The AI Super Bowl is here! Jensen Huang unveils the next-generation Blackwell | WPN World People News

NVIDIA CEO Jensen Huang took the stage at the GTC Technology Conference in the early hours of March 19 (Taiwan time) to discuss the development path of artificial intelligence (AI) and unveil a new lineup of chips. Huang described this year’s GTC as the “Super Bowl of AI,” emphasizing that in this incredible AI competition, everyone is a winner.

GTC originated from GeForce, NVIDIA’s personal computer graphics processing unit (GPU). GeForce has been on the market for 25 years and has completely sold out worldwide. Not only has GeForce driven the development of CUDA technology, but it has also enabled revolutionary breakthroughs in AI within the GPU field.

Huang noted that AI has made remarkable progress in just 10 years, with the past five years focused on generative AI, which transforms text into images, videos, and more. The latest advancement, Agentic AI, not only understands problems and performs reasoning but can also take action.

At the GTC Technology Conference, Huang introduced the next-generation NVIDIA Blackwell Ultra GPU. This chip enhances the inference scalability during both the training and testing phases by allocating more computational resources to improve accuracy. This technology will help industries worldwide accelerate AI-driven inference applications. He emphasized that the computational performance of inference and Agentic AI is several orders of magnitude higher than current levels. The Blackwell Ultra is an upgraded version of NVIDIA’s next-generation Blackwell GPU.

“AI has already achieved extraordinary progress,” Huang stated. NVIDIA’s data center construction spending is estimated to reach $1 trillion by 2028. NVIDIA plans to launch the Blackwell Ultra chip in the second half of 2025 to pave the way for the AI inference era. In addition to the Blackwell Ultra chip, the highly anticipated GB300 series chips were also unveiled at GTC.

Huang further announced that the next-generation AI superchip, Vera Rubin, will be launched in the second half of 2026, followed by the Vera Rubin Ultra in the second half of 2027. Reports indicate that Blackwell’s performance is 68 times that of Hopper, while the Vera Rubin Ultra, which integrates four GPUs, boasts performance up to 900 times that of Hopper.