A government-supported artificial intelligence lab in Hong Kong is set to introduce a new model based on DeepSeek that operates exclusively on Chinese-manufactured chips. The Hong Kong Generative AI Research and Development Centre (HKGAI) is gearing up to launch its HKGAI-V3 model in the first half of the year, aiming to extend its “sovereign AI” capabilities to international markets. This system, built on the DeepSeek V4 architecture, has undergone comprehensive parameter fine-tuning for localization and is optimized to function on both mainstream and local hardware, including Huawei Technologies’ Ascend 910C chips, as explained by director Guo Yike during a Tsinghua University alumni event in Hong Kong. HKGAI, founded in 2023 under the government’s InnoHK initiative, is dedicated to developing large language models tailored to Hong Kong’s specific needs. It introduced the city’s first Cantonese-enabled chatbot, HKChat, in November, offering services like real-time bus information and guidance on local laws. In a research paper from July, HKGAI characterized its initial model, HKGAI-V1, as a significant step towards creating sovereign AI, defined as the capability for a region to develop and operate AI systems using its own infrastructure. Researchers highlighted that a sovereign large language model like HKGAI-V1 aims to secure critical technologies, ensure data sovereignty, and align with local language, culture, and institutional requirements.
How is Hong Kong innovating with AI using Chinese chips?
