The human-shaped world argument
Factories, warehouses, offices, and homes were all designed around human bodies: doorways sized for human shoulders, shelves at human reach height, stairs sized for human legs, tools gripped by human hands, pedals and steering wheels built for human limbs. The core bet behind humanoid robotics is that a robot shaped roughly like a person can walk into any of these spaces and use the infrastructure that already exists, without anyone having to re-engineer a building, a vehicle, or a workflow around the robot. That is a genuinely appealing pitch to investors and operators, because retrofitting physical environments for a non-human robot design is slow and expensive at scale.
There's a second piece to the bet: generality. A wheeled cart robot or a fixed robot arm is usually built to do one class of task well. A humanoid, in theory, could be trained to do many different tasks across many different sites, because it has hands, arms, legs, and a torso capable of the same broad repertoire of motion humans use for nearly everything. If that generality can actually be achieved, a single hardware platform could in principle be sold into many industries instead of needing a custom robot design for every job.