Nvidia Corp (NASDAQ:NVDA, ETR:NVD) kicked off its GTC 2025 conference with a flurry of announcements, unveiling its latest AI hardware innovations, expanding its automotive partnerships, and setting ambitious revenue targets for the future.
The firm introduced the Blackwell Ultra, a successor to its Blackwell AI chip, touting a 1.5x performance increase and a 50x data center revenue opportunity compared to its Hopper chip. The company emphasized the chip’s capabilities for AI reasoning, a form of AI processing designed to mimic human thought patterns.
Blackwell Ultra will integrate into Nvidia’s NVL72 rack server, featuring 72 GB300 superchips for improved efficiency. Nvidia claims the system can process 1,000 tokens per second when running DeepSeek’s R1 model, a tenfold increase over Hopper’s 100 tokens per second. This translates to response times dropping from 1.5 minutes to just 10 seconds.
Vera Rubin
Nvidia also provided a first look at its next-generation Vera Rubin AI superchip, which is expected to launch in late 2026. The chip, like its predecessor Grace Blackwell, features a combined CPU (Vera) and GPU (Rubin) architecture.
Key upgrades include 4.2 times more memory than Grace, 2.4 times higher memory bandwidth, 88 CPU cores delivering twice the overall performance of Grace Blackwell, and 288GB of high-bandwidth memory 4 (HBM4) in the Rubin GPU.
Beyond Vera Rubin, Nvidia teased its 2027 follow-up, the Vera Rubin Ultra, which will feature Rubin Ultra GPUs that combine four GPUs into a single unit.
Automotive AI with GM
In a major partnership announcement, General Motors (GM) will integrate Nvidia’s DRIVE AGX hardware into its future autonomous and driver-assistance systems.
“The time for autonomous vehicles has arrived,” said CEO Jensen Huang. “We’re looking forward to building with GM AI across manufacturing, enterprise, and in-car AI.”
GM will also leverage Nvidia Omniverse and Nvidia Cosmos to train AI models for factory planning and robotics.
Feynman and the road ahead
Huang emphasized that AI is at an “inflection point,” predicting $1 trillion in data center revenue by 2028. He also announced that Nvidia’s next-generation AI chip, codenamed “Feynman,” will debut in 2028.
With rapid advancements in AI hardware and expanded industry partnerships, Nvidia positioned itself at the forefront of the AI revolution, setting the stage for what Huang called a fully AI-assisted future for software engineers.
Shares of Nvidia were down around 2.6% following Huang’s presentation at the conference.