Humanoid Robots, Second Batch of Consolidated Delivery
When the second factory places an order, engineers may still have to start all over again. They need to connect the systems, adjust the working rhythm, handle exceptions until the acceptance check is completed; the work that was done for the first order may not be saved when moving to a new site.
Humanoid robots can be mass-produced by model, but their delivery has to be carried out in different factories. As orders increase, how many procedures and experience the enterprise can reuse from the previous delivery determines how many additional on-site staff it needs to recruit.
The first project usually requires the supplier and the customer to explore together. The customer is familiar with the process and knows when the equipment will affect the entire production line; the supplier is familiar with the robots and needs to align the requirements of both sides. After the two parties complete the commissioning, they will leave interface programs, action modules, test methods and fault handling experience. If these can be incorporated into the product and delivery process, the subsequent engineering team that takes over can use them directly.
Whether these assets can be reused depends on how closely they are tied to the original factory. The process data provided by the customer has specific usage conditions, and the interface written for one production line needs to be modified to adapt to other equipment. A solution that can only be debugged with the help of the original engineers requires far more manpower than a solution that another team can install following mature methods.
Manufacturing costs can be gradually reduced through bulk procurement, design improvement and higher production efficiency. Delivery costs depend more on on-site workload. Some enterprises turn project experience into standard products, so they do not need to expand their engineering teams at the same speed when orders increase; other enterprises make profits through customized implementation and services, and can also operate sustainably as long as the contract price covers the corresponding input.
In 2025, BMW tested the Figure 02 at its Spartanburg plant in the United States. In 2026, BMW further promoted the AEON pilot with Hexagon at its Leipzig plant in Germany. The two suppliers correspond to different factories and tasks. BMW also disclosed that the group has gradually unified its production IT and data architecture in recent years, and established a Physical AI Competence Center, where the internal team is responsible for technology assessment, pilot and deployment, gradually forming modules for its factories to use.
Robot suppliers can vary, but the access experience accumulated by BMW itself can continue to be used. The engineering team familiar with the production system, unified data environment and test methods can provide a foundation for the subsequent introduction of equipment. This part of the customer's capabilities improves the flexibility of procurement options, and may also reduce the coordination work that needs to be completed after new suppliers enter the site.
In 2024, U.S. logistics enterprise GXO signed a multi-year robot-as-a-service agreement with Agility Robotics. Digit robots were deployed in GXO's warehouses and connected to the Agility Arc platform at the same time. The platform is responsible for site mapping, process setting, operation management and fault handling, and the robots also need to cooperate with the existing automatic mobile equipment and conveyor lines in the warehouse.
Agility hopes to manage equipment and processes at different sites through the platform. New warehouses still need to adjust workstations and connect existing systems. If the configuration tools and fault handling methods that have been written into the platform can be used directly, engineers can reduce redundant work, and customers do not have to start the docking process from scratch every time.
When suppliers send fewer engineers, customers sometimes need to arrange more in-house technical personnel. For the entire project, the work may only be transferred to the other party. If the customer establishes a mature access system, the supplier's installation work may be reduced accordingly. The benefits brought by efficiency improvement ultimately depend on how much work each party undertakes.
In the investor materials submitted by Agility in October 2026, robot sales, one-time deployment, and subsequent software and maintenance are accounted for separately. In the illustrative model for Digit 5, Agility sets the one-time deployment cost for a single robot at about 15,000 U.S. dollars, and the annual software and maintenance delivery cost at about 15,000 U.S. dollars. After the installation is completed, continuous investment is still required for maintenance and software services.
The decline in robot manufacturing costs can increase the profit margin of each unit of equipment; fewer deployment personnel can reduce installation costs; software and maintenance generate continuous revenue, but corresponding service input also needs to be taken into account. Customers may request price reductions in the next contract, and system integrators may also take over part of the work. How much value a robot enterprise can capture depends on delivery responsibilities, service contents and contract prices.
To observe whether a robot enterprise can make its delivery process more efficient as it completes more projects, it is best to compare subsequent projects with similar tasks and independent customers: how much of the interface program is rewritten, how many days engineers need to stay on site, how many technical personnel the customer invests, and who is responsible for acceptance and subsequent maintenance. Adding equipment in the same factory can reuse existing facilities and teams; entering a new independent customer can better test how much the experience is still useful after leaving the original production environment.
Even with the same order growth, the operating results of robot enterprises can be very different. If interfaces and commissioning methods can be reused across different customers, the engineering input required for new projects will have the opportunity to decrease; if re-adaptation is required every time a new factory is accessed, the increase in revenue will also continuously bring higher delivery costs. Even if on-site work is reduced, factors including customers' price reduction requests, integrators' participation in labor division, and subsequent maintenance input will affect which party the saved costs ultimately belong to. After the second order, whether the delivery efficiency can be converted into the enterprise's own profits will gradually show the scale effect.