Author: Chainalysis
Translated by: Deep Tide TechFlow
Deep Tide introduction: As global controls on computing power and algorithms for cutting-edge AI continue to tighten, an underground network of brokers specializing in reselling overseas AI model APIs to restricted regions has taken shape. Chainalysis recently conducted field research and on-chain tracking of 100 AI brokerage services.
The report says that around 40% of brokers directly support cryptocurrency top-ups and settlements, primarily in stablecoins, and use third-party custodial payment channels to obscure the flow of funds. On-chain monitoring also shows that whenever a leading US AI model is released, payment activity from China-linked wallets surges.
Key Takeaways
· The shadow market is helping Chinese users circumvent U.S. AI export controls. Proxy providers offer access to restricted frontier models, enabling companies to distill and integrate the capabilities of advanced U.S. models into their own systems at a fraction of the cost of training models from scratch.
· Cryptocurrency is helping this market operate. Chainalysis research examines the proxy ecosystem’s evolving supply chain and payment infrastructure, finding that 40% of the proxy providers surveyed accept cryptocurrency payments, primarily in stablecoins.
· Wallet activity linked to China rises sharply after the release of U.S. frontier AI models. Data monitoring shows that whenever the United States releases a new model, the transfer and payment activity of wallets suspected of being linked to China rises significantly above normal levels, pointing to concentrated demand for access to new model capabilities—even though aggregate on-chain data alone cannot directly confirm the specific motive for each use.
The emergence of underground channels for accessing overseas large models
Developing frontier AI models requires enormous investment and has significant national security implications. With this in mind, the U.S. government has explicitly prohibited users and companies in mainland China from accessing leading AI tools such as Claude and ChatGPT, in an effort to limit a major competitor’s access to these technologies.
However, research by the Chainalysis Global Intelligence team found that users linked to China who are subject to these restrictions can still frequently access the models. They rely on an underground proxy economy designed specifically to evade export controls, with many channels even reselling access at a discount to official prices.
Once they have access, developers can extract outputs from leading U.S. models at very low cost and use them to enhance their own models, saving enormous amounts on foundational research and development.
These intermediaries, known as “proxy services,” route requests to large models hosted overseas while concealing the requests’ true origin. They accept payments through local fiat channels such as WeChat Pay and Alipay, and have also integrated on-chain assets on a large scale.
Chainalysis surveyed a sample of 100 such AI proxy platforms and found that 40% also accept cryptocurrency deposits, with stablecoins serving as the settlement medium in the vast majority of cases.
Figure: Account top-up page of an AI proxy platform surveyed in August 2026, showing support for Alipay, WeChat Pay, and USDT deposits on the TRC20, BSC, and Arbitrum networks
Cryptocurrency payments are likely a deliberate business strategy for these proxy services: they can obscure fund flows, help users circumvent foreign exchange controls, and preserve an “off-grid” access channel for privacy-sensitive customers.
Chinese companies use proxy interfaces to advance model training, weakening the impact of controls
By providing indirect access to U.S. frontier models, these proxy services continue to undermine the access barriers established by U.S. laboratories on national security grounds.
Using knowledge distillation, Chinese AI labs can train and strengthen their own competing models directly on the outputs of advanced models, at far lower cost than building a training system from scratch.
Anthropic recently publicly accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of operating 24,000 fake accounts through proxy services to systematically extract Claude’s capabilities.
As Anthropic warned in its report, derivative models produced through illicit distillation often lack the safety guardrails built into the original models, and could readily spread high-risk technological capabilities. Just last month, the National Security Agency (NSA) also issued a formal warning, stating that this activity poses significant risks to critical infrastructure, the military, and the security of the technology economy.
Once scattered and covert gray-market channels have now evolved into a clearly structured commercial ecosystem with distinct upstream and downstream roles, supported by dedicated compute-account credit top-ups and financial infrastructure.
In this emerging commercial network, cryptocurrency payments are rapidly gaining traction, offering a ready-made alternative for customers seeking to avoid the real-name trails of domestic digital payment systems.
On-chain payments from wallets suspected of being linked to China surge after U.S. large-model releases
To verify the financial patterns in this chain, we tracked on-chain activity from wallets suspected of being linked to China before and after each major U.S. AI model release. If developers are indeed using proxy services to gain early access to new models and paying with cryptocurrency, a clear spike in transfers should follow the release of each new model.
The results support this hypothesis: during the first three days after each U.S. frontier AI model release, payment activity from wallets suspected of being linked to China to the proxy payment channels described above surged, rising significantly above the average for other wallets across the network.
Specifically, from day 0 to day 2 after a frontier model’s release, payment activity from China-linked wallets was 55% above its normal baseline. Around day 13 after release—typically the window when Chinese open-weight models follow suit—activity instead fell 21% below the baseline.
Figure: Changes in payment activity from wallets suspected of being linked to China during each three-day period following the release of a U.S. frontier AI model (Source: Chainalysis)
This staggered timing closely matches the industry’s urgency to register for and use new U.S. model capabilities, and supports the view that crypto assets are serving as an underlying payment channel for model distillation. Chainalysis also notes, however, that aggregate transaction data alone cannot directly identify the precise use behind each payment.
On-chain payments are often routed to proxy providers through custodial channels
Geographically, crypto wallets making payments to AI proxy providers are heavily concentrated in mainland China, with significant activity also in Japan, Hong Kong, Singapore, and South Korea. This is consistent with the common route used by Chinese proxy operators to access overseas computing resources via corridors in neighboring parts of Asia.
Further tracing also found that the vast majority of AI proxy providers do not receive customers’ cryptocurrency payments directly at their own addresses. Instead, they commonly integrate third-party custodial crypto payment processors.
This layered architecture provides three levels of protection for the proxy ecosystem: first, it creates a buffer between end buyers and the actual proxy operators; second, it helps circumvent banks’ cross-border foreign exchange controls; and third, it preserves offshore, covert access channels for highly privacy-sensitive users.
Given mainland China’s strict ban on cryptocurrency payment transactions, using third-party payment providers to collect and disburse funds can also significantly reduce the legal and operational risks proxy platforms would face if they built their own payment nodes.
However, these nested payment channels also create hidden compliance exposure for regulated, compliant cryptocurrency exchanges. Funds from proxy services, laundered through multiple intermediaries, ultimately still flow to compliant trading platforms, while their direct links to underground AI proxy providers are difficult to identify from the books alone.
How blockchain intelligence can help penetrate and address the proxy shadow market
For U.S. law enforcement and regulators seeking to block the outflow of leading models, AI proxy services pose a serious enforcement challenge. Now that the underground resale ecosystem operates on an industrial scale, simply blocking access to official websites at the source is no longer effective.
The proxy industry’s multilevel supply-and-demand structure makes enforcement even more difficult: upstream suppliers use account farming and centralized account management, as well as model-channel swaps, to obtain access quotas; midstream and downstream proxy providers aggregate compute pools, forward requests, and sell subscriptions on a usage basis. Even if a channel provider at one link is shut down, other suppliers can quickly step in.
Against this backdrop, blockchain intelligence tools can help investigators follow on-chain fund flows through multiple layers of obfuscation and accurately reconstruct the network of connections between underground proxy operators, paying users, payment collectors, and the trading platforms where the funds ultimately settle.
By precisely mapping these hidden financial connections, regulators and AI research organizations can identify and target critical bottlenecks in the industry chain, systematically dismantling the underground funding networks that sustain the unauthorized use of U.S. frontier AI capabilities.
Frequently Asked Questions
What are Chinese AI proxy services?
They are commercial platforms that resell access to the APIs of large models such as Claude, ChatGPT, and Gemini in restricted regions. Most proxies use a pay-as-you-go subscription model: users can register for free, purchase a quota of tokens, and submit inference requests directly.
Why are regulators and AI companies so concerned about these services?
Beyond ordinary use by individual users, the greater concern is that Chinese AI research teams could use these services to extract capabilities from U.S. frontier models at scale, then fine-tune or distill them into their own systems. This would let them avoid comparable investment in foundational research, circumvent export controls, and potentially produce derivative models without adequate safety safeguards.
To what extent do AI proxy services use cryptocurrency?
Cryptocurrencies, particularly stablecoins, allow proxy providers to collect payments outside the traditional banking system, reducing payment friction and limiting customer identity verification. A Chainalysis survey of 100 proxy platforms found that 40% had integrated cryptocurrency payments.
How do proxy providers typically handle on-chain fund settlement?
Most providers do not receive payments directly into their own wallets. Instead, they rely heavily on third-party custodial cryptocurrency payment processors, creating an additional buffer between buyers and operators while circumventing local financial regulations and reducing the risk of funds being traced.
