Nvidia (NVDA) chief executive Jensen Huang shared updates on the chipmaker’s next artificial intelligence chips as part of his announcements at the company’s GPU Technology Conference on Tuesday.
Huang also discussed the company’s work on reasoning models, agentic AI, photonics, and robotics at the annual developer conference, also known as the GTC.
Here are some takeaways from Huang’s keynote.
Huang said Nvidia’s Blackwell chips — which he announced at last year’s GTC — are now in full production.
The next Blackwell Ultra NVL72 chips, which have one-and-a-half times more memory and two times more bandwidth, will be used to accelerate building AI agents, physical AI, and reasoning models, Huang said. Blackwell Ultra will be available in the second half of this year.
Nvidia’s next-generation GPUs named after astronomer Vera Rubin will be released in the second half of next year, Huang said. Rubin has a new CPU that has twice the performance of its previous chip, as well as more memory and bandwidth.
The second half of 2027 will bring Rubin Ultra, Huang said, which will contain multiple GPUs connected together.
Huang said Nvidia’s next architecture will be named Feynman, and is coming in 2028.
Nvidia Isaac GR00T N1 is the first open and fully customizable AI foundation model for building humanoid robots in the world, Huang said.
The company will pre-train GR00T N1 on synthetic and real data, and release the open family of models to robotics developers.
Huang unveiled Nvidia Dynamo — “essentially the operating system of an AI factory” — to accelerate and scale reasoning models efficiently.
He also announced the Llama Nemotron family of reasoning models which will enable developers and enterprises to build AI agents. The new reasoning models are open.
Huang introduced Nvidia DGX personal AI supercomputers, which are powered by the company’s Blackwell AI platform. DGX Station and DGX Spark, formerly known as Project DIGITS, will enable everyone from developers to students to prototype, fine-tune, and run AI models from a personal desktop.
