The R&D cost of Unitree is unexpectedly far lower than that of LEGO.

Looking through Unitree Robotics' prospectus, there is a number that feels quite counterintuitive.

In 2025, Unitree Robotics' R&D expenditure reached 145 million yuan. Back in 2024, the figure was merely 70.02 million yuan. Although it more than doubled year-on-year in 2025, the number is still surprisingly low when compared with large global corporations.

Take the Lego Group as an example.

Lego's R&D expenditure in 2025 hit 2.3 billion Danish kroner. Converted at a roughly equivalent exchange rate against RMB, that amounts to over 2 billion yuan, with the absolute value roughly more than ten times that of Unitree. In 2024, Lego's R&D spending was even higher, reaching 2.476 billion Danish kroner, while Unitree's R&D expenditure that year was only 70 million yuan, creating a gap of more than 30 times between the two sides.

Lego's annual report clearly defines this caliber: Other operating expenses basically refer to research and development expenses, including salaries, consumables and external R&D costs.

Thus a very interesting scenario emerges:

A rising Chinese tech star that is developing humanoid robots, quadruped robots and embodied intelligence large models spends far less on R&D in a single year than a company that "sells building blocks".

Of course, this comparison cannot simply lead to the conclusion that "Unitree does not attach importance to R&D".

In 2025, the proportion of Unitree's R&D expenditure in its revenue was 8.53%, while that of Lego was only about 2.8%. Unitree's R&D expense ratio was even as high as 31.39% in 2023, and stood at 17.83% in 2024.

The reason why the ratio suddenly dropped to 8.53% in 2025 is that its revenue grew extremely fast: Unitree's 2025 revenue has reached 1.699 billion yuan, while its R&D expenditure increased by 107% year-on-year.

01

What is truly worth observing is another matter:

Unitree is likely one of the companies with the most exaggerated R&D efficiency in the global technology industry at present.

As of September 2025, Unitree had only 480 employees in total, 175 of whom were R&D personnel. With an R&D team of less than 200 people, it has developed a full range of robot products including the Go series robotic dogs, B-series industrial quadruped robots, G1, H1 and other models.

If you break down the 145 million yuan R&D expenditure, you will find that it is even somewhat "modest".

85.04 million yuan goes to salaries of R&D personnel, 28.97 million yuan to direct materials, 15.15 million yuan to cloud computing power and cloud services, and 7.78 million yuan to depreciation and amortization.

In other words, the core competitiveness that Unitree has built over the years is not the typical "large factory-style R&D".

It has no research institute with thousands of employees, no annual GPU budget of billions of yuan, nor does it burn money wildly on model training like OpenAI and Anthropic.

It is more like a company with an extreme engineer culture.

It integrates all capabilities including motors, reducers, control algorithms, structural design, motion control, supply chain and mass production within a very small organization, and iterates products at an extremely high speed.

This also explains why the Unitree case easily makes the public, including people in Silicon Valley, feel uncomfortable.

A large number of basic researches in the robotics field are the results of decades of joint accumulation by the global academic circle. Papers, algorithms and engineering experience of projects such as MIT Mini Cheetah have been publicly available for a long time. Unitree did not invent all robot technologies from scratch. Its outstanding advantage is that it combines public knowledge, China's supply chain and its own engineering capabilities, and quickly achieves low cost, high reliability and mass production.

When tracing the history of quadruped robot technology, we can see this kind of technology diffusion from the U.S. scientific research system to China's commercialization system.

Therefore, it may not be accurate to simply regard Unitree as a "robot scientific research company". Nor is it a toy company like Lego.

02

It is more like Xiaomi in the robotics era. The logic makes perfect sense when you understand it this way.

No wonder Lei Jun's Shunwei Capital invested in Unitree in its early days.

Unitree knows what must be developed independently, what can be purchased, what academic achievements can be directly transformed into engineering applications, and finally how to reduce the cost of a robot that originally cost 1 million U.S. dollars to tens of thousands of U.S. dollars.

Looking back at Lego from another perspective, you will find that the phenomenon that "a toy company has higher R&D expenditure than a robotics company" is not that absurd.

In 2025, Lego's revenue reached 83.5 billion Danish kroner, with more than 33,000 employees worldwide, and it launched 868 new products in the year. Continuous R&D investment is required in fields including material science, molds, automated production, digital products, games, software, child safety and environmentally friendly materials.

The Danish team is very rigorous in their work.

One is a global consumer goods industrial system that has been operating for decades, and the other is a robotics startup with less than 500 employees.

Up to now, the humanoid robot industry has unexpectedly produced a world top-tier company that only spends 145 million yuan on R&D per year.

However, this state may not last for long.

There is a notable small figure in Unitree's prospectus: in 2024, its cloud computing power and cloud service expenses were only 1.23 million yuan; in 2025, the number suddenly rose to 15.15 million yuan, a surge of more than 10 times in one year. The reason is written very clearly — the company has begun to strengthen technical R&D in the "large model direction".

This may be the real watershed for Unitree.

The core of robotics competition in the past decade has been mechanical, electronic control, motion control and supply chain, which are typical engineering problems.

But if the future competition shifts to VLA models, world models, foundational robot models, massive real machine data and reinforcement learning, then its competitors will no longer be only Boston Dynamics, but also Tesla, Google DeepMind, and a growing number of AI companies with huge computing power budgets.

At that time, whether 145 million yuan of R&D expenditure can still create miracles like today will be the most noteworthy issue after Unitree goes public.

After truly entering mature industrial competition, R&D is an extremely costly endeavor.

The most incredible thing about Unitree is that it has not been that costly so far.

This is also the point that has been widely criticized by stock investors.

This article is from the WeChat official account "Wu Duidui" (ID: esnql520), written by Wu Duidui, and published with authorization from 36Kr.