Article · AI translation
Global training, global delivery
Chinese physical AI companies now have a global training ground.
On November 5, Cainiao held an AI and embodied intelligence industry cooperation launch event at its Hong Kong Smart Gateway, announcing that it would open its global warehouse network to AI and embodied intelligence companies to create a cluster of real logistics settings for training AI and embodied intelligence.
The range of logistics tasks available to embodied robots is quite comprehensive. From warehouse picking and palletizing and depalletizing to autonomous navigation and delivery and inspections at height, it covers many stages of logistics operations.
Cainiao has also put together an overseas expansion package for these robots.
At the event, Cainiao launched its “three-day delivery for technology going global” service for AI computing power and embodied intelligence companies, extending its global three-day delivery capability to high-tech products entering overseas markets.
The full set of services covers cross-border transportation, overseas deployment, spare-parts replenishment, returns for repair, and local operations and maintenance.
Cainiao is giving embodied robots real warehouse settings in which to practice. At the same time, AI hardware and embodied intelligence products going overseas can obtain a full range of logistics support here, from cross-border transportation and overseas deployment to returns for repair and maintenance.
By using its global warehouse network and logistics system, Cainiao is meeting physical AI’s needs from R&D validation through commercial deployment, connecting the first and second halves of industrialization.
This time, Cainiao is tackling both sides.
What settings is Cainiao opening to robots around the world?
Cainiao has prepared quite a few “global training positions” for robots this time.
Drawing on a logistics network covering more than 200 countries and regions and warehousing infrastructure numbering in the tens of thousands, Cainiao will open logistics settings across dozens of industries, including retail, 3C, aviation, automotive, fashion, and fast-moving consumer goods, to industry partners.
Cainiao has identified dozens of broad categories and more than 100 proof-of-concept (PoC) scenarios for productivity gains, covering warehouse picking, autonomous navigation and delivery, inspections at height, security operations and maintenance, aviation logistics, and auto parts.
More specifically, the tasks robots can take part in fall into roughly 3 categories.
In-warehouse operations: frequent tasks requiring hand-eye coordination, such as feeding parcels, applying labels, picking individual items, and palletizing and depalletizing;
Mobile robot tasks: dynamic movement tasks such as autonomous navigation and delivery, inspections at height, and security monitoring;
Vertical-industry settings: specialized fields with higher barriers to entry, such as aviation logistics and auto parts.
From simple repetitive actions to intricate multistep operations and long-range planning in complex environments, these tasks cover several key directions in which embodied intelligence is expanding from basic operations to complex tasks.
Simulation can handle pretraining, but the continuous operation, changing conditions, and variety of parcels in a real warehouse are key to generalization.
Xiong Wei, president of Cainiao International, gave a very specific example in a media interview.
In a real warehouse, a robot may need to run continuously for 1,000 to 2,000 hours and handle tens of thousands of parcels a day.
In addition, changes in lighting, temperature, humidity, dust, and other on-site conditions provide different data inputs for a robot’s hardware and model.
In a real warehouse, a robot may need to run continuously for 1,000 to 2,000 hours and handle tens of thousands of parcels a day.
In addition, changes in lighting, temperature, humidity, dust, and other on-site conditions provide different data inputs for a robot’s hardware and model.
In his view, the generalization capability gained from such real-world settings could be more than 1,000 times that gained in a laboratory environment.
The value of on-site data also lies in its record of a robot’s successes and failures.
Data from a robot’s successful operations can be used to train models, and data from correcting an operation after an error is also valuable learning experience.
As Xiong Wei sees it, a general-purpose model may already have a capability score of 80 after pretraining. Once it enters a specific industry setting, parameter adjustments, post-training, and contextual capabilities can continue to refine it, gradually raising the score to 95 to 98, and then toward 99 and 99.99.
Moving from initial usability to stable operation at an industrial level requires extensive refinement through real tasks. By opening its warehouse network, Cainiao is providing precisely this kind of training and validation environment.
During a panel discussion, Li Qiang, founder of Kunteng Power, spoke about what makes logistics settings distinctive: “Logistics tasks allow some room for error. If a robot fails to pick up a parcel the first time, it can try again. At the same time, parcels in real warehouses come in a wide variety of forms, providing a rich set of operational examples.”
This helps explain why logistics has become one of the first industry settings explored by embodied intelligence companies.
Of course, the value Cainiao can offer also comes from its logistics technology capabilities built up over many years.
From its early work on unmanned warehouses to bin shuttles and goods-to-person systems, and then coordination between large and small vehicles and climbing robots, Cainiao has invested in logistics automation for more than 10 years.
As of October 2026, Cainiao Logistics Technology had implemented more than 1,100 projects in 29 countries worldwide, serving more than 400 customers.
Cainiao has years of industry experience across the process from robot R&D and pilot programs to engineered deployment and operations at scale, and some of its robot warehouses are already operating.
Cainiao continues to expand its global robot warehouse network, providing cross-border e-commerce platforms and merchants with local warehousing, order dispatch, and delivery support. Robot warehouses are already in use in Hong Kong, China, and a climbing-robot warehouse has entered routine operation.
In Europe, a robot warehouse in the Netherlands is about to open, while another is under construction in Spain.
At a cross-border warehouse where Cainiao’s self-developed ZeeBot climbing robot has been deployed, storage density has increased by 40%, and labor productivity in warehouse operations has risen by 100%.
Another advantage for Cainiao is that its warehouses are not empty sites waiting for robots to be tested.
Years of serving manufacturers and global e-commerce have given it experience with large volumes of products, complex orders, promotional demand surges, and cross-regional operations, allowing it to offer robots diverse examples and working conditions closer to real production.
Xiong Wei also noted in his speech that logistics environments differ markedly between China and overseas markets.
Take express parcel sorting: China has developed a large-scale system for separating large and small parcels, but parcel sizes and weight distributions may be entirely different across overseas markets. These differences affect a model’s recognition and operating strategies, as well as a robot’s load-bearing capacity.
A rich variety of products, sufficiently complex working conditions, diverse settings built up through years of global operations, and a continuously running production environment of the kind robots need: all are valuable training resources for embodied intelligence companies seeking to improve generalization.
Getting the equipment there is only part of the job; overseas deployment and after-sales service must also be arranged
Opening the warehouse network addresses the need for settings in which to develop and validate robots. Cainiao has also prepared a solution for the delivery challenges high-tech products face once they go overseas.
The consumer robots, AI computing equipment, and embodied intelligence products that Cainiao currently focuses on helping enter overseas markets each present different supply-chain challenges.
Consumer robots and 3D printers must handle both bulk orders and scattered orders, as well as peak-season surges;
AI computing equipment is high-value and bulky, with delivery closely tied to the progress of data center construction;
Embodied intelligence products involve prototypes, trade shows, and bulk deliveries, as well as longer-term needs such as spare parts, returns for repair, and local operations and maintenance.
Xiong Wei said Cainiao has been exploring longer-term coordinated overseas expansion plans with a number of embodied intelligence companies. As robots move from one-off needs for research and trade shows toward ongoing industrial applications, overseas services also need to extend from single shipments to long-term support.
To address these issues, Cainiao has upgraded and launched “three-day delivery for technology going global,” drawing on 3 main capabilities—global express services, global air freight, and its global warehouse network—to support technology products at different stages of overseas expansion.
Global express services mainly address the need for time-sensitive delivery.
Whether for R&D testing, a customer demo, an overseas trade show, or small-batch replenishment, transport methods can be combined flexibly according to the cargo’s characteristics. Even a single robot prototype can receive end-to-end service from pickup to overseas delivery.
Global air freight is aimed mainly at large, high-value, and overweight cargo such as AI computing equipment, control systems, and core modules.
These products place high demands on transport safety and delivery times. Cainiao can mobilize global air freight resources to provide customized door-to-door service and support detailed steps such as scheduled delivery, moving equipment upstairs into place, and unpacking and handover.
The global warehouse network serves established companies that need to operate in overseas markets for the long term.
Drawing on its worldwide warehousing network, Cainiao can provide regional stock positioning, warehousing across multiple countries, local delivery, and management of after-sales spare parts. It can also link outbound fulfillment with reverse logistics for repairs, helping companies build more stable inventory and service systems overseas.
Together, these 3 types of service cover more than shipping equipment from China overseas: they span the entire process from prototype testing and bulk delivery to overseas inventory management and after-sales support.
This service package is already being used in actual business operations.
In the 3D printer business, Cainiao planned warehouse capacity and processing capacity in North America and Europe in advance, helping a customer reduce slow-moving inventory by more than 20%, lower final-mile costs by 15%, and improve delivery times by 1 to 3 days.
In the robotic vacuum cleaner business, Cainiao uses serial-number-level management, separate storage, and insurance arrangements to track individual products and dispatch items from larger lots.
The service for AI computing equipment shows the level of detail in this system even more clearly.
Faced with irregular shipping frequency and a diverse range of equipment, Cainiao must coordinate domestic ports, air cargo capacity, overseas customs clearance providers, data centers, and equipment manufacturers so that everyone stays aligned on timing and processes.
After the equipment arrives, a specialist last-mile team must complete scheduling, delivery, loading, and unloading in accordance with the customer’s requirements.
According to data Cainiao disclosed on a related AI computing project, the end-to-end on-time performance rate exceeded 95%, the customs clearance time target attainment rate exceeded 98%, the customer complaint rate was 0%, and logistics costs fell by more than 10%.
Cainiao has its own business case for doing this
After all this work, Cainiao naturally has its own calculations.
In the past, Cainiao primarily used its global warehouse network, time-sensitive delivery network, and supply-chain capabilities to serve cross-border e-commerce and the global flow of goods.
Now, the AI computing and embodied intelligence industries are bringing new customer needs, and Cainiao is extending its existing logistics technology, operational settings, and supply-chain capabilities to the hard-tech sector.
Xiong Wei revealed that relatively mature technology categories such as consumer robots already account for 10% of Cainiao International’s revenue, with growth doubling this year.
Combined revenue from the overseas expansion of consumer robots, AI computing, and embodied robots has grown by more than 100% from last year.
Xiong Wei described Cainiao’s future global growth as driven by both e-commerce and technology expanding overseas.
For Cainiao, opening its warehouse network and working with embodied intelligence companies also serves its own interests.
The primary goal remains improving its own productivity. Faced with increasingly complex and varied operational needs in logistics, Cainiao hopes to introduce more robotics technology so that robots can better assist workers and improve the efficiency of human-robot collaboration.
Specifically, Cainiao can combine the real logistics tasks and operational experience it has accumulated with partners’ models, hardware, and operating capabilities to conduct joint R&D and business validation.
This can help robotics companies refine their products in real-world settings, while also creating opportunities to upgrade Cainiao’s own dispatch systems, in-warehouse automation, and robots’ operational capabilities.
At the same time, the new needs of hard-tech companies expanding overseas are gradually becoming a business area worth investing in for Cainiao.
Embodied intelligence involves a long technology stack, from the robot itself and foundation models to vertical models, agents, and harnesses, and then to standard operating procedures (SOPs) integrated with actual operations. Each layer has different technology investment needs.
Cainiao hopes to invest in technology areas closely related to its own operational settings and work with partners in complementary roles, avoiding duplicate development while jointly refining solutions that can keep working on site.
Cainiao currently operates millions of square meters of logistics facilities worldwide, has deployed more than 10,000 specialized robots, and has hundreds of thousands of frontline employees on site.
Over the next 10 years, Cainiao hopes robots will assist in at least 50% of the tens of thousands of frontline physical work environments, taking on more complex, temporary, and high-intensity work to help people become more productive.
Industrial support for physical AI is expanding from R&D and training to validation, delivery, and operations and maintenance. By opening its global warehouse network and upgrading its three-day delivery service for technology going global, Cainiao is linking 2 important stages in the industrialization of physical AI.
Refine the technology through real business operations, then use the global supply chain to reach broader markets.
Refine the technology through real business operations, then use the global supply chain to reach broader markets.
For Chinese AI and embodied intelligence companies, beyond R&D, there are now more settings and logistical resources they can draw on.
As for how big this business can become, Cainiao’s own books already offer a first indication: revenue from taking technology overseas more than doubled, and its profitability was even higher than that of its cross-border e-commerce business.
A win-win for everyone~
As more technology companies and upstream and downstream partners join forces with Cainiao, their collective, open collaboration could, over time, build a lasting support network for industrialization and global expansion.
From R&D validation to overseas delivery, capabilities that companies once had to build one by one themselves can gradually be filled out through industry partnerships. Technical teams can then devote more energy to solving problems with the products themselves.