Surging demand for AI agents fuels Hong Kong’s race to expand computing power

As leading Chinese chip executives predict that autonomous AI agents will trigger an unprecedented explosion in computing demand, Hong Kong is accelerating efforts to expand its artificial intelligence computing capacity.Calling token consumption in the era of AI “far beyond our imagination”, Zhang Jianzhong, founder and CEO of graphics processing unit designer Moore Threads, said at an AI summit in Hong Kong on Saturday that the exponential surge in demand had made it impossible for anyone to accurately predict future needs.Zhang argued that competitive edge was no longer about sheer volume; rather, it was to do with cost efficiency. “Whoever offers the cheapest tokens will be able to build better infrastructure for everyone,” he said.Chen Weiliang, chairman and general manager of chip designer MetaX, echoed Zhang’s views.Citing a recent SemiAnalysis report, Chen attributed a 6,400 per cent increase in average daily token consumption to the industry’s transition from basic large language models to agentic AI workflows.Zhang Jianzhong, founder and CEO of graphics processing unit designer Moore Threads. Photo: HandoutHong Kong currently has 5 ExaFLOPS of computing power, compared with 60 ExaFLOPS in Beijing and 120 ExaFLOPS in Shanghai, according to Zhang, another speaker at the AI Summit.