--- title: "Why Investors Warn of an AI Bubble While Betting on It So Heavily" author: "Unique Research" sourcePublication: "Unique Research Substack" originalPublishedAt: "2026-04-10T10:18:14+00:00" canonical: "https://ffcap.cn/en/research/src-20260410-03html" source: "https://uniqueresearch.substack.com/p/src-20260410-03html" language: "en" --- # Why Investors Warn of an AI Bubble While Betting on It So Heavily _Original · Unique Research · 2026-04-10_ _Editor's note: This translation preserves the original author's first-person narrative and the speakers' statements, not the translator's firsthand experiences. The opening and the transcript heading give different panel titles; both are retained pending source verification. Relative dates and the future-tense reference to “this year's” Spring Festival Gala remain as written in the April 2026 source. Investment histories, institutional English names, founder biographies and valuations require confirmation. References to doctoral study and dropping out do not establish that a degree was completed. The 95% professor-startup failure figure is a claim the speaker attributes to articles, not a verified statistic; the IRR comparison is also not independently audited. Valuation passages that omit a currency remain unspecified. Market views and predictions are attributed opinions, not investment recommendations or guarantees._ Unique Awards This Wave of AI Really Does Resemble the Early Days of a New Internet When investors are no longer looking at just one sector, but betting on a new era of infrastructure, many things suddenly make sense. At Hangzhou AI WEEK, I listened to a panel titled “Different Narratives in Chinese and US AI Development, and the Implications for Capital.” Frankly, a topic like this can easily turn into a flavorless industry report. Comparisons between China and the United States, diverging capital flows, generational shifts in technology—before long, it can become a collection of correct but empty statements. You cannot find anything wrong with it, but after reading, you think: Fine, and then what? But there was a moment in this roundtable that genuinely struck me. It was not a guest delivering an explosive conclusion. It was Ren Bobing, discussing this wave of AI investment, suddenly saying that he thought it resembled the early internet—and might move even faster. The remark seemed simple, perhaps even a little clichéd. But listening through the rest of the panel, I realized it connected almost everything the participants agreed on. Whether they were looking at large models, an Agent, OpenClaw, embodied intelligence, world models, AI hardware, AI animated short dramas, AI music, dexterous hands, or MaaS, these investors were essentially facing the same question. They are not investing in a clearly defined sector. They are betting on a new internet era. That is a very big thing. If you understand today's AI as an isolated technological upgrade, much of what is happening will be hard to explain. Why do some fields already command valuations above ten billion when they have not truly entered everyday life? Why do so many projects race through financing round after financing round when commercialization is still at an early stage? Why have some funds made fewer investments over the past few years, only to suddenly make repeated bets on AI? Why are people warning of a bubble while simultaneously fearing they will miss out? Traditional sector-investment logic often struggles to explain this. But put it in a larger context, and many things fall into place. This Does Not Feel Like a Mature Industry. It Feels Like a Frontier. Today's AI looks very much like new infrastructure growing outward. What was most fascinating about the early internet? It was not that mature products already existed. Quite the opposite: Many key pieces were incomplete. Payments were awkward, communication was awkward, collaboration was awkward, security mechanisms were incomplete, and the mobile experience was terrible. Everyone knew it was immature, yet everyone could sense a new ecosystem emerging. Ren Bobing put it particularly well. He said that today, you can use OpenClaw on a PC, but the mobile experience remains far behind. There is no particularly good payment mechanism. Security is immature, communication cannot truly broadcast to everyone, and collaboration between people, between a person and an Agent, and between one Agent and another remains unresolved. That does not feel like a mature industry. It is the characteristic feel of a frontier. There are plenty of problems. And plenty of opportunities. So the hardest part of this investment wave is not understanding a product. It is judging whether these fragments could come together to form the next generation of infrastructure. That immediately raises the demands on investors. You need more than foresight. You also need courage. Early internet investing was about trends. AI investment today is often about trends too, but they move even faster. A thesis that seemed rock-solid three months ago may be overturned by a different form three months later. Today it is a lobster; tomorrow a goldfish might arrive and swallow the whole seafood platter. The Hard Part Is Not Seeing the Opportunity. It Is Daring to Bet First. I will remember Lu Hongyu's analogy for a long time. He said that in the AI industry today, your company may be very popular, but a crisis could be just around the next corner. That sounds a little harsh. But my own feeling is that it is true. For many years, one of the most common illusions in private-market investing was that once a business model became established, its moat would keep deepening. AI does not work that way. The cruelty of the AI era is that your moat may not be gradually eroded. A new paradigm, interface, or interaction layer may suddenly bypass it altogether. That is why, after listening to the panel, my strongest impression was not that investors had become more aggressive. It was that they were being forced to behave more like gamblers in the early internet era. Of course, by gamblers I do not mean people making random bets. I mean people who know the system is unfinished, know there will be bubbles, and know many metrics are unstable, yet must still identify, amid incomplete information, what could become the next entry point, protocol, or consumer habit. They are not investing in a clearly defined sector. They are betting on a new internet era. Those Most Likely to Break Through Either Control a Foundational Advance or Win a Lead in Time This made me pay particular attention to another question. What kinds of founders are more likely to break through in this wave? I thought Zhao Peizhou's answer was particularly representative. He did not simply say professors were better, or young people were better. He offered two profiles: academic founders deeply engaged in research and capable of foundational innovation, and very young founders who are bold enough to act and seize the timing early. Those profiles look different, but the underlying logic is the same. Either you control a foundational breakthrough, or you secure a time-based barrier to entry. That time advantage is extremely important. He was direct: Many of the companies with the strongest growth did not enter after AI became popular. They were already working on it before the boom. When market enthusiasm rose and demand took off, they already had a relatively mature product, team, and position. Think about it: Isn't that a classic internet-era story? Later, people remember only that the company became a unicorn. But what determined its fate was often not what happened at peak excitement. It was the work it was already doing when few people understood the opportunity. That is also why Zhao Peizhou emphasized young founders. Not because youth itself is valuable. But because the window in this AI wave is so short. You need to act while everyone else is still debating whether it can work. To seize that moment, people may drop out, abandon further doctoral study, or return to China to start a company. They trade a planned life path for a lead in time. That is very risky. But it also resembles how central figures in a transformative era make their moves. It Is Not That Business Logic Has Stayed Still. The Way KPIs Appear Has Changed Completely. Returning to investment firms themselves, I also thought Lu Hongyu's comments on KPIs deserved careful consideration. He was candid: Internally, they repeatedly discuss how to assess these new AI companies. During the internet and mobile internet eras, whether looking at SaaS or video, they tracked retention, time spent, subscriptions, and repeat purchases. Those metrics gradually became relatively stable. AI companies today are different. Do you look at users? Yes. Do you look at PMF? Yes. But many metrics now appear in different forms: Token consumption, API calls, or usage frequency hidden inside a workflow. They are less immediately visible, even though the underlying business fundamentals have not truly changed. I think that distinction matters a great deal. It is easy to slide into one of two extremes today. One says a new AI era has arrived and all past business logic is obsolete. The other says to forget the new terminology because business fundamentals never change. My own feeling is that each statement is only half right. Of course the fundamentals remain: Do users like it? Does the product have value? Will the market pay? Those questions never disappear. But AI is transforming the way those things manifest. With an App, downloads, daily active users, retention, and time spent used to be straightforward to observe. Now much of what is genuinely valuable may be hidden in an enterprise workflow, a chain of calls, or a collection of invisible prompt instructions, Skill capabilities, context, and automated tasks. So investors are not ignoring KPIs. They first have to reinvent them. That is why, by Series A and Series B, investors increasingly emphasize the concept and PMF alongside the team. It is not that concepts matter more than products. When new infrastructure is just beginning to emerge, you first need to judge whether a direction could become a key node. Once that node is secured, many other things may grow around it. Embodied Intelligence, AI Animated Dramas, and AI Music May Look Unrelated, but They Bet on the Same Thing Talking about key nodes inevitably brings us to embodied intelligence. This sector has felt surreal over the past two years. Many people probably share the same confusion: Why have valuations soared for something that does not yet seem to have fully entered either daily life or factories? I understand that confusion very well. From an ordinary person's perspective, what you see is still demonstration footage: a robot taking a few steps, grasping something, or doing a flip. Widespread deployment still seems some distance away. Why, then, are investors willing to pay so much? I found it interesting to consider Bu Liangyuan's and Lu Hongyu's perspectives together. One side looks at the team, differentiated technology, and whether the company can secure a real position in the industry chain. The other looks at selling shovels. If you believe a mining boom is coming, investing in dexterous hands, underlying hardware, and capability modules is not about needing to see robots everywhere today. It is about securing a component that future robots may all need. This is much like a gold rush. Not everyone finds gold, but the people selling jeans and shovels often make money first. Once again, the logic returns to the early days of a new internet. Many people focus on whether something works well right now. Investors are looking at who will become indispensable once the entire chain develops. Of course, not everyone has to chase the hottest thing. I thought Wu Jiabing's remarks were particularly valuable. He represented a completely different kind of rationality. You do not have to invest in every hot sector. If you do not understand it, do not invest. In today's environment, that statement feels unusually rare. Especially as the CVC arm of a listed company with a core business, organizational boundaries, and strategic-synergy requirements, you cannot chase whatever is fashionable. Investing in ShengShu, animated-drama platforms, or the content-production chain is not about following other investors. It is about how those elements connect to the industry position you are building. I genuinely think that respect for one's limits matters. Discussions of AI investing today easily create the illusion that the whole world is chasing the next big thing and anyone moving slowly will disappear. But that is not actually the case. The most mature participants are often clearer about which markets are theirs and which are not. Not understanding embodied intelligence does not mean you are falling behind. If you understand AI animated dramas, AI music, and the content-production chain—and can connect them with your products, customers, channels, and ability to co-develop underlying models—that is a more grounded battle for you. The Real China–US Difference May Be Not Just Technology, but the Narrative and Where It Lands This also increasingly makes me think that the most interesting difference between the Chinese and US AI industries today may not simply be their technological paths. The deeper difference lies in their narratives and where they put them into practice. In the United States, many narratives naturally lean more toward infrastructure, To B markets, and foundational platform capabilities. People tend to assume that you will first build out productivity and the technological foundation. In China, of course, investors also back world models, large models, embodied intelligence, and hard technology. But at the same time, attention quickly returns to another question. When will this become a concrete use case? Can it become an application? Can it become something everyone can experience? Can it become an entry point like OpenClaw that people can immediately try for themselves? Can it take a form such as AI animated dramas, AI music, or AI hardware that ordinary people understand as soon as they use it? My own feeling is that China's greatest market strength may never have been being first to invent an underlying concept. It is the ability to rapidly bring something into social use, concrete scenarios, consumer markets, and commercial activity. That may sound unsophisticated. But I have always thought it is a powerful capability. Not every technological turning point becomes an industry. What bridges the gap is use cases and organizational capability. So when the panel ended with everyone naming, in one sentence, the AI niche they were most optimistic about for 2026, the answers were particularly interesting. Some continued to focus on the upstream and downstream ecosystem around an Agent and OpenClaw. Some continued to focus on world models. Some looked at AI hardware and wearables. Some looked at multimodal generation and physical AI. Some directly chose AI animated dramas and AI music. At first glance, the picture looks scattered. But it is not scattered at all. All those directions are betting on the same thing. AI will not remain confined to a model website. It will become new infrastructure, entry points, consumer media, production systems, and hardware forms. Whoever becomes indispensable at one specific node first has a chance to capture the largest gains from this wave. The Hard Part Is Knowing When to Push Forward and When to Hold Back If you ask me what I felt most strongly after listening to this panel, I would say it was not that investors had gone mad. It was that they increasingly understood they were not facing a collection of disconnected sectors, but a new internet era unfolding. At such a moment, the hard part is not seeing opportunities. It is daring to place a bet before the system is finished. That really does go against human instinct. People like certainty. But a new world is often most uncertain at the beginning. That is precisely why money is not the only valuable thing. So are foresight, courage, judgment about timing, and knowing when to push forward, when to wait, when to follow, and when to simply admit: This is not my game. What an era, my friends. More from the Conversation Unique Awards · Hangzhou AI WEEK Trend Roundtable Panel: “Differences in AI Startup Opportunities and Cross-Generational Collaborative Innovation” Guests: Wu Jiabing, General Manager of Strategic Investment at Wondershare | Zhao Peizhou, Partner at Xiaomiao Langcheng | Lu Hongyu, Director and Senior Partner at DT Capital | Ren Bobing, General Manager of the Frontier Technology Fund at Sinovation Ventures | Bu Liangyuan, Listed-Company Investment Department at Times Bole Moderator: Li Jinxiang, Founder & CEO of Xiniu Data Li Jinxiang: I'm the founder of Xiniu Data, and my name is Li Jinxiang. Xiniu Data primarily provides data and AI-tool products to financial institutions. Our distinctive dataset is private-market investment and financing data. If you are looking for investment opportunities, I strongly recommend trying our APP; you can experience a different kind of product. I'll leave most of the remaining time to our five guests. Looking them up in our Xiniu Data APP shows how carefully this panel was selected. In terms of stage, they cover everything from angel and seed rounds to Series A and Series B, as well as listed-company-backed CVC. Another important point is that many investment institutions may have made relatively few investments over the past two or three years. But among these five guests, I can see that four firms made at least ten investments in the past year. So today's discussion should provide particularly practical insights, from the perspective of actual investments, into what is different about this year's—or this wave's—AI investing. Ren Bobing: Here is how I see this wave of AI entrepreneurship. We positioned ourselves quite early in the AI 1.0 era. For companies such as Fourth Paradigm, Megvii, Momenta and WeRide in the autonomous-driving wave, and Horizon Robotics, we generally invested in the first or second round. The approach was quite clear then: Find young talent and bet on a young scientist early in a technological shift. You could more or less see a trend that would progress through its first, second, third, and fourth steps. But this wave is clearly different. There is already substantial infrastructure in place. In ’23, we saw large models, GPUs, and extensive infrastructure across AI. The application-company ecosystem in the United States is essentially B2B, while Chinese applications, both B2B and B2C, have progressed somewhat more slowly than their US counterparts. But looking at this year and the next few years, we see multimodal technology and foundational developments such as embodied intelligence entering a period of vigorous growth. On the B2C side, architectures such as OpenClaw have essentially brought coding capabilities to a point where they can evolve relatively autonomously. Combined with advances in memory and tool use over the previous two years, they bring many elements together. Personally, I think this creates a substantial challenge for investors. It resembles the early internet era. We can now use OpenClaw on a PC, for example, but the mobile experience is still quite far behind. When we want to communicate with others, we cannot be sure they are online in real time. In the QQ era, you would ask whether someone was there; in the WeChat era, that is unnecessary. The period we are entering also lacks good payment mechanisms and good security mechanisms. Even in communication, we cannot broadcast fully to everyone, and we do not know how a group of connected people should collaborate. With so many unresolved questions, I think investors need considerable foresight and courage to face this surging wave in the early days of a new internet era. It may move even faster than before. A great deal of activity may create some bubbles, but you still need the courage to place bets. Zhao Peizhou: We began early-stage investing around two universities: Shanghai Jiao Tong University and East China Normal University. We can also be considered a Jiao Tong-affiliated fund. Around half our investments have a Jiao Tong background, including 30 companies founded by its faculty. Our main investment areas are artificial intelligence, new manufacturing, and frontier technology. In AI, our newest fund is a partnership with Shanghai's Guotou, and that fund will also invest in artificial intelligence. From investing in the previous AI generation in ’16 and ’17 to the ChatGPT boom in ’22 and ’23, we have invested in many companies, building a systematic portfolio from foundational computing chips through embodied intelligence to AI applications. First, we focus deeply on startups founded by accomplished researchers. There is a misconception here. Some people read articles saying you should never invest in professors' startups, or that such startups in the United States have a failure rate as high as 95%. But we have now gone through this process ourselves. Jiao Tong began formally allowing faculty entrepreneurship under a compliance framework in 2019, with a very comprehensive set of procedures. We have invested in around 30 such companies so far, and some good businesses have emerged. Although I cannot discuss DPI, their IRR is higher than that of some market-oriented founders we have seen. But I would also like to share another emerging profile: young founders. I'll give two examples. One is Songyan Dynamics, which everyone will see performing a sketch with Cai Ming at this year's Spring Festival Gala. Its founder, born in ’98, was a Tsinghua doctoral student who dropped out to start the company. He was very determined and felt the timing had arrived, positioning himself among the first wave of embodied-intelligence founders. We invested at the angel stage and have followed on repeatedly. Its valuation has now reached unicorn level. It has been about two and a half years since it was founded, and it has developed very quickly. The other company works on optical computing. It has two founders, born in ’96 and ’97. The majority shareholder was a doctoral student at Oxford who also dropped out and returned to China to start a business. To seize that window, they traded a planned life path for a lead in time. So I see two features in this AI wave: Young people have drive and determination, and timing is paramount. To lead an industry, the barrier you build comes from that time advantage. Lu Hongyu: We often discuss internally how to assess AI companies today. People may remember that in the internet or mobile internet era, we had many company KPIs. For SaaS, you looked at recurring revenue and annual subscriptions. For video, there were daily time spent, seven-day visits, retention, and many other metrics. I believe those metrics were also uncertain in the early internet days. It was only later, as people observed companies growing, that they became established. We really are on an entirely new path that no one has traveled before, and we do not know what will happen. For example, how will the Agent companies people discuss today, including OpenClaw, operate in the future? We keep discussing that ourselves. But I believe some things will never change. As a company, we aim to make a profit. Is our product welcomed by the market? First, you need users, right? Additional purchases, repeat purchases, usage frequency, and time spent remain relevant. It is just that in the AI era, usage and the way it appears may differ. It may be Token consumption or the number of API calls. Many things are invisible or less obvious than before, but I believe the broad principles remain the same. So at our current investment stages, if we must identify KPIs, we still look at the team. At Series A or Series B, we may focus more on the Concept. By Series B, we look at Product Market Fit—PMF. The product must be more than a prototype; it must be something that can genuinely be brought to market. We still examine these things. In short, their presentation may differ slightly, but business fundamentals have not changed. Bu Liangyuan: On embodied intelligence, why are valuations already so high when it does not seem to have fully entered everyday life or factories? We focus more on the backgrounds of the founding teams and technological differentiation. As an institution backed by a group of listed companies, we can provide many upstream and downstream connections. So we care more about whether these companies can actually reach concrete use cases. Often, investment logic is not about whether robots are already everywhere today. It is about whether an entire industry chain will develop and whether you hold a key position in it. If you sell shovels, that part of the chain may make money before the prospectors emerge at scale. Wu Jiabing: Embodied intelligence is very popular, but we do not necessarily understand every industry. We do not invest in a sector simply because it is hot. I invest only in sectors I understand. If I do not understand one, it may be someone else's market, not ours. We are strategic investors, which differs from financial investing: We prioritize strategic synergies. We invest more because we hope to strengthen one another's businesses, or to work together over a longer horizon—short, medium, and long term—rather than pursue returns on short-term financial metrics. For example, investing in ShengShu Technology or animated-drama platforms is not fundamentally about chasing trends. It is about building an industry position. We need deep co-development and mutual support with underlying-model vendors, while strengthening our own advantages in the AIGC content-production chain. We have always had two identities: entrepreneurs and strategic investors. So we cannot engage in too much speculation. We focus more on long-term synergies. Ren Bobing: For the AI niches I am most optimistic about in 2026, I will continue looking at the Agent field, including upstream and downstream ecosystems around OpenClaw, as well as world models. We have already invested in some companies in both areas. Zhao Peizhou: Although hardware is not necessarily our strongest area, I think AI hardware offers opportunities. There may be a new hardware medium in every home or everyday life that takes you away from phones and PCs. It could handle vertical tasks, whether companionship or improving work efficiency. There may be hardware for each of these. Lu Hongyu: I genuinely like OpenClaw. As a personal assistant or for collecting information, it really feels like an employee who does not cost much. Of course, you spend some money on Token usage, but it truly is an employee. If you use it well, it can help you a great deal. Bu Liangyuan: On the software-oriented side, I think there are still opportunities in physical AI and multimodal generation and interaction. In hardware, I am also optimistic about wearables, including AI combined with smart hardware in various forms and scenarios. Wu Jiabing: First, I strongly agree with Mr. Ren on world models and large models; we remain optimistic about them. I also agree on the Agent direction. OpenClaw has generated two kinds of reactions: Overseas users are very excited, while users in China are generally anxious because they feel it is replacing people and wonder whether they will lose their jobs tomorrow. Those are two different attitudes. But from a highly vertical investment perspective, I favor two sectors. The first is AI animated dramas—short dramas—which I see as a relatively certain opportunity. The second is AI music. Why do I view AI as relatively certain? Because AI's contribution to film and television is already producing very clear effects and results, while major changes to the music industry are gradually unfolding. I think we may soon find that many people can release and write songs, and perhaps even become stars from ordinary backgrounds. Any field that allows ordinary people to step onto a stage under the spotlight has tremendous potential, in my view. --- Original publication: https://uniqueresearch.substack.com/p/src-20260410-03html On-site reading page: https://ffcap.cn/en/research/src-20260410-03html