Nvidia (NVDA, Financial) CEO Jensen Huang kicked off GTC 2025 by calling it the "Super Bowl of AI." He emphasized how AI is evolving fast, with agentic AI—systems that can make independent decisions—demanding 100 times more computing power than experts predicted last year.

To meet this surge, Amazon Web Services, Google Cloud, Azure, and Oracle Cloud have already snapped up 3.6 million Blackwell GPUs, far surpassing last year’s 1.3 million Hopper GPUs. Huang expects data center investments to top $1 trillion by the decade’s end.

Next-Gen Chips: Blackwell Ultra, Rubin, and Feynman

Huang confirmed Blackwell Ultra is coming later in 2025, offering twice the memory and bandwidth of Blackwell. Rubin will launch in 2026, followed by Rubin Ultra in 2027, which boasts 15 times more inferencing power. The next big leap? Feynman, expected in 2028.

Major Partnerships and AI Software

Nvidia is teaming up with Cisco (CSCO, Financial), T-Mobile (TMUS, Financial), and Cerberus ODC to build radio networks in the U.S. Meanwhile, General Motors (GM, Financial) has tapped Nvidia to power its self-driving car fleet.

Huang also introduced Dynamo, a system designed to optimize AI inferencing performance, calling it the “operating system of an AI factory.”

New Networking and Optical Tech

Nvidia launched the Spectrum-X "Supercharged" Ethernet switch, arriving in 2026, and Quantum-X silicon photonics, built with Taiwan Semiconductor (TSM), rolling out in 2025.

With AI evolving rapidly, Nvidia is making sure it's leading the charge in computing, networking, and automation.