
Prime Highlights:
Nvidia shifts to an annual chip release cadence, moving away from its previous two-year cycle, beginning with the Blackwell Ultra and Vera Rubin announcements.
Vera, Nvidia’s first custom-designed CPU, will deliver twice the performance of the previous Blackwell chips, tailored to support the new Rubin GPU.
Key Background:
At its annual GPU Technology Conference (GTC), Nvidia made significant announcements regarding the future of artificial intelligence (AI) hardware. CEO Jensen Huang introduced the Blackwell Ultra family of chips, which will be released later this year, and unveiled Vera Rubin, the company’s next-generation graphics processing unit (GPU), slated for launch in 2026.
These new chip announcements mark Nvidia’s shift to an annual release cycle, moving away from the previous two-year cadence. Blackwell Ultra chips, an upgrade to the existing Blackwell family, are designed to handle time-sensitive AI applications, offering cloud providers the ability to deliver premium AI services. These chips can generate more content per second, increasing efficiency and providing up to 50 times the revenue potential compared to the previous Hopper generation. Blackwell Ultra will be available in various configurations, including a version paired with a Nvidia Arm CPU and others with multiple GPUs in a single server.
The Rubin GPU, named after astronomer Vera Rubin, represents Nvidia’s first custom-designed CPU, called Vera, which will be paired with the next-gen Rubin GPU. The Rubin system will deliver double the performance of current Blackwell chips, with a 50-petaflop capacity for inference and support for up to 288 gigabytes of memory. Rubin will also introduce a new chip design, utilizing multiple dies, with plans to expand this further with a “Rubin Next” chip in 2027.
These announcements come amidst a surge in Nvidia’s sales, largely driven by the AI boom sparked by OpenAI’s ChatGPT. Nvidia’s chips have become essential for training and deploying AI models, securing its position as a key player in the industry. The company’s focus on improving performance and efficiency with these new chips aims to maintain its strong market position, as cloud companies such as Microsoft, Google, and Amazon continue to build data centers based around Nvidia’s technology. Nvidia’s commitment to innovation was further highlighted by plans for future chip families, including a new series named after physicist Richard Feynman, slated for release in 2028.