Dialogue with Gudi Robot's Hu Chunxu: Computing Power Must Balance Applications, Not Just Be the Largest; Continuous Effort for a Future Era of Robots Counted in 'Hundreds of Millions'

On November 21, the DDC 2025 DiGua Robotics Developer Conference was held in Shenzhen. Leading enterprises and developers from fields such as AI and robotics gathered to discuss the development of the Embodied AI industry.

On the day of the conference, Hu Chunxu, Vice President of Developer Ecosystem at DiGua Robotics, shared the company's ecosystem layout and operational progress, and accepted interviews from multiple media outlets including 42 Hertz.

He noted that some key changes have occurred in the robotics industry over the past two years: more companies are increasingly focusing on consumer robotics, while other large humanoid robots are exploring how to implement applications in more scenarios. Consequently, a large number of new companies and technologies emerge daily.

Currently, AI Infra (AI infrastructure) is undergoing large-scale construction worldwide. Comparing AI Infra with robotics Infra, Hu Chunxu believes the biggest difference lies in "physics." The terminal application scenarios for the two differ significantly. When mentioning the latter, people first think of various types of robots such as humanoid and quadruped models, rather than model algorithms.

However, the integration of hardware and software is the core method driving the continuous evolution of robotic intelligence. In this regard, Hu Chunxu stated that DiGua Robotics has designed the BPU intelligent computing architecture as an AI algorithm acceleration engine for robots, enabling algorithm acceleration within chips to provide high computational density and processing efficiency.

In terms of commercialization, in the early stages of the Embodied AI industry's development, the entire sector is still exploring the balance between R&D investment and commercial profitability.

In this industry context, Hu Chunxu introduced that there are currently mass-produced products, such as robot vacuums and robotic lawn mowers, which constitute the basic business areas of Digua Robotics.

Moreover, compared to last year, the number of customers of this type has doubled. The mass production of their products ensures that DiGua's basic business will grow rapidly. On this basis, DiGua will be able to devote more energy to ecosystem building.

Overall, Hu Chunxu stated that Digua Robot adopts a two-pronged strategy: one leg ensures the foundational business, while the other explores future growth, namely the potential era of robots measured in "hundreds of millions".

After this conversation with Hu Chunxu, we can actually gain a clearer understanding of Digua Robot's industry positioning and developer ecosystem.

For this dialogue, No. 42 Radio edited the content without changing the original meaning. The following content is for reference only.

Q: For the humanoid robot field, what is the core demand for computing power chips currently?

A: The technology for humanoid robots has not yet converged, and there is no definitive conclusion on the exact computing power requirements from various manufacturers, including ourselves. Therefore, this is a matter that we are all exploring together.

Our judgment is that computing power is not necessarily better when it is larger. There is an impossible triangle of cost, power consumption, and performance; all three cannot be maximized simultaneously. The current situation is that the higher the computing power, the higher the power consumption and the higher the cost. Therefore, this determines that the selling price of future humanoid robots will also be higher. Higher selling prices may lead to a smaller market. Therefore, a balanced compromise must be chosen in between.

So as you can see, the S600 we released today has 560 TOPS (INT8) computing power. At this stage, we believe that 560 TOPS, or a level of 500-600 TOPS, is the optimal choice for implementation in the next 3 to 5 years.

Q: What key changes have occurred in the robotics industry over the past two years?

A: In my personal view, there are two profound directions of change. The first is in consumer scenarios, where numerous companies are exploring various products, such as companion robots and wearable devices. However, no scenario has yet confirmed a large-scale landing opportunity.

The second change involves embodied AI, specifically full-size, medium-size, and wheeled-arm morphologies of embodied robots. People are exploring which scenarios embodied robots can land in and what the model size and computing power requirements are. Therefore, many technologies in this area have not yet converged.

Consumer scenario technologies are relatively converged, but their application scenarios are divergent. These two changes are very intuitive to us and iterate rapidly, with many new companies and technologies emerging every day.

Q: What are the differences and connections between AI Infra (AI infrastructure) and Robotics Infra?

A: The biggest difference lies in physics. The AI Infra we see more often consists of models and algorithms currently in use, which are relatively software-oriented. Unlike robots, when people think of robots, they do not immediately envision a specific algorithm, but rather images of humanoid robots, companion robots, robot vacuums, and so on.

These two infrastructures correspond to terminal application scenarios that are vastly different. There are many forms of robots, algorithms, and scenarios, whereas AI Infra is relatively more convergent. For instance, large language models focus mainly on language interaction functions, but they lack a clear physical carrier. Therefore, I believe this is the biggest difference point, which in turn leads to differences in all toolchains, ecosystem operations, and company structures. Moreover, AI Infra can be considered a subset of robot infra, providing significant computing power support for it.

Q: How does Sweet Potato Robot balance the high investment in technological R&D with the short-term goal of market profitability?

A: The entire industry is currently still in a relatively early stage, but there are already products achieving mass production, such as robotic vacuums, lawn mowers, and pool cleaners. As you can see in the exhibition area, there are many lawn mowers and robotic vacuums; these constitute our basic business scope.

The number of customers has doubled compared to last year. The mass production of their products ensures that Sweet Potato's core business continues to grow rapidly, while also allowing me to invest more energy and capital into ecosystem construction. From an industry perspective, we believe that ten years from now will undoubtedly be the era of robots, where the number of robots could reach hundreds of millions, billions, or even tens of billions.

So, how do we tap into these scales of billions and tens of billions? I believe the opportunities lie within ecosystem creativity. This is the significance of our ecosystem efforts, and we must adhere to long-termism. After sustained long-term investment, we will reap higher returns in the future. In short, Sweet Potato follows a two-pronged approach: one leg ensures the core business, while the other explores future growth.

Q: In terms of commercial implementation, which areas does Sweet Potato Robot's ecosystem focus on?

A: We divide the current robot market into three main segments. The first is the relatively mature market for robotic vacuums and lawn mowers. The second includes innovative consumer products, such as companion robots. The third covers categories like humanoid robots and large quadruped robots. Among these three major segments, robotic vacuums and lawn mowers are relatively mature and form the core business of Sweet Potato Robot. We will steadily secure our market share in this segment and even achieve continued growth.

The other two parts are relatively less converged, so we will make significant investments in the ecosystem in these two areas, including startups and even laying out plans starting from the primary school stage. We believe that the ideas of primary school students could potentially become reality in the future; thus, this is a fundamental direction of our ecosystem.

Q: Does Diwa Robot have a specific focus in its strategy for supporting or assisting developers?

A: It essentially divides into two areas. The first focuses on commercial implementation, driven by commercial value, such as lawn mowers and robotic vacuums. The second involves companionship robots, new-form robots, and embodied AI robots. In terms of the ecosystem, we do not use commercial value as the core guiding principle. Instead, we generally evaluate several dimensions.

On one hand, it resembles venture capital investment; I invest my products, human resources, and ecological resources, so I also need to assess their future growth potential and changes. On the other hand, I look at how much they embrace Diwa. If their embrace level is low, it is difficult for me to support them. Next, I examine the team's technical control and capabilities, including whether their judgment is unique and accurate, which is very similar to investment.

Additionally, I look at whether the company's products are sufficiently innovative. We not only judge their commercial value but, more importantly, assess their overall future growth potential.

Q: Are there any emerging fields currently visible that could be the next big area like robotic vacuums?

A: To be honest, it is currently impossible to determine exactly what the next 'robotic vacuum' equivalent will be. Our current logic is to provide support to developers based on assessments of overall future growth potential.

We have encountered many bold ideas during this process. Although they may seem baffling at present, we believe the future is truly exciting and these concepts may evolve.

Q: Many robot manufacturers have announced mass production plans for the coming year or two in the second half of this year. How does Digua Robot plan to seize the opportunity for increased volume and capture a larger market share?

A: We pursue this through our core business fundamentals. We have established a complete system for tracking commercial leads, supported by dedicated personnel to achieve breakthroughs on a point-to-point basis. Another crucial approach is leveraging ecosystems. The conference serves as a platform to embed our ecosystem—spanning from hardware products to platforms—into people's minds, demonstrating our ongoing efforts. We can reach over 1,000 companies in the market, effectively covering most of them, while we continue to closely monitor many startups.

*Effective TOPS with 1/2 Sparsity, TPP(Total Processing Performance)<4800