StartseiteArtikel

In weniger als 3 Jahren erreichte der Unternehmenswert über 15 Milliarden Yuan. Die physische Ambition des Geniestudenten von Huawei

中国企业家杂志2025-08-07 09:48
ZhiYuan Robotics hat sich entschieden, die gesamte Technologie-Stacks von Roboterkörpern, Kleinhirnen und Gehirnen eigenständig zu entwickeln.

In 16 trading days, with 11 consecutive daily limit up of 20%, the stock price soared by 10.83 times, and the market value skyrocketed by more than 34 billion yuan. This is the record set by Shangwei New Materials Co., Ltd. (688585.SH), a listed company on the Science and Technology Innovation Board, in July. Therefore, Shangwei New Materials has become the first and so far the only "ten - bagger" this year.

Behind this capital carnival is a less - than - 3 - year - old embodied intelligence company, Zhiyuan Robotics.

On July 8th earlier, Shangwei New Materials issued an announcement stating that Shanghai Zhiyuan New Creation Technology Co., Ltd. [the operating entity of Zhiyuan Robotics, now renamed Zhiyuan Innovation (Shanghai) Technology Co., Ltd.] plans to acquire no more than 63.62% of Shangwei New Materials' shares at a price of 2.1 billion yuan through a share - holding platform jointly funded by the company and its core team.

After the acquisition is completed, the controlling shareholder of Shangwei New Materials will be changed to the entity jointly held by Zhiyuan Robotics and its management team, and the actual controller will be changed to Deng Taihua. The core team includes Peng Zhihui and others.

Zhiyuan Robotics first caught the attention of the capital and the industry because of Peng Zhihui, a former "Genius Youth" at Huawei. Born in 1993, Peng Zhihui, with the online name "Zhi Huijun", is one of the top 100 UP owners on Bilibili, with 2.82 million followers. He used to work on Ascend AI chips and AI algorithms at Huawei.

During his time at Huawei, Peng Zhihui was praised by Ren Zhengfei in a speech, who said that he was "the driving force for Huawei's innovation". However, at the end of 2022, Peng Zhihui announced his departure from Huawei and his participation in founding Zhiyuan Robotics, where he serves as the CTO.

Subsequently, the founding team of Zhiyuan Robotics continued to expand. In March 2025, the legal representative of Zhiyuan Innovation was changed to Deng Taihua, who was a former vice - president at Huawei, leading Huawei's 5G campaign and building the Ascend AI ecosystem. Deng is not only Peng Zhihui's former supervisor at Huawei but also his alumnus from the University of Electronic Science and Technology of China. In April 2025, Luo Jianlan became the chief scientist. Luo was a research scientist at Google DeepMind and a core member of the robotics expert Sergey Levine's team at the University of California, Berkeley. He led the development of the world's first super - human robot reinforcement learning system. In 2024, Yao Maoqing joined as a partner and the president of the embodied business department. Yao used to be in charge of autonomous driving algorithms at Waymo and NIO.

Photo: Zhao Dongshan

The financing speed and amount of Zhiyuan Robotics have set a record in the industry. In less than a year since its establishment, its valuation quickly reached 1 billion US dollars, making it the world's fastest - growing embodied intelligence company to become a unicorn. The investors include well - known institutions such as Hillhouse Capital, Matrix Partners, CDH Investments, and BYD.

So far, Zhiyuan Robotics has completed 11 rounds of financing. In March 2025, Zhiyuan Robotics received a new round of financing led by Tencent. In May 2025, JD.com and the Shanghai Embodied Intelligence Fund participated in its new round of financing, and many old shareholders such as SAIC increased their investments. After this round of financing, the valuation of Zhiyuan by a third - party research institution has reached as high as 15 billion yuan. On August 1, 2025, Zhiyuan Robotics announced another strategic investment from an international group, led jointly by LG Electronics and Mirae Asset.

Recently, Zhiyuan Robotics was once again pushed into the spotlight after jointly winning the first 124 million - yuan order in the humanoid robot field with Unitree.

According to the procurement and bidding website of China Mobile, Zhiyuan Robotics won a 78 - million - yuan bid for a humanoid robot procurement project of a wholly - owned subsidiary of China Mobile. For Zhiyuan Robotics, this large order of 78 million yuan is not only a breakthrough in short - term revenue but also a key catalyst for promoting the comprehensive upgrade of its technology, supply chain, and market strategy.

"Specifically, it is a bipedal humanoid robot customized for China Mobile, which will mainly be used in the stores and business halls of China Mobile operators to perform some interactive tasks such as reception and explanation. In the future, we will also see similar service - reception scenarios in operators, hotels, banks, etc., where there is actually a huge demand." Regarding the details of the order, Yao Maoqing told "China Entrepreneur".

Currently, Zhiyuan Robotics has entered the "stage of centralized commercial delivery". "The number of robots mass - produced offline by Zhiyuan has exceeded 2,000, and the expected shipment volume this year will reach several thousand. It has achieved breakthrough implementation in four major scenarios: industrial manufacturing, warehousing and logistics, power inspection, and interactive guidance." Yao Maoqing said. In addition, he revealed that Zhiyuan has launched an overseas expansion plan and has achieved a layout in markets such as North America, Europe, the Middle East, Japan, South Korea, and Southeast Asia.

Full - stack Layout

While some entrepreneurs in the robot field are seeking breakthroughs in single points, Zhiyuan has chosen a different path. Like Apple and Tesla, it conducts a thorough full - stack optimization and closed - loop layout of software and hardware.

In terms of product layout, Zhiyuan Robotics has three major robot families: Expedition, Elf, and Lingxi, and its products cover a variety of commercial scenarios. Among them, the Expedition series is positioned for industrial - grade applications; the Elf series is for commercial and light - industrial scenarios; the Lingxi series focuses on scientific research and frontier exploration.

In terms of technology layout, Zhiyuan Robotics chooses to conduct full - stack self - research on the robot body, cerebellum, and brain to build the core capabilities of the robot in three aspects: interaction, operation, and motion intelligence. This is not an easy path. On the one hand, the capital investment is huge; on the other hand, bottlenecks in any link may become a constraint for the whole company.

Photo: Zhao Dongshan

Regarding why they choose a full - stack layout, Yao Maoqing told "China Entrepreneur": "The implementation of intelligent robots is a tightly - closed flywheel, and it is difficult to outsource one part completely. In the process of practice, it is a cycle. From body design, to data collection, and then to model iteration, after the deployment is completed, you will find many new areas that need improvement. Therefore, at present, the team is making saturated investments in all directions."

In terms of capital operation, Zhiyuan Robotics has also taken a completely different path. On the one hand, Zhiyuan Robotics has received 11 rounds of external investment in two years; on the other hand, Zhiyuan Robotics has also actively laid out the upstream supply chain through investment.

According to the statistics of "China Entrepreneur", Zhiyuan Robotics has currently invested in more than 10 embodied intelligent robot - related enterprises, such as Digital Huaxia, Lingchu Intelligence, Qianjue Robotics, and Fuxing Electromechanical. Among them, there are upstream companies such as robot components, embodied intelligent systems, and bionic robots, as well as joint - venture companies jointly established by Zhiyuan and listed companies such as Boson Precision, Dafeng Industry, Wolong Electric Drive,isoftstone, and Fuling Precision.

"Most of our investments are in the upstream supply chain, such as sensors, joint modules, etc. And some of the investors and shareholders we have introduced are actually scenario - based parties, including the automotive and 3C electronics industries." Yao Maoqing told "China Entrepreneur".

Regarding the progress of cooperation with upstream and downstream partners, Yao Maoqing said: "We have introduced the components of some upstream supply chain enterprises into our complete machines, and the downstream scenarios have also been opened for us. The two sides are jointly building some POC (Proof of Concept) projects, and there is a chance for some deliveries in the second half of the year."

Build Its Own Data Warehouse

Currently, the bottleneck in the development of embodied intelligent robot technology lies in the extreme shortage of high - quality scenario data. This is also the key reason why the performance of humanoid robots varies greatly in different scenarios.

For this reason, Zhiyuan Robotics has chosen to "build its own data warehouse".

In September 2024, Zhiyuan Robotics built a data collection factory of more than 3,000 square meters, containing more than 3,000 real objects. At present, through this factory, Zhiyuan has formed the world's largest data set, AgiBot World, and has made it open - source, covering hundreds of scenarios such as households, laboratories, and retail stores.

 "Facing the desert of embodied intelligent data, we choose to plant the first tree, hoping it will grow into a forest." When talking about the original intention of creating the data set and making it open - source, Yao Maoqing said.

In March 2025, Zhiyuan Robotics released its first general embodied base model, "Zhiyuan Qiyuan Large Model GO1". This model can learn by combining Internet videos and real human demonstrations, enhancing the model's understanding of human behavior. In Zhiyuan's design concept, this model is a general robot strategy model that can be migrated between different robot forms and can be seamlessly adapted to heterogeneous robot bodies such as bipeds, wheeled robots, and robotic arms, greatly reducing the cost of intelligent migration.

Embodied intelligence means enabling AI models to have a "body" to perceive, make decisions, and act in the real physical world. Therefore, while humanoid robots are showing off their skills, many people are also concerned about when the "ChatGPT moment" of the embodied intelligence industry will come.

In Luo Jianlan's view, "Reaching the 'ChatGPT moment' doesn't make much sense. The goal is to reach the 'DeepSeek - R1 moment' after training and optimization. Enabling robots to have both generalization ability without sacrificing performance and reducing the hallucination rate, with the success rate of each task approaching 100%, is what can really make robots useful in the real physical world."

"If you only know the principle of tying shoelaces but have only a 10% success rate in actual operation, you won't be able to leave home in three hours." Luo Jianlan explained to "China Entrepreneur".

For this reason, at the 2025 World Artificial Intelligence Conference, Zhiyuan Robotics released the world's first open - source platform for the world model of a real - world dual - arm robot, Genie Envisioner. Based on the collected AgiBot - World data set, this platform uses more than 1 million and nearly 3,000 hours of synchronized video streams of the head and dual - arm wrists to capture the spatial layout, action evolution, and semantic intention in robot operation tasks, which can help robots complete cross - modal migration from visual perception to robot action execution.

"If the GO1 model solves the problems of 'what to do' and 'how to do it', the world model solves the problems of 'how the environment changes' and 'what the consequences of actions are'. The former helps robots achieve multi - task decision - making and action execution, while the latter helps robots build the dynamic deduction ability of the physical environment." An embodied intelligence practitioner told "China Entrepreneur".

Put into Practical Use

In the embodied intelligent robot industry, the pros and cons of a robot are mainly evaluated by two indicators: generalization ability and performance. Generalization ability refers to an individual's ability to apply skills in different situations, while performance refers to the success rate and speed of a robot in completing tasks.

"Different from large models, robots have very high requirements for performance. If the task execution success rate is only 50% or 60%, such robots cannot be applied in the real world. For example, if a robot spills water every other time when pouring water for you, surely no one will accept it." Luo Jianlan told "China Entrepreneur".

In 2025, many practitioners in the embodied intelligence industry call it the "year of mass production" and the "year of commercial implementation" of humanoid robots. Luo Jianlan also believes that "the robot industry has passed the stage of showing demos and is evolving from showing off skills to an industrial closed - loop model."

Currently, the entire application scenario and market direction of Zhiyuan Robotics are mainly focused on the B - end, and it has achieved breakthrough implementation in four major scenarios: industrial manufacturing, warehousing and logistics, power inspection, and interactive guidance.

Photo: Zhao Dongshan

During the 2025 World Artificial Intelligence Conference, Zhiyuan Robotics also jointly conducted a global live - broadcast with Dematic. Zhiyuan's robots acted as couriers and sorted express parcels on - site. According to Yao Maoqing, after less than a month of cooperation between Zhiyuan Robotics and Dematic, the sorting speed of the Elf G1 has been increased to one parcel every six seconds, which can already meet the needs of some clothing e - commerce enterprises.

However, although humanoid robots have had some practical applications in real - world work, there are still certain scenario limitations.

In the past year, Yao Maoqing has clearly felt the change in customers' attitudes. "Previously, they thought they could use the robots directly by just plugging them in. Today, they realize that it actually requires a two - way effort. This is a data - driven process, which requires robots to be trained in real scenarios and continuously iterated. Secondly, some production line transformations are also needed to better connect the robots. At the same time, from the perspective of customers' funds, they are willing to invest a lot of money in scenario verification with us."

Wang Chuang, the president of the general business department of Zhiyuan Robotics, is responsible for the commercial implementation of Zhiyuan Robotics. He said, "In terms of scenario implementation, Zhiyuan Robotics follows the principle of 'from easy to difficult'. First, solve the scenarios that traditional intelligence cannot handle but embodied intelligence can easily achieve, such as material box transfer, and then tackle more complex scenarios such as material sorting that require generalization ability."

Yao Maoqing is optimistic about robots truly being put into practical use. "In the first half of this year, Zhiyuan has signed some contracts for flexible scenarios for implementation. In about a year, we will see many such cases. Factories are just the first step. The next step is the retail service industry, and ultimately, in a few years, robots will enter households."

However, Yao Maoqing also admitted that "the current supply chain, especially the upstream joints, motors, reducers, etc., is still in the process of overall improvement to achieve large - scale production while ensuring product consistency. At present, the suppliers we are in contact with are mostly small and medium - sized enterprises, and we are still accompanying them in their growth. And the real big players in the electromechanical field are still eager to try."

This article is from the WeChat public account