JD.com has put a vast procurement target behind its effort to connect artificial intelligence with physical logistics. Under a new Physical AI Acceleration Plan, the company reiterated that it intends to procure 3 million robots, 1 million autonomous vehicles and 100,000 delivery drones over five years.
The plan, announced at JDDiscovery 2026 in Beijing, reaches from warehouse handling to transport and last-mile delivery. It also leaves crucial financial questions unanswered: JD disclosed neither the expected total cost nor a network-wide return-on-investment target.
A Robot Fleet Built Around Logistics Work
JD Logistics introduced a Wolf Robot series for warehousing, sorting, transportation and delivery. The systems include equipment designed for temperatures as low as minus 20 degrees Celsius, automated pharmacy dispatch, autonomous delivery vehicles and drones. That range shows a strategy based on specialized operational roles rather than a single universal machine.
The company already has a substantial automation base. As of June 30, its LangzuTech goods-to-person warehouse system operated in more than 30 Chinese warehouses, with deployments also launched in the UK and Germany. JD Logistics said its wider network comprised more than 1,800 self-operated warehouses and over 2,000 third-party cloud warehouses, covering more than 36 million square metres.
Thousands of unmanned vehicles were in regular operation across more than 20 Chinese provinces by the end of June. More than 100 domestic drone routes served parcel and food delivery, emergency-medicine transport and disaster relief. In Zizhong, Sichuan, a drone network launched in June covers 78 administrative villages; JD said some mountain-village deliveries can take as little as seven minutes.
Meta Brain Connects Planning With Machines
The hardware is being coordinated through Meta Brain, JD’s AI system for warehouses, transport and delivery. The company claims Meta Brain 3.0 can calculate optimal routes for hundreds of millions of parcels in seconds, where the task previously took minutes. The source does not provide an independent benchmark or describe the operating conditions behind that comparison.
Meta Brain also powers the LangzuTech Packer robotic arm. Using multimodal sensor data, the arm tracks, grasps and places parcels of different shapes. JD said the system applies reinforcement learning in simulated environments to improve parcel-placement sequences and loading layouts. Force-control upgrades introduced in the second quarter were intended to make cage loading more precise, and JD reported that the Packer was running around the clock at multiple logistics parks by June.
Computing capacity is another part of the plan. JD Cloud intends to work with Chinese chipmaker Moore Threads on a 100,000-GPU cluster for model training, inference and embodied-AI workloads. The companies previously collaborated on a 10,000-GPU cluster, but they have not disclosed which GPU models the larger system would use. JD Cloud also plans to collect more than 10 million hours of video showing real-world human activity over the next two years for embodied-AI training.
Scale Brings Costs and Workforce Questions
JD Logistics spent RMB2.3 billion on research and development in the first half of 2026, up 23.7% from RMB1.9 billion a year earlier. The company attributed the rise to technology and innovation but did not isolate spending on AI or robotics. Purchases of property, equipment and investment properties reached RMB3.09 billion, while depreciation and amortization increased 18.7% to RMB2.6 billion. Those figures cover the broader logistics business, not only the new plan.
The workforce implications are similarly mixed. JD founder Richard Liu has said robots will eventually assume parcel-delivery work now performed by couriers. At the same time, JD has agreements with around 120 educational institutions to retrain workers for roles including robot repair and maintenance. It operates eight robot repair centres in China and expects expanded after-sales capabilities to support more than 100,000 robotics service engineer jobs.
The procurement targets signal extraordinary ambition, but deployment numbers are not the same as measured productivity or financial returns. JD has shown that many components already operate at commercial scale. The unanswered test is whether millions more machines can be integrated reliably, economically and with credible paths for the people whose work they change.
