Harry Wang's Counter-Cyclical Bet — Entering the Game --- When the venture capital industry was retreating en masse, Harry Wang made a move that seemed to defy the tide. It was late 2023. The Chinese tech investment landscape lay in near-freeze — fundraising had collapsed, deal flow dried to a trickle, and even seasoned investors were sitting on their hands or pivoting to safer harbors. At Linear Capital, where Wang had spent years as a founding partner, the firm was also tightening its belt, shifting toward more conservative deployment. Wang chose the opposite path. He left. Not to retire. Not to wait out the winter. But to start something new — a fund built for exactly this moment of pessimism, when valuations had cratered and founders with real substance were being overlooked in the panic. "Everyone's talking about risk-off," Wang told me when we met in Shanghai. "But that's precisely when the best opportunities emerge. The question is whether you have the stomach for it." The new vehicle, which Wang has been quietly assembling over the past year, represents a

"Enter the Game" is a regular column of Waves. It stems from our observation that once-reliable operating models are facing new challenges, and the industry rules inherited from west to east have been disrupted. People urgently need a new map and new order for innovation and capital. "Entering the game" is the most precious posture of all. "Enter the Game" was born amid transformation. To summarize its subject in one sentence: we hope to find new players and new moves better adapted to a changing environment. This is the sixth article in the column.

By Zhiyan Chen

Edited by Jing Liu

"We don't invest in large models."

At the very start of our meeting, Huai Wang, founder of Linear Capital — a VC fund branded around "technology" — stated this position unequivocally to Waves. His reasoning was simple: AI large models are infrastructure, not suitable for early-stage investors.

The battle over Chinese AI large models rages on. Whether Wang's judgment proves accurate remains to be seen. But amid the chaos, he is among the rare few with clear, sharp convictions. Some of his choices appear to run against the current — saying no to AI large models, for instance, or advising a "don't rush, take it slow" approach to RMB fundraising.

Meanwhile, Wang has decided to pursue other directions more aggressively: since 2023, Linear Capital has focused on global opportunities in Europe, opening an office in Munich, Germany. And he is embracing AI and young people in his own way.

Recently, Linear Capital announced a new investment program specifically for AI application startups: Linear Bolt (hereinafter "Bolt").

New technological trends have given rise to new investment paradigms. Bolt literally means "lightning." Wang told Waves he hopes to use Bolt to invest in "early-stage, globally-oriented AI application teams" with a lighter touch.

As "dedicated capital," Bolt's typical target company looks like this: a small founding team of 3-4 people; not necessarily with a "fixed office"; active on Instagram, X, Jike, and Xiaohongshu; working alongside AI; capable of going from idea to prototype in two months and shipping a product in four.

"Compared with typical 'Linear investments' — which focus on hard tech and the intersection of frontier technology with industry — Bolt focuses on applications, investing in application builders," Wang said.

Linear Capital has also assembled a younger, more nimble three-person Bolt investment team. Their mandate in practice is threefold —

First, more flexible check sizes. Linear's typical target range is $3 million to $10 million, with flexibility for larger or smaller amounts.

Second, faster decision-making. Through 3-4 meetings with founding teams, decisions in principle within two weeks, with overall execution not exceeding one month.

Third, more founder-friendly terms. A commitment to the cleanest possible terms that still match market convention.

In short: "Lighter, faster, more flexible."

Over the past month, Bolt has completed four investments: Xinguang, an AI life-logging product; Final Round, a native AI tool for job seekers; Xbuddy, a copilot for overseas university students; and Cathoven, an AI-driven language teaching platform. Several additional projects are also in the investment pipeline.

An engineer by training, Wang before becoming an investor was the second Chinese engineer and first engineering manager at Facebook headquarters. He believes deeply in advanced technology — past investments like Agile Robots, Horizon Robotics, and Sensors Data all carry clear tech DNA. But in AI, what concerns him more is how technology finds application scenarios, and whether users can understand and use it.

Recently, Waves spoke with Wang. Here, you'll see the choices of a classic dollar-denominated VC in a transformative era.

The conversation follows —

Right Problem, Right People

Waves: Typical Linear deals require rigorous technical due diligence, but Bolt's core is speed and efficiency. Is this a compromise to the AI wave?

Wang: That's not how to understand it. As baseline model intelligence improves, more and more AI-empowered applications are emerging. While there are certain technical challenges in model fine-tuning, most of these applications' core capabilities are provided by AI large models. So in technical evaluation, we spend more time on what the team has actually built — assessing team capability, product direction, and whether they can quickly find product-market fit. At the same time, developers who can leverage AI well are dramatically more efficient in their use of resources and capital than in the past. This means they need less capital to validate products and achieve goals, and their fundraising preferences change accordingly. So Bolt was created to embrace this change with the right investment approach and team configuration.

Waves: What are the keys to AI application entrepreneurship?

Wang: Similar to the early mobile internet era, AI applications will definitely show tool-like characteristics early on. In the long run, to build stickiness and repeat usage, the critical question is: how to break out of the tool chain and accumulate data. I believe that to raise barriers in AI applications, you need to accumulate three types of data to enhance user stickiness: first, contextual data — AI's understanding of context; second, personal data — AI's understanding of the user; third, collaborative data — AI applications should exist within workflows and interactions, not as standalone tools.

Waves: Are applications what Chinese entrepreneurs do best?

Wang: After all these years of the internet, China has accumulated the best product managers and the most intensely competitive engineers — both advantages for AI application entrepreneurship. But Bolt wants to invest in globally-minded young entrepreneurs whose markets aren't necessarily just China. Decades of development have given us many people with international vision and execution capability who understand how to build and operate global applications. Second, the internet and mobile internet have made direct global user reach possible. Plus, the proliferation of electronic payments worldwide has made it possible for people everywhere to pay for universal applications — market barriers have been broken down in the virtual world. Consequently, a commercial application distributed through internet and mobile channels can potentially be a global native app from day one. Meanwhile, empowered by AI, applications can undergo entirely new, major upgrades. What was previously "matching information with people" — more about "searching and matching" — now with AI compressing the world's accumulated knowledge into models, it's no longer about finding information but "generating." The efficiency and creativity possible could rise to another level entirely. So we see this as an opportunity the era has given us.

Waves: The mobile internet lesson was that consumer applications go through "burn money for market share" to eventual "winner takes all." Will this be the commercial path for AI applications?

Wang: I don't think so. AI applications won't burn money for market share, and there won't be winner-takes-all dynamics. First, the economics don't work. For traditional consumer apps, once you reach a certain scale, the marginal cost of serving the next user isn't high. So scale was what mattered. But in AI, each time you serve a user, the cost increases linearly because every interaction consumes tokens. The world knowledge compressed into the model is actually the fixed cost. Second, traditional consumer apps were more homogeneous, so individual user stickiness wasn't high. Price-based customer acquisition mattered, and capital played a big role. But in AI, while model capabilities differ, these differences get amplified by varying amounts of contextual data, personal data, and collaborative data. Of course, accumulating these three data types does depend on capital, but whether someone else's capital can steal your users — I think that's harder than in the mobile internet era. Third, "burning money for market share" previously meant traffic, with business models based on advertising or e-commerce. But AI applications answer questions and provide services, not traffic monetization. Burning money for market share makes no sense here. The focus is on building a good product, accumulating user data, refining the model, getting users to go deep, and getting them to pay — because paying users pay for value, and the wool comes from the sheep's back. I think this is a more reasonable and robust approach.

Waves: Bolt emphasizes a "lighter, faster, more flexible" investment approach. Why does this matter for investing in AI applications now?

Wang: Bolt is dedicated capital for AI applications. Linear traditionally emphasized TPF — technology-problem fit. But for AI applications, the technical barrier isn't high because core technology is provided by large models. The application layer demands more from founders' ideas, and requires faster action and faster team-building. After going through several decision processes, we identified problems with our old approach. So we decided: find the right problem, find the right people, and invest — don't overcomplicate things.

Waves: After ChatGPT emerged in late 2022, AI-related entrepreneurship surged. Why launch Bolt only now?

Wang: If Bolt had launched in the first half of last year, suitable opportunities probably wouldn't have existed — many capabilities weren't ready yet. Starting in the second half of last year, infrastructure large model capabilities and meta-toolchains (like in-model calls to external mature capabilities) matured. I believe the timing is right. As more capabilities become available, workflows and applications become genuinely useful. From a consumer perspective, AI applications will upgrade everyone's skill set; from an enterprise perspective, AI applications will bring truly new quality productive forces.

Waves: How will you ultimately judge whether Bolt succeeds?

Wang: I hope that among the dozens of projects Bolt invests in, we can produce AI-native super global apps — corresponding to 1-2 companies valued above $1 billion. That would be Bolt's success.

Playing Devil's Advocate

Waves: Among early-stage dollar funds, Linear hasn't invested in a single AI large model. Why?

Wang: AI large models are infrastructure with extremely high capital requirements. Now, price wars have broken out before any financial returns, and such capital intensity isn't friendly to early-stage investors. Over the years, Linear has maintained a calm stance toward such "price wars." And Linear's strength is understanding AI technology itself. After looking around, we genuinely felt that rather than competing in large model pricing — unfriendly to early-stage investors — we'd rather deploy resources where we're stronger.

Waves: Then in AI, what kind of money does Linear want to make?

Wang: Everyone talks so much about large model capabilities, but in work and life, what problem-solving methods have actually been replaced by AI? Take hotel booking — to this day, no AI application can compress away comparison shopping, selection, personalization, and execution for users. These are the opportunities Linear wants to participate in. We still want to make money from what we understand. Even if we fail in the future, it will be because our understanding fell short, not because we weren't bold enough.

Waves: Currently, application companies clearly have lower ceilings than large models. How do you explain this investment decision to LPs?

Wang: We communicate regularly with LPs, and through ongoing dialogue we've accumulated trust and understanding. Specifically for Bolt, because individual check sizes are flexible and there will be many investments, I tell LPs they can view Bolt as a basket. The total eventual investment (including follow-on) will be $20-50 million. Among these projects, whichever teams ultimately succeed in accumulating contextual data, personal data, and collaborative data and keep evolving with the times — the returns won't be bad. Moreover, the total value created by application companies in the future will far exceed that of large models.

Waves: Is the capital war around AI large models an "overreaction" by dollar funds?

Wang: There's definitely FOMO (fear of missing out) psychology at play. At the same time, most investors also want to participate in this new wave. Another fact is that in the current market environment, after concentrated fundraising in 2021-2022, dollar funds objectively still have capital on hand — in certain quantities — it's just that everyone is extremely cautious in deploying it. This year's AI large model investment boom actually corroborates this from the side.

Waves: When discussing technology investing, betting on the right technical route is considered an important measure of investor capability. In AI, are the technical route divisions not that significant?

Wang: We pay close attention to foundational model innovation and believe there's still much innovation possible at the technology base. It certainly won't just be about single large models getting bigger, faster, and stronger. There will also be much innovation in multimodality, expert network collaboration, black-box and white-box models, edge AI, and other directions. Future inevitable technical directions, including AI combined with hardware especially intelligent robotics — Linear remains very focused on these, though individual capital commitments will be larger. Regarding consensus, my view is: respect consensus and leverage it. For early-stage investing, you want to bet before or at the early formation of consensus. If an investor's view of the future later becomes consensus, then consensus is your opportunity to succeed. So embrace consensus. But don't follow consensus. For example, investing in cloud-based large models today — that's following consensus. For an investment firm, what you see in the short term is "intensity" — many want something spectacular. But in the long term, endurance is what matters.

Waves: "All in on technology" has been a distinctive label for Linear in the industry. Do you still hold to this today?

Wang: Linear never uses the phrase "all in," and I fundamentally dislike it. In fact, we don't all in on any single technology either. The beta we believe in is: frontier technology represented by AI transforming various industries, ultimately achieving productivity gains.

What Game Are You Playing?

Waves: You mentioned earlier that top dollar funds aren't short of money, but in recent years we've sensed intense anxiety among them. What's the source?

Wang: Not being able to deploy is a problem; deploying badly is also a problem. Raising dollars is a problem; raising RMB is also a problem. Exits are a problem; management is also a problem. So fundraising, investing, managing, exiting — basically everything is a problem. Dollar funds aren't short of money in the near term; they just haven't hit the survival line yet. But everyone is anxious inside.

Waves: Yet to this day Linear, while dual-currency, is still dollar-denominated. What's your thinking on RMB fundraising?

Wang: Linear has RMB experience. For our new RMB fund — Linear RMB Fund V — many RMB investors have proactively approached us for partnership, including relatively mainstream LP types. Our current thinking is to take it slowly, not rush, while investing well the capital we already have. I'm an engineer by background; what I most want to do is transform advanced technology into things that are fully applied in life and industry, household names. Using a gaming analogy: are you playing a strategy game or a role-playing game? Every firm has its own choices and emphases.

Waves: In the China market, there are plenty of players investing in both games.

Wang: Of course, some do play both well. But everyone has their own capability boundaries. Willingness doesn't equal ability. Linear is a VC with a typical engineering background. We think earlier than most about what we like to do and what we're good at.

Waves: Since last year, Linear has been actively laying out in Europe.

Wang: Yes. Europe is an investment bug — we believe European technology + Chinese scale + global markets present huge opportunity. There are only two unified large markets in the world — the US and China — and only these two places have private markets that support entrepreneurship and innovation. Silicon Valley gathers 80% of the world's best entrepreneurs and the most investment institutions. As a Chinese VC, it's very difficult to invest in the highest-quality entrepreneurs in the US. Looking at Europe: technologically, Europe is on par with the US. The US has deindustrialized, leaning more toward information technology like AI, while Europe has a very long manufacturing history with technology pushing toward the cutting edge. China has three major advantages: manufacturing advantage, engineering advantage, and talent advantage. On this foundation, combining Europe's best technology with China's advantages creates opportunity for global markets. Similar to our past investment in Agile Robots, we hope to extract core elements and replicate them in unique ways going forward. While every success is different, the key elements behind them are things a fund with global capabilities can capture.

Waves: Globalization also seems to have become consensus among dollar funds.

Wang: Our strategy is called: China Amplified Global Arbitrage Strategy — simply put, "based in China, global arbitrage." We hope to leverage China's three major advantages to profit in global markets. Because if we detach from China's core capabilities, overseas we have nothing special beyond some money. Beyond these considerations, Linear has historically invested in companies like Agile Robots and DiSi Medical, whose founders studied and worked in Europe and started businesses locally or after returning. So in 2023 and 2024, we visited many European universities and labs, held events with local students and entrepreneurship communities to deepen understanding, and became more certain that we can use China's three advantages to help European startups grow — even in subsequent fundraising, collaborating with Chinese funds that have global investment capabilities, and using our Silicon Valley network to help them secure US investment and enter the US market. So overall, on globalization: Linear's primary focus is on entrepreneurs who can build global markets from China, while our overseas investment focus is on Europe.

Image source: Pexels | Layout: Nan Yao