Summary
Highlights
Jensen Huang welcomes attendees to GTC, highlighting its focus on technology and platforms, including CUDA-X, accelerated systems, and the new AI factories. He emphasizes NVIDIA's rich ecosystem, with 450 sponsoring companies and a thousand technical sessions covering every layer of AI development, from hardware to applications. The 20th anniversary of CUDA is celebrated, underscoring its long-standing architectural dedication to accelerating computing.
Huang details the 'CUDA flywheel,' explaining how its vast installed base drives developer engagement, leading to breakthroughs like deep learning and new markets. This continuous cycle accelerates growth and reduces computing costs. He traces NVIDIA's journey from GeForce 25 years ago, which led to the programmable shader and ultimately CUDA, supporting the rise of AI. The presentation introduces neural rendering, a fusion of 3D graphics and AI, showcasing DLSS 5 as an example of combining structured data with generative AI for realistic and controllable content.
Huang explains the integration of structured data (like SQL data frames) with unstructured data (such as PDFs and videos) using AI. He announces NVIDIA's foundational libraries, cuDF for structured data and cuVS for unstructured data, which are crucial for accelerating data processing. Partnerships with IBM (Watsonx.data), Dell (AI data platform), and Google Cloud (BigQuery) demonstrate significant speedups and cost reductions, showcasing the real-world impact of accelerated computing on enterprise data.
Huang outlines NVIDIA's unique strategy: vertical integration complemented by horizontal openness. This approach allows NVIDIA to deeply understand applications and algorithms across diverse verticals like automotive, financial services, healthcare, industrial, and telecommunications. By developing domain-specific libraries and working with a broad ecosystem, NVIDIA ensures its technology can be deployed anywhere, from data centers to the edge, making it a foundational platform across all AI life cycles.
Huang discusses the three key catalysts of the recent AI surge: generative AI (ChatGPT), reasoning AI (O1), and Agentic models (Claude Code). He highlights the exponential increase in computing demand, estimating a 1 million-fold rise in the last two years. This surge signals the arrival of the 'inference inflection,' where AI's ability to think, reason, and act drives massive computation. Huang projects a minimum of $1 trillion in demand for NVIDIA’s Blackwell and Rubin platforms through 2027, driven by the need for efficient token generation.
Huang introduces the Vera Rubin platform, emphasizing its design for Agentic AI, with innovative features like NVLink 72, NVFP4 precision, and significant performance-per-watt improvements. He details the collaboration with Groq, integrating their LPUs to create a disaggregated inference system that optimizes for both high throughput and low latency, dramatically increasing token generation speeds. This co-design approach ensures industry-leading performance and cost efficiency for AI factories globally.
NVIDIA introduces its DSX platform, an Omniverse-based digital twin blueprint for designing and operating AI factories. This platform enables virtual collaboration for mechanical, thermal, electrical, and network simulations, ensuring maximum token throughput and energy efficiency. It integrates with industry tools and partners like PTC, Dassault Systèmes, and Siemens, making it possible to build and operate complex AI infrastructures at a planetary scale.
Huang announces NVIDIA's support for OpenClaw, an open-source project enabling the creation of personal AI agents, likening its impact to that of Linux or HTML. NVIDIA introduces NemoClaw, an enterprise-ready reference design for OpenClaw, incorporating security and privacy features. He then showcases NVIDIA's frontier open models, including Nemotron for language, Cosmos for physical AI, GR00T for general-purpose robots, BioNeMo for biology, and Earth-2 for climate forecasting, all designed to enable specialized sovereign AI.
The keynote culminates with a focus on physical AI and robotics. Huang announces new partnerships for NVIDIA's robotaxi-ready platform with major automotive manufacturers (BYD, Hyundai, Nissan, Geely) and Uber. He highlights the role of simulation with Isaac Lab and Newton physics solver in training robots, showcasing examples like Peritas AI and Skild AI. The presentation concludes with a live demonstration of a walking Disney robot (Olaf) powered by NVIDIA's physical AI, emphasizing the seamless integration of AI into physical embodiments and underscoring the massive potential of AI to transform diverse industries.