NVIDIA CEO and founder Jensen Huang.
NVIDIA
The AI Revolution is accelerating, driven by billion-parameter models of reasoning needed to develop physical and agentic AI. NVIDIA CEO Jensen Huang explained in his GTC keynote that the shift from training to full production inference is driving AI compute demand skyrocketing as data centers around the world transform into AI factories to efficiently and effectively process millions of user queries. NVIDIA unveiled at GTC major advancements to meet this $1 trillion opportunity – from the Blackwell Ultra AI Platform and an operating system designed for AI factories, to advancements in robotics, networking, and accelerated computing.
Blackwell has already been in full production and delivers an incredible 40x performance increase over Hopper. This architecture redefines AI model training and inference and makes AI applications more efficient. Blackwell Ultra, a powerful GPU with expanded memory that supports the next generation AI models, will be available in the second half of 2025.
NVIDIA is committed to a yearly AI architecture refresh. NVIDIA Vera Rubin was designed to boost AI data center performance.
Beyond the GPU, AI infrastructure is undergoing an unprecedented shift, with innovations in photonics and AI-optimized data storage, as well as advanced networking. These breakthroughs will enhance scalability and efficiency across massive AI data centres.
According to Huang, physical AI in robotics and industry represents a $50 trillion opportunity. AI-powered automation will reshape industries in manufacturing, logistics, healthcare and more. NVIDIA’s Isaac and Cosmos platforms will be at the forefront of the next era in AI-driven robotics.
NVIDIA Roadmap : The NVIDIA roadmap includes Vera Rubin which will be released in the second quarter of 2026. Vera Rubin Ultra will follow in 2027. The Rubin servers and chips boast improved speeds in data transfers, which is critical for large AI systems that have many chips. The next architecture, Feynman,is scheduled to be released in 2028, and will use next-gen HBM Memory.
DGX Personal AI Computers: Powered on the NVIDIA GRACE BLACKWELL platform, DGX Spark, and DGX Station, are designed to develop large models, fine-tune them, and infer their conclusions. These computers will be produced by several companies, including ASUS Dell and HP.
Spectrum X AND Quantum X( )[]networking[ ]platforms: Silicon photonics networking switch helps AI factories connect millions GPUs across multiple sites and reduces energy consumption dramatically. Quantum-X Photonics InfiniBand switch will be available in the second half of this year. Spectrum-X Photonics Ethernet switch will arrive in 2026. Dynamo Software Open-source Dynamo, released for free, helps speed up the process of multistep reasoning in AI factories, improving efficiency and reducing the time to innovation. NVIDIA Accelerated Quantum Research Center : A Boston-based center will provide cutting edge technologies to advance quantum computation in collaboration with leading software and hardware makers.
NVIDIA ISAC GR00T N1:A foundational model of humanoid robotics, the GR00T N1 model is the first fully customizable, open foundation model in the world for generalized humanoid skills and reasoning. It has a dual-system similar to reasoning models for both fast and slower thinking.
Newton Physics Engine : NVIDIA announced a collaboration between Google DeepMind, Disney Research and NVIDIA to develop Newton. This open-source physics tool allows robots to learn how to perform complex tasks more accurately.
Here are the highlights: Don’t miss NVIDIA’s live blog for the complete GTC recap
A sponsored article is content that has been produced by a company who either pays for the post or VentureBeat’s business relationship. They are always clearly marked. For more information, contact
