China's Humanoid Robots Are Profitable. Korea Is Still Counting Screws.

The Numbers That Should Worry Seoul

UBTech reported a 66.8% gross profit margin on its humanoid robot business in the first half of this year. That is not a prototype studio running on venture capital. That is a company shipping thousands of units out of a factory in Liuzhou, watching per-unit costs collapse as volume climbs.

Humanoid sales grew 15.4 times year over year. Units sold grew more than 20 times. The business, which accounted for less than 3% of UBTech’s revenue just two years ago, now generates nearly half of the company’s total revenue. UBTech is still net-loss-making at the company level — H1 loss narrowed 23% to 339 million yuan — but analysts say adjusted EBITDA could flip positive as early as the second half, with net profit coming next year.

Unitree tells the same story from a different angle. Its average selling price dropped 56% in 2024 alone and another 36% the year before. Gross margin, meanwhile, climbed from 43.6% to 60.3%. The mechanism is blunt: Unitree manufactures its own motors and actuators, the most expensive components in a humanoid robot, and pushed annual sales above 5,000 units. Every additional unit drives the fixed cost per unit lower. Price cuts become possible without cutting into margins.

What makes these numbers significant is their directionality. They are not static snapshots of companies sitting at some equilibrium — they are trajectories moving away from it. Chinese humanoid robot makers are not approaching profitability. They are already inside it, on the core product line, at scale.

Who Controls the Market Right Now

Counterpoint Research puts global humanoid shipments at over 22,000 units in H1 — triple the year-ago period. The top five sellers are all Chinese. AegisBot led with roughly 9,700 units. Unitree followed with about 7,000. Galbot, UBTech, and Lightning Robot completed the list. Together they command 86% of global shipments.

The remaining 14% is fragmented across Western and Japanese firms that are still largely in demonstration mode or early commercial deployments. The gap is not incremental. It is structural.

AegisBot’s dominance is particularly striking. The company, backed by investors including Alibaba, has moved aggressively into manufacturing capacity, opening a 100,000-square-meter facility in Guangzhou designed for mass production. Its robot, the ZhiYuan, has found early traction in warehouse and logistics applications where the value proposition is clearest: a machine that can navigate human-designed environments and perform tasks that rigid industrial robots cannot.

Unitree, by contrast, has competed on price. Its H1 series was marketed at under $9,000 for the consumer-facing model and under $16,000 for the more capable G1. Those prices forced the industry to confront a question that had been lurking: how much of a humanoid robot’s cost is genuinely structural, and how much is just the inefficiency of low-volume manufacturing? Unitree’s answer is that most of it is the latter.

Korea’s Pilot Trap

South Korea’s situation reads like a checklist of what not to do when building an export-intensive hardware industry.

Rainbow Robotics reported 865 million won in H1 revenue from its wheeled humanoid robot but can only produce around 120 units per year. The company posted an operating loss of 359 million won, up 3.9% from a year earlier. It is generating revenue, yes, but at a scale that cannot compete with Chinese manufacturers driving costs down through volume.

Robozee turned profitable last year, which sounds strong until you read the breakdown. Of its 27.2 billion won H1 revenue, 98% came from actuator components — the parts inside humanoid robots, not the robots themselves. Wheeled humanoid revenue accounted for just 2%. Robozee is arguably the strongest case among Korean makers: it plans to begin mass-producing its bipedal AI Sapiens K1 next year at a target of 1,000 units annually. But 1,000 units is a fraction of what UBTech and Unitree move in a single month.

Rostlos, which sells its bipedal Igris-C robot, has produced 36 units to date and sold 18 of them. That is not a business. That is a workshop.

Samsung Electronics has also entered the humanoid space with ambitions announced at its Samsung Next division, targeting 10,000 units by 2026. But Samsung’s entry follows the same pattern that has defined its historical weakness in this domain: internal capability without external velocity. The company is developing in-house but has no production track record comparable to the Chinese operators. Its 10,000-unit target, if achieved, would still be dwarfed by what UBTech shipped in the first half of this year alone.

An industry executive put it plainly: Chinese companies are building finished robots by the thousands, using volume to drive down component costs, refine assembly lines, and accumulate real-world operational data simultaneously. Korean companies are strong on components but lack experience scaling whole-body production. The question is how quickly they can convert pilot plans into actual sales and cost reductions.

The second-order effect of this dynamic is already visible. Korean component suppliers — including companies that produce the very actuators Robozee sells — are beginning to face a dilemma. Their largest and fastest-growing customer may be the Chinese robot manufacturers they supply. Building for Chinese buyers reinforces Chinese scale, which in turn makes it harder for Korean system integrators to compete. The supplier becomes the competitor’s ally without any intention of being one.

Why This Matters Beyond Asia

Humanoid robots sit at the intersection of three strategic industries: AI software, advanced manufacturing, and export-grade hardware. China is positioning itself to dominate all three by using robotics as the application layer that validates its AI stacks and consumes its component supply chain on the way to global markets.

The margin dynamics favor the volume player. Once a company crosses the threshold where fixed costs are spread across thousands of units rather than hundreds, the economics shift permanently. Chinese firms are entering that zone now. Korean firms are still walking toward it.

The component advantage Korea possesses — actuators, sensors, precision gears — is real but fragile. Component suppliers earn steady margins but do not capture the system-level value that accrues to the company integrating those parts into a complete robot. If Korean firms remain actuator vendors rather than becoming robot builders at scale, they will see their margins compressed by the very Chinese companies they supply.

There is also a geopolitical dimension that extends beyond the bilateral rivalry. Western governments have been quietly alarmed by China’s speed in commercial robotics. The same concerns that drove semiconductor export controls are now applying, in softened form, to dual-use AI hardware. Korean companies could find themselves caught between a China that wants them as suppliers and a West that does not want them as alternatives to Chinese domination of the category.

What Comes Next

The next twelve months will determine whether Korean humanoid robotics remains a component story or becomes a system story. Robozee’s 1,000-unit annual target for its AI Sapiens K1 is the closest any Korean company has come to a meaningful production commitment. If Robozee delivers on that timeline and sees actual shipment volumes climb, it could prove the model works. If it stalls — as pilot programs often do — the gap will widen further.

UBTech and Unitree are not waiting. Their factories are running. Their margins are expanding. Their component supply chains are locking in through vertical integration. Every month of delay compounds the cost disadvantage for late entrants. Unitree’s price cuts force competitors to choose between matching them and losing market share, or holding prices and accepting thinner margins — both bad outcomes.

The export war over AI hardware has not been declared in most Western coverage. It is already underway in Liuzhou, Shenzhen, and Hangzhou. Korea’s challenge is not whether it can build a competent humanoid robot. It can. The challenge is whether it can build one fast enough, at scale enough, and cheap enough to matter — and whether its suppliers will let themselves become the scaffolding for someone else’s victory.