Goldman Sachs adds robots to Xiaomi's watchlist: How a self-owned factory boosts AI content

On July 14, Xiaomi released a new performance report. A standout highlight was that four months after robots entered the automotive factory, the success rate of installing self-tapping nuts increased from 90.2% to 98%. Two new tasks were added: installing center console side covers and folding material bins, both with a success rate of 90%.

Subsequently, on the 21st, a publicly released Goldman Sachs research report discussed the factory's progress alongside Xiaomi-Robotics-U0 and Xiaomi-Robotics-1, maintaining a target price of 40 HKD. The market is now paying attention: Will Xiaomi's robots remain internal automation tools, or will they become a business with external orders and independent revenue?

From "capable of movement" to "capable of working," factories are rewriting the industry threshold for embodied intelligence

In the robotics industry, running, somersaults, and complex motion demonstrations easily attract attention. However, manufacturing enterprises care about a different set of metrics: continuous operation time, cost per task, fault recovery speed, cycle time matching, and yield rates.

In June 2026, the Ministry of Industry and Information Technology (MIIT) and the State-owned Assets Supervision and Administration Commission of the State Council launched a special action for practical training in humanoid robots and embodied intelligence. They proposed refining over 100 high-value scenarios within the year to form implementation capabilities at a scale of tens of thousands of units. Policy documents listed production manufacturing, inspection, maintenance, and warehousing logistics as key areas. The direction of industrial development has shifted from prototype demonstrations to routine operations.

China possesses dense soil for verification. Data from the International Federation of Robotics shows that in 2024, China added 295,000 industrial robots, accounting for 54% of global deployments; the electronics and electrical industry installed 83,000 units, and the automotive industry installed 57,200 units, ranking first and second domestically. Xiaomi's mobile phone and automotive businesses are situated in these two highly automated industries. Its internal factories can directly provide real workstations, engineering teams, and stable sources of tasks.

Traditional industrial robots can already work efficiently in standardized processes such as welding, spraying, and handling. The incremental space for humanoid robots mainly comes from flexible tasks: workstations require frequent adjustments, material forms are not uniform, or the existing environment is designed according to human body structures, making the cost of rebuilding dedicated automation equipment high. Xiaomi chose final assembly, part loading, and bin processing, avoiding the fixed processes where mature robotic arms are strongest, and placing robots in positions closer to actual needs.

Therefore, Xiaomi's factory currently plays a dual role. On one hand, robots improve the level of internal automation; on the other hand, the factory converts technical issues into recordable data: which materials are most prone to errors, which types of movements require manual intervention, and how precision changes after continuous work. Demonstration videos rarely form continuous training materials; real production lines generate failure samples every day. For embodied intelligence companies, these failure samples are often more valuable than a single successful demonstration.

The value of the four types of scenarios is not the same; Xiaomi needs to convert terminal scale into the learning efficiency of robots

Goldman Sachs focuses on the integration of Xiaomi's "hardware, data, and models," a judgment based on a rare combination of businesses. In the first quarter of 2026, Xiaomi shipped 33.8 million smartphones, its AIoT platform connected 1.1187 billion devices, and it delivered 80,856 vehicles. Mobile phones, home appliances, and cars provide user entry points and product distribution, while the automotive factory provides motion data, failure samples, and engineering validation. The coexistence of these four assets within one company reduces the costs for the robot team to find initial scenarios and coordinate with external customers.

However, terminal volume cannot be directly equated with embodied data. There is a significant data gap between mobile phone clicks, home appliance controls, and vehicle driving data, and robotic arm grasping and bimanual coordination. The roles of the four types of scenarios also differ: mobile phones and home appliances are closer to interaction entry points, cars can contribute spatial perception and intelligent driving R&D experience, but the factory is the direct training ground for robot motion models.

The two models recently released by Xiaomi target data generation and motion learning respectively. Xiaomi-Robotics-U0 has 38 billion parameters and generates multi-view, cross-entity embodied scenes through world models. Papers show that after adding synthetic data, the success rate of the π0.5 model in out-of-distribution real operation tasks increased from 36.9% to 63.2%. This route attempts to reduce the problems of expensive real-machine data and slow collection speeds.

Xiaomi-Robotics-1 uses over 100,000 hours of real operation trajectories for pre-training, covering over 1,700 household, commercial, industrial, and outdoor scenarios; post-training also used over 7,200 hours of real robot household data. Xiaomi disclosed that for new tasks such as mobile phone packaging, printer refilling, laundry, and boxing, when using less than 10 hours of demonstration data per task on average, the overall success rate reached 75%, compared to 40% for the π0.5 model. Lower data adaptation costs help reduce the project-based burden of redeveloping for each additional workstation.

The factory continuously generates real machine data, world models supplement scarce and long-tail samples, and motion models return to the production line for verification; failure cases enter the next round of training, forming a cycle of model iteration and workstation expansion. Xiaomi's internal demand for automotive, mobile phone, and home appliance manufacturing allows this cycle to continue operating, whereas pure model companies usually need to look outward for real robots and production scenarios.

The difficulties also center on the word "migration." The standardized workstations of an automotive factory are far apart from open home environments; adapting the same model to different mechanical structures, different supplier equipment, and different safety standards also increases engineering costs. If entering every factory requires collecting data again, adjusting algorithms, and modifying production lines, the robot business could fall into low-margin custom delivery. The applications with higher certainty for Xiaomi at this stage remain its own factories and industry chain partners; home robots belong to a more distant choice.

Only when robots walk out of Xiaomi's factory can a new business potentially form

Xiaomi's financial statements show expansion capacity, but also constraints. In the first quarter of 2026, the company's revenue was 99.1 billion yuan, and adjusted net profit was 6.1 billion yuan, down 10.9% and 43.1% year-on-year respectively; R&D expenses were 8.95 billion yuan, up 33.4% year-on-year, and capital expenditures were 3.27 billion yuan, up 20% year-on-year. Revenue from smart electric vehicles, AI, and other new businesses was 19.9 billion yuan, with an operating loss of 3.1 billion yuan. Mobile phones and AIoT still bear the main cash flow, and investment in new businesses has already formed visible pressure on profits.

In the short term, robots are more likely to manifest as internal cost reduction: reducing labor demand for highly repetitive and intensive positions, improving continuous operation capabilities, and lowering quality fluctuations. From Xiaomi's latest financial report and this round of public information, the company has not yet disclosed the deployment cost per workstation, payback period, downtime rate, or the number of deployed robots, nor has it announced independent orders for external customers. Lacking these indicators, robots still belong to the R&D and manufacturing system and are temporarily difficult to calculate revenue and profit separately.

External commercialization must still cross two engineering hurdles. Factory solutions need to achieve standardized replication; the deployment process cannot rely long-term on Xiaomi engineers being on-site for debugging. The revenue structure also needs to extend from one-time 本体 sales to model licensing, skill software, maintenance, and performance-based payment. Hardware expands revenue scale, software and services improve gross margin structure, and external repurchases determine business independence. The home scenario has greater space, but environmental complexity, safety responsibilities, and after-sales costs are also higher, giving it a lower priority in the short term than industrial scenarios.

Observing Xiaomi's robots does not require rushing to predict the launch time of general-purpose home robots. More effective observation indicators include the number of workstations, continuous operation duration, manual takeover rate, single-task adaptation cost, and external orders. If these indicators continue to improve, Xiaomi will gradually move from a robot user to a robot supplier; if progress remains long-term stuck at model releases and local workstation tests, the contribution of robots to the company's fundamentals will mainly reflect manufacturing efficiency.

Goldman Sachs including robots in its scope of observation for the company carries significance here. Xiaomi already possesses models, terminals, and factories, yet lacks a robot business that can be independently accounted for. Own factories can shorten the verification cycle, but may also keep technology long-term as internal tools. The key change in the coming years is whether Xiaomi can organize its self-use capabilities into standard products and then hand those standard products over to external customers.

Staying in the factory, robots improve costs; walking out of the factory, robots have the opportunity to change Xiaomi's revenue structure.