--- title: "Three Months After OpenClaw Went Viral, Only Three Business Models Have Truly Worked" author: "Unique Research" sourcePublication: "Unique Research Substack" originalPublishedAt: "2026-06-09T11:00:46+00:00" canonical: "https://ffcap.cn/en/research/src-20260609-03html" source: "https://uniqueresearch.substack.com/p/src-20260609-03html" language: "en" --- # Three Months After OpenClaw Went Viral, Only Three Business Models Have Truly Worked _Original · Unique Research · 2026-06-09_ _Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the complete narrative analysis of three business models, the highway/transportation metaphor, the core viewpoint summary, and the full roundtable Q&A transcript. All named speakers, companies, roles, numbers and claims 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 Observation After OpenClaw, Who Is Actually Making the Money from AI Agents? Consensus of Three AI Entrepreneurs: Big Players Eat the Meat, I Pick Up the Crumbs They Drop on the Floor " The real commercial opportunity of AI Agents is not in the hands of those who build Agents, but in places Agents can't reach. At the start of 2026, OpenClaw completed Chinese market education in an almost absurd way. It did nothing. No advertising, no PR, not even a proper Chinese website. But in Shenzhen, Huaqiangbei geeks took old Lenovo boxes, stuck printed lobster stickers on them, installed preloaded systems, and sold them for 1,700 yuan each—with deliveries stretching past Chinese New Year. Sun Xuefeng was one of those who spent 1,700 yuan. He's the CEO of Xiaoshui Intelligence (小水智能), based in Beijing, selling 1 million children's watches per year. "We folks in Beijing are the ones getting fleeced," he self-deprecatingly said. "Shenzhen launched the 'lobster shell' immediately." But the 1,700 yuan was well spent. Because from this "lobster," he saw a fact most people overlook: the real commercial opportunity of AI Agents is not in the hands of those who build Agents, but in places Agents can't reach. Big Players' Blind Spots Are Small Teams' Livelihoods Sun Xuefeng told a very typical scenario. After Chinese New Year, a large number of central/state-owned enterprises and government clients came to him saying they wanted to use "lobster," but internal security regulations wouldn't allow installation. "Leadership feels OpenClaw is unsafe, data will leak out." The demand is real, but supply was cut off with one stroke. His approach was simple and direct: on the open-source foundation of OpenClaw, he built a "Honeycomb" (蜂巢) system. Architecturally, it implements permission tiering—the "queen bee" (蜂王) manages global security policy, the "queen consort" (蜂后) manages permission allocation, and Agents for specific business scenarios are called "worker bees" (小蜜蜂)—a customer service role can only do customer service things and can't touch other data. The entire product architecture was built by his CTO in two weeks with his own "lobster." There's an easily overlooked detail here: big players are not incapable of doing private deployment—they're unwilling to bend over for individual clients. Central/state-owned enterprises want dedicated service—on-site debugging, hand-holding teaching, on-demand customization. ByteDance and Tencent salespeople won't squat in a client's server room modifying configs for a few-million-yuan deal. But for a team of dozens, a million-level private deployment contract is a year's livelihood. Sun Xuefeng extended this logic further to Huaqiangbei. He went to Shenzhen and found that traditional consumer electronics bosses making recording pens, speakers and watches—with single-product shipments in the millions—don't have a single AI engineer on their teams. "These people don't not want to do AI; they can't afford to hire an AI team." His solution: deploy the AI middle platform for free, help them build consumer-facing Agent scenarios, and take a cut of incremental revenue. "Facing this wave of AI and competition from big players, mid-tier enterprises must band together. Hardware has the Huaqiangbei brothers, AI industrialization has Beijing's talent—combine them and go global." Behind this statement lies a more essential judgment: the technology window is only 3 to 6 months, but a service moat can last for many years. Large models iterate rapidly, and any model's technical advantage will be caught up to on a six-month scale. But the service relationships you build with clients, the depth of your understanding of industry scenarios, and every line of customized code you leave in a client's server room—these are the true moats. The Community Grew Big, but the Money Came from Another Business Bruce may be one of the people closest to the OpenClaw ecosystem in the past three months. He started playing with Clawdbot in January, held Asia's first offline Meetup in Hong Kong in early February, then went on to Shenzhen, Shanghai, Beijing, Tokyo... Next week at the BEYOND Expo in Macau's Venetian, he's hosting ClawCon, the official OpenClaw community event. But if you ask him how much money he's made from the community, he'll tell you: the community itself doesn't make money. "We also made a one-click installation tool in February, but after Chinese New Year when we saw big players like Tencent also starting to offer lobster-installation services, we directly stopped," Bruce said frankly. Going head-to-head within the big players' firing range is a dead end. What truly showed him opportunity was the same pain point repeatedly mentioned by community users: some models work in one region but not in another. He calls this the "cross-border e-commerce" of the AI era. Upstream model companies are like different brand SKUs—GPT, Claude, DeepSeek, Zhipu—each leading by 3 to 6 months, with huge differences in price and performance. Downstream demand is varied: some want the most expensive inference quality, some want the cheapest batch calls. Users in some regions simply can't access Claude at all—like consumers in certain countries who can't buy a certain brand. So he pivoted from community operations to an API relay station. "Essentially it's a Marketplace—solving matching, payment and scenario complexity problems." He also admits there will be shakeouts and price wars in the short term. But he's betting on a structural judgment: niche services that large model companies can't swallow are the structural opportunity for entrepreneurs. Large model companies want all developers to directly call their APIs, but in the real world, clients want "help me get it done" rather than "here's the interface, you figure it out yourself." This is essentially the same logic as Sun Xuefeng's—both are doing "the last mile." A 6-Person Team Produces 200 Hours/Day, Not Because AI Is Strong, but Because It's Used in the Right Place Wang Weiyang's case is the most extreme. Zenkit, 6 people, office in Shenzhen, making overseas ToC software and North American ToB productivity tools. The team keeps 3 OpenClaw instances, all given female names. These three "virtual employees" each have their roles: one monitors GitHub trending projects, one monitors Steam trending games, and one serves as a stock query bot in a 10,000-person community. He mentioned a metric called "parallel multiplier"—actually working 8 hours at the computer, the system shows output equivalent to 200 hours of work. But he specifically emphasized one thing: Token consumption is not the measure. "The Token consumption generated by community questions is extremely low—compared to AI coding it's a drop in the bucket. But in operations and customer deployment, it brings a hundredfold or thousandfold efficiency improvement. One person serving a 10,000-person community—that's true leverage." This statement precisely punctures an industry bubble: everyone is staring at model capability, Token prices and inference speed, but what truly determines whether AI Agents can land is not how strong the model is, but where you use it. Wang Weiyang did two things to support this judgment. First: he rewrote OpenClaw's original Gateway, Database and Memory entirely in Go and Rust. The original TypeScript had poor performance under high concurrency; after the rewrite, it ran for two and a half months with zero downtime. This action shows he's not a "just use it" user, but a "modify it" Builder—which happens to be something big-company employees rarely do. Second: going global. The same product—28 yuan per month in China, 1,000 to 1,500 yen in Japan, $10 in North America. A price difference of 3 to 5 times. "Supply in Japan is severely insufficient. What you think is brutally competitive in China, taking it to Tokyo is an overwhelming advantage." Bruce confirmed this when holding events in Tokyo: Japanese users are still buying books to learn "how to use ChatGPT to make PPTs." It's not that Japanese people are slow—it's that market simply hasn't experienced the level of competition in China.Just take a domestic AI application solution that's already been validated, and it's a scarce product in that market. The Highway Is Built—Who's Running the Transport? String these three cases together, and a main thread emerges. Qualcomm announced at Computex in Taipei that "2026 is the year of the AI Agent." Jensen Huang wants to build Agent factories. ByteDance placed orders for millions of chips. NVIDIA, Intel, Arm—chip giants reached a rare consensus in the same time window: AI is shifting from "answering questions" to "doing things for you." They're building the highway. But after the highway is built, the ones who truly make money are not the road builders, but the people running transport on the road. Sun Xuefeng's private deployment is the dirty work big players are unwilling to bend over for. Bruce's API relay is the niche market large model companies can't cover. Wang Weiyang's going-global plus tech-based leap is using capabilities honed through domestic competition to attack markets with insufficient supply. The three people's paths are completely different, but the underlying business logic is identical: don't fight big players head-on on the open battlefield. Go to places big players look down on, can't reach, or are too lazy to manage—use your dedicated service and scenario understanding to eat the cake crumbs they drop on the floor. Those crumbs mean nothing to giants. But for a team of dozens, they're everything needed to survive. And the biggest structural advantage of this wave of Chinese AI application-layer entrepreneurs is—the domestic market is too competitive. Capabilities honed through competition, taken to Tokyo, Singapore, Bangkok, are crushing-level. Bruce said something worth savoring: "Don't just stare at places that are too competitive. Look farther away. Your capabilities can easily find demand elsewhere." The most valuable question in the Agent era is not "how strong an Agent can I build," but "how many places big players can't reach can I deliver Agents to." Core viewpoint: The commercial opportunity in the AI Agent era lies not in "building Agents" itself, but in "the last mile"—private deployment, API relay, going-global leapfrog. The niche markets big players can't reach, look down on, or are too lazy to do are where small teams can truly strike gold. More Conversation Details Guests: Zenkit Founder — Wang Weiyang OpenClaw Asia Founder — Bruce Xiaoshui Intelligence CEO — Sun Xuefeng Moderator: HKU ICB Visiting Mentor — Chen Yunfeng Tim I. Guest Self-Introductions and Project Progress Chen Yunfeng: I'm very happy to host this roundtable today. I'm in Guangzhou, with a joint venture with Baidu, mainly serving the 2B and 2G AI track in Guangzhou. At the same time, I serve as an AI mentor at several universities including Sun Yat-sen University and South China University of Technology. Today we have one Beijing guest who couldn't attend for personal reasons, so our three guests will discuss together. First, Mr. Sun, please introduce yourself. Sun Xuefeng: I'm Sun Xuefeng, CEO of Xiaoshui Intelligence. Xiaoshui Intelligence has recently had two core developments in the AI field: Consumer electronics hardware (children's watch track): We currently achieve annual shipments of 1 million units. We've introduced AI Agents and AI-native systems into children's watches, achieving complete AI-native interaction. We're currently researching Agent-ification of children's watches, attempting to bring OpenClaw into the watch environment so children can raise their own AI agents. We've been competing with big players in this track. The teams behind big players are usually just a few dozen people; if you can make investment more precise and bring in the world's best technology, you can completely achieve surpassing them. "Honeycomb" security system for central/state-owned enterprises: During Chinese New Year, we noticed the huge commercial opportunity of OpenClaw, but found that government, central/state-owned enterprises and classified enterprises weren't allowed to install it directly due to security concerns. For this reason, based on OpenClaw we built a security system exclusively for these enterprises—the "Honeycomb" system. The system establishes "queen bee" and "queen consort" to control permissions and security, with "worker bees" at the bottom specifically responsible for customer service and other specific business. Chen Yunfeng: Mr. Sun's enterprise in Beijing is large-scale and has very good profit margins. Next, let's give the floor to Bruce, founder of OpenClaw Asia. Bruce: Hello everyone, I'm Bruce. I started researching OpenClaw in January, initiated Asia's first offline event in Shenzhen in February, then held multiple community events in Shanghai, Beijing and Tokyo, Japan—even meeting the "father of lobster" Peter in Tokyo. Next week at the BEYOND Expo in Macau, I'm also hosting the official OpenClaw community event. My own company is in Hong Kong, with a technical background. I previously worked at Microsoft and Alibaba, returned from the US 10 years ago to do FinTech entrepreneurship, and later did AI going-global applications in Hong Kong. After OpenClaw exploded, we decisively stopped our ongoing projects. In the past, finding PMF (Product-Market Fit) for entrepreneurship required a lot of effort to educate the market, but now OpenClaw has completed market education across the entire network. We found this is an excellent growth channel—entrepreneurs just need to do customized screening for clients like Mr. Sun. In a moment I can share in depth my observations in Hong Kong, Southeast Asia and Japan. Chen Yunfeng: Later we can expand around two points: first, your observations as a "global digital nomad" in Hong Kong, the Greater Bay Area, Japan and Southeast Asia; second, the commercial opportunities and dividends brought by OpenClaw. Next, Mr. Wang. Wang Weiyang: Hello everyone, I'm Wang Weiyang. I currently mainly do going-global software, with a company in Hong Kong deeply engaged in overseas general-user communities doing ToC software, while simultaneously operating a ToB productivity tool project Zenkit in the US. We're currently a small team of 6 people, and OpenClaw is an indispensable piece of our team's puzzle. I started fully researching OpenClaw in February and deployed it on a server in my apartment. Internally, we keep three OpenClaw virtual assistants that help us research GitHub trending and Steam trending projects every day, continuously monitoring public preferences for software, technology and games. Externally, we serve the WaytoAGI community of over 10,000 people—many users like to use my "lobster" to query stocks. Recently with the spread of high-concurrency models, we've also integrated these high-concurrency capabilities into the system. Chen Yunfeng: You mentioned a "parallel multiplier" monitoring metric in the North American productivity tool—this is equivalent to using AI to let 2 people do the work of 50 people? Wang Weiyang: Yes. We have a parallel multiplier monitor for coding and work. This metric is very clear—for example, you actually work 8 hours a day, but the system monitor shows you've produced output equivalent to 200 hours of work. That's the efficiency amplification brought by AI. II. City AI Genes: Differences in Atmosphere Between Beijing, Shanghai, Guangzhou, Shenzhen and Overseas Chen Yunfeng: Today the four of us represent four cities: Guangzhou, Hong Kong, Beijing and Shenzhen. I'm very interested—how do you view the AI entrepreneurship atmosphere in different domestic cities, and even overseas cities? Sun Xuefeng: I'm in Beijing, but I also frequently come to Shenzhen for exchanges. Beijing atmosphere: Beijing AI enterprises, when starting out, tend to value industrial opportunities, future listing value and grand visions—one could say stronger "To VC" capability. Shenzhen atmosphere: Shenzhen is extremely pragmatic and fast-acting. As soon as OpenClaw went viral, Shenzhen immediately launched "lobster shell" hardware boxes with systems preinstalled for direct sale. We in Beijing also need to learn from this pragmatic geek spirit. Chen Yunfeng: Beijing indeed has inherent political and capital advantages, suitable for high-profile, "end-as-starting-point" 0-to-1 work. But for scaling up, you often need to come south or to places with better policies. Bruce, from a global perspective, what do you think? Bruce: I have deep experience with this topic—developers and market performance in different cities are completely different: Beijing: Event registrants have extremely high education and IQ levels, with many PhDs. It brings together top AI large model companies, discussing world-changing problems. Shenzhen/Guangzhou: Extremely pragmatic, the core is "how to make money first." At the Shenzhen event, many super-individuals directly sold courses or offered on-site installation services. Hong Kong: Relatively few engineers, but many people with finance and Web3 backgrounds. Shanghai: We held an event at Pudong Software Park, where people tend to do deep business micro-innovation within enterprises. Overseas markets (Japan and US): Japan is an excellent enterprise-service going-global market, with high user willingness to pay and loyalty, but domestic AI supply is severely insufficient—people still learning ChatGPT through traditional methods (like buying books). And in the US, outside Silicon Valley tech circles, the general public actually doesn't care much about OpenClaw. This shows that AI applications in mainland China are deeply competitive—everyone can broaden their horizons and go overseas to find overwhelming-advantage opportunities. Wang Weiyang: Yes, North America and Japan indeed have better payment acceptance. The same product—domestic monthly subscription might only be 28 yuan, in Japan it can sell for 1,000 to 1,500 yen, and in North America $10. From the perspective of free team growth, this wave of OpenClaw dividends is too competitive in Beijing with limited spots. But Shenzhen is extremely inclusive—our Shenzhen office has expanded several times over. In Shenzhen, even a small team of a few people can live very well through Bootstrap (self-funded entrepreneurship). Beijing is suitable for high-profile approaches, while Shenzhen is more suitable for free growth. III. Surrounded by Giants, How Can Startups Build Moats? Chen Yunfeng: Mr. Sun, please talk in depth about your product form, sales channels and specific market approach surrounded by traditional big brands. Sun Xuefeng: We've been deeply engaged in the communications industry for nearly 10 years. Our core approach: private deployment + scenario co-creation + hardware empowerment Customized "Honeycomb" system: We hand security-related permissions to enterprise management ("queen bee" and "queen consort"), limiting unwarranted consumption of high-priced APIs. This system was led by our CTO and completed a prototype in only two weeks combined with OpenClaw. Dedicated service and client co-creation: Big players can't do dedicated service for vertical enterprises, but we can directly teach client teams how to play with AI and use "worker bees" to co-create business scenarios. Empowering traditional hardware makers: Huaqiangbei has a large number of traditional consumer electronics enterprises making recording pens, speakers and watches—with shipments in the millions but lacking AI teams. We deploy the AI middle platform for free via private deployment, helping them build consumer-facing Agent scenarios and share subsequent incremental revenue. In the wave of mid-tier AI entrepreneurship, we must band together. Combining Huaqiangbei's hardware manufacturing capability with the north's engineer talent dividend to go global—that's a huge opportunity. Chen Yunfeng: The model Mr. Sun shared hits the nail on the head. Technology usually only provides a 3 to 6 month window; the true moat must be formed through action, capital accumulation and user accumulation. Bruce, you built the OpenClaw community—what commercial path ultimately emerged? Bruce: Building the community was initially unintentional, but in the process we discovered users' real pain points. After Chinese New Year, seeing big players like Tencent also launching one-click installation tools, we decisively abandoned the original tool development and pivoted to the API relay station (API Router) business. We found the API relay station is essentially the "cross-border e-commerce model" of the AI era: Upstream: Each major model is like a different brand SKU, with each model's leading advantage lasting only about 3 to 6 months. Downstream: Client needs are varied, and different regions face network or payment restrictions. What we do is a Marketplace, solving matching, payment and scenario complexity problems. The industrial intelligence era has just begun, 90% of applications will be rewritten, and niche services that large model giants can't swallow are the dividends for entrepreneurs. Chen Yunfeng: Relay stations have indeed been very popular recently. Do you think this is a short-term dividend or a medium-to-long-term opportunity that can last one or two years? Bruce: In the short term it will definitely experience shakeouts and price wars, but in the long term it's a structural opportunity. Entrepreneurs need to find their comparative advantage, survive first, and build it into a true Marketplace. IV. Summary: The Essence of OpenClaw and the Productivity Revolution Chen Yunfeng: Finally, Mr. Wang, please summarize in one or two sentences: What exactly is OpenClaw? What dividends has it released in your business model? Wang Weiyang: From the application level, OpenClaw helps us do all the dirty work. It connects to over 30 IM (instant messaging) platforms globally, becoming a bridge between large models and instant messaging, capable of doubling and amplifying the capabilities of founders and engineers. It greatly lowers the threshold for ToB technical services. Previously clients felt integrating APIs and SDKs was too high a threshold; now through the Agentic deployment mechanism, you can directly pull it up with commands. Technical refactor: The original OpenClaw was written in TypeScript, with performance and stability struggling under 10,000-person community concurrency. Our team spent a week rewriting its Gateway, Database and Memory entirely in Go and Rust. After the rewrite, it never crashed for two and a half months, stably supporting 10,000-person group chats. Efficiency dividend: Users interacting with OpenClaw in Feishu groups to query stocks generate Token consumption that's a drop in the bucket compared to our AI coding, yet it greatly frees up our operations staff's energy. It brings a hundredfold, thousandfold leverage improvement in operations and customer deployment. --- Original publication: https://uniqueresearch.substack.com/p/src-20260609-03html On-site reading page: https://ffcap.cn/en/research/src-20260609-03html