Winnie Tang

Opinion | Hong Kong’s supercomputing centre plans must be strategic to succeed

As Hong Kong looks to set up an AI supercomputing centre, it must watch out for construction costs, define the goals of the project clearly and ensure the city has enough local talent to realise its potential

Reading Time:3 minutes

Why you can trust SCMP

A view of Kowloon at night in 2020. Hong Kong’s budget this year proposes to study the feasibility of setting up an AI supercomputing centre to promote scientific research. Photo: Sam Tsang

Winnie Tang

Published: 6:15am, 12 Mar 2023

Computing power has become a core driver of the development of the digital economy. The white paper on China’s computing power development index released by the China Academy of Information and Communications Technology in November last year estimates that every yuan invested in computing power would increase economic output by three to four yuan.

Moreover, computing power has a significant correlation with gross domestic product. Of the top 20 countries by computing power in 2021, 17 are among the top 20 economies in the world. The top five countries by GDP – the United States, China, Japan, Germany, and the United Kingdom – were also the top five in terms of computing power.

Hong Kong’s budget this year proposes to study the feasibility of establishing an artificial intelligence supercomputing centre to promote AI development and scientific research. The study will be completed by 2024. For the plan to succeed, the government should have two major considerations: the positioning of the centre and talent training.

Sun Dong, secretary for innovation and technology and industry, said that if one made reference to the supercomputing centres in neighbouring areas, the investment involved could be to the tune of billions of dollars. Further, the globally tight supply chain of hi-tech products, especially computer chips currently, adds to the complexity of the undertaking.

To make good use of public funds, we could learn from the mainland’s experience.

First, beware of price chaos. It has been reported that the construction cost of the same AI computing centre – with computing power of 100 PFlops (FP16) – could differ by as much as six times between mainland cities, costing 75 million yuan (US$10.7 million) to build in one city and 460 million yuan in another. This, coupled with low price transparency, means extra caution is warranted to ensure funds are wisely spent.

A worker monitors the Shenwei TaihuLight supercomputer at the National Supercomputer Center in Wuxi in eastern China’s Jiangsu province on August 29, 2020. Photo: AP