Published on Aug. 18, 2026

Ten kilometers off the coast near Shanghai, Chinese engineers are testing the world's first underwater data center powered by offshore wind. The rising demand for AI compute and data centers has increased power consumption in the grid of Shanghai’s Lingang Special Area by nearly 30 percent in the first five months of 2026, which in turn is driving this search for innovative ways to reduce data center cooling costs, the business magazine Caixin reported.

Beneath the waves, a submerged cylindrical cabin houses a high-density computing hub, next to more than 50 offshore wind turbines that supply its power. The facility, built by Shenzhen HiCloud Data Center Technology, is cooled naturally by the seawater. It consumes almost no freshwater and has a power usage effectiveness (PUE) of 1.15, according to the report.

The underwater data center's design also solves a problem of an earlier pilot. In 2020, a fully submerged model in Hainan province required costly engineering ships for routine maintenance, and operating expenses ballooned. The Shanghai facility keeps its power equipment on a surface platform and only the 1,950-ton server cabin under water, which drastically cuts maintenance costs.

The underwater data center is a testing ground for a strategy Beijing calls “computing-power synergy,” or suandian xietong. In 2026, the term entered China’s Government Work Report for the first time and was written into the outline of the 15th Five-Year Plan. It means closely linking powerful computing infrastructure with green electricity from wind and solar and coordinating them intelligently, the National Energy Administration explains. China’s industrial policy thereby ties the build-out of AI capacity to the build-out of renewable energy.

On August 12, SAIC-GM’s Cadillac announced its all-new XT5 PHEV as the first production model to carry Momenta’s mass-produced R7 world model, reported the Chinese auto portal Gasgoo on August 14, 2026.

Momenta is a Suzhou-based Chinese developer of driver-assistance systems that counts Mercedes-Benz among its investors.

In this new model, a novel electronic architecture links perception, decision-making and execution so that the car can carry out the world model’s driving intentions with greater precision, wrote Gasgoo.

According to Cadillac, conventional driver-assistance systems focus primarily on identifying the road and the obstacles immediately ahead. The R7 World Model “can not only ‘see the world’ but also ‘understand the world,’” says the company’s announcement, published by the news portal Sina on August 12, 2026 (in Chinese).

Cadillac lists some situations it has in mind, morning rush-hour commuting, parking, passing on narrow roads and reversing along a previously driven path. It promises new ways of handling surprises such as a delivery rider appearing out of nowhere or a car door opening into the vehicle’s path.

Huawei and Tesla currently dominate the market for luxury SUVs in China with powerful driver-assistance systems. Cadillac’s own announcement names them, Huawei’s Qiankun ADS and Tesla’s FSD, and says the upgraded Super Cruise will join them in the top tier of driver assistance in China, in what it calls a new “Three Kingdoms era,” a nod to the three-way contest for power in third-century China. The XT5 also becomes the first real-world test of whether world models drive better than the systems now on the road.

The Chinese AI chip designer Cambricon has adapted its processors to run inference for five major domestic large language models, chairman and general manager Chen Tianshi said at the company’s half-year earnings briefing on August 12, reported the tech portal IT Zhijia (IT Home) on August 13, 2026 (in Chinese). The five are GLM by Zhipu AI, DeepSeek by DeepSeek, Qwen by Alibaba, Kimi by Moonshot and MiniMax by MiniMax.

Chen also said that a sixth-generation processor microarchitecture and instruction set are under development, optimized primarily for the training and inference of large models. The aim is to improve programming flexibility, performance, energy efficiency and chip-area efficiency.

Cambricon’s revenue rose 108 percent from a year earlier to nearly 6 billion yuan (roughly US$890 million) in the first half of 2026, and net profit attributable to shareholders climbed 123 percent to 2.3 billion yuan (roughly US$340 million).

The adaptation work has implications for the AI race between the U.S. and China. Washington bans the export of advanced AI chips to China, and American AI companies such as Anthropic have started to bar Chinese developers from using their models. China’s response is “local substitution,” the effort to make itself less dependent on American AI hardware and software. When Chinese AI models are better integrated with domestic silicon, that substitution gets a direct push.

Shanghai-based AgiBot has surpassed its domestic rival Unitree Robotics to become the world’s largest vendor of humanoid robots in the first half of 2026, reported CCTV Finance, the business channel of state broadcaster CCTV, on August 11, 2026 (in Chinese), quoting a report by the research firm Smart Analytics Global.

AgiBot shipped roughly 8,400 humanoid robots between January and June and captured 44 percent of the global market, according to the new data. Its shipments rose 562 percent from a year earlier. Hangzhou-based Unitree, which had led the field, fell to second place with about 5,900 units and a share of 31 percent.

AgiBot’s shipments rose on the back of a versatile portfolio of full-size bipedal A-series models, compact X-series units and wheeled G-series robots, analysts believe. Unitree’s humanoid shipments remain concentrated in its flagship G1 model, with education, research and performance deployments making up a substantial share.

Both companies are currently preparing for an IPO. Unitree is close to a listing on Shanghai’s STAR Market, an offering “widely seen as a bellwether for mainland China’s emerging humanoid sector,” wrote the South China Morning Post. AgiBot is pursuing an IPO in Hong Kong. The frenzy around Unitree’s listing, and how founder Wang Xingxing built the company, is the subject of this week’s Must Read.

The gap between the two rivals is small, and both sales figures are still relatively low. Roughly 8,400 units against 5,900 in six months is a vanity number. While they say little about the long-term prospects of both companies, market data suggest huge opportunities for the two of them.

Global humanoid shipments jumped 272 percent from a year earlier to about 19,100 units in the first half, and Chinese manufacturers supplied more than 97 percent of them, according to the research quoted by CCTV.

Industrial and commercial applications took more than 70 percent of shipments, up from about 50 percent a year earlier, as factory automation and logistics operators become the main buyers.

The Second Affiliated Hospital of Zhejiang University School of Medicine in Hangzhou is using AI to control the temperature of medicines delivered by drones. This system, called “AI Temperature Control + Drone Medication Delivery,” is the first application of a joint laboratory the hospital founded with Huawei, and first trial results are in, reported the Chinese news portal Caiwen on August 14, 2026 (in Chinese).

This new cold chain for temperature-sensitive medicine is designed for deliveries between main hospitals and smaller clinics, or to supply schools that might urgently need a specialist medication for students or teachers.
"The drones are equipped with AI-controlled temperature regulation boxes that keep specialty medicines such as insulin at a constant 4–5°C,” writes the Zhejiang channel of the news portal Zhongguo Xinwen Wang on August 14, 2026 (in Chinese).

The trial service, a first in China, currently uses 14 public drone takeoff and landing sites in Hangzhou, plus a number of temperature-controlled “micro-warehouses.”

Within Hangzhou’s main urban area, no more than 30 minutes pass between a patient’s online prescription request and the drone’s landing, the reports said. Throughout, AI makes sure the temperature of the drug stays constant.

Micro-warehouses could fundamentally reshape how drugs are distributed in China’s tiered healthcare system. Main hospitals would no longer need to stock every category of medicine at every primary-care location, and smaller facilities would be spared the inventory costs and expiration losses of rarely used specialist and emergency drugs. A region-wide network with rapid response and end-to-end traceability would serve institutions at every level.