来源：内容来自 风传媒 。 台积电作为一家纯晶圆代工制造厂，为无晶圆厂芯片设计公司提供了可靠的制造产能。更重要的是，台积电由一位了解产 业游戏规则、曾任职美国半导体主流企业的资深经理人经营，唯一的问题是技术落后。台积电成立时，其制程技术大约比 产业领先的垂直整合制造商落后了两个制程节点，还无法为无晶圆厂芯片设计公司的创新产品提供服务。因此，台积电必 须从捡拾垂直整合制造商的"剩菜"开始，而不是和无晶圆厂芯片设计公司结盟。当时大多数的美国无晶圆厂芯片设计公司 都是委托日本的半导体制造商代工，因日本的制造品质和良率享有盛名。 创业维艰 虽然业绩欠佳，但台积电致力于技术学习。在成立约四到五年后，即1991至1992年间，台积电与市场领先者之间的技术差 距已缩小至一个制程节点，此后无晶圆厂芯片设计公司的订单即逐渐涌入。创业维艰，1987年至1992年，对台积电最终的 成功来说，是既辛苦又关键的阶段，在这段艰难的岁月中，工研院和飞利浦的技术根基和经营团队的人脉网络是公司重要 的资产。 台积电最早、也是最重要的"剩菜"订单来自英特尔，它委托台积电代工制造MCU 80C51芯片，采用1.5微米制程。曾繁城 从电子所的VLS ...

Group 1 - The core viewpoint of the article is that the demand for data centers in the Asia-Pacific region is expected to double in the next three years due to the robust development of AI and the increasing demand for cloud services [1] - The report highlights that mature markets such as Japan, Australia, and South Korea will continue to see strong demand, while Singapore, despite limited supply, remains a focal point of interest [1] - It is projected that by 2028, the Asia-Pacific region will face a power shortage of 15 to 25 gigawatts due to limited power supply and a lack of infrastructure to support AI [1]

Group 1 - The core viewpoint of the report is that the demand for data centers in Hong Kong is strong due to the AI boom and digital transformation across various industries [1] - The report predicts that the supply of data centers in the Asia-Pacific region will double in the next three years, but there will be a power shortfall of 15 to 25 gigawatts by 2028 due to power supply limitations and lack of infrastructure supporting AI [1] - Hong Kong is positioned as a critical entry point for data center traffic in the Asia-Pacific region, attracting investment despite rising operational costs and infrastructure constraints [1] Group 2 - The Asia-Pacific region faces similar challenges, with strong demand for server hosting and large-scale data centers driven by the growth of AI and cloud services [2] - The rapid adoption of AI and cloud services is increasing the demand for next-generation data centers, but many existing and upcoming data centers are not designed to support AI workloads [2] - Data centers focused on AI require more than double the power density per server rack, advanced cooling systems, structural reinforcement, and low-latency network connections, posing significant challenges for developers [2]