Original · Unique Research · 2026-05-17
Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, the full interview narrative, and the complete Q1–Q10 selected Q&A. Company scale, profitability, product capability, and market-projection figures are source or speaker claims, not independently audited findings. Company names (Saibo Chuangli / 赛博创力, Saibo Grand Stage / 赛博大舞台) and personal names (Zhou Yixu / 周羿旭) are working English renderings where official English forms remain unverified. The Disney Olaf robot, Cambridge University research, and China's "Interim Measures for the Management of AI Anthropomorphic Interactive Services" are attributed references. The term "00后" is rendered as "post-2000s." The source is dated May 17, 2026.
Unique Awards · Guest Interview
When domestic users get an AI hardware device, their first sentence is usually: "What can it do?"
"
When domestic users get an AI hardware device, their first sentence is usually: "What can it do?"
When overseas users get the same thing, they ask: "Is it cute? Will it remember me?"
When Zhou Yixu told me this contrast, I paused for a moment—but anyone who has done going-overseas has more or less felt this difference, and he has also bet his entire company's product logic on the latter question.
He is the founder of Saibo Chuangli, from the Zhejiang University AI Research Institute, doing multimodal large-model research. He started the business in 2024, with a single-digit team, mostly post-2000s. The product is not AI glasses, not AI earphones—it's AI plush toys and desktop robots. They have alreadybatch shipped overseas through the ToB model and achieved positive profitability.
AI Doesn't Lack Intelligence, It Lack Presence
When Zhou Yixu was doing research at Zhejiang University, his biggest feeling was not that the technology wasn't good enough, but that the technology had no "body."
"AI already doesn't lack 'intelligence,' but it lacks 'presence.'"
He said that when people use AI, it's basically still typing in a chat box. But humans' truly natural interaction方式 isn't like that. You look at something, touch it, put it on your desk, hug it to sleep, take it out with you.
"So later I increasingly felt that the truly big opportunity for large models is not just software, but whether, after AI enters the physical world, a new type of character-based product will be born."
Another observation that pushed him to start a business came from young people. He found that many post-2000s and post-2005s young people are already very naturally "building relationships with AI." Not treating AI as a search engine, but as a companionship object, an emotional outlet, even some kind of "digital life."
This isn't his speculation. Character.AI's user data, the retention curves of various AI social products—all confirm the same thing: humanity's need for "being continuously attended to by another being" is much stronger than we think.
"Devices That Can Talk" and "Characters That Live in Devices"
Many people's approach to AI hardware is: take a speaker or robot shell, plug in a large model, add voice interaction, and done.
Zhou Yixu feels this path won't go far.
"Hardware is functional; character is relational."
A speaker is a tool—you won't develop feelings for it. But if something has personality, has memory, and will proactively respond to you—even if it's just a plush doll—you'll start to develop an emotional connection with it.
"We often discuss a sentence internally: users won't long-term fall in love with a function, but they will long-term fall in love with a character."
After hearing this sentence, I felt it was quite accurate. Think about it—how many "efficiency tools" have you deleted from your phone? But that stuffed doll from your childhood might still be sitting in some cabinet somewhere.
So Saibo Chuangli chose product forms of plush toys, desktop robots, and smart-education cameras—things that inherently carry emotional and companionship attributes. Their core product is called "Saibo Grand Stage," a palm-sized smart base that, through NFC stickers, can make any figurine or plush toy "come alive," start talking to you, remember you, and respond to you.
Why not make "cooler" categories like AI glasses or AI earphones?
"Because glasses and earphones are essentially tool augmentation. But a child won't hug a smart speaker to sleep—he'll hug a plush character."
His judgment is: the first step of AI entering the real world is not an efficiency revolution, but an emotional revolution.
How Hard Is It to Make a Character "Come Alive"?
It sounds romantic, but doing it is all engineering hell.
Zhou Yixu said AI hardware is much harder than pure software, and the core reason is: users' fault tolerance for hardware is extremely low.
If software freezes for a moment, you just refresh the page. But if a physical character responds slowly, doesn't understand what you say, or suddenly stops responding—the user will immediately "break character." Because interaction in the real world is continuous, and any breakage shatters the illusion that "it's alive."
"This is not just a model problem, but an end-to-end system problem."
For an AI character hardware to be truly usable, he listed six keywords: hear clearly, understand visually, respond fast, have memory, have emotional expression, and have physical feedback.
The first three are basic capabilities; the latter three are what make a character "have a soul."
"Physical feedback is especially important. Even just a light change, a small movement, a vibration, a changed expression—all greatly enhance the feeling that 'it's alive.'"
And among all these dimensions, the one point he repeatedly emphasized is response speed.
"The rhythm of conversation in the real world is millisecond-level. If a child asks a question and has to wait three or four seconds, the experience drops sharply. The child won't understand 'the model is thinking'—he'll just feel: it's ignoring me."
"Latency itself is part of the personality experience."
Think about it—if a person always responds half a beat late, you'll feel he's absent-minded. It's the same for AI characters. Response speed is not just a technical metric; it directly affects the user's perception of "does this character care about me."
So Saibo Chuangli internally一直 obsesses over real-time interaction: VAD (Voice Activity Detection), streaming voice, multimodal low latency. These technical points that sound dry are, in their context, "personality infrastructure."
Children's Scenario: Not Answering Questions, but "Being Seen"
When talking about the children's market, Zhou Yixu's judgment differs from many people making education hardware.
Most people think: what can AI help children learn? What questions can it answer? What homework can it tutor?
He feels this has it backwards.
"Children often don't need a correct answer—they need to be responded to, listened to, and encouraged."
Natural language interaction is inherently friendly to children; kids don't need to learn to type or understand interface logic—they just open their mouths and speak. And children are more likely than adults to treat AI as a "friend" rather than a tool. Adults will constantly remind themselves "this is just a machine"; children won't.
This is also why the safety boundary in children's scenarios is particularly important.
"Children will develop emotional dependence on AI, so the industry definitely needs a higher sense of responsibility than traditional internet products."
This isn't empty talk. 2025 has already been called the "inaugural year of AI toys"; while the global market size surges, controversies over children's privacy, emotional dependence, and content safety are also heating up. Cambridge University research found that existing AI toys have significant limitations in understanding preschool children's complex emotions; China issued the "Interim Measures for the Management of AI Anthropomorphic Interactive Services" in early 2026, specifically setting thresholds for minor protection.
Zhou Yixu's attitude toward this is: safety boundaries are not constraints, but capabilities that products must build in. Large models have creativity, but also hallucinations and instability—placed in children's scenarios, these are not "occasional bugs" but trust issues.
The Future of IP: From "Being Consumed" to "Companionship Existence"
Saibo Chuangli has always paid close attention to the AI + IP direction. The reason is simple:
"We increasingly find that what users truly interact with long-term is not 'AI capability,' but 'characters they like.'"
Think about it—would you chat with an AI every day because its model parameters are strong? Probably not. But if that AI is your favorite anime character, game character, or even a virtual avatar you set yourself—you might genuinely be willing.
Zhou Yixu's judgment is: in the future, IP will shift from "content" to "relationship."
"In the past, when you followed a character, you watched its story. In the future, it might be that this character truly knows you. It will remember you were in a bad mood yesterday, remember your birthday, your habits, what music you like."
This means IP is no longer just a content product that is "consumed," but may become a kind of "companionship existence." The consistency of character personality, worldview, and expression becomes more important than the technology itself.
And the potential user base is broader than many people imagine: children, 18–35 female users, ACG (anime/comic/game) community, three-dimensional celebrity fans, designer-toy collectors—these people have already built emotional relationships with "characters." AI is just the first time these characters have truly "come alive."
What's the hardest part of making IP characters AI-ified? It's not technology.
"It's character consistency. Personality, worldview, expression, content safety, user-relationship boundaries—if these aren't done well, users will immediately feel 'this isn't that character.'"
Why This Worked Out Overseas First
Saibo Chuangli took the overseas route very early. The starting point was the Canton Fair.
"At the time the team was very small, the product was still constantly iterating, but fortunately, we got our first batch of overseas orders at the Canton Fair."
Later they exhibited successively at CES and MWC (4YFN), coming into contact with a large number of overseas purchasers, channel partners, and IP collaborators.
This process gave him a strong feeling: the overseas market's acceptance speed of "AI character-based hardware" is much faster than expected.
"Many overseas purchasers don't understand it as a 'toy that can talk,' but as a new interactive consumer product, a new IP carrier."
Going back to the contrast at the beginning—domestic users ask "what can it do," overseas users ask "will it remember me"—behind this are two different consumption logics.
Users in Europe, America, and Japan are accustomed to building relationships with pets, building relationships with IP characters, and paying for emotional value. They inherently understand "why a character needs memory, personality, and long-term interaction capability."
And choosing the ToB model to go overseas has another layer of practical consideration:
"AI hardware is different from pure software—it ultimately must go through prototyping, production, shipping, after-sales, certification, and cost control. We very early hoped we wouldn't be 'PPT AI hardware,' but products that can truly be delivered inbatch."
ToB can faster verify users' willingness to pay, whether the product can scale, and whether the supply chain is stable. For a single-digit team, this is much more pragmatic than burning money on a C-end brand.
Shenzhen Supply Chain: Competition Domestically, Endorsement Abroad
When talking about the advantages of Chinese teams doing AI hardware going overseas, Zhou Yixu's passage I found quite interesting:
"Domestically, there are many teams in Shenzhen doing hardware, and everyone may be competitors. But once you go abroad, you suddenly find: Shenzhen's complete supply chain system and Chinese manufacturing capability are actually a very strong endorsement for us."
He said many overseas teams may have concept, design, and model capability, but Chinese teams can truly achieve rapid prototyping, rapid iteration, cost control, module integration, small-batch trial production, and rapid mass production.
"This capability is actually very scarce. Especially AI hardware is still an industry highly dependent on 'software-hardware collaboration.'"
This is completely different from the logic of pure software going overseas. Software going overseas competes on product and growth; hardware going overseas competes on the full-chain capability from design to mass production. And this happens to be what Chinese teams are best at.
Will AI designer toys become a new entry point for AI application going overseas? Zhou Yixu thinks it's very possible.
"AI designer toys are essentially notsimply hardware, but AI + IP + emotional consumption + long-term companionship. Compared with pure software subscriptions, it更容易 form character assets and brand recognition."
The Singularity Is Not a Moment, but Five Things Converging
Zhou Yixu said he believes "the AI hardware singularity has arrived," but his understanding of "singularity" is not a single breakthrough.
"It's several conditions arriving simultaneously: large-model capability has arrived, voice interaction has matured, hardware costs have dropped, users have started accepting AI characters, and global consumer electronics channels are also looking for new stories."
Five thingssuperimposed together, and truly AI-native products start to appear.
"Many times, a technological breakthrough itself doesn't immediately change the world. What truly matters is: whether technology, user habits, supply chain, and business environment happen to converge at the same point in time."
As for the product form that will first break out in the next three years, he is optimistic about "embodied-intelligent toys."
He gave the example of Disney's Olaf robot. This character robot, developed by Disney Imagineering in collaboration with NVIDIA and DeepMind, has been常态化 operating at Disneyland Paris since March 2026—it can walk, maintain balance, perform classic moves, and visitors can have natural conversations with it.
"Because it inherently同时 possesses character感, motion feedback, emotional expression, companionship capability, and long-term interaction capability. And humans will inherently develop emotional projection toward characters that 'can move, can express, can respond.'"
"What truly breaks out in the future may not be the AI device with the strongest parameters, but—the first AI character that truly makes users develop emotional dependence."
Young Team's Intuition
Saibo Chuangli's team is very young—early single-digit scale, mostly post-2000s technical personnel.
I asked him what he thinks about young teams doing AI hardware, after all AI hardware involves algorithms, embedded systems, supply chain, design, content, and sales—very high complexity.
"The biggest advantage of a young team is natural sensitivity to new interactions. Many post-2000s young people actually don't treat AI as a tool, but will truly treat AI as a 'digital character.' This kind of intuition is very important."
But he's also very clear-headed: must not try to do everything yourself.
"Hold the truly critical parts in your own hands—AI character capability, multimodal interaction, character system, product definition. But for supply chain, production, and some hardware modules, heavily rely on ecosystem partners."
The key for a small team making complex products is not being all-capable, but knowing which things determine life or death and which things can be leveraged.
I finally asked him: if you summarize in one sentence, what is the true opportunity of AI-native hardware?
"Not putting a large model into a device, but through AI + IP, letting humans for the first time truly be willing to build a long-term relationship with a character they like, in the real world, develop an emotional connection, and even live together."
Honestly, after hearing this sentence, my feelings were a bit complicated. On one hand, I feel this direction genuinely makes sense—humanity's need for companionship is the mostrigid ofrigid needs, and the technical conditions have genuinely arrived. On the other hand, there's an indescribable unease: when children start to develop genuine emotional dependence on a plush toy, when "characters" understand how to respond to your emotions better than real people—Is this progress, or something we should think seriously about?
Maybe both. Maybe this is what every technological inflection point looks like: opportunity and hidden worry are always two sides of the same coin.
Selected Q&A
Q1: If you didn't use the official introduction, what exactly are you doing?
Zhou Yixu (Founder & CEO of Saibo Chuangli): Very simply, it's bringing characters that people originally could only like and imagine, truly into real life. In the past, whether anime characters, idols, game characters, or plush toys and figurines, they were more just existences that are "watched." We hope AI can let these characters start to truly know you, remember you, and accompany you. So what we're making is not an "AI hardware," but more like making "an entry point for AI characters in the real world."
Q2: Why leave the Zhejiang University AI Research Institute to start a business making hardware?
Zhou Yixu: When doing multimodal large-model research, my biggest feeling was that AI already doesn't lack "intelligence," but lacks "presence." People still use AI in a chat box. But humans' truly natural interaction happens in the real world—looking at it, touching it, putting it on the desk, hugging it to sleep. The truly big opportunity for large models is not just software, but new categories after AI enters the physical world.
Q3: For AI plush toys to make children willing to interact long-term, what's the hardest part?
Zhou Yixu: It's not answering questions. It's personality, tone, memory, motion feedback, safety boundaries, sustained novelty—these together constitute the experience that "it's alive." Taking any one item alone isn't enough.
Q4: What do you think about the "response speed" issue in AI hardware?
Zhou Yixu: Very important. The rhythm of conversation in the real world is inherently millisecond-level. If a child asks a question and has to wait three or four seconds, the experience drops sharply. He won't understand "the model is thinking"—he'll just feel "it's ignoring me." Latency itself is part of the personality experience.
Q5: For making IP characters AI-ified, is the hardest part technology or character consistency?
Zhou Yixu: Character consistency. Personality, worldview, expression, content safety, user-relationship boundaries—if these aren't done well, users will immediately feel "this isn't that character." Technology is the foundation, but the character's sense of soul is why users stay.
Q6: What's the difference in acceptance of AI toys between overseas and domestic markets?
Zhou Yixu: Overseas users have higher acceptance of AI companionship. Users in Europe, America, and Japan are accustomed to building relationships with pets, building relationships with IP characters, and paying for emotional value. Many domestic users currently still first ask "what can it do?" while overseas users will ask "is it cute?" "will it remember me?" These are two very different consumption logics.
Q7: Why early on choose ToB and overseas purchasers as the breakthrough point?
Zhou Yixu: AI hardware is different from pure software—it ultimately must go through prototyping, production, shipping, after-sales, certification, and cost control. ToB can faster verify whether users will pay, whether the product can scale, and whether the supply chain is stable. We very early hoped we wouldn't be "PPT AI hardware," but products that can truly be delivered inbatch.
Q8: What's the biggest difficulty in AI hardware going overseas?
Zhou Yixu: Certification, compliance, supply chain, after-sales, content safety, children's privacy, localized characters and language—none of these is easy. But the most core is still: can your product truly run in overseas users' hands. Not that the demo works, but that daily use works.
Q9: Is the core competitiveness of AI hardware ultimately model capability or end-to-end experience?
Zhou Yixu: Ultimately it's end-to-end experience. The model is important, but users won't long-term like a product because it uses a certain model. What users remember is "does it have a sense of soul."
Q10: What do you think about the judgment that "the AI hardware singularity has arrived"?
Zhou Yixu: It's several conditions arriving simultaneously—large-model capability has arrived, voice interaction has matured, hardware costs have dropped, users have started accepting AI characters, and global consumer electronics channels are also looking for new stories. These thingssuperimposed together, and truly AI-native products start to appear. A technological breakthrough itself doesn't immediately change the world; what truly matters is whether technology, user habits, supply chain, and business environment happen to converge at the same point in time. I think this is happening now.