Original · Unique Research · 2026-04-10
Editor's note: This edition preserves the original author's narrative and the panelists' statements, not the translator's firsthand experiences. OODA appears in the original headline but is not explained in the extracted text; no missing explanation has been invented, and the source images still require inspection. The narrative credits the closing “let things play out” remark to Hu Bin, whereas the transcript credits it to Cao Wei. Both attributions are retained pending verification. Relative dates refer to the original April 2026 discussion. Revenue-growth examples and the hypothetical ten-million market do not specify a currency, so none has been added. Fund and portfolio figures, factory and labor conditions, technical references and anonymous employment anecdotes remain source-attributed. Broad market comparisons, investment forecasts and cross-border structuring suggestions are the speakers' views, not independently verified conclusions or legal or investment advice.
Unique Awards
At AI's Current Pace, Rushing to Judgment Is Often a Mistake
OpenClaw, one-person companies, overseas expansion, and the China–US gap: this roundtable explored almost everything Chinese AI founders care most about.
"
Let things play out a little longer.
A roundtable at Hangzhou AI WEEK discussed the topics that concern Chinese AI founders most: OpenClaw, one-person companies, overseas expansion, and the gap between China and the United States.
Let's start with Wayne, Managing Partner at Argo Venture Partners—a new fund with a very particular focus.
He said they invest only in Physical AI and do not touch other sectors.
What is Physical AI? He divided it into three layers:
The first is Intelligence. He said the “cerebellum” is now quite good, but the “cerebrum” remains very weak. That is a priority.
The second is embodiment. It depends entirely on China's supply chain, making it a good opportunity to expand overseas together. Mr. Cao of Fuyao Glass has been operating in the United States for 10 years. The factory is profitable, workers' wages have risen, and there is still no union.
The third is safety and security. OpenClaw was rather wild when it first appeared. It is better now, but safety and security are the next major challenge—and a major opportunity.
Wayne's logic was clear: Chinese founders and engineers should aim to win the North American market. Not intensify competition at home, but earn dollars abroad.
Everyone Is Getting into “Aquaculture”
The discussion then turned to OpenClaw.
Shen Dongliang put it most directly: “Everyone is getting into ‘aquaculture.’”
He thought the biggest change was that applications might no longer be built for people, but for machines.
Previously, the debate was whether to sell tools or results. Now that OpenClaw has appeared, many companies are trying to deploy their own versions. They can solve relatively shallow application needs themselves. When the barrier is not deep enough, it can easily be flattened. Every niche will become more competitive, making differentiation harder.
What, then, constitutes a barrier? An industry foundation plus accumulated, differentiated industry data.
Cao Wei added another angle. He said OpenClaw is itself a framework with three core elements: accumulating skill capabilities, aligning underlying context, and a mechanism for working 24 hours a day, 7 days a week.
His team's internal discussions noted that large companies iterate quickly on the framework itself: There is already a “battle of a hundred lobsters.” But the skill could become the real node where value resides.
It needs three elements:
First, long-tail know-how.
Second, resource paths.
Third, access permissions.
He said OpenClaw is currently like an “unfinished apartment,” still far from a place where users can simply “move in with their bags.” He had written more than a thousand lines of code for a skill himself, and configuring resource paths and networks was a headache. So far, the value it brings does not seem proportionate to the time it consumes.
There Will Be More One-Person Companies, but Not Every Company Will Become One
The conversation became livelier when it turned to one-person companies.
Piruze said she was completely certain that in the near future, one person would be able to build a technology company and launch a SaaS product or personalized software. A year from now, we would see more of these emerge in open source.
But she also offered a caution.
In enterprise services, when a founder brings a product to market, the buyer asks: “Can I build this myself?”
Tools now make building a software company much easier. But some software companies are evolving: moving into hardware, establishing data advantages, keeping data secure, and solving extremely complex workflow problems.
The implementation layer is becoming more creative. Enterprises want to move quickly and see AI's impact as soon as possible, while concentrating on what they do best. So, at least outside China, they are very willing to buy this software. Some of these companies can grow revenue from 1 million to 200 million within six months or a year.
Hu Bin's view was more direct.
“At the micro level, I don't know what to invest in right now. Things are evolving so quickly that our traditional way of working—meeting once a week—has been completely upended. The most rational thing to do now may be to let things play out a little longer and observe.”
“But that absolutely does not mean we are pessimistic. In fact, we are very optimistic.”
He said that after the Spring Festival, investors had to run OpenClaw themselves. Most had never done Coding or worked with servers before. But if this is supposedly something one person can do, failing to try it yourself means falling behind.
“I've run it myself, but found that the value for the cost is indeed not very high right now. Of course, some very capable people may already be much further ahead than I am.”
“Previously, people felt FOMO, afraid of falling behind at the starting line of a generational shift. But looking at how technologies iterate, there is no need to rush so much. People who move a little more slowly can still catch up and benefit from what others have done. We used to say Chinese companies worked hard and competed intensely, but now you cannot outcompete machine life through sheer effort.”
What Is Venture Capital Actually Investing In?
Tina asked a very interesting question.
If there will be so many one-person companies, how will venture capital firms find opportunities and judge whether they are investable?
Piruze answered candidly.
Her job is to find founders with immense ambition: people who want to build at scale and create great companies that endure across eras. That essential task has not changed.
Those companies may start with only two to ten people, and research and development costs may fall. But if they want to train their own models and own their own servers, those are still enormous costs. The challenge is not just building a product. Business development in a highly competitive market becomes much more complex, and implementing software globally still requires people who understand customers' problems and can persuade them to work together. So they need capital to grow faster.
Cao Wei said their approach was to “bet on both sides.”
Large-model teams themselves are already working on interactions with the most widely used software and the most frequent skill functions. But ultimately, user needs must lead. For things involving personal privacy, he wants local storage and local models, keeping personal and cloud-based work relatively separate.
For long-tail, unusual interactions with traditional software, even a large model needs a learning process based on a source of data. In those areas, the playing field is relatively more equal for large and small companies.
Wayne's view was more pragmatic.
“I think this round resembles the previous one, when people first built all kinds of small tools and automatic website builders. There will not be that much in the middle that can produce capital gains. We make money through capital appreciation; I cannot make money from cash flow.”
He said one-person companies were a good thing. The employment system is an outdated model built for manufacturing 100 years ago: This model is broken.
A large company can collapse in an instant. The US market is brutal: A company can fall to zero overnight. Why has every billionaire gone back to work, turning a traditional Corporate organization back into a startup? Whether you are a billion dollar company or a one-person company, everyone is in founders mode, afraid of being disrupted.
A one-person company does not need funding as long as the economics of its product work. But many of these “seafood products” quickly go out of season. An improvement in large-model capabilities could make you obsolete.
“Before technology converges, you need differentiation. If you alone understand a market worth ten million, you have 100% market share and very high gross margins. That is a characteristic of capital value creation in the new era. You must understand it exceptionally well to survive.”
Everyone Should Expand Overseas
Piruze outlined what Chinese founders get right: a vision to expand globally, supportive investors with a global outlook, an appropriate corporate structure, a problem with universal relevance, and enough data points or technological advantage to differentiate.
“How do I turn my technology into commercial success? That is usually a skill that can absolutely be learned.”
Cao Wei considered several layers.
In robotics hardware, China is already far ahead in physical manufacturing and locomotion. But Silicon Valley still does better in frontier manipulation capabilities, model architecture innovation, and training-paradigm innovation. China may be more focused on engineering, or more constrained in training resources.
In software, China's flourishing variety of solutions and sensitivity to embracing new productive capabilities are much stronger than overseas.
At the foundational-model layer, leading people overseas are exploring new formations and architectures without backpropagation. But Chinese companies such as DeepSeek and Kimi do very well in the engineering framework. Overseas approaches are mostly closed source, while China continues to pursue open source. If China's open models become dominant, their influence in open-source ecosystems outside the United States will be enormous.
Wayne was more direct.
“On overseas expansion, I think everyone should do it. Just look at GDP: All the consumer spending is overseas. If you are not in a sensitive sector subject to sanctions, you should start planning for North America now. The growth, gross margins, and return and exchange rates there are all far beyond those in China.”
He also mentioned a crucial point: capitalization structure. The Chinese and US businesses need to be clearly separated, moving from a parent–subsidiary structure to a sibling structure, to avoid future geopolitical complications.
Let Things Play Out a Little Longer
At the end of the Panel, each person summed up in one sentence.
Shen Dongliang: This is a tremendous opportunity. We are all fortunate to witness such a huge leap in productive capability. We should accept it and embrace it.
Piruze: This is a once-in-a-lifetime opportunity, and China has some of the very best talent in the world.
Hu Bin: We are very excited, but we can let things play out a little longer.
“Let things play out a little longer.” This is not passive waiting. It is clear-headedness.
At AI's current pace, rushing to judgment is often a mistake. Stay curious, keep experimenting, and remain sensitive to change—but do not let FOMO take control.
Time will provide the answer.
More from the Conversation
Unique Awards · Hangzhou AI WEEK Trend Roundtable Panel: “Different Narratives in Chinese and US AI Development, and the Implications for Capital”
Guests: Wayne, Managing Partner at Argo Venture Partners | Piruze Sabuncu, Square Peg | Hu Bin, Founding Partner at INCE Capital | Cao Wei, Partner at BlueRun Ventures | Shen Dongliang, Founding Partner at Yuanshu Ventures
Moderator: Tina, Unicorn Interview Room
Tina: I'm delighted to moderate this panel today and discuss the different narratives surrounding AI industry development in China and the United States. We are also very pleased to have invited so many funds that invest in unicorns to explore Global AI trends together. First, could each guest introduce themselves, their fund, and what distinguishes their AI investments? Shall we start with Wayne?
Wayne: Thank you, Tina. We are a new fund, but our theme is that we are a Physical AI specialist. We do not pursue other sectors. Our primary market is actually North America. I think that among today's Chinese founders, including researchers in large North American technology companies, most are also Chinese AI researchers. So we believe that, as capable Chinese founders and engineers, we should work to win the North American startup market. That is our theme.
Tina: Could you briefly explain Physical AI?
Wayne: We divide Physical AI into three layers. The first is what we call Intelligence, which may be the priority. At present, the “cerebellum” is quite good, but the “cerebrum” is very weak, so Intelligence is a major focus. The second is embodiment, which relies entirely on China's supply chain. I think this creates a good opportunity for us to expand overseas together. We are looking at how to move China's advanced supply chain into “friend-shoring” arrangements accepted by the North American market—whether building factories in Mexico or Texas, or doing contracting. People are already doing all of these things. Our founders have been overseas for a long time. Mr. Cao of Fuyao Glass, for example, has been there for 10 years. His factory is profitable, workers' wages have increased, and there is still no union. So I think the US environment is becoming increasingly ready. The third layer is safety and security. OpenClaw was still rather wild when it appeared last month, for example; it is better now. Safety and security may be a major challenge or investment opportunity facing the entire AI environment next. We see considerable opportunity there.
Tina: Let's invite Mr. Hu from INCE Capital.
Hu Bin: Good afternoon, everyone. I'm Hu Bin from INCE. We are a dual-currency VC fund investing in US dollars and renminbi, primarily in startups around Series A to Series B. Our main investment directions are AI and consumer technology, with some healthcare and gaming activity as well. We mainly back Chinese founders, but many of their businesses also operate overseas.
Tina: We have also invited Piruze from Square Peg. Please introduce your fund and what distinguishes your AI investing.
Piruze Sabuncu: I'm Piruze, and I live in Singapore. I'm from Square Peg Capital. Square Peg invests in AI and fintech founders who start in Asia and Australia and expand globally. We have been investing for the past 10 years and are investors in companies including Canva, Airwallex, and Supabase. We have a US$550 million fund investing in founders with a global outlook and the ambition to build great businesses—ideally starting in this part of the world but looking globally.
Tina: Next, let's invite Cao Wei from BlueRun Ventures to introduce himself.
Cao Wei: I'm Cao Wei from BlueRun Ventures. We focus on early-stage technological innovation. We have invested in China for more than 20 years and have over 300 portfolio companies. We invest around RMB 1.5 billion to RMB 2 billion annually, with the ability to invest in both renminbi and US dollars. AI has consistently been a major focus: from the chips underlying AI infrastructure, through the model layer—such as Moonshot AI and Westlake Xinchen—to AI applications on top of models, such as Genspark, then products combining software and AI, and finally hardware combined with AI, including various robots. We began investing in robotics in ’16 and have now invested in more than 20 leading projects across different types of robotics.
Tina: Thank you. Let's invite Shen Dongliang from Yuanshu Ventures.
Shen Dongliang (Yuanshu Ventures): Hello, everyone. I'm Shen Dongliang from Yuanshu Ventures. We mainly invest in AI applications, AI hardware, and AI businesses expanding overseas. We are a group of founders who previously started businesses in the software industry and formed this renminbi investment fund after exiting. Our investments are therefore mainly early stage, from angel rounds to Series A.
Tina: Let's turn to an interesting and very current topic: How do you see the opportunities OpenClaw has created for founders? My own observation is that the large companies followed very quickly after this wave of OpenClaw appeared. With Coding, it feels as though their speed in following popular products has gone from several months or half a year to less than a month, or even two weeks. There are probably many OpenClaw founders here today. I'm curious how you see the startup opportunities from this wave. Would you actually invest, and in what directions? Let's start with Mr. Shen.
Shen Dongliang: I think the impact is substantial, and it is certainly very popular. It feels as though everyone has started “aquaculture.” I see several major changes. First, it has a huge impact on existing founders, especially at the application layer. We began investing in application-layer AI projects around ’22. Initially, people built tools, then moved toward delivering results. There was a debate about whether application businesses should sell tools or results. After OpenClaw appeared, the biggest change was that our applications might no longer be built primarily for people, but for machines. We also used to say the application layer could be thin and lack deep competitive strength. Now many companies are trying to deploy their own versions of OpenClaw, allowing them to solve many relatively shallow application needs themselves. If an application startup's barrier is not deep enough, it is easily flattened. Every niche will therefore become more competitive, and differentiation will be harder.
Tina: What barriers do you think application-layer companies can establish?
Shen Dongliang: I think industry expertise is the foundation, followed by accumulated, differentiated industry data. That is an important focus when we assess application projects: Do they have unique data? Only data offers the possibility of training something with different capabilities. But frankly, building a meaningful gap is not easy.
Tina: Mr. Cao, would you like to share some views?
Cao Wei: OpenClaw really is a major topic this year. There is a hot topic at the beginning of every year. Last year, it was the Reasoning capabilities of DeepSeek V3; this year, it is OpenClaw. We have spent a lot of time discussing what it is. Our internal conclusion is that it remains a framework with three core parts. One is that it can accumulate skill capabilities. Second, it provides an underlying structure for context alignment, collecting various kinds of information and aligning them beneath the surface. Third, it has its own system mechanism that lets it work 24 hours a day, 7 days a week. Given those characteristics, we have been discussing where the core value lies. First, the framework itself. But everyone can see how quickly the major companies iterate, whether Anthropic or the various models recently launched by major Chinese companies using OpenClaw's open-source framework. We used to have a battle of a hundred models; now it is a “battle of a hundred lobsters.” Third, we see another important element within the framework: the skill, which could become the real node where value resides. First, it needs long-tail know-how. Second, it needs to know the paths to resources. Third, it needs permission to access those resources. With those three elements, it can become a skill that genuinely produces value.
As for OpenClaw's current problems, the first is security and the second is usability. Without security guarantees, there can be no meaningful discussion of usability. The basic version is really only a proof of concept. It is like an “unfinished apartment,” still far from a user-friendly place where someone can simply “move in with their bags.” I also wrote more than a thousand lines of code for a skill myself and found resource-path and network configuration very troublesome. So far, the value it provides has not been particularly proportionate to the time it consumes. Finally, this kind of agentic framework will significantly affect existing organizational structures, workflows, and team collaboration. These are all things we are considering.
Tina: I think this is an excellent topic. I've also heard people say that OpenClaw feels like the arrival of Linux: a new agent OS on which people can build new startup opportunities. Others say its engineering is very incomplete, leaving large companies plenty of room to build cloud products and Desktop versions, offering closed-source products with more complete engineering. From your description, do you see a way to catch specific opportunities, or is this more about customizing your existing portfolio?
Cao Wei: I think there are definitely opportunities. First, find value around use cases and look for things where sustained work can compound over time. You may not be able to match the engineering capabilities of large companies. But if you can find a long-tail use case and use OpenClaw to replace difficult or complex tasks, with it working 24 hours a day, 7 days a week, there is an opportunity. Second, there are opportunities around the physical hardware itself. I previously installed my OpenClaw on a laptop, but found that when it was working, I could not work. It felt as though there was another body. So I really think it should have its own computer. How to accumulate data with compounding value around that hardware may be a window of opportunity for startups.
Tina: Does that necessarily require hardware, or could cloud-based sandbox solutions address it too?
Cao Wei: We have also looked at many overseas sandbox solutions, such as the cloud computer based on Ubuntu launched by JetBrains. But there are two points. First, in some scenarios, you do not want personal private data uploaded to the cloud. Second, people have shifted from being “execution nodes” in the workflow to “authorization nodes.” Previously, I did all the work. Now it is about permission: I'm going to delete this file—is that OK? I'm going to use your account to buy water—is that OK? How can the authorization itself be more secure, and where would you feel safer keeping the alignment data? I believe that as a tool, it will ultimately remain use-case-driven and have different layers. There will be a cloud-collaborative “scenario version” and a “local version” based on users' own hardware. So the local version could be a startup opportunity.
Tina: Excellent. Let's welcome Piruze into the discussion. We have just talked about technological breakthroughs around open source. What do these mean for founders' futures and the spread of AI?
Piruze Sabuncu: I think “open” represents a moment that interests many people, stimulates creativity, and makes them very excited. Several things follow from that. First, as others have said, deploying and releasing an idea so that it actually works requires a lot of work and is very complex. We already know that intermediary layers and companies are emerging to help people put those ideas into practice. My only concern is that because people are currently using it in unsafe ways, there could be bad incidents that discourage adoption or lead governments to intervene and stop it. But I think that probably will not happen. Other things are happening too: A second version of OpenClaw, for example, is entering companies in a safer way. Many CEOs and business owners now expect their teams to find ways to become more creative and productive. I think that will encourage many employees to adopt these tools more safely. I'm completely certain that the scenario in which one person builds a technology company and launches a SaaS product or personalized software is very close to reality in the near future. A year from now, we will see more of these things emerge in open source.
Tina: How do you see startup opportunities in enterprise services? What kinds of products should founders bring to market?
Piruze Sabuncu: I'll be very candid. Two years ago, we were saying that building a large-model company was too difficult and that vertical AI and AI-enabled software were the way forward, because there were so many problems to solve. But now those software companies are finding growth more difficult, because tools make building a software company so easy. When you sell to an enterprise—say you have built a company solving legal, customer-service, or sales problems—the buyer asks: “Can I build this myself? Will my needs change? Will the technology change?” The company says it needs its own team to do it. That said, some software companies are evolving. They are moving into hardware, building data advantages, collecting interesting data from multiple companies, keeping that data secure, and solving very complex workflow problems. So there is now an AI layer and a software layer, while the implementation layer is becoming more creative too. Enterprises want to act quickly, see AI's impact as soon as possible, and focus on the business they do best. So at least outside China, they are very willing to buy this software. We have seen some of these companies grow revenue from 1 million to 200 million within six months or a year.
Tina: Very interesting. You also mentioned one-person companies. How do you see their future? If there will be so many, how will a venture capital firm managing a substantial pool of capital find opportunities and assess whether they are investable?
Piruze Sabuncu: As venture investors managing funds of a certain scale, our job is to find founders with enormous ambition who want to build at scale and create great companies that endure across eras. So the essence of my job has not changed. That does not mean there will no longer be successful one- or two-person companies; they will have their own business models. But there will also be ambitious founders who want to build infrastructure for those one-person companies or create consumer applications. There will always be enough interesting problems for people to build enormous companies around. Those companies may begin with only two to ten people, and research and development costs may fall. But suppose they want to train their own models and own their own servers: Those are still enormous costs. The challenge is therefore not just building a product. Business development in a highly competitive market will become much more complex. Or, when implementing software globally, someone still needs to understand customers' problems and persuade them to work with the company. So they need capital to grow faster.
Tina: Agreed. Let's invite Mr. Hu to share his views.
Hu Bin: At the micro level, for me personally, I don't know what to invest in around OpenClaw right now. Things are evolving far too quickly. Our traditional way of working—meeting once a week—has been completely upended. So the most rational approach now may be to let things play out a little longer and observe. But that absolutely does not mean we are pessimistic; in fact, we are very optimistic. At a broader level, after the Spring Festival, investors had to run OpenClaw themselves. Most of us had never done Coding or worked with servers, but if this is supposedly something one person can do alone, failing to try it yourself means falling behind. I've run it myself and found that the value for the cost is indeed not very high right now. Of course, some very capable people may already be much further ahead than I am. Previously, people felt FOMO—the fear of missing out—afraid of falling behind at the starting line of a generational shift. But if you look at how technologies iterate, there is no need to rush so much. Those who move a little more slowly can still catch up and benefit from others' work. We used to say Chinese companies were hardworking and competed intensely, but now you cannot outcompete machine life through sheer effort.
On OPCs, I think it resembles the early internet era of grassroots entrepreneurship. Previously, you had to find front-end and back-end developers and technical experts to get things done. With these tools, a few people—or even one—will be able to accomplish a great deal. The advantage large companies once had through money, staffing, and specialized division of labor may become less pronounced. Future disruptive developments could come from even more unexpected places, and there will definitely be more opportunities for individual founders. Many people will be freed to build what they want. That is a positive outlook.
Tina: I suddenly have a thought. On the One-person company, I was in Silicon Valley a couple of days ago and happened to speak with an employee of a large company. They require employees to write their workflows as skills: If they do not, they are laid off immediately; after writing them, they are laid off three months later anyway. It made me wonder whether the process of writing a skill will ultimately be absorbed by large-model companies, as the accumulated insights of many humanities graduates gradually enter the models. If so, would you investors prefer to increase your investments in large-model companies or invest in a One-person company?
Hu Bin: The biggest beneficiaries in this direction will definitely be the large-model companies. Token consumption will unquestionably become a normal part of future infrastructure. I had never bought API calls before. Now I wake up and 100 yuan is gone, without knowing what it did. Large models will certainly receive the biggest share of the benefits. This includes a lot of hardware: In the not-too-distant future, homes may run several intelligent assistants managing different things. Capital markets have also given their answer—these companies are growing very rapidly. For an individual, at worst you lose a job and become a freelancer. For a large company, the entire company could disappear. In the future, one Agent will communicate and evolve with another. If the safety layer can be handled well, what they create could fundamentally disrupt large companies. So individuals need not worry too much; these concerns are minor by comparison.
Tina: Would you like to add anything, Mr. Cao? You have the perspective of investing in large models and may also look at the One-person company. Would you choose to keep adding to large-model investments, or something else?
Cao Wei: Our approach is to bet on both sides. First, large-model teams themselves are already working on interactions with the most widely used software and the most frequent skill functions, such as deep research and news organization. Everyone is developing computer use solutions. If something does not require computer use, large models can already do a great deal. But I think user needs will ultimately lead. For things involving personal privacy, I want local storage and local models. I would keep personal and cloud-based work relatively separate. For long-tail, unusual interactions with traditional software, even a large model needs a learning process based on a source of data. So in those areas, the playing field is relatively more equal for large and small companies.
Tina: OK. Wayne, let's hear your views.
Wayne: I agree with Hu: Let things play out a little longer. I think this round resembles the previous one, when people first built all kinds of small tools and automatic website builders. There will not be that much in the middle that can produce capital gains. We all make money through capital appreciation; I cannot make money from cash flow. I think one-person companies are a good thing. The employment system is an outdated model designed for manufacturing 100 years ago. This model is broken. A large company can collapse in an instant, particularly in the brutal US market, where it can fall to zero overnight. Why has every billionaire gone back to work, turning a traditional Corporate organization back into a startup? Whether you are a billion dollar company or a one-person company, everyone is in founders mode, afraid of being disrupted. A one-person company does not need funding as long as the economics of its product work. But many of these “seafood products” quickly go out of season. As large-model capabilities improve, your product may become obsolete. Before technology converges, you need differentiation. If you alone understand a market worth ten million, you have 100% market share and very high gross margins. That is a characteristic of capital value creation in the new era. You must understand it exceptionally well to survive.
Tina: OK. Let's discuss China and the United States. Something that made me particularly proud in Silicon Valley this time was that US funds there finally started asking whether I could get them an allocation in Kimi. I really feel Chinese startups have achieved something in overseas expansion. I'd very much like to hear how you see the opportunities for Chinese AI companies expanding abroad.
Piruze Sabuncu: We are fortunate to work with the group we call “Chinese founders,” who are building global businesses from here. Several things these founders get right are important. First, if they have a vision to expand globally, they need to ensure they have supportive investors with a global outlook; an appropriate corporate structure so they are not constrained later; a problem that is universally relevant; and enough data points or technological advantage to differentiate. For example, suppose you build a very distinctive search API tool in a very popular area. Even if you developed similar technology inside Alibaba and have a distinctive way of building it, the global market is enormous, and major technology companies outside China may use your product. You therefore need to ensure your team can work smoothly with them. You need proper documentation, robust data privacy protection, and the corresponding settings. That is the first point. Second, if you are building products for overseas developers or building infrastructure, you need to know how to tell your story and build a following, and then start thinking about commercializing the company. A large gap we often see among Chinese founders is: “How do I turn my technology into commercial success?” But that is usually a skill that can absolutely be learned.
Cao Wei: We look at this on several levels. First, in robotics hardware, we consider four core capabilities: locomotion, navigation, interaction, and manipulation. China is already far ahead in physical manufacturing and locomotion. But in frontier manipulation capabilities, model architecture innovation, or training-paradigm innovation, I think Silicon Valley still does somewhat better. China may be more engineering-oriented or relatively short of training resources. In software, China's flourishing range of solutions and sensitivity to embracing new productive capabilities are much stronger than overseas. Previously, AI was the Copilot helping me work. Now AI does the work, and I am the Copilot authorizing it. Productive forces determine relations of production. Once that relationship reverses, China's responsiveness is extremely strong. At the foundational-model layer, leading people overseas, such as Ilya, are exploring new formations and architectures without backpropagation. But Chinese companies such as DeepSeek and Kimi do very well on the engineering side of the framework. Overseas approaches are mostly closed source, while China continues to pursue open source. If China's open models become dominant, their influence in open-source ecosystems outside the United States will be enormous. We are catching up in many areas.
Wayne: On overseas expansion, I think everyone should do it. Just look at GDP: All the consumer spending is overseas. If you are not in a sensitive sector subject to sanctions, you should start planning for North America now. The growth, gross margins, and return and exchange rates there are all far beyond those in China. Another issue is capitalization structure. The Chinese and US businesses need to be clearly separated, shifting from a parent–subsidiary structure to a sibling structure. That is how you avoid future geopolitical complications and do not repeat the experience of certain companies that reached several billion US dollars but faced restrictions because their structure had not changed. There are definitely approaches in North America that can be planned in advance.
Tina: Thank you. Let's have one final sentence from each person, and then we will close.
Shen Dongliang: I think this is an enormous opportunity. We are all very fortunate to witness a huge leap in productive capability, so we should accept it and embrace it.
Piruze Sabuncu: I think this is a once-in-a-lifetime opportunity, and I also believe China has some of the very best talent in the world.
Cao Wei: As we said earlier, we are very excited, but we can let things play out a little longer.