Detailed Analysis
Anthropic's decision to crack down on the practice of "efficient" or discounted Claude API resale has become a flashpoint in the developer community, particularly among programmers in China who have built businesses and workflows around arbitraging access to Claude's coding capabilities. The 36Kr article's framing—that programmers worldwide have been "giving Anthropic free money" and that the company can "no longer stand by"—points to a pattern in which third-party resellers and middlemen packaged Claude API access (often through mirrored endpoints, shared accounts, or bulk-purchased tokens) at prices below what Anthropic charges directly, undercutting the company's own revenue capture while still relying entirely on Anthropic's underlying models. Anthropic appears to have moved to shut down or restrict these gray-market channels, a move that directly affects the many independent developers, small dev shops, and coding-tool startups that had come to depend on cheaper access to Claude, especially for tools like Claude Code.
This matters because Claude has become the preferred foundation model for a large swath of the AI coding tools ecosystem—including Cursor, Windsurf, and countless homegrown agentic coding wrappers—precisely because of its strong performance on software engineering benchmarks and its favorable price-to-capability ratio compared to GPT-4-class or Gemini models. When resellers found ways to offer that access more cheaply than Anthropic's official pricing, it created a shadow economy that both expanded Claude's reach into developer communities (particularly in markets like China, where direct API access can be complicated by payment, compliance, or regional availability issues) and quietly siphoned margin away from Anthropic. The company standing by while this happened for a period likely reflected either limited enforcement bandwidth, tacit tolerance because it grew Claude's developer mindshare, or difficulty distinguishing legitimate reseller partnerships from unauthorized arbitrage. The reported crackdown suggests Anthropic has now decided the revenue leakage and lack of control over usage, rate limits, and terms-of-service compliance outweighs the growth benefits.
The episode is emblematic of a broader tension in the AI industry between compute costs and developer demand for cheap, high-volume access to frontier models. As coding assistants and agentic AI tools proliferate, the marginal cost of running inference at scale is substantial, and companies like Anthropic—which do not have Google's or Microsoft's balance sheets to subsidize usage indefinitely—are under pressure to tighten monetization. This mirrors dynamics seen with OpenAI's periodic crackdowns on API key sharing, rate-limit evasion, and unauthorized wrapper services, as well as broader industry moves toward usage-based pricing, stricter authentication, and enterprise-tier gating to prevent value leakage to intermediaries.
Finally, the situation underscores how central Chinese developer communities have become to the global AI coding tools market, even as geopolitical and regulatory friction limits official mainland access to leading US models. Programmers in China have shown a strong willingness to find workarounds—VPNs, resold API keys, mirrored gateways—to access Claude specifically, which speaks to the model's reputation there as best-in-class for coding tasks, arguably ahead of domestic alternatives from Alibaba, DeepSeek, or Zhipu. Anthropic's tightening of enforcement, if confirmed, will likely accelerate interest in open-weight or domestically hosted alternatives among price-sensitive developers, while also signaling to investors that Anthropic is serious about converting its technical lead into sustainable revenue rather than allowing arbitrage to erode its unit economics during a period when frontier model training and inference costs remain extraordinarily high.
Read original article →