The Humanoid Market Is Early. The Best Entry Point May Be a Bottleneck.
China-based companies accounted for roughly 85% of global humanoid and quadruped robot shipments in 2025. Yet only around 10% of humanoids were actually deployed in real-world applications.
That gap should make corporate leaders cautious—but not necessarily patient.
The obvious response to an immature market is to wait for demand to become clearer. In robotics, that may be the wrong assumption. While the end market is still uncertain, the industrial positions underneath it are already being formed.
For companies considering physical AI, the more useful question is therefore not, “Should we invest in humanoid robots?”
It is:
If robotics scales, which bottleneck could become strategically valuable—and can we prove its value in a real workflow today?
The market can be early while the supply chain is already moving
China has built formidable advantages in robotics: manufacturing scale, deployment volume, batteries, vision technologies, and a rapidly developing domestic AI ecosystem.
But the physical AI stack is far from settled.
Research from Taiwan’s DSET points to continuing dependencies in areas such as frontier edge AI chips, simulation software, high-end precision components, machine tools, and specialized sensors. At the same time, the report argues that Taiwan and other non-China supply-chain participants may have only a three-to-five-year window to establish stronger positions before the ecosystem becomes harder to reshape.
That creates an unusual strategic situation.
Companies do not need to know which humanoid manufacturer will ultimately dominate. They need to identify which capabilities are likely to remain difficult, scarce, and valuable regardless of who wins at the system level.
For an established company, that may be a much more credible opportunity than trying to become a robot company itself.
The better opportunity may sit one layer below the robot
Taiwan illustrates the problem clearly.
DSET identifies roughly 180 Taiwanese companies and organizations already participating in, or developing capabilities relevant to, robotics. These capabilities include semiconductor manufacturing, EMS and system integration, precision motion components, industrial computing, sensors, and manufacturing scale-up.
On paper, that looks like an obvious industrial opportunity.
In practice, many suppliers remain hesitant.
The constraint is not simply technical readiness. Companies interviewed by DSET pointed to uncertain humanoid demand, limited financial incentives, and technology that is still evolving quickly.
This matters because corporate innovation teams often overestimate the value of having the “right capability.”
A capability becomes a business only when someone has a sufficiently important reason to pay for it.
In emerging technology markets, the scarce resource is often not engineering talent or manufacturing capacity. It is credible demand.
That is why building capacity because robotics appears inevitable can be dangerous. The technology may indeed grow dramatically while a particular component, architecture, or use case fails to become commercially important.
Start with a workflow, not a market forecast
BMW’s experimentation with Figure’s humanoid robots offers a more useful model for corporate decision-making.
BMW did not begin by making a broad prediction about the size of the humanoid market. It started with a specific manufacturing task.
The company tested the technology in the lab, moved it into a real factory environment, and observed what happened under production conditions.
During a ten-month pilot at BMW’s Spartanburg plant, Figure 02 robots operated for roughly 1,250 hours, moved more than 90,000 components, and supported production of more than 30,000 BMW X3 vehicles.
Those numbers matter. But the more valuable outcome may have been everything BMW learned around them.
Real deployment exposed requirements involving safety, IT infrastructure, connectivity, shop-floor logistics, production processes, and integration with existing systems.
Each of those requirements can become an innovation opportunity.
The useful question was never simply whether a humanoid could perform a task.
It was whether the task created enough operational value to justify solving all the surrounding problems required for deployment.
That distinction should shape how companies explore physical AI.
What should companies explore now?
For companies with existing strengths in industrial technology, electronics, software, manufacturing, logistics, or services, the best opportunity may be a narrow enabling position.
System integration. Edge computing. Motion components. Sensing. Safety systems. Certification. Deployment infrastructure. Maintenance. Workflow software.
The specific answer will differ by company.
What matters is the intersection of three things: a capability the company can credibly extend, an operational problem customers need solved now, and a robotics layer that is likely to become more important as adoption grows.
This is why physical AI should initially be treated as a venture discovery problem, not automatically as a major technology investment.
Early resources should buy evidence.
Can an existing corporate capability solve a sufficiently valuable problem in a real environment?
Will a customer change behavior, redesign a workflow, integrate a new system, or pay to remove that constraint?
If the answer begins to become clear, the company has something more valuable than a robotics strategy.
It has a foothold.
The humanoid market may still be early. The most defensible positions inside it may not wait for the market to mature.
If humanoid robotics grows substantially over the next five years, which bottleneck could your company credibly own—and what real customer problem could you use to validate that position before everyone else sees it?
Comments