AI is beginning to directly impact Midea's profits

AI is starting to directly affect how much money Midea makes.

In 2025, Midea improved cost reduction through AI, raising it from 180 million yuan to 700 million yuan.

This kind of scale shift has, in the past, more often shown up in internet and technology companies. Traditional manufacturing cost reduction has long relied on compressing headcount, optimizing materials, and simplifying processes; but at Midea, AI has already deeply penetrated the entire chain—production scheduling, energy consumption, inventory, processes, and more—creating value across multiple links, and ultimately reflecting in the financial statements.

Behind this is the result of sustained investment.

A month ago, at the Foshan high-quality development conference, Midea Group Chairman Fang Hongbo disclosed the corresponding investment structure: an AI talent team of more than 400 people, and R&D spending in the frontier fields of more than 60 billion yuan over the next three years.

More tangible implementation is the more than 13,000 agents that are running stably every day across Midea’s business processes. They receive data in real time, make autonomous judgments, and iterate continuously, directly influencing production and business decision-making.

Decision-making shifts from people to systems

At Midea’s most core ends—user demand and production manufacturing—AI has already been involved, mapping respectively to home scenarios and manufacturing scenarios.

On the home side, the change shows up in the interaction method.

In the past, smart appliances followed a one-way model: “the person issues instructions, the device carries them out.” Now it is being broken by AI. In 2025, Midea launched “XiaoMei AI,” and in multiple product categories deployed AirAgents based on the “MeiYan” large model. Systems no longer rely entirely on explicit instructions; they can now start to make context-aware autonomous judgments.

For example, in the whole-home air scenario, when a user says it feels “a bit stuffy,” the system will integrate data such as temperature, humidity, and air quality, then autonomously decide to turn on fresh air, adjust the air conditioner mode, and control humidity—no longer requiring the user to operate item by item. According to the disclosure, in those scenarios, user operation frequency drops by about 74%, and delegated tasks are executed more than 5 million times within two months.

Similar capabilities are being rolled out across multiple categories. An AI refrigerator can identify ingredients, and also dynamically adjust temperature-control strategies based on each ingredient’s freezing point and the environment. A dishwasher automatically adjusts wash duration and water usage based on how dirty the dishes are. An air conditioner can achieve ±0.3°C-level comfort temperature control accuracy under different climate conditions.

Users only need to express their status; decision-making is completed by the system, execution is coordinated by devices, forming a loop of “sensing—decision-making—execution—iteration.”

Currently, Midea’s entire product range of 500 million home appliances has connectivity capability. Globally, more than 140 million smart appliances are already networked, more than 150 million smart users are connected, and an AI rollout covering more than 150 categories of home appliance products has been completed. More importantly, some decision-making can be completed directly at the device level—by deploying lightweight models on WiFi modules, without relying on network connectivity and without increasing hardware costs.

Midea’s “people–car–home” linkage is a typical example of implementation: when the user’s vehicle enters a 500-meter radius around home, the system automatically completes preparations such as pre-cooling the air conditioner, heating the water heater, and turning on lights. Midea has already completed system integration with automakers such as BYD, Huawei’s HiCar, and NIO.

On the manufacturing side, AI almost takes over the factory’s overall scheduling logic.

Taking Midea’s Hubei Jingzhou washing machine factory as an example: this factory is already unified-scheduled by its “factory brain” M.bot. With 14 agents covering 38 core production scenarios, it runs through key links such as production scheduling, quality inspection, logistics, operations and maintenance, and energy management.

Midea Group Chief Digital Officer Zhang Xiaoyi calls it a “brand-new species.” In his description, every production element in the factory—people, machines, materials, methods, environment—is no longer an isolated unit. Robotic arms, AMRs (autonomous mobile robots), injection molding machines, cameras, sensors, and more are all endowed with sensing, understanding, decision-making, and action capabilities.

The results are already reflected in efficiency. In the Jingzhou factory, production scheduling response time has been reduced from the hour level to the second level. Scheduling efficiency has increased by about 90%, and average efficiency gains across multiple manufacturing scenarios exceed 80%.

But more importantly, the change is the full-process information integration. Quality inspection exceptions can directly trigger process adjustments. Changes in device status simultaneously affect production scheduling and maintenance plans. What used to require humans to connect the workflow step by step has been compressed into a single system-level continuous reaction.

This is also the core meaning of the “factory brain.” By using a multi-agent architecture, it unifies decision-making across different links into the same set of systems, enabling the factory to shift from “segmented optimization” to “overall dispatching.”

Within this system, robots become the execution end. The humanoid robot “Meilo” can already handle tasks such as first-piece inspection and delivery for inspection,巡检 (patrol inspection), and equipment maintenance. The “Yutu” patrol inspection robot enables autonomous patrol inspection and diagnosis, increasing patrol inspection frequency by 100% compared with human work.

In the past, Apple used iOS to integrate consumer-end devices, while Siemens used industrial software to unify production scheduling and dispatching. And at Midea, these two logics are integrated into the same system: it both controls the C-end user entry and also has B-end manufacturing scheduling capabilities.

These moves ultimately affect profits. The AI cost-reduction scale amplified by more than 4 times within a year, and is planned to continue increasing in 2026. At the same time, AI also brings new pricing premium space for products: “One-click good air” has sold 983k sets cumulatively and contributed 1.53 billion yuan in profit; “AI-managed services” related products have achieved 1.29 billion yuan in profit.

The growth ceiling is raised

After AI affects Midea’s income statement, a more critical question emerges: does this capability only serve home appliances, or can it be reused across industries?

Midea’s recent layouts in healthcare, energy, logistics, and other fields have already provided the answer. Although industries differ, the core logic is consistent: data enters, models make judgments, devices execute, and the approach is scaled and replicated across different scenarios.

In 2025, Midea’s healthcare business unit—Wandong Medical—together with Wanli Cloud, Midea’s AI Research Institute, and Alibaba’s DAMO Academy, launched the “industry’s first DR agent.”

This agent relies on Wanli Cloud’s over 4 million cross-domain chest DR imaging cases to build a multimodal diagnostic large model. It introduces a “deep thinking” reasoning engine that can automatically locate key signs, associate 13 types of common diseases, and generate structured diagnostic conclusions with coherent medical logic.

In high-load scenarios such as health check centers and emergency departments, single-case chest X-ray analysis is compressed to the minute level. More than 70% of doctors’ repetitive reading work is freed up, while consistency and accuracy of reports improve. The imaging interpretation workflow is also being adjusted: the system first provides an initial conclusion, and doctors complete the final confirmation.

In fields with even higher technical barriers, Midea has achieved breakthroughs as well.

In 2022, Midea launched China’s first “liquid-helium-free superconducting MRI,” breaking China’s long-standing “helium-free shortage” dilemma in the MRI domain. The technical foundation comes from Midea’s existing accumulation in the home appliance industry—low-temperature control, mechanical structure, and imaging capabilities—which, in the new industry, are recombined and amplified.

In addition, Midea Healthcare launched the “Kunlun AI smart imaging platform,” which boosts scanning speed by up to 35 times and reduces dosage by 70%.

This also means that Midea’s recognition, judgment, and quality-control logic accumulated in its industrial system has been brought into higher-precision, higher-risk medical scenarios.

In the energy sector, this logic is amplified further. By integrating resources such as KELD Electronics and KUKA, Midea builds a virtual power plant system, using AI to make real-time predictions and dispatch decisions for electricity prices and load. It extends dynamic control capabilities at the device and building level to the complex system level of the power grid.

Similar reuse is also evident in logistics. Ande Zhilian uses AI to re-architect its inventory network, integrating thousands of dispersed nodes into an efficient distribution system. It optimizes inventory layout, order routes, and transportation scheduling, significantly improving fulfillment efficiency. These capabilities originally come from the manufacturing system; they are simply scaled up in logistics scenarios in terms of scale and complexity.

When these capabilities are further abstracted and standardized, Midea enters a new stage. Midea Yunzhishuo, built on Midea’s AIGC platform (AIGC engine), unifies capabilities from R&D, manufacturing, and supply chain to office work into an AI agent system, enabling large-scale deployment internally and externally.

The core significance of this step is: AI capabilities shift from being internal efficiency tools to being externally deliverable products and services.

More importantly, once this model works across industries, Midea’s growth boundaries are redefined. This is also the key differentiation point between traditional home appliance companies and Midea: the former still expands product categories around product expansion, while the latter uses a set of AI-driven decision systems to continuously extend its business boundaries.

More than a decade of sustained investment

This seemingly sudden surge in value is actually the result of Midea’s sustained planning over more than ten years.

In 2012, Midea internally launched the “632” digitalization project, unifying dispersed systems and data. At the time, this initiative served management efficiency more than anything else, but looking back today, it feels like pre-laying the groundwork: integrating data structures so that business lines began to have conditions for reuse, and models could be built on stable inputs.

This kind of foundation often exists implicitly; its value only becomes apparent when technology is layered further on top.

In 2017, Midea completed its acquisition of KUKA, directly moving into core industrial robot segments; then in 2022, the Blue Orange Laboratory was approved, and R&D continued to sink deeper into critical components. Equipment capability gradually shifted from “usage” to “definition,” and the control chain began to be pulled back into the system itself, improving controllability of key processes.

In the even deeper underlying domain of chips, Midea’s planning spans more than a decade. The project can be traced back as early as IPM R&D in 2010. Then in 2018, Midea Ren Semiconductor was established. Until mass production of MCU in 2021, the entire process spans more than ten years. By 2024, related products’ cumulative sales exceeded 983k units, the defect rate is controlled within 5ppm, and supply to external manufacturers has begun.

A long investment cycle confirms the difficulty of cultivating core control capabilities endogenously, and also builds barriers that are hard to replicate.

These capabilities ultimately landed on the production site. In 2025, Midea’s self-developed humanoid robots entered the Jingzhou washing machine factory, participating in quality inspection and patrol inspection, and directly connecting to the scheduling system to run. Currently, in robot R&D, Midea covers three directions: humanoid-like, full humanoid, and超人形 (super-humanoid). Deployment pathways primarily focus on industrial scenarios first, validating capabilities in controllable environments, then gradually expanding outward.

At the same time, Midea’s R&D organization itself is also changing. Over the past five years, Midea’s cumulative R&D investment exceeded 60 billion yuan, and it will maintain the same intensity over the next three years. The proportion of research-oriented PhDs exceeds 40%, and the AI team is rapidly expanding. Engineers from different backgrounds, such as materials and food engineering, have begun participating in product development. The boundary between research and engineering continues to shrink, and technology transformation efficiency has improved significantly.

Seeing these changes together makes it clearer: data is integrated first, hardware capabilities gradually become endogenous, the execution end is continuously strengthened, and the organizational structure is adjusted accordingly. Under all those layers of groundwork, Midea’s AI capabilities expanding across industries have solid support.

The market is still valuing with old logic

Contrary to outside impressions, Midea is no longer a pure home appliance company. Yet the market’s valuation logic still stays in the past.

As of March 30, 2026, Midea’s P/E ratio is 12.26x, still within the home appliance industry range. This corresponds to a familiar traditional home appliance pricing logic: it sells air conditioners and washing machines; growth depends on the real estate cycle; and profits depend on the cost of raw materials.

But the reality is that Midea’s revenue structure has changed.

Midea’s 2025 financial report shows that Midea achieved total operating revenue of 458.5 billion yuan, up 12.1%; and net profit attributable to shareholders of listed companies was 43.95 billion yuan, up 14% year over year.

Among them, its ToB business revenue exceeded 122.8 billion yuan, up 17.5%. Multiple lines—buildings, robots, industrial segments, healthcare, and energy—were expanded, and business boundaries continued to extend outward.

Taking the robot business as an example. In 2025, demand for industrial robots in automotive, semiconductors, AI servers, and other fields exceeded expectations, directly driving growth in businesses such as KUKA. According to MIR Rui industrial statistics, in 2025 KUKA industrial robots’ domestic shipments exceeded 32k units, up more than 30% year over year. Market share reached 9.6%, ranking stably in the top three of the industry.

If valued by segments: KUKA and Midea Yunzhishuo are benchmarked against the robotics and automation sector (P/E 25–30x). Weilink and KELD Electronics are benchmarked against the new energy sector (P/E above 20x). With the current P/E around 12x, it is clearly undervalued.

The root cause of this mismatch lies in the way value is realized that has changed.

Hardware is still being sold, but the value-realization cycle has been greatly extended. Once equipment enters homes and factories, AI systems run continuously, optimize dispatching, and deliver iterative decisions, creating value throughout usage. Revenue and profit extend from one-time product sales to long-term operations.

This brings two core changes:

First, marginal costs decrease. New users, factories, and scenarios do not bring significant cost pressure, but they accumulate more data and improve system accuracy, forming a positive feedback loop;

Second, the growth model shifts from linear to compounding. Growth no longer relies entirely on large capital inputs. Instead, it expands outward based on the existing AI system, exhibiting the compounding characteristics of a technology platform.

This is exactly the boundary between home appliance companies and technology companies. The former relies on product sales to complete value realization, and growth is constrained by cycles; the latter extends value into the usage stage by building ecosystems and systems, earning returns over a longer cycle.

Referencing Nvidia’s valuation re-assessment path: although the GPU hardware business has not disappeared, when computing power becomes infrastructure, its valuation is determined by its ecosystem position, expansion space, and whether these capabilities can continue to translate into revenue growth.

By comparison, Midea’s pace is slower, but its structure is already formed. Overseas markets are also re-evaluating Midea. In the past, when China-based manufacturers expanded overseas, they were often viewed as extending into contract manufacturing, leading to relatively low valuations. But Midea, by acquiring KUKA, building its own R&D system, and promoting high-end brands like COLMO overseas, has upgraded its globalization to “technology output + brand output.”

Overall, when AI enters the home appliance company’s profit and loss statement, it looks like an unexpected technical surprise, but in reality it is the result of more than ten years of deep accumulation and refinement. From digitalization projects to self-developed chips, from acquiring KUKA to humanoid robots entering factories, Midea continues to invest in underlying areas where short-term returns are hard to see. Now, these seemingly scattered capabilities are being strung together into a line in the AI era, forming a decision system that can self-iterate and can be replicated across industries.

When this system becomes the new competitive advantage, Midea’s corporate nature has changed. It is no longer just a manufacturing company that sells home appliances, but a technology company whose value is being re-recognized by the market—and whose valuation urgently needs to be reassessed.

大量资讯、精准解读，尽在新浪财经APP