China’s Humanoid Robots Face Their Hardest Test Yet: Can They Make Money?

China’s humanoid-robot industry has demonstrated that its machines can dance, fight and run. The harder question is whether they can perform useful work repeatedly, safely and cheaply enough to create more value than the labour, supervision and infrastructure they require.

That is the commercial test now facing the sector: not whether a robot can complete a spectacular routine once, but whether it can operate in an untidy workplace with limited human intervention and produce a credible return on investment. 1

The industry is shifting from spectacle to useful work

The 2026 World Robot Conference in Beijing highlighted that change. More than 300 companies displayed over 2,000 exhibits, including robots sorting parcels, packing mobile phones and assisting with household tasks.

The demonstrations are designed to answer a more practical set of questions. Can a robot handle unfamiliar objects? Can it maintain performance over a full shift? Can it detect and recover from mistakes? And can it complete a task without a human handler stepping in whenever conditions change?

The World Humanoid Robot Games made a similar shift. Alongside athletic events, competitors were tested in scenarios involving warehousing and packing, assembly and material feeding, retail and office services, electric-vehicle charging, and cable or tool use. Those tasks put sustained operation, dexterity, perception and autonomy under pressure in ways that a short, scripted performance does not. 1

Unitree’s IPO shows enthusiasm—but not proven demand

Investor appetite has moved even faster than commercial adoption. Unitree raised 6.1 billion yuan, or about $904 million, in its Shanghai IPO, becoming the first mainland-listed humanoid-robot maker, according to reporting on the offering. 3

When trading began, Unitree shares closed at 845 yuan, 460% above the 150.8-yuan IPO price, giving the company a valuation of about $50 billion. 2

That market response demonstrates confidence in China’s robotics ambitions and gives Unitree capital to expand. It does not, by itself, prove that customers are buying robots for profitable production. Unitree is profitable, but reporting has also noted that relatively few of its robots are used commercially and that many sales have been one-off purchases by research institutions. 2 4

The distinction matters. A strong IPO can fund manufacturing, software and deployment programs, but the long-term business case still depends on recurring customer revenue and measurable productivity gains.

The cost gap is still substantial

For a humanoid robot to replace a worker economically, its total cost must include more than the purchase price. Maintenance, deployment, charging, safety systems and human support all affect the calculation.

Guotai Securities estimated that an industrial humanoid, including maintenance, would need to cost about 160,000 yuan to pay for itself within two years against a worker earning 80,000 yuan annually. Reported real-world prices of comparable robots are closer to 300,000–500,000 yuan. 12

That gap does not make every deployment uneconomic: a robot may be valuable where work is dangerous, repetitive or difficult to staff. But it does mean that broad labour substitution requires either lower prices, better productivity, longer operating hours, or a combination of all three.

Reliability and hidden human labour may decide the outcome

Hardware performance is only one part of the problem. A commercially useful robot must function safely in changing environments, recover from errors and avoid requiring a specialist or remote operator at every failure.

Current machines can still struggle with basic tasks outside controlled settings, including pouring a glass of water or plugging in a cable. Unitree’s own prospectus acknowledged limited commercial applications for its quadruped and humanoid products, with scientific research and education accounting for much of its sales. 4

Customers must also provide the surrounding system: suitable facilities, redesigned workflows, charging arrangements, safety procedures, maintenance and service coverage. If a robot needs a human supervisor nearby for much of its operating time, its effective labour cost may be far higher than its sticker price suggests.

Some deployments may be for data, not productivity

Early factory and service installations could therefore serve two purposes. They may perform useful work, but they may also collect the real-world data needed to improve perception, dexterity and control systems.

That strategy can be commercially rational if the data helps a company build a better product later. It should not, however, be confused with proof that humanoid robots already form a self-sustaining labour market. A deployment that depends on silent human intervention may be an R&D expense rather than a productivity success.

The most meaningful evidence will come from repeatable results: hours worked, intervention rates, downtime, safety performance, maintenance costs and the value of completed tasks.

China has scale advantages, but global access is not guaranteed

China’s position gives its companies several advantages. Its large manufacturing base, component supply chains, policy support and domestic market provide opportunities to build, test and improve machines at scale. China is estimated to account for roughly 85% of the global humanoid-robot market.

That lead can create a valuable learning loop: more deployments can generate more operating data, which can improve products and lower costs. But market share is not the same as durable commercial leadership. The decisive advantage will belong to companies that turn manufacturing scale into reliable systems that customers can operate profitably.

Geopolitics adds another constraint. The U.S. Federal Communications Commission has restricted new imports of foreign-made advanced robots, including humanoid and quadruped systems, citing cybersecurity and national-security risks.

The immediate effect may be limited because adoption of humanoid robots in the United States remains low. Over time, though, the restriction could limit Chinese suppliers’ access to a major future market and encourage separate technology, supply-chain and standards ecosystems.

The metric that matters next

China’s humanoid-robot sector has already shown engineering speed, manufacturing capacity and investor appeal. Its next milestone is less cinematic: a robot completing valuable work for long periods, at a predictable cost, with few interventions.

The winners will not necessarily be the companies with the most impressive demonstrations. They will be the ones that can document dependable uptime, low supervision requirements and a return on investment that survives real workplace conditions.