting 10x lower inference costs for specialized agentic workloads. There’s also speculation about optical-compute chips or Co-Packaged Optics (CPO) switches to solve energy efficiency problems at gigawatt-scale AI factories.
Physical AI and Robotics
Sessions will cover advances in robotics and physical AI - how simulation, digital twins, and foundation models are moving systems from virtual training environments into real-world deployment. Nvidia has been building this capability for years through its Omniverse platform; GTC will show where it stands.
How to Watch
Jensen Huang’s keynote streams free at nvidia.com on Monday, March 16 at 11 a.m. PT. The pre-show “GTC Live” starts at 8 a.m. PT, hosted by Sarah Guo (Conviction), Gavin Baker (Atreides Management), and Alfred Lin (Sequoia Capital), featuring CEOs from Perplexity, LangChain, Mistral AI, Skild AI, and OpenEvidence.
The conference runs through March 19 with 700+ sessions, 150+ researcher poster presentations, and 70+ hands-on training labs.
What to Watch For
The hardware announcements will dominate headlines, but the real story is Nvidia’s expansion beyond chips into software platforms. NemoClaw represents a strategic shift - making Nvidia essential to enterprise AI even on competitors’ hardware.
If NemoClaw succeeds, Nvidia captures value at the application layer while maintaining its GPU dominance. If the N1X delivers on gaming performance claims, it challenges Qualcomm’s Windows on Arm lead. If Vera Rubin hits production targets, it maintains the multi-year performance advantage over AMD and Intel.
GTC has become the event that sets AI infrastructure direction for the year. This week will show whether Nvidia can maintain its position as the entire industry scales toward gigawatt-class compute.