Claude tokens at bargain prices: China's grey market

The price of Claude tokens on Anthropic's official API is not within reach for everyone, especially in markets where access to US-based paid services is complicated. An article published this week on ChinaTalk describes in detail how users and developers in China gain access to Claude, often with significant discounts from list price, through resellers, intermediary platforms and local cloud providers acting as an unofficial distribution layer.

The piece is not about a marginal anecdote: it points to a structured ecosystem, with forums, messaging groups and shops commercialising Anthropic API credits opaquely. The discussion on Hacker News that brought the news to our attention remains modest in comments, but the issue has more layers than it first appears.

What's actually happening

China has access restrictions on Anthropic services: neither the direct API nor Claude.ai are officially available to users with Chinese IPs or payment methods. This does not prevent access, but rather creates incentives for third parties to arbitrage: they purchase API capacity with foreign accounts and cards and resell it to local users, generally through tokens or prepaid credits, at prices that can be 30 to 60 percent cheaper than official pricing, according to the article.

The mechanism is familiar in other contexts: it occurs with SaaS subscriptions, software licenses and now also with LLM inference capacity. What's particular here is that this is not just curious individual users, but development teams integrating Claude into actual products and looking to reduce operational costs.

Why it matters beyond price

There are at least three relevant angles for the ecosystem.

The first is compliance. Anthropic's terms of service prohibit unauthorised resale of API access. Developers building on resold capacity assume real risk: if Anthropic detects patterns of abuse or decides to shut down intermediary accounts, their integrations could become unusable overnight without warning. This is not a theoretical risk; similar platforms have suffered abrupt cuts in the past with other providers.

The second is competitive. The fact that there is enough demand to sustain a structured resale market in China says something about Claude's technical appeal against local alternatives like Qwen, DeepSeek or Doubao. If Chinese models were perceived as equivalent for relevant use cases, the motivation to navigate the complexity of accessing Claude would disappear. The arbitrage is, in a sense, a signal of preference.

The third is strategic for Anthropic. The company does not operate directly in China, but its technology does circulate. That raises questions about future expansion, how to manage the use of its models in jurisdictions where it has no legal presence, and whether a more explicit distribution strategy in those markets, similar to what some cloud providers do through regional partners, would make commercial and regulatory sense.

Who should care about this

If you are a developer or technical lead with operations in China or teams there, the full ChinaTalk article is worth reading: it describes the concrete channels, price ranges and associated risks in enough detail to make informed decisions. If you are evaluating LLM providers for projects in that region, the existence of this parallel market should enter your vendor dependency risk analysis.

For everyone else, it's a useful window into how technology geopolitics and distribution restrictions create friction and opportunities that product teams and investors do not always have on their radar.

---

From our perspective, what stands out most is not the discount itself, but that there is enough technical demand to justify a parallel distribution chain. That is useful information about where real interest in the Claude ecosystem actually lies, regardless of what official active user figures say.