How this ‘underground market’ is giving Chinese coders access to Claude and Gemini

Chinese developers are increasingly turning to a sprawling underground network of “shadow APIs” to access leading US artificial intelligence models such as Anthropic’s Claude and Google Gemini, despite tightening restrictions aimed at cutting off users in mainland China.According to a report by South China Morning Post, relay platforms hosted outside mainland China are quietly becoming essential tools for programmers who want access to frontier AI systems for coding, debugging and image generation. These services effectively act as middlemen, routing requests through overseas infrastructure and masking the user’s location.Listings advertising “no-VPN” access to Claude, ChatGPT and Gemini have become common on Chinese marketplaces operated by Alibaba Group, including Taobao and Xianyu. Sellers promise support for tools such as Cursor and VSCode, along with access to large context windows and supposedly unrestricted versions of the models.One seller cited in the report claimed to have fulfilled more than 2,200 orders for Claude access alone.The demand reflects a broader perception among many Chinese developers that top Western AI systems still outperform domestic alternatives in complex engineering tasks. A Hangzhou-based programmer told SCMP that Claude produced more reliable coding results and required fewer bug fixes than Chinese rivals, which he said were still more prone to hallucinations and unwanted outputs.The grey market has also become increasingly sophisticated. Some providers now advertise unified API access to hundreds of models through a single interface, mirroring similar services emerging globally. Others undercut official pricing by offering API access at sharply discounted rates.But there is a catch: users do not always get what they are paying for.Some operators reportedly substitute expensive overseas models with cheaper Chinese systems such as Qwen or MiniMax without disclosing it to customers. One reseller admitted that many users lack enough experience with frontier AI systems to tell the difference.Meanwhile, access itself is becoming harder to maintain. Over recent months, Anthropic has tightened enforcement measures, expanding restrictions across mainland China, Hong Kong and Macau while introducing stricter identity verification checks. Sellers told SCMP that account bans surged after new know-your-customer requirements were introduced earlier this year.The report highlights how export controls and regional AI restrictions are increasingly fuelling parallel access markets rather than eliminating demand. In practice, the harder major AI companies try to lock down access, the more valuable these unofficial relay networks appear to become.