Bloomberg published a piece this month with an argument that cuts against a lot of the current excitement about humanoid robots: for most factory assembly line work, the humanoid form factor is engineering overkill. Factory tasks are typically designed around fixed workstations, specific tooling, and well-defined motion sequences. A purpose-built robotic arm does those tasks more reliably, more cheaply, and with less engineering complexity than a bipedal robot with hands that can open a door. The humanoid shape solves a problem — operating in spaces designed for human bodies — that most industrial environments do not actually have, because those environments were designed around machines from the start.
That is a sound engineering observation. It has not, however, slowed investment or deployment timelines in any visible way. The reason is that the humanoid form factor solves a different problem than factory efficiency: it solves the retraining problem. A purpose-built robot arm is optimised for one task. When that task changes — when the product line shifts, when the assembly process is redesigned — the robot becomes a stranded asset. A humanoid robot that can be directed through natural-language instructions or demonstration learning can, in theory, be redeployed without a hardware refresh. The flexibility premium is what the investment thesis is actually buying, even if it is not what gets emphasised in the press releases.
China's position in this market deserves attention that it hasn't fully received. TechCrunch's February analysis of the humanoid robot landscape found China winning the early market by a significant margin, driven by a combination of government industrial policy, vertically integrated manufacturing supply chains, and aggressive production targets from companies like UBTech — which plans to ship 5,000 units in 2026 and 10,000 in 2027. Beijing has placed humanoid robotics at the centre of its strategic manufacturing ambitions in a way that mirrors its earlier bet on electric vehicles: use domestic demand and state support to build production scale before the technology is fully mature, then leverage that scale once the market develops globally. The EV playbook worked; there is reasonable logic to attempting it again.
Western deployments are moving more cautiously. Hyundai Motor Group announced plans to deploy humanoid robots at its Georgia manufacturing facility starting in 2028 — a two-year runway that reflects the gap between demonstration capability and production-ready reliability. The 1X robotics company struck a deal to ship up to 10,000 of its Neo humanoid robots to industrial customers across EQT's portfolio between 2026 and 2030, concentrating on manufacturing, warehousing, and logistics. These are real commitments, but they are scaled to the uncertainty: measured deployments in controlled environments rather than wholesale replacement of human workforces.
The number that frames the investment case is McKinsey's estimate of a $370 billion market for general-purpose robotics by 2040, with global humanoid robot shipments potentially reaching 2.6 million units by 2035. Those projections require a lot to go right — continued improvement in AI-based manipulation and locomotion, resolution of safety certification questions, and a production cost decline steep enough to make the economics work outside high-value assembly. None of that is guaranteed. But the scale of investment — from NVIDIA's physical AI infrastructure to Boston Dynamics' continued development to the Chinese government's industrial programmes — suggests that enough capital is committed that some version of a large humanoid robot market is likely to emerge, even if it looks different from what the current hype implies.
The Bloomberg critique is correct that the humanoid form is not the obvious engineering solution for most of today's factory problems. It is probably also correct that the first wave of humanoid deployments will be more about demonstration and learning than about production efficiency. What it undersells is the possibility that the right way to think about humanoid robots is not as factory tools but as general-purpose physical agents whose value is precisely their non-specificity — and that industrial deployment is just the most legible early market, not the terminal one.