China grants AI export licenses to DeepSeek and MiniMax

Licensing for major players

Two of China's prominent artificial intelligence companies, DeepSeek and MiniMax, have officially cleared the path to take their models into international markets. According to a post by Yuyuan Tantian, a WeChat account linked to the state broadcaster CCTV, both firms completed the necessary registration to secure open-source and overseas licensing rights.

These approvals come in the wake of a policy announcement on October 16 regarding China's AI export controls. While the regulations are designed to manage the flow of technology, the state media outlet noted that most AI technology will remain free to export. The licensing process ensures that these specific companies can operate within the legal framework of the new rules.

Neither DeepSeek nor MiniMax provided an immediate response to requests for comment regarding the licensing. However, the move highlights a growing trend of major Chinese firms seeking formal state approval to participate in the global AI economy. This process allows them to navigate the complexities of national security requirements while pursuing international growth.

The three pillars of control

The commerce and science ministries are working to add 23 specific AI technologies to a list of items that are banned or restricted from export. These categories include large models, agents, and computing clusters. The government intends to use three primary pillars to determine what can leave the country: data, sensitive use, and computing power.

Regarding data, the rules would ban the export of large models trained on China's "core data." Models trained on "important data" would be restricted regardless of their size. This suggests a focus on protecting specific national information from being integrated into models that might be used by foreign entities.

Sensitive use and computing power form the other two pillars of the framework. Agents that exceed a certain performance level or are built for defense, core government affairs, or critical infrastructure face a ban. Additionally, the rules target AI chip design and high-speed links for large computing clusters, setting specific thresholds for what constitutes a restricted model.

Defining restricted models

The government has established specific technical numbers to identify which large models fall under these export restrictions. The catalogue includes models with training compute above 1026 floating-point operations or those with more than 2 trillion parameters. It also covers models that possess capabilities ranking in the global top ten.

By setting these specific metrics, the government aims to create a clear boundary for what constitutes a high-end AI capability. This allows for a distinction between general-purpose AI and the more powerful systems that could have significant strategic implications. Companies must register before they can export any technology that meets these criteria.

The rules are designed to be specific rather than broad. By defining the technical limits of parameters and compute, the state can monitor the movement of the most powerful technologies while allowing smaller or less capable models to move more freely. This distinction is central to the current regulatory environment in China.

Commercial paths and exemptions

Despite the new restrictions, the policy includes several exemptions to ensure that commercial activities are not stifled. The rules specify that offering a restricted model through API calls, web pages, SaaS, or lightweight local deployments does not count as a technology export. This means that many standard ways of providing AI services to users remain unaffected.

Open-source technology is also granted automatic licensing, which the state media described as "immediate reporting and export." Similarly, Chinese companies' own computing centers located abroad, chip tape-outs overseas, and joint research on international standards are permitted. Technology that is already public through academic papers or open-source releases also falls outside the new rules.

The goal of these exemptions is to allow AI companies to continue their commercialization efforts as usual. By exempting common service models like APIs, the government allows for the continued use of Chinese AI in global business applications. This creates a dual-track system where the underlying high-end technology is controlled while the resulting services remain accessible.

A different path from the West

The state media outlet noted that China's approach is distinct from the strategies used by the United States. The post specifically cited the US Commerce Department's AI diffusion framework from January 2025 and an "alliance strategy" attributed to Anthropic's CEO, Dario Amodei. The report suggested that China did not take that specific path in its regulatory design.

Legal experts who participated in the consultations, including Liu Jing and Liu Hongmei of Beijing Dacheng Law Offices, noted that the rules were refined based on company feedback. Liu Hongmei stated that restricted technologies can still go abroad with a license, emphasizing that the rules would not disrupt the overseas plans of Chinese companies.

The regulatory environment is evolving as overseas internet users have reportedly urged other Chinese open-source models to follow the lead of DeepSeek and MiniMax. As the public comment period for the new rules closes on November 15, the industry will watch closely to see how these licenses are applied to other major players. The focus remains on balancing the export of high-end capabilities with the broader goal of international commercial expansion.