JD.com is pursuing a robotics strategy that looks less like a bet on one humanoid robot and more like an attempt to become the industry’s commercialization layer. By combining demand aggregation, shared components, real operating environments, data collection, repairs, and lifecycle services, the company wants to help robots move from laboratory demonstrations into repeatable use across retail, logistics, health, industry, homes, and public-facing services. 56

The scale of the plan is substantial: JD says it will commit RMB 10 billion in robotics resources by 2028, support 100 robot brands in reaching more than RMB 1 billion in independent sales each, and help bring robots into one million endpoint scenarios. Those are announced objectives, not independently verified outcomes. 56

The central idea: robotics needs infrastructure, not just better machines

A robot can work in a controlled demonstration and still fail as a commercial product. Real deployment requires reliable components, buyers, installation, training, software updates, spare parts, repairs, recycling, and enough operating data to improve performance.

JD’s premise is that robotics companies are often specialized: one may build the hardware, another the control system, and another the model. Few can independently create a large sales channel, a multi-industry deployment network, standardized procurement, global maintenance, and a large real-world dataset. JD is attempting to spread those fixed costs across many brands and use cases. 4710

1. Aggregating demand across consumers and enterprises

JD already says it works with more than 200 robot brands and core-component suppliers through its self-operated business. Its coverage includes consumer and household robots as well as applications in retail, health, industry, and logistics. The company has also organized robots on its platform into 16 categories based on user needs. 146

That categorization gives JD more than a merchandising function. It can help clarify who a product is for, expose brands to consumer and enterprise demand, and return signals from searches, consultations, purchases, and reviews to product teams. One report says JD’s platform activity reflects demand from about 700 million consumers and 8 million enterprise customers. 4

For smaller robotics companies, this could reduce the cost of finding customers and testing product-market fit. For JD, it creates a broader view of which tasks buyers actually want automated—rather than relying only on what manufacturers choose to build.

2. Standardizing components, starting with batteries

JD is using batteries as an example of how pooled demand could address fragmented robotics supply chains. The company is working with more than 20 robotics partners on common battery dimensions, interfaces, communication protocols, and safety standards. The model also involves centralized purchasing, inventory preparation, and staged delivery. 51517

The intended benefits are practical:

• lower purchasing and component costs through greater volume;
• fewer bespoke designs and less duplicated engineering;
• more predictable spare-parts availability;
• simpler replacement and maintenance; and
• shorter product-development and service cycles.

Battery standardization does not solve every interoperability problem, and the sources do not establish the size of any realized savings. But it illustrates JD’s broader approach: identify shared requirements across competing robot makers, then use purchasing and logistics scale to make those requirements easier to support.

3. Putting robots into recurring real-world work

JD is opening parts of its commercial footprint to robot applications rather than treating robotics as a stage demonstration. Reported examples include mall guidance and shopping assistance, shelf replenishment, coffee preparation, warehouse handling and sorting, autonomous delivery, health-pharmacy packaging, and elder-care support. Its service activities also include diagnostics, repairs, battery replacement, maintenance, appraisal, and recycling. 16182022

These environments matter because they expose robots to changing conditions: different layouts, objects, customers, workflows, and failure modes. A robot that operates repeatedly in a store or warehouse can generate more useful feedback than one that performs a short scripted demonstration.

JD describes these business environments as a form of ongoing practical training. The strategic value is two-sided: robots receive operating experience, while JD learns which tasks are economically useful, where human intervention is still needed, and what buyers expect from a deployed system. 1517

4. Building a repair and lifecycle network

After-sales support may be one of the least glamorous—and most important—parts of commercial robotics. A machine that cannot be diagnosed, repaired, upgraded, or safely recycled can become an expensive liability even if its core technology works.

JD says it has established eight domestic robot repair centers and introduced a “120” service target: a one-minute response, an on-site visit within two hours, and same-day repair. Its repair capabilities have also expanded into Europe, the Middle East, and North America. 415

The company plans to build 80 RoboBase facilities over five years and develop after-sales coverage in more than 100 countries and regions. These bases are intended to combine display and delivery, maintenance, research and design, pilot assembly, data collection, manufacturing, and product upgrades. 57

In theory, this gives robot makers access to services that would be costly to build independently. It also gives customers a clearer answer to a basic adoption question: what happens when the robot breaks?

5. Creating a real-world data flywheel

Embodied AI depends on data that connects perception to action: what the robot sees, how it moves, what it touches, and how it responds to unexpected conditions. JD Cloud says it plans to collect more than 10 million hours of real-world scenario data within two years and use that foundation to develop an internally controlled AI-model stack. 5611

Earlier reporting describes a data-collection pipeline spanning collection, annotation, training, and validation, with scenarios including logistics, industrial manufacturing, health care, household services, and urban operations. JD has also discussed collecting both human real-world video and robot-body data. 2426

The proposed flywheel is straightforward:

1. robots operate in stores, warehouses, homes, and other working environments;
2. those operations produce examples of successful actions, failures, and edge cases;
3. the data helps improve models, interfaces, and task planning; and
4. improved systems can be deployed back into the same network for further testing.

The caveat is that a large volume of video is not automatically a high-quality robotics dataset. Data must be relevant, well-labeled, legally usable, and connected to the actions and outcomes that models need to learn. JD’s 10-million-hour figure is a target, so its eventual technical value cannot yet be assumed.

Why JD believes the shared-network model could work

Economies of scale

Pooling demand can improve bargaining power and make inventory easier to plan. Shared specifications may also reduce duplicated design work and spare-parts complexity, particularly for common components such as batteries. 1015

Reliability as an operating capability

Robotics reliability is not only a question of whether a machine can complete a task. It also depends on uptime, diagnostics, replacement parts, field technicians, software updates, and end-of-life handling. A lifecycle network can make those requirements part of the product rather than an afterthought. 78

Faster market feedback

JD’s commerce and operating businesses can connect product discovery, customer questions, transactions, reviews, deployments, and service incidents. That feedback can help brands identify promising use cases and prioritize technical improvements faster than isolated pilots allow. 47

A way around collective-action problems

Standardization and shared data create benefits for an entire ecosystem, but no single small manufacturer may have enough incentive or resources to build them alone. JD is trying to occupy that gap by acting as an organizer, buyer, operator, service provider, and data platform at the same time.

What remains unproven

The strategy is ambitious, but its most important claims are forward-looking. The 80 RoboBases, coverage in more than 100 countries, 10 million hours of data, RMB 10 billion commitment, 100 high-selling brands, one million endpoint scenarios, and more than 100,000 projected after-sales jobs are announced plans or targets. 567

The available reporting does support several existing building blocks: JD’s relationships with more than 200 robot brands, eight domestic repair centers, standardized battery work with more than 20 partners, and reported deployments across retail, logistics, maintenance, health, and household-service settings. 41518

But those building blocks do not yet prove that the complete network will deliver lower costs, higher reliability, or mass adoption at global scale. Execution will depend on whether competing brands accept common standards, whether real-world data translates into better robot performance, whether service economics work across markets, and whether customers continue paying for robots after initial pilots.

Bottom line

JD.com is betting that the bottleneck for embodied AI is not simply intelligence inside the robot. It is the surrounding system that makes robots affordable to buy, practical to deploy, easy to repair, and capable of learning from real work.

Its strategy therefore combines four forms of scale: demand from a large commerce platform, procurement and logistics for shared components, operating environments for deployment and data, and a service network for the robot’s full lifecycle. If JD can execute, it could help turn fragmented robotics products into a more supportable industry. For now, however, the strongest evidence is that JD is assembling the infrastructure—not that the promised global robotics network has already been achieved.

JD.com is pursuing a robotics strategy that looks less like a bet on one humanoid robot and more like an attempt to become the industry’s commercialization layer. By combining demand aggregation, shared components, real operating environments, data collection, repairs, and lifecycle services, the company wants to help robots move from laboratory demonstrations into repeatable use across retail, logistics, health, industry, homes, and public-facing services. 56

The scale of the plan is substantial: JD says it will commit RMB 10 billion in robotics resources by 2028, support 100 robot brands in reaching more than RMB 1 billion in independent sales each, and help bring robots into one million endpoint scenarios. Those are announced objectives, not independently verified outcomes. 56

The central idea: robotics needs infrastructure, not just better machines

A robot can work in a controlled demonstration and still fail as a commercial product. Real deployment requires reliable components, buyers, installation, training, software updates, spare parts, repairs, recycling, and enough operating data to improve performance.

JD’s premise is that robotics companies are often specialized: one may build the hardware, another the control system, and another the model. Few can independently create a large sales channel, a multi-industry deployment network, standardized procurement, global maintenance, and a large real-world dataset. JD is attempting to spread those fixed costs across many brands and use cases. 4710

1. Aggregating demand across consumers and enterprises

JD already says it works with more than 200 robot brands and core-component suppliers through its self-operated business. Its coverage includes consumer and household robots as well as applications in retail, health, industry, and logistics. The company has also organized robots on its platform into 16 categories based on user needs. 146

That categorization gives JD more than a merchandising function. It can help clarify who a product is for, expose brands to consumer and enterprise demand, and return signals from searches, consultations, purchases, and reviews to product teams. One report says JD’s platform activity reflects demand from about 700 million consumers and 8 million enterprise customers. 4

For smaller robotics companies, this could reduce the cost of finding customers and testing product-market fit. For JD, it creates a broader view of which tasks buyers actually want automated—rather than relying only on what manufacturers choose to build.

2. Standardizing components, starting with batteries

JD is using batteries as an example of how pooled demand could address fragmented robotics supply chains. The company is working with more than 20 robotics partners on common battery dimensions, interfaces, communication protocols, and safety standards. The model also involves centralized purchasing, inventory preparation, and staged delivery. 51517

The intended benefits are practical:

• lower purchasing and component costs through greater volume;
• fewer bespoke designs and less duplicated engineering;
• more predictable spare-parts availability;
• simpler replacement and maintenance; and
• shorter product-development and service cycles.

Battery standardization does not solve every interoperability problem, and the sources do not establish the size of any realized savings. But it illustrates JD’s broader approach: identify shared requirements across competing robot makers, then use purchasing and logistics scale to make those requirements easier to support.

3. Putting robots into recurring real-world work

JD is opening parts of its commercial footprint to robot applications rather than treating robotics as a stage demonstration. Reported examples include mall guidance and shopping assistance, shelf replenishment, coffee preparation, warehouse handling and sorting, autonomous delivery, health-pharmacy packaging, and elder-care support. Its service activities also include diagnostics, repairs, battery replacement, maintenance, appraisal, and recycling. 16182022

These environments matter because they expose robots to changing conditions: different layouts, objects, customers, workflows, and failure modes. A robot that operates repeatedly in a store or warehouse can generate more useful feedback than one that performs a short scripted demonstration.

JD describes these business environments as a form of ongoing practical training. The strategic value is two-sided: robots receive operating experience, while JD learns which tasks are economically useful, where human intervention is still needed, and what buyers expect from a deployed system. 1517

4. Building a repair and lifecycle network

After-sales support may be one of the least glamorous—and most important—parts of commercial robotics. A machine that cannot be diagnosed, repaired, upgraded, or safely recycled can become an expensive liability even if its core technology works.

JD says it has established eight domestic robot repair centers and introduced a “120” service target: a one-minute response, an on-site visit within two hours, and same-day repair. Its repair capabilities have also expanded into Europe, the Middle East, and North America. 415

The company plans to build 80 RoboBase facilities over five years and develop after-sales coverage in more than 100 countries and regions. These bases are intended to combine display and delivery, maintenance, research and design, pilot assembly, data collection, manufacturing, and product upgrades. 57

In theory, this gives robot makers access to services that would be costly to build independently. It also gives customers a clearer answer to a basic adoption question: what happens when the robot breaks?

5. Creating a real-world data flywheel

Embodied AI depends on data that connects perception to action: what the robot sees, how it moves, what it touches, and how it responds to unexpected conditions. JD Cloud says it plans to collect more than 10 million hours of real-world scenario data within two years and use that foundation to develop an internally controlled AI-model stack. 5611

Earlier reporting describes a data-collection pipeline spanning collection, annotation, training, and validation, with scenarios including logistics, industrial manufacturing, health care, household services, and urban operations. JD has also discussed collecting both human real-world video and robot-body data. 2426

The proposed flywheel is straightforward:

1. robots operate in stores, warehouses, homes, and other working environments;
2. those operations produce examples of successful actions, failures, and edge cases;
3. the data helps improve models, interfaces, and task planning; and
4. improved systems can be deployed back into the same network for further testing.

The caveat is that a large volume of video is not automatically a high-quality robotics dataset. Data must be relevant, well-labeled, legally usable, and connected to the actions and outcomes that models need to learn. JD’s 10-million-hour figure is a target, so its eventual technical value cannot yet be assumed.

Why JD believes the shared-network model could work

Economies of scale

Pooling demand can improve bargaining power and make inventory easier to plan. Shared specifications may also reduce duplicated design work and spare-parts complexity, particularly for common components such as batteries. 1015

Reliability as an operating capability

Robotics reliability is not only a question of whether a machine can complete a task. It also depends on uptime, diagnostics, replacement parts, field technicians, software updates, and end-of-life handling. A lifecycle network can make those requirements part of the product rather than an afterthought. 78

Faster market feedback

JD’s commerce and operating businesses can connect product discovery, customer questions, transactions, reviews, deployments, and service incidents. That feedback can help brands identify promising use cases and prioritize technical improvements faster than isolated pilots allow. 47

A way around collective-action problems

Standardization and shared data create benefits for an entire ecosystem, but no single small manufacturer may have enough incentive or resources to build them alone. JD is trying to occupy that gap by acting as an organizer, buyer, operator, service provider, and data platform at the same time.

What remains unproven

The strategy is ambitious, but its most important claims are forward-looking. The 80 RoboBases, coverage in more than 100 countries, 10 million hours of data, RMB 10 billion commitment, 100 high-selling brands, one million endpoint scenarios, and more than 100,000 projected after-sales jobs are announced plans or targets. 567

The available reporting does support several existing building blocks: JD’s relationships with more than 200 robot brands, eight domestic repair centers, standardized battery work with more than 20 partners, and reported deployments across retail, logistics, maintenance, health, and household-service settings. 41518

But those building blocks do not yet prove that the complete network will deliver lower costs, higher reliability, or mass adoption at global scale. Execution will depend on whether competing brands accept common standards, whether real-world data translates into better robot performance, whether service economics work across markets, and whether customers continue paying for robots after initial pilots.

Bottom line

JD.com is betting that the bottleneck for embodied AI is not simply intelligence inside the robot. It is the surrounding system that makes robots affordable to buy, practical to deploy, easy to repair, and capable of learning from real work.

Its strategy therefore combines four forms of scale: demand from a large commerce platform, procurement and logistics for shared components, operating environments for deployment and data, and a service network for the robot’s full lifecycle. If JD can execute, it could help turn fragmented robotics products into a more supportable industry. For now, however, the strongest evidence is that JD is assembling the infrastructure—not that the promised global robotics network has already been achieved.