Huawei, Tencent, Baidu Join the Fray: The Full-Scale Battle for Robot Brains Begins
Original by Xinmu · Author | Li Xiaodong
Just two days ago, the gong sounded at the Hong Kong Stock Exchange for the "first stock of the robot brain," as Xian Gong Intelligent (XG Intelligent) went public, surging over 38% intraday on its first day. Notably, during the subscription phase, the public offering saw nearly 6,000 times oversubscription, with only a 5% chance of securing one lot. Eight cornerstone investors, including Hillhouse and Yuanbao Family Office, collectively poured HK$462 million.
A company established only six years ago, with an annual revenue of RMB 442 million and still in a loss-making state, managed to attract such fervent capital market attention. Its core selling point is not the robot body itself but the controller known as the "robot brain." Data from Zhuoshi Consulting shows that by sales volume, XG Intelligent's robot controller has captured 24.8% of the global market and 45.2% of China's market, ranking first in both.
XG Intelligent's listing coincides precisely with an inflection point in the industry.
Over the past two years, most attention in the embodied intelligence track has focused on humanoid robot bodies—whose robots walk more steadily, move more flexibly, and look closer to humans. But entering 2026, the wind has shifted.
Capital is now bypassing the "body" and rushing directly toward the "brain." According to QbitAI, in the first half of 2026, total funding in China's embodied intelligence sector reached approximately RMB 43.8 billion, with over half flowing to "brain"-focused companies, while body manufacturers received less than 20%.
This marks a shift in industrial value center as embodied intelligence moves from concept validation to large-scale implementation. When hardware bodies mature and supply chains stabilize, the core variable determining what robots can do and how well they do it is shifting from mechanical structure to intelligent systems.
01 Bypassing the Body, Heading Straight for the Brain
Back in 2024, the hottest companies in the embodied intelligence track were still humanoid robot integrators.
Ubtech listed on the Hong Kong Stock Exchange, while Zhiyuan, Yushu, Fourier, and others sequentially released new products, each launch drawing mass attention from the tech circle. The narrative then was straightforward: embodied intelligence is the next computing platform, humanoid robots are the ultimate form, and whoever mass-produces the body first secures the lead.
But in just over a year, capital's focus has shifted.
The funding structure in the first half of 2026 is telling. Brain-focused companies received half of the funds, while full-stack, core components, and body companies remained off-center. Viewed differently, "brain"-related companies collectively captured nearly 70% of funding, with pure hardware body companies pushed to the edge.
The pace of funding is also accelerating. Many brain companies complete a round on average once a month, with the fastest two companies having only two weeks between rounds. Such speed is rare in hard tech, typically seen only in internet and pure software tracks.
The logic behind this is not complicated. After two years of explosive growth, the technical barrier for robot bodies is rapidly lowering. Yushu Technology announced that its bipedal humanoid robot Unitree R1 dropped from RMB 39,900 to RMB 29,900, with spot sales open.
Previously, humanoid robots were generally considered products priced at hundreds of thousands of yuan. This price drop reflects the maturity of the supply chain and the start of large-scale production—core hardware costs for joint modules, motors, and reducers are all falling quickly, showcasing the advantages of Chinese manufacturing.
When "bodies" are no longer scarce, the value of "brains" becomes prominent.
The principle is simple: the same robot hardware, equipped with different brains, can accomplish vastly different tasks. Low-end controllers only allow robots to repeat tasks along preset paths, while high-end intelligent controllers enable robots to perceive environmental changes, autonomously plan paths, coordinate multi-machine operations, and even understand natural language commands through large models. The former sells hardware, the latter sells capability.
Gross margin differences best illustrate value stratification. XG Intelligent's prospectus shows its controller business has a gross margin of 79.8%, software business 89.3%, while robot integrators only achieve 38.4%, and parts business just 15.7%. Nearly 80% gross margin is extremely rare in the hardware-dominated robot industry, closer to the profit levels of pure software companies.
This is the core reason capital pursues brain companies—compared to heavy asset, low-margin, long-cycle hardware bodies, controllers and intelligent systems have lower marginal costs and stronger scaling effects. Once ecosystem barriers are established, profit margins become very attractive.
Of course, the term "brain" in the current context has a much richer connotation than a few years ago.
Early robot controllers were essentially motion control boards, responsible for directing motor rotation and coordinating joint actions, with technical barriers mainly in real-time performance and stability. But today's "robot brain" has evolved into an intelligent system integrating perception, decision-making, and control—incorporating SLAM navigation, visual semantic recognition, reinforcement learning, multi-machine scheduling, and even starting to access large language models and world models.
According to XG Intelligent's prospectus, the global robot controller market for independent controller suppliers grew from 6,000 units in 2021 to nearly 50,000 units in 2025, and is expected to exceed 300,000 by 2030. In terms of revenue, the global market size grew from RMB 700 million in 2021 to RMB 2.4 billion in 2025, and is expected to reach RMB 8.4 billion by 2030, with a CAGR of 28.8% from 2026 to 2030.
Some industry reports also indicate that in 2025, the robot brain controller market reached RMB 2.236 billion, while the robot motion control system market hit RMB 6.073 billion. Including AI-enhanced smart controllers and supporting software, algorithms, and cloud services, the market could be much larger.
A report by UK-based Future Market Insights shows the global physical AI market is expected to grow from approximately USD 383 billion in 2026 to USD 3.26 trillion by 2040.
Physical AI essentially means equipping various physical entities with intelligent brains. Robots are just one of the most typical forms.
02 Cloud Providers Enter the Fray, Vying to Become Robot Brain Suppliers
If the startup funding frenzy is just an internal signal, the collective entry of internet giants and cloud providers elevates the robot brain war to the ecosystem level.
This year, Huawei, Tencent, Baidu, and Alibaba have successively released their own embodied intelligence platform products, and without exception, they have chosen the route of "no body, only brain"—not building robots themselves but providing intelligent systems, development tools, and cloud service foundations for robot manufacturers.
Huawei moved fastest and most thoroughly. At this month's Huawei Cloud INSPIRE Developer Conference, Huawei Cloud officially launched CloudRobo, an embodied intelligence development platform, positioned as the "world's first full-process embodied intelligence one-stop development platform." According to official statements, this platform covers the entire chain of data synthesis, model development, simulation verification, and cloud-edge-device deployment, with built-in million-level data assets and over 20 Ascend-friendly models, enabling "robot cloud onboarding in hours, model deployment in minutes."
Huawei's thinking is clear: use cloud capabilities to lower the threshold for robot intelligence. Previously, robot manufacturers had to develop algorithms, train models, and set up simulation environments themselves, with high costs and long cycles. Now they can directly use the Huawei Cloud platform to complete everything from data to training to deployment, allowing them to focus on hardware bodies and scenario implementation.
On the release day, more than 20 companies including Youiai Zhihe and Huayan Robot announced they would be the first to join the CloudRobo platform. Huawei also launched an ecosystem cooperation plan called "Hundred Models, Thousand Scenarios, Cloud Gathering for Win-Win." Combined with Huawei's layout in Ascend chips, industrial software, and 5G networks, CloudRobo essentially builds an "end-edge-cloud" collaborative embodied intelligence infrastructure.
Tencent chose a lighter approach. Tencent Robotics X Lab unveiled its complete technical matrix for the first time, releasing the Tairos embodied intelligence open platform, open-sourcing the Hunyuan embodied large model HY-Embodied series, and showcasing robot interconnection technology RoboFusion.
Zhu Yajuan, head of the Tairos product ecosystem at Tencent Robotics X, stated on site that Tencent positions itself as an "indispensable titanium screw" in the robot industry, not building robot bodies but focusing on software and cloud services. This statement is interesting—it acknowledges its boundary of not doing hardware while emphasizing its core value at the system level.
Specifically, the Hunyuan embodied large model solves the "understanding and thinking" problem, allowing robots to perceive environments, understand commands, and autonomously plan tasks; the Tairos platform addresses development efficiency by providing standardized toolchains; and RoboFusion solves interconnectivity between different robots. Together, these form Tencent's version of the robot brain solution.
Baidu follows a "data + model + infrastructure" route. At the Create 2026 Baidu AI Developer Conference in May, Baidu Intelligent Cloud clearly stated it would increase investment in three areas: AI Infra infrastructure, scenario connectivity, and industry standard building. In April, Baidu had already teamed up with multiple robot companies to launch an "embodied intelligence data supermarket," building a hierarchical data labeling system.
On the investment front, Baidu is also deeply involved. Zhoupingfang's RMB 1 billion Series B and Beijing Humanoid Robot Innovation Center's RMB 700 million Series A both have Baidu's backing. Baidu's logic is to use the capabilities of the Wenxin large model to bind hardware companies, completing the AI closed loop from virtual to physical worlds.
In February this year, Alibaba DAMO Academy released the RynnBrain embodied foundation model, open-sourcing seven models at once, including the industry's first 30B MoE architecture embodied model. Official materials state that this model equips robots with spatiotemporal memory and spatial reasoning abilities for the first time, achieving SOTA on 16 embodied evaluation benchmarks, surpassing Google's Gemini Robotics ER 1.5.
ByteDance has not individually released an embodied intelligence platform, but it has listed world models as its top priority for 2026. At the June 23 Volcano Engine FORCE conference, ByteDance mentioned that its Seedance video generation model can be applied to data synthesis for embodied intelligence. Given ByteDance's technical accumulation in multimodal large models and video generation, as well as its Volcano Engine cloud service capabilities, its eventual entry into the embodied intelligence brain track is almost inevitable.
The collective choice of big companies to "build brains, not bodies" is essentially a financial calculation.
While hardware bodies are intuitive, they involve heavy assets, low margins, complex supply chain management, and likely price wars—as repeatedly proven in the smartphone industry. The brain and platform layer, though requiring high upfront R&D investment, rapidly reduces marginal costs once an ecosystem forms and can generate recurring revenue through cloud services.
More importantly, cloud providers have natural advantages in making robot brains. Embodied intelligence training requires massive computing power, large-scale simulation environments, and multimodal large model foundations—all of which are cloud providers' core competencies. For robot body manufacturers to build an AI infrastructure from scratch is neither economical nor realistic.
From this perspective, the future embodied intelligence industry may form a layered structure similar to smartphones: the bottom layer is chips and computing power, the middle layer is operating systems and intelligent brains, and the upper layer consists of various robot hardware forms and scenario applications. Cloud providers are competing for the core system position in the middle layer.
03 High-Margin Brains: Why Can't They Support Revenue?
While the brain story sounds compelling, returning to business reality reveals challenges.
XG Intelligent's prospectus is worth careful reading. On one hand, the controller's 79.8% gross margin and software's 89.3% gross margin fully demonstrate the commercial value of robot brains. On the other hand, the controller business generated only RMB 85 million in 2025, accounting for 19.3% of total revenue, while the low-margin robot integrator business generated RMB 300 million, accounting for 67.9%.
That is, the most profitable business has the smallest scale, while the revenue driver has low margins. This is a common dilemma for all third-party controller manufacturers.
The reason is straightforward. At this stage, robot manufacturers—especially industrial robot firms—prefer to purchase complete integrated solutions rather than buy controllers separately. First, integrated delivery allows quick deployment and direct use, requiring no additional integration from customers. Second, as a core component, if controllers are purchased separately, customers need strong secondary development capabilities, which is too high a threshold for many traditional manufacturing clients.
Thus, controller manufacturers often must ship controllers through integrated solutions—first installing controllers into their own complete robots, then selling to end customers. The brain is good, but it must be attached to a body to be sold.
Xu Zhaoyun, a partner at Lihai Investment, commented to the media that domestic humanoid robot integrators are generally in the ramp-up phase of large-scale production, with the core demand being control of hardware BOM costs. Third-party controller manufacturers can only bind customers through low-price volume sales, causing software value to be discounted by the hardware carrier. This leads to an awkward situation: controllers have high technical barriers and high gross margins but can't be sold alone; integrators have relatively low technical barriers and low gross margins but are the main revenue source.
XG Intelligent has not escaped this pattern.
From 2023 to 2025, the company's revenue grew from RMB 249 million to RMB 442 million, a three-year CAGR of 33.2%. However, during the same period, net losses were RMB 47.7 million, RMB 42.3 million, and RMB 47.07 million, respectively, with a cumulative loss of approximately RMB 137 million over three years. Adjusted net losses narrowed—from RMB 20.91 million in 2023 to RMB 2.87 million in 2025—but true profitability remains elusive.
This is not unique to XG Intelligent but a stage-specific characteristic of the entire industry.
Currently, embodied intelligence is still in the early stage of implementation. Customers are more willing to pay for tangible hardware, while willingness to pay for software, algorithms, and intelligent systems has not yet been fully established. Similar to early smartphones, consumers initially paid for hardware; the value of software and services became apparent only after the mobile internet ecosystem matured.
Another challenge comes from in-house development by large companies. With cloud providers like Huawei, Tencent, and Baidu launching their own embodied intelligence platforms, will the survival space for third-party independent controller manufacturers be squeezed?
At present, the positioning still differs. Cloud provider platforms are more focused on general-purpose large models, training tools, and cloud scheduling, while companies like XG Intelligent are more focused on underlying real-time motion control and on-site execution—complementary rather than substitutive. However, in the long term, it's unclear whether the boundaries will blur.
In the international market, the traditional big four industrial robot companies—Fanuc, ABB, Yaskawa, and Kuka—all develop controllers and bodies integrally, never selling controllers separately, tightly holding their core technology. This is a key reason they maintain high profit margins. As domestic robot manufacturers move upmarket, they will likely also pursue in-house controller development.
Thus, the true opportunity for independent third-party controller manufacturers may not lie in selling to leading body manufacturers but in serving a vast number of small and medium-sized customers and long-tail scenarios.
XG Intelligent's "Nebula" open system embeds over 1,000 robot models, covering forklifts, bin-handling robots, wheeled humanoids, robot dogs, and more, allowing modular customization. This "Lego-style" solution is very attractive to SMEs that want to avoid starting from scratch but need quick deployment.
By the end of 2025, XG Intelligent has served over 2,000 customers, with products sold to 35 countries and regions, and end clients including Tesla, BYD, Foxconn, and Siemens. A large enough customer base and wide industry coverage form the foundation for an independent controller manufacturer to build ecosystem barriers.
XG Intelligent's listing is a landmark event. It signifies that the capital market has begun to recognize the independent value of the "robot brain" and indicates a deepening division of labor in the embodied intelligence industry.
In previous years, the industry debate focused on "can robots do it"—can they walk, run, grasp? Now, these basic questions are gradually being answered, and the industry is entering the second half: "how smart can robots be?" The winner in this second half will not be determined by whose mechanical structure is more delicate, but by whose brain is smarter, whose system is more open, and whose ecosystem is more prosperous.
Capital flowing to brains, big companies laying out platforms, and startups delving into vertical scenarios—these three threads are interweaving and advancing. In the short term, hardware bodies will still be the main revenue driver, with brain value more reflected in technical barriers and differentiated competitiveness. But in the long run, as robot hardware becomes commoditized, intelligent systems will be the true profit center and ecosystem entry point.
Of course, it's too early to talk about the endgame. Embodied intelligence is still very early-stage—Omdia data shows that global general-purpose embodied intelligence robot shipments were only 3,000 units in 2024, expected to reach 2.6 million by 2035. From three thousand to 2.6 million, there's still a long road ahead.
But the direction is clear. In the second half of embodied intelligence, the brain calls the shots.
