JD.com is positioning its next AI push around the physical world: data captured from real environments, models that interpret those environments, and robotics infrastructure that can put the resulting systems to work. The company says it will open its full-chain self-developed AI capabilities to global partners, linking a large retail and logistics business to an industrial strategy built around JoyAI models, the EgoLive dataset, and planned RoboBase sites.
Pandaily reported that JD.com founder Richard Liu said the company’s full-chain AI would be opened to global partners. The report describes programs spanning data, models, robots, and logistics. Separately, Futunn’s financial-results coverage reported that JD.com’s first-half 2026 research-and-development spending rose 53% year over year.
Those disclosures make the announcement more concrete than a general corporate pledge to use AI. JD.com is describing an operating stack with identifiable components. It says it has open-sourced EgoLive, a dataset based on first-person perspectives; it is building physical-AI infrastructure; and it has begun a RoboBase project in Guangzhou. The company also says it plans more than 80 RoboBase sites nationwide over the next five years. Those are JD.com’s plans and disclosures, not independently verified deployment outcomes.
JD.com’s JoyAI Strategy Connects Data, Models, and Robots
JD.com’s physical-AI strategy begins with its JoyAI model family. In the financial-results coverage, the company said its work covers language, speech, image, video, real-time interaction, world models, and embodied intelligence. That breadth is relevant because physical AI requires systems to handle more than text. A robot or automated system may need to interpret images, understand instructions, track objects, and act in a real setting.
The company’s description of a “full-chain” AI stack links those models to data and infrastructure. JD.com says it has open-sourced EgoLive and that more than 100 universities and research institutes have applied to use the dataset. That application count is a JD.com disclosure. It indicates interest in the project but does not show how many organizations are actively using the data or what results they have obtained from it.
EgoLive’s role is central to JD.com’s pitch because embodied systems need information about how people interact with objects and environments. A first-person dataset can capture actions from a human perspective. The source material does not establish that the dataset solves every training-data problem, but it shows JD.com treating data collection as a strategic asset rather than a narrow internal function.
EastFrontier has previously covered JD.com’s robot-ambulance service, which focused on maintenance and repair for robots. The current program is broader. JD.com is describing a structure in which data, models, hardware, logistics, and external partners can connect. That is a different ambition from deploying one robot service inside an existing logistics network.
RoboBase Plans Turn Corporate AI Into Industrial Infrastructure
The physical footprint of JD.com’s plan is the RoboBase program. JD.com says its first RoboBase project began construction in Guangzhou during the second quarter and that it intends to deploy more than 80 sites nationwide over five years. The report does not provide a confirmed construction budget, detailed site list, or operating schedule for all those bases. The appropriate reading is that JD.com has announced a national rollout plan, not that 80 locations are already operating.
RoboBase is significant because it implies an infrastructure model for robotics development and deployment. A physical-AI system needs places to collect data, test equipment, run simulations, validate behavior, and connect machines with business workflows. JD.com says its approach covers collection, storage, labeling, training, evaluation, simulation, and testing. That is an ambitious full-cycle description, and its value will depend on execution at individual sites.
The company’s choice to connect the plan to logistics gives it a practical starting point. JD.com already operates warehouses, delivery systems, and supply-chain services. Those environments can provide repeated tasks and real operational constraints. The sources reviewed here do not quantify how many robots will be trained through the program or how much commercial revenue it will generate. They show that JD.com intends to use its existing industrial settings as a foundation for physical AI.
This model fits a wider Chinese push to build embodied-AI capacity around real industrial clusters. EastFrontier has reported that Guangdong is developing embodied-AI industrial clusters. JD.com’s contribution is different because it is a company-led stack that begins with retail, logistics, and its own data and model programs.
Opening the Stack Will Test JD.com’s Partner Strategy
JD.com’s stated plan to open its AI capabilities to global partners is the commercial test. Opening a stack can mean several things: sharing models or data, offering tools, creating partner programs, or allowing external companies to build on a company’s infrastructure. The sources say JD.com intends to open its full-chain AI to partners, but they do not provide a complete public set of access terms, pricing, or partner names.
The company’s R&D figure shows the scale of its investment direction. A 53% year-over-year increase in first-half R&D spending is a specific signal that JD.com is allocating more resources to technology. It does not isolate how much of that spending went to physical AI, EgoLive, RoboBase, or each JoyAI model. The number is best understood as context for the company’s broad AI investment, not as a budget for one project.
JD.com’s strategy will now be judged by measurable evidence: whether partners use its tools, whether the planned RoboBase sites are built, how the EgoLive dataset is adopted, and whether the company’s AI systems improve operating results in physical environments. The announcement supplies a map of the intended stack. It does not yet show the commercial results of opening that stack.
For China’s AI sector, JD.com’s announcement reinforces a broader point. Physical AI is becoming a contest over data infrastructure, model capabilities, test environments, and distribution channels, not only robot hardware. JD.com is trying to combine all four using assets it already has. Its next disclosures will show whether that combination becomes a partner ecosystem or remains primarily an internal corporate program.