_________

Foldable smartphones need AI to become really useful. That is the trend emerging from three premium model releases by top smartphone makers within 72 hours.

Huawei launched a tri-fold device that expands into a tablet-sized screen. Xiaomi introduced a similar smartphone. Shortly after that, Apple revealed its first foldable iPhone, the iPhone Duo.

While the foldable screens captured most of the attention, the three companies all emphasized their progress in AI. They highlighted AI agents that operate on behalf of users and showcased in-house chips capable of running AI directly on the device.

Analysts describe it as a "system-level" competition. Such AI agents can fill shopping carts with gifts, compare prices, and book restaurant tables using voice commands.

Foldable or larger screens alone cannot justify prices of roughly US$1,600 to US$3,000. The devices must prove their usefulness. Foldables must be combined with AI that saves users time if they are to become a new type of computing device. This makes chips and operating systems as crucial as displays.

_________

UBTech Robotics has launched a new factory in Liuzhou where robots manufacture other robots. Some reports suggested that the facility can produce one robot every ten minutes, aiming for over 10,000 units annually.

These figures describe the plant’s design capacity, however. Four Cruzr humanoids work on the plant’s own shop floor, unloading pallets, stacking goods, and carrying material to the line. UBTech is growing very fast, but it sold only 1,079 full-size humanoids in 2025.

Skeptical observers stress that a factory with a 10,000-unit capacity incurs costs before orders arrive. UBTech reported 2 billion yuan (about US$296 million) in revenue for 2025 but posted a net loss of 700 million yuan (about US$104 million).

The new plant shows that humanoid production is becoming a serious industrial process.

Empyrean Technology, China’s top electronic design automation (EDA) software maker, is shifting from “AI-assisted tools” to “AI-autonomous design,” said Chairman Liu Weiping. Empyrean’s Andes FPD tool now automatically generates display panel layouts, reducing a four-week task to one week, according to Liu.

Future pricing may depend on AI agent usage, Liu noted. The Ministry of Industry and Information Technology released an “AI + Software” action plan, calling for deeper AI integration with industrial software, including EDA.

U.S. export restrictions have pushed Beijing to seek self-sufficiency in semiconductors, including control over chip-design software.

BAIC Foton unveiled its first AI mobile medical vehicle, designed to carry AI-assisted diagnostics to remote areas in China.

The 30-square-meter bus houses ten AI-powered medical applications, including fundus screening, ECG monitoring, ultrasound scans and an AI agent for digital radiography.

Trained on a domestic dataset of 4 million cases, the agent is designed to handle everything from image capture to report writing, effectively placing a radiologist on board.

For BAIC Foton, a commercial-vehicle maker, the bus tests a business model built on software.

UniForce AI, a Shanghai startup, raised nearly 50 million yuan (about US$7.4 million) in an angel-plus round led by the Shanghai Future Industry Fund and the city’s Science and Technology Innovation Investment Group. The company specializes in AI for engineering, with nuclear fusion as its first field of application.

Traditional plasma simulations take hours or days. UniForce AI trains neural operators and diffusion models to predict directly how physical fields evolve, and claims they work tens to hundreds of times faster.

If these models prove themselves in fusion research, UniForce AI plans to expand into oil and gas, nuclear power, electronics, and aerospace.

CaoCao, the ride-hailing platform backed by the carmaker Geely, now allows passengers in Beijing, Hangzhou, and Suzhou to order rides through Doubao, ByteDance’s AI app.

Users enter their destination by text or voice. Doubao plans the route, offers vehicles and prices, and processes payment within the app.

This week’s Must Read explores how AI agents are becoming new entry points for online purchases and how China’s superapps are pushing back.

This service brings CaoCao orders from AI channels, reducing reliance on traditional ride-hailing aggregators.

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

_________

_________

AI models handled more than 500 trillion tokens a day on average as of June 2026, the Ministry of Industry and Technology said on September 11, 2026.

Source: Ministry of Industry and Information Technology

Hebei province showcased 100 model use cases for "AI+Manufacturing" at the 2026 China Computing Conference in Langfang on September 12. The iron and steel sector alone features 32 cases, robotics has 12, and chemicals account for 7. The list also includes examples from the pharmaceutical, building materials, and electronics sectors.

Sources: Hebei Daily (Hebei Ribao)

Agricultural Bank of China, Bank of Beijing, and four other banks have extended nearly 2 billion yuan (around US$296 million) in credit lines to AI firms in the Beijing Economic and Technological Development Area.

The "Yiqi Token Loan" program evaluates borrowers based on token generation and consumption and the value of their computing service contracts.

Traditional loans rely on collateral and profits, which many tech companies still in development lack. These represent the first token-based loans in Beijing, reported Beijing Daily on September 14.

Source: Beijing Daily (Beijing Ribao)

_________

China’s Supreme Court has introduced initial guidelines for AI-related lawsuits, although the country still lacks a dedicated AI law. On September 7, the court released its Opinions on the Adjudication of Disputes Involving Artificial Intelligence in Accordance with the Law at a press conference.

The document has 24 articles. It covers unauthorized AI face swaps, voice cloning, AI hallucinations and doxing. It also addresses discriminatory pricing based on customer data, plagiarism in AI-generated content, infringement involving open-source software and liability for autonomous driving accidents.

In May 2023, criminals in Baotou exploited AI face-swapping to impersonate someone during a video call, stealing 4.3 million yuan (about US$630,000) in just 10 minutes, according to the party newspaper Renmin Ribao.

The Opinions assign differentiated liability to developers, platforms and users.

This compliance document indicates the route formal legislation will most likely take. Key issues include discriminatory pricing and unauthorized training data use.

China’s industry ministry wants AI integration inside the software engineers use to design products and chips. The Ministry of Industry and Information Technology issued Implementation Plan for the “AI + Software” Special Action on September 11, 2026, reported the Chinese technology portal IT Zhijia.

The plan sets clear targets for 2028. Intelligent programming tools and development platforms are to reach 20,000 software companies above a certain revenue threshold. And 100 benchmark applications of agentic software are to be created in key industries.

The plan calls for deep integration of AI with computer-aided design, computer-aided engineering and electronic design automation.

Two approaches are visible. “AI-assisted” adds intelligent functions to existing architectures. “AI-native” lets AI models take on core software capabilities, noted the industrial portal Zhineng Zhizao Wang.

Western firms dominate CAD, CAE, and EDA. Chinese media view the plan as a strategy to bypass them.

Chinese regulators addressed the loss-of-control scenario two years ago. It could not be ruled out that future AI might autonomously obtain external resources, replicate itself, develop self-awareness and seek external power, warned an AI safety framework released in September 2024 under the guidance of the Cyberspace Administration of China.

A 2025 update added a governance principle of “trusted application, preventing loss of control”. In May 2026, rules for AI agents identified “operational loss of control” as a security risk.

Developers must enhance their ability to detect, intervene, block, and recover from improper behavior.

China has begun drafting a mandatory national standard for AI agent safety, likely to become the first globally.

President Donald Trump and Chinese leader Xi Jinping are expected to discuss AI governance on September 24, 2026.

China is writing the rules for agentic AI. A national standard would establish compliance requirements for any AI agent in China. Foreign firms using agents in their Chinese operations should prepare for scrutiny of discovery, intervention, blocking, and recovery.

_________

While the U.S. and Europe are debating frontier models and their risk to humanity, China’s AI sector keeps laser-focused on physical AI and its monetization. The focus of the industry is shifting away from competition in large models and computing power toward “large-scale deployment of AI agents and application monetization,” says a new report by the China Telecom Research Institute quoted by CCTV News.

The enthusiasm of Chinese companies and individuals for AI agents that empower them to “complete complex end-to-end tasks without needing specialized technical expertise” is currently ushering in a “new stage of development” for the AI industry, the research institute argues.

While Chinese companies are shifting away from globally developed closed-source models, “agents are becoming the dominant form” of AI use in the country, according to the report quoted by CCTV.

As a result of this shift toward agentic AI, the demand for tokens is exploding. “Over the next two to three years, computing demand in China is expected to increase by nearly tenfold on average each year,” CCTV reports.

The institute forecasts annual token consumption to reach 100 quadrillion in 2026 and exceed 35 quintillion by 2030. Slowing down clearly is for others. China is hitting the accelerator button when it comes to AI.

With the excitement over generative AI slowly waning, agentic AI is clearly the next big thing in China. Industrial agents may unlock a new competitive advantage for China, which has a large and very diverse manufacturing base.

The next corporate success stories may very well not come from powerful AI models, but from the machines and systems that put those models to work.

What these agents already do in Chinese factories and ports, which Chinese newcomer is carving out a closed loop of services from perception to action rivaling Siemens and Huawei, and where the hype is still ahead of reality, are the subjects of this week’s Deep Dive.