UBTECH’s Humanoid-Robot Bet Moves From Shipments to Revenue and Repeat Factory Orders

UBTECH used the 2026 World Robot Conference in Beijing to make a commercial argument: humanoid robots should increasingly be judged by the quality of the business they create, not simply by how many units manufacturers claim to ship.

That means tracking revenue, gross margin, progress toward profitability and—above all—whether industrial customers place repeat orders after initial trials. The shift matters because UBTECH’s results show both sides of the story: its humanoid business is becoming a meaningful source of revenue, but the company remains loss-making.

From robot counts to commercial proof

The conference ran from August 19 to 23 under a broader industry push toward practical deployment. Organizers said more than 300 companies attended, displaying more than 2,000 exhibits and launching over 150 products. 1

Against that backdrop, UBTECH presented its robots less as isolated demonstrations and more as systems intended for production environments. Its industrial Walker robots appeared in a 1:1 reproduction of a customer production line, with coordinated workflows including automotive sheet-metal handling, palletizing and logistics sorting. 3

The commercial question is not whether a humanoid can complete a carefully prepared stage routine. It is whether the robot can execute tasks repeatedly, cope with variation, recover from errors and integrate safely into an operating factory. A replicated customer line is therefore a more useful signal than a simple shipment tally—although a demonstration alone does not prove long-term production performance.

What UBTECH’s 2025 figures show

UBTECH’s 2025 annual results provide the financial foundation for its new narrative:

• Total revenue reached RMB2.001 billion, up 53.3% year on year.
• Revenue from full-size embodied-intelligence humanoid products and services rose from RMB35.6 million in 2024 to RMB820.6 million in 2025, an increase of about 2,203.7%.
• The full-size humanoid category accounted for 41.1% of total revenue and included 1,079 units sold.
• Group gross margin increased to 37.7%.
• Net loss narrowed by 31.9% to about RMB789.8 million.

The numbers support a substantial change in revenue mix. Full-size humanoids became UBTECH’s largest reported revenue stream, rather than a small experimental business. But the company was still not profitable. The more cautious interpretation is that UBTECH has demonstrated revenue growth and improving gross economics, not completed the transition to a self-sustaining humanoid business.

The U1 added a consumer and companion-robot angle

UBTECH’s industrial displays were paired with the UWORLD U1, a hyper-realistic humanoid aimed at interaction and emotional companionship rather than factory automation. Reports described lifelike silicone skin and facial micro-expressions, with potential applications including emotional-support services and elderly care. 11

The company said cumulative U1 orders across channels had exceeded 13,361 units. Another report described the total as more than 13,000 bookings and said initial deliveries were planned for September 16. 815

Those figures need careful reading. Channel bookings are not necessarily end-user deliveries, and they do not automatically represent recognized revenue. Pricing reports also vary by product version: one report cited a price of about RMB99,800, while another identified a consumer U1 Pro variant at RMB169,800. 89 The safest conclusion is that U1 demand was significant at the order-book level, while the conversion of those bookings into deliveries, revenue and repeat purchases remained an open question.

Why factory reliability is the harder test

Humanoid robots face a more difficult deployment problem than conventional automation because they must operate in environments designed for people, where objects, layouts and workflows can change.

For an industrial customer, useful performance includes more than a high success rate on a fixed demonstration. It includes sustained operation, tolerance for task variation, safe interaction with equipment and workers, recovery when an action fails, and compatibility with existing production systems.

That is why UBTECH’s emphasis on customer-line replication and recurring industrial orders is strategically important. An initial pilot can show technical promise. Repeat orders suggest that a customer found enough operational and economic value to continue buying or expanding the deployment. They are still not proof of profitability, but they are stronger evidence of product-market fit than publicity-driven shipment totals.

Embodied AI’s data bottleneck

UBTECH also framed real-world data as a central constraint for embodied AI. Physical-task data are slower and more expensive to collect than internet-scale language or image data, and factories differ substantially from one another. A model trained in one environment may not generalize cleanly to another.

The company’s proposed response combines data from deployed robots with simulation, world modeling and on-device inference. The goal is to make a robot’s learned behavior more transferable across tasks and sites while reducing the computing burden required to run the system locally.

UBTECH described a three-layer Thinker architecture: the Thinker perception model, the Thinker-WM world model and the Thinker-VLA execution layer. The company said on-device optimization improved inference efficiency by 176% and reduced GPU-memory requirements from 64GB to 32GB, while moving computation from x86 systems to lower-power ARM boards. These are company-reported performance claims rather than independently audited benchmarks.

The MetaX chip partnership

UBTECH is also trying to control more of the hardware stack. It partnered with MetaX to establish Xixuan Chuangzhi Technology (Wuxi), a joint venture focused on researching and mass-producing chips for embodied intelligence. Public filings cited by Gasgoo said the venture had registered capital of RMB100 million, with MetaX and UBTECH each holding 35.01%.

The strategic objective is local, more efficient inference for robots: lower-cost computing, reduced power requirements and less reliance on general-purpose accelerators. The partnership does not remove the technical and manufacturing risks involved in developing a new chip platform, but it shows that UBTECH views edge computing as part of the product’s economics—not merely a supplier decision.

A fast-growing market with disputed shipment totals

UBTECH’s argument arrives as China’s humanoid-robot industry expands rapidly, but the market data are not consistent.

One Smart Analytics Global estimate put global H1 2026 humanoid shipments at approximately 19,100 units, up 272% from 5,100 in the same period a year earlier. It estimated that Chinese manufacturers accounted for more than 18,500 units, or over 97% of the global total, and projected close to 60,000 units worldwide for the full year.

Other figures released around the conference were much higher: an industry report said China alone shipped more than 40,000 units in the first half of 2026. Counterpoint Research was also reported to have counted more than 22,000 global H1 shipments.

These estimates cannot all describe the same scope or methodology. They may differ in how they define a humanoid robot, count shipments and distinguish prototypes, trials, channel inventory or commercial deliveries. The disagreement is itself a reason not to treat any single shipment number as a definitive market census.

The next metric is whether growth becomes profitable

UBTECH’s interim 2026 results sharpen the issue. First-half revenue reached RMB1.269 billion, up 104.2% year on year, while gross margin rose to 44.7%. Full-size humanoid revenue increased 1,445% to RMB590.3 million on sales of 921 units. Yet the company still recorded a RMB339 million net loss. 714

That combination explains the change in emphasis at WRC. Shipments can establish manufacturing scale and market presence, but they do not show whether each deployment creates attractive economics. Revenue growth, gross margin and narrowing losses are more informative, while repeat industrial orders are the clearest available test of whether customers find the robots useful enough to keep buying.

UBTECH’s conference message was therefore not that shipment counts no longer matter. It was that they are no longer sufficient. The decisive question for the company—and for the humanoid sector more broadly—is whether demonstrations become dependable factory work, bookings become delivered products, and rising revenue eventually becomes durable profit.