Recently, General Catering Equipment Robot Company Yingzhi XBOT announced the completion of two consecutive rounds of financing worth hundreds of millions of yuan. Among them, the Series A financing of 200 million yuan was contributed by Hong Kong Jiankun Capital GPTX; The B-round financing is 300 million to 500 million yuan, jointly contributed by multiple government funds, US dollar funds, and industry investors.
The helmsman Tang Mu is a true product veteran. In 2003, he joined Tencent and led the establishment of the Tencent User Research and Experience Design Center (CDC) over the next decade. He personally created classic QQ avatars and is also known as the "father of QQ avatars" by the media.
In 2013, Tang Mu chose to join Xiaomi, which was in a period of rapid growth, completing a crucial leap from software interaction to intelligent hardware. At Xiaomi, he served as the General Manager of the Intelligent Products Department and Vice President of the Ecological Chain, leading the development of two billion dollar product lines, Xiaomi Router and Xiaoai Smart Speaker, from scratch.
In 2022, Tang Mu left the Xiaomi system to start an independent business and founded Beijing Yingzhi Technology Co., Ltd. in November. His core judgment is to use the concept of consumer electronics products to create a universal catering robot, transforming the robot from an "exhibit" in the laboratory into a mass-produced and profitable "product" in commercial scenarios.
The team configuration of Yingzhi XBOT is different from typical robotics companies, more like a combination of a mature consumer goods company and an AI company. The robotics technology sector is led by Wang Jiali, a Ph.D. in mechanical and electrical engineering from Harbin Institute of Technology who has served as a senior executive in companies such as Aerospace Science and Technology, Sany Group, and Xinsong Robotics. The operation sector is composed of the Luckin team, including Cao Ruikun, Yu Tao, and others who have been responsible for the operation of large-scale store chains and the establishment of brand standardization systems in the region.
At the beginning of his entrepreneurship, Tang Mu received angel round investments from friends such as Zhang Xiaolong, Senior Vice President of Tencent, and Li Wanqiang, co-founder of Xiaomi. As of now, the coffee robots of Yingzhi XBOT have been deployed in over 100 cities worldwide, producing over 4 million cups of coffee. By 2025, the revenue has exceeded 100 million yuan.
Tang Mu revealed that the company's order amount in 2026 has approached 300 to 500 million yuan.
The following is the full text of the dialogue, with slight deletions:
Zhike ZhiKer: What fundamental questions did you consider when deciding to start a business in 2022?
Tang Mu: At that time, I pondered three questions repeatedly.
The first question is whether to innovate or follow. In China, innovation is particularly difficult, and it's not easy to come up with an idea that big companies will kill you on the beach as soon as they follow up. But completely following is equivalent to having no chance at all. So I chose innovation.
The second question: general or specific. Everyone is making humanoid robots, but it is very difficult to industrialize a fully intelligent and versatile humanoid robot. Rarely in history has a product been completely universal, you can't turn a refrigerator into a washing machine and an oven. So I chose to specialize.
The third question: To VC or To C. To VC can quickly get money, but that's short money, fast money, or it could be debt. So I choose To C, even if it's a bit difficult at the beginning, as long as the product meets real needs in real scenarios, it can go far.
Zhike ZhiKer: Why did you ultimately choose the coffee robot category?
Tang Mu: Firstly, food is of paramount importance to the people. Regardless of the economic situation, people must eat and drink. Coffee is a rapidly growing market in China, and even with Luckin Coffee and Kudi, the per capita annual coffee consumption in China is still much lower than in developed countries.
Secondly, there is an opportunity for an efficiency revolution. The barista works from morning till night, without even having time to go to the bathroom, and has to face various demands from customers. We hope to use robots to liberate these people, not replace them.
Thirdly, and most importantly, the standardization level of coffee is sufficiently high. The variables of a latte are relatively controllable, including extraction time, milk froth temperature, and latte action, which can be structured and digitized. Compared to Chinese stir frying, coffee is more suitable for robots to enter first.
Zhike ZhiKer: Do you want to replace baristas?
Tang Mu: Our customers are mainly non coffee chain cross industry customers, including luxury stores, home furnishing stores, 4S stores, and cultural and museum venues. These scenarios originally did not have coffee services, or coffee was just a supporting service. What they need is to use coffee to retain customers, not to make money with coffee.
In addition, the clients of Yingzhi XBOT include JD Seven Fresh Coffee, FICO Robot (FICO Coffee), Jilv Holdings (Jilin Ice and Snow Culture and Tourism), Museum Tianxing (Cultural and Expo Venue), etc., and jointly explore overseas markets with Asian coffee brands that have more than 4000 stores.
For these customers, robots are not replacing a barista, but creating a new service touchpoint. This logic is similar to Xiaomi's approach to IoT back then, which is not to replace traditional home appliances, but to redefine demand with new scenarios.
Zhike ZhiKer: What is the biggest help for making robots from software to intelligent hardware?
Tang Mu: All my experiences have taught me one thing: to make products with real user needs in real scenarios, the core is to meet the real needs of users. No matter what type of products you are working on, traditional software, Internet software, intelligent hardware, robots, as long as you grasp this point, the general direction will not be biased.
Zhike ZhiKer: The mainstream founders of the robotics industry have a technical background. Do you feel like you are an "outsider"?
Tang Mu: Indeed, many founders of robotics companies at the forefront of this era come from the fields of technology and engineering. As someone who makes products, I feel a bit like an outsider. But I have been thinking: if embodied intelligence crosses the gap and enters the commercial scene, is this a technical issue or a product issue?
My answer is that it could be a product issue. Technology is always evolving, and we see many high-precision and cutting-edge technologies that can be stored for a decade or two because we cannot find a commercial path. To turn embodied intelligence into a business, the core issue to be solved is still the product problem.
Zhike ZhiKer: You have been emphasizing "making robots in a consumer electronics way", what does this specifically mean?
Tang Mu: Nowadays, robots are mostly in a laboratory oriented, exhibition oriented, and showcase oriented way. At present, our XBOTC3 coffee robot is an industrial grade 6-axis robotic arm robot that can asynchronously make two cups of coffee with its left and right arms. It can make 10 cups of coffee in 10 minutes, and the entire coffee shop is miniaturized in a 2.5-square robot box.
At Xiaomi, the hallmark of a successful consumer electronics product was the shipment of millions of units, as only mass production would face challenges such as whether the production line was complete, whether industrial workers had been trained, whether robots had been validated in various usage scenarios, and whether they had undergone safety certification and 3C certification. In 2025, we took this step forward.
The way of consumer electronics refers more to product thinking, ultimate cost-effectiveness, rapid iteration, and prioritizing user experience. For example, our AI printed coffee allows users to upload photos and robots to create exclusive prints on the coffee. The instinctive reaction of consumers after receiving it is to take photos and post them on social media. The free traffic brought by this emotional value is something that traditional coffee equipment will not consider.
Based on the coffee robot, we have generalized three products:
The XBOT I3 ice cream robot was made in just one month, priced at 179000 yuan, and can make 60 cups of ice cream per hour. The design lifespan of an ice cream robot is 250000 cups. Calculated at an average price of 10 yuan, the revenue of 2.5 million yuan is nearly 20 times its own value.
The XBOT X1 universal catering humanoid robot is equipped with a dual arm seven axis structure and a local computing power of 500 TOPS. It can operate stably in a disconnected state, covering the entire process of food collection, production, and delivery. It can complete various operations such as cooking noodles and mixing drinks. It is expected to be mass-produced and sold by the end of 2026.
The XBOT CUBE robot fast food truck is equipped with a 20 degree battery and covers an area of 8 square meters. It supports functions such as coffee, ice cream, cocktails, and sausages. The launch date is yet to be determined.
Zhike ZhiKer: What were the impressive scenes during the landing of 1000 robots?
Tang Mu: The first robot to enter Tiananmen Square was on the eve of National Day in 2024. The leaders of the Tiananmen Management Committee saw our robot through a video account and told us that they needed to provide coffee for tourists in the square for National Day.
The other one is the first robot in Hong Kong, bought by the Independent Commission Against Corruption. I have been watching TVB TV dramas since I was young, and they often say, 'The Independent Commission Against Corruption invites you to have a cup of coffee.' Now, the Independent Commission Against Corruption's coffee service is provided by our XBOT.
Zhike ZhiKer: Why doesn't the universal food robot XBOT X1 have legs?
Tang Mu: Resolutely refrain from doing leg exercises. Legs under the bar counter in the catering industry are invisible to others, and the legs themselves require extra computing power and electricity to stand firm. We use a base and swivel wheels to complete the movement, and the storage compartment serves as the lower body. After completing the task, we directly place it in the storage compartment and transport it to the customer.
Zhike ZhiKer: Did you design clothes for XBOT X1?
Tang Mu: Dress the robot. We may be the first company in the world to take this issue seriously. If humanoid robots really want to enter various work scenarios, it is impossible to do so without wearing clothes. They will get dirty, splash oil, need to protect their circuits, and ensure food safety. We have found a designer from Beijing Institute of Fashion Technology (the designer of the entry plan for the Chinese delegation to the Milan Winter Olympics) to design different work uniforms for each profession, with baristas, bartenders, dessert chefs, and pastry chefs each having their own exclusive clothing.
Zhike ZhiKer: What are the differences in data collection and training practices of Yingzhi XBOT?
Tang Mu: We turned the real landing points of 1000 robots into data collectors and collected data directly at commercial sites. Using war to sustain war not only enables better data collection and product optimization, but also eliminates the need to spend extra money on hiring remote control personnel.
We went through three stages, and in the first stage, it only took 200 hours of data collection for the robot to learn how to make coffee from scratch. We invited top baristas to the laboratory and recorded their movements with 16 sets of cameras, using pixel level replication to teach the robotic arm how to pull flowers.
In the second stage, 1000 robots will land and 4 million cups of data will be fed back.
In the third stage, we use Human Preference Alignment (RLHF) technology to continuously strengthen the large model and robot based on human feedback.
Zhike ZhiKer: Apart from hardware, what is the core competitiveness of Yingzhi XBOT?
Tang Mu: We call it 'One Brain with Multiple Forms, Dominated by Ten Thousand Stores'.
The self-developed XOS 3.0 "One Brain Multi Form" technology architecture enables cross ontology skill transfer, and the learned operational abilities can still be reused after changing the robot form, truly achieving "one-time research and development, versatile reuse".
Another one is the AI Agent brand "Aibao Store Manager". If XOS 3.0 is the "brain" responsible for learning and evolution, the store manager of Aibao is the "person" who brings this brain to the front of users. Responsible for both front-end user interaction and back-end operational management. A store manager who has operated 1000 stores and knows tens of thousands of loyal customers may not be able to hire such a manager at all. But we can distill the store manager and provide it to every partner who owns our robots.
Zhike ZhiKer: Is the business model a one-time sale?
Tang Mu: Implement the RaaS model (Robot as a Service), where partners can obtain robots at low cost and pay monthly material fees and AI store manager assisted operation subscription fees, instead of paying high robot purchase fees from the beginning.
This includes an AI chain operation system, which has reconstructed the location selection of new retail chains using AI. Traditional location selection is like looking at feng shui, but now the point data generated by large models has a consistency rate of over 97% with on-site survey results. At the same time, we have also reconstructed the smart store system with AI, continuously learning store operation habits and providing suggestions throughout the process.
(This article is the first release of the Titanium Media APP, written by Zhike Zhiker, author Guo Hongyun, editor Yang Lin)
For more exciting content, follow the Titanium Media WeChat account (ID: taimeiti), or download the Titanium Media App