Moving beyond superficial traffic tactics, JD.com's AI strategy demonstrates industrial depth. Twenty-two years later, Richard Liu has placed another major bet. In early 2004, when China's retail industry was dominated by offline chain models, Liu made a decision that surprised all his employees. He shut down the dozen-plus offline stores he had painstakingly built and staked his entire fortune on a website called "Jingdong Multimedia Network." He firmly believed that internet technology was reshaping the experience, cost, and efficiency structure of retail, and JD.com had to position itself at the source of this transformation. His idea initially received little support, but subsequent events proved his foresight.
Today, 22 years later, while most people are engrossed in chatting with AI and most companies are still fixated on competing for AI traffic gateways, Liu has realized that the true value of AI lies in the physical world. This time, however, he sees not just a new retail channel, but a new form of productivity capable of serving the entire society. AI is reshaping the fundamental logic of myriad industries in a deeper way.
On May 12, JD.com released its Q1 2026 financial results. During the reporting period, revenue reached 315.7 billion yuan, a year-on-year increase of 4.9%. As of the end of the quarter, annual active users exceeded 740 million, marking the tenth consecutive quarter of double-digit growth since Q4 2023. Beyond the operational figures, an "AI portrait" is unfolding: from open-source large models to digital humans capable of "mukbang" (eating broadcasts), from the "Wolf Pack" robots covering the entire logistics chain to the world's largest embodied intelligence data collection center, AI has been integrated into every business line of JD.com.
As the entire AI industry evolves from competing on computing power and model parameters to competing on ecosystems and self-contained loops, JD.com has also established a business model around AI spanning model computing power, industrial applications, and user experience: the value of artificial intelligence = Model × Experience × (Industrial Depth)². Behind this seemingly simple formula lies the full confidence accumulated from JD.com's 22-year deep cultivation of its supply chain.
**1. R&D Growth Exceeds 50% for Two Consecutive Quarters; AI Infuses Every Business Line**
Unlike other companies that broadly discuss AI visions in their earnings reports, JD.com's Q1 report focuses more on demonstrating the specific changes AI has brought to its various business lines. Supporting these changes is, first and foremost, the continuously increasing R&D investment. In Q1 this year, JD.com's R&D expenses grew by 48.6% year-on-year to 6.9 billion yuan. This marks the second consecutive quarter with growth exceeding 50%, following a 52% year-on-year increase in R&D investment in Q4 2025. Prior to this, JD.com's R&D investment growth rate was mostly below 30%. Such high-intensity R&D investment is uncommon within the JD.com system, known for its "restraint and efficiency."
CEO Sandy Xu stated on the earnings call, "R&D expenses will continue to grow steadily in the coming period. These investments will gradually translate into operational dividends, achieving AI-driven efficiency improvements and driving an overall improvement in the cost structure." This indicates that JD.com is not blindly following the trend in AI investment but has seen the tangible output of AI technology.
These returns are reflected across business lines. In the core retail business, JD.com's self-developed AI Agent "Jingyan" underwent a comprehensive upgrade during the period. Xu revealed on the call that in Q1, Jingyan's quarterly active users increased by over 200% year-on-year, and user interactions grew by over 300% year-on-year. When users make purchasing decisions, AI is increasingly replacing steps that previously relied on manual searches and comparisons. Furthermore, the multi-agent life service assistant "JD AI Buy" has also been launched. It can provide instant responses to lightweight requests and, when faced with complex demands, coordinate multiple specialized agents to complete tasks collaboratively, forming a complete closed loop from "understanding needs" to "completing orders and delivery."
Changes in logistics are more evident in efficiency gains. JD Logistics released the upgraded embodied intelligent device "Yi Wolf" robotic arm in Q1, evolving from a single-arm to a dual-arm design specifically for parcel grasping and palletizing. The upgraded arm can significantly improve parcel sorting efficiency within warehouses. To date, JD.com's "Wolf Pack" series of robots covers the entire logistics chain and has been deployed at scale globally.
In the healthcare sector, while most AI medical assistants remain at the level of simple Q&A life assistants, JD Health has achieved medical-grade AI capabilities. The AI doctor can not only provide advice but can also directly connect users to online consultations, at-home rapid testing, nurse home visits, and medicine delivery, forming a complete service loop. In Q1, JD Health also launched its first medical device product equipped with JoyInside and announced plans to enable AI capabilities for one million elderly-friendly devices within the next year, deeply serving the needs of an aging society.
Finally, in the industrial sector, leveraging JoyIndustrial, the industry's first supply chain vertical large model, JD Industrial has compressed the processing time for hundreds of thousands of SKUs from "months" to "hours," improving human efficiency by over tenfold.
Across the four business lines of retail, logistics, health, and industry, JD.com's application of AI essentially follows the same logic. "Regardless of how technology evolves, its core always lies in providing users with a better experience, lower costs, and higher efficiency to meet their continuous pursuit of better products, better prices, and better services," Xu said. This is precisely the core model JD.com has built over the past 20-plus years through its deep cultivation of the supply chain.
In addition to letting users experience the convenience brought by AI, JD.com is also using AI internally to improve efficiency. According to management, JD.com is integrating and upgrading various work segments through the application of Agents, ultimately connecting the entire workflow. Agents have replaced many inefficient intermediate steps, helping the group achieve operational efficiency gains. JD.com is not using AI to create a flashy new story but to more efficiently propel the company's development with the experience accumulated over more than two decades.
**2. Deconstructing the A-B-C Positive Cycle: JD.com's AI is Not a Traffic Converter**
AI is merely a tool. Some companies choose to place this tool at the front end to achieve more precise recommendations and higher payment conversion rates. But JD.com has taken a different path, building a mutually reinforcing positive cycle between AI (A), industry (B), and users (C).
In this cycle, AI capability is the starting point and the foundation for the cycle to run. Currently, from open-source large models to a localized computing power closed loop, JD.com has established a complete technology stack. In Q1 this year, JD.com open-sourced the Instruct version of its foundational large model, JoyAI-LLM Flash, for the first time. It introduced the FiberPO reinforcement learning method based on "fiber bundle" theory, placing it in the industry's first tier across 19 authoritative benchmarks. Its greatest distinguishing feature is not its parameter count but its efficiency. When completing the same task with comparable accuracy, JoyAI-LLM Flash consumes only 20% to 25% of the Tokens used by other major industry models. Currently, Tokens have become the primary means of production in the AI era. Completing the same task with fewer Token consumption means stronger economic benefits and higher work efficiency.
Model capability is equally dependent on underlying AI infrastructure support. JD.com's self-developed JoyScale heterogeneous computing platform and JD Cloud Sea AI storage system form a complete full-stack support. These achievements demonstrate that JD.com's AI capability is a complete, self-built system from the underlying computing power and intermediate models to the upper-layer agents.
Moreover, JD.com has not monopolized these technological capabilities within its own business system. Instead, it has chosen to solidify them into productivity open to the entire industry. This process is the transition from "A" to "B." The most representative example is the digital human livestreaming service. JD.com offers the JoyStreamer digital human to merchants for free, launching "Free-State Digital Humans" for five major industries including home appliances & furniture and fashion apparel. This breaks the limitations of traditional digital humans' stiff movements and fixed postures, supporting natural walking, flexible posing, and camera following.
To date, the service has covered over 70,000 merchants. Merchants can use this technology to quickly set up 24/7 non-stop livestreaming rooms at one-tenth the cost of using real humans. For the recently hot embodied intelligence sector, JD.com has also combined its strengths in industry and AI technology to address the industry pain point of scarce training data, establishing the world's largest embodied intelligence data collection center. According to JD.com's plan, the data collection center will accumulate 5 million hours of real human scenario video data within the next year, exceeding 10 million hours within two years, while simultaneously collecting 1 million hours of robot本体 data. This tackles the core bottleneck of "scarce high-quality real-scenario data" in the embodied intelligence field from the source.
AI entering the B-side is only the first step in this cycle theory. For the cycle to continuously generate value, it must also penetrate C-end users. This is also the layer where JD.com's AI capability is most easily perceived by the general public. The air conditioner used daily at home or an exquisite plush toy may have the support of JD.com's AI capabilities behind them.
The Xinfei AI Voice Air Conditioner equipped with JD.com's JoyInside embodied intelligence was exclusively launched on JD.com. It significantly lowers the operational barrier for the elderly and children through voice interaction. Although it seems like a simple voice control scenario, it involves the complete integration of large model capabilities, appliance protocols, and home scenario understanding.
Based on insights into family companionship scenarios, JD.com, through its Jingzao team, also created an AI toy product that conducts voice interactions tailored to the emotional needs of different age groups: children, adults, and the elderly. The product went from project initiation to delivery in less than a month. Within six months of launch, the AI plush toy topped the list for highest positive review rate in JD.com's plush toy category.
More crucially, all this real interaction data from the user end is continuously feeding back into JD.com's own models, forming a genuine closed loop: AI capabilities empower industries, industrial products improve user experience, and real user usage nurtures stronger AI capabilities.
**3. Unique Competitiveness Shaped by the Super Supply Chain**
Not everyone can successfully run the "A-B-C" cycle. The fundamental reason JD.com can integrate AI into the capillaries of industry is that its super supply chain, cultivated over 20-plus years, forms a moat that is difficult to replicate.
What is a supply chain? In the retail industry, the long, complex chain comprising product sales forecasting, inventory management, integrated warehousing and distribution logistics systems, and after-sales customer service constitutes the main body of the supply chain. Over the past 20-plus years, JD.com has pushed the depth and breadth of this supply chain to the extreme.
The breadth of JD.com's supply chain covers multiple real-world scenarios such as retail, logistics, health, and industry. In the AI era, algorithms are not scarce; the scarcest resource is training soil from the real world. While most companies are still training large models with publicly available internet data, JD.com already possesses scenario depth that others find difficult to access. JD Logistics' "Wolf Pack" robot army, covering over ten countries globally and more than 20 provinces nationwide, generates real embodied intelligence training data with every grasp, move, and sort. JD Industrial's years of deep cultivation have accumulated vast industrial data covering the entire chain of supply, demand, and fulfillment, providing JoyIndustrial with professional production knowledge that general large models cannot access.
From JD Retail's consumer decision data and JD Logistics' full-chain fulfillment data to JD Health's medical interaction data and JD Industrial's manufacturing and procurement data, together they form a data pool precipitated by real transactions, real fulfillment, and real services. This breadth directly raises the ceiling for JD.com's AI capabilities.
Beyond breadth, this system also has depth. Behind the JD AI Doctor is a complete supply chain covering consultation, testing, delivery, and nursing. Behind AI Buy is the full chain composed of JD.com's self-operated products, platform merchants, and delivery system. Behind the AI plush toy is the production capacity support from over 3,000 cooperative suppliers and 800 core partners of Jingzao. JD.com's ability to mobilize over 100,000 internal employees to carry out the "largest-scale human data collection operation" also relies on decades of heavy asset investment and operation in logistics and supply chain. Pure internet companies would find it difficult to dispatch such a "digital workforce" that truly penetrates the physical world.
Returning to the initial value formula: the value of artificial intelligence = Model × Experience × (Industrial Depth)². The model layer might be caught up with, and the experience layer can be easily imitated. However, the depth of real scenarios extending from retail, logistics, health, and industry, along with the industrial thickness built by the supply chain system accumulated over 22 years, cannot be replicated overnight. This is why JD.com dares to square this term. In a multiplicative relationship, squaring means exponential amplification.
In this Q1 report, the "industrial squared effect" has already found concrete validation: Driven by AI, JD Retail's operating profit margin increased from 4.9% in the same period last year to 5.6%, with quarterly operating profit reaching 15 billion yuan. JD Logistics revenue grew by 29% year-on-year, and its operating profit margin significantly improved from 0.3% in the same period last year to 1.7%. The average user shopping frequency increased by 37% year-on-year. These achievements are, to varying degrees, attributable to AI's enhancement of operational efficiency and improvement of user experience.
When AI is no longer an abstract future narrative but is genuinely realized in the form of corporate operating profit margins, user stickiness, and efficiency curves, JD.com's positioning should also shift from a traditional e-commerce player to the world's largest physical world operations center. Twenty-two years ago, Richard Liu saw clearly the potential of "Internet + Retail." Twenty-two years later, JD.com is placing its bets on "AI + Super Supply Chain." Because for JD.com, AI is not just computing power; it is new productivity oriented towards the physical world.