Is CATL eyeing the AI business?
In the late 19th century, the founders of Standard Oil, the predecessor of ExxonMobil, began buying shares of railway companies in their personal names. Jabez Bostwick, one of the founding partners, later simply served as the president of the New York and New England Railroad Company, while another partner Camden guarded the oil-producing regions for Standard Oil and financed the Ohio River Railway at the same time.
Over thirty years, these individual shareholdings quietly formed a network. In 1904, investigative journalist Ida Tarbell wrote the well-known assertion in The History of the Standard Oil Company: Standard Oil "holds shares in most of the large railway systems and has representatives on the boards of directors of almost all major railway systems".
At that time, Standard Oil under the management of John D. Rockefeller was the largest crude oil extraction, gasoline and lubricant production enterprise in the United States. For Standard Oil, taking shares in railways could not only ensure that it could obtain cheap and safe oil transportation routes, but also make railway companies themselves major customers for lubricant sales. It can be said that this method guaranteed Standard Oil's position in the industry to a certain extent.
The same story is still happening today.
On October 6, 2026, Bloomberg, citing people familiar with the matter, reported that DeepSeek's new round of financing is nearing completion, with a scale ranging from 800 billion yuan to 1 trillion yuan. Tencent and CATL have once again become one of the investors with the highest capital contribution. It has been only more than four months since CATL first appeared on DeepSeek's shareholder list.
As a leader in power batteries, CATL used to invest its capital in mining, materials and automobile companies: securing supply in the upstream and connecting customers in the downstream. Now, its investment tentacles are beginning to extend to large models, data centers, computing power operation, and embodied intelligence.
Why does a battery manufacturer invest in AI? The story of Standard Oil provides a perspective to understand CATL.
What CATL is looking for is not a new story unrelated to batteries, but the next batch of customers for batteries and electric energy, as well as new capabilities required to access these customers.
In those days, oil needed to be transported by trains; today, electricity needs to be sent to data centers and supplied to robots. What CATL is trying to do is to become a shareholder at the table before the next round of demand really expands.
Is the Battery Giant Chasing the AI Trend?
If you make a list of CATL's investment targets in the past few years, you will find a change: in the early days, the list was densely populated with mining, materials, battery equipment and automobile companies, while now large models, data centers and robots are occupying more and more positions.
This is not a sudden cross-border move. In the middle of this year, Zeng Yuqun, founder of CATL, was asked in an exclusive interview with Caijing why CATL's business boundaries are expanding wider and wider. His answer was: "All our expansions are centered on our core capabilities."
One of the previous priorities was to carry out industrial layout around electric vehicles. By 2026, the CATL ecosystem has invested in more than 10 automobile enterprises in total.
For example, in 2021, CATL contributed 770 million yuan to participate in the capital increase of AVATR and became its second largest shareholder. In the same year, it also participated in Zeekr's Pre-A round of financing. In addition to equity investment, CATL has also established joint battery companies with many automobile enterprises, advancing the cooperation from procurement relationships to joint R&D, production and supply.
AVATR is a typical example of this strategy. In the cooperation system composed of Changan, Huawei and CATL, CATL does not just wait for new cars to be launched to receive orders, but participates in product development earlier as a shareholder and technical partner. For battery suppliers, this means the opportunity to integrate their own technologies into customers' solutions at the vehicle model planning stage.
At the same time, the venture capital institutions under the CATL ecosystem, including Puquan Capital, Wending Investment, Chendao Capital and Mindong Times, have also expanded to both ends of the industrial chain: moving upstream to mining, materials and equipment to ensure supply and technical support; extending downstream to automobile enterprises, battery swapping and recycling to find markets and new business models for batteries.
Direct investment and fund investment together form a network, and the role of capital has extended from obtaining returns to establishing cooperative relationships and complementing industrial capabilities.
In the AI era, CATL has also begun to shift its focus.
An early signal came from Qianxun Intelligence. In November 2024, Qianxun Intelligence announced the completion of the angel+ round of financing exclusively invested by Borey Capital, with an amount of tens of millions of yuan.
Borey Capital is not an affiliate of CATL, but a venture capital firm established with personal capital from Li Ping, co-founder and vice chairman of CATL. On the other hand, CATL's presence can be seen in the business dealings of Qianxun Intelligence: CATL provides batteries for Qianxun's robots, and Qianxun has obtained the identity of "CATL Ecological Enterprise".
Subsequently, in the field of embodied intelligence, CATL invested in Galaxea, Songyan Power, Zhongqing Robot and Weita Power. Among them, in addition to the direct investment from CATL, Puquan Capital and Chendao Capital, the industrial investment platforms under the CATL ecosystem, also participated in the investment in Galaxea and Songyan Power.
Apart from embodied intelligence, a larger energy consumption scenario in the AI era is fixed data centers. In terms of computing power infrastructure, CATL has invested in Silan Technology, Zhongheng Electric and 21Vianet. In terms of energy AI and computing-power-electricity coordination, it has investments in Damao Technology and Deep Intelligent Control.
CATL's most well-known investment in the AI sector is its two participations after DeepSeek opened its investment round. JPMorgan Chase believes that CATL will "transform from a battery supplier to a strategic energy infrastructure partner for AI computing power construction".
CATL's reach may extend further in the future. In September 2026, CATL disclosed that it participated in two funds, Shidai Yiyuan and Shidai Shenyuan, with a total amount of 1.62 billion yuan, and both funds clearly listed artificial intelligence and embodied intelligence as their investment directions.
From the perspective of CATL's AI investment direction, the targets are spread across the upstream and downstream of the AI industry, and are strongly connected to its main business. At the same time, CATL has not invested in the popular AI applications on the market, which also shows that the characteristics of this round of investment are closer to industrial layout rather than financial investment. The logic behind this is to find a suitable entry point based on the hottest current demand for electricity, and carry out a comprehensive closed-loop layout of the industrial chain.
Meanwhile, we also noticed that CATL's layout method is neutral. On the one hand, the shares held by each company are mostly no more than 10%, and it does not seek control of any single enterprise; on the other hand, CATL often invests in many enterprises with competitive relationships. Compared with taking sides, it is more like growing together with the entire AI industry.
Growing Pains of the "Ning King"
From automobiles to robots, the product form has changed, but the energy problem has not disappeared.
When talking about robots in the exclusive interview, Zeng Yuqun did not first talk about how smart their "brains" are, but mentioned the robot marathon: the problem of not being able to finish the whole race lies behind the problems of energy efficiency and energy density. This perspective is very "CATL-style": when the outside world pays attention to how many actions robots can perform, it cares more about what supports these actions to be completed continuously.
Zeng Yuqun also talked about three conditions for entering new fields: the market is large enough, the entry threshold is high enough, and the new technology is adjacent to existing technologies.
As a leading enterprise, CATL also has its own troubles, which come from its main business.
The story of high growth in power batteries has begun to be questioned. The growth rate of the electric vehicle market is slowing down, and automobile enterprises are also increasing their suppliers, building their own battery production capacity and investing in new technologies to improve the autonomy of the supply chain, trying to reduce their dependence on CATL.
For example, Li Auto and XPeng are promoting diversified procurement, GAC has built its own Inbattery power battery, SAIC has invested in Qingtao, and Tesla and Toyota are also following the route of self-developed cells.
For CATL, extending to AI is to continue expanding the growth space of its energy business beyond power batteries. At this time, the AI demand also happens to come knocking. According to the estimation of the International Energy Agency, global data centers consumed about 415 billion kWh of electricity in 2024, and the figure will nearly double to 945 billion kWh by 2030, with the increment mainly driven by AI.
At the same time, AI's demand for electricity is not just "more power", but a package of demands for cells, energy storage, green power, heat dissipation and dispatching systems, many of which match CATL's existing capabilities in batteries and energy storage.
AI hardware also requires high-quality power supply. The current battery life of embodied intelligence is generally between 2 and 4 hours, and to meet the needs of industrial production, it is inevitable to assemble batteries with higher energy density and longer cycle life. From the perspective of the power demand of AIDC, the application of lithium batteries for short-term backup power in data centers is increasing, and energy storage has begun to undertake more functions such as load regulation and green power supporting.
Zeng Yuqun has made a clear explanation for this path. He said in the exclusive interview with Caijing that CATL does not intend to compete with AI enterprises, but hopes to contribute to the energy solutions of AIDC. Earlier, he also proposed that the development and operation of zero-carbon power grids may have a far larger business space than power battery supply.
More importantly, the new story is not just about "AI is short of power, and CATL sells power". AI and robots are not just demand sides, they themselves are becoming production tools for CATL.
Robots entering factories is one of the most obvious trends. In December 2025, CATL disclosed that the humanoid robot of Qianxun Intelligence entered the PACK final inspection production line of CATL's Zhongzhou base. The industrial robots of Galaxea are also operating on a regular basis in CATL's battery factories.
In the long run, what CATL actually wants is an energy operating system in the intelligent era: binding green power and chip computing power in the upstream, locking power supply equipment and energy storage dispatching in the midstream, penetrating into computing power operation and embodied intelligence terminals in the downstream, plus a battery bank and recycling closed loop. Eventually, it will become a platform provider that delivers "reliable power consumption" and grows together with the industry.
Where is the Boundary of the "Seller of Shovels" Logic?
Investing in an AI company only means entering its shareholder list. Whether it can enter the procurement list and continue to get orders are two other separate matters.
This is the first reality that CATL has to face after extending to the AI sector.
Industrial investment can bring earlier technical exchanges and deeper scenario cooperation, but it cannot replace product competition. When robot companies choose batteries, they will ultimately compare weight, power, safety, service life and cost; when data centers choose energy solutions, they will also calculate the power supply reliability, construction cycle and long-term operating costs.
The automotive industry has already demonstrated the complexity of this relationship. The fact that automobile enterprises increase battery suppliers and build their own production capacity does not necessarily mean that they stop cooperating with CATL, but to find a balance between supply guarantee, technology selection and bargaining power. After entering the AI industry, the same considerations still exist: customers may welcome CATL to provide capital and energy capabilities, but they will also reserve space to choose other suppliers.
In addition, the "seller of shovels" is not a universal solution for business, and not taking sides is also a kind of stance: core suppliers supply to competitors, which is a test of business tolerance for any manufacturer.
The larger the volume, the harder it is to remain neutral, which is a paradox inherent in the "seller of shovels" logic.
Another test is when demand can be converted into recyclable revenue.
New energy vehicles have completed the penetration from 0 to 10 million units in ten years, and the demand is certain.
AI is very different. Embodied intelligence is still in the stage of "many prototypes, few mass-produced products" in 2026, and the capital expenditure of AIDC is also highly cyclical. Once the model training efficiency or the structure of computing power supply and demand changes, the scale of demand will be affected.
There are also several technical problems that cannot be bypassed. For example, robot batteries need to meet high energy density, high rate, long cycle and high safety at the same time, and the safety certification and backup power switching standards for lithium battery energy storage in data centers are still being improved.
The real test for CATL has never been whether AI will succeed, but an older question: whether and how to bind the fate of the shovel to that of the gold mine?
Automobile enterprises have given the answer for electric vehicles, and large model vendors, cloud manufacturers and robot companies are giving the answer for AI.
This article is from the WeChat official account "Baobian" (ID: baobiannews), written by Zhang Jingwei, edited by Xing Yun, and published with authorization from 36Kr.