Cyberport Eyes Chinese GPUs to Cut Nvidia Reliance

Cyberport, Hong Kong’s government-backed tech incubator, is exploring the adoption of Chinese-made GPUs at its AI Supercomputing Centre to reduce reliance on Nvidia amid ongoing China-US technology tensions.

The centre, which currently operates on Nvidia H800 chips delivering 1,300 petaflops of computing power, plans to expand capacity by 1,700 petaflops by the end of 2025, reaching 3,000 petaflops.

Cyberport has tested GPUs from four domestic Chinese chipmakers at its AI lab, evaluating both performance and cost before making procurement decisions. The move reflects a growing industry trend where geopolitical considerations now outweigh purely technical choices, as Hong Kong seeks to strengthen supply chain resilience.

Key Highlights

• Cyberport tests four Chinese GPU suppliers to diversify beyond Nvidia
• AI capacity to expand to 3,000 petaflops by end of 2025
• Hong Kong aims for 15,000 petaflops AI computing power by 2030

The incubator is also building a 60,000 sq. ft. data centre, dedicating two-thirds of the space to AI computing, with a long-term goal of reaching 15,000 petaflops capacity by 2030.

Chinese GPU makers such as Biren Technology and Moore Threads have recently demonstrated competitive performance, with some benchmarks showing domestic GPUs outperforming Nvidia in specific simulations, including flood defense modeling. This maturing ecosystem makes cost the key differentiator rather than performance.

Also Read: Foxconn and NVIDIA to Construct AI Supercomputer in Taiwan

Nvidia itself has responded to US export restrictions by designing region-specific chips, including the B30 chip, which delivers 80 percent of its Blackwell performance for China, underlining the shift toward localized solutions. Cyberport’s diversification strategy ensures Hong Kong remains a bridge between Chinese innovation and international AI infrastructure development.