Chinese artificial intelligence developers DeepSeek and MiniMax have secured state approval to offer their AI models for use around the world, a decision that signals Beijing is willing to let its homegrown technology compete on the international stage even as it tightens its grip on the most sensitive pieces of the AI stack. The approval was reported by a social media account affiliated with state broadcaster CCTV, which said the green light covers global deployment of the two companies’ models. Neither DeepSeek nor MiniMax responded immediately to requests for comment, leaving the precise scope of the approval and the conditions attached to it largely unspoken.
The news landed alongside another detail that may prove just as telling. In a separate post, the same CCTV-affiliated account, Yuyuan Tantian, noted that internet users overseas had begun pressing other Chinese open-source large language model developers to pursue similar approval. That observation cuts against the assumption that demand for Chinese AI flows only in one direction, driven by state promotion rather than genuine interest. Developers and businesses abroad have spent the past two years watching Chinese labs release capable models at prices that undercut their Western counterparts, and the enthusiasm for those tools appears to be creating its own momentum. Whether regulators in Beijing see that enthusiasm as an opportunity or a risk is the question that now hangs over the sector.
The approval did not arrive in a vacuum. Just a week earlier, China proposed export controls on 23 AI technologies, a list that includes models, agents and computing clusters. The commerce ministry framed the proposal with care, stressing that most AI technologies would remain exportable and that open-source exports would face simplified licensing, particularly for Chinese companies operating their own computing centres overseas. The proposed restrictions would target overseas deployment of AI models containing restricted technology, part of a broader effort to keep tools deemed vital to national security from flowing too freely amid intensifying competition with Washington. Read together, the two developments describe a deliberate two-track approach: open the door for models to travel, but keep a firm hand on the infrastructure and know-how that give those models their strategic value.
For anyone who has followed China’s AI sector, the timing carries a certain logic. The country has spent years cultivating open-source model development, and companies like DeepSeek have earned international attention for releasing capable systems at remarkably low cost. That strategy has paid dividends in reputation, if not always in revenue. Allowing global use of those models while simultaneously tightening controls on sensitive technology lets Beijing claim the mantle of openness while preserving its leverage over the pieces of the stack it considers non-negotiable. It is a familiar playbook in Chinese technology policy, and it has generally served the country well in areas from solar manufacturing to electric vehicles.
Yet the balance is easier to declare than to maintain. Approvals for global use could accelerate adoption of Chinese models among international developers and enterprises looking for affordable alternatives to American offerings, but export controls, even simplified ones, add friction and uncertainty. A developer in Southeast Asia or Europe weighing whether to build on a Chinese model now has to factor in the possibility that licensing rules could shift, that a model approved today might face new conditions tomorrow. That kind of ambiguity has a way of cooling enthusiasm, particularly among businesses that need predictable access to the tools their products depend on.
The companies themselves sit in an awkward position. DeepSeek and MiniMax have built their reputations on openness, publishing research and releasing weights that allow outsiders to inspect and adapt their work. That openness is precisely what made them attractive to the global developer community, and it is also what makes them subject to scrutiny from regulators who see unrestricted dissemination as a vulnerability. The approval to operate globally is a vote of confidence, but it also binds the companies more tightly to a regulatory framework that can expand or contract with the political weather.
Overseas users urging other Chinese developers to follow DeepSeek and MiniMax’s path may find that approval is granted case by case, shaped by considerations that have as much to do with diplomacy as with technology. The Chinese government has shown that it is comfortable moving quickly when it sees strategic advantage, and equally comfortable slowing down when it does not. For developers watching from outside, the lesson is that access to Chinese AI is likely to remain conditional, subject to reviews that are rarely transparent and always political.
What remains genuinely unresolved is whether China’s openness to exporting models can coexist with its instinct to control the technologies it considers essential. The commerce ministry’s assurances that most AI technologies remain exportable are reassuring on their face, but the list of 23 restricted technologies is broad enough to give regulators considerable discretion. How that discretion is exercised over the coming months will determine whether the approval granted to DeepSeek and MiniMax becomes a template for a more open era or an exception that proves the rule.
There is also the matter of public perception, which cuts in several directions at once. Supporters of open-source AI will see the approval as a win, proof that Chinese labs can compete globally without sacrificing the collaborative ethos that animates the field. Skeptics will note that state approval is not the same as state neutrality, and that a model permitted to operate globally can just as easily be restricted later. And users themselves, the developers and small teams who ultimately decide which models gain traction, tend to care less about geopolitics than about whether a tool works, whether it is affordable, and whether it will still be available next year. On all three counts, the picture remains incomplete.
For now, the approval stands as a meaningful moment for two companies that have done more than most to shape China’s reputation in open-source AI. It suggests that Beijing sees value in letting those companies compete abroad, and that it believes the benefits of global adoption outweigh the risks of proliferation, at least for now. But the broader contest between openness and control is far from settled, and the next round of export control decisions, licensing reviews and company announcements will say more about China’s direction than any single approval ever could.
