NVIDIA's next AI GPU architecture is codenamed Rubin, to be available by 2026

While the world is still fascinated by NVIDIA Blackwell, the company is reportedly working on the next-gen AI GPU architecture, codenamed Rubin. Designed to deliver superior performance at lower power consumption, NVIDIA Rubin gives us a sneak peek into the company’s future AI-focussed GPU plans.

The developments have been shared by Mich-Chi Kuo from TF International Securities, according to TechPowerUp.

Mich-Chi Kuo tells us that the first AI GPU based on the Rubin architecture, named R100, will enter mass production by Q4 2025 and should be available for evaluations as early as Q1 2026. However, we may have to wait up to a year before the R100 is available on a mass scale.

As per the report, the R100 will use the TSMC-N3 node as opposed to Blackwell’s TSMC-N4P. The AI GPU will feature a 4x reticle design with TSMC’s CoWoS-L (Chip-On-Wafer-On-Substrate-L) packaging.

The design of the Rubin-based R100 revolves around lowering power consumption, which has become one of the biggest concerns for enterprises scaling up their operations. Reports suggest that ChatGPT consumes 17,000 times the energy compared to a US household. As for Microsoft AI, its water consumption is more than 670 families a year.

Efficient AI-focussed GPUs are the need of the hour, and NVIDIA’s latest Rubin architecture used in R100 is certainly a step in the right direction. We can’t comment on the performance until the GPU is available for initial evaluation. And that’s roughly two years from now!

What are your first thoughts about NVIDIA Rubin? Share with our readers in the comments section.