On September 25th, JD.com held its JDDiscovery-2025 Global Technology Explorer Conference in Beijing under the theme "Enjoy AI." The company conveyed several key messages during the event: releasing an AI panorama that systematically showcases its overall AI strategic layout, along with a commitment to continue investing over the next three years to drive various industries toward forming a trillion-yuan artificial intelligence ecosystem.

Industry application has become the keyword for JD.com's AI implementation and exploration.

"At the 2023 JDD conference, we proposed a formula: the value of large models = algorithm × computing power × data × the square of industry depth. After two years of practice and exploration in artificial intelligence, we have updated our understanding of AI value. We now believe that the value of artificial intelligence = model × experience × the square of industry depth," said Xu Ran, Vice Chairman of JD.com's SEC and CEO of JD.com Group, in his keynote speech. He emphasized that JD.com does not pursue trendy AI for its own sake, but rather seeks sustainable development and AI that truly creates value for industries.

**Industry Foundation Reflected in Value Formula**

The squared effect of "industry depth" stems from JD.com's long-term cultivation in retail, logistics, healthcare, and industrial sectors. From nationwide warehousing and logistics networks to management experience with tens of millions of SKUs, from massive consumer behavior data to industrial supply chain service practices, these are not simulated laboratory data but valuable assets that have been tested through commercial operations.

JD.com adheres to a pragmatic approach of "validating through internal business first, then providing external output after maturity," and has simultaneously launched deep applications in core B2B sectors including retail, logistics, healthcare, and industrial services.

In logistics, the fully upgraded Logistics Super Brain large model 2.0 has comprehensively moved toward multimodality, marking the logistics "super brain's" evolution from decision support to embodied execution. Through collaboration among multiple intelligent devices, it significantly improves logistics supply chain operational efficiency, with standardized operation levels increasing by 15% and human-machine collaborative efficiency improving by 20%. The newly released "Alien Wolf" embodied intelligent robotic arm represents another breakthrough innovation in the self-developed "Wolf Series," capable of solving automated palletizing challenges for massive non-standard packages, greatly improving logistics warehousing efficiency and accuracy. Currently, "Wolf Series" intelligent equipment has been deployed in over 500 warehouses globally, covering the full chain of storage, handling, and sorting, marking a generational leap in next-generation logistics intelligence.

In retail, JD.com publicly unveiled its e-commerce innovation AI architecture system Oxygen for the first time. This system relies on the Joy AI large model to create rich system capabilities and diversified intelligent agents, using AI technology to build intelligent application scenarios for e-commerce shopping and supply chain management.

In October, JD.com will release the intelligent search and recommendation feature "AiGou" in JD App version 16.0. This function uses the semantic controllable generative recommendation model OxygenRec and the e-commerce multimodal understanding large model OxygenVLM as core supporting technologies combining fast and slow thinking. Through natural language interaction, it precisely understands user needs, achieving breakthrough efficiency improvements that revolutionize traditional shopping methods and providing users with truly personalized shopping experiences.

In healthcare, JingYi QianXun 2.0 can understand medical reports and listen to symptom descriptions. As a trusted reasoning model, it can strictly simulate doctors' clinical diagnostic thinking paths, deeply integrating medical knowledge systems with real diagnostic logic to comprehensively serve complex specialized disease diagnosis and treatment. Meanwhile, JD Health AI Hospital 1.0, carrying next-generation model capabilities, has officially launched, bringing together AI JingYi intelligent agents that include simulated doctors, pharmacists, nutritionists, and other professional roles, as well as over a thousand expert doctor intelligent agents that simulate individual expert diagnostic experiences, providing users with AI triage, AI pharmacy, AI medical center, and other capabilities.

In the industrial sector, the industrial supply chain large model JoyIndustrial drives an industrial supply chain efficiency revolution through "data + intelligence," providing core engines for cost reduction and efficiency improvement across industries. This model is trained on data from over 57.1 million industrial SKUs and accumulated experience from 40+ subdivided industries, serving over 10,000 key industrial enterprises and precisely addressing "high cost, low efficiency" pain points. The industry's first industrial supply chain trillion-yuan cost reduction model released jointly by JD Industrial and the National Research Institute for Big Data shows that supply chain digitalization and intelligence will bring 6.77 trillion yuan in cost reduction space for the industry. JD Industrial's nine major industrial industry and robot-specific scenario solutions will help Chinese industry achieve "trillion-yuan cost reduction."

Additionally, JD.com is gradually opening logistics, healthcare, and other industrial scenarios and supply chain data to support industry model training and application testing, strengthening B2B ecosystem empowerment. It has launched Digital Person Platform 4.0 (industry brand spokesperson digital persons, reducing live streaming costs to 1/10 of real persons), JoyAgent 3.0 intelligent agent platform (100% open source, with over 30,000 internal intelligent agents operating), JoyCode 2.0 code platform (integrating intelligent agent technology, shortening product development cycles by 30%), and other tools to further help enterprises reduce costs and improve efficiency.

Cao Peng, Chairman of JD.com Group's Technology Committee and President of JD Cloud, stated that facing the new era of artificial intelligence, deep applications are the fulcrum for crossing into super intelligence. "We believe that AI application development models will definitely differ from previous application development. They will become more agile and efficient, with AI nativeness. The integration of AI intelligent agents and code will become the new paradigm for next-generation AI development, which is also a very critical scenario for building deep AI applications for enterprises."

This industry-rooted strategic choice has allowed JD.com to avoid the trap of "technical idle running" in the AI competition and find the core driving force for sustainable development.

**Making AI Usable, Perceivable, and Synchronizable**

JD.com's thousands of scenarios generate calls for over a trillion tokens daily.

Thanks to self-developed investments and co-creation with partners, JD.com has formed complete capabilities across smart computing infrastructure, models, platforms, scenarios, and products, achieving AI usability.

AI application perception is reflected in frontend interactions with consumers.

The newly released personal AI product "JingXi" is the super portal for next-generation shopping and lifestyle services. This soon-to-be-launched app integrates AI technology throughout the entire shopping process, understands user needs, deeply comprehends products, can select the most desirable items for users, and makes the checkout and ordering process more intelligent and smooth. Beyond shopping, ordering food, buying airline tickets, and booking hotels can all be completed through voice commands on the app.

"JD.com's light conversational shopping format combined with deep thinking shopping format may represent a new possibility for future e-commerce formats. To achieve this, we rewrote all inference engines and matched them with corresponding underlying computing resources to achieve optimal effects. The investment in this area is enormous," Cao Peng said.

He mentioned rebuilding the next-generation recommendation engine using new generative recommendation methods. The reconstructed integrated search and recommendation architecture can controllably address users' shopping concerns according to their needs while continuously integrating large model world knowledge.

This means intelligent agents can also serve more scenarios. For example, "TaTaTa" is a universal digital assistant. In this app, there's a digital intelligent agent called Universal Doctor that can answer almost any user question and chat with users like an old friend, from checking weather and asking for news to ordering takeout and discussing finance. "TaTaTa" also supports users creating their own intelligent agents and finding like-minded friends through AI circle interactions. Additionally, users can install "TaTaTa" into intelligent devices such as robot dogs and AI toys, giving these hardware devices a soul.

"We don't want technology to remain only in PowerPoint presentations. We hope to deliver the most advanced digital person large models directly to every C-end user. The TaTaTa app released today includes an AI same-style function and a zero-frame digital person mini-program. We enable every user to instantly generate movie-quality digital persons. This is technology that truly achieves 'everything can speak' in the industry and the first commercializable movie-quality digital person product," said He Xiaodong, Senior Vice President of JD.com Group and Deputy Director of JD.com Group's Exploration Research Institute.

Furthermore, the launched JoyInside embodied intelligent agent demonstrates JD.com's ecosystem openness, breaking enterprise boundaries and achieving AI synchronization. Whether it's software-hardware integration for various intelligent device scenarios like robots, AI toys, and smart glasses, or better anthropomorphic dialogue with users through coordinated speech and action to provide better product experiences, smart hardware accessing JoyInside sees users initiating conversations more actively, with conversation frequency increasing by over 120% on average. The latest version has been successively integrated into hardware products from over 30 leading brands and cooperates with over 10 leading robot brands including Zhongqing and Unitree.

AI technology is transitioning from "conceptual hype" to "deep implementation zones." JD.com has proven through a complete strategic layout and practical results that true AI value lies not in parameter heights or demonstration brilliance, but in whether it can take root in industrial soil and solve real problems. JD.com is forging a path of AI development that "creates no gimmicks, only does practical work." This industry-rooted, value-centered approach not only injects growth momentum into JD.com itself but also provides a valuable model for the sustainable development of China's AI industry.