Nvidia GTC: Jensen Huang announces new AI chips, deal with GM and robot partnerships

Blackwell Ultra will be released this year and Vera Rubin chips in 2026

Jensen Huang, chief executive of Nvidia, announces a new partnership with GM during the company's GTC conference. Photo: Bloomberg

Nvidia chief executive Jensen Huang on Tuesday introduced new chips and spoke of his vision for Agentic AI, autonomous vehicles, robotics and accelerated computing during his company's much-anticipated GTC conference.

"They used to call this the Woodstock of AI, and now some have called this the Super Bowl of AI. The only difference is that unlike the Super Bowl, everybody wins here," Mr Huang told those attending.

In his trademark leather jacket, he addressed the packed SAP Centre in San Jose, California, detailing to the audience how he planned to keep his company's AI-fuelled momentum going.

Nvidia chief executive Jensen Huang speaks at Nvidia GTC. AP

He began by talking about what he believes is the coming wave of AI technology, Agentic AI, which can be defined as an artificial intelligence implementation that has agency and can also reason and take action to help humans with day-to-day tasks.

"We're talking about trillions and trillions of tokens to train those models," he said, highlighting the need for increased computing power that he believes Nvidia is ready to deliver.

Mr Huang also touched on transport, saying that his company has been helping with self-driving car research for more than a decade. He announced that Nvidia would be teaming up with General Motors to help the company build its fleet of self-driving cars.

"The time for autonomous vehicles has arrived," he said. "We're looking forward to building with GM AI in all three areas … AI in manufacturing so they can revolutionise how they manufacture, AI in enterprise so they can revolutionise how they work, and then of course AI in the car. I'm super excited about this."

Mr Huang said that, for the first time, Nvidia was dedicating an entire day of the multi-day conference to the growing field of quantum computing.

Unlike conventional computers, which process data in bits and have a binary value of zero or one, quantum computers can process digits simultaneously using a two-state unit called a qubit. That means incredibly complex computations can be done in a fraction of the time.

“We’re working with just about everybody in the ecosystem, helping them research on quantum architectures, quantum algorithms and in building accelerating quantum heterogeneous architectures,” Mr Huang said.

A little more than an hour into the address, the Nvidia founder announced what many in the audience had been waiting for: the company's latest AI chips and graphics processing units (GPUs).

Nvidia's Jensen Huang speaks on the company's Vera Rubin chips, scheduled for release in 2026. AP

Nvidia's Blackwell Ultra, a family of chips designed for Agentic AI and Physical AI, will begin to be released towards the middle of this year.

“AI has made a giant leap. Reasoning and Agentic AI demand orders of magnitude more computing performance,” said Mr Huang. “We designed Blackwell Ultra for this moment – it’s a single versatile platform that can easily and efficiently do pre-training, post-training and reasoning AI inference.”

He announced that the company's coming advanced chip family, Rubin, named after the American astronomer Vera Rubin, would be available in the middle of 2026.

Mr Huang also gave a tantalising glimpse at 2028, when he said that the company's Feynman chips, which will have unprecedented power and capability, would begin to be released. The chips are named after American physicist Richard Feynman.

The Nvidia founder wrapped up his speech by talking about what he described as Physical AI – where AI understands spatial relationships and the physical behaviour of humans – and how it would merge with robotics research.

"It is moving so fast and everybody should pay attention to this space because this could be the largest industry of them all," Mr Huang said, before announcing Newton, a partnership between Google's DeepMind, Disney and Nvidia that helps to develop robots with an open-source physics engine.

He took a robot on stage to display some of Newton's abilities.

Although it is still riding a wave of success, this GTC comes at a time when some technology and business analysts are starting to ask questions about potential vulnerabilities faced by Nvidia.

Those vulnerabilities mainly come from DeepSeek, a China-based company that claims to have created an AI platform without the need for the GPU-heavy infrastructure that helps to increase Nvidia's bottom line.

Jensen Huang said Nvidia will help General Motors to develop self-driving vehicles and work with the telecoms industry on new 6G networks. Bloomberg

Shortly after DeepSeek made its debut in late January, downloads of the AI chatbot soared as the reported efficiency of the app sent many US technology and energy stocks down.

Yet in recent weeks, Nvidia has not been shy about embracing DeepSeek to an extent, issuing various press releases showing how Nvidia hardware works with the relatively new Chinese-made AI platform.

Other speakers at Nvidia's GTC conference include Meta's chief AI scientist Yann LeCun, Rivian founder and chief executive RJ Scaringe, Mistral AI chief executive Arthur Mensch and OpenAI research scientist Noam Brown.