Hong Kong AI Lab to Launch Sovereign Model Based on DeepSeek Architecture

A Hong Kong government-backed artificial intelligence lab is preparing to launch a new model built on DeepSeek that can run entirely on Chinese-made chips.

The Hong Kong Generative AI Research and Development Centre (HKGAI) plans to unveil its HKGAI-V3 model in the first half of this year as it looks to export its “sovereign AI” capabilities to overseas markets.

The system was based on DeepSeek V4 architecture with “full-parameter fine tuning for localisation”, and had been optimised to operate on both mainstream and domestic hardware, including Huawei Technologies’ Ascend 910C chips, according to director Guo Yike, who spoke on the sidelines of a Tsinghua University alumni event in Hong Kong on Sunday.

HKGAI, established in 2023 under the government’s InnoHK programme, focuses on developing large language models tailored to Hong Kong’s needs. It launched the city’s first Cantonese-enabled chatbot, HKChat, in November, offering services such as real-time bus information and guidance on local laws and regulations.

In a research paper published in July, HKGAI described its first model, HKGAI-V1, as a “crucial and pioneering step” towards building sovereign AI – broadly defined as the ability for a jurisdiction to develop and operate AI systems on its own infrastructure.

The researchers said that a sovereign large language model such as HKGAI-V1 aimed to “ensure the security of critical technologies, the mastery of data sovereignty, and alignment with local language, culture and institutional requirements”.