Infrastructure creates possibilities. That was one of Jeff Bezos’ key convictions in Amazon’s early years. Amazon Web Services, or AWS, grew out of a similar realization: many companies did not lack ideas; they lacked the ability to build the underlying infrastructure themselves.
The embodied intelligence industry now faces a similar problem. Its technological capabilities are advancing rapidly, but integrating a robot into the physical world requires far more than capable hardware and models.
At the recent World Robot Conference, running, dancing, and backflips were no longer novelties. Robots were also becoming more adept at everyday tasks such as handing over water and grasping objects. Over the past few years, advances in humanoid robots’ motion control, interaction, and task execution have become increasingly visible.
But beyond staged demonstrations lies a world far more complicated than the exhibition floor: the overlapping noise and crowds of a shopping mall, the strict operating cadence of a warehouse, the endless variety of objects inside a home, and the long product lifecycle stretching from delivery and configuration to repair and recycling.
A stage can demonstrate the outer limits of capability. Industry tests whether those capabilities can work reliably over time. Between the two lie wide gaps in supply chains, data, delivery, operations and maintenance, and service systems.
That is one way to understand JD.com’s latest push into robotics.
Rather than mounting a technology showcase around a single brand, JD.com presented what amounted to a broader vision of a future city. It included robot training systems spanning data and models, component procurement, and repair services; industrial settings covering manufacturing, logistics, retail, and home services; and visions of daily life that mixed cyberpunk aesthetics with ordinary human needs, including exoskeletons that could help older people climb mountains and small humanoid robots designed to care for and accompany children.
When will this vision become reality: in ten, 30, or 50 years? No one knows. Technological maturity is only part of the answer. Just as important is when the industrial capabilities needed to bring robots into the real world at scale will mature. JD.com’s goal is to shorten that distance.
Once robots leave the laboratory and deployments expand from one unit to 1,000 or 10,000, the nature of the problem changes. Data, model training, component procurement, sales, fulfillment, and after-sales service are scattered across the industry value chain. If every company continues to build all of these capabilities independently, economies of scale will be difficult to achieve.
The key to moving robots from something companies can build to something that can operate at scale is organizing those fragmented capabilities into reusable industry infrastructure.
Scaling begins with infrastructure
“Two years ago, people were asking whether this sector could work at all. Now they care more about how to make it work well,” the head of JD.com’s intelligent robotics business told 36Kr, summarizing the shift in industry sentiment.
The market has already produced some encouraging signals. According to JD.com, sales of humanoid robots on its platform during this year’s 618 promotional season increased more than tenfold from a year earlier.
For a new category with low market penetration, surging sales are more a sign that market attention has arrived than proof of sustained commercial demand. Whether product value, user experience, and service systems can hold up over the long term remains to be tested.
The business head was reluctant to call the current moment a commercialization inflection point. A more accurate description, he said, is that “scaling has only just reached the starting line.”
But a starting line does not mean every type of robot is beginning from the same position. Different categories vary significantly in technological maturity, application scenarios, and commercialization progress. Based on JD.com’s platform data, he believes robot commercialization can be divided into several layers.
Consumer robots for education and family companionship have already entered homes and are beginning to face tests of price, interaction quality, and after-sales service.
Delivery robots, cleaning robots, and some industrial robots can already perform specialized tasks, but customization, deployment, and maintenance costs still weigh on returns on investment.
General-purpose robots capable of operating in open household environments and handling long-horizon tasks remain constrained by model capabilities, data quality, and coordination among cognition, motion control, and physical hardware.
Yet regardless of their technological maturity, these products encounter the same problem once they enter the market: the industrial support layer between a prototype, a commercial product, and a reliable service remains thin.
That gap was less visible in the past. During the technology validation stage, robotics projects were relatively limited in scale, and founders and engineers stationed on-site could solve many problems themselves. As orders increase, however, costs that were once hidden begin to surface.
Put another way, once an industry begins to scale, problems have to be solved by systems rather than individuals.
Compared with mature consumer electronics, robotics faces complexity at both the production and usage ends of the chain.
Upstream, component specifications are not yet fully standardized, procurement volumes remain limited, and mass production still faces challenges in cost and supply chain efficiency.
Downstream, once a robot arrives at a customer site, it often requires site surveys, adaptation, and ongoing maintenance. Completing the transaction is merely the beginning of the service cycle.
The problem is that robotics remains relatively low-volume even though it already requires an asset-heavy, service-intensive network. These capabilities share a similar cost structure: a network must be built first, and only sufficiently high usage density can spread the cost.
A repair center still needs premises, engineers, testing equipment, and spare parts even if it services only a small number of machines. A physical store still has to cover rent, demonstration units, and specialist sales support even if customer traffic is limited.
Even with leading robotics companies now delivering significant volumes, those machines are scattered across China and, increasingly, overseas. That makes it difficult for any individual service location to achieve the density needed to operate efficiently.
If each brand independently builds its own repair sites, warehouses, sales channels, and service teams, limited sales volumes will struggle to absorb the high fixed costs. Network utilization will remain low, while per-unit service costs stay high.
Handing the work to fragmented third-party providers creates another set of problems: uneven technical capabilities, inconsistent access to parts, and varying service standards.
Duplicated infrastructure creates more than a cost problem. It can also drain resources from robotics companies’ R&D efforts.
The core competitiveness of a robotics company still comes from its robot hardware and technology. Commercialization is supposed to provide cash flow for continued development. But as a company gets closer to the market, sales, channels, delivery, and services consume more resources. At an early stage, commercialization can end up straining the organizational capacity of a technology company instead.
When a growing number of companies struggle with the same difficult leap from technology to market, the problem no longer looks like a weakness in individual businesses. It begins to reveal gaps in infrastructure.
Who, then, should build the underlying infrastructure that every company needs but no single brand can easily create on its own?
Who builds the infrastructure?
The value of a platform company is not that it does everything on behalf of robotics companies. It is that it can connect common functions that were previously fragmented across individual brands.
The first step is organizing market demand that has yet to fully take shape.
Late last year, Unitree Robotics opened what it said was its first offline experience store worldwide inside the Shuangjing branch of JD Mall in Beijing.
The companies combined JD.com’s first-party online retail operations with an offline physical experience. Unitree supplied products, demonstration units, and specialist technical support, while JD.com provided space, customer traffic, transactions, warehousing and distribution, and after-sales service.
On the surface, this was a channel partnership. But in a category whose definition is still changing quickly, the value of a channel extends beyond providing a place to sell products.
What are consumers willing to pay for? Which groups are different robots best suited to? Which use cases actually exist?
All of these questions still need to be tested.
A platform therefore sits between brands and products on one side and real users on the other. Its role is to turn fragmented demand into something more identifiable.
According to Unitree’s prospectus, JD.com was one of its largest customers during the reporting period. Their partnership has also expanded from sales of consumer robots into a broader range of commercial scenarios.
That offers a direct illustration of a larger shift: large platforms are no longer merely sales channels. They are also becoming aggregators of demand.
That is what makes JD.com a case worth watching.
JD Retail’s hybrid model, combining first-party retail and third-party merchants, has already turned warehousing, logistics, supply chains, and service networks into systems that can be shared across brands.
Its entry into robotics effectively extends that organizational capability into a new product category that has yet to scale.
JD.com said it is already working with more than 200 robotics brands through its first-party retail model, with the platform handling merchandise operations, customer service, warehousing and distribution, fulfillment, and parts of after-sales service.
Searches, inquiries, transactions, and reviews generated by roughly 700 million consumers and eight million enterprise customers can, in turn, be translated into insights about product demand, according to JD.com.
What kind of robots do users actually need?
One telling detail is that JD.com has already divided robots on its platform into 16 categories based on user needs.
Sometimes, a brand has not yet decided which customer segment it should target, and JD.com helps clarify that positioning first. In this sense, the company is trying to act as an early interpreter of demand in an emerging consumer market.
Once demand has been organized, the next step is scaling supply. JD.com has therefore extended its efforts further into the supply chain, services, data, and other parts of the industry.
Upstream, JD.com is trying to aggregate fragmented demand for components that many companies need.
Batteries are one example. Different manufacturers currently define their own requirements for safety, capacity, dimensions, and communication protocols. Because individual brands purchase relatively small volumes, they have limited bargaining power and can struggle to maintain stable inventories.
As part of the initiatives announced in its latest robotics push, JD.com is working with more than 20 industry partners to promote standardization of robot batteries. It plans to aggregate demand through centralized procurement, inventory stocking, and staggered collection by customers.
For an industry that has yet to reach annual shipments in the millions, standardization may be more than a result of maturity. It may be one of the conditions required for scaling to happen at all.
Then comes another question: how can robots operate reliably in the real world over the long term?
JD.com said it has already built eight robot repair centers in China, while its repair services have expanded to Europe, the Middle East, and North America.
Over the next five years, it plans to establish 80 RoboBase robotics facilities. These centers are intended to bring component production, pilot production of complete robots, secondary development, maintenance and repair, and recycling into a full-lifecycle service system, with related capabilities covering more than 100 countries and regions worldwide.
This approach resembles JD.com’s earlier approach to logistics infrastructure: invest heavily to establish a network first, then allow multiple brands and product categories to share it, creating service scale that no individual brand could support at the current stage.
Whether this service capability can work as intended will still need to be tested in demanding field environments, where unexpected failures can occur frequently.
At the World Humanoid Robot Games, JD.com also deployed robots to support on-site operations, putting its capabilities to the test under real operating conditions.
But before robotics reaches mass adoption, another problem still needs to be overcome: data.
Embodied intelligence requires large amounts of high-quality data from the physical world, but obtaining real-world data is far more difficult than collecting internet data.
Internet-based models can learn from existing text and images. Robot data, by contrast, often has to be generated as machines operate in real environments.
JD.com has access to many such environments, including JD Mall, 7Fresh, logistics and supply chain operations, and unmanned pharmacies.
High-quality data derived from human actions can also be useful. For example, the movements and procedures performed by JD.com’s home service workers as they clean, organize, and carry out other real-world tasks could be collected and annotated to create training examples for robots.
JD.com has identified this as an opportunity. It aims to collect more than 10 million hours of real-world scenario data over the next two years and use capabilities including JoyInside to support multimodal interaction and adaptation to different environments.
Viewed together, JD.com’s robotics strategy is not simply a collection of separate initiatives.
On one side, it is reusing its existing retail, supply chain, and service networks to connect with market demand.
On the other hand, it is building infrastructure specifically for the robotics industry, including standards, RoboBase facilities, and real-world scenario data.
The latter covers much of the process of bringing a robot from production into practical use.
Scenarios and data address how robots can be trained and iterated. Standards and supply chains address how they can be manufactured at scale. RoboBase is intended to address how they can be deployed, maintained, and operated over the long term.
As robots move toward scaled deployment, the industry’s need for infrastructure is becoming more apparent. That infrastructure has to do two things: reduce the cost of commercialization and shorten the distance between technological iteration and market feedback.
Robots enter the real world
As robots move into practical applications, competition is beginning to shift from isolated capabilities toward the ability to perform real-world tasks continuously.
Embodied intelligence ultimately has to create value through action.
Hardware carries the capability. Whether a machine can keep moving goods, cleaning spaces, guiding visitors, or performing production tasks is what determines whether it is useful in practice.
Only after robots enter real environments can they continuously expose problems, generate data, and improve the user experience. That is how commercialization and technological iteration can begin to reinforce each other.
Judging from the strategy JD.com has disclosed, its goal is not merely to become the largest robotics sales platform.
It wants to provide the supply chains, operating environments, data, and services that products need after they enter the real world.
Robotics companies will continue to determine the capability boundaries of robot hardware and models. JD.com, meanwhile, is trying to help those capabilities find demand faster, enter real-world environments, and improve through continuous operation.
Its investments in supply chains, service bases, and real-world scenario data all come into play after robots leave the laboratory and enter the physical world.
Consistent with that goal, JD.com also announced that by 2028 it plans to commit RMB 10 billion (USD 1.5 billion) in resources to robotics, help 100 brands each surpass RMB 1 billion (USD 148.4 million) in sales, and push the robotics industry deeper into end-use scenarios.
Over the next five years, it also plans to create more than 100,000 jobs for robot maintenance engineers to support the industry’s after-sales service needs.
The RMB 10 billion commitment is substantial. But what may matter more for the industry is that JD.com has tied long-term investment to brand growth and end-market penetration.
Only if real orders continue to expand can the capabilities being built today develop into a functioning industrial system.
As in many other industries, once common infrastructure is in place, individual companies no longer need to build every capability from scratch. They can redirect more resources toward products and technology, potentially lowering the cost of bringing robots to market.
That does not mean a platform can help the industry clear every hurdle.
Robot hardware, motion control, and models will still require continued advances from technology companies. The reliability of robots in open environments will also take time to prove.
What is changing is that competition is no longer confined to capabilities developed inside the laboratory.
Getting products into real operating environments at lower cost, then keeping them running reliably over the long term, is becoming another major front.
The running, dancing, and backflips at the World Robot Conference showed how far robot capabilities have advanced.
But between a successful demonstration and reliable operation lies a long road built from supply chains, data, real-world environments, and services.
JD.com’s latest move is an attempt to build more of that road.
Only as those capabilities mature can robots move from the stage into shopping malls, warehouses, factories, and homes.
That distance, between technological capability and reliable long-term operation, is where JD.com is placing its bet.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Xi for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.74 = USD 1 based on estimates as of August 28, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.