NVIDIA’s Soaring AI Chip Demand Forces TSMC to Increase 3nm Production by 50%, Ready for Next-Gen Rubin Series

Jensen Huang, CEO of NVIDIA, is currently visiting Taiwan, where he aims to secure a substantial portion of TSMC’s 3nm chip production tailored for AI applications. This meeting emphasizes NVIDIA’s commitment to meeting the soaring demand for its next-generation products, particularly those in the Vera Rubin AI lineup.

NVIDIA’s Strategic Moves to Capture TSMC’s 3nm Production

The ongoing AI surge is propelling NVIDIA into a unique position, particularly as it ramps up production of its Blackwell Ultra chips. The increased demand for high-performance semiconductors is evident. As noted in a report by UDN, Huang’s visit to TSMC’s fabrication plant in Tainan centers around negotiations to increase the capacity allocation for TSMC’s N3 process. This move is crucial to ensure there’s enough supply to satisfy the anticipated rush for chips intended for the upcoming Vera Rubin lineup.

Reportedly, TSMC plans to elevate its 3nm fabrication capacity at the Southern Taiwan Science Park from 100, 000 wafers per month to approximately 160, 000 wafers—a nearly 50% increase. Notably, a significant portion of this enhanced capacity is anticipated to be allocated exclusively to NVIDIA. This aligns with NVIDIA’s belief in the colossal demand for its Rubin AI chips, prompting early supply reservation to accommodate future needs. For TSMC, this increase in 3nm production is expected to significantly drive revenue for multiple forthcoming quarters.

NVIDIA circuit board displayed on stage shows TWW 2538 on chips. — NVIDIA’s Vera Rubin Superchip

The Rubin AI series is poised to be a transformative leap for NVIDIA, introducing an architecture that promises comprehensive enhancements and a significant boost in computational power. In addition to TSMC’s advanced N3P process, the Rubin lineup will integrate cutting-edge HBM4 technology, elevating the overall performance capabilities of its products. Impressively, NVIDIA has already lined up customers for the Rubin chips, indicating strong market interest even before mass production officially begins in about two quarters.

For TSMC, high-performance computing (HPC) clients like NVIDIA contribute considerably to its revenue stream. The company is actively implementing newer manufacturing processes and rapidly advancing towards the adoption of A16 (1.6nm) technology. This ongoing evolution underscores the critical nature of TSMC’s partnership with NVIDIA, as repeatedly highlighted by Jensen Huang during his visits to Taiwan.

Source & Images