El mercado sombra detrás del auge de la destilación de modelos de IA en China
Chainalysis investiga cómo un mercado sombra permite a usuarios chinos eludir los controles de exportación de IA de EE. UU. mediante servicios proxy, descubriendo que el 40 % de estos servicios aceptan criptomonedas. El informe destaca un ecosistema creciente habilitado por cripto que podría socavar las restricciones de seguridad de EE. UU.
Puntos clave
- El 40 % de los servicios proxy de IA encuestados aceptan cripto, principalmente stablecoins.
- La actividad de carteras vinculadas a China se dispara tras los lanzamientos de nuevos modelos de EE. UU.
- Los pagos suelen fluir a través de procesadores custodiales, ocultando a los usuarios finales.
- La cadena de suministro de los proxies está en capas, lo que dificulta su interrupción.
- La inteligencia blockchain puede mapear los flujos financieros para atacar puntos críticos de presión.
- Servicios proxy que aceptan cripto
- 40%
- Servicios proxy encuestados
- 100
What the report asks
Chainalysis set out to map the emerging “shadow market” that enables Chinese developers to access U.S. frontier AI models despite export bans. The core question is how these proxy services operate, how they are paid for, and whether cryptocurrency plays a role in evading both U.S. restrictions and Chinese capital controls.
Data and methodology
The Global Intelligence Team surveyed 100 AI proxy providers that sell access to models such as Claude, ChatGPT, and Gemini. They examined payment methods, wallet activity, and the broader supply chain, using blockchain analytics to trace crypto flows. The analysis also compared transaction volumes before and after major U.S. model launches to spot anomalous spikes.
Main findings
Chainalysis estimates that 40% of the proxy services accept cryptocurrency, predominantly stablecoins, allowing users to top‑up accounts without using domestic payment rails. Activity from wallets flagged as China‑linked rises noticeably in the days following a new model release, suggesting rapid attempts to acquire fresh capabilities. Payments rarely go directly to the proxy operators; instead, most funds pass through custodial crypto processors, adding a layer of anonymity and reducing exposure to Chinese regulatory scrutiny.
Caveats and limitations
The report stresses that wallet spikes cannot definitively prove the purpose of the payments, nor can blockchain data alone confirm that the funds are used for model distillation. Moreover, the analysis relies on heuristic tagging of wallets as “Chinese‑linked,” which may include false positives or miss covert actors.
Implications for readers
For policymakers and compliance teams, the findings underscore a new enforcement challenge: blocking direct API access is insufficient when a mature resale ecosystem exists. Blockchain intelligence offers a way to trace the financial lifelines of these networks, identify high‑risk intermediaries, and potentially disrupt the flow of funds that fuels unauthorized AI training. Crypto‑focused firms should be aware that their platforms may inadvertently process payments for services that facilitate export‑control violations, exposing them to regulatory risk.