JD Unveils 100,000-GPU Domestic Compute Plan and JoyAI World Model at Beijing Event

The event, themed “JoyAI · Leap into the Physical World,” was held as JD described its goal of building the “world’s largest physical-world operations center.” The company said it is combining compute, data and model capabilities with its supply chain and applying them in retail, logistics, technology, health, industrial, property development, finance, delivery and housekeeping scenarios. It also said it will open those capabilities to industry and society.

JD Group Technology Committee Chairman and JD Cloud President Cao Peng said JD’s AI was not born in papers but was ground out order by order in production lines, warehouses, delivery stations and the full supply chain process. He said JD’s supply chain is the best training ground and first stop for AI to move into the physical world.

JD’s “super AI supply chain” is based on six modules—cloud, data, model, device, scenario and chain—and integrates compute, data and models with intelligent terminals and real scenarios. According to the company, scenarios generate data, data trains models, models drive terminals, and task feedback pushes AI to continue evolving, forming an “intelligence flywheel.” The supply chain connects research and development, manufacturing, sales, delivery and repair, lowering the threshold from validation to productization and large-scale application, forming an “industry flywheel.”

In compute, JD Cloud has built a domestic 10,000-GPU cluster with partners including Moore Threads and plans to build a 100,000-GPU cluster. JD described it as an independently controllable ultra-large-scale compute foundation for AI cloud in China, providing stable support for large-model training and real-world applications. In data, JD Cloud said its 10-million-hour human data collection effort, described as the largest of its kind, is progressing. It has built embodied data infrastructure covering collection, storage, annotation, training, evaluation, simulation and testing, turning human operations in real scenarios into data fuel for embodied model training. The first open-source dataset, EgoLive, is available, and more than 100 universities and research institutions in over eight countries have applied to use it.

In models, JD has formed a JoyAI foundation model matrix covering multimodal models, world models and embodied models. At JDD, it released JoyAI-Echo WM, a real-time interactive world model that scored 81.6 and ranked first in WBench Navigation, a public benchmark for interactive world models, reaching what JD called a top industry level. JD also released vertical models including JD Logistics Super Brain 3.0, JD Industrial JoyIndustrial2.0 and JD Health Jingyi Qianxun 3.0. JD said Super Brain 3.0 is the logistics industry’s first industrial-grade large-model application integration covering the full chain of warehousing, transportation and delivery, with unified AI decision-making, collaboration and execution. It said the model combines decision AI, process AI and physical AI, compresses end-to-end optimal route planning for billions of packages from minutes to seconds, and achieves a 96.7% success rate for embodied intelligence models in multi-task logistics scenarios. JoyIndustrial2.0 can complete mechanical design from simple language instructions and quickly print finished products in 3D, according to JD.

JD launched a “physical AI acceleration plan” for the robot industry, saying it will support the full chain and full lifecycle of robot training, research and development, manufacturing, sales, application and service. The company aims to achieve six “world firsts” within several years. It said JD Cloud will build the world’s largest embodied intelligence data collection center, gathering more than 10 million hours of real-scenario human video data in two years to address a data shortage in embodied intelligence. In the next five years, JD plans to deploy more than 80 RoboBase robot industry bases across China, creating what it calls the world’s largest robot industry base with integrated services for display and delivery, repair and maintenance, R&D and design, pilot assembly, data collection, manufacturing and iterative upgrading.

JD Industrial launched a “Robot Component Industry Development Alliance,” aiming in the next three years to help 100 robot body manufacturers reduce costs across all categories and 100 component suppliers double their performance, making JD the world’s largest robot component supplier. JD Retail plans to invest 10 billion yuan in resources before 2028, help 100 brands achieve independent sales of more than 1 billion yuan each, and push the robot industry into one million terminal scenarios covering tens of millions of users, becoming the world’s largest robot retail channel. JD Logistics aims to build the world’s largest embodied robot application legion, planning to purchase 3 million robots, 1 million unmanned vehicles and 100,000 drones in five years to advance unmanned logistics. Its Super Brain model handles intelligent decision-making and collaborative scheduling in complex scenarios, while its Wolf Clan robot legion has deployed “nine wolves” in warehousing and delivery, including the recently added Mother Wolf, Warehouse Wolf, Smart Wolf low-temperature version and the upgraded embodied dexterous arm Alien Wolf.

JD Logistics also plans to build the world’s largest robot maintenance service network. It has built eight major maintenance centers in China and plans for its robot after-sales service capacity to cover more than 100 countries in the next five years, creating more than 100,000 jobs for robot after-sales service engineers to support large-scale deployment and continuous operation.

In retail, JD launched a new version of its app with an AI shopping assistant called Dongdong. JD said the assistant combines AI understanding of user needs, moving from helping users find a product to understanding what problem they need solved and dispatching goods and services to meet that need. In health, Jingyi Qianxun 3.0 and JD Health’s first fully intelligent renowned-doctor agent allow users to input symptoms, medical history and test reports; the agent asks about the condition, reads indicators, gives medical direction and reminds users about follow-up medication and visits. Through AI, hardware and services, JD connects glucose meters and blood pressure monitors to personal health records and links online consultation, home rapid testing and nurse home visits. In the home, JD’s embodied intelligence JoyInside implants an AI brain into hardware, pushing smart devices from single-product responses to whole-house proactive service. It has technology cooperation with more than 200 brands covering appliances, home, toys and robots, among other categories.

JD contrasted its approach with companies that first develop technology and then look for scenarios, describing that path as “making shoes before measuring feet.” JD said it grows technology from scenarios, or “measuring feet before making shoes,” first using AI to maximize its own supply chain efficiency and then opening refined capabilities to industry partners and customers. Under its “world’s largest physical-world operations center” goal, JD said it will act as a connector and operator with industry partners to push AI into the physical world and contribute to the real economy. QbitAI said the material was provided by JD and that the views belong to the original author.

Editor's Summary JD.com used its 2026 JDD event to outline a physical AI strategy centered on a planned 100,000-GPU domestic compute cluster, a 10-million-hour embodied data effort, the JoyAI-Echo WM world model and vertical models for logistics, industry and health. It also launched a robot-focused physical AI acceleration plan that includes targets for data collection, robot industry bases, component supply, retail channels, logistics deployment and global maintenance services. The announcements show JD positioning its supply chain as both a training ground for AI and a channel for bringing embodied intelligence into warehouses, deliveries, stores and homes.