Rubin was first introduced at GTC 2025 in Washington. Then, the head of the company, Jensen Huang, showed the Vera Rubin Superchip, a solution combining two powerful GPUs, a new "Vera" CPU, and LPDDR memory located around the structure. This chip is being created for the largest data centers, where the need for AI computing is rapidly growing.
After Huang's visit to the TSMC factory, it was confirmed that Rubin has already been launched on the production lines. Until recently, only test samples existed, which indicates an extremely fast pace of development. Against the background of record demand for Blackwell and Blackwell Ultra, TSMC increased production using the 3nm process by 50%.
In parallel, NVIDIA is testing HBM4 memory, ordering samples from SK Hynix, Micron, and Samsung to avoid supply shortages. SK Hynix became the first company to start supplying HBM4 and once again outstripped competitors.
Mass production of the Rubin GPU is scheduled for the third quarter of 2026, but may begin earlier. The Rubin architecture will become the new base for AI accelerators and is already associated with a large-scale contract between NVIDIA and OpenAI to supply chips for future data centers.
Rubin, according to NVIDIA, will become the "new standard" for AI infrastructure and the most powerful GPU in the company's history.