China locked its AI models, then cleared DeepSeek and MiniMax for the world
China proposed strict AI export controls, then immediately let DeepSeek and MiniMax register for global use. It is a strategy, not a ban.
Policy · Source: Reuters
What happened
On October 16, China proposed adding 23 AI technologies to its restricted export list. This draft catalogue covers large models, AI agents, computing clusters, and information processing. A week later, state media account Yuyuan Tantian reportedly disclosed that DeepSeek and MiniMax completed a new registration process. This process was introduced alongside the proposed update. It allows the two model developers to proceed with open-source releases and overseas licensing while the revision remains open for public comment.
The Ministry of Commerce drew clear lines on what actually counts as an export. Serving a restricted model via API calls, web apps, or software-as-a-service products does not count as exporting large-model technology. The ministry gives the same treatment to lightweight deployment on a device. Furthermore, technologies already made public through open-source releases or academic papers are proposed to sit entirely outside export management. The ministry stated that the great majority of AI technologies will remain freely exportable.
Local deployment overseas is where the rules tighten. MiniMax established a formal application route for regions excluded by its earlier model license, including the United States, the European Union, the United Kingdom, and South Korea. An organization in these regions must pass a compliance review and obtain a license before deploying the models locally. Meanwhile, DeepSeek moved fast to test the open-source exemption. They released DeepInfer, their large-scale inference framework, and a companion toolkit called deepseek-recipe right in the middle of this consultation period.
Key facts
- 23 — AI-related items added to the draft export catalogue
- October 16 — Date the Ministry of Commerce opened the draft revision for public comment
- DeepInfer — The large-scale inference framework open-sourced by DeepSeek
Why it matters
If you build products with Chinese open-weight models, your access is not disappearing anytime soon. You can still hit their APIs or use their web interfaces without triggering export controls. The Chinese government is making it incredibly easy for their top AI labs to serve the global market remotely. They want global reliance on their infrastructure. This is a deliberate strategy to ensure Chinese models remain competitive and widely adopted by developers worldwide. They are choosing market share over isolation.
The friction hits enterprise adoption and local infrastructure. If you want to deploy a restricted model like MiniMax locally on your own servers in the United States or Europe, you now face a formal application route. You have to pass a compliance review from the Chinese side. This creates a massive wedge in the market. Startups and indie hackers will keep using these models via API because it is cheap and accessible. But large western enterprises will hesitate to put their infrastructure and data through a Chinese compliance review just for local hosting. This splits the market into remote consumers and heavily regulated local deployers.
For builders
Safe to build on Chinese APIs
The draft rules explicitly exempt API calls and SaaS products from export restrictions. You can integrate DeepSeek or MiniMax via API without worrying about sudden cutoffs from the Chinese government. You pay for the API usage, and you get the intelligence without the regulatory headache.
Local deployment requires compliance reviews
If your enterprise clients demand local hosting of MiniMax models in the US or EU, you face a massive bottleneck. You must apply for a license and pass a compliance review. This adds severe legal overhead and delays for enterprise builders who need on-premise solutions.
Open source remains untouched
Open-source releases are proposed to sit outside export management entirely. DeepSeek just dropped DeepInfer to prove this framework works in practice. Builders relying on open weights and open frameworks win here because the code remains freely accessible.
My take
I look at this and see a masterclass in geopolitical strategy. China is not locking its AI models away from the world. They are locking the world into their AI ecosystem on their exact terms. They give you the API for free innovation but put a tight leash on enterprise local deployments.
Original reporting: Reuters. This is my rewrite and opinion.