On March 16, 2026, Google made headlines across Hong Kong with a long-awaited announcement: Gemini, its flagship AI assistant, was finally being opened to all users in the city. No VPN required. No enterprise subscription needed. Just visit gemini.google.com and sign in with a personal Google account.
The response was immediate. The Gemini app shot to the top of the App Store charts within days of launch. For a city that had spent years watching Western AI tools pass it by — or using VPNs to access them in a legal gray zone — the rollout felt like a turning point.
But there’s a catch that most of those headlines missed entirely: the Gemini API is still blocked from Hong Kong IP addresses. If you’re a developer trying to build anything with Gemini — a chatbot, an automation workflow, an AI-powered application — you’re still getting the same error message you’ve always gotten.
"User location is not supported for the API use."This isn’t a glitch. It’s a deliberate policy split — and understanding it reveals something important about how geo-restrictions actually work, and why a VPN remains essential for anyone doing serious AI work in Hong Kong.
What Google Actually Opened (And What It Didn’t)
Before March 2026, Gemini was essentially off-limits for individual users in Hong Kong. For years, the city had been grouped alongside mainland China as a region where major Western AI companies — Google, OpenAI, Anthropic — chose not to operate their consumer AI products. The reasons were a mix of regulatory uncertainty, compliance risk, and the chilling effect of Hong Kong’s post-2020 legal environment.
Some residents found workarounds: third-party aggregators, shared accounts, or VPNs that made their traffic appear to originate from a supported region. But these were unofficial routes, not something you’d build a business or a development workflow on.
Google’s March 2026 rollout changed that for the consumer product. The web app at gemini.google.com is now accessible to personal Google accounts registered in Hong Kong. The mobile app followed shortly after, landing first on Android before rolling out to iOS. As of now, Gemini is the only major Western AI assistant with a free tier officially available to individual users in the city — ChatGPT and Claude remain blocked for personal use.
What the rollout did not change:
- Gemini API via Google AI Studio — still blocked for Hong Kong IPs
- Google AI Studio itself — still inaccessible from Hong Kong
- Programmatic access to Gemini models — still requires either a workaround or an enterprise route
According to Google’s official documentation (last updated March 2026), the Gemini Web App now covers 230+ countries and regions, while the Gemini API covers 200+ — but Hong Kong appears on neither list for the API tier. The web app and the API are governed by separate regional policies, and they’re being opened on separate timelines.
The Error Developers Keep Hitting
For developers, the practical consequence is straightforward and frustrating. Any API call to the Gemini API originating from a Hong Kong IP address returns a 400 error with the message: User location is not supported for the API use (FAILED_PRECONDITION).
This has been documented in Google’s developer forums for over a year — developers running servers in Hong Kong, or whose cloud infrastructure routes traffic through Hong Kong data centers, hitting the same wall regardless of where they personally are located. The restriction is enforced at the network level, not at the account level.
The Cloudflare Workers case makes this especially visible. Developers based in supported regions like Australia or the US have reported their Workers deployments returning the same “User location is not supported” error — because Cloudflare’s routing placed those Workers on its HKG edge node in Hong Kong. Your account, your location, your billing address — none of it matters. What matters is the IP address that makes the outbound request.
This creates real friction across a range of common developer scenarios:
- Applications deployed on cloud platforms with Hong Kong infrastructure
- Cloudflare Workers routed to HKG edge nodes
- Self-hosted services running from local Hong Kong servers
- CI/CD pipelines hosted in Hong Kong data centers
- Any automated workflow that calls the Gemini API at scale
Why the Split? The Logic Behind Two Separate Policies
It might seem strange that Google would open the consumer product while keeping the developer API closed. But the split actually follows a clear internal logic — one that shows up across how AI companies think about regional risk.
Consumer access and API access carry fundamentally different risk profiles. A user chatting with Gemini through a browser generates one conversation at a time. An API key can power thousands of automated requests per hour, integrated into applications that might serve users anywhere in the world. From a compliance and abuse-prevention standpoint, these aren’t the same thing.
The post-2020 legal environment in Hong Kong remains a factor. The regulatory and compliance uncertainties that led companies like OpenAI and Anthropic to block Hong Kong IP addresses in the first place haven’t fully resolved. Google appears to be making a calculated move: open the consumer product to build goodwill and market presence, while maintaining tighter controls on programmatic access until the legal picture is clearer.
Misuse risk is higher at the API level. In early 2026, Anthropic publicly reported that Chinese AI laboratories had used fraudulent accounts to extract capabilities from Claude at industrial scale — over 16 million interactions. Google has reason to be cautious about API access from Hong Kong IP addresses, even if individual users in the city are perfectly legitimate.
This also explains why Google’s enterprise route through Vertex AI has always been available in Hong Kong, even when the consumer product wasn’t. Enterprise customers go through vetting, sign contracts, and operate under explicit terms of service — a very different risk profile from anonymous API key usage.
What Developers Can Actually Do Right Now
If you need Gemini API access from Hong Kong, there are three realistic paths. Each has tradeoffs.
Option 1: Vertex AI (the official enterprise route)
Google Cloud’s Vertex AI platform provides full API access to Gemini models — including Gemini 2.5 Pro and Flash — through Google Cloud infrastructure, which operates under different regional policies than the consumer Gemini API. Vertex AI is available in Hong Kong, no VPN required.
The endpoint is OpenAI-compatible, meaning tools like LangChain, Cursor, and most AI frameworks can connect to it with minimal changes. The tradeoffs: Vertex AI pricing runs roughly 20–40% higher than the direct Gemini API, and you’ll need a Google Cloud billing account with a payment method on file. It’s the right path if you’re building something production-grade and want a fully official, contractually supported route.
Option 2: Route through a supported region
If your infrastructure allows it, routing outbound API calls through a server or proxy in a supported region — Japan, Singapore, the US — bypasses the geo-restriction cleanly. This is the developer equivalent of what individual users do with VPN browser access, applied at the infrastructure level.
For teams already running multi-region cloud infrastructure, this can be as simple as ensuring that AI API calls are routed through nodes in supported regions. The call still comes from your application; it just exits through an IP address that Google’s systems recognize as valid.
Option 3: VPN at the server level
For developers running self-hosted infrastructure in Hong Kong, configuring a VPN at the server level — rather than just on a browser or personal device — routes outbound API calls through a supported region’s IP address. This is the most flexible option for teams that want to use the standard Gemini API at standard pricing, without migrating to Vertex AI or restructuring their cloud architecture.
A reliable VPN that supports server-level configuration, stable IP addresses in supported regions, and consistent uptime is the key requirement here. Consumer VPNs marketed primarily for streaming access won’t always meet these needs — you want something built for infrastructure use.
The Broader Pattern: Geo-Restrictions Don’t Disappear Overnight
The Gemini Hong Kong situation is a useful case study in how geo-restrictions actually work in practice. When a company announces that a product is “now available” in a region, it rarely means all access layers open simultaneously. Consumer products, developer APIs, enterprise plans, and embedded integrations often operate under separate regional policies — updated on different timelines, governed by different legal and risk assessments.
Hong Kong’s AI access landscape as of March 2026 reflects this layered reality. The Gemini web app is open. The Gemini API is not. ChatGPT remains officially unavailable for personal accounts. Claude has no official personal-use launch in the city. DeepSeek is available but carries well-documented concerns about content restrictions on politically sensitive topics. Grok, Perplexity, and Meta AI through WhatsApp fill some gaps, but none offer the full developer access that serious builders need.
For developers and technical users working from Hong Kong, this means a VPN isn’t just a tool for bypassing regional blocks on consumer apps — it’s a foundational piece of infrastructure for anyone building with AI. The consumer access story may be improving, but the developer access story is lagging well behind, and there’s no announced timeline for when the Gemini API will open to Hong Kong IPs directly.
Until that changes, the options are Vertex AI, routing through supported regions, or a reliable VPN. For most developers, some combination of these three will be the practical reality for the foreseeable future.
What to Watch For
The March 2026 consumer rollout does suggest that Google is actively reassessing its Hong Kong policy. The company’s stated goal of advancing Hong Kong signals genuine intent to expand access over time, not just a one-off gesture.
The question for developers is whether API access follows the consumer rollout on a similar timeline, or whether it gets treated as a separate decision requiring a separate round of compliance review. Given how differently the two tiers are treated today, there’s no guarantee that opening one automatically leads to opening the other.
Watch Google’s available-regions documentation page (ai.google.dev/gemini-api/docs/available-regions) — it’s updated periodically and will be the first place any change to API regional availability appears. Until Hong Kong shows up on that list for the API tier, the workarounds above remain the practical path forward.