Ask AI · How does Midea’s DR intelligent agent reshape radiology workflows?

Less than a week after releasing its 2025 financial report, Midea Group’s core technology research institute, Midea Central Research Institute, led an online media exchange.

The exchange was themed around the industry’s first DR intelligent agent, mainly applied in the radiology department within the medical field. With several key terms placed side by side, it’s hard to tell how closely it is connected to the image many people have of Midea Group as a home appliance powerhouse.

“In today’s world, ToC business is the moat for survival and growth, while ToB business is the future safeguard that carries you through cycles.” In its 2025 annual report, Midea Group explained its current development logic. The 2024 annual report described it as: “Promoting and realizing balanced development of ToC and ToB.”

In 2025, Midea Group achieved revenue of 458.5 billion yuan, up 12.1%; net profit attributable to shareholders was 43.95 billion yuan, up 14%. While some companies in the home appliance industry saw stage-by-stage fluctuations in performance, double-digit growth like this is not common.

Among them, the smart home business achieved revenue of 299.93 billion yuan, up 11.3%. At the same time, ToB business revenue reached 122.75 billion yuan, up 17.5%, which is higher than the overall growth rate.

On the profit side, the smart home segment’s overall profit margin was 29.9%; the overall gross profit margin for ToB business was about 20.8%. Within that, the building technology business’ gross profit margin was 30.6%, already surpassing the C-end business.

Behind these two sets of figures is the fact that Midea is adjusting its business structure to reduce reliance on a single industry and respond to cyclical risks.

Midea Group also reiterated in its annual report that it plans to continue investing more than 60 billion yuan over the next three years, with a focus on cutting-edge areas such as AI and embodied intelligence, and to fully transition its business toward AI. As for the DR intelligent agent for the medical industry, it is just one of the areas it covers.

“Midea is trying to tear off the label of ‘home appliance company,’” Liu Buchun, a senior industry observer in the home appliance sector, told Phoenix WEEKLY Finance.

ToB business becomes a new growth curve

Although the home appliance business still provides the scale and profit foundation for Midea, the driver of incremental growth at present is truly the B-end business.

Breaking down Midea Group’s B-end business, in 2025, its three major core segments—industrial technology, building technology, and “robots and automation”—recorded revenues of 27.23 billion yuan, 35.79 billion yuan, and 31.01 billion yuan, respectively, representing year-on-year growth of 10.2%, 25.7%, and 8.1%.

The financial report shows that the three B-end items together generated 94.13 billion yuan. Adding other business revenue, ToB business revenue reached 122.75 billion yuan, accounting for more than a quarter of total revenue. At the same time, the building technology business’ gross profit margin was 30.6%, already higher than that of the smart home (C-end) business, forming a stable profit model—indicating that Midea’s ToB business has entered a “capability realization phase.”

As the business continues to extend upward, in areas such as industrial automation and building systems, Midea will inevitably face direct competition with international leading players such as Siemens, Schneider Electric, and Honeywell.

“In sub-segments where it directly competes head-to-head with international leading companies, it may be difficult for Midea to win in the short term purely by technology benchmarking,” Chen Jingjing, founder of Jingjie Brand Consulting and a brand strategy expert, told Phoenix WEEKLY Finance. “International top-tier companies have already taken the high ground in technology and standards, but Midea can make up the gap through engineering integration to achieve off-cycle competition.”

In Chen Jingjing’s view, Midea’s ToB business is moving from scale expansion toward deeper capability development. In the short term, growth still depends on industry dividends and market penetration; in the long term, it depends on whether it can make the leap from engineering capabilities to technology and system capabilities. What’s worth noting is that Midea still needs to improve its leadership in core technologies, and it must build sufficient technological barriers and industry depth; otherwise, it risks remaining a “highly efficient integrator” in ToB for the long run, without breaking into the high end of the industrial chain.

Robots unlock the upper limit of space

To outsiders, the key to Midea’s next capability leap may lie in its robotics business.

Back in 2017, through its acquisition of KUKA, Midea already entered the core field of industrial robots.

According to MIR Rui Industrial statistics, in 2025 KUKA industrial robot shipments in the domestic market exceeded 32,000 units, up over 30% year on year; the market sales share reached 9.6%, keeping it firmly in the top three in the industry. KUKA’s domestic market share for robots with a load capacity of more than 300 kg was as high as 47.4%.

While traditional industrial robots have maintained shipment volumes, the variable in Midea’s robotics field may come from humanoid robots.

The financial report mentions that by 2025, Midea had developed three generations of humanoid robot products, with a total of five humanoid robot models, covering both industrial and commercial scenarios.

On the industrial side, the humanoid robot “Meiro” has been deployed at the Jingzhou washing machine factory. Under the scheduling of the “factory brain,” it can perform tasks such as 3D quality inspection, equipment patrol, and sheet metal loading. On the commercial side, “Mela” can carry out operations such as retrieving items, heating food, and making coffee, and is expected to be deployed in stores in 2026.

Midea states that it has set three major R&D directions: humanoid, fully humanoid, and super humanoid robots. In the short term, it focuses on industrial scenarios—replacing repetitive labor and completing complex tasks such as precision assembly—to address bottlenecks in manufacturing efficiency. In the medium to long term, it looks ahead to entering broader scenarios such as home and healthcare, and ultimately becoming the next-generation intelligent terminal connecting “smart home” and “smart manufacturing.”

The core logic behind this plan is to upgrade robots from “automation tools” to “general-purpose execution capabilities.” Once this capability matures, its application scenarios will no longer be limited to manufacturing and may extend into broader service systems.

AI capabilities permeate all business

Beyond robots, Midea’s AI business is also becoming a foundational capability that runs through all of its operations.

Currently, Midea has assembled an AI R&D team of more than 400 people. It deploys more than 13,000 intelligent agents across multiple business scenarios every day, covering manufacturing, home, office, medical, and logistics.

In 2009, Midea Group, Wandong Medical, and Alibaba jointly invested in and established Wanli Cloud Medical Information Technology.

In February 2026, Wandong Medical, Midea AI Research Institute, Midea Central Research Institute, and Alibaba Damo Academy jointly announced the release of the WanDong DR intelligent agent.

This product is the industry’s first DR intelligent agent. Based on a large model trained with 4 million imaging cases, it enables full-process automation, reshapes radiology workflows, and drives intelligent radiology and medical homogenization.

Guo Hongyu from Midea Central Research Institute’s Intelligent Department told Phoenix WEEKLY Finance that the DR intelligent agent is built on a general large model and is developed in cooperation with Alibaba. It optimizes performance through model switching and fine-tuning. During training, the team introduced medical knowledge systems such as clinical diagnosis knowledge bases and expert consensus, integrating them into model training and reinforcement learning to enhance the model’s professionalism and reliability.

Huang Jiajang, general manager of Wanli Cloud, further introduced that the development cycle of the intelligent agent has exceeded 10 months, and its overall performance is currently good. Previously, the system had already been trial-tested inside Wanli Cloud’s imaging platform, covering about 200 institutions, with the trial period approaching three months. From the results of the current evaluation, the model’s overall performance has reached around 80 points.

This suggests that Midea’s AI not only serves its own business, but is gradually being converted into a technology product for external use.

In Liu Buchun’s view, to achieve a “technology” positioning, Midea needs to explore and establish a development path with its own characteristics and core competitiveness. As things stand, there is still significant room for improvement.