NVIDIA plans to launch four new AI chips annually, including Feynman, from the latter half of this year through 2028. The company also introduced Halos, its autonomous driving system, at the event.
On Tuesday, NVIDIA held its annual developer conference, GTC 2025, in San Jose, California, reaffirming its commitment to becoming a full-scale AI enterprise. CEO Jensen Huang stressed the rapid advancements in AI over the past year, building on the company’s previous vision.
Opening his two-hour keynote, Huang said, “If last year’s GTC was a pop concert, this year’s event felt like the Super Bowl of AI.”
He projected that NVIDIA’s data center infrastructure revenue will reach $1 trillion by 2028, driven by surging GPU demand from major cloud service providers.
Huang also announced the upcoming release of Blackwell Ultra this fall, which will replace the company’s top-tier AI chip. He outlined a roadmap featuring Vera Rubin next year, its enhanced version Vera Rubin Ultra in 2027, and Feynman in 2028. Vera Rubin is expected to deliver three times Blackwell’s performance.
In addition, Huang revealed a partnership with General Motors to supply AI-based technologies for autonomous vehicles. He also announced collaborations with Disney, Google DeepMind, and Newton to advance research in NVIDIA’s physical AI systems.
Huang explained the concept by stating that Physical AI will enable robots to perceive and reason about their environment through slow thinking, while fast thinking capabilities will drive action.
The event also unveiled new AI-powered laptops and desktops, including the DGX Spark and DGX Station, designed to run large-scale AI models like Meta’s Llama and DeepSeek.
Moreover, Huang introduced updates to networking components, enabling hundreds or thousands of GPUs to operate as a single unit. He also showcased Dynamo, a software package designed to optimize chip utilization.
