FA hat sein Büro neben DJI verlegt.

Everyone knows that the AI hardware market is booming. But just how hot is it? I heard that some financial advisors (FA) have even moved their offices next to DJI. It's not just FAs. In the Liuxiandong Headquarters Base area in Shenzhen where DJI is located, institutions like the incubation fund of Shenzhen Institute of Science and Technology Innovation initiated by Hillhouse and Li Zexiang are also based here.

A friend who works as an FA complained to me that FAs riding on the wave of AI are mostly doing the most basic tasks, such as creating WeChat groups and listing a bunch of US dollar funds. They basically don't need to provide in - depth services. However, he has seen some people change their WeChat nicknames to "5 term sheets (TS) a day".

Speaking of which, this is quite different from the trend in previous years when industry investment was popular and the focus was on high - quality FAs. The reason behind this is that in the AI - related tracks, those with the halo of big companies have an easier time raising funds. The strategy of FAs has become more straightforward, which is to target people who have left big companies.

The people FAs focus on are where the capital is flowing. The unwritten rule is to target ByteDance for software and DJI for hardware. Some FAs even set up their offices next to DJI, encouraging their staff to get to know DJI employees and keep a close eye on the latest personnel dynamics. Some FAs have also set KPIs for their employees, requiring them to meet at least one or two managers from ByteDance and DJI every week.

Of course, if you take a closer look at the recently disclosed financing situations, you'll find that DJI is just one of the sources. Roborock, Xiaomi, and other hardware manufacturers, as well as executives who have worked on hardware in large Internet companies, have all become the targets of investors. In just the past month, the following AI hardware companies have disclosed financing news:

Robopoet, a company focusing on AI companionship, has completed a tens of millions of yuan angel + round of financing, led by Sequoia China, with existing shareholders GSR Ventures and Zero2IPO Ventures following. Founder Sun Zhaozhi was formerly the head of product design at XPeng Robotics.

Nothing has announced the completion of a $200 million Series C financing. Its goal is to build an AI - native platform that combines software and hardware. The founder is Pei Yu, the former co - founder of OnePlus. This round of financing was led by Tiger Global, with follow - on investors including existing shareholders such as Google Ventures, Highland Europe, EQT, Latitude, I2BF, Tapestry, as well as new strategic investors Nikhil Kamath and Qualcomm Ventures. This globally - oriented company also has some domestic VCs among its investors. For example, Mingshi Capital appeared on the investor list of its Series A+ round in 2021, and Gao Rong Capital also participated in the investment in 2022.

Ropet, an AI machine pet company, has completed a tens of millions of yuan Series A financing, led by the Beijing Artificial Intelligence Industry Investment Fund, with Fengrui Capital following. Co - founder and CEO He Jiabin has previously worked at Microsoft Research Asia, Baidu, Ling Technology, and ByteDance.

Earlier, Looki, which launched a multi - modal AI wearable device, completed three rounds of financing (angel, angel +, and Pre - A) within half a year, with a total financing amount exceeding $10 million. The latest round was led by Zhongding Capital, with existing shareholders BAI, Alpha Startup Fund, and Tongge Venture Capital over - subscribing.

One of the two founders was the former head of Meituan's smart hardware, and the other was the former head of Meituan's autonomous driving algorithms.

Another AI toy company, Haivivi, has officially announced the completion of a 200 million yuan Series A financing series, led by funds under CICC Capital, Sequoia China Seed Fund, Huashan Capital, and Joy Capital. Founder Li Yong is the core founder of Tmall Genie and has witnessed the process from scratch to 30 million units. This round of financing came just four months after the previous Series A2 round and set a record for the highest financing amount in the AI toy industry at that time.

Of course, the above are just some prominent PR messages. Some even say that US dollar funds have fully invested in consumer hardware. This logic is not hard to understand. Firstly, backed by China's strong supply chain, combining with hardware is an easy choice for entrepreneurs. Secondly, the listing of Insta360 this year has undoubtedly further stimulated investors' enthusiasm for hardware.

But AI hardware is a broad and vaguely - defined category. Specifically, what exactly are people investing in? What's the underlying logic?

AI Toys: AI First or Toy First?

"Investors are flocking to invest in embodied and AI - companion toys. Last year it was embodied toys, and this year AI toys are the most popular investment track among investors," an AI industry friend Eric (pseudonym) told me. The valuation of most AI toy companies in their first round of financing is over 100 million yuan, and for more mature ones, it can reach over a billion yuan - similar to that of mid - tier humanoid robot companies.

Whether you call them AI toys or AI - companion robots, this is probably the fastest - growing segment in the AI hardware field. According to IT Juzi data, 96 investment institutions have participated in the AI toy track, including first - tier investment institutions and large Internet companies such as Sequoia, Shunwei Capital, Wuyuan Capital, ByteDance, and JD.com.

The core driving force lies on the supply side. In addition to supply - chain capabilities, technological capabilities, especially in the field of voice, have also reached a certain level. Zhu Xiaohu once said that voice agents have been commercially available on a large scale, and customer service, sales, and AI toys are all extensions of this scenario. However, there are also many non - consensuses in this track. An FA friend told me that many investors he has contacted believe that AI toys are highly homogeneous, and they want to know how AI is reflected in these toys. Another group of investors believes that simply adding AI for the sake of it has little value, and AI toys are essentially toys, belonging to the category of consumer goods. For startups, a key question is whether to develop hardware for the sake of AI or to develop a good product and then integrate AI technology into it?

At a recent salon event, Li Feng, the founding partner of Fengrui Capital, said bluntly that his team has spent a lot of time looking at AI hardware this year, including AI - companion toys, but they were a bit disappointed. "I said these are all toys made for the sake of AI, which doesn't seem very consumer - oriented, so I've decided not to look at them anymore," he said.

His view is quite representative. Annie (pseudonym), an investor from a strategic investment firm, spent three months researching the AI toy track. She communicated with at least 70% of the companies in the market, regardless of whether they are in the form of robots, pendants, or smart dolls, and also tried out many products herself. In the end, she didn't plan to invest in any of them.

Her experience of using these products is as follows: In the first week, there is a strong sense of novelty. The product will create a dedicated social circle and automatically generate a diary after each interaction, recording the content of the communication, such as chatting about the day's weather or sharing a gloomy mood. At night, it will summarize "who talked to me about these things today". During this period, users will feel that "it trusts me and can talk to me openly". However, this sense of novelty is hard to sustain. After a week, users will clearly feel that the functions of these toys are homogeneous, and the initial sense of novelty will basically disappear.

This phenomenon is also confirmed by industry data: "This year, we can see in many industry reports that the user retention rate in the toy track has dropped significantly. For projects like Xingye, which are more virtual and focus on emotional value of companionship, the user retention rate has now dropped back to the level of a year ago, and the daily number of new users is very small."

Looking further at the usage of toys by different audiences, for toys targeting children, even if some brands are very popular at the moment and focus on educational entertainment for children, some parents are willing to buy them directly. However, in fact, it is the children who use the toys, and most children don't like these products. Since it is the parents who pay for them, a group of customers can be accumulated in the early stage, but it is difficult to retain customers based on the children's willingness to use the products later.

For companion toys targeting adults, that is, young people in their twenties and thirties, Annie found that the sense of novelty also fades quickly, and the products cannot meet the deeper needs of companionship, so it is difficult to generate the motivation for continuous use.

Going back to the essence of AI toys, if they are regarded as toys, we need to consider the problem from the perspective of consumer goods. Either the IP capabilities should be strong, or the channel capabilities should be strong. In addition, we also need to consider the needs of different groups. "The explosion of trendy toys is the result of the accumulation of social and demographic structures to a certain extent. For this wave of AI toys to explode, we need to wait for the AI - native generation to grow up. These people will be used to obtaining information and establishing connections through interacting with AI. It's still too early now," Eric said. In his opinion, even a company that has received a lot of financing has just benefited from starting early and having a relatively high - quality product.

Zhu Xiaohu also tends to view this industry from the perspective of toys rather than AI. "AI is just a gimmick and a stepping - stone. The most important thing is the 'cultivation' gameplay. It's like the Japanese electronic pets in the 1980s. The technology was very primitive, but they became popular globally," he revealed in a recent interview with Silicon Valley Geek. It is said that Mr. Zhu only took a dozen minutes to decide to invest in Robopoet.

After some disappointments, Fengrui Capital finally invested in Ropet, mainly because they recognized the founder's idea - to first develop a product that users like, and then consider what role AI can play. With its product - defining ability and a keen understanding of user psychology, Ropet became the focus of the 2025 Consumer Electronics Show (CES) with its cute appearance and interactive design.

Ultimately, AI toys are a segment that is relatively easy to validate and attract customers to buy. In the long run, several investors and entrepreneurs I've talked to are optimistic about this field, but they also believe that the final product form may not be the same as it is now.

Phones, Glasses, or a New Form?

AI needs new intelligent terminals to be implemented, so everyone wants to replace the phone, but what form will it take? Investors are still taking bets.

Pei Yu said in a public financing letter that the most powerful aspect of smartphones lies in their unique ability to acquire context information and user cognition. Therefore, he believes that in the AI era, smartphones will still be one of the most important devices.

However, OpenAI wants to replace the phone. In May this year, CEO Sam Altman announced that OpenAI would acquire io Products, a company founded by former Apple designer Jony Ive. Altman described the process of interacting with an AI chatbot in a video, saying it was like opening a laptop, completely ignoring the functions of the phone. It is said that this cumbersome process prompted Altman to decide that OpenAI should build its own AI device.

On the one hand, OpenAI is recruiting hardware personnel from Apple and using Apple's supply chain to advance these efforts. On the other hand, it has also reached a strategic cooperation with Luxshare Precision, the leading company in China's "Apple supply chain", to jointly develop a revolutionary AI device for consumers. This device, which is in the prototype development stage, is designed to be pocket - sized and portable, equipped with an advanced context - aware system, and deeply integrated with the large - language model behind ChatGPT to achieve more natural human - machine interaction.

Meta is betting that glasses will eventually replace phones as people's main devices. In September, Zuckerberg took the stage at the company's Connect conference and declared: "Glasses are the only device that can allow AI to see what you see, hear what you hear, and talk to you all day long." Meta's newly launched Ray - Ban smart glasses are equipped with a built - in display that can show applications, alerts, and directions on the right lens and can handle many tasks that users traditionally perform on smartphones.

Despite investing billions of dollars in virtual reality headsets, for Meta, AI - driven smart glasses now seem to be the most promising way to connect with users.

"Glasses are definitely a huge market. In the future, the new generation of computing devices will be glasses replacing phones," an AI investor said firmly. In his view, compared with the so - called AR and VR of the previous generation, the technology of this era is different, and AI will play an extremely important and decisive role in glasses.

However, when it comes to the smart glasses track, there is still no consensus. There are still many investors who are not optimistic, each with their own reasons. For example, some investors believe from the perspective of human behavior habits that computers and smartphones are like the stones held by apes, which are in line with human biology, but smart glasses are not. Not everyone has the habit of wearing glasses. Some investors have further compared the usage habits of Chinese and American users, pointing out that the myopia rate in the US is low, and people have the habit of wearing sunglasses while driving. In contrast, in China, the myopia rate is high, and women like to wear colored contact lenses, which reduces the possibility of wearing glasses. Some investors have also told me from a technical perspective that there has been no breakthrough in physics, and the waveguide technology is not mature, so they are not optimistic about all current optical solutions.

Regarding the investment opportunities brought by AI, Li Feng revealed that a key perspective is: when new technologies emerge, which previously unexploited needs can now be solved. Based on this perspective, he sees two types of opportunities: The first type is "upgrading opportunities" - using technologies such as AI to optimize and upgrade things that already exist widely and do not require market education, such as improving efficiency or experience in mature scenarios. The second type is "defining opportunities" - creating a brand - new category that may not have existed before or could not be accurately defined, which is equivalent to creating new needs from scratch.

The underlying logics of these two types of opportunities are completely different: the capabilities they require and the final growth paths are significantly different. Among them, the first type of opportunity may be more familiar to everyone, while the second type, although not easy to see clearly, is also highly exploratory. Of course, in addition to these big and imaginative stories, there are also huge market opportunities hidden in some specific and minor needs.

For example, plaudAI, which has seized the demand for AI meeting minutes, is regarded as the most successful AI hardware company. It has been established for only three years, and its shipment volume has exceeded one million units. "We want to go back to the purest way of product development 10 years ago: find the factor that causes the greatest change, and then provide the greatest value to users based on this factor," the person - in - charge of plaud in the Chinese market said in a recent media interview.

For example, Annie believes that concepts like "companionship" are difficult to define precisely, but there are some clear and rigid - demand scenarios within them, such as companion education for special children. Currently, she is mainly focusing on two types of projects:

The first type is rehabilitation - assistance projects targeting special groups, mainly serving children with autism or those with slower progress in language expression and thinking training due to diseases. The core goal is to help patients improve their language - expression abilities. The key to these projects is that the technology needs to be combined with professional medical resources, and professional doctors need to be involved to accurately interpret the deep - seated needs and states conveyed by the expressions and actions of sick children and convert this information into structured data, and then build targeted vertical models. From the demand side, if the technology can mature, professional institutions such as hospitals will have clear demands. For high - net - worth families with relevant needs, the goals of these solutions are also very clear, but there are currently no mature products on the market.

The second type is intelligent interaction projects focusing on the fan economy. The core is to build an "intelligent base" with strong IP recognition and interaction capabilities: No matter what form the IP peripherals are, such as cards or dolls, as long as they have a clear IP image and are placed on the intelligent base, the base can automatically identify the corresponding IP. More importantly, users can interact with the IP based on its exclusive voice. If the IP is G - Dragon, users can talk to the IP carrier using G - Dragon's own voice, meeting the personalized interaction needs of fans with their favorite idols.

This article is from the WeChat official account "China Venture Capital", author: Liu Yanqiu, published by 36Kr with authorization.