Shanghai-based Agibot is probably the largest humanoid robot maker in the world. The company shipped roughly 5,100 units in 2025 — a 39% share of the global market — and crossed 10,000 cumulative units earlier this year, surpassing its first three-year total in just three months. It now offers humanoid robots and robots-as-a-service in more than 17 countries.
I recently had the opportunity to interview the president of Agibot’s embodied business unit, Dr. Yao Maoqing, via email.
I wanted to know where its thousands of shipped robots are being employed. When we’ll hit a tipping point and the humanoid robot era will truly begin. What the biggest blockers are today, how Agibot sees the global market – including potential emerging protectionist forces – vertical integration of hardware and software, plus how the company sees robots and humans working together in the future.
In this interview Dr. Maoqing argues that the humanoid robotics industry is right now crossing a far more important threshold than production capacity: the shift from demos to deployment, and he frames it as a move from what he calls the “X curve” of technology exploration to the early stage of the “Y curve” of real-world deployment growth.
He’s also candid about what differentiates Chinese robotics firms … and it’s not what most Western observers assume.
Koetsier: The scale you recently announced is impressive. Where are those robots going: labs? factories? homes? offices?
Dr. Maoqing: At this stage, humanoid robots are first entering scenarios where demand is relatively clear, the environment is more controllable, and a closed-loop value creation can be established. These include industrial manufacturing, logistics and warehousing, commercial services, security and inspection, research and education, data collection, and scenario validation.
From our perspective, the embodied intelligence industry is gradually moving from the “X curve” into the early stage of the “Y curve.” The X curve is essentially the technology exploration phase, where the industry focuses on proving whether robots can move, see, and understand. The Y curve is the deployment growth phase, where the focus shifts to whether robots can truly enter workflows, operate continuously and reliably, and create real productivity value.
So at this stage, large-scale deployment will not begin in homes. It will begin in industrial, commercial, and service scenarios. These scenarios involve higher-frequency tasks, clearer ROI, and are more likely to generate a real data flywheel through deployment, which in turn improves reliability, intelligence, and generalization.
Homes are certainly a very important long-term direction. But the home environment requires much higher levels of safety, cost efficiency, interaction quality, long-term stability, and generalization. The technology stack needs further maturation before humanoid robots can enter homes at scale.
For us, the significance of scale is not simply about how many robots have been produced. It is about how many robots can actually enter real-world scenarios, operate continuously, and form deployment models that are replicable and scalable. We believe that is the real sign that embodied intelligence is beginning to become a form of productivity infrastructure.
Koetsier: You hit 1,000 units after roughly two years, then 1,000 → 5,000 in about a year, and 5,000 → 10,000 in another year. What’s your internal target for the next 10,000? 100,000?
Dr. Maoqing: This is no longer just about production volume.
Over the past few years, we have indeed made rapid progress in R&D, manufacturing, supply chain, and delivery capabilities. The industry was also more focused on proving the technology itself: whether robots could walk, run, or complete certain complex movements. But starting from this year, one very clear change is that more and more customers are seriously discussing how to deploy robots, how to replicate deployments, and how to calculate ROI. That shift is very important.
In the past, many projects were more demonstration-oriented or validation-oriented. But now we are seeing some customers move from one robot, or a few robots, to one factory, one production line, or even deployments across multiple locations. They are beginning to treat robots as productivity tools, rather than as novel devices.
Internally, we are not overly fixated on a specific number for the next stage.
Compared with the number itself, we care more about whether robots are truly entering workflows, whether customers will continue to repurchase, whether scenarios can be replicated at scale, and whether robots become more stable and more intelligent as they are deployed.
For the humanoid robotics industry, the real inflection point is not only when production capacity increases. It is when demand begins to form a positive cycle.
Today, many people are still asking when robots will become widely adopted. But what we are seeing is that once robots begin to create stable value in real-world scenarios, demand itself may grow faster than many people expect. So whether the future is 100,000 units or more will not be determined only by supply. More importantly, it depends on when the industry truly completes the shift from “selling robots” to “delivering outcomes.” And we are already beginning to see that trend.
Koetsier: Are we at the tipping point for humanoid robots? Is production globally just about to explode?
Dr. Maoqing: I do think the industry is now very close to an important inflection point. But this inflection point does not mean that, all of a sudden, robots will enter every scenario overnight. It means that many of the key conditions that used to develop separately are, for the first time, are converging simultaneously.
Over the past few years, the industry has been waiting for several things. First, whether foundation model capabilities are strong enough. Second, whether the robotic body itself is reliable enough. Third, whether a continuous data flywheel can be formed in real-world scenarios. These elements existed before, but they did not come together at the same time. Sometimes AI capability was strong, but the robot was not stable enough. Sometimes the robot could move, but there was not enough real deployment data. Or the supply chain and cost structure were not ready to support scalable delivery.
But over the past two years, a very clear change is that large-model intelligence, robotic-body reliability, and the data flywheel are beginning to appear in the same time window. We are witnessing a fundamental shift in the industry narrative from “what can robots do?” to “can robots truly create productivity?”
We are already seeing more customers seriously discussing deployment, instead of just looking at demos. In areas such as industry, logistics, retail, and commercial services, real long-term operating scenarios have already started to emerge.
So I would say the industry is now on the eve of industrialization. But the real explosion will not come from one single technological breakthrough. It will come from the joint maturity of the entire value chain, including AI models, hardware, supply chain, data, scenarios, and business models.
In the next few years, I believe humanoid robots will first scale in a small number of high-value and replicable scenarios. Then they will gradually expand into more industries, more complex environments, and eventually broader social and home scenarios.
This process will not happen overnight, but the trend is already very clear.
Koetsier: What are the biggest blockers for humanoid robots right now? What areas need to improve the most?
Dr. Maoqing: I think the biggest bottleneck and challenge for the industry today is not one single technology. It is how to integrate many different capabilities into a system that can operate stably in the real world over a long period of time.
There is a massive gap between lab demos and real-world deployment. In the lab, you can control the environment, the task, and the variables. But the real world is dynamic, open-ended, and full of uncertainty. A robot must not only know how to perform actions; it must also understand the environment, understand human intent, handle unexpected situations, and operate reliably for long periods of time.
So we increasingly believe that humanoid robotics is not simply a software problem or a hardware problem. It’s a systems integration challenge at scale. There are several particularly important areas.
The first is still AI generalization. Many models today can already deliver very impressive demos, but there is still a long way to go before they can truly adapt to complex real-world environments. Robots will not ultimately work in fixed environments. They will have to deal with a large number of long-tail situations.
The second is data. One major difference between embodied intelligence and internet AI is that real-world data is extremely expensive to obtain. Robots need real physical data involving motion, manipulation, interaction, failure, and feedback, not just text data. That is why real deployment matters so much. Only when robots actually enter real scenarios can the data flywheel begin to form.
The third is reliability. I think this is often underestimated. It is not difficult for a robot to complete a demo once. The real challenge lies in sustained, fault-tolerant operation. Once a robot enters a factory, a store, or a public space, what customers care about most is not how smart the robot can be at its best, but whether it will fail at its worst.
Then there are also cost, supply chain, and maintenance systems. Humanoid robots will ultimately not be a niche device. They will become a large-scale industry, so they must be manufacturable, maintainable, deliverable, and replicable.
But the most important point is this: robots must ultimately solve real problems.
Over the past few years, the industry has seen many exciting technical demonstrations. But what will determine whether the industry can truly scale is whether robots can continuously create real value, where their productivity lies, and whether they can truly enter workflows instead of remaining at the stage of “showing robots.”
Koetsier: Are you concerned that protectionist legislation will block access to specific markets? How do you see the global market evolving?
Dr. Maoqing: I think any new foundational technology will face different requirements around safety, data, compliance, and industrial development when entering global markets. This is a normal process.
This is especially true for humanoid robots. They are not technologies that exist only in the digital world. They are physical AI systems that will enter factories, shopping malls, public spaces, and eventually even homes, working alongside people. So different markets will naturally pay close attention to reliability, safety, and localization.
But in the long term, I believe humanoid robotics is inherently a global industry, because the problems it addresses are fundamentally shared across the world.
Many countries today are facing changes in labor structure, the need to improve manufacturing efficiency, aging populations, and labor shortages in service industries. Whether in industry, logistics, retail, or services, the demand for automation and intelligence is continuing to grow.
Of course, different markets will develop at different speeds and in different directions. Some markets may prioritize industrial and warehousing scenarios. Some may focus more on services, elderly care, or homes. Others may place greater emphasis on research, education, and innovation ecosystems.
So what we care about is not simply entering one specific market in the short term. We care more about building long-term, trusted, and localized collaboration capabilities. Humanoid robotics is ultimately not an industry based on simply exporting hardware. It requires local deployment, local service, local operations, and long-term collaboration with local ecosystem partners.
I believe truly valuable technologies will ultimately enter global markets in a safe, compliant, and transparent way. For companies, what matters more is continuously improving product capabilities, building trusted systems, and truly solving customer problems.
Koetsier: If some of that legislation is based on safety/privacy concerns, should humanoids be offered with open-source models, inspectable source code, or perhaps should manufacturers offer models that don’t send data back home?
Dr. Maoqing: I think this is a very reasonable topic, and one that will remain important for a long time. Humanoid robots are different from many internet products in the past. They will eventually enter real physical spaces, including factories, commercial environments, and even homes. So concerns around safety, privacy, and data governance will only become more important.
The debate isn’t binary between open-source and proprietary. Open source and auditable code may be one way to build trust in certain scenarios. But what matters more is whether the entire system is verifiable, auditable, controllable, and able to meet the data governance requirements of different markets and customers.
In the future, different customers will have very different needs. Some customers will want fully localized deployment. Some will require private cloud solutions. Some will care more about edge computing, data isolation, or access control. And some industries will require very strict security audits and compliance systems.
So I think what humanoid robotics companies ultimately need to provide is not just a robot product, but a trusted technical architecture and governance capability. Customers need to clearly understand what data is collected, how the data is processed, what data stays local, and how the system is managed and audited.
When Agibot enters each market, we will respect local requirements around data, safety, and compliance, and provide corresponding deployment and governance solutions based on local regulations and customer needs.
For Physical AI, I believe the key question in the future will not be who is more “closed” or more “open.” It will be who can build a long-term trusted, transparent, and locally compliant deployment system. Trust will be the ultimate moat.
Koetsier: LG is an investor and the CEO toured your facilities recently. Is there a chance for LG to become a major customer? Would you partner on their CLoID robot?
Dr. Maoqing: LG is an outstanding global technology company with deep expertise in consumer electronics, smart home, manufacturing, and robotics. We value our dialogue with global leaders like LG.
One very clear trend is that more and more global technology companies are paying serious attention to humanoid robots and Physical AI. They recognize that this is not only about a robot product. It is an important foundation for the next generation of intelligent terminals, intelligent manufacturing, and service systems.
This industry also naturally requires long-term collaboration. Behind humanoid robots are AI, foundation models, robotic bodies, chips, supply chains, manufacturing, scenarios, service networks, and many other capabilities. The complexity demands ecosystem-level collaboration. So in the future, we will definitely see more cross-industry and cross-regional collaboration.
As for specific projects or specific cooperation, we will refer to information officially disclosed by both sides. But in general, we are always willing to work with leading global companies to explore the real-world deployment of robots, including opportunities in technology, scenarios, products, and global markets.
Koetsier: Sharing or renting robots looks like a monetization/ownership model that might work going forward. Your thoughts? What might a robot cost per day or per month?
Dr. Maoqing: I think the robotics industry will definitely see more rental, subscription, and RaaS, or Robot-as-a-Service, models in the future. This is a very natural direction.
Many customers do not actually need to “own a robot.” What they need is continuous and stable automation capability, and a real productivity outcome. Especially in the early stage of humanoid robotics, many companies will begin by validating value in specific scenarios and smaller-scale deployments, such as night-time inspection, material handling, guidance, retail services, or data collection. In those cases, compared with one-time purchases, RaaS lowers the barrier to entry for enterprises.
Also, a humanoid robot is not just a hardware product. Behind it are model upgrades, software maintenance, task adaptation, remote operations, and continuous data iteration. So the future business model of the industry may increasingly resemble cloud services or intelligent services, rather than traditional equipment sales.
At present, Agibot has already started to offer rental models in more than 17 countries and regions. For example, in the United States, the starting price for robot services is already around USD 2,000 per day. Of course, this usually includes not just the robot itself, but also deployment, operations and maintenance, software services, and sometimes even on-site support.
That said, at this stage, there are still significant differences across robot types, task complexity, and deployment environments, so the industry has not yet formed a unified pricing system.
I believe that as mass production scales, supply chains mature, and AI capabilities improve, the overall usage cost of robots will continue to decline. Ultimately, the core of industry competition will not simply be how much a robot sells for, but who can continuously and reliably deliver productivity value.
Koetsier How do you see robots and people working together in the future? As robots do more, will humans lose jobs, or do different jobs, or get some level of basic income? A Chinese court ruled recently that companies couldn’t fire humans because of AI … thoughts on similar approaches for robots?
Dr. Maoqing: I believe that, in the long run, the relationship between robots and humans will be much closer to collaboration than simple replacement.
In many industries today, The challenge isn’t overpopulation, but demographic shifts. It is that many jobs are becoming harder to recruit for, more physically demanding, or less attractive to younger workers over the long term. Examples include night-time inspection, repetitive handling, work in hazardous environments, and long periods of standing in service roles. These are exactly the types of scenarios where robots are well suited to enter first.
So in the early stage of the industry, humanoid robots will more often supplement labor shortages rather than directly replace people. At the same time, technological progress will certainly change job structures. This has happened in every industrial revolution.
In the future, some jobs will be redefined, and new types of roles will also emerge. For example, robot training, operations and maintenance, scheduling, safety management, data annotation, and scenario development are already beginning to appear.
In the longer term, I think society may also rethink what human work really means. If robots can take on more repetitive physical labor in the future, people may be able to spend more time on creativity, communication, companionship, decision-making, art, research, and other capabilities that are more uniquely human.
As for specific laws or policies, I do not think it is appropriate for me to comment on individual cases. But I do believe that a healthy process of industrial development should not only advance technology, but also help society manage the transition more smoothly. This is not only a corporate issue. It also requires education systems, industrial policy, and social security systems to gradually adapt.
Our mission is to augment human potential, not replace it. It is to allow robots to take on work that is not suitable for people to do over long periods of time, while helping people move toward more creative and more valuable work.
Koetsier: You have a very holistic approach: hardware, software, cloud, etc. Do you think vertical integration will win in the first few generations of the development of the robot ecosystem?
Dr. Maoqing: I think in the early stage of humanoid robotics, vertical integration is a strategic necessity in this nascent stage, and to some extent inevitable. The reason is that this industry has not yet formed stable divisions of labor or standardized interfaces.
A robot is not simply a hardware product. It is a highly coupled system. AI models, motion control, robotic body design, sensors, data systems, cloud scheduling, and task software are all still evolving together. For example, a change in the action model may directly affect body design. Data feedback from a real-world scenario may, in turn, change model training and system architecture. So in the early stage of the industry, if these capabilities are fragmented, it becomes very difficult to iterate quickly.
I think this is also why many robotics companies are now trying to build full-stack capabilities. Competition is no longer just about one component or one single technology. It is about who can form a complete closed loop across data, models, robotic bodies, and deployment more quickly.
But I do not think the industry will remain completely closed in the long run. On the contrary, as the industry matures, there will definitely be clearer ecosystem specialization. In the future, there will be operating system platforms, model platforms, robotic body platforms, supply chain platforms, scenario solution providers, and even robot application development ecosystems.
So I see vertical integration more as a stage-specific capability in the early phase of the industry, rather than the final form. Companies with long-term competitiveness will not only build deep capabilities internally, but also open up those capabilities so that more partners can help grow the entire ecosystem together.
Koetsier: What’s the question Western journalists should be asking Chinese humanoid robotics CEOs that none of us are actually asking?
Dr. Maoqing: I think one question that deserves more serious discussion is this: What is driving the accelerated deployment velocity of Chinese robotics firms?
Very often, when people outside China discuss the Chinese robotics industry, they first focus on simply cost advantages. But I believe the real core of competition in humanoid robotics is not who can make a demo. It is who can put robots into the real world faster and continue iterating from there.
Humanoid robotics is essentially a deployment-driven industry. Only when robots enter factories, malls, warehouses, stores, and public spaces can they obtain real-world data, form a data flywheel, and then use that to continuously improve models, hardware, and system capabilities.
China has a very unique combination of factors: a wide range of real-world scenarios, a complete supply chain, fast engineering iteration, and a large number of customers willing to try new technologies. When these factors come together, they can significantly accelerate the transition of robots from the lab to real deployment.
So I think the more important question for the future may not be who can create the most impressive video first. It is who can build real-world deployment capability, operational capability, and a data closed loop the fastest.
Because what will ultimately determine this industry may not be model capability alone, but who can continuously bring Physical AI into real workflows.
Koetsier: Thank you for your time.
