_________

The robotics business of electric carmaker XPeng has raised more than US$900 million in its first financing round. It was the largest single private financing China’s embodied AI sector has seen so far.

The list of investors is as striking as the sum involved. Alibaba and Tencent backed the same company in the same round, which is a rarity in the history of China’s internet industry. Those two internet giants are fierce competitors almost everywhere else, but they see a common opportunity here.

Ample financing like this has implications for the future trajectory of the robotics industry. XPeng plans to start mass production of its humanoid robot IRON by the end of this year. Deliveries are to follow in China and overseas in 2027. This means that XPeng is racing Tesla to be one of the first to deploy humanoid robots at scale.

Skeptics still question whether humanoid robots will ever play a big role in factories or households. Capital of this size, partly from strategic investors who control real deployment scenarios in retail and logistics, makes humanoid commercialization considerably more likely. It is an enabler for the whole nascent industry.

The big financing round confirms an industry trend with relevance for international carmakers. The sensors, chips and AI models developed for autonomous driving are the same ones that power robots. The automobile industry has a huge opportunity here.

Why the two industries are converging, and what XPeng plans to do with the money, is the subject of this week’s Must Read:

_________

Hangzhou-based aircraft maker CoolFly has unveiled Urban, a single-seat electric aircraft it calls the world’s first “standing-posture manned flying vehicle.” The pilot stands on a flat, flower-shaped platform roughly 3.5 meters across. Four ducted fans are built into petal-shaped sections around that standing area. An AI-powered gravity-sensing system follows every weight shift of pilot and machine and redistributes thrust across the rotors in real time. As a result, flying the vehicle feels similar to riding a self-balancing scooter. Urban and Dream have already found their first insurer. CoolFly says it is moving toward mass production this year.

It is progress for the industry that such a vehicle can now be insured, sold and legally flown in China.

The new joint venture Creativity (Changzhou) Robotics is training robots of the type AgiBot G2 in workshops with CNC machines. They are humanoids with two arms, but move on wheels. Machine tools are a demanding workplace for robots. To load and unload a CNC machine, a robot must place each workpiece precisely. It also has to work hand in hand with clamping fixtures and keep pace with the machining cycle, all in real time.

Machine-tool tending has resisted full automation because workpieces, fixtures and cycles vary from job to job.

For several days in late August, an anonymous model named Ox Alpha topped the charts of the AI marketplaces OpenRouter and OpenCode. People were speculating about who the developer was. On August 26, Beijing-based Zhipu announced that the mystery model was its newly released GLM-5.3-Flash. One detail made international headlines. All of that traffic ran on a cluster of more than 100,000 AI accelerators produced domestically in China, Zhipu disclosed. The company prices the model in the same low-cost bracket as DeepSeek, and its weights are open.

If the claims hold, a hit AI model has for the first time been served at scale entirely without American hardware.

iFlytek Healthcare, often described in China as the country’s first listed medical LLM company, reported first-half revenue growth of 49.4 percent year on year. The gross margin rose above 50 percent. The company’s shares jumped more than 22 percent at one point after these figures were released.

Its AI-assisted diagnosis system now covers more than 77,000 primary-care institutions across all 31 provincial-level regions. China’s National Healthcare Security Administration added 12 AI-assisted diagnostic services to the nationally unified medical insurance catalog earlier this year.

China has designated its national medical insurance system as a payer, and iFlytek Healthcare’s results are the first hard evidence that this arrangement produces real revenue.

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

_________

_________

According to the 2026 industry development report released in Guiyang on August 29, 2026, China's data sector grew 15.7% year-over-year to reach 6.78 trillion yuan (roughly US$1 trillion) in 2025.

Registered data firms climbed 17.9% to 482,000, with capital heavily targeting large language models, simulation, synthetic data, and embodied AI.

Source: 21st Century Business Herald (21 Shiji Jingji Baodao)

A company prospectus shows that industrial uses generated just 9% of Unitree Robotics’ humanoid robot revenue in early 2025, while research and education drove 73.6%.

Globally, Counterpoint data shows that over 60% of the 22,000+ humanoid robots shipped in H1 2026 went to entertainment and data research, compared to 13% for smart manufacturing and 5% for logistics.

Sources: The Economic Observer (Jingji Guancha Bao), Counterpoint

By the end of July 2026, China's total capacity for intelligent computing reached 2.45 million PFLOPS (FP16), according to a government official.

Source: National Data Administration, IT Home (IT Zhijia)

_________

Who is responsible for a car that runs a red light with nobody at the wheel? China’s top legislature has begun to address this question. It plans to make carmakers or importers liable when autonomous driving functions are switched on. The Standing Committee of the National People’s Congress started a draft revision of the Road Traffic Safety Law. For the first time, the law will contain a dedicated chapter on autonomous vehicles and advanced driving systems. Responsibility switches the exact moment a driver switches the system on. The new liability comes with an upside for everybody who has to shoulder it. A manufacturer that must answer for its software also gains the right to sell fully autonomous driving as a product on public roads nationwide. The draft names importers alongside carmakers. This means foreign manufacturers that ship cars with automated driving functions into China face liability, even when they do not manufacture there.

The world’s largest car and EV market takes on a question that politicians must ultimately settle everywhere. When the system drives autonomously, someone must bear responsibility. Chinese lawmakers place this burden on the OEM.

Beijing E-Town, the capital's high-tech industrial development zone, has published what it calls China's first dedicated AI4Chip policy. AI4Chip means applying artificial intelligence along the whole semiconductor value chain, from chip design and manufacturing to packaging, testing, equipment and materials. More than 400 integrated circuit companies operate in Beijing E-Town, with an output of more than 130 billion yuan, which equals roughly 19 billion US dollars. By 2028, the zone wants to cultivate three to five internationally influential AI4Chip ecosystem leaders and more than 10 benchmark applications that the rest of the industry can copy.

For European makers of chip equipment, the crucial factor is data. Whoever collects and models that data will end up holding important know-how.

Guangzhou is about to turn its AI policy into a law. The Standing Committee of the city's People's Congress published a revised draft of the Guangzhou Regulations on Promoting Artificial Intelligence Development for public comment. An expert committee with Chinese and international members and professionals from leading companies will meet once a year to advise the city. Computing power and data are the two production factors the municipal government identifies as most crucial. The city plans to build a unified platform that coordinates and schedules computing power across Guangzhou. Every year, the city will publish a catalog of public data opened for access or authorized use.

A regulation passed by the local legislature outlasts any mayor and gives companies a solid foundation to hold the city to.

China's National Energy Administration set up an “AI + Energy” working group within the cooperation network of the Belt and Road Energy Partnership. The group supports the digital transformation of member countries' energy sectors through joint research, exchanges on industrial applications and demonstration projects. Topics include renewable power forecasting and optimized grid operation. A critical item is standards. The group conducts joint research on digital-energy standards, technical certification, green-computing assessment, and the governance of energy-data cooperation.

BBelt and Road countries present a vital growth market for grid technology, storage and data centers. A working group that writes the specifications for green computing and energy data across those countries influences what their utilities will tender for.

_________

It was a strikingly sober remark about humanoid robots from a freshly minted billionaire. His company Unitree Robotics had just witnessed a blockbuster trading debut, with its shares soaring 460 percent on their first day on the Shanghai stock exchange.

One day later, Wang Xingxing stepped onto the main stage of the World Robot Conference in Beijing and seemed to question what some people call “hype”. Humanoid robots, said the founder of Unitree Robotics, are still less efficient than human workers and must be retrained for every new task.

The breakthrough in his industry, which he calls its “ChatGPT moment”, may arrive in two to three years in the best case, Wang said. It could also take up to ten. Unitree’s stock, which had just completed one of the most spectacular market debuts China has seen, fell sharply the same day.

Coming from the man whose dancing and kickboxing specimens made humanoids famous around the world, the no-nonsense assessment deserves attention. It also matches what engineers report from factory floors. Robots can now handle short, repetitive tasks almost as well as people. They tend to stumble when the jobs get longer or the environment suddenly changes.

We are currently being offered two stories about humanoid robots. One says they will replace factory workers within a few years. The other says they are just toys that dance at trade fairs. Both versions are wrong, and expensive decisions about automation, supply chains and competition from China depend on a more precise picture.

Wang has just given us an honest yardstick. Where humanoids already match human workers, why their success rates collapse in unfamiliar situations, and what has to happen before they reach their “ChatGPT moment”, is the subject of this week’s Deep Dive: