I keep coming back to the same question whenever new shipment numbers drop. How did one country end up with almost the entire early market for humanoid robots while everyone else is still figuring out the basics? The latest figures show Chinese makers delivered roughly 97 percent of global humanoid units in the first half of the year. That is not a rounding error. That is near-total control of the first meaningful production wave.
The Numbers That Should Make Every Policy Maker Pause
Global shipments of humanoid robots tripled in the first six months compared with the same period a year earlier. The total reached about 19,100 units. Forecasts now point toward roughly 60,000 units for the full year and as high as half a million by the end of the decade. Those are not speculative slides from a conference. They line up with the more cautious institutional projections that sit in the high hundreds of thousands.
What stands out is not simply the growth rate. It is the concentration. When nearly every unit leaving a factory carries a Chinese label, the rest of the world is not competing on the finished product. It is competing on whether it can even assemble one without importing the critical pieces.
I have followed industrial policy long enough to recognize the pattern. Years of targeted funding, patient capital, and coordinated direction tend to produce these kinds of results. The same approach that accelerated electric vehicles and drones is now visible in bipedal machines that can walk, grasp, and increasingly operate with less human supervision.
Why the Supply Chain Matters More Than the Finished Robot
A humanoid is only as good as the stack of parts underneath the polished exterior. High-torque motors, precision actuators, force and torque sensors, compact batteries with high energy density, and the software that turns sensor data into fluid motion. Most of those components are currently manufactured in Asia, with a heavy tilt toward Chinese suppliers.
This is the real choke point. You can ban finished robots at the border. You cannot easily ban the motors that make them move if those motors also appear in other industrial equipment. The recent decision to restrict certain Chinese humanoid and quadruped imports, along with selected components, acknowledges the risk. Yet it also highlights how thin the domestic alternative still is.
In my view the most interesting part of the current moment is not the robots themselves. It is the sudden recognition that physical AI will depend on a reliable, high-volume hardware base. Software can be written anywhere. Precision metal and rare-earth magnets cannot.
What Years of Policy Support Actually Looked Like
China did not wake up one morning and decide to dominate humanoids. The foundation was laid through sustained investment in robotics, sensors, and advanced manufacturing. Local governments competed to attract factories. Universities funneled talent into mechatronics programs. State-backed funds took early positions that private capital might have avoided.
The result appeared on the exhibition floor of the recent World AI Conference in Shanghai. Dozens of companies showed machines that moved with noticeably better balance, faster recovery from disturbance, and more natural manipulation of objects. Dexterity that looked experimental two years ago now looks closer to commercial readiness.
That kind of visible progress creates its own momentum. Engineers see what is possible. Investors see a market that is starting to exist. Factory managers begin to plan pilot deployments. Once the flywheel turns, catching up becomes harder with every quarter.
The American Response So Far
The current administration has made clear that humanoid robots sit at the intersection of industrial capacity and national security. Import restrictions on new Chinese humanoids and certain related components are one concrete step. The stated concerns include cybersecurity risks to AI infrastructure and the broader strategic dependence that would follow from relying on foreign hardware for critical tasks.
What has not yet appeared is a dedicated executive order or large-scale industrial program focused specifically on accelerating domestic humanoid production. That absence is noticeable. Other sectors have received targeted attention. Humanoids remain largely inside broader technology and manufacturing conversations rather than at the center of a focused push.
I suspect that is changing. The shipment numbers are difficult to ignore. When nearly every unit in the early market comes from one country, the strategic implications become obvious even to people who prefer not to think about industrial policy.
Where the Real Bottlenecks Sit
Let’s be specific. The hard parts are not the outer shells or the basic processors. They are the actuators that deliver high torque in a compact package, the sensors that give the machine a sense of force and balance, the battery systems that keep a walking platform running for hours rather than minutes, and the control software that turns all of those inputs into reliable motion under real-world conditions.
Many of the specialized motor and actuator companies sit in Asia. Some of the most capable sensor producers are concentrated in the same region. Battery chemistry expertise for high-power, high-cycle applications has followed the electric vehicle supply chain. Replicating that depth takes years, not months.
This is why simple bans, while understandable, are incomplete. Restricting finished robots without building the underlying component base simply shifts the problem one layer deeper. The machines still need to be assembled from something.
- High-torque, low-inertia actuators remain difficult to source outside current production hubs
- Force-torque sensors at the required price and performance points are scarce
- Battery packs that balance energy density with safety under dynamic loads are still specialized
- Real-time control stacks that handle balance, gait, and manipulation simultaneously are software-heavy but still depend on reliable hardware feedback
Each of those items represents a multi-year development and manufacturing ramp. Waiting until the market is already large makes the climb steeper.
Physical AI and the Next Industrial Layer
People sometimes talk about AI as if it lives entirely in the cloud. That view is incomplete. The next phase involves machines that move through the physical world, interact with objects, and operate in environments designed for humans. Warehouses, factories, construction sites, and eventually homes. Those machines need bodies. The bodies need reliable components. The components need a supply chain that can scale without geopolitical interruption.
I have found that the conversation often jumps too quickly to the finished product. Investors ask which robot company will win. Policy makers ask which country will lead. Both questions matter, yet the quieter question is more decisive: who will control the motors, the sensors, and the batteries that every serious humanoid will require?
That is where the current imbalance is most pronounced. Finished robots can be designed in many places. The industrial capacity to produce the critical sub-systems at volume and cost is far more concentrated.
What Closing the Gap Would Actually Require
Catching up is possible, but it is not automatic. It requires more than research grants and demonstration projects. It requires patient capital willing to fund the messy middle stage between prototype and high-volume production. It requires skilled technicians who can keep advanced manufacturing lines running. It requires standards that allow different suppliers to interoperate without forcing every company to reinvent the same actuator.
Perhaps the most practical step is to treat the component layer with the same seriousness currently applied to semiconductors or advanced batteries. Identify the specific motor topologies, sensor types, and battery formats that appear in the leading designs. Map the current suppliers. Then decide which pieces can be on-shored or friend-shored at reasonable cost and timeline.
Some of that work is already happening quietly inside existing industrial firms. Others will need new entrants or partnerships. The timeline matters. Every year of delay allows the current leaders to deepen their process knowledge and lower their costs further.
The Risk of Waiting for the Market to Mature
There is a natural tendency to wait until the technology is more proven before committing large resources. That instinct is understandable. Early humanoids still have limitations in battery life, reliability under continuous operation, and cost. Yet the shipment trajectory suggests the early market is forming faster than many expected.
Once a country or region locks in the early production volume, the learning curve becomes self-reinforcing. Yields improve. Design iterations accelerate. Customer feedback loops tighten. Later entrants face both higher technical barriers and lower margins.
I have watched similar dynamics play out in other hardware-heavy fields. The window for meaningful participation does not stay open indefinitely. The 97 percent figure is a warning light, not a permanent scoreboard. But ignoring it invites a future in which critical physical AI hardware is sourced from a single geography under all conditions.
Labor, Productivity, and the Longer View
Beyond the immediate industrial competition sits a larger question about how work gets done. Humanoid form factors are attractive precisely because they can operate in spaces designed for people. Existing buildings, existing tools, existing workflows. That flexibility could matter in aging societies facing labor shortages in manufacturing, logistics, and care.
Whether those machines ultimately complement or displace human workers will depend on how they are deployed and regulated. The more immediate issue is whether the countries that will rely on them most heavily will also control a meaningful share of their production.
In my experience, technology that arrives exclusively from abroad tends to arrive with strings attached, whether formal or informal. Having domestic or allied capacity creates options. Options matter when supply chains become political.
Policy Tools That Could Still Move the Needle
Several practical levers remain available. Targeted procurement that creates early demand for domestically assembled humanoids. Tax treatment that favors capital investment in precision manufacturing equipment. Streamlined permitting for new production facilities. Collaboration between national laboratories and private firms on the hardest actuator and sensor problems.
None of these steps require reinventing the entire industry overnight. They require treating the component layer as strategic infrastructure rather than a pure market good. Markets are excellent at optimizing once the foundation exists. They are less reliable at building the foundation when the returns are distant and the risks are high.
The absence of a standalone executive order focused on humanoids does not mean nothing is happening. It does mean the effort remains diffuse. Concentration of attention often produces faster results than diffusion across many overlapping programs.
What the Next Few Years Are Likely to Bring
Shipments will continue to rise. The question is how the geographic distribution evolves. If current trends persist, the early market will be almost entirely supplied from one region. If policy and private investment respond with urgency, a second production base could begin to form.
I do not expect parity overnight. The lead is real and the process knowledge is deep. What I do expect is that the strategic importance of physical AI will become harder to overlook with every new data release. The 19,100 units shipped in the first half of the year are still a small absolute number. The direction of travel is not small.
Companies that control the actuators and sensors will quietly accumulate influence regardless of whose logo sits on the finished robot. That is the layer worth watching most closely.
A Personal Note on Timing
I have spent enough years around technology cycles to recognize when a capability is moving from laboratory curiosity to industrial reality. Humanoids are crossing that threshold now. The polish is improving. The failure modes are better understood. The unit economics, while still challenging, are moving in the right direction for certain high-value tasks.
The risk is not that the technology fails. The risk is that the production base remains so concentrated that the rest of the world becomes a permanent customer rather than a participant. Customers have limited leverage when supply is scarce or when political conditions change.
Closing the gap will require more than rhetoric. It will require factories, skilled people, and a willingness to fund the unglamorous work of making motors and sensors at scale. That work is already underway in some places. Whether it accelerates fast enough remains the open question.
The shipment data is public. The policy tools are known. The only missing ingredient is the decision to treat the problem with the seriousness the numbers already justify.
Looking Past the Headlines
Most coverage focuses on the finished machines walking across stages or performing tasks that look impressive on video. Those demonstrations matter. They show what is becoming possible. Yet the durable advantage will sit one layer below, in the ability to produce the enabling hardware at volume and cost.
Countries that master that layer will shape the physical AI era more than countries that simply buy the end products. The current concentration of production is therefore not a temporary curiosity. It is an early indicator of where the industrial center of gravity may settle.
Whether that center of gravity remains so heavily tilted depends on decisions being made right now about capital allocation, workforce training, and industrial strategy. The window for influence is still open. It will not stay open forever.
The 97 percent figure should not be treated as destiny. It should be treated as a clear signal that the race has already begun and that one participant has a substantial head start. The response, if it comes, needs to match the scale of the challenge rather than the scale of the current news cycle.
That is the real story underneath the shipment numbers. Not just who is leading today, but who is building the capacity to lead tomorrow.
