On 9 September 2026, JD.com announced its Physical AI Acceleration Plan at the JD Global Technology Explorer Conference in Beijing, setting six "world's largest" goals across embodied AI data collection, robotics industry bases, service networks, logistics robot fleets, component supply, and retail channels. The plan was outlined by Cao Peng, Chairman of JD.com's Technology Committee and President of JD Cloud, under the theme "JoyAI: Leaping into the Physical World."

Confirmed

  • Embodied data centre: JD Cloud aims to gather more than 10 million hours of real-world human activity video over the next two years, combining internet, simulated, and robot data into an integrated "understanding-generation-action" model.
  • RoboBase hubs: JINGDONG Property plans to establish more than 80 robotics industry bases across China over five years, integrating demonstration, delivery, R&D, pilot assembly, data collection, manufacturing, maintenance, and iteration.
  • Service network: JINGDONG Logistics currently operates eight repair centres in China and plans to extend after-sales capabilities, targeting more than 100,000 robotics service engineer jobs over five years.
  • Logistics robot fleet: JINGDONG Logistics aims to build the world's largest-scale application of embodied robots and advance autonomous operations across the entire logistics process.
  • Component supply: JINGDONG Industrials launched an alliance connecting robot manufacturers with component suppliers, targeting the world's largest robotics component supply network.
  • Retail channel: JINGDONG Retail commits RMB 10 billion by 2028 to help 100 robotics brands each achieve RMB 1 billion in independent sales and reach 10 million users.
  • Procurement targets: Over the next five years, JD.com plans to procure more than 3 million robots, 1 million autonomous vehicles, and 100,000 drones, according to conference disclosures reported by BigGo Finance.
  • Compute infrastructure: JD Cloud has partnered with Moore Threads to build a domestically produced 10,000-GPU cluster, with plans for a 100,000-card cluster.
  • R&D investment: System-wide R&D spending grew 53.2% year-over-year in the first half of 2026, marking three consecutive quarters of accelerating investment.

Unknown

  • Independent verification of the 10 million-hour data collection timeline and the operational readiness of the 80 RoboBase hubs within five years.
  • Whether the 3 million robot procurement target reflects firm orders or aspirational demand signals across the supply chain.
  • Cost per robot, deployment schedules, and the fraction of logistics workflows that can be autonomously operated within the plan period.
  • Revenue contribution and margin profile of the RMB 10 billion retail commitment by 2028.
  • Technical maturity of the JoyAI-Echo WM world model and its integration latency with Meta Brain 3.0 in production logistics.

Our take

JD.com is using its own logistics network as both the primary training ground and first commercial deployment target for embodied AI at million-unit scale. The data flywheel — 10 million hours of real warehouse video feeding models that then control robots in those same warehouses — could close the sim-to-real gap faster than lab-only approaches. The 100,000 service-engineer job target also signals JD treats after-sales as a structural cost, not an afterthought.

Sources