_________

At China’s first Physical AI and World Models Conference in Nanjing, Jiangsu province, international experts including Turing Award winners Yann LeCun and Andrew Yao talked about world models and the future of robotics and autonomous driving.

China wants to move physical AI from research labs into factories as quickly as possible, and Jiangsu is positioning itself as a major supplier of data. A new alliance of ten cities aims to turn the Yangtze River Delta, a manufacturing stronghold, into a national powerhouse for annotating data for machine learning.

Jiangsu has recognized the business opportunity in AI models’ appetite for data. It is putting its industrial base to work as a training ground for robots, factories and vehicles.

The China AI2X Briefing editorial team attended the conference in Nanjing and followed all the presentations, including the video appearances by Yann LeCun and Andrew Yao.

This week’s Must Read examines Nanjing’s plans in detail. We also explore why leading researchers such as Harry Shum describe the human hand as “the last bastion of human dignity.”

_________

SAIC Roewe unveiled the Jiayue 07 on October 23. The company calls it the “world’s first mass-produced AI-native car.” Developed jointly with ByteDance’s Volcano Engine, the SUV uses AI to coordinate the entire car, from the cockpit voice assistant to driving assistance functions and the chassis.

The model illustrates an idea Xiaomi founder Lei Jun expressed at the World Intelligent Connected Vehicles Conference. Combining AI and cars is no longer just about adding functions, he said, but about a “brain transplant.”

A national pilot base for AI applications led by China Mobile unveiled the automotive industry’s first shared AI platform on October 23. It brings together 112.7 EFLOPS of computing power, as well as data and models. Six AI agents are reportedly already in use at more than ten state-owned automakers.

Smaller automakers can test AI in engineering, production and automated driving without having to build expensive infrastructure themselves.

Empyrean Technology, China’s leading domestic provider of electronic design automation (EDA) software, has unveiled its .aix series of AI agent platforms. Each chip design tool is paired with an AI assistant that works around the clock. Some analog circuit design tasks that used to take weeks now reportedly take less than a day.

Faster development processes could help Chinese chipmakers catch up with foreign competitors in the breadth of their product portfolios, China hopes.

Tairex, a startup that emerged from Tsinghua University, has unveiled a a new system with 54 AI doctor agents. They learn the diagnostic reasoning of individual doctors. Then, each doctor gets his or her digital twin. In nine pilot hospitals, almost 200 doctors have already received their own AI twins.

The agents can take over routine tasks and bring specialist knowledge to primary care.

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

_________

_________

On October 26, 2026, China’s Ministry of Industry and Information Technology proposed 34 more smart factories for the highest level of support. They include Xiaomi’s AI-controlled electric vehicle plant. China distinguishes four levels of smart factories. The title “pioneer factory” marks the highest. At the end of September 2026, China had more than 13,000 advanced-level factories, over 500 at the excellence level and just 15 pioneer factories.

Source: Xinhua

This is the computing power of the first shared AI platform for China’s automotive industry. It was unveiled on October 23, 2026, by a national pilot base led by China Mobile. Its first six AI agents are already in use at more than ten state-owned automakers.

Sources: Keji Ribao (Science and Technology Daily)

According to iFlytek, this is the share of patient follow-up care handled by AI at Shandong University’s Qilu Hospital. The company released the figure on October 22, 2026, when it introduced its medical AI model Spark Medical V3.7.

Source: : Gelonghui

_________

China’s State Council has issued guidelines that set out 18 tasks to make AI adoption more employment-friendly, the state-run news agency Xinhua reported on October 23, 2026.

Authorities are to tap AI’s job-creation potential, but mitigate job displacement caused by AI applications and develop a mechanism to assess the employment risks of AI adoption. Workers that lose their job due to AI should get support from the government.

Beijing still wants rapid AI adoption. These legally non-binding guidelines issued by the highest organ of the central government are meant to cushion the negative impact of AI on the job market.

China has set up the National Data Group Company, a central state-owned enterprise in Xiong’an New Area tasked with developing the country’s data resources and serving AI development, the newspaper Jiefang Daily reported on October 27, 2026.

Founded on October 20, the group is backed by China’s three state telecom carriers China Telecom, China Unicom and China Mobile as well as the two grid operators State Grid Corporation of China and Southern Power Grid.

Together, these state-owned companies own vast amounts of telecom and energy data relating to a large share of the Chinese population. While the National Data Administration sets policy, the new state-owned data group is now tasked with turning raw data into marketable data products.

China’s data reforms are moving from rule-making to operations. This step means a further centralization. The government says the state-owned data group was established with the aim of building a “national champion.” Central access to a big amount of data of a population of more than one million people is expected to enable the development of AI applications that can compete with the biggest U.S. frontier developers.

Nanjing’s municipal legislature proposed regulations comprising 62 articles on October 28 to promote the AI industry and its applications, the newspaper Yangzi Wanbao reported on October 29, 2026.

The regulations provide comprehensive support for AI-driven one-person companies (OPC) and help for founders whose ventures fail.

Nanjing attempts to support local AI pilot schemes with binding rules. The planned fund evaluation rule removes a common reason for state-backed investors to avoid risky, long-term AI projects and is also meant to reduce the risks faced by founders of small companies.

_________

GPT-6 Astra was not initially marketed as a robot brain. When OpenAI introduced the model in early September, the company talked mainly about coding and science. Within days, however, videos showed Astra controlling robotic arms without additional training. The surprise has sparked a lively debate in China’s physical AI community.

Some familiar questions have suddenly become urgent again. How much of the physical world can a machine learn from language alone? Will large language models and world models, which learn how the real world behaves, merge? Or will they remain separate modules that complement each other?

Chinese AI researchers and entrepreneurs have discussed these questions at several recent conferences. They are divided into three broad camps. Let us call them the LLM camp, the world-model camp and the camp favoring a division of labor.

In the United States, the debate between two camps has hardened. Sam Altman’s OpenAI is betting that language models will take over robotics. Yann LeCun and the world-model camp see that as a dead end. Many Chinese researchers and robot makers refuse to join either camp and are pursuing hybrid solutions.

In this week’s Deep Dive, we hear from all three camps, explain their arguments and examine what the first tests of GPT-6 Astra in China show.