---
title: "AI Hardware Going Global: Is Less Than 80% Gross Margin a Loss?"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-05-30T12:21:26+00:00"
canonical: "https://ffcap.cn/en/research/src-20260530-01html"
source: "https://uniqueresearch.substack.com/p/src-20260530-01html"
language: "en"
---

# AI Hardware Going Global: Is Less Than 80% Gross Margin a Loss?

_Original · Unique Research · 2026-05-30_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the complete narrative analysis, all section headers, and the full roundtable Q&A transcript with the moderator and both panelists. All named speakers, companies, roles, numbers and pricing details are preserved. Company, personal and product names are transliterated where official English forms remain unverified. Market projections, performance claims and company-specific figures are source or speaker attributions, not independently verified findings._

AI Industry Truth

AI Hardware Going Global: Is Less Than 80% Gross Margin a Loss?

The day hardware is sold is not the endpoint—it's the starting point

Two people, one tennis robot prototype weighing dozens of kilograms, carried all the way from Los Angeles to New York.

On the court, they stopped Americans playing tennis: "Want to try our robot?" Before crowdfunding even launched, 200 people had already used this machine.

Around the same time, another entrepreneur spent three to five thousand dollars renting a small venue in Silicon Valley to hold a launch event. Took a few photos, bought some media coverage, came back and wrote on his WeChat Official Account: "We launched in Silicon Valley."

They were both heading to the same place—getting overseas users to pay ten-plus dollars a month in subscription fees.

This is a story I heard at an AI globalization roundtable in Shenzhen. The two protagonists: Chao Guang, co-founder of Yisi Tennis Robot, and Duan Ran, CEO of Xingqiong Fangzhou. Their product forms, customer unit prices and target users barely intersect. But by the end of the conversation, I found they were actually answering the same question: after the hardware is sold, where does the money come from?

Don't Make a Better Tool—Make "I Want It"

Let me start with a counterintuitive fact: tennis ball machines have existed for decades, but their penetration rate is extremely low.

It's not because the technology isn't good. Chao Guang explained the real reason—the value ceiling is too low.

"Take a pool cleaning robot, for example. You're just getting the pool clean. Its ceiling is the value of a person cleaning it manually." Ball machines are the same—you buy one, use it twice, and once the novelty wears off, it gathers dust in the garage.

He gave an extremely precise distinction:

"What you're buying with a tool-type product is actually its value ceiling, but what you're buying with a consumer-grade product is only one reason—'I want it.'"

Breaking it down: with a tool, you buy what it can maximally help you do—the ceiling is fixed. With a consumer product, you buy the desire it creates in you—the ceiling can be raised. An iPhone can make calls and send texts, but nobody buys an iPhone for that.

So what Yisi is making is not a "better ball machine." What Chao Guang wants is to use AI to change human-machine interaction from "interact once per match" to "interact on every single rally." As interaction density goes up, highlight moments increase. An amateur who plays only once a week—if the pleasure in that one session is dense enough, he won't let the machine gather dust.

"This way, when expanding the circle, some users may not be enthusiasts. They may not play often or with high frequency. So you need the highlight value in one or two sessions to be high enough to break through to them."

Duan Ran's angle is completely different, but the underlying logic is the same.

He doesn't make a single hardware product. Instead, he offers multiple SKUs—pendants, glasses, rings, necklaces—plus a unified model interface layer. "In Shenzhen's Huaqiangbei, mold opening costs 100,000-plus yuan, and mass production can be covered by pressing some payment for goods." The hardware itself is not a barrier. It's an entry ticket.

His entry target is very clear: people with high offline information density.

"Focus on sales, consulting—people who communicate with others every day. Let a salesperson do not just sales, but also pre-sales consulting (SA), customer success, and solution architecting."

Hearing this, I thought of an analogy. CRM used to be a piece of software installed on a computer. Now what Duan Ran wants to build is a "physical CRM" hanging around your neck—recording, organizing and automatically generating proposals for you 24/7, then charging monthly through the AI capability behind it.

The two people's product forms are worlds apart. One is a large scenario robot costing thousands of dollars; the other is a small pendant costing just over a hundred dollars. But the endpoint of their business models is the same: hardware is one-time revenue; AI interaction and software subscription are recurring revenue.

An Arithmetic Problem Most People Won't Discuss Publicly

When talking about going global, entrepreneurs usually only say "we're planning to enter North America." How to make money, how much you need to make to not lose money—almost nobody elaborates in roadshows and interviews.

Duan Ran laid the numbers on the table at this roundtable.

"Hardware gross margin needs to be at least 60%, normally 80%."

Why so high? Because the cost structure of going global is completely different from domestic. Tariffs are one layer, cross-border shipping is another, overseas warehousing is another, and after-sales is even harsher—basically "replace, don't repair," and the return shipping costs more than sending a new unit. Stack all these layers, and if your hardware gross margin is only 40% or 50%, you're losing money on every unit sold.

How to set prices? Duan Ran gave a very practical coordinate system:

In the U.S. market, around $100 is a comfortable range for users—they "don't have much cognitive burden." But he immediately followed with: "I mean the U.S. The U.S. and Northern Europe are at the same table; Southeast Asia and Russia are at another table." Different markets have completely different wallets and psychological accounts.

But relying solely on the one-time hardware sale can't sustain things. Duan Ran calculated another account: the bandwidth and computing power costs behind large models are extremely high. "Domestically, Doubao says it's free, but the bandwidth costs behind it are very high—it can't stay free forever."

Conclusion: going-global AI hardware must be paired with SaaS subscription. $9.9 to ten-plus dollars per month. American users are already accustomed to software subscriptions and won't blink at this price point. This is not a "bonus item"—it's a "lifesaving measure."

Chao Guang's side has also validated the same path, and has gone further.

Early on, a user asked a seemingly naive question: "If I play tennis with this robot, does that mean I can never win?"

This question forced out an entire product design. If users can never win against the machine, then this becomes an endless torment. "So based on this, we extended out a 'teach-compete-practice system'—your interaction is purposeful, embedded into the entire system of teaching, competition and practice."

Goals, progression, feedback. Based on this system, Yisi implemented subscription-based pricing—the first functional hardware product to charge for software.

This is almost exactly the same path as Plaud AI. In 2023, this AI recording hardware company crowdfunded over $1 million on Kickstarter. By 2025, it had sold over 1 million units globally, with annual revenue of approximately $250 million. What it validated is precisely this path: hardware acquires customers, subscription makes money.

AI companies still relying on pure hardware price differences for going global in 2026 are no different from those selling power banks overseas ten years ago. They can survive, but they can't grow big.

Two Going-Global Paths: Dozens-of-Kilograms Prototype vs. Three-to-Five-Thousand-Dollar Launch Event

This is what I found the most interesting segment of the entire roundtable. The two people's going-global paths are completely opposite.

Chao Guang's path: heavy.

Before crowdfunding launched, he and a colleague flew to the U.S. with a tennis robot prototype weighing dozens of kilograms and spent about half a month there. From university courts in Los Angeles to the U.S. Open finals site in New York, they did only one thing—"street challenge."

"We'd take the machine directly to tennis courts and ask people playing tennis if they wanted to try our machine."

200 people used the product before crowdfunding launched. The feedback from these 200 people directly drove subsequent product definition.

He added a very sobering remark: "Don't overestimate technological progress. Everyone predicted pure vision for lawnmowers back in 2020. Six years later, we're still in the LiDAR era. You are, after all, a mature ToC product."

Duan Ran's path: light.

"You can take the product to Silicon Valley and hold a launch event. It's actually not expensive—about $10,000, or even $3,000 to $5,000 can do it. Take a few photos, buy some media, promote it on the Chinese internet as 'launched in Silicon Valley'—sounds very high-end."

Then find people on Twitter to write positive reviews—export for domestic sales.

He put it bluntly: "It's still a viable path."

Both paths are real. But Duan Ran himself drew the dividing line: products priced under $150 can go straight to TikTok or Twitter for video promotion; higher-priced products must go offline for hands-on experience—CES, IFA, even grabbing a photo op at Apple events—to get product endorsement.

A hundred-plus-dollar pendant, users might place an order after watching a video. A several-thousand-dollar tennis robot? No matter how many tweets you post, he won't pay. He has to touch it, play with it, feel it.

Chao Guang added what I consider the most valuable judgment of the entire event during the audience Q&A:

"For complex products, early users shouldn't be treated as users—they should be treated as 'partners.' If someone follows you at such an early stage, they must be strongly related to the industry."

Translation: someone who leaves their email to follow you before the product launches is eight out of ten times a tennis coach, club owner or sports rehabilitation therapist themselves. They don't just want to buy a machine—they want to use your product to do business. Treating these people as "consumers" and blasting them with marketing emails is wasting the scarcest resource.

He also gave the timing rhythm for conversion:

"The essence of operations is giving them expectations, fulfilling expectations, and continuously building trust. Around the seventh touch, they'll buy."

My judgment is clear: for light-decision products under $100, Duan Ran's approach is low-cost and fast-acting—worth copying. But for complex, high-ticket AI hardware, the halo bought through information asymmetry won't survive the first wave of user reviews. Only by hitting the streets to withstand real feedback and building a proprietary system based on that feedback—like the teach-compete-practice system—is this a cycle-traversing approach.

Learn strategy foundations from Chao Guang, borrow tactical leverage from Duan Ran.

Finding People vs. Finding Courts: Globalization Is Not Copy-Paste

Finally, let's discuss an easily overlooked issue. The same product may solve fundamentally different contradictions in different markets.

Chao Guang has a deep feel for this. "Overseas, especially in North America, there are many courts, and transportation is door-to-door. So the main contradiction overseas is 'finding people'—need to find playing partners and smart companions." The tennis robot in North America solves the sparring partner need, directly ToC.

What about domestically? "Population and courts are concentrated in coastal areas, and courts are the main contradiction." This is also why a new format called "ball machine gym" has emerged in recent years—using machines in office buildings to solve the problem of having no court to play. So domestically, it goes through channel partner cooperation, through venues, clubs, training camps—essentially ToB.

The same machine, sold to individuals overseas, sold to venues domestically. It's not that the product changed—it's that the contradiction changed.

Duan Ran's thinking on privacy is also worth mentioning. Taking AI wearables with cameras to Europe and America is extremely sensitive. His solution is not to pick one option and stick with it, but to give users choices.

"Some users want higher efficiency and don't care—add visual shooting, as long as you don't take it to the bathroom. Some are sensitive to visuals and only use audio. Some don't want information sent to the cloud at all—we can also provide local or edge-side services."

"Believe that users today are smart, with independent choice and thinking ability. You can't deceive them—appropriate promotion is fine, but you can't lie to them."

As for how to choose channels in different regions, this roundtable gave some very specific paths:

North American tech hardware goes through Kickstarter to create hits; traditional hardware (like power banks) goes through Amazon.

The Middle East goes through family-run offline small department stores—some have been open for 20-30 years, with high community trust.

Offline retail takes a heavy cut: Amazon takes 15% not including traffic fees; Best Buy / Target take 35% not including logistics and warehousing.

One audience member came with a specific question: their product is priced at $6,000–$8,000, and they've received 800 overseas waitlist emails. How to convert?

Duan Ran's answer was direct: "800 isn't enough. For this kind of product, I estimate it takes 100 emails to convert one." He gave two suggestions: first, stop using email outreach—pull them into a Discord community for daily communication; second, provide visualizable videos—work progress, CAD renders, vision videos. "Users willing to buy such high-ticket products have high income and high cognition, no time to read text. A two-minute video that makes them think this is so cool—then ask them to put down a deposit."

Chao Guang followed up: "The essence of complex products is converting one-time purchases into private domain. It's not about calculating traffic funnels—it's about capturing them within your information coverage range and reaching them multiple times through community."

After Entry

Returning to the opening scene.

One person carrying a dozens-of-kilograms prototype, chatting with strangers on New York tennis courts. Another person taking photos and posting to Moments in a small Silicon Valley venue. One looks clumsy, the other clever—but they're calculating the same account in their heads: the day hardware is sold is not the endpoint, it's the starting point.

Huaqiangbei makes hardware production costs lower and lower. Mold opening for 100,000-plus yuan, pressing some payment for mass production, and a small device that can record, photograph and connect to the internet is built. As hardware becomes cheaper and more similar, only two things remain truly valuable: how deeply you understand user interaction scenarios, and how thick the software system you build based on that understanding is.

Chao Guang built the teach-compete-practice system because a user asked "can I never win?" Duan Ran built a 24/7 AI hardware that automatically fills in information because he saw salespeople still having to manually organize meeting notes after talking with clients.

Anyone can afford the entry ticket. The game after entry has only just begun.

More Dialogue Details

Guests:

Yisi Tennis Robot Co-founder—Chao Guang

Xingqiong Fangzhou CEO—Duan Ran

Moderator: InnoVoxa Technology Brand Going Global Partner—Michael

Michael: My name is Michael, from a Singapore company called NW. We're an organization that helps Chinese consumer brands enter North American offline supermarkets. Before our roundtable begins, I'd like to ask both guests to take about a minute to introduce themselves and their company's business, okay? Chao, you go first.

Chao Guang: Hello everyone, I'm Chao Guang. What we're making now is tennis robots, which broadly belong to scenario robots. Humanoid robots are a more generalized, fully intelligent existence, but scenario robots lean more toward intelligent devices for a specific scenario. So first, the difference between us and general robots is that we're scenario robots. Second, why do we need to do intelligent upgrades on existing automated equipment? Essentially, we want to shift from a pure tool attribute to a consumer product attribute, thereby expanding the circle and satisfying and providing greater user value. Third, our previous team made various service robots, so we went from pure productivity tools to a robot that combines productivity tools with emotional value, lean more toward consumer products. So we mapped out seven vertical depths across technology, product, supply chain, interaction, emotion, data and so on, making full-market judgments from different vertical depths, and then selected this track. I myself was previously in investment, mainly investing in smart hardware—that's roughly my background.

Duan Ran: Let me briefly introduce myself too. I don't know if there are friends here from companies like Huawei, Tencent, Alibaba. Previously, we mainly provided solutions for large companies similar to Huawei, ByteDance, Alibaba, including the Shenzhen Science and Technology Museum, McKinsey, PwC—this series of large companies. Now B-end is hard to do, as everyone can understand, so we're now making our own products. Our AI hardware is also in cooperation with Volcano Engine, yes, and their Official Account also reported on it. We mainly solve one thing: frankly, everyone now wants to intelligentize older-generation equipment, but can't use it, can't get it working. This thing requires you to fill out forms yourself, write a bunch of background information. What we focus on is 24/7 recording and 24/7 shooting, helping everyone automatically fill in physical world information, so that what you ask is what you get—that's it, I'm done.

Michael: Duan has explained the logic behind what he's doing, but could you also introduce the product form to everyone? Which hardware form did you choose?

Duan Ran: We're not making a single hardware product, because Shenzhen's Huaqiangbei is now very good. If you're local in Shenzhen, you know that mold opening costs aren't very expensive—maybe just over 100,000 yuan, yes. And for mass production, pressing some payment for goods, everyone can afford it. Under these circumstances, we actually include pendants, glasses, rings, necklaces—making multiple AI hardware products and a model interface layer.

Michael: The title of our roundtable today is "From Arena to Market: The Going-Global Playbook of Vertical-Scenario Hardware." I think there are three perspectives we can explore. The first is of course the vertical-scenario entry point, which is our product form. So on this perspective, I'd like to ask Duan first. We know that making AI wearables—whether AI glasses or AI pendants—in 2025 and the first half of 2026 is actually a very crowded track. Overseas, Meta mentioned an AI pendant called Limitless, right? Domestic devices include Rokid and so on—everyone has their own different approaches. So what's the logic behind your choice to enter this scenario? Is it targeting a specific demographic, a specific usage action, or a specific wearing culture?

Duan Ran: The question actually comes down to two points. First, so many companies are making AI hardware now, which shows it's a hot track, right? A hot track means there's money to be made, traffic and customers. Second, it's not yet like phones and watches with shipments of tens of millions or hundreds of millions of units, which means the ceiling is high but there's no leader yet. So this is easy to say—a track with a high ceiling and currently no leader, so everyone's competing for territory, and we should hurry over to get a piece of the meat. The second part is the specific demographic. We actually target people with high offline information density, like sales, like consulting—people who communicate with others every day. Because people like this may have high or low salaries, but they're hardly from coding backgrounds. Under these circumstances, programmers have GitHub as reference cases for collaboration, but these people don't have such a product that can take one thing—meeting materials from chatting with clients—and directly close the loop with the entire company's internal documents, letting them directly deliver a proposal after chatting with clients. Yes, what we want to do is simply this: let a salesperson do not just sales, but also pre-sales consulting (SA), customer success, and solution architecting—all customer support positions except writing code.

Michael: Duan, can I understand it this way—you chose a relatively mature hardware track, believing that elephants can't crush ants, right? The top players have their volume, and you making things like pendants and glasses can definitely find a more niche vertical, and can definitely move some volume yourself—is that the understanding?

Duan Ran: That's the idea.

Michael: Then second, I'd like to ask Chao the same question. Tennis robots in recent years, especially on some crowdfunding platforms like Kickstarter, have actually appeared a lot, right? I looked it up—the unit price is basically between $600 and $1,500, and the highest one should be a crowdfunding that just ended last month, raising about $2.4 million. Traditionally, there have also been ball machines—although they might be called tennis robots, everyone's perception of traditional ball machines is that the price is like a smartphone but the configuration is basically a landline, right? There's this common problem. From your perspective on AI tennis robots, you more have the sparring function, right? Do you think its essence is a sports training device, or an early commercialization sample of mobile embodied intelligence?

Chao Guang: The big logic is definitely that we think AI is rising with the tide, but the barrier is actually in hardware—the complexity of hardware and the entire systems engineering is a barrier. Second, for tennis—although some folks have already made products on Kickstarter with good results—what we want to make is not a better tennis ball machine, but a consumer-grade product.

Michael: What's the difference between these two?

Chao Guang: A better tennis ball machine is essentially still a tool attribute, a functional attribute. The penetration rate of ball machines is actually very low. The first reason is that they can't provide more value; the second reason is that they can't achieve higher usage frequency. So most ball machines, once bought, gather dust. If you play tennis, you can actually perceive what kind of person can use a tennis ball machine. We previously made various service robots—lawnmowing, pool cleaning, commercial cleaning—so on one hand, we've accumulated capabilities in perception and mobility; on the other hand, many of the products we made before were tool attributes. For example, pool cleaning—you're just getting the pool clean, and its ceiling is the value of a person cleaning it manually. So after exploring for a while on service robots with tool attributes and productivity tool attributes, we found that we should look toward more emotional value, or an integrated product that combines emotional value and productivity value. So when looking at the entire full-scenario robot landscape, we felt tennis was a very good scenario—precisely applying the accumulated mobility and perception capabilities to ball machines, deepening their interaction with people. From interacting once per match in the past, to now interacting on every rally. This deepened interaction amplifies user value—from an average user value to a point with particularly high highlight moments. This way, when expanding the circle, some users may not be enthusiasts, may not play often or with high frequency, so you need the highlight value in one or two sessions to be high enough to break through. To summarize: what you buy with a tool-type product is actually its value ceiling, but what you buy with a consumer-grade product is only one reason—'I want it.' This is the difference we want to make, and why we're doing this even though ball machines already exist and are doing well.

Michael: So what you're saying about developing toward emotional value is also a way to maintain stickiness with users. You just mentioned the original intention behind doing this, which leads to our second question: how do we cultivate the first batch of seed users from 0 to 1? Taking this opportunity, I'd like to ask Chao. We know that tennis, both domestically and overseas, is a middle-class sport. This group has two typical characteristics: first, relatively high consumption level; second, they're a very picky group. How did you cultivate your first batch of overseas seed users? Did you find B-end club coaches or KOL players, or were they the first early adopters from your early crowdfunding? Based on their feedback, did you adjust subsequent product definition?

Chao Guang: First, we chose an interest-based track. Interest-based tracks aren't as large as general tracks, but they have their own barriers, which is very conducive to building your own base—especially digging the moat of technical depth and product depth. So for our seed users, first of all, we ourselves are typical users—we ourselves are tennis enthusiasts who have been playing all along. So for interest-based tracks, your own personal identification is actually very important, just like if you don't actually use a hardware product, your perception is a bit off. Second, in this process, you must perceive users' emotions—you need to get that emotional resonance before you can improve and solve their pain points. We should be one of the few who, before crowdfunding, took prototypes to the U.S. for a journey. Before going on crowdfunding, two of us took a prototype weighing dozens of kilograms to the U.S. and ran around for about half a month. We did something very interesting at the time—we did 'street challenges.' We'd take the machine directly to tennis courts and ask people playing tennis if they wanted to try our machine. We went from Los Angeles to New York, from schools in Los Angeles to the U.S. Open finals site in New York. Before crowdfunding, about 200 people should have used our product and given us a lot of feedback. Third, before defining early products, we actually spent more time thinking about what product definition changes are needed to shift from productivity tools to vertical-track robots.

Chao Guang: Actually, when making products, we borrowed a lot from previous lawnmower experience. For lawnmowers, everyone already predicted pure vision back in 2020, but it's been six years now, and I think we're still in the LiDAR era. So don't overestimate technological progress, because you are, after all, a mature ToC product. Additionally, early users' suggestions were very helpful to us, but they more helped us advance one step within the architecture of our original product definition. For example, early on there was a very interesting question—a user asked us: 'If I play tennis with this robot, does that mean I can never win?' This triggered a lot of our thinking and exploration. If you play against a robot and can never win, that means it's an endless thing. So based on this, we extended out a 'teach-compete-practice system'—your interaction is purposeful, embedded into the entire system of teaching, competition and practice. Currently, only our company's product has this on the market, and based on this system we implemented subscription-based pricing. This should be the first functional-transformation hardware product to charge for software.

Michael: Chao, from your expression, Yisi is a company that pays great attention to interacting with frontline consumers—including going to the U.S. for roadshow activities before crowdfunding to find the first batch of seed users, then improving product definition based on feedback. I think this is worth everyone learning. I'd like to ask Duan the same question. Everyone knows that the hardest barrier for AI wearables going global is wearing culture. Whether AI rings, pendants or AI glasses, overseas culture is different from domestic—especially when your glasses or pendants have cameras, this kind of thing is particularly sensitive in Europe and America. During the product definition stage, did you first make it for Chinese users then adapt for overseas, or the other way around? And finally, how do you pre-design the going-global version in hardware design?

Duan Ran: Pre-designing the going-global version. Before answering this question, let me make a suggestion—think about it, because everyone sitting here are real friends, and I believe you may have come with some questions and ideas to listen. Let's leave the last five minutes for everyone to ask questions, and we'll try to answer what we can. Back to this question—it actually involves two major points. First, the issue of product design positioning caused by differences between Chinese and American culture. I think this is hard to solve, just like Beijingers and Shenzhen people have different habits—Shenzhen eats roast goose, Beijing eats instant-boiled mutton. So between China and the U.S., we generally have specific versions—your Chinese version won't directly migrate to the U.S. Based on each country's user cultural habits, we'll design different appearance shapes, because in the U.S. many religious factors influence things—if you design a cross and it breaks, you don't even need to go to market, but domestically there aren't really many taboos in this regard. Second, on functional design, this is actually relatively consistent—whether for privacy protection or security considerations, I think it's about giving users more choices. Just like earlier this year, OpenClaw, right—you can deploy it in the cloud to execute tasks, or you can put it on your local end and give it all your passwords and chat records, though you also bear the risk of privacy leaks. The hardware logic is the same—give users a choice: how much privacy right are you willing to give up to get better service? This is worth weighing and providing to users. Some users want higher efficiency and don't care—add visual shooting, as long as I don't take it to the bathroom. Some are sensitive to visuals and only use audio. Some don't want information sent to the cloud at all—we can also provide local or edge-side services, letting information go to their own NAS or computer for processing. Various solutions have pros and cons—must be open for users to choose. Believe that users today are smart, with independent choice and thinking ability. You can't deceive them—appropriate promotion is fine, but you can't lie to them. This is the design rationale for appearance and functionality.

Duan Ran: The second part is overseas data buying points. Small companies have small-company playbooks—like going on Kickstarter or Indiegogo crowdfunding, buying some traffic is normal; large companies just retweet on Twitter. I think what Chao just said is very good—you can take the product to Silicon Valley and hold a launch event. It's actually not expensive—about $10,000, or even $3,000 to $5,000 can do it. Take a few photos, buy some media, promote it on the Chinese internet as 'launched in Silicon Valley'—sounds very high-end. Then find people on Twitter to write some positive reviews, so export-for-domestic-sales is still a viable path. I find that many good hardware companies in Shenzhen now go global through Kickstarter but don't take products over themselves, still going the TikTok video promotion path. I think if your product is under $150 or $100, you can go straight to TikTok or Twitter; but if the price is relatively high, I suggest you hold a launch event in Silicon Valley, or go to CES, IFA, even grab a photo with Cook and Zhang Yiming at Apple events—the product gets relatively strong endorsement and will have better sales performance.

Michael: Duan's methodology is actually very down-to-earth and very replicable. I think it's a very good path for early-stage enterprises—don't think so grandly, don't think about making a standardized thing that can be used globally. The most important thing is to do some localized operations. Last question—we just talked about from 0 to 1, and the future will definitely face the process from 1 to N, which is about global channels, pricing and sustainable purchase discussion. I'd like to ask Duan: how do you price the AI pendant you're mainly promoting recently—competing on features or competing on price? Second, how much premium do you think overseas consumers are willing to pay for Chinese-brand AI wearables?

Duan Ran: How about we take it one question at a time—my context window is limited, I don't have an H200 in my head.

Michael: First question, do you think different-form products compete on price or features?

Duan Ran: Normally speaking, for Shenzhen or Chinese companies, it's 'want it all'—want good performance and want to compete you to death on price. But this can be looked at from two aspects: first, you use low prices—for example, we price at $128 overseas, domestically it's a third cheaper. Overseas, around $100 is a comfortable range for users, they don't have much cognitive burden (I mean the U.S.—the U.S. and Northern Europe are at the same table, Southeast Asia is actually at the same table as Russia). Here, when calculating costs, hardware gross margin needs to be at least 60%, normally 80%. Because for export, you not only consider tariffs, but also shipping, warehousing, after-sales repair (mostly replace-don't-repair)—if gross margin can't reach 60%, you're even losing money. So I suggest not only high gross margin, but preferably with some SaaS subscription, because the U.S. is very accustomed to SaaS subscriptions—ten to twenty dollars. Look at domestically, Doubao says it's free, but the bandwidth costs behind it are very high—it can't stay free forever. So adding some SaaS subscription fees like $9.9 can generate recurring revenue, which is a reasonable profit method. Second is a high-end product, like around two to three hundred dollars, which can have enough profit margin to guarantee the iterative R&D of next-generation products. This requires you to have relatively good traffic in the North American market—not just Amazon review manipulation, but more importantly making hit videos on TikTok to give users a very reliable feeling before they'll buy.

Michael: Actually Duan has talked about some very practical going-global issues. Everyone does more online Amazon—the platform itself takes 15% not including traffic fees; offline like Best Buy, Target take 35% not including logistics and warehousing. So going global still needs to be segmented by audience—there are hundred-plus-dollar ones and one-to-two-hundred-dollar ones. One last supplementary question: Duan, are you already doing Middle East going global?

Duan Ran: The Middle East is mainly through channel partner cooperation, because we haven't figured out online either. Many wearable devices in the Middle East are sold in offline stores—a family or community runs a store maybe for 20-30 years. Entering those small department stores to sell is relatively good.

Michael: Community retail—actually North America has that too.

Duan Ran: Also including chain retail. For North America online, we're not going through Amazon, because we're tech hardware—we still go through things like Kickstarter to create hits. If you're traditional hardware like power banks, Amazon is better; but if you have sci-fi (Sci-Fi) concepts, like AI Agent, brain-computer interfaces, I suggest going through Kickstarter.

Michael: Last question for Chao. Tennis robots are now generally priced at 5,000 to 10,000 RMB, a typical high-ticket, low-frequency product. Is your customer structure more B-end clubs or C-end individual players? How do you differentiate channels domestically and overseas?

Chao Guang: Overseas and domestic, our scenarios have some differences. Overseas, especially in North America, there are many courts, and transportation is also door-to-door (garage to court), so the main contradiction overseas is 'finding people'—need to find playing partners and smart companions, so overseas we're mainly ToC. Domestically, because population and courts are concentrated in coastal areas, the court issue is the main contradiction. This is also why a new format called 'ball machine gym' has extended out in recent years—solving the tennis-playing problem in office buildings. So for the domestic court contradiction, we mainly cooperate with channel partners, going through professional channels (venues, clubs, training camps). Besides the ToB and ToC difference, overseas there's also a new model, somewhat like a Global Store. Because the product emphasizes hands-on experience, you don't necessarily need to find big KOLs—instead find coaches who have courts to do online-offline combination, directing hesitant users to offline demos to drive transactions, and offline outlets can also get profit sharing—this is our differentiated approach.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260530-01html
On-site reading page: https://ffcap.cn/en/research/src-20260530-01html