Does Hong Kong Have Its Own Large Language Model?

That distinction is the story: Hong Kong has crossed into a localised LLM-plus-applications phase, but the evidence does not yet prove a fully autonomous frontier-model stack.

First, define “Hong Kong’s own model”

There are two very different meanings hidden inside that phrase.

The first is a localised LLM: a model developed, fine-tuned, deployed or operated by Hong Kong institutions for local government, education, finance or enterprise workflows. HKGAI V1, HKPilot and related local LLM research fit this category on the public record.

The second is a from-scratch frontier foundation model: a base model with public evidence of its own pretraining, compute, data, benchmark performance and commercial deployment strong enough to compete directly with the world’s leading foundation-model developers. The public material on HKGAI V1, HKPilot, Cyberport’s Artificial Intelligence Supercomputing Centre (AISC) and the government’s AI Plus agenda does not establish that Hong Kong has reached that level.

HKGAI V1: the strongest evidence for a local LLM

HKGAI V1 matters because it is not just a research announcement. HKUST says the tool is available to its staff and students, and that HKGAI V1 had previously been piloted by civil servants before being extended to the university setting.

For readers outside Hong Kong, HKUST is one of the city’s major research universities, while HKGAI is the Hong Kong Generative AI Research and Development Center. The government’s Legislative Council material says HKGAI is funded by InnoHK, a government-backed research-cluster programme, and that HKGAI V1 was released in February 2025.

The caveat is equally important. The same Legislative Council page records reports that HKGAI V1 was based on DeepSeek full-parameter fine-tuning and continuous training. So the most precise wording is: Hong Kong has a locally developed and locally deployed LLM for Hong Kong scenarios; public evidence does not justify presenting HKGAI V1 as a wholly from-scratch, globally leading base model.

Real use cases are appearing first in public services and education

A model name alone does not make an AI ecosystem. Deployment does.

The public-sector use case is the clearest. Government material identifies HKPilot as a generative-AI document-processing copilot and says it has been put on trial in more than 70 government departments. The same government material says HKGAI is researching a series of open-source foundation models, including a local LLM and HKPilot based on that LLM.

Education is the next visible channel. HKUST says it is the first local university to trial HKGAI V1 and that staff and students can access it for free.

Finance also has a formal testing route. The Hong Kong Monetary Authority (HKMA), the city’s monetary and banking regulator, announced a second cohort of its GenA.I. Sandbox, showing that generative AI trials are being channelled through a regulated financial environment. That supports the case for an institutional AI ecosystem, but it does not mean every sandbox project uses HKGAI V1.

The ecosystem story is about compute, funding and company density

The more durable question is not whether Hong Kong can advertise a “Hong Kong GPT”. It is whether the city is building the compute, capital, talent pipeline, companies and application demand around AI.

On compute, the Innovation, Technology and Industry Bureau says the first-phase facility of Cyberport’s Artificial Intelligence Supercomputing Centre began operation in December 2024, with the aim of supporting local computing-power demand and improving Hong Kong’s R&D capability across multiple technology fields.

On funding, Cyberport says the 2024-25 Budget allocated HK$3 billion for a three-year, multi-pronged AI support arrangement to support Hong Kong’s AI ecosystem development.

On demand, the South China Morning Post reported Cyberport as saying more than 90% of Hong Kong’s AI supercomputing capacity was already in use. That is a useful demand signal, but it should not be overread: high utilisation of an AI supercomputing centre is not the same as proof that Hong Kong can train a global frontier model from scratch.

On company formation, a Legislative Council paper says that since 2023 the government has facilitated about 500 leading or promising innovation-and-technology enterprises to set up or expand in Hong Kong, across strategic industries including life and health technology, AI and robotics, advanced manufacturing and new energy. Cyberport also says its community includes around 400 startups specialising in AI and big data. Those figures do not mean every company is building foundation models, but they do support the more measured conclusion that a local AI ecosystem is taking shape.

Policy direction: more AI adoption than model-size race

The government’s 2025 Policy Address language, as described in a Legislative Council reply, points to “AI Plus”: broadening AI applications to empower industries while consolidating Hong Kong’s strengths in AI research, talent, funding and data.

That emphasis matters. Public policy language looks less like a single-minded race to build the largest model and more like an effort to spread AI across public services, research, regulated industries and business workflows.

What this means for founders and product teams

For companies building in Hong Kong, the public evidence points to nearer-term openings around document-workflow copilots, education tools, regulated financial-AI trials and services linked to supercomputing demand.

The pragmatic product lesson is not that foundation-model research is irrelevant. It is that the clearest current opportunity appears to be at the application and integration layer: local deployment, trusted workflows, domain-specific tools and models adapted to Hong Kong use cases. Based on the published projects, Hong Kong’s visible path is a combination of localised models, AI compute infrastructure and vertical applications.

Bottom line

If “Hong Kong’s own large model” means an LLM developed by local teams, deployed in local government and university settings, and adapted for Hong Kong needs, the answer is yes: HKGAI V1 is the clearest example.

If the question is whether Hong Kong has a local AI ecosystem, the answer also leans yes. The evidence includes Cyberport’s AISC, HK$3 billion in three-year AI support, HKPilot trials across more than 70 government departments, HKUST’s HKGAI V1 trial, the HKMA GenA.I. Sandbox and a cluster of AI and big-data startups.

But if the claim is that Hong Kong already has a fully autonomous, from-scratch foundation model that can compete head-to-head with the global frontier, the public evidence is not enough. The fairer conclusion is that Hong Kong is building an AI ecosystem centred on localised LLMs, supercomputing infrastructure and vertical applications. HKGAI V1 is an important starting point, not the final proof.