Nvidia explains its roadmap for AI, reveals name of next-gen architecture
Table of Contents
Nvidia has been at the forefront of the seemingly unstoppable rise of AI technology. The GTC keynote speech by Jensen Huang indicates that Team Green wants to continue to be at the cutting edge of AI development with the reveal of its AI roadmap.
Despite the issues with the launch of the Blackwell series, with even the commercial and business side seeing shipping delays and low yield rates, Blackwell has seen huge adoption. Nvidia CEO Jensen Huang was firm in his belief that this was only the beginning.
Nvidia reveals roadmap for Blackwell, Rubin, and Feynman
The GTC keynote set out figures that showed 3.6 million Blackwell GPUs have been shipped to the top for CSPs in the USA. This does not include start-ups and other smaller AI companies. This equates to a threefold increase in sales figures compared to the ‘Hopper' GPUs
Huang's keynote speech stated that AI computing was at an “inflection point” and the demand for hardware was set to grow rapidly. Nvidia predicts that data center builds will reach over $1 Trillion in expenses by 2027, almost four times more than current figures.
Nvidia plans to capitalize on this vast surge in demand and already has plans for two generations of new GPU architecture. As of 2026, Nvidia will be upgrading Blackwell into Blackwell Ultra, which will offer up to 288 GB of HBM3e memory. After that, Nvidia revealed the Vera Rubin system, which is set to launch in the second half of 2026.
It seems Team Green has some very long-term aspirations, however, and is already laying the groundwork for the architecture after that. This has been called Feynmen after a pioneer in the field of quantum mechanics. Like the Vera Rubin, the Feynman will use the advanced HBM memory
The full specs of these new GPU platforms have not yet been revealed, and it is unknown what plans Nvidia currently has for the consumer GPUs.
About the Author