A Wuhan startup called GigaAI has just become the first Chinese company to put humanoid robots inside ordinary households at any scale. Its SeeLight S1 model is now installed in 100 real homes for a multi-month trial, reportedly the largest at-home test of a general-purpose domestic humanoid to date. The robots arrived after less than a month of on-site training. In a demonstration apartment in Wuhan, two of them retrieved food, heated chicken in a microwave, cleared dishes, loaded the dishwasher, removed clothes from the dryer, folded them and put them away in a wardrobe.
That is impressive, and also, on close inspection, slow. Reports from the trial say organising a small stack of books takes the robot several minutes. Folding a single piece of clothing can take over ten. The S1 has reportedly struggled to hold cups without spilling. The point of the deployment is not to deliver a finished product. It is to harvest the messy, varied, unpredictable data that homes generate and factories do not, then feed it back into the next model.
The trial lands in the middle of an ongoing argument about whether China's humanoid boom has any commercial floor underneath it. Chinese makers shipped roughly 85 percent of all humanoids sold globally last year, by Barclays' estimate. The country's Ministry of Industry and Information Technology counted over 140 manufacturers and 330 models active in 2025, with Beijing publicly warning of a bubble. Production keeps climbing regardless: Unitree and AGIBOT each shipped more than 5,000 units last year, while US rivals Figure AI and Tesla each shipped a few hundred or less. Morgan Stanley expects Chinese humanoid shipments to roughly double this year to 28,000 units.
The question every analyst keeps returning to is who actually wants to buy them. State-owned enterprises are doing much of the early purchasing, for power plants, data centres, postal sorting and entertainment work. Coffee chains and hotels show up on the order books at Shanghai-based Matrix Robotics. But Samm Sacks at the New America think tank put the bear case bluntly: "humanoid robots remain expensive to produce, fragile in operation, and dependent on highly structured environments to function." That is exactly the opposite of what a household needs.
GigaAI's response is to attack the problem from the other end. Most humanoid demos rely on what the company calls the robot's "cerebellum," meaning whole-body coordination for impressive but rehearsed routines like backflips and dancing. Household work, the company argues, is a "brain" problem. The S1 uses what GigaAI describes as an embodied foundation model, processing spoken instructions, interpreting its surroundings, planning actions and adapting when furniture moves. The benchmark is no longer the choreographed demo. It is the unstructured chaos of someone's apartment.
The economic logic is real. Roland Berger estimates the operational cost of an advanced humanoid at roughly two US dollars an hour, set against US warehouse wages closer to 28 dollars. Citibank's math suggests a 25,000-dollar humanoid working a 16-hour shift could pay for itself in 36 weeks at the US minimum wage. Morgan Stanley expects the average price to fall from 46,000 dollars last year to around 21,000 by 2050, with some Chinese models already shipping below 6,000.
Even so, the test is no longer whether robots can run a demo. It is whether they can survive a Tuesday morning with kids, pets, a moved chair and a wet towel on the floor. GigaAI's 100-home pilot is the first serious attempt to find out, conducted at the scale only China can currently mobilise. The honest answer is that nobody yet knows what comes back from inside those homes, and the company will spend the next year mostly watching its robots struggle. That is the point of the deployment-first strategy, and also its largest risk.