---
title: "Japanese People Flying Drones Indoors? The Two Extremes of AI Going Global"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-04-17T10:01:35+00:00"
canonical: "https://ffcap.cn/en/research/src-20260417-02html"
source: "https://uniqueresearch.substack.com/p/src-20260417-02html"
language: "en"
---

# Japanese People Flying Drones Indoors? The Two Extremes of AI Going Global

_Original · Unique Research · 2026-04-17_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, repeated examples and full panel, including all named speaking turns and their continuation paragraphs. Product, market, user-behavior and technical figures are source or speaker claims, not independently audited findings. Company, personal and product names are transliterated where official English forms remain unverified. Regional user-behavior observations are attributed to speakers, not independently established market data. The source is dated April 17, 2026._

Unique Awards

AI Hardware and Companionship Going Global: New Value and Global Growth Opportunities

Don't Use APP Thinking to Ruin AI Hardware

When making the same piece of hardware for users around the world, the deadliest illusion is "the whole world is the same."

You might think everyone buys a flying camera to go skiing and surfing at snow mountains and by the sea. But when you get to Japan, because of strict local aviation laws (drones over 100 grams require a three-day exam and registration), local users force you into a 99-gram "special edition." Even more surreal, because it is lightweight, Japanese users start flying drones indoors to follow them skateboarding in narrow spaces.

You think when making an AI companionship product, everyone wants a caring big sister. But thedeployment data will slap you hard: East Asian guys just want to "raise a child," Latin American tough guys are busy creating avatars in cyberspace to express their taste, while European and American elites don't want to chat with you at all—they just want you to remind them on schedule to check in on Duolingo.

This is the truemicrocosm of the current AI hardware and companionship going-global track—absurd, torn apart, yet full of real new dividends in gold and silver.

At a closed-door roundtable at Unique Awards · Hangzhou AI WEEK, four frontline operators—Huang Kang, Head of Sales at HOVERAir; Lin Yi, Founder of MuMuverse; Professor Xing Guoliang, Co-founder of ThingX; and Zhou Yixu, Founder of Cybertron—pierced through this layer of window paper. Some truths simply cannot be seen in press releases.

Don't Use APP Thinking to Ruin AI Hardware

Over the past year, AI hardware has gone from myths everywhere to feathers everywhere. First there was AI Pin, mocked across the internet for its terrible experience; then countless pieces of hardware with a large-model shell gathered dust in the first month after going on sale.

Why? Because many entrepreneurs have been brainwashed by internet thinking.

"Hardware and software are very different." Professor Xing Guoliang pointed out thiscrux very restrainedly. Making APPspopular "launch as soon as possible then iterate quickly"—as long as the core function is there, it doesn't matter if the UI is rough. But in the hardware world, this way of playing is suicide.

"The usage cost of hardware is extremely high." Xing Guoliang put it very bluntly, "If the user experience is poor at first launch, this hardware has basically failed."

Supply chain is not a threshold in China at all. You can find a foundry to knock out a shell in minutes. Where is the real competition? In data understanding, processing latency, and even that tiny bit of power consumption control. Many manufacturers, to grab a "first launch" in front of investors, dare to sell algorithms that aren't even half-finished.

At this point, what truly makes users unable to live without you is not how awesome your hundred-billion-parameter model is, but whether you have solved that damn real pain point.

Huang Kang gave a particularly interesting example. As an executive at HOVERAir (a one-takeoff flying camera that follows you), why does he always bring his own product when traveling? Because his wife wants to take family photos, and he doesn't want to keep holding a heavy camera. "When the kids are riding bikes, my wife has to follow behind with a big camera to take photos—she even fell once."

When something truly rescues you from a painful scenario, you don't need to educate users at all—they will naturally take it out with them.

The End of Companionship Is Not "Chatting" but "Connection"

After being baptized by AI, everyone wants to make a virtual companion like JOI from Blade Runner 2049. But many people, as they make it, narrow the path—turning it into a GPT voice assistant with a plush shell.

If you lock a person and a good friend in a room to chat for seven days, after seven days they will definitely have nothing left to say. This is why many AI hardwarefocus on "chatting" get thrown into the storage closet in the third week.

Zhou Yixu of Cybertron hit the nail on the head: "The person you chat with cannot be locked in a room with you with closed information—it has to have its own life."

This returns to the ultimate solution for AI companionship: it is not you unilaterally extracting emotions from it, but you watching it also living its own "cyber life." Zhou Yixu's approach is to build a virtual social universe for these AI characters behind the hardware. When you haven't seen it for a few days and it speaks to you again, it will tell you what new friends it met yesterday and what adventures it had.

This stickiness brought by "shared experience" is a dimensional-reduction strike.

Lin Yi of MuMuversesimilarly deeply believes this. In his view, traditional IP is a business that must be restrained, because "the life of IP is consumed away." But when AI technology comes in, this logic reverses—the consumption process of the product itself becomes the production of content.

Someone cried while holding a plush toy and said, "Oh my god, you saved my life." In that moment, it was no longer a circuit board with a microcontroller—it was an emotional enclave.

Going Global Is a Precise Dimensional Reduction of Human Nature

When the product is made, how do you sell it to the world? The core experience these four operators gained from stumbling overseas is: thesatisfaction points of foreigners is completely different from what you think.

Lin Yi has handled a large number of overseas growth portfolios. When he peels apart global demand for AI companionship, the picture is very interesting:

East Asian market (China, Japan, South Korea): everyone is particularly obsessed with "raising." That is, watching a character grow from small to big, even developing an intimate bond like "raising a child."

Latin American market: Latin Americans desperately need an avatar. Over there, Character.ai is played as a face-creation game. They create a beautiful or highly personalized substitute to socialize in cyberspace, essentially projecting their own aesthetic preferences—"Look, this is my taste."

Europe and US domestic: European and American guys have extremely strong subjectivity. They almost don't want too clingy bond relationships; they more need a "healing-type tool person," such as a ruthless but gentle reminder machine for scheduled Duolingo check-ins.

These collisions brought by culturalundertone occur not only at the software level, but also at the compliance level of physical hardware. In Japan, the weight requirement is stuck at 100 grams; in Europe and the US, the limit is a relatively lenient 250 grams. This directly leads to the same brand selling hardware in two markets that belongs to completely different application scenarios (indoor following vs. outdoor off-road).

The "Non-AI" Underlying Password

At the end of the discussion, when asked what advice they have for new entrants, an interesting consensus emerged across the board: this group of people who understand AI best tells you, when making AI hardware, to look more at "non-AI" things.

Many geek founders always feel that as long as AI capability is maxed out, users will pay. But what truly determines the life or death of an AI companionship toy may not be your algorithm at all.

Zhou Yixu left a heartfelt sentence: "Everyone often focuses on high-frequency scenarios or using AI to express emotion, but what is easily overlooked is timely feedback interaction. For example, if you touch it or shake it, can it immediately give you a vibration and sound that makes you feel comfortable?"

Because your competitors are no longer just the tech maniacs in Silicon Valley doing chips and models. On this battlefield for users' sofas and bedside tables, your real enemies are those traditional toy factories and established consumer goods companies that extremely understand human nature and interactiveuser experience.

Technology will be caught up to, even overturned. But the moment a real person's fingers glide across the hardware in the physical world, and that "good morning" carrying real memory—those are the ultimate answer for AI hardware totraverse cycles.

More Conversation Details

Unique Awards · Hangzhou AI WEEK Trends Roundtable Panel

"AI Hardware and Companionship: New Value and Global Growth Opportunities"

Guests:

HOVERAir Smart Flying Camera — Head of Overseas Market and Sales — Huang Kang

MuMuverse — Founder & CEO — Lin Yi

ThingX — Co-founder & Professor, Department of Information Engineering, CUHK — Xing Guoliang

Cybertron — Founder — Zhou Yixu

Moderator: SmallWOD — Co-founder — Zhao Xiaochun

Zhao Xiaochun: The topic we are discussing today is AI hardware and companionship, new value and global growth opportunities. I feel that when we talk about hardware today, we actually cannot avoid talking about things related to AI. Whether it is the camera you make, Mr. Huang, or possibly some companionship hardware, AI is reshaping the skeleton behind hardware. Our four guests today are also exploring AI directions from different angles. How about we first ask the four guests, starting with Mr. Huang, to introduce your own businesses and overseas layouts?

Huang Kang: Hello everyone, my name is Huang Kang, and I am currently responsible for overseas market and sales for HOVERAir flying cameras. This camera is very close to our theme today, because for over 100 years, the most important original intention of the camera industry has been to capture very precious moments in people's lives. But the problem with most traditional cameras is that the moment you take out the camera, you come out of that moment—because the person taking the photo cannot enjoy life. So the product we launched is themed around intelligent flight, turning the camera into a photographer. With just one tap to take off, it can automatically follow you, or shoot from different angles in the sky, just like you brought a private photographer to follow you and take photos. This product is currently sold in over 200 countries worldwide, with the main market in Europe and the US; it is also sold domestically. I am currently mainly responsible for overseas market and sales.

Lin Yi: The project I am currently working on is an IP called MuMuverse. You can understand it as letting IP not end when you buy it, but hoping that when a product appears, combined with the most fun technology on the market today (whether AI or AR), the product itself becomes a space to be consumed again. We all know IP is a restrained business—why must it be restrained? Because the life of IP is consumed away. But if our product itself is content? So what we mainly explore is creating a new-technology experimental laboratory, through which our own products themselves are content, themselves are fun, and can be used to play AR and all the interactions on the market that can make users' hearts move. We have also done global layout—whether on Twitter or Instagram, we have started making our own things. Overseas, without any paid promotion, we have million-level exposure and views; domestically we already have 150,000 followers. We also want to further explore the differences between domestic and overseas users, better handle various different relationship forms, and let IP truly accompany users.

Xing Guoliang: Hello everyone, I am Xing Guoliang. I currently teach in the Department of Information Engineering at the Chinese University of Hong Kong, and I also co-founded an AI hardware company called ThingX. Based on years of research in embedded AI hardware at the CUHK AIoT Laboratory, last year we launched the world's first smart necklace based on an on-device large model. I am also wearing one—welcome everyone to come find me toexperience later. Based on multimodal sensor recording, this smart necklace can monitor a person's heartbeat, record your emotions and various scene events in your life. Our product recently won the iF Design Award, has already launched on Amazon US, and we plan to launch the domestic product line in May this year. Our major update is totargeting all life scenarios, building a wearable product that can continuously remember information. Our main technical breakthroughs lie in on-device intelligence and on-device large models with multimodal sensors. Hope to have the opportunity to further交流 with everyone, thank you.

Zhou Yixu: Hello everyone, I am Zhou Yixu from Beijing Cybertron. What we are doing is the AI companionship hardware track. We have currently made several different product forms, including companionship plush toys and AI trendy figurines. Hardware is actually a connection we build; behind the hardware we build a character-based universe. Actually, it is very similar to the now-popular OpenClaw concept—in our universe, every character, every agent, is exploring its own life, building its own stories, and generating social interaction between characters. Our hardware is actually the carrier connecting these two universes, allowing the characters we like in the AI universe to interact and communicate with our users in the real world. This is what we are mainly doing now.

Zhao Xiaochun: Understood, thank you four guests for the introductions. Let's enter the first discussion topic. In the past, when people talked about hardware, many talked about parameters, but now everyone perceives that AI has brought great changes to interaction and human perception. So I want to first ask Professor Xing—at the intersection of academia and industry, what do you think the recent AI technology dividend for hardware looks like?

Xing Guoliang: Recently we can see several important basic trends:

First, more and more large models can natively support multimodal data, from the initial text and images to more complex user data (including various sensor information).

Second, Agents are very hot, including technologies similar to OpenClaw, which can better record or understand human needs and help people complete various tasks—essentially integrating the AI toolchain with existing APPs and other tools.

Third, AI capabilities are increasingly shifting from the cloud to the device side, including mobile devices and wearable devices, which now have on-device AI capabilities.

The combination of these three trends brings huge impact to AI hardware. From the past single modality (for example, past bracelets and watches only depended on a few sensors), in the next one or two years there will be a leap to understanding people's full-scenario life status (such as health, emotions, work status, productivity, etc.). AI hardware is crossing from a simple recording tool to a tool that understands people's life, health and productivity.

Zhao Xiaochun: Eason, your product actually also验证 Professor Xing's point about from cloud to device. Regarding the concept of companionship—I am myself an old otaku, who wouldn't want to have a Rei Ayanami or Nezuko? But when people talk about AI smart hardware providing emotional value, they often question what depth it can provide. What do you think about the issue of emotional depth that AI赋予 hardware?

Zhou Yixu: Actually, the part about depth is also what we have been exploring:

The first step is ultra-high还原度. For example, when you make a character, when chatting, the voice, personality and even catchphrases must feel very similar.

The second step is to have accompanying growth memory. The first day I chat with it, and after chatting for a week then saying good morning to it, its answer should be different.

The third step is building a virtual universe. Even if it chats very well, if you lock a good friend in a room for seven days, after seven days there will be no new content to chat about. This is why we put AI characters in a virtual universe to generate new social interactions. The person you chat with is not locked in a room with you with closed information—it also has its own life. This is what a friend is—when you haven't met, each has had more stories happen.

Zhao Xiaochun: Understood, Nezuko can't always be a demon, there will always be a day she becomes human—got it. Next question I want to ask Mr. Huang first—HOVERAir has gone global. In your observation, are there gaps in usage habits among users in different regions?targeted at these differences, what local marketing tactics can you provide for the Founders here?

Huang Kang: We indeed found very large differences when marketing globally. On one hand are differences at the user end, and on the other hand there are significant differences in the business environment and compliance policies, for which we have made great localization adjustments.

Take Japan in Asia as an example: Japan's business environment and user habits are very different from Europe and the US. For example, in Japan, aircraft over 100 grams require registration, and registration takes two to three days for an exam, which is very cumbersome. If not registered, it can only be used indoors. Based on this, we specially developed a flying camera under 100 grams (99 grams), which everyone can use directly without registration. But even so, we were still quite surprised to see many users using this product for indoor roller skating, skateboarding and other sports following shots.

In contrast, in Europe and the US, everyone is relatively more open, and the takeoff weight limit is 250 grams. On the other hand, everyone has a stronger sports gene. Their usage habits in cycling, skiing and various events are very diverse, and the scenarios are very different from Japan.

Zhao Xiaochun: This point is so interesting, I didn't think Japan would have this kind of indoor scenario before. Next I want to ask Lin Yi—I know you are helping many AI companies operate growth. From the perspective of growth strategy and IP, are the "satisfaction points" of users in different markets for AI companionship very different?

Lin Yi: First, to be very accurate, no matter which country there is this kind of demand, but the倾向 relationship is different.

East Asian market (such as China, Japan): more倾向于 a raising-type relationship. We can see the very popular IP games last year, essentially the relationship between me and the character. We want to watch the character experience things, exploring in-depth raising and even an intimate relationship like "raising a child."

European and American markets: Europe and the US are very different—there are many functional companionship APPs (such as letting you check in on Duolingo). Europeans and Americans have strong subjectivity; they more hope this is a healing-type tool, but don't necessarily need how deep a relationship, more hope to independently be themselves.

Latin American market: Latin America is also very different. For example, Character.ai is very popular in Latin America—it is actually me creating an avatar to socialize instead of me. Latin Americans need to express aesthetics on their avatars, a bit similar to the mood when we buy trendy toys: "The taste of this thing represents me."

There are specific differences in each market, but actually everyone's true mood is hoping to be cared for, loved, accompanied, seen and understood.

Zhao Xiaochun: What we just talked about is how to make the product reach users' hearts. Now many AI hardware are "peak at debut," and there is a gap between going viral and forming a brand. At this moment, what kind of AI hardware can make users feel unable to live without it? How to cross this gap? Starting with Professor Xing?

Xing Guoliang: I very much like toexperience new hardware (including early smart glasses Glass, Leap Motion, AI Pin, etc.). My experience is that hardware and software are very different. Software thinking is launch as soon as possible then iterate quickly; but the usage cost of hardware is very high—if the user experience is poor at first launch, this hardware has basically failed. Many hardware manufacturers easily step into a pit: to launch as soon as possible, they launch products when data and algorithms have not accumulated to a breakthrough state, and the experience is extremely poor. Our domestic supply chain is not a problem at all, even leading the US; but for personal products like sleep rings and watches, the competition is in "internal skills" like data understanding, processing latency and power consumption. What time to launch what experience product is very, very important.

Zhao Xiaochun: I very much want to ask Mr. Huang—I feel you have actually crossed this GAP.

Huang Kang: A very key point is to see whether users will spontaneously use the product, which behind means whether it has solved a real pain point. Take myself as an example: my wife likes photography, and our family of four rarely has family photos when traveling because no one takes them for us. Because it solves this pain point, we will definitely bring it when traveling. Also, when the kids are riding bikes, my wife has to follow behind with a big camera to take photos—she even fell once. With it, there is no need for someone to keep holding the camera behind. If you solve users' real needs in daily life, you don't need to激励 users and they will use it, which basically can cross a relatively long cycle.

Zhao Xiaochun: Eason, this should also be a question you often think about?

Zhou Yixu: Our productbiased向 companionship—it is not as strong as functional demand, sometimes needed and sometimes can be without. From our perspective:

First, it must appear in high-frequency scenarios, for example more portable, so it is convenient for users to carry.

Second, build relationships and memory. Besides me personally holding it and chatting, maybe I can hold it and touch it with a friend, bringing real-world social relationships, or go somewhere with it to check in.

On one hand, let users carry and use it at high frequency; on the other hand, if I occasionally forget it, I can still think of it.

Zhao Xiaochun: The emotional card is a very good card. I also want to ask Lin Yi's view on this—making IP is also a process of insight into human nature.

Lin Yi: We are essentially a technology-empowered content company, and the logic is more like making IP. This is a symbol of spiritual power—for example, you may think of the character when sad or happy. Combined with our story, what users think of is not just the image, but "the story between me and him, he is re-empowering my life." I visited fans before, and someone said while holding a plush toy, "Oh my god, you saved my life." Because of the deep感动 brought by real empathy, I believe IP can definitelytraverse cycles.

Zhao Xiaochun: Now going-global costs are very high. In the process of doing overseas markets, how do you use AI itself (whether marketing tools or other methods) to reduce growth costs? Starting with Lin Yi.

Lin Yi: A lot of our internal materials are actually made with AI, mainly to save manpower and improve efficiency. But in distribution strategy, it still very much tests people's control over content and hot topics. Hot topics come and go immediately—for example, the hot topic of pinching cat tofu, you must discover it yourself and tell AI. Currently AI is more like a tool to assist me in rapid production, and distribution strategy is still done by humans.

Huang Kang: The place where cost saving is greatest for us is material production. We originally needed to shoot on location in places with beautiful scenery in Europe and the US (such as Canada, Switzerland, New Zealand), which was extremely costly. At first we tried direct AI generation and found the "AI feel" was very strong and fake; later we found a middle way: take back the real angles of the shot product, and let AI extend on that basis. For example, change snow to maple leaves, winter to summer, then use AI to help with lighting and bokeh, which greatly reduced our content production costs.

Zhou Yixu: We are building a system for processing To B client information and emails. We adjusted the internal structure, letting technical personnel solve deeper technical problems, while this system, after theunderlying is built, is completely constructed by marketing students themselves according to demand. We are trying to use AI to replace miscellaneous affairs in more aspects.

Xing Guoliang: For going-global enterprises, how to scientifically obtain user experience data is a pain point. Everyone attaches great importance to Early Adopters (early users), which is correct; but many new productstargeting the general public, and Early Adopter data may lead the product astray. Enterprises or product managers need extraordinary insight to extract from limited and noisy data the features that can truly change user experience and put them into the product—this is a very powerful capability.

Zhao Xiaochun: Finally, I would like to ask everyone to give the friends here who are doing overseas a most practical advice standing at the moment.

Zhou Yixu: Regarding companionship products, I add a pitfall-avoidance advice. Everyone often focuses on high-frequency scenarios or using AI to express emotion, but what is easily overlooked is timely feedback interaction (for example, touching or shaking it can timely give sound or vibration), which may not have much to do with AI, but is extremely important for experience. Our competitors are no longer another tech company, but another product company or even product manager. Don't focus too much on technology—pay special attention to users' final interactive experience.

Xing Guoliang: In entrepreneurship, use first principles to think about how to solve users' real needs, rather than simply chasing hot topics. Before hardware becomes a mature product, costs are extremely high and input-output ratio is very low—think clearly before starting a business.

Lin Yi: Being yourself is the most vertical track, and being yourself is the hardest to be shaken. Only when you find something you truly like will you be willing to invest huge costs for it, because maybe your whole life is here for this.

Huang Kang: I suggest everyone carefully recall and sort out their current work, see what can be disassembled and modularized and completely handed over to AI—this can greatly improve human efficiency and quality. AI can help a lot with the online part; in the future, manpower should be more placed in real physical-world interaction (just like looking at Iceland online is different from experiencing it in person). Whatever value AI cannot replace, put manpower there; what can be replaced by AI, completely release to AI.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260417-02html
On-site reading page: https://ffcap.cn/en/research/src-20260417-02html