"JD.com does not blindly pursue campaign-style AI, but seeks sustainable development and AI that truly creates value for industries," stated Xu Ran, CEO of JD.com, revealing the underlying logic of JD's AI strategy. In her view, AI should not merely be a temporary buzz and fanfare, but should have long-term planning and tangible industrial contributions.

On September 25, at the 2025 JD Global Technology Explorer Conference (JDD), Xu Ran further elaborated on JD's unique understanding of artificial intelligence value. She pointed out that the value of artificial intelligence should be jointly defined by "Model × Experience × Industry Depth²." This means that artificial intelligence must not only have powerful models as support, but also focus on user experience while being deeply rooted in industries, achieving deep integration and collaborative development with various sectors.

The "industry depth" emphasized by Xu Ran represents JD's core advantage in AI deployment. Years of deep cultivation in complex scenarios including retail, logistics, healthcare, and industrial sectors have accumulated data resources and business insights that make JD's AI capabilities more aligned with actual scenario needs.

Industry observers note that the current AI sector generally faces challenges such as disconnection between technology and scenarios, and difficulties in commercial implementation. JD.com has chosen to open high-value scenarios, supply chain data, and co-build ecosystems with industries, allowing AI to subtly help various industries achieve digital transformation needs.

**Building a Trillion-Scale Artificial Intelligence Ecosystem**

At the JDD conference two years ago, JD.com proposed that the value formula for large models was: Algorithm × Computing Power × Data × Industry Depth². After nearly two years of practice, JD's understanding of AI value has evolved to: "Model × Experience × Industry Depth²."

This transformation reflects the prevalent pain point in the current AI industry of disconnection between technology and scenario implementation. Currently, pursuing artificial general intelligence and creating phenomenal applications has become an industry focus, while large models, as disruptive new technologies, are still in early development stages. The influx of numerous enterprises and capital has also generated certain bubbles.

As investment enthusiasm gradually recedes, how to truly serve the real economy with AI technology and achieve commercial implementation remains a core issue that the industry urgently needs to address. Taking the currently popular robotics technology as an example, its ultimate goal should be integration into actual life and improving production efficiency, rather than merely remaining in the research demonstration phase. If AI cannot deeply integrate into the real economy and optimize supply chain efficiency at all levels, it will ultimately remain theoretical.

In response to this situation, JD.com has chosen to focus on actual technical experience and integration effects with industries, rather than unilaterally pursuing model parameter scale. Whether in retail marketing, logistics warehouse optimization, or health consulting, AI's value lies in genuinely integrating into and optimizing existing business processes.

Regarding JD's artificial intelligence planning, Xu Ran indicated that JD's goal is to continue investing over the next three years, driving the formation of a trillion-scale artificial intelligence ecosystem. Specifically for JD's AI layout path, she presented a comprehensive JD AI landscape map.

In this landscape, the foundation is "JD Cloud Smart Computing" infrastructure; above that is the large model matrix; the next layer consists of AI platforms and tools; and the top level connects to specific application scenarios and user products in retail, healthcare, logistics, and industrial sectors.

Around this layout, JD.com released multiple innovative achievements at this JDD conference. Among them, the new large model brand JoyAI covers full-size parameters from 3B to 750B, including voice and digital human large models. Through algorithmic innovation, it achieves 1.8x faster inference speed and 70% reduced training costs.

According to reports, JoyAI large model scored 76.3 on the Rrbench large model evaluation benchmark established jointly by Tsinghua University and Stanford University (as of 0924), with reasoning capabilities ranking among the top domestically and second globally.

Simultaneously, JD.com launched three AI products: Jingxi App, Tata, and JoyInside 2.0, deepening AI applications in retail, logistics, healthcare, and industrial scenarios, and showcasing multiple technical achievements including digital humans, intelligent agents, and coding platforms.

Industry observers believe that through strengthening the combination of technology and industry, JD's AI strategy has formed deep integration with cloud as the foundation, models as the core, and industrial scenarios as value outlets. JD's strategic direction emphasizing "experience" and "industry depth" not only helps drive AI from technological concepts to commercial effectiveness but also provides important reference for the entire industry to explore sustainable development paths.

**Accelerating AI Scenario-Based Implementation Applications**

From JD's AI landscape map, it's evident that the AI achievements JD showcased are not laboratory technologies gathering dust, but directly address very challenging problems in industrial implementation, making AI technology transition from usable to user-friendly while genuinely helping industries or consumers solve problems.

For example, in e-commerce material generation, Jingdiandian helps retail merchants improve material refinement iteration efficiency by over a thousandfold through industry-leading product consistency maintenance, fine design element control, and autonomous iteration capabilities driven by AB test data.

In supply chain management, JD.com applies large models with multi-agent collaboration, transforming point optimization into full-chain coordination, improving overall operational efficiency. Similar product logic is already reflected in JD's AI layout.

In the retail sector, JD.com developed the Oxygen intelligent architecture as the core of e-commerce innovation. Features like "AI Shopping" natural language recommendations and "TryOn" virtual fitting not only optimize consumer experience but also substantially improve merchants' operational efficiency.

In logistics scenarios, the Super Brain Large Model 2.0 achieves multimodal autonomous decision-making. The "Wolf Pack" series products have been deployed to over 500 warehouses globally, helping improve frontline operational efficiency by nearly 20%.

The healthcare business launched "Jingyi Qianxun 2.0" with reliable reasoning and full-modal capabilities, implementing comprehensive services from triage and consultation to home testing for AI hospitals.

In industrial scenarios, the JoyIndustrial large model, based on 57.1 million industrial product SKU data, provides intelligent services to over 10,000 enterprises.

While promoting industrial intelligence, JD.com is also reconstructing user-end AI experiences through three products, with all three centered on being conversational, capable, and warm.

The soon-to-launch Jingxi App serves as the next-generation lifestyle service portal, incorporating voice interaction for seamless experiences in shopping, dining, and hotel booking. The Tata digital human assistant, centered on the Almighty Doctor, supports intelligent Q&A, lifestyle services, and hardware integration. Users can also create intelligent agents and participate in community interaction. At the event, Xu Ran demonstrated the complete process of ordering takeout through voice activation of Tata's Almighty Doctor, from voice ordering, intelligent recommendations, and voice confirmation to payment completion, taking only seconds.

JoyInside, as a software-hardware integrated solution, has been integrated with over 30 robot brands, significantly improving interaction frequency and user experience of hardware devices through humanized interaction.

The market generally believes that JD's AI practice is problem-oriented, not pursuing flashy technological demonstrations but focusing on solving real pain points in industrial digitalization. Whether addressing uncontrollable generated content or fragmented system optimization, this technology forged from business frontlines can better feed back into industries and become JD's core competency in AI competition.

**Opening Scenarios and Data to Drive Industrial Digital Transformation**

Since this year, open source and openness have become key trends in the AI industry, with leading companies building ecosystems through technology sharing. JD.com is not limited to pure technology output but actively transitions to an industrial co-builder role, opening its core scenarios and data resources to promote deep integration between AI technology and the real economy.

At the JDD conference, JD.com announced it will gradually open multiple industrial scenarios including logistics and healthcare, as well as supply chain data, to large model and embodied intelligence enterprises. This covers high-value segments such as warehousing, sorting, delivery, pharmaceutical management, and online consultations. Leveraging JD Cloud infrastructure, these high-quality datasets will provide model training and testing support for the industry, helping enterprises overcome data bottlenecks.

In fact, the lack of high-quality data has become a major obstacle constraining AI implementation. Many enterprises struggle with effective model validation due to lack of real scenario data, preventing technology from scaling up applications. By opening data resources accumulated in complex supply chain scenarios, JD.com essentially provides industry-validated data samples, significantly lowering the threshold for AI technology implementation.

Beyond data opening, JD.com is accelerating ecosystem co-building through technology open-sourcing. In the smart hardware field, unlike the currently popular embodied intelligence, JD.com proposed the concept of "attached intelligence," opening software capabilities to hardware companies like robots, positioning itself as a ladder builder. Only with good interactive experiences will users truly pay, thereby enabling the robotics industry to develop.

At the technical capability level, JD JoyAI large model has achieved full-size coverage, multimodal support, and algorithmic innovation, with domestic computing power also achieving substantial progress. JD.com has thus upgraded three major AI infrastructure platforms.

The Digital Human Platform 4.0 industry-first launched brand spokesperson digital humans, reducing live streaming costs to one-tenth of human costs. JoyAgent 3.0 intelligent agent platform achieved 100% open source, with over 30,000 intelligent agents running internally. JoyCode 2.0 coding platform, combined with intelligent agent technology, shortens development cycles by 30%.

Industry observers believe that JD's open path essentially shares technology derived from its years of accumulated supply chain capabilities with the industry. Through opening scenarios, open-sourcing tools, and co-building ecosystems, converting effectively validated AI models into foundational industrial capabilities not only reduces innovation costs for industries but also accelerates digital transformation across all industrial segments.