Ask AI · How does UBTECH leverage embodied intelligence technology to achieve a qualitative change in its revenue structure?

From “technological showmanship” to “value realization,” UBTECH is the first to break into real industrial applications with the “full-stack” capabilities of its full-size embodied intelligent humanoid robots.

On March 30, “the first stock of humanoid robots,” UBTECH released its 2025 performance announcement.

According to the data, in 2025 UBTECH achieved revenue of 2.001 billion yuan, a year-on-year increase of 53.3%, demonstrating strong growth characteristics; meanwhile, gross profit recorded 754 million yuan, up 101.56% year-on-year, corresponding to a gross margin of 37.7%, indicating a clear improvement in profitability quality.

However, expanding in scale is not the whole story. The special point of this financial report is that: the company’s full-size embodied intelligent humanoid robot business has seen an all-around explosion.

Over the past year, driven by the wave of physical AI, the market has been looking for validation signals for the “commercial implementation” of humanoid robots, and this time UBTECH’s financial report provides the answer. In 2025, UBTECH’s revenue from full-size embodied intelligent humanoid robots reached 821 million yuan, surging 2203.7% year-on-year.

The breakout of the core business released two positive signals: first, the humanoid robot industry has reached a key turning point for commercialization and “value realization”; second, over the past 14 years, UBTECH’s deep accumulation in the embodied intelligence field has successfully converted into delivery capabilities at large scale, bringing the company into a new phase of high growth.

Boosted by strong performance, in the two trading days after the financial report was released, the company’s share price kept rising, with a cumulative gain of nearly 20%. Citibank also raised UBTECH’s target price from HKD 155 to HKD 190 and reaffirmed a “Buy” rating.

So what is the underlying logic behind UBTECH’s business explosion? Does it have investment value?

Qualitative Change in Revenue Structure

The biggest highlight of this financial report is that UBTECH’s revenue structure has undergone a qualitative change: growth in its full-size embodied intelligent humanoid robot business is significant, and it has become the company’s largest revenue source.

According to the financial report data, in 2025 UBTECH’s shipment volume of full-size embodied intelligent humanoid robots reached 1,079 units, up 35,866.7% year-on-year, driving revenue to rise from 0.36 billion yuan to 821 million yuan in 2025, with a year-on-year growth rate of 2203.7% and revenue share of 41.1%.

Meanwhile, because the value per unit of full-size embodied intelligent humanoid robots is higher—gross margin reached 54.6%—as the share of revenue increases, the company’s overall gross margin also rose from 28.7% to 37.7%, and its profitability capability has moved into an upward channel in parallel.

The scaling up of the full-size embodied intelligent humanoid robot business benefited from a comprehensive acceleration of the company’s scaled scenario applications over the past year. According to the disclosure in the financial report, currently more than 80% of UBTECH’s product application scenarios are anchored in the industrial sector. The company mainly focuses on demanding tracks such as automotive manufacturing, smart logistics, 3C electronics manufacturing, semiconductor manufacturing, aerospace manufacturing, and industrial data collection, which require strict autonomy, stability, precision, and safety. Its customer roster includes multiple Fortune 500 companies, such as Texas Instruments, Airbus, BYD, Foxconn, Audi FAW, and others.

These industry giants are willing to pay because UBTECH’s full-size embodied intelligent humanoid robots have “brains” that can autonomously recognize, make decisions, and execute tasks. They are capable of addressing core pain points in workstations such as high personnel turnover, difficult management, and strong work repetition, including handling, sorting, and quality inspection.

In handling scenarios, Walker S2 features autonomous pallet-stacking decision-making, whole-body coordinated control, and obstacle-avoidance in narrow areas, enabling it to complete dense multi-layer, multi-row bin handling work. In sorting scenarios, Walker S2 can perform stacked grasping and flexible material operations in constrained environments, and can handle complex processes such as smooth-bin rack sorting and stripping bolt protective sleeves. In quality inspection scenarios, Walker S2 has both visual and operational capabilities, allowing it to perform material completeness checks, plug-in quality inspections, and air-conditioner leak detection. It also connects to the production system in real time, achieving closed-loop management of quality data.

Goodbye “remote-control toys,” reshaping industry standards

The surge of UBTECH’s full-size embodied intelligent humanoid robot business is not simply a category expansion, but rather a standard reshaping of humanoid robots—from “remote-control toys” to “production tools.”

In its financial report, UBTECH proposed a standard for humanoid robots: robots that are not remote-controlled, not toys, and have a height of 160cm or above are defined as full-size embodied intelligent humanoid robots; robots that are remote-controlled, pre-programmed, and have a height below 160cm are defined as non-full-size, non-embodied intelligent humanoid robots. Most humanoid robots in the market belong to the latter category, but the industry often confuses the concept and lumps remote-control toy types together as humanoid robots.

Although these products have a humanoid appearance, at their core they still rely on pre-programming or remote-control to execute fixed action sequences. Even though their deployment threshold is low, because they lack environmental understanding and autonomous decision-making, most of them are limited to simple scenarios such as performances and welcoming guests. Their unit value is low and the growth ceiling is clearly visible.

UBTECH’s core value comes precisely from its global leadership in the full-size embodied intelligent humanoid robot track. According to disclosures, the company is the only enterprise in the world that can achieve large-scale delivery of full-size embodied intelligent humanoid robots with real revenue and high gross margins.

Different from the aforementioned humanoid robots, full-size embodied intelligent humanoid robots are taller than 160cm. They not only have a human-like form factor, but also possess a “embodied brain.” Therefore, they do not require remote control, can perform autonomous work like humans, and can achieve deep understanding and autonomous decision-making in complex dynamic environments.

It is precisely because of this that the R&D and deployment threshold for full-size embodied intelligent humanoid robots is extremely high, yet they can be adapted to application scenarios with higher value—meaning the corresponding unit value per unit is also higher. According to the financial report data calculations, if based on sales volume of 1,079 units and revenue of 0.82 billion yuan, the average value per unit is as high as 760,000 yuan.

So why can UBTECH achieve value realization first? This cannot be separated from the company’s continuous, high-intensity R&D, which has built a soft-and-hard coordinated barrier. In 2025, the company’s R&D investment exceeded 500 million yuan, of which investment in the full-size embodied intelligent field exceeded 270 million yuan—nearly 16 times that of non-embodied products in the same period.

These huge R&D efforts have been transformed into tangible technical advantages and engineering capabilities. On the software side, based on hundreds of millions of high-quality industrial scenario data, UBTECH has independently developed multiple large models, including the open-source embodied intelligence large model Thinker, which won nine global “firsts”—this is the foundation for building physical AI. It also has a closed-loop feedback visual-language action model Thinker-VLA, as well as a world model Thinker-WM oriented for industrial embodied intelligent humanoid robots.

The upgraded group-brain network 2.0, powered by the company’s self-developed collaborative agents Co-Agent and Thinker-VLM, gives robots the ability to respond quickly, perceive spatial information accurately, and achieve environmental cognition and forward-looking decision-making—an “industrial lobster,” enabling a sense of coordination and unspoken teamwork among robots.

The company’s self-developed humanoid dual-eye vision technology delivers accuracy beyond RGBD cameras, achieving 99.5% precision in industrial scenarios, giving robots “true eyes.”

On the hardware side, UBTECH introduced the 4th and 5th generation dexterous hands, which have 19 fully actuated, active rotational degrees of freedom. Together with tactile and vision multimodal sensors, this greatly improves the precision of industrial operations. At the same time, it innovatively defined an on-body hot-swappable autonomous battery replacement system: when the robot runs out of power, it can complete a battery swap in 3 minutes on its own, laying the foundation for 7×24 hours of continuous operations.

Overall, the large-scale revenue generation of full-size embodied intelligent humanoid robots marks a critical leap for UBTECH—from “technological showmanship” to “value realization.” Relying on soft-and-hard coordinated capabilities, the company not only builds a deep technological moat, but also, in the process of moving toward AGI, is among the first to lock in the premium benefits of a trillion-scale market.

Defining the endpoint of physical AI with a “super brain”

The essence of this premium is that the competitive focus of humanoid robots is shifting from “limb development” to “brain evolution.”

Over the past year, the pace of humanoid robot evolution has been evident to all. From the boxing demonstration of Unitree robots to the smooth soccer-kicking moves of Expedition A1, embodied intelligence is carrying market expectations for physical AI in an unprecedented way.

However, a careful look shows that most humanoid robots currently on the market still linger in lightweight scenarios like entertainment performances. “Looks good but isn’t practical” is the core bottleneck preventing them from scaling up into commercial use.

The key constraint behind this is that many robot companies get trapped in a “small-brain shortcut” in real life: they overly rely on high-frequency sensor feedback and complex dynamics algorithms to pursue motion control, while ignoring understanding the commands themselves. A robot that can dance smoothly may be visually impressive, but this “advanced” performance is actually the result of highly preset behavior—so the robot cannot truly understand the task requirements behind it.

For robots to work deeply and at scale in real scenarios, they must have a crucial capability: understanding human language and doing human tasks. This requires endowing them with a powerful “embodied brain,” enabling them to accurately understand language, the environment, and task objectives—shifting from “doing by seeing” to “executing after understanding.”

Unlike most companies that focus on the “small brain,” UBTECH, while deeply engaged in motion control, has also proactively tilted core resources long term toward the “brain.” Over four years, the company’s total R&D investment exceeded 1.9 billion yuan, and it has already covered the planned funding that peers will have for embodied large models in the coming years. This also helps UBTECH secure a key position in the “intelligence competition” of future humanoid robots.

As the core engine of the “super brain,” UBTECH builds an embodied intelligence large model Thinker using independently developed algorithms and a rich supply of scenario data. The model has physical-layer task execution capabilities such as task-instruction understanding, environment perception, decision-making, and task planning, providing humanoid robots with strong “mental power” from “getting moving” to “figuring it out.” In authoritative global rankings of embodied intelligence brain models with parameters below 10B (ten billion), it won 9 first places.

It is worth noting that Thinker’s core competitive advantage lies in the authenticity of its training data. The model is trained and fine-tuned based on hundreds of millions of high-quality industrial data accumulated from the Walker S series. Since its data distribution shares the same source as real-world scenarios, it can directly convert into efficiency and stability when robots operate in factories.

In addition, by introducing RAG (retrieval-augmented generation) technology, UBTECH enables rapid training for specialized work types. This significantly improves robots’ decision-making ability and generalization capability across multiple workstations, laying the foundation for large-scale applications of full-size embodied intelligent humanoid robots.

In summary, the breakout of UBTECH’s full-size embodied intelligent humanoid robot business is not only a breakthrough in the company’s performance, but also a milestone showing that the industry has officially said goodbye to the “show stage” and stepped into the first year of industrial commercial deployment. In the face of future industrial real-world operations, the battle of winners and losers has evolved from “body agility” into a deep game of “brain intelligence.”

With long-term accumulation in embodied intelligence models and being “fed” with hundreds of millions of real industrial data points, UBTECH has built a solid competitive moat. Its abundant cash reserve of 4.92 billion yuan also provides ample assurance for subsequent model iteration.

Looking ahead, in the golden track of physical AI, the company not only enjoys the scarcity-premium of being an industry leader, but also gains a first-mover advantage by achieving a qualitative change in its revenue structure—showing strong certainty in growth and profit elasticity. Benefiting from this, the company is expected to enter a stage where performance and valuation resonate, a “Davis double play” phase, and its long-term investment value will become even more apparent.