Not a loophole: Singapore AI export controls let China tap US AI legally

American AI technology is reaching Chinese tech giants through a route that US export controls were never designed to close: Singapore. The city-state sits outside the geographic scope of US restrictions targeting Mainland China, and that single regulatory gap has quietly made it a pivotal node in the global AI supply chain — one where Alibaba, Baidu, and Tencent can access frontier American models that would otherwise be off-limits to them.

Key takeaways

• US export controls restrict AI technology exports to Mainland China but do not apply to Singapore, creating a legal gap that Chinese tech firms are actively exploiting.
• Alibaba, Baidu, and Tencent use their Singapore-incorporated subsidiaries — legally treated as Singaporean companies — to access American AI models.
• Alibaba Cloud already offers OpenAI-compatible APIs routed through its Singapore infrastructure.
• OpenAI committed over S$300 million (~$234 million) to establish an applied AI lab in Singapore in 2026; Google DeepMind opened a regional research hub there the same year.
• The US Commerce Department could broaden Entity List restrictions to cover subsidiaries of blacklisted firms in Singapore — a regulatory shift that would disrupt current arrangements overnight.

Singapore as a Neutral AI Hub

Singapore’s appeal here is not accidental. It has spent years positioning itself as Southeast Asia’s premier technology and financial center, attracting global companies with stable governance, strong rule of law, and deep infrastructure investment. What it did not anticipate — or perhaps did — is that this neutrality would place it at the center of a geopolitical fault line in AI.

US Export Controls and Geographic Scope

The architecture of US export controls on AI is more selective than it appears. Restrictions target specific entities and specific geographies. Mainland China is firmly in scope. Singapore is not. That distinction, straightforward on paper, has enormous practical consequences when Chinese cloud providers have been quietly building out their Singapore operations for years — data centers, engineering teams, and now API offerings.

The controls were designed to prevent the most sensitive American technology from flowing directly into Chinese hands. But the policy targets entities, not capabilities. As a result, the capabilities flow through whichever entity is not on the restricted list — and Singapore-based subsidiaries qualify.

Legal Status of Singapore Subsidiaries

This is where the legal distinction becomes decisive. A Singapore-incorporated subsidiary of a blacklisted Chinese firm is, on paper, a Singaporean company. It operates under Singaporean law, pays Singaporean taxes, and can enter contracts that its parent in Shenzhen or Hangzhou legally cannot. Under current US rules, selling an AI model to that subsidiary does not trigger the same restrictions as selling directly to the Chinese parent.

That gap is real, and it is being used.

Chinese Tech Giants’ Access to American AI Models via Singapore

All three of China’s dominant tech platforms — Alibaba, Baidu, and Tencent — maintain substantial operations in Singapore. The question of whether those operations serve genuine local business needs or primarily function as a routing mechanism for restricted technology is one that US policymakers are increasingly asking.

Alibaba Cloud’s OpenAI-Compatible APIs

The most concrete example is Alibaba Cloud’s offering of OpenAI-compatible APIs through its Singapore infrastructure. Developers building on Alibaba’s platform can access models architecturally identical to what OpenAI sells directly — routed through a Southeast Asian intermediary rather than sourced from San Francisco. The practical effect is that Chinese developers get access to frontier AI capabilities through a compliant channel.

This matters beyond just Alibaba. It signals that Chinese platforms are embedding interoperability with American models into their core infrastructure, not treating it as a temporary workaround.

Operations of Baidu and Tencent in Singapore

Baidu and Tencent have followed similar paths, expanding Singapore footprints that provide the same structural advantage: a locally incorporated entity that can enter into agreements unavailable to the Chinese parent. The extent to which each is accessing American AI models through these structures is less publicly documented than Alibaba Cloud’s API offering, but the legal framework enabling it applies equally to all three.

Major AI Investments in Singapore

The irony is that American AI companies have been actively deepening their own Singapore presence at the same time, making the city-state an even more important node in the global AI ecosystem.

OpenAI’s Applied AI Lab

In 2026, OpenAI committed over S$300 million — approximately $234 million — to establish its first applied AI lab outside the United States in Singapore. That investment signals OpenAI’s own strategic interest in the region, not just as a distribution point but as a genuine base for applied research and development.

Google DeepMind’s Regional Research Hub

Google DeepMind opened a regional research hub in Singapore in the same year. The concentration of top-tier American AI infrastructure in a jurisdiction that Chinese tech subsidiaries can freely access is not a coincidence — it reflects Singapore’s deliberate positioning, but it also creates structural conditions that complicate US export control enforcement.

Export Control Circumvention and Regulatory Risks

Microsoft’s Azure Model Distribution Inside China

Perhaps the most instructive precedent is Microsoft. The company has maintained partnerships that allow it to offer OpenAI-powered models inside China itself, despite the restrictions that prevent OpenAI from operating there directly. Microsoft holds exclusive commercial licensing rights to OpenAI’s models, and its Azure cloud platform serves as the delivery mechanism. Through existing Chinese partnerships, Microsoft can distribute those models in ways OpenAI structurally cannot.

This creates a dynamic where the same US policy simultaneously restricts Chinese access to American AI and enables it — the outcome depending entirely on which corporate structure is doing the selling.

US Export Controls Targeting Entities, Not Capabilities

That dynamic is not a loophole in the colloquial sense — it is a direct consequence of how the controls are written. The policy targets entities listed by the US Commerce Department, not the capabilities themselves. So capabilities migrate to unlisted entities. The Singapore subsidiary model is the most visible expression of this, but the Microsoft Azure arrangement demonstrates that the same principle applies even within China’s borders when the right corporate structure is in place.

Potential US Commerce Department Regulatory Changes

The chip export controls offer a sobering parallel. They started narrow and expanded significantly over multiple iterations as the Commerce Department closed off workarounds that emerged after each round of restrictions. If the department decides that selling AI models to Singapore subsidiaries of blacklisted Chinese firms violates the intent of its Entity List restrictions, the entire current arrangement could collapse without warning. That is not a remote scenario — it is the documented trajectory of how chip controls evolved.

Investment Implications of Regulatory Uncertainty

For investors with exposure to AI-adjacent assets — whether in Alibaba Cloud, Microsoft’s Azure business, or companies building on OpenAI’s API ecosystem — the Commerce Department’s rulemaking calendar deserves close attention. The current arrangement generates real commercial value for multiple parties. But its continuation depends on regulators not acting, and the history of US tech export controls suggests that regulatory inaction has a shelf life.

The underlying tension is structural: US export controls on AI were designed for a world where frontier models lived on servers in American data centers. The shift to cloud distribution, API access, and global subsidiary networks has made that geography much harder to police. Singapore’s emergence as the central node in this system is less a deliberate evasion than an artifact of how modern AI infrastructure actually works — and that is precisely what makes it so difficult to address without disrupting legitimate commerce on both sides.

FAQ

Why does Singapore serve as a hub for Chinese firms to access American AI models?

Singapore is not subject to US export controls restricting AI technology to Mainland China. Subsidiaries incorporated in Singapore are legally treated as Singaporean companies, which means they can enter into agreements with American AI providers that their Chinese parent companies cannot.

How do Alibaba, Baidu, and Tencent access American AI models despite US restrictions?

They use their Singapore-incorporated subsidiaries, which fall outside the scope of US Entity List restrictions. Alibaba Cloud, for example, already offers OpenAI-compatible APIs hosted through its Singapore infrastructure, giving developers on its platform access to models architecturally identical to OpenAI’s direct offerings.

What regulatory risks exist regarding AI technology exports through Singapore?

The US Commerce Department could broaden its Entity List restrictions to explicitly cover subsidiaries of blacklisted firms operating in neutral jurisdictions like Singapore. Given how chip export controls expanded over time to close similar workarounds, this is a credible risk that could disrupt current AI distribution arrangements rapidly.

How does Microsoft distribute OpenAI models inside China despite export restrictions?

Microsoft holds exclusive commercial licensing rights to OpenAI’s models and uses its Azure cloud platform and existing Chinese partnerships to distribute them within China. Because US controls target entities rather than capabilities, Microsoft’s unlisted Chinese operations can distribute models that OpenAI itself is prohibited from selling there directly.

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.