NVIDIA founder Jensen Huang launched the company’s AI developers meeting on Tuesday and told thousands of people that AI is experiencing a “turning point.”
At GTC 2025 (known as “The Super Bowl of AI”), Huang focused his keynote on the company’s advancements in AI and its predictions about how the industry will grow in the next few years. He said demand for GPUs from the top four cloud service providers is surging, adding that he expects NVIDIA’s data center infrastructure revenue to reach $1 trillion in 2028.
Huang’s highly anticipated announcement reveals more details about NVIDIA’s next-generation graphics architecture: Blackwell Ultra and Vera Rubin, named after famous astronomers. Blackwell Ultra is scheduled to be launched in the second half of 2025, and its successor, Rubin AI chip, is expected to be launched by the end of 2026. The Rubin Ultra will take office in 2027.
During his speech that lasted more than two hours, Huang outlined the “extraordinary progress” made by AI. He said that over a 10-year period, AI graduated from generated AI from perception and “computer vision” and is now AGIC AI or AI with reasoning capabilities.
“AI understands the context, understands our requirements,” he said. “Answers can now be generated.”
He said the next wave of AI is happening: robotics.
He said robots driven by so-called “physical AI” can understand concepts such as friction and inertia, causality, and object permanence.
“Each of these phases opens up new market opportunities for all of us,” Huang said.
The key to this physical AI, as well as many other announcements by Huang, is the concept of model training using synthetic data generation (AI or computer-created data). He said AI requires digital experiences to learn, and it allows humans to learn at the speed of humans in the training cycle.
“There is only too much data and we can do too much human demonstration,” he said. “This is a big breakthrough in the past few years: strengthening learning.”
He said NVIDIA’s technology could help AIs when they attack or try to solve problems step by step.
To this end, Huang announced the Isaac GR00T N1, an open source basic model designed to help develop humanoid robots. The ISAAC GR00T N1 will be paired with the updated COSMOS AI model to help develop simulation training data for the robot.
Benjamin Lee, a professor of electrical and systems engineering at the University of Pennsylvania, said the challenge of training robotics is data collection because real-world training is time-consuming and expensive.
Simulated environments have long been the standard for enhanced learning, so researchers can test the effectiveness of their models, Lee said.
“I think it’s really exciting,” Lee said. “More and more researchers can start using this comprehensive data, not only large players in the industry, but also academic researchers.” ”
Huang introduced the Cosmos series AI model that can generate cost-effective photo realistic videos and then train robots and other automation services at CES earlier this year.
An open source model, a physics simulation tool, in collaboration with NVIDIA’s Omniverse, can create more realistic videos and promise much cheaper than traditional collection training formats, such as having cars record road experiences or having people teach robots to repetitive tasks.
Huang said U.S. automakers generally plan to integrate NVIDIA technology into their new self-driving cars. The two companies will work together to use Omniverse and Cosmos to train AI manufacturing models to build custom AI systems.
The NVIDIA head also launched the company’s Halos System, an AI solution around the safety of cars (especially autonomous driving).
“I believe we are the first company in the world to have all the code security,” Huang said.
At the end of the speech, Huang is an open source physics engine for robot simulation, called Newton, which is being developed by Google DeepMind and Disney Research.
A small robot named Blue joins him on the stage and pops out of a hatch on the floor. It beeps to Huang, following his commands, standing next to him as he stands at the thoughts of him.
“The era of generalist robotics is here,” Huang said.
