TSMC executive outlines seven actions for Taiwan's AI scale-up
At a Taiwan Semiconductor Industry Association annual conference, TSMC business development deputy general manager Yuan Liben said AI infrastructure has entered an industrial-scale deployment cycle, increasing the difficulty of chip and system validation. He proposed seven actions covering early ecosystem design, shared test facilities, data infrastructure, talent, regulation, funding, and localized supply chains, while warning that future NVIDIA NVL144 racks could require 1,500 TB/s of scale-up bandwidth, more than 600 kW of power, and full liquid cooling.
The remarks connect AI infrastructure growth with concrete engineering and policy requirements, including CPO, liquid cooling, early validation, and supply-chain coordination, based on observations attributed to one TSMC executive.
台積電業務開發副總經理袁立本指出，全球 AI 產業已從實驗室階段的「創新週期」，全面邁入大規模部署的「產業化規模週期」。隨著 AI 伺服器算力百倍成長，產品開發與量產時程急遽縮短，晶片與系統驗證難度呈指數型攀升。袁立本強調，台灣具備全球最完整的 AI 資料中心生態系，應透過早期協同創新與產業鏈升級，從「全球 AI 致能者」（AI Enabler）跨越為「全球 AI 樞紐」。
#TSMC#Taiwan Semiconductor Industry Association#NVIDIA#NVL8#NVL72#NVL144#Rubin#DGX#Hopper