Why the most useful humanoid robots are the small ones

In the first half of 2026, the world shipped almost 25,000 humanoid robots, up 432.1% year over year, according to IDC data reported by Yahoo Finance (IDC via Yahoo Finance). The vendor at the top of that leaderboard, Agibot in Shanghai, doesn't sell the towering industrial robots most people picture when they hear "humanoid." Its commercial product line targets roughly one-meter form factors, around half the height of the 1.6-to-1.7-meter industrial units that anchor most U.S. demos. Agibot shipped more than 8,600 units in H1 2026, about 35% of the global market, with Unitree in second at around 5,000 units (Smart Analytics Global; Humanoid Guide). Nine of the top ten vendors are Chinese, Chinese factories built more than 95% of every humanoid shipped globally, and Chinese customers bought about 78% of them.

The unit-count tables miss the form factor shift. Most of the vendors atop the H1 2026 leaderboard, including Agibot and Unitree, are explicitly targeting roughly one-meter-tall humanoids. A Seeking Alpha column flagged this trend as a contrarian investment angle (Seeking Alpha). The smaller form factor changes the addressable market in concrete ways.

Smaller humanoids are cheaper to build. Smaller actuators, lighter batteries, and reduced structural mass compound across the bill of materials in ways a redesign at full size cannot recover. They are safer to operate near people because a falling 30-kilogram robot does less damage than a falling 70-kilogram one, and a lower center of gravity reduces the chance of a tip-over in tight quarters. They also fit where the work actually lives: apartment hallways with 90-centimeter doors, restaurant kitchens with low clearances, retail back rooms with narrow aisles, warehouse lanes between racks, and the kinds of unstructured spaces a 1.7-meter industrial humanoid cannot physically enter. None of those are heroic environments, but they are the spaces where most service and residential labor actually happens.

Embodied AI's addressable market is gated by where the robot can stand, reach, and recover. A 1.7-meter humanoid can lift a heavier box but cannot pivot inside a 90-centimeter corridor; a one-meter unit can. Half-size humanoids trade payload and reach for accessibility, and for most service, retail, and residential use cases, accessibility is the binding constraint.

Half-size units carry less payload, reach less far, and remain constrained by current perception and manipulation stacks. A one-meter robot with a 5-kilogram arm is not going to unload a truck. Smaller form factors also read as less impressive to investors and policymakers who anchor their mental model on the towering demos of a few years ago, which can distort capital allocation away from where the units are actually shipping.

IDC's H1 2026 data splits the humanoid category in two. Full-size industrial humanoids target controlled factory environments. Roughly half-size humanoids target the much larger set of spaces where most people actually live and work. The shipment numbers say the second category is the one scaling first. Whether a one-meter robot can do enough useful work to justify its cost in those spaces will become clear over the next four quarters of deployment data.