_________

Momenta has secured a place in the development of new Peugeot and Jeep models for China, Europe and other markets. The Chinese supplier of driver-assistance technology has just inked a strategic deal with the joint venture Dongfeng Stellantis Automotive Technology in Wuhan.

Stellantis gains access to Momenta's software and development speed for its established global brands. Momenta can benefit from access to Peugeot and Jeep’s vehicle programs and global sales reach.

This partnership, if successful, opens a new route abroad for Chinese driver-assistance systems. So far, developers could either hope to have their systems exported in Chinese cars. Or supply driving assistance at home for the China-only models of multinational carmakers.

In our Must Read we examine how the chance to define product development together with an international OEM could give Momenta an edge over the competition, how Chinese AI world models would fit in with French chassis tuning and American off-road experience, and which regulatory and political hurdles await this courageous bet on the future of autonomous driving in Europe.

_________

DeepSeek has released open-source software for Huawei’s Ascend chips, targeting a costly obstacle to adoption of Chinese AI hardware. The package adapts key parts of its AI software infrastructure to Huawei’s platform. It offers a route away from Nvidia’s ecosystem, though independent performance benchmarks are still missing.

For companies considering Chinese chips, the cost of moving software may matter as much as the hardware.

China Southern Power Grid and Unitree have agreed to co-develop robots for dangerous grid work. The partnership gives the robot maker access to demanding tasks and real operating conditions, from inspections to removing ice from high-voltage lines.

Utilities could give robot makers the repeat business and training environments they need to grow.

Robots are helping automate garment production in Ningbo, where manufacturers are testing embodied AI on repetitive factory tasks. Local component makers are also moving into the robot supply chain, building on their existing industrial skills.

Ningbo is testing whether AI can keep traditional industries competitive as wages rise and birth rates fall.

The Chinese Academy of Sciences has launched Panshi ScienceOne to connect research data, AI models, scientific facilities and autonomous laboratories. Nearly 100 organizations have joined its ecosystem, with a program to share computing capacity and tools.

Shared research infrastructure could reduce duplicated work and help companies turn discoveries into engineering applications.

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

_________

_________

China used just 36.8% of its available intelligent computing supply in 2025, according to data from the China Academy of Information and Communications Technology. The government wants the infrastructure in place once the demand rises. It has, however, started to rein in local governments racing ahead too fast.

Source: Zhengquan Shibao (Securities Times)

About 70% of doctors surveyed by Fudan University had seen patients who brought AI-generated health information to their consultations. After “Dr. Google” (or, in China, “Dr. Baidu”), medical doctors now have to learn to live with “Dr. AI.”

Sources: Shangguan Xinwen (Shanghai Observer)

Chinese AI models on OpenRouter’s ranking processed 62.22 trillion tokens in the week of September 21–27, compared with 14.2 trillion for U.S. models. This was their 22nd consecutive weekly lead on that platform.

Source: : Meiri Jingji Xinwen (National Business Daily)

_________

China should devote more regulatory resources to high-risk AI applications and leave lower-risk innovation more room, argues a September 28 column on Renmin Wang, the People’s Daily website. The article presents risk-based oversight as a way to support adoption without applying the same controls to every use of the technology.

Its example is the AI hallucination, a plausible but incorrect answer from a model. In casual conversation, that may impair the user experience. In medical diagnosis, financial services or industrial production, the consequences can be far more serious.

The column cites the 15th Five-Year Plan’s proposals for tiered management of technology applications and AI risk controls across the full life cycle, including monitoring, warnings and emergency response. It also argues for predictable rules and warns that overly detailed requirements imposed too early can restrict experimentation.

China is not neglecting the risks of AI, but believes it can manage them. The tiered approach advocated here will likely be reflected in many laws and regulations governing AI in the months and years to come.

While this is still under discussion, it nevertheless gives companies a useful planning signal. Businesses deploying AI in China should distinguish low-consequence tools from systems whose errors could harm patients, customers or production, and be ready to explain how their controls reflect those differences.

China’s Ministry of Industry and Information Technology and four other agencies are inviting recommendations for national green computing facilities for 2026. The notice covers AI computing, general computing and supercomputing facilities. The Ministry of Industry and Information Technology (MIIT) published it on September 21, 2026.

Applicants must have clear ownership, their own power and cooling systems, and more than a year of continuous, stable operation as a complete facility. Large and very large facilities built after 2022 should, in principle, sit in designated national computing-network hubs or pilot zones coordinating computing and electricity.

Facilities generally need at least 3,000 standard racks and rack occupancy of at least 60%. Public-institution, financial-sector and AI computing facilities are exempt from the minimum rack count. The notice also requires computing-resource utilization at an advanced industry level.

Facilities must reach at least grade 2 under the green data center standard and level 2 under the data center energy-efficiency standard. Their share of renewable electricity must meet or exceed the quota of the province where they operate.

The central government in Beijing is tightening regulations because some data centers in certain provinces have been mostly idle, while others had to reject orders. More oversight is meant to speed up an orderly build-out of AI computing powered by green energy.

Ping An Bank has adopted internal AI management rules that place oversight with its board’s strategy committee. The bank told financial news service Cailianshe on September 30 that the measures implement guidance issued in June by the National Financial Regulatory Administration, the news agency Cailianshe reported on October 1, 2026.

“Whoever uses it is responsible” is one of the principles in the bank’s rules, according to the report. The measures provide for controls throughout the AI life cycle, covering data security, algorithm risk, ethics review and tracing responsibility. AI risk becomes part of the bank’s overall risk management, classified by category and tier.

By the end of June, Ping An Bank had deployed more than 450 large-model application scenarios, including precision marketing, robo-advice, compliance, risk control and office support. The regulator’s guidance requires boards at banks and insurers developing and using AI to designate a special committee to oversee it, the report said.

For AI vendors selling to Chinese banks, technical performance is only part of the requirements. Banks must also be able to govern the systems they buy. Ping An’s approach suggests suppliers should be prepared to document risks, support reviews and help customers trace responsibility when an application fails.

China and the United States have agreed to establish an AI dialogue and a communication channel for AI incidents. The arrangement will allow the two countries to exchange views on the technology’s risks and benefits, according to an October 1 commentary in the newspaper Jingji Ribao, republished by Zhongguo Jingji Wang.

While this agreement was reached during the recent U.S.-Chinese summit in Washington, Beijing still wants the United Nations to lead global AI governance. China is also willing to communicate with Washington through an intergovernmental AI dialogue, said Foreign Ministry spokesperson Guo Jiakun on September 28, as reported by the newspaper Zhongguo Qingnian Bao via Sina. He presented bilateral contact as compatible with a wider international framework.

Both superpowers are signaling that they are aware of the cybersecurity and military risks of AI. They have agreed upon an initial, bare-minimum mechanism to manage potential crises. The U.S., however, has rejected China's proposal for broader joint oversight.

_________

For China, artificial intelligence is a new type of infrastructure. Beijing’s planners place computing capacity alongside the electricity grid, water supply, roads and train lines. That approach was confirmed by the country’s monetary policy this week.

Low-cost credit from the People’s Bank of China is now available to the national “computing network”, alongside five other “networks” such as the electricity grid, underground pipelines and water pipes.

A day earlier, the cabinet led by Premier Li Qiang had called for large-scale construction to begin as soon as possible. Planned spending on this network alone amounts to hundreds of billions of U.S. dollars, most of it to be shouldered by companies.

The plan, in one senior official’s words, is to build “moderately ahead of demand.” The underlying logic is as simple as it is convincing. Just as you may need to build a bridge to a remote valley before businesses can settle there, you need to build data centers to provide industry with the tokens it needs to grow.

Silicon Valley and much of the Western world are fascinated by (and scared of) of frontier models. Beijing focuses on AI in factories, ports and hospitals and builds the capacity now before customers or patients even ask for it.
Should this plan work out, manufacturers in China will have access to AI as a cheap utility. That would also change the competitive position for any foreign enterprise that operates in China or competes with Chinese industry.

How many data centers are currently unused in China while others report zero remaining capacity, how Beijing links computing power with green energy, and what all of this means for foreign manufacturers, is the subject of this week's Deep Dive.