It took ten years, but UBTECH finally managed to sell their robots.

Text | Rongzhong Finance

On the evening of March 31, 2025, UBTECH released a year-end report that made them proud.

Revenue of 2 billion yuan, up 53% year-over-year. For the first time, humanoid robots surpassed educational robots and consumer-grade robots to become the company’s largest source of income. A total of 1,079 units were delivered throughout the year, generating 821 million yuan in revenue, with a gross profit of 448 million yuan—an extremely impressive profit margin, causing the stock price to soar 17% the next day.

This company had been waiting a long time for this day.

At the end of 2023, founder Zhou Jian, along with Walker S, rang the Hong Kong Stock Exchange’s bell. It was a widely circulated scene—a person standing shoulder to shoulder with a robot at the exchange’s entrance. Zhou Jian said this symbolized “using robots to ring in a new chapter of human history.” The applause was enthusiastic, and the capital market’s imagination for this track had already been ignited.

But once the applause faded, reality was cold.

In its first year listed, UBTECH delivered only 3 full-sized humanoid robots throughout the year. Not 300, not 30, but 3. At that time, the company’s losses were still widening, and doubts were pouring in from all directions: after burning so much money over the years, who exactly are these robots being sold to? How long can this story go on?

No one knew the answer.

By 2025, those 3 units had become 1,079. A year-over-year increase of 35,866%.

The words Zhou Jian said back then now finally have numbers to respond to the doubts.

From 3 units to 1,079

To understand how important this year-end report is, we must first go back to 2024.

That year, UBTECH’s situation was somewhat awkward. With the halo of being the “first stock of humanoid robots,” its stock price fluctuated throughout the year after listing, and market doubts never ceased: after burning so much money, who exactly are these robots being sold to?

The answer was, very few units were actually sold.

In 2024, UBTECH delivered only 3 full-sized humanoid robots—this is inferred from the year-over-year growth figures disclosed in the 2025 report. A broader statement in the financial report is “a total of 10 units delivered,” which includes different models and sizes, but the true industrial-grade full-size standard products numbered just 3 units.

Behind these 3 units, at best, were highly customized “training machines,” accompanied by many engineers providing on-site services, used for factory “internships,” more akin to technical validation than commercial delivery. At that time, UBTECH’s humanoid robot revenue was 35 million yuan, with an average price of 3.5 million yuan per unit—buyers weren’t purchasing robots; they were buying a set of customized technical service packages.

This model was impossible to scale, everyone knew that.

So throughout 2024, UBTECH was doing one thing: sending robots into factories, running repeatedly, failing repeatedly, and iterating repeatedly. Names like BYD, Geely, FAW-Volkswagen, Audi FAW, and Foxconn appeared on UBTECH’s cooperation list, but behind the list was essentially “training,” not procurement. Robots stood on production lines, with efficiency only about 30% of workers, and when errors occurred, engineers monitored nearby. The entire process resembled a long product debugging session.

The outside world saw a company that was unable to commercialize for a long time, while UBTECH internally accumulated 22 months of real scene data.

This accumulation paid off in 2025.

In April, Dongfeng Liuzhou Motor officially signed a procurement contract, and Walker S1 units began mass production on automotive assembly lines—this was the world’s first mass production of humanoid robots in an automotive factory, not for display or training, but truly operational. Subsequently, orders flooded in: in July, winning a 90 million yuan project from Miyi Auto; in September, securing a single contract worth 250 million yuan, breaking global records; in November, the first batch of hundreds of Walker S2 units officially entered mass production and delivery, with a monthly capacity exceeding 300 units.

By the end of the year, 1,079 units, 821 million yuan in revenue, and 1.4 billion yuan in orders in hand.

From a custom service priced at 2B yuan per unit to a standardized product at 760k yuan per unit, prices dropped 80%, while volume increased by 358 times. This curve itself is a microcosm of humanoid robots moving from laboratory experiments to productization—when something begins to be mass-produced and sold in large quantities, it ceases to be just a prototype.

UBTECH took thirteen years to reach this point.

Who is buying those 1,079 units?

After discussing the quantity, it’s necessary to talk about money and where it came from.

1,079 units, 821 million yuan in revenue, averaging about 760k yuan per unit. This number is higher than many imagine—a single industrial humanoid robot costs more than a Porsche Cayenne.

Why so expensive?

UBTECH isn’t just selling hardware; it’s offering a complete solution that includes the robot itself, software systems, scene adaptation, and ongoing maintenance. Walker S2, standing 1.72 meters tall, weighing 76 kilograms, equipped with 41 servo joints, capable of handling tasks like material handling, sorting, and quality inspection, and can autonomously change batteries and operate 24/7. In real-world tests at Geely’s 5G smart factory, Walker S2 participated in battery module assembly, shortening batch production cycles by 15%. The pricing logic of this package is closer to industrial equipment procurement than consumer products.

But behind the 760k yuan average price, the customer structure warrants closer examination.

In 2025, UBTECH secured nearly 1.4 billion yuan in orders, with some of the most notable: in September, a single contract worth 250 million yuan, once the largest humanoid robot order globally; in November, winning a 264 million yuan project in Fangchenggang, Guangxi, again setting a record.

Looking at the buyers of these large orders, a pattern emerges: Liuzhou in Guangxi, Zigong in Sichuan, Jiujiang in Jiangxi, the Greater Bay Area in Huizhou—most of these projects’ main parties are “embodied intelligence data collection and testing centers” led by local governments.

This is a special procurement logic. When local governments buy these robots, it’s not to replace workers or cut labor costs, but to collect real scene data, build industrial demonstration bases, and respond to national embodied intelligence industry policies. The robots are mainly used for “training,” not “doing work.”

This is very different from automakers directly deploying robots on production lines.

Automaker orders do exist and are more valuable. Dongfeng Liuzhou, BYD, Geely, Foxconn—these collaborations represent the commercialization of humanoid robots in real industrial environments, which is the story UBTECH most wants to tell externally. But frankly, in the 1.4 billion yuan order pool, government-led data collection center projects account for a significant portion.

This isn’t unique to UBTECH. According to the Humanoid Robot Scenario Application Alliance’s statistics in the first half of 2025, among publicly known humanoid robot orders in China, the main purchasers include universities, research institutes, and demonstration centers, with actual market-based procurement from industrial end customers being relatively small. This reflects the true landscape of the industry in its first year of commercialization—policy-driven, market following, with a pace still lagging behind.

So, looking at these 1,079 units, it’s helpful to distinguish: the ones on automotive lines are part of commercialization; those in data collection centers are in the transitional phase of industrialization cultivation. Both are real deliveries, but their future order sustainability has different reference values.

Another number worth noting: currently, the efficiency of UBTECH humanoid robots in industrial handling scenes is about 30% to 40% of human workers. The official target is to increase to 50% by 2026 and to break through 80% by 2027.

This may mean that a 760k-yuan robot does less work than a monthly-paid worker earning 5,000 yuan. At this stage, customers are willing to pay for data accumulation, scene validation, and future efficiency expectations—not for immediate return on investment.

This logic was valid in the early industry stage, but it relies on a key premise: that efficiency must truly improve on schedule, and products must become cheaper and cheaper.

A comparison: Yushu Technology

After discussing UBTECH, it’s necessary to mention a company—Yushu Technology.

Not to compare who is better, but to put both together, which can better illustrate the current real situation of this industry.

Also in 2025, Yushu’s report card is: revenue of 1.7 billion yuan, net profit of 288 million yuan, nearly 600 million yuan after deducting non-recurring gains and losses. Sales of humanoid robots reached 5,500 units, with an average price of about 170k yuan per unit, and a gross profit margin of nearly 60%.

Placed side by side with UBTECH’s figures:

Revenue is similar (760k vs. 760k), but one lost 800 million yuan, and the other earned 600 million yuan. One sold at 3.5 million yuan per unit, the other at 170k yuan. One delivered 1,079 units, the other 5,500 units.

This contrast has sparked much discussion in the industry—some say Yushu proves humanoid robots can make money, others say UBTECH’s approach is off track. But it’s not that simple.

How does Yushu achieve a 60% gross profit margin? The prospectus reveals a detail: for humanoid robots like H1 and G1, hardware costs are around 60,000 to 70,000 yuan, but the selling price is 170k yuan. They control costs tightly, hold strong pricing power, and scale effects are already at play. Meanwhile, Yushu’s expense structure is very lean—R&D expenses in the first three quarters of 2025 were only 7.7%, and sales expenses 6.5%, with almost no marketing efforts, mainly relying on brand organic traffic and reputation.

This model yields a net profit margin close to Wuliangye’s, which is quite unusual in a generally loss-making track.

But Yushu also faces a problem: its customers are mainly universities and research institutes, with few cases of entering industrial production lines and replacing workers—most of the robots sold are used for research, secondary development, and technical demonstrations—essentially selling a “high-performance hardware development platform,” not an industrial solution.

Thus, the two companies represent two different business models, each with its own boundaries:

UBTECH pursues high unit prices, heavy services, and deep industrial customer engagement—each order is substantial, with long delivery cycles and high service costs, but once established in automakers, the moat is deep. Yushu pursues low unit prices, light services, and volume-driven market expansion—high gross margins, good cash flow, but customer base mainly in research, still far from industrial substitution.

One resembles an early industrial software company; the other, an early DJI.

Neither has reached the end of the road nor been disproven. But one thing is certain: regardless of the model, the ultimate question remains—can this robot truly replace a worker, and is it a worthwhile investment?

In 2025, no one can fully answer this question.

Finally: 2026 is the real test

UBTECH set a goal in its annual report: to increase industrial humanoid robot annual production capacity to 10,000 units—almost a tenfold increase. Whether this can be achieved depends on several factors happening simultaneously: ramping up capacity, maintaining quality yield, coordinating supply chains, and most importantly, continuous order inflow. Of the 1.4 billion yuan in orders in 2025, how many are one-time policy window purchases, and how many will turn into stable repeat customers? The answer will become clearer in 2026.

Another efficiency curve needs to be realized: from 30% to 50%. It may seem just a numerical change, but behind it are breakthroughs in algorithm iteration, scene data accumulation, hardware reliability improvements, and a series of engineering challenges. UBTECH has accumulated two years of real scene data in factories like Geely, BYD, and Dongfeng Liuzhou, which is one of its most valuable assets. But asset realization takes time.

Pricing is also a hurdle. The current average price of 760k yuan supports the current revenue scale, but this price cannot open up a larger market. Industry consensus suggests humanoid robot prices need to drop to 200,000–300k yuan to truly trigger large-scale industrial substitution. This means UBTECH must continue to reduce costs while scaling up—doing both simultaneously is a huge test for supply chains and manufacturing systems.

Money-wise, there’s no need to worry for now. With 4.8 billion yuan in cash and strong financing ability in the Hong Kong stock market, UBTECH has enough room for trial and error. But capital’s patience is limited. Whether burning money to buy time can lead to real large-scale delivery in 2026 remains the question everyone is waiting for.

Ten years of sharpening a sword, and it’s already unsheathed.

But whether the sword is sharp enough remains to be seen in another year.