OpenAI and Google sell AI to China via Singapore, sidestepping US sanctions

Three major Chinese tech companies on the US Department of Defense blacklist have found a way to buy advanced AI from OpenAI and Google. The trick is simple: they buy through Singapore, where export controls are looser than in the United States itself. The companies purchase access to cloud APIs via Singaporean intermediaries, so the American providers technically stay within the law.

AI-processed from TNW; edited by Hamidun News
OpenAI and Google sell AI to China via Singapore, sidestepping US sanctions
Source: TNW. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTNW ↗
Editor-in-Chief
Key metric #1Three companies on the US blacklist get access to GPT and Gemini
Key metric #2Direct supply is banned, but secondary resale is permitted
Key metric #3Singapore acts as the intermediate link thanks to more flexible regulation

Three Chinese technology companies on the US blacklist continue to get access to advanced AI from OpenAI and Google. The trick is simple: they buy through Singapore, where export restrictions are less strict.

How the sanctions loophole works

US export control law explicitly prohibits supplying AI services to blacklisted companies. However, it does not regulate secondary sales through third countries.

The scheme works like this: a company in Singapore obtains a license and access to the OpenAI or Google API. It then re-aggregates that access: end users connect through the Singaporean intermediary rather than directly. For the American providers, they are formally selling a cloud service to a legitimate Singaporean company. What that company does with the access afterwards is something for which their responsibility is minimal.

  • Three companies on the US blacklist get access to GPT and Gemini
  • Direct supply is banned, but secondary resale is permitted
  • Singapore acts as the intermediate link thanks to more flexible regulation

Why cloud services are hard to control

The core problem lies in the architecture of modern cloud platforms. OpenAI and Google do not closely monitor the end users behind each API request. The system is simple: you pay for tokens, you use them.

An intermediary can buy huge volumes and resell access to many customers. The provider is not obliged to track who stands behind the end usage. The alternative is a geographic block at the IP-address level or a requirement for additional verification. But that would slow the service for legitimate users and could hurt business in Singapore.

A systemic export control problem

History shows that technology sanctions work only if they are coordinated. During the Cold War, the US coordinated chip export controls through the COCOM alliance (later the Wassenaar Arrangement). No such mechanism exists for AI today.

Without the agreement of Singapore, India, the UAE and other neutral countries on common rules, the loopholes will remain. And China will use every one of them.

What this means

The current AI export control system has turned out to be leakier than the US government assumed. Cloud architecture, anonymous API tokens and the presence of intermediary countries create gaps through which China gains access to advanced models. If the US and its allies really want to restrict a competitor, they need tougher technical enforcement of controls or international agreement.

Story

Key Questions Answered

How the sanctions loophole works?

US export control law explicitly prohibits supplying AI services to blacklisted companies. However, it does not regulate secondary sales through third countries. The scheme works like this: a company in Singapore obtains a license and access to the OpenAI or Google API. It then re-aggregates that access: end users connect through the Singaporean intermediary rather than directly. For the American providers, they are formally selling a cloud service to a legitimate Singaporean company. What that company does with the access afterwards is something for which their responsibil

Why cloud services are hard to control?

The core problem lies in the architecture of modern cloud platforms. OpenAI and Google do not closely monitor the end users behind each API request. The system is simple: you pay for tokens, you use them. An intermediary can buy huge volumes and resell access to many customers. The provider is not obliged to track who stands behind the end usage. The alternative is a geographic block at the IP-address level or a requirement for additional verification. But that would slow the service for legitimate users and could hurt business in Singapore.

What are the key details on A systemic export control problem?

History shows that technology sanctions work only if they are coordinated. During the Cold War, the US coordinated chip export controls through the COCOM alliance (later the Wassenaar Arrangement). No such mechanism exists for AI today. Without the agreement of Singapore, India, the UAE and other neutral countries on common rules, the loopholes will remain. And China will use every one of them.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…