JD.COM Targets 3 Million Robots in Physical AI Logistics Push

JD.COM plans to procure three million robots over five years as it expands physical AI across logistics, retail and services. The headline scale matters, but the deeper strategy is the combination of machines, real-world training data, computing infrastructure and maintenance hubs—an attempt to turn JD.COM’s operating network into both a customer and development environment for embodied AI.

The Three Million Figure Is a Procurement Target

AI News reported that JD.COM is reiterating a five-year target to procure three million robots, one million autonomous vehicles and 100,000 delivery drones. These are future purchasing goals, not the number of machines currently operating. The targets were presented alongside a Physical AI Acceleration Plan at JDDiscovery in Beijing.

JD.COM’s official event summary frames the plan across the robotics lifecycle, from data and development to deployment and servicing. The scale could give suppliers long-term demand while letting JD.COM standardize equipment across warehouses, delivery networks and customer-facing services. Comparable deployment questions appear in Business 2.0’s coverage of large-scale humanoid infrastructure.

Wolf Robots Target Multiple Logistics Tasks

JD Logistics introduced its industrial Wolf Robot series for warehousing, sorting, transport and delivery. Reported examples include machines built to work at temperatures as low as minus 20°C, automated pharmacy-dispatch equipment, autonomous delivery vehicles and drones. This is less a bet on one universal humanoid than on specialized systems designed around repeatable operational tasks.

JD.COM already has experience with unmanned warehouses and delivery machines, including earlier autonomous delivery trials. The advantage is access to high-volume routes, facilities and demand patterns where robots can be measured against human cost, reliability and service targets. Business 2.0 has seen similar task-specific logic in public-sector robotic process deployment and autonomous logistics partnerships.

Data and Compute Are Central to the Plan

JD Cloud aims to collect more than 10 million hours of high-quality video from real human activity over two years for embodied-AI training. The company also outlined a 100,000-card computing cluster with Chinese chipmaker Moore Threads, according to Caixin Global. Together, those assets are intended to improve models that perceive environments, plan movement and control physical systems.

The strategic asset may be the operating data rather than the machines alone. Warehouse picking, cold-chain handling and last-mile delivery generate repeated examples of edge cases that simulations may miss. Yet collecting video at that scale requires clear governance for workers, customers and commercially sensitive environments. Independent coverage from DigiTimes and Digital Today also emphasizes the link between deployment and training data.

RoboBase Hubs Could Reduce Deployment Friction

JINGDONG Property plans to establish more than 80 RoboBase facilities across China over five years. These hubs are intended to combine demonstration, delivery, research, pilot assembly, data collection, manufacturing, maintenance and product iteration. JD.COM previously outlined this lifecycle strategy at the World Robot Conference, alongside RMB 10 billion in resources through 2028.

Maintenance is often the gap between a successful robot demonstration and profitable fleet operation. Regional facilities could shorten repair times and help vendors refine machines against real failures. The model has relevance to emerging robotics companies such as those covered in quadruped delivery services and commercial service-robot expansion.

Execution Will Determine Whether Scale Creates an Advantage

Industry reporting from CEP Research and a syndicated JDDiscovery account confirms the programme’s breadth. JD.COM’s JoyInside platform separately expects more than 10 million devices and robots to connect to its ecosystem by the end of 2026; that figure should not be confused with owned logistics robots.

The business test is whether automation lowers fulfilment costs and improves reliability after hardware, integration, maintenance and replacement expenses. A three-million-unit procurement target creates leverage with suppliers, but also exposes JD.COM to technology obsolescence and uneven performance. Physical AI becomes defensible when deployed machines produce better data, which improves models and future machines. JD.COM now has to prove that this cycle works across operations, not just in controlled demonstrations.