PetroChina holds a 60% controlling stake and has co-founded a company with Unitree and Fanjie. The acceptance right for oil and gas robots cannot be handed over to external manufacturers.
CNPC holds a three-party joint venture, anchoring the standard definition of oil and gas embodied robots.
Energy Insider noticed that recently, Beijing Municipal Administration for Market Regulation publicized a new establishment case of a joint venture: China National Petroleum Corporation Kunlun Manufacturing Co., Ltd. and its concerted actor Kunlun Capital have signed an agreement with Unitree Robotics and Beijing Fanjie Intelligent Technology Co., Ltd., planning to jointly set up a joint venture engaged in the business of industrial embodied intelligent robot solutions. The equity structure is 60%, 20% and 20%, with the Kunlun faction holding absolute controlling stake.
Three months ago, Kunlun Capital just participated in the strategic placement of Unitree on the Sci-Tech Innovation Board with 136 million yuan. The pace from financial shareholding to joint establishment of a controlled operating entity is not slow.
But what is more noteworthy than "CNPC making robots" is the figure of 60%. A state-owned energy enterprise uses absolute controlling stake to put hardware companies and AI model companies into the same legal entity, which points not to the initiative of technology R&D, but to another thing, that is, what kind of machines can enter CNPC's stations and who defines it.
Who is Fanjie Intelligence, and what can the world model do at oil and gas sites
Unitree does not need too much introduction. The B2-Ex explosion-proof quadruped robot launched by Unitree has been applied to scenarios such as natural gas gas gathering stations and oil and gas production device areas, undertaking route patrol and anomaly inspection, and the capabilities at the hardware level are available in existing products. What is really unfamiliar is Fanjie Intelligence.
This company was established only in April 2026, with a registered capital of 1.1111 million yuan. It was incubated by the team of Academician Qiao Hong from the Institute of Automation of the Chinese Academy of Sciences, and Qiao Hong directly holds 9% of the shares. Its core direction is "the integration of world model and embodied foundation model", and CEO Han Jinchi comes from Qiao Hong's team. To put it more plainly, what Fanjie wants to do is to install a "brain" that can understand physical causality for robots, not purely relying on data training, but using mechanism models to constrain algorithm output, trying to make robots make more reliable judgments in scenarios with sparse samples and dense long-tail risks.
This exactly aligns with CNPC's long-standing pain points in oil and gas inspection. Real leaks, fires and emergency states cannot be exhausted by data, and pure learning-based models are naturally limited in generalization ability when facing rare working conditions.
Fanjie's path is logically feasible, but its actual verification volume at oil and gas sites is almost zero. By including Fanjie in the equity structure, the joint venture essentially allows "algorithm interpretability" and "site access" to be run in at the same operating entity, shortening the cycle from prototype to acceptance.
CNPC does not lack robots internally, what it lacks is "replicable standard parts"
A fact that is easy to ignore is that the robot layout within the CNPC system is no longer a blank sheet of paper.
In Xinjiang, the explosion-proof inspection robot "Xiaozhi" developed by Kunlun Digital Intelligence has obtained the Ex d IIC T6 certification, running continuously for 216 hours during the Spring Festival holiday in the environment of minus 40℃ in Xinjiang, replacing more than 350 person-times of manual inspection, with an identification accuracy rate of over 99.5%.
The explosion-proof quadruped robot of the Pipeline Bureau Communication Company also completed the first quadruped inspection implementation in the oil and gas sales field at Xigu Oil Depot in May 2026, solving a series of on-site engineering problems such as signal occlusion, pipe ditch crossing and intrinsic safety circuit transformation. These are real operation records, not demonstrations.
The problem is not "whether there are robots", but "fragmentation". Xiaozhi is wheeled, with limited traffic capacity in complex terrain with dense pipe ditches and undulating Gobi; the quadruped pilot of the Pipeline Bureau is project-based, with hardware, algorithms and perception equipment coming from different suppliers, leading to high cost of system interconnection and unified data scheduling.
As the technical support unit for CNPC's digital and intelligent transformation, Kunlun Digital Intelligence excels in software, large models and system integration, but does not take equipment manufacturing as its main business. Kunlun Manufacturing, established in Xiong'an in October 2023, is CNPC's energy equipment manufacturing platform, focusing on the intelligent upgrading of equipment products.
The division of labor between the two determines that CNPC needs a carrier between "digital intelligent R&D" and "equipment manufacturing": integrating the hardware body, AI model and oil and gas acceptance standards into the same product definition, to form a set of standard solutions that can be mass-produced and included in the procurement catalog, instead of assembling a separate set for each station.
The joint venture controlled by Kunlun Manufacturing with 60% stake exactly fills this position. It is not building a new stove outside Kunlun Digital Intelligence and the Pipeline Bureau, but trying to gather the scattered robot capabilities from various lines into an operating entity that has both equipment manufacturing genes and scenario discourse power.
The real test lies not in technology, but in the internal procurement logic
The competitors faced by the joint venture are not outside in the first place.
There are already multiple robot technology routes running in the CNPC system, including the wheeled inspection of Kunlun Digital Intelligence, the quadruped pilot of the Pipeline Bureau, and the stock equipment of external suppliers such as Qiteng that has accumulated more than 8 million hours of operation in the industrial explosion-proof inspection scenario, all competing for the same procurement budget.
To get orders, the new platform must present a more robust account in terms of total cost of ownership, inspection coverage and fault warning capability than existing solutions. Procurement in the oil and gas industry is driven by operation and maintenance procedures, after a robot passes the pilot acceptance, there are several years of stability assessment before it can enter the framework procurement. A technology can be regarded as truly entering the regular procurement list only after it is written into the formal operation and maintenance procedures.
This means that 60% controlling stake solves the right of product definition, but cannot solve the accumulation speed of operation data. The hardware reliability of Unitree and the model interpretability of Fanjie need to run enough on-site hours in the same legal entity, so that the "acceptance standard" in the hands of Kunlun Manufacturing can be turned from paper into a reusable procurement basis.
ANYmal X went through four years of engineering R&D and more than ten patents before launching a deliverable product that meets both ATEX and IECEx requirements, with the first batch delivered to Equinor; BP, Shell, Petronas, Petrobras and others have participated in early tests at hundreds of onshore and offshore facilities since 2023.
According to the public product parameters, the nominal operating temperature range of Unitree B2-Ex is from minus 20℃ to 55℃, while the temperature in Xinjiang Oilfield often drops below minus 40℃ in winter. Xiaozhi can run there relying on the high-energy low-temperature battery specially developed for extreme cold and the full-link low-temperature reinforcement design. The winter working condition may be the first physical constraint that the joint venture needs to supplement in its product definition.
Explosion-proof certification is the entry ticket, and operation records are the real bargaining chip. Whether the joint venture can gain a firm foothold in the CNPC system mainly depends on whether the operation report after the first batch of Unitree bodies equipped with Fanjie's model runs through a complete winter in a certain oilfield is written into the technical specification for the next procurement. After all, standards cannot generate data, but data can define standards.