Compared to other major tech companies, JD.com's investments and layout in the AI field focus on its core business, demonstrating a strong pragmatism.

Food delivery and AI are the sectors where Liu Qiangdong currently invests the most energy.

In March this year, Liu Qiangdong visited the Hong Kong University of Science and Technology, toured the Cheng Kar-Shun Robotics Laboratory and the drone noise testing platform, and exchanged ideas with the technical team of the multimodal large language model Mini-Gemini V2.

At the same time, JD.com also adjusted its AI R&D departments. JD Explore Academy, originally under JD Technology, became a directly managed department of JD Group, with Liu Qiangdong personally serving as dean. Explore Academy became JD.com's core R&D department in the AI field.

The original dean, He Xiaodong, became the vice dean, but he was also promoted from vice president to senior vice president of JD Group, becoming the de facto head of JD's AI R&D.

According to Digital Frontier, the upgrade of Explore Academy to a directly managed department of the group was not only due to increased emphasis on AI by senior management but also because the results achieved by Explore Academy in recent years have been recognized by Liu Qiangdong. For example, in April 2024, Liu Qiangdong used a digital human developed by JD Explore Academy for his first live streaming product promotion. "He believed that the scenario was already good enough for him to be willing to use his own image as a spokesperson."

In addition, JD.com also unified its large model Yanxi into JoyAI. The reason for unifying the AI brand image was that the previous Yanxi was the brand of Explore Academy, but various business segments had both vertical large models and some applications of large models, leading to internal confusion regarding the AI brand. This year, JD internally solicited a unified name for AI, hoping that people could immediately recognize it as JD's AI brand. Joy has always been JD's official mascot, so most subsequent JD AI applications will be prefixed with "Joy."

Behind these actions, what kind of AI game plan has Liu Qiangdong, who has returned to the forefront, laid out for JD.com?

01

Investing to Pave the Way, Betting on Embodied Intelligence

The investment boom in embodied intelligence has sparked curiosity about JD.com's AI layout. JD.com is the most active internet giant in embodied intelligence, investing in six projects in less than three months, a speed and density that surprised the outside world.

The latest investment occurred in early August when embodied intelligence equipment and sensor company "Pasini" announced the completion of a new round of Series A funding, led by JD.com. The founding members of this company come from the Robotics Laboratory of Waseda University in Japan, and the company has completed RMB 1 billion in funding in just four months.

On July 30, embodied intelligence startup RoboScience announced the completion of nearly RMB 200 million in angel round funding, led by JD.com. What really attracted a lot of attention from the outside world was on July 21, when three embodied intelligence companies, Qianxun Intelligence, Zhujidongli, and Zhongqing Robotics, announced financing news on the same day. Among them, Qianxun Intelligence completed nearly RMB 600 million in Pre-A+ round funding; Zhongqing Robotics consecutively completed Pre-A++ and A1 rounds, raising nearly RMB 1 billion with a latest valuation of RMB 7 billion; Zhujidongli did not disclose the specific funding amount.

However, the lead investor for all three projects was the same company - JD.com. Earlier in May, JD.com led the Series B+ round of funding for Zhiyuan Robotics. As one of the hottest embodied intelligence projects at the moment, Zhiyuan is backed by a large number of industrial capitalists and has a valuation exceeding RMB 20 billion.

According to He Xiaodong, JD.com's investments in the field of embodied intelligence "will not stop here," and there will be further investment announcements.

JD.com's bet on embodied intelligence is not surprising. JD.com has always been interested in the robotics sector in its early years, including logistics distribution and unmanned warehousing, where it has accumulated relevant technologies. For example, JD.com's Asia No. 1 warehouse uses many unmanned warehousing technologies.

As early as the 2023 World Artificial Intelligence Conference (WAIC), He Xiaodong revealed that there are two directions for future general AI: one is multimodal, and the other is embodied intelligence.

Since then, JD.com's layout in the field of embodied intelligence has focused on solving the challenge of training data.

In April 2024, JD Explore Academy constructed JD ManiData, China's first open-source dataset for dual-arm mobile robot manipulation, including proposing an end-to-end spatio-temporal interactive understanding multimodal large model. In February this year, JD.com, in collaboration with a number of enterprises and universities such as SLAMTEC, Digua Robotics, and the University of Science and Technology of China, launched the industry's first embodied intelligence "atomic skill library" framework, which alleviates the scarcity of embodied intelligence data by decomposing tasks into reusable skill modules.

It is not unusual for large companies to build their business landscapes through strategic investments, but such intensive moves in the same sector within a short period are not common in JD.com's past investment history.

Prior to this, JD.com's investments in the AI field were actually not numerous. During the last wave of large model investment fever, the "AI Six Dragons" received heavy investments from Alibaba, ByteDance, and Tencent, with JD.com barely participating.

In 2017, JD.com began its technological transformation, and Liu Qiangdong summarized JD.com's strategy for the next 12 years as "technology, technology, technology" and repeatedly mentioned the impact of AI technology on life. JD.com also significantly increased its pace of external investments.

However, JD.com's investments and input in AI during this round almost entirely revolved around the logistics sector. For example, JD.com established the X and Y business divisions around 2016, with the former focusing on logistics robots and the latter building smart supply chains through big data and AI.

The same is true for the investment projects. In 2017, JD.com participated in the RMB 300 million strategic financing of iFlytek, which specializes in speech recognition technology; in 2018, JD.com invested in Keda Intelligent, a manufacturer of logistics robots and warehouse automation, and the logistics sector has always been at the forefront of JD.com's AI layout. In February 2019, JD.com invested in autonomous driving technology service provider AutoX. Later, JD.com made another bet by investing in Inceptio Technology, which focuses on autonomous truck driving.

In the following years, JD.com hardly made any more moves in the AI field.

The most recent investments in the AI field were in December 2023 and July 2024, when JD.com made strategic investments in cooking robot company Xianglu Technology in two separate rounds. Besides capital, the two parties also cooperated at the business level. Today, some of Xianglu Technology's cooking robots are used in JD.com's 7FRESH Kitchen, which is part of its food delivery service.

Interestingly, both Meituan and JD.com have invested in a large number of AI enterprises, but their investment targets hardly overlap, as if intentionally avoiding each other. The last time both invested in the same company was in 2021, when both Meituan and JD Logistics invested in Inceptio Technology, an autonomous truck driving company.

However, JD.com's investment strategy in the field of embodied intelligence in this round is quite different from before, shifting more from ToB to ToC, hoping that technology can be more directly targeted at the broader family user market.

"From industrial applications to companionship and household services, I believe this space will be much larger, possibly one to two orders of magnitude larger, and the user experience will be the most direct," said He Xiaodong. JD.com's embodied intelligence brand Joy Insight also focuses more on serving end products, including AI toys, robots/robot dogs that emphasize companionship and education.

02

High Salaries to Attract Talent, Stirring Up the AI Talent Market

Besides investing money, JD.com is also poaching talent from all over.

At the beginning of this year, JD.com launched the "Top Young Technology Talent Program," offering heavy remuneration to recruit top AI-related talent globally, with no salary cap and even offering monthly salaries exceeding RMB 10,000 to interns in production and research positions.

JD.com posted 3,000 job openings on Maimai, and the urgent positions were all related to AI, including multimodal algorithm engineers and large model algorithm engineers, requiring only one or two years of work experience, offering 19 months of compensation, with annual salaries basically around RMB 1 million.

Currently, several AI experts have successively joined JD.com. These include Duan Nan, head of the video generation model at Jieyue Xingchen and Tech Fellow, and Yan Zhijie, head of the speech team at Alibaba's Tongyi Lab, who have joined JD.com to lead the vision and multimodal lab and speech team, respectively. Gu Simiao, chief architect of Baidu Search, was also reported to have joined JD.com as the head of the AI Application and Innovation Department.

Embodied intelligence is also an important area for JD.com to poach talent. In April 2025, media outlets revealed that Shen Hui, former VP of SenseTime, joined JD.com as the head of embodied intelligence. JD.com also recruited many talents from SenseTime, including some technical backbones from the "Yuanluobo" project who followed Shen Hui to JD.com.

However, the person officially designated as the head of embodied intelligence at JD.com is Dai Wenjun. Dai Wenjun previously served as the director of visual intelligence products at Alibaba's DAMO Academy.

The last time JD.com recruited AI talent so intensively was after it established its technology transformation strategy in 2017, including Zhou Bowen, the chief scientist of IBM, who joined JD.com in 2017 as vice president of JD Group, responsible for JD.com's AI research and platform-related business, becoming the leader of JD.com's AI field.

Since then, a group of technical experts, including He Xiaodong, the chief NLP researcher at Microsoft's Redmond Research Institute, Bo Lefeng, the chief scientist at Amazon, Pei Jian, a professor in the Department of Computing Science at Simon Fraser University in Canada, Peng Jian, an assistant professor at the University of Illinois at Urbana-Champaign, Zheng Yu, the head of urban computing at Microsoft, and Mei Tao, a senior researcher at Microsoft Research Asia, have successively joined JD.com.

Zhou Bowen, He Xiaodong, and Mei Tao are considered the leading figures of JD.com's early AI department. He Xiaodong and Mei Tao were also selected as IEEE Fellows in early 2019, marking the first time that incumbent scientists at JD.com have been promoted to IEEE Fellow.

In 2020, JD.com established JD Explore Academy. Liu Qiangdong set a goal for this department to recruit 18 global top scientists within three years. Moreover, shortly after the establishment of JD Explore Academy, Tao Dacheng, an internationally renowned scholar in the field of artificial intelligence and information science, also joined JD.com as the first dean of Explore Academy.

However, among the industry experts who joined JD.com during the last AI boom, only He Xiaodong, Zheng Yu, Xue Chao, and others are still in core positions, with many others having left the company successively.

According to a person familiar with JD.com, after many technical experts entered JD.com, there was a disconnect between their concepts and JD.com's development needs. For example, many technical experts from universities wanted to maintain their academic influence or use corporate resources to continue their research and publish academic papers. However, JD.com required them to devote themselves fully, requiring an overall assessment of investment and returns, so that technology could be truly applied in business.

Today, many of these AI experts have returned to their university positions after leaving JD.com.

This conflict of ideas is also reflected in the focus of JD Explore Academy. In its early days, JD Explore Academy mainly focused on three areas: quantum machine learning, trustworthy AI, and deep learning. However, with the wave of technological iteration, the latest introduction of JD Explore Academy has been replaced by large models, embodied intelligence, and multimodal intelligence.

Behind these transformations lies a single point: JD.com's technological investment is more focused on its core business.

03

Serving the Core Business, JD.com's Ultimate Pragmatism

Like other major tech companies, JD.com has constructed an AI technology landscape encompassing the full stack from underlying computing power to models, and then to tools and applications.

At the bottom layer is the AI computing power infrastructure JD Cloud, including distributed storage Cloud Sea, container platform Cloud Ship, data platform Deep Sea, large model integration machine, etc.

At the upper layer is the full-size base large model JoyAI, covering 3B, 10B, 81B, and the latest 750B, encompassing various modalities such as language, speech, image, and video.

Above that are various AI platforms and tools, including the Agent platform (already open source), auxiliary programming platform, model development platform, embodied intelligence platform, digital humans, etc.

At the top are various AI applications, focusing on JD.com's advantageous business segments, including intelligent logistics sorting, intelligent customer service, retail sales forecasting, marketing material generation, health intelligent inquiry, and other scenarios.

Internet companies like JD.com share a common characteristic: the rapid development of their own businesses gives rise to the demand for technology. For example, JD Cloud's high-performance storage Cloud Sea was initially developed to meet JD.com's internal high-concurrency needs.

It is not difficult to find that when choosing directions and implementing technologies in AI, JD.com will prioritize satisfying its core business and focusing on industrial applications.

The layout of embodied intelligence is a typical example. Therefore, at this year's WAIC exhibition, JD.com displayed a large number of hardware products related to robots and embodied intelligence, such as AI toys, companion robots, chess-playing robots, robot dogs, etc. One of JD.com's goals is to turn these intelligent terminal devices into popular items on JD.com's e-commerce platform, forming a user mindset of "buy robots on JD.com."

JD.com specifically proposed the concept of JoyInside embodied intelligence. JD.com's AI intelligent agent serves as the brain, implanted in various terminal devices. On the one hand, it builds a commercial landscape for embodied intelligence through investment; on the other hand, it attracts brand owners to settle in through traffic incline and technology empowerment.

The most typical example is various AI toys. Currently, chess-playing robot Yuanluobo, AI fashion brand Fuzozo, Huohuotu, and others have settled on JD.com through the JD JoyInside model. Even large humanoid robots like Zhujidongli and Zhongqing Robotics have opened flagship stores on JD.com, with prices reaching RMB 180,000. However, the current number of JoyInside AI toys is limited, and the prices are relatively expensive, so no true blockbuster has emerged yet.

In the retail sector, AI technology is integrated into JD.com's retail business processes, covering shopping guides, marketing, customer service, etc. For example, the Jingyan intelligent shopping guide assistant can provide consumers with more accurate shopping suggestions, and the Jingdiandian AIGC platform can help merchants quickly generate product images, operational copywriting, short videos, etc.

Digital humans are one of the areas where JD.com has invested heavily, used in live streaming scenarios on e-commerce platforms. Liu Qiangdong also completed his first live streaming product promotion in April 2024 in the form of a digital human, generating over RMB 50 million in sales in less than an hour. Today, over 9,000 merchants are already using JD digital humans for live streaming, achieving large-scale industrial practice.

In the health sector, JD Health's "AI Jingyi" has launched over 500 expert doctor intelligent agents, providing AI diagnosis and treatment services, medical record summaries, medication reminders for patients, etc.

In the industrial sector, JD Industry has also applied large models to scenarios such as commodity sourcing and compliance management, including the creation of the industrial product commodity standard library "Mercator," which standardizes industrial product coding and significantly improves the efficiency of supply chain management.

These applications that JD.com focuses on correspond to JD.com's core business segments - retail, logistics, finance, health, and industry. This is also JD.com's advantage. With a large amount of scenarios and data accumulated in these fields and industries, JD.com can quickly implement AI.

For example, JD.com is currently heatedly competing in the food delivery sector and needs to recruit a large number of riders. Therefore, JD.com internally launched a recruitment officer Agent, allowing AI intelligent agents to replace HR in completing the entire process of interviews, information verification, approval for employment, etc. This Agent has already interviewed over 100,000 people.

"We are more concerned about whether it is truly commercially viable and whether it can create tangible closed-loop value for partners and JD.com," said He Xiaodong. He revealed that there are already over 1,000 scenarios within JD.com using JD.com's large models. He also specifically mentioned that JD.com's internal businesses can freely choose large models, including closed-source or open-source models.

An insider also expressed the same view to Digital Frontline, "There is a big difference between JD.com and other companies, that is, all JD.com's technologies must serve the main business of the supply chain." This is also Liu Qiangdong's positioning of JD.com, with all businesses centered around the supply chain.

JD.com's investment in technology shows a strong pragmatism, "No matter how cool the technology is, it must be able to be implemented in business."

One important reason behind this is that compared to other major Internet technology companies, JD.com is a company with very low profit margins, unable to support many purely theoretical research. Investments need to see returns. Taking 2024 as an example, JD.com Group's net profit was 41.3 billion yuan, with a net profit margin of 3.3%; while Tencent's net profit was 222.7 billion yuan, with a net profit margin as high as 33%; Alibaba's net profit was 125.9 billion yuan, with a net profit margin of 12.6%.

However, since JD.com's transition to technology in 2017, over the past eight years, it has accumulated investments of more than 140 billion yuan in research and development. Now, with Liu Qiangdong returning to the forefront, JD.com's layout in AI will take further actions around its main business.