At the GTC conference, Nvidia CEO Jensen Huang introduced the latest AI chips: Blackwell Ultra and Vera Rubin, designed to push the boundaries of AI performance.
Blackwell Ultra: More Power for AI Training & Inference
Set for release in the second half of 2025, Blackwell Ultra delivers up to 20 petaflops of AI power. With enhanced efficiency, it allows cloud providers to increase revenue by up to 50 times compared to previous generations.
Vera Rubin: The Next Leap in AI Processing
Launching in 2026, Vera Rubin is a new architecture integrating Nvidia’s first custom Vera CPU with the Rubin GPU. It boasts 50 petaflops of inference power, more than twice that of Blackwell Ultra, and supports up to 288GB of high-speed memory for larger and faster AI models.
Nvidia’s Vision for the Future
Huang emphasized the growing demand for AI computing, especially for agentic AI—a new era of autonomous systems. Nvidia expects its data center revenue to hit $1 trillion by 2028, further solidifying its dominance in AI infrastructure.
