NVIDIA Seeks 30% of TSMC's 3nm Output for Rubin AI Chips
Huang mentioned on his visit to Taiwan that the company is negotiating for about 30% of TSMC's future output for use in producing the firm and performance hungry Rubin AI products. Report suggest that NVIDIA is moving for a considerable chunk of TSMC's 3nm production capacity. Huang is at Taiwan to get first-hand information from TSMC on the current status of the 3nm fab line within the Tainan site.
With the potential rise in the number of productive 3nm wafers at the Southern Taiwan Science Park from 100,000 to 160,000 wafers per month, a significant portion is expected to be kept for NVIDIA alone; a fact reflecting the confidence of future demand on its Rubin AI chips.
Pragmatic to Rubin's Current Architecture Announcement
As the driving platform for NVIDIA into the next generation of supercomputing, Rubin AI promises to be a significant compute leap. Key technological features of the Rubin architecture include:
- Using TSMC's N3P Process, which employs advanced 3nm-class manufacturing node.
- Integrating the next generation of High Bandwidth Memory for enhanced performance.
Currently, interest in the platform is so high that according to reports, customers are almost signed up for the Rubin lineup in advance of mass production.
Importance of NVIDIA-TSMC Partnership
Such negotiations strongly underline the exceptionally vital tie that exists between NVIDIA and TSMC. NVIDIA has been instrumental in bringing High-Performance Computing (HPC) customers, a major chunk of TSMC's business. This is the reason why Jensen Huang keeps emphasizing on such ties during his visits to Taiwan, given that both companies tend towards future process technologies. For example, TSMC is planning to develop the A16 (1.6nm) node.
