JD.com Sets Three Million-Robot Target For Physical AI Logistics Push
JD.com is expanding physical AI across logistics with robot, autonomous vehicle and drone targets, backed by live warehouse deployments and higher R&D spending.
AI News reports that JD.com is turning physical AI into a larger logistics programme, pairing a target to buy 3 million robots over five years with 1 million self-driving delivery vehicles and 100,000 drones.
The plan, launched at JDDiscovery 2026 in Beijing, moves the company’s robotics work from isolated automation tools toward a broader operating system for warehouses, transport and delivery.
JD Logistics also introduced its industrial Wolf Robot series for warehousing, sorting, transport and last-mile tasks.
Some of the equipment is designed for specialised logistics environments.
The source describes cold-chain machines rated for minus 20 degrees Celsius, automated pharmacy dispatch, driverless delivery vehicles and aerial delivery systems.
The procurement target builds on automation already running inside the company’s network.
By June 30, JD Logistics had put its LangzuTech Goods-to-Person system into more than 30 Chinese warehouses and had also begun deployments in the UK and Germany.
Its wider footprint included more than 1,800 self-operated warehouses and more than 2,000 third-party cloud warehouses on the Open Warehouse Platform, together spanning over 36 million square metres.
Autonomous delivery is also operating beyond small pilots in parts of the network.
By the end of June, unmanned vehicles were in regular use in over 20 provinces in China.
JD’s drone network had more than 100 domestic routes, carrying parcels, meals, emergency medicine and relief supplies.
In Shenzhen, JD Logistics has added overnight autonomous routes so vehicles can keep running through the full day.
The software layer is JD’s Meta Brain system, which links physical equipment across warehousing, transportation and delivery.
The latest version has cut route-planning time for hundreds of millions of parcels from minutes to seconds.
The LangzuTech Packer robotic arm applies the same AI layer with multimodal sensing so it can identify parcels, grip them and place varied package shapes.
A first-quarter regulatory filing adds detail on how the Packer is being trained.
In simulation, the arm runs parallel reinforcement-learning routines that test placement order and loading patterns before they are used in live sorting work.
The goal is better sorting productivity and fuller use of carrier space.
After a second-quarter force-control upgrade for cage loading, Packer units were running continuously at several JD Logistics parks by June.
JD is also building computing and data capacity around the push.
JD Cloud plans to work with Moore Threads on a 100,000-GPU cluster for large-model training, inference and embodied-AI workloads, after earlier work on a 10,000-GPU cluster.
For training data, JD Cloud intends to gather above 10 million hours of video showing human activity in real settings over a two-year period.
The financial signal is visible but not fully itemised.
Research and development spending at JD Logistics reached RMB2.3 billion in the first half of 2026, a 23.7% increase from RMB1.9 billion a year earlier.
Depreciation and amortisation tied to property, equipment and intangible assets rose 18.7% to RMB2.6 billion, and purchases of property and equipment and investment properties reached RMB3.09 billion, though those figures cover the wider logistics business rather than the new physical AI programme alone.
A total cost for the five-year procurement plan and a network-wide return target are absent from the published plan.
Its disclosed base still gives the programme a large operating field: robot repair centres, live warehouses, autonomous routes, drone networks and about 700,000 delivery and logistics personnel across JD and its ecosystem.
