Global AI Computing Power Shortage: NVIDIA Relies on Taiwan Ecosystem to Scale Production

(Central News Agency, Reporter Wu Jiahao, Taipei, June 3) AI chip giant NVIDIA has begun full-scale mass production of its new-generation architecture, Vera Rubin. Shar Narasimhan, Director of AI and Data Center GPU Product Marketing at NVIDIA, said in an interview today that market demand for AI computing power has far exceeded the total computing capacity currently available. He stated that NVIDIA needs the assistance of the Taiwan ecosystem to expand production capacity to the mass production stage in order to supply global AI factories.

Speaking to Taiwanese media at the GTC Taipei exhibition area today, Narasimhan pointed out that the Taiwan ecosystem has been a great partner, walking alongside NVIDIA from the very beginning, and has enabled many incredible technologies that allow NVIDIA to scale the new systems on display.

He emphasized that the Taiwanese supply chain possesses two core strengths: being 'hard-working' and having 'very advanced technology,' making it an indispensable partner. NVIDIA has also significantly simplified the design of the Vera Rubin platform. A manufacturing step that took 2 hours on the previous Blackwell architecture can now be completed 20 times faster on the Vera Rubin platform, greatly improving production throughput.

Narasimhan stated that the evolution of AI technology is driving a transformation in hardware architecture. To meet the workload demands of Agentic AI, every inference must be verified. To this end, NVIDIA has equipped the new platform with an independent central processing unit (CPU) featuring 88 cores, specifically designed to handle tool invocation and answer verification, and to send feedback in real-time to the graphics processing unit (GPU), forming a more efficient computing synergy system.

Regarding NVIDIA CEO Jensen Huang's famous quote evolving from 'The more you buy, the more you save' three years ago to today's 'The more you buy, the more you earn,' Narasimhan offered his own interpretation.

He analyzed that these two phrases represent the combined benefits of revenue, profit, and cost reduction. For current AI factories, there is a need for the Vera Rubin platform to achieve the lowest Token cost on one hand, and the highest Token throughput on the other.

'In the current market, anyone with Tokens to sell is in a state of supply shortage,' Narasimhan said. He added that global demand for AI computing power has far exceeded the total computing capacity currently available, which is why NVIDIA needs the Taiwan ecosystem's help to scale production capacity to the mass production stage to supply global AI factories. (Editor: Zhang Liangzhi) 1150603