Key facts
- Hong Kong's installed compute capacity reached 5,000 petaflops in 2026, up from roughly 60 petaflops in 2023, per China Daily Hong Kong.
- The report said Hong Kong plans to scale to 180,000 petaflops by 2032, a level it said matches Ulaanqab in Inner Mongolia.
- Forrester analyst Charlie Dai said he has not seen severe compute oversupply in Hong Kong, per the report.
- University of Hong Kong's Mao Zhenhua described a structural mismatch: too much mid- and low-end capacity, too little high-end capacity for large-model training.
- China Mobile launched a global intelligence center in Tseung Kwan O in March 2026, the report said.
Hong Kong's installed artificial intelligence compute capacity rose from roughly 60 petaflops per second in 2023 to 5,000 petaflops in 2026, with plans to reach 180,000 petaflops by 2032, China Daily Hong Kong reported on November 11, 2026. The outlet said the 2032 target would match the scale of Ulaanqab, a computing hub in China's Inner Mongolia region powered by wind and solar generation.
The report said China Mobile, which set up its global network center in Hong Kong's Tseung Kwan O district in 2014, launched a global intelligence center in the city in March 2026, describing it as one of the special administrative region's largest stand-alone intelligent computing facilities.
Charlie Dai, a vice-president and principal analyst at research firm Forrester, told the outlet he has not seen severe compute oversupply in Hong Kong because demand from financial institutions, multinational companies, cloud providers and AI enterprises remains healthy, though some newer facilities may see slower utilization during ramp-up. Dai said premium facilities with strong power, connectivity and graphics processing unit capabilities continue to draw customers, and that the primary risk for Hong Kong is demand-supply misalignment rather than structural oversupply.
Mao Zhenhua, a professor of practice in economics at the University of Hong Kong Business School, described a structural mismatch in computing infrastructure across three dimensions: an oversupply of mid- to low-end capacity against a shortage of high-end clusters needed for large-model training, AI enterprises and compute demand concentrated in different areas, and technology giants holding most high-end compute resources while small and medium-sized enterprises face high costs and limited access.