Hong Kong's internet is completely open. ChatGPT and Claude are not available here — Gemini is. Here is the accurate picture, and what your options are if you just want to use AI.
Availability differs by company, and the difference matters:
| Service | Hong Kong | Why |
|---|---|---|
| ChatGPT (OpenAI) | Not available | Hong Kong, Macau and mainland China are absent from OpenAI's supported-countries list |
| Claude (Anthropic) | Not available | Absent from Anthropic's supported regions; location is re-checked during use, not only at sign-in |
| Gemini (Google) | Available | Hong Kong and Macau appear on Google's supported list; mainland China only through Workspace |
| Copilot (Microsoft) | Available | Routes to GPT models under Microsoft's own regional terms |
None of this is a government block, and none of it is censorship. Your connection is not filtered — every other service loads normally. These companies choose which territories they serve, and the lists differ from one to the next. Check the companies' own pages rather than any article, including this one: verified 25 October 2026.
Taiwan, incidentally, is listed by all three. People often assume one regional policy covers all of Greater China; the published lists do not work that way.
It is worth being precise, because the usual framing is wrong in both directions:
Using a VPN or a foreign payment method to obtain an account in an unsupported region breaks the service's terms. Accounts get closed, and money paid is typically lost. Some third-party services resell access; those are between you and them, and your data goes wherever they send it.
We do not provide circumvention instructions — not out of caution about the technology, but because there is no version of it that is compliant, and building a habit on an account that can be closed at any time is a bad foundation for work you depend on.
The distinction worth keeping clear: availability can change with a policy update, and a service you built a habit on can be withdrawn. A file on your own disk cannot.
In most places, running AI locally is a preference about privacy or cost. In Hong Kong it is also about access: a capability that cannot be withdrawn by a policy update.
The trade-offs are real, and we would rather state them than sell past them. A local model on an ordinary laptop is slower than a cloud service and less capable at hard reasoning. It needs 8 GB of memory minimum, and 16 GB to be comfortable. What you get instead is that it works, needs no account, costs nothing after the download, and keeps working in five years.
Free tiers of cloud assistants are genuinely usable, but they come with limits — smaller models, message caps, and your text going to a company. If your work involves client information, that last point is the one that decides it. Local removes the question rather than answering it.
29 pages, plain language, no account needed. What a language model actually is, which kinds exist, what you are really paying for, and how to run one yourself offline.