_________

China seeks tighter control over AI technologies its companies export. The government suggested adding 23 AI technologies to its export-control list, including models, agents and computing clusters. It plans to keep overseas access open via apps, APIs and SaaS-applications, however.

The planned regulations, which have been published as a draft and are currently open for public comments until Nov. 15, would add a number of core AI technologies to China's technology export catalog. They are part of a broader set of new technologies, including biomedical and photovoltaic ones, which are all new on the list. AI accounts for 23 of the 40 proposed additions.

The new AI export controls distinguish between technologies deemed relevant to China's national security and economic development and less sensitive AI services made available abroad. Exports of AI models trained on “core data” would be prohibited, while the export of others surpassing certain capability thresholds (2 trillion parameters, for example) would need licenses.

Access to Chinese AI models such as DeepSeek or Qwen via API or consumer-facing apps does not fall under the new export controls. Officials stress that the proposed regulatory changes are not an AI “export ban,” just an update of the existing tech export catalog, and that most of the country's AI will remain free to export.

By exerting tighter control over which AI technologies can be exported, Beijing wants to protect national security and industrial expertise without hindering the overseas expansion of its AI developers.

The new control measures come amid an intensifying rivalry with Washington over AI. While there is a temptation to misread these new export controls purely as a reaction to U.S. boycotts of AI chips and other sanctions, they follow an internal Chinese logic which precedes both Trump administrations and the U.S. chip war by decades.

Our Must Read this week explores why China believes that such controls are needed to preserve its sovereignty, after once falling behind in science and technology and being partially colonized, and why the new export catalog is better understood as an extension of Deng Xiaoping's policy of reform and opening since 1978 and his quest for a strong China. It is those principles that are recognizable in the choice of concrete AI technologies and specifications that fall under the new licensing regime.

_________

At the Paris Automotive Summit on October 15, XPeng launched XPENG AI Engineering, a new business that lets automakers worldwide license its full vehicle platforms, individual modules such as its driving system, AI chip and fast-charging battery, and its production toolchains.

Under pressure at home, Chinese EV makers are starting to sell their AI know-how, not just their cars. XPeng points to its partnership with Volkswagen as proof that the model works.

XHEART, a physical-AI chip designer less than three years old, has closed a strategic funding round. Renault invested for the first time, Chery added to its stake, and memory maker CXMT joined as well.

With memory prices soaring this year, access to memory increasingly decides chip contracts. CXMT already mass-produces automotive-grade memory and could give XHEART an edge on price and supply.

State-owned CHN Energy has built an AI power trading platform that supports the whole workflow, from price forecasting to bid submission. In a two-month trial, the AI bought power more cheaply than human traders on 59.5% of trading days.

China is building a unified national power market by 2030, with more complex rules and wider price swings. For generators, AI-assisted trading is becoming a core capability.

At the CIIF industry fair in Shanghai, Black Lake Technologies presented more than ten AI agents that handle tasks from quoting and scheduling to quality inspection. The company claims its platform is already used in 30,000 factories.

Co-founder Xiao Zhe says the strongest demand comes from small and medium-sized manufacturers, who are willing to pay as long as the AI solves real problems on the shop floor.

For details and more analysis, please click here to read the full, free version of this week’s Highlights:

_________

_________

DeepSeek suffered an outage lasting around five hours on October 15. The disruption left its V4-Pro and V4.1-Flash models and chat services unavailable.

Source: Nikkei Asia

The 140th Canton Fair is displaying 235,000 AI products. More than 20,000 exhibitors use industrial internet, AI and other new technologies in their businesses.

Sources: Meiri Jingji Xinwen

Nvidia’s share of China’s market for data-centre AI chips fell to 34 percent in the first half of 2026, according to IDC. Huawei followed with 28 percent as domestic suppliers gained ground.

Source: : IDC

_________

China’s Ministry of Commerce and Ministry of Science and Technology proposed adding 40 technologies in AI, biomedicine and photovoltaics to the country’s technology export catalog on October 16, 2026. Fourteen would be prohibited and 26 restricted, newspaper Guangming Ribao reported on October 18.

The AI proposals distinguish between tools that generate and modify malicious code, which would be banned from export, and AI-assisted vulnerability discovery, which would be restricted. Alongside the consultation, the ministries introduced interim pre-export registration requirements for the relevant technologies.

Companies taking AI technology abroad need to distinguish between the proposed controls, which could take months before they come into effect, and the interim registration requirements already introduced. The latter add an administrative step before covered exports while the final scope of the catalog remains under consultation.

China’s Ministry of Industry and Information Technology (MIIT) released a compilation of more than 1,000 industrial AI application scenarios on October 15, 2026. It covers 20 subsectors, including petrochemicals, steel, automobiles and consumer electronics, science newspaper Keji Ribao reported that day.

Compiled by research institutions and companies, the map draws on actual manufacturing practice. It addresses a practical obstacle to wider adoption. Experts cited in the report point to companies blindly following trends, duplicating projects and struggling to transfer successful applications to other settings.

The ministry has also organized nearly 30 events matching industrial AI suppliers with potential users, involving more than 5,000 enterprises.

Manufacturers gain a more concrete starting point for choosing AI projects. The commercial test is whether an application can be adapted to another production environment at a reasonable cost. A catalog of examples helps identify opportunities. It does not remove the implementation hurdle.

A new framework for evaluating AI security in industry and information technology was released on October 15, 2026. Led by the China Artificial Intelligence Industry Alliance and compiled by 31 organizations under MIIT guidance, it covers models, applications and their operating environments, according to an October 19 report from the China Academy of Information and Communications Technology.

The framework addresses risks involving data, models and infrastructure, as well as AI agents, embodied AI and specific industrial and telecommunications uses. Its methods combine benchmark, adversarial and scenario-based testing. The evaluation process includes checking results, correcting problems and retesting.

The framework offers industrial buyers guidance for examining an AI system in the setting where it will operate. Suppliers can use it to document how they identify and remedy security weaknesses across the system. A strong model benchmark alone would leave part of that assessment unanswered.

China’s Ministry of Commerce and eight other departments have issued guidelines for using technology to strengthen commerce and domestic demand. The policy, published on the ministry’s website on October 19, calls for AI and other emerging technologies to support the development of the entire commercial and trade value chain.

The guidelines cover goods consumption, services, commercial distribution and future business models. They place AI alongside big data, the Internet of Things, cloud computing, blockchain, digital twins and embodied intelligence. Separate sections address the business environment and measures to support implementation.

Beijing is connecting technology adoption to its effort to strengthen the domestic market. For AI suppliers, that widens the potential field beyond factory productivity to the businesses that sell, distribute and service goods.

_________

Wang Xingxing admitted a big mistake in front of a big audience. The founder of Unitree Robotics, probably China's best-known humanoid robot maker, stood on stage at the China International Industry Fair when it opened on October 12. He had not believed in AI enough, he said.

His confession came during a roller-coaster ride for his shareholders that week. Since Unitree's listing on Shanghai's STAR Market on August 19, the company temporarily lost 281.1 billion Yuan (about 37.4 billion Euro) in market value, observers noted. That was a drop of more than 60 percent from its post-IPO peak.

Similarly inflated, then rapidly collapsing valuations had already set off alarm bells in Beijing. Just days before this speech, Beijing had issued new guidelines to prevent an AI bubble.

The central government published new guidelines warning officials that they will be held accountable if they “rush headlong into mass action or make blind investments.” In the same document, the Chinese leadership confirmed its “AI Plus” initiative to rapidly deploy industrial AI.

While an “AI bubble” is a real concern in China just as it is in other countries, both the famous robotics founder and the central leadership reached the same conclusion.

While hyped expectations distort the financial market, the potential of physical AI is real and its progress will only accelerate. What is needed is a better investment focus placing quality over quantity.

➤➤➤ Read China Insights Weekly