#NvidiaGTC2026ConferenceBegins

NVIDIA GTC 2026 Conference Begins

The annual NVIDIA GTC conference officially kicked off on March 16, 2026, at the SAP Center in San Jose, California, transforming the venue into the epicenter of the artificial intelligence world. Drawing over 30,000 attendees, the four day event serves as a barometer for the entire tech industry, showcasing how AI is transitioning from digital novelty to physical infrastructure. NVIDIA founder and CEO Jensen Huang took the stage to a capacity crowd, delivering a more than two hour keynote that laid out an ambitious vision spanning next generation hardware, open source software, and the rise of physical AI.

The Dawn of the Inference Era and Trillion Dollar Ambitions

Huang opened the keynote by reflecting on the 20th anniversary of CUDA, the platform he described as the flywheel driving accelerated computing and supporting every phase of the AI lifecycle. He emphasized that the computing demand for NVIDIA GPUs is off the charts, noting a millionfold increase in demand over the last few years. This explosion is fueled by the shift from AI training to inference, where models are deployed to generate tokens and make decisions in real time. To underscore this transition, Huang boldly projected that NVIDIA expects to generate at least one trillion dollars in revenue from 2025 through 2027, driven by the insatiable need for AI factories worldwide.

Vera Rubin and Feynman: The Next Generational Leap

The centerpiece of the hardware announcements was the Vera Rubin platform, a full stack computing system designed for agentic AI. Unlike previous iterations, Vera Rubin is not just a chip but a complex system comprising seven chips, five rack scale systems, and a supercomputer architecture. The platform includes the new Vera CPU, which Huang claimed is fifty percent faster than traditional rack scale CPUs with twice the efficiency, making it ideal for large scale AI services and enterprise agents. Complementing the CPU is the Rubin GPU, built to deliver up to 10 times more performance per watt than its predecessor, Grace Blackwell.

In a move that surprised many, NVIDIA announced a strategic integration of Groq technology into its roadmap. The company unveiled the Groq 3 LPU, or Language Processing Unit, a specialized inference chip resulting from a twenty billion dollar licensing agreement with Groq. The LPU is designed to handle the decode phase of inference with ultra low latency, a task for which traditional GPUs can be less efficient. By combining the Rubin GPU for prefill and attention tasks with the Groq LPU for feed forward network decoding, NVIDIA claims throughput efficiency can improve by 35 times. The Groq 3 LPX Rack, housing 256 LPUs, will be deployed as part of the Vera Rubin rack scale system later this year.

Looking even further ahead, Huang previewed the Feynman architecture, named after the legendary physicist Richard Feynman. This next generation platform will feature the Rosa CPU, named in honor of Rosalind Franklin, alongside the LP40 GPU and advanced networking components like BlueField 5. Notably, Feynman will support both copper and co packaged optics for scale up connectivity, signaling that while optics are the future, copper remains a viable, simpler option for customers in the near term.

DLSS 5: The GPT Moment for Computer Graphics

Before the AI boom, NVIDIA was synonymous with gaming, and Huang paid homage to this heritage by announcing DLSS 5. Described as the most significant breakthrough in computer graphics since the invention of the programmable shader, DLSS 5 represents a fusion of traditional 3D rendering with generative AI. The technology uses a real time neural rendering model to infuse pixels with photorealistic lighting and materials, enabling 4K performance on local hardware. Huang called it the GPT moment for graphics, blending handcrafted rendering with AI to deliver a dramatic leap in visual realism while preserving artistic control. DLSS 5 is expected to ship in the fall, with support from major publishers like Ubisoft, Tencent, and Warner Bros Games.

NemoClaw and the Rise of Agentic AI

One of the most significant software announcements centered on the explosion of agentic AI. Huang praised OpenClaw, an open source project from developer Peter Steinberger, calling it the most popular open source project in the history of humanity. OpenClaw allows developers to build autonomous AI agents, or claws, that can interact with local file systems and data to execute complex tasks. However, the power of these agents raises security concerns, as they require deep access to user systems.

To address this, NVIDIA introduced NemoClaw, a software toolkit designed to run claws safely within enterprise contexts. NemoClaw acts as a policy engine, providing network guardrails, privacy routing, and a contained virtual environment to prevent agents from going rogue. This stack includes the NVIDIA OpenShell runtime, ensuring that as every company develops an OpenClaw strategy, they can do so securely. Huang predicted that this shift will transform every Software as a Service company into a GaaS, or Agent as a Service company.

Physical AI: Robots, Robotaxis, and Orbital Data Centers

Huang dedicated a significant portion of the keynote to physical AI, extending intelligence beyond the digital realm into machines that navigate the real world. He declared that the ChatGPT moment of self driving cars has arrived. NVIDIA announced a major expansion of its automotive partnerships, adding BYD, Hyundai, Nissan, and Geely to its robotaxi ready platform. In a groundbreaking collaboration with Uber, NVIDIA plans to launch a fleet of autonomous vehicles powered by its Drive AV software in 28 cities across four continents by 2028, with Los Angeles and San Francisco leading the way in 2027. These vehicles will leverage the new Alpamayo 1.5 model, a reasoning vision language action model that provides advanced navigation guidance.

Beyond transportation, NVIDIA is pushing the boundaries of computing off planet. In a visionary announcement, Huang revealed NVIDIA is going to space with the Vera Rubin Space 1 module. Designed for orbital data centers, these systems are engineered for size, weight, and power constrained environments. Working with partners like Starcloud and Axiom Space, NVIDIA aims to bring AI compute to satellites, enabling real time data processing in orbit. Huang acknowledged the engineering hurdles, such as radiation and the lack of convection for cooling, but framed this as the next frontier for edge AI.

To cap off the physical AI demonstrations, Huang brought out a surprise guest: Olaf, the snowman from Disney's Frozen, who walked onto the stage driven by NVIDIA's physical AI stack and the Newton physics engine. The whimsical demo underscored a serious point: the robots and characters of the future are being trained and simulated in NVIDIA Omniverse before they ever step into the physical world.

The AI Factory and Ecosystem Expansion

Throughout the keynote, Huang reinforced the concept of the AI factory, where data centers become token generation plants. He introduced the Vera Rubin DSX AI Factory reference design, which allows companies to deploy liquid cooled, high density racks at scale, turning raw power and data into intelligence. This infrastructure is the bedrock upon which the next generation of AI applications will be built. As GTC 2026 continues, the message from San Jose is clear: the era of physical AI and ubiquitous inference is not coming it is already here, and NVIDIA is providing the pickaxes and shovels for the gold rush ahead.