Claude at ten percent of list: China's proxy market makes geoblocking a billing control

Farming free credits does not pay for this on its own. Anthropic's $5 signup credit ^[11], resold at a tenth of list ^[1], grosses fifty cents; an operator would need roughly 400 registered accounts to gross $200 ^[1]. Registration is not free labour either, since each account needs a foreign phone number bought from an SMS verification platform ^[7]. The margin has to come from the other levers Qian lists: enterprise and education discounts, and a single $200 Max plan sliced across many users by token quota ^[11]. And from cards that were not the buyer's, whose share of the account pool the analysis cannot size ^[12].

Model swapping is the part that should interest anyone building on resold access. Because the station sits between the client and the API, it decides which model answers ^[2], and operators substitute cheaper models for expensive ones without telling the buyer ^[5]. Someone paying ten percent of list ^[1] for what is sold as a frontier model has no way to verify what produced the tokens. That includes the buyers Qian believes are distilling Western outputs to improve weaker models ^[13]. A distillation run bought through a station is trained on whatever the operator's routing policy picked that day, not on the model named in the listing.

Look at where Anthropic's controls actually sit. Phone numbers, foreign credit cards, billing addresses, majority-ownership screening, and an ID plus live selfie for selected users ^[4] all cluster at the moment an account is opened, and account brokers have industrialised exactly that moment ^[7]. None of it observes the person running inference, because that person never holds an account. Verification of this kind establishes who can be billed.

Qian is careful about her evidence, and the read should be too: parts of the account rest on informal conversations and publicly available sources ^[15]. The KYC bypasses she describes cover generated IDs, deepfaked liveness checks, and recruited people in low-income countries ^[9], but without a price attached to Claude-specific verification. The nearest hard figure is her Worldcoin comparison, where iris scans from Cambodia and Kenya changed hands for under $30 ^[10]. Treat the mechanisms as documented and the volumes as unmeasured.

Her framing correction matters as much as the pricing. In the US these proxies are discussed as a security problem, while the buyer list runs through students, researchers, developers, tech employees, companies and hobbyists ^[14]. Ten percent of list is a ninety percent discount, the top end of the 70 to 90 percent range Qian describes ^[2], and a discount that large with a customer base that wide is a functioning consumer market. Markets of that shape are not closed by tighter checks at signup. They are closed, if at all, by raising the cost of the account supply feeding them.