Despite China's official ban on US AI services, a thriving underground market has emerged where developers access ChatGPT, Claude, and other large language models at heavily discounted rates through a network of intermediaries and proxy systems. This grey market operates largely invisible to Anthropic and OpenAI, revealing how geopolitical restrictions on AI technology create unintended economic consequences and enforcement challenges.

How Does China's Underground AI Token Market Actually Work?

The system relies on what researchers call "transfer stations," middlemen who purchase API tokens from legitimate sources and resell them at discounts to Chinese developers who cannot access these services directly. According to reporting on the grey market, "the logs are the product," meaning the transaction records themselves become valuable data that intermediaries can analyze and monetize. This creates multiple layers of economic activity beneath the surface of official token requests that Anthropic and OpenAI track.

The infrastructure supporting this market includes several key mechanisms:

• VPN and Proxy Networks: Developers route requests through Southeast Asian servers, particularly Singapore, to mask their location and appear as legitimate users outside China's restricted zones.
• Token Reselling Intermediaries: Third-party brokers purchase API access at standard rates and resell tokens at 30-50% discounts, profiting from arbitrage while keeping prices accessible to Chinese teams.
• Fraud and Testing Layers: The market includes significant fraud risk, with some intermediaries testing token validity and authenticity before resale, creating quality assurance mechanisms within the grey market itself.

Why Is Singapore Becoming the Hub for China's AI Activity?

Singapore has emerged as the critical junction point for this shadow economy. The city-state's position as a regional financial and technology hub, combined with its proximity to mainland China and permissive regulatory environment, makes it the natural gateway for traffic flowing through VPNs and proxy services. Data shows that people in Singapore are using Claude at exceptionally high rates, a pattern that reflects the concentration of mainland Chinese developer activity being routed through the region.

This geographic arbitrage creates a peculiar situation where Southeast Asia becomes the de facto access point for US AI services that are officially unavailable in China. The arrangement benefits Singapore's tech infrastructure providers while simultaneously undermining the enforcement mechanisms that Anthropic and OpenAI have implemented to comply with US export controls and Chinese government restrictions.

What Does This Mean for AI Companies and Geopolitical Tensions?

The existence of this grey market highlights a fundamental tension in AI regulation. While governments attempt to control access to advanced AI models through official channels, decentralized reselling networks create workarounds that are difficult to police. Neither Anthropic nor OpenAI can easily distinguish between legitimate international users and Chinese developers accessing services through proxies, making enforcement nearly impossible at scale.

The broader context matters here. China has blocked its population from accessing US AI services as part of a larger strategy to develop domestic alternatives like DeepSeek and Moonshot AI. However, the existence of cheap, accessible alternatives through grey markets may actually slow adoption of homegrown Chinese models among developers who can afford the discounted US services. This creates an ironic outcome where geopolitical restrictions intended to boost domestic AI development instead subsidize foreign competitors through arbitrage pricing.

The situation also reflects deeper questions about how AI regulation functions in a globally connected internet. When services are restricted in one jurisdiction but accessible through proxies in another, the line between compliance and circumvention becomes blurred. Developers operating in this grey market are not technically violating Chinese law, since they are using legitimate API access purchased through legal channels, even if those channels involve intermediaries and geographic routing tricks.

As tensions between the US and China over AI technology continue to escalate, this underground market represents a form of technological friction that neither government fully controls. The existence of cheap, reliable access to ChatGPT and Claude in mainland China, despite official bans, suggests that geopolitical AI restrictions may be less effective than policymakers assume, and that market forces often find paths around regulatory barriers.