Nvidia’s next AI architecture, Rubin, has begun development at Taiwan-based chipmaker TSMC. The company’s CEO, Jensen Huang, described Rubin as the “most advanced” architecture and stated that this project will be a major breakthrough in the computing world. Production of six new Rubin-based chips has begun.

Nvidia Rubin to be manufactured by TSMC

According to Huang’s statement during his visit to TSMC, Nvidia has put six different Rubin chips into production, including new CPUs and GPUs, the scalable NVLink Switch, and a silicon photonics processor. The Rubin architecture incorporates innovations across Nvidia’s entire technology stack. HBM4 memory provides a significant performance increase over the existing HBM3E standard.

This new architecture utilizes TSMC’s 3nm (N3P) manufacturing process and CoWoS-L packaging technology. Also notable is its chiplet design and 4x reticle approach, a first for Nvidia. This represents a significant improvement over the 3.3x reticle design in the previous Blackwell architecture. Experts say Rubin will represent a major generational leap, similar to the Hopper architecture.

Nvidia’s move with Rubin further intensifies competition in the AI ​​hardware market. The company’s significant investment in next-generation AI chips demonstrates its efforts to provide the high processing power required for future AI applications.

This development is particularly exciting news for large data centers and AI research institutes. The Rubin architecture is expected to be released in 2026-2027. This move demonstrates Nvidia’s continued commitment to consolidating its leadership in the industry.