Conversation with Zhang Mingchen of Xinglian Capital: Hunting for Hidden Gems at Tsinghua, Observing the Professors' Tiger Valor, Monkey Agility and Bear Resilience

By Hairuo Jing

Another Tsinghua-affiliated fund has raised new capital, this time it's Xinglian Capital founded by Zhang Mingchen.

The young Xinglian Capital invested in Zhipu AI's angel+ round when it debuted in 2021, and kept increasing its stakes afterwards. 40% of the capital in its first fund was allocated to Zhipu AI. The story of aggressive stake increase also happened to Jiliu, Shengshu Technology and Mianwu Intelligence.

For Jiliu, they invested in five or six consecutive rounds and became its largest external shareholder. When Jiliu was raising financing at a valuation of 1.5 billion yuan, the previously agreed investor temporarily backed out. Zhang Mingchen said, "No problem, we will lead the round." In less than a year, Jiliu's valuation rose 10 times.

Before meeting Zhang Mingchen, I thought he might be a radical investor who dares to bet heavily, but he described himself as "quite conservative" and rarely goes all-in even when playing cards. He believes that heavily investing in companies with the potential to deliver a home run and pursuing high returns at manageable high risks is the standard play for US dollar funds, and he calls Xinglian Capital a "US dollar fund in RMB clothing". Now, they have successfully raised US dollar capital.

According to exclusive information from Waves, Xinglian Capital completed the first close of its first US dollar fund in October; the first close of its RMB growth fund in August; and the first close of its third early-stage RMB fund in April. Since its establishment, "one third of the AI projects invested by its first fund are on the IPO track."

The first layer of the "conservatism" Zhang Mingchen refers to means not betting on unproven hypotheses. If a complex task is split into 8 steps, he will not enter the market when only 2 steps are solved; instead, he will wait until the academic community has completed 6 steps, leaving only 2 of the hardest problems unsolved, before investing and joining the game.

When talking about AI technology, he said straightforwardly, "Philosophy is useless, we go straight to quantitative analysis." When a new overseas chip was widely discussed in the market, he did not participate in qualitative discussions, but directly contacted several large model manufacturers to ask about their actual usage results — once he got the facts, the discussion ended.

It is a typical way of thinking for science and engineering backgrounds.

In Zhang Mingchen's view, Xinglian's courage to heavily allocate comes from "being close enough to the portfolio companies".

Zhang Mingchen completed both his bachelor's and master's degrees in the Department of Automation at Tsinghua University, so he can naturally integrate into the circle of Tsinghua's tech entrepreneurs. When Jiliu Technology was just founded with no customers, Zhang Mingchen helped them coordinate a batch of GPUs through friends to complete technical verification. Investment is only the starting point. AI companies change rapidly, and he gets along with the founding teams like colleagues, as "brothers in the same trench". He has full visibility of the key portfolio companies' orders, product pipelines and potential demands, and these "clear cards" make him dare to follow-on invest.

There is often hesitation in the investment circle about "investing in professors", with a hundred reasons both for and against such investments. Zhang Mingchen is a firm advocate of this strategy. He defines Xinglian Capital as a Researcher Fund, and has firmly adhered to the principle of "not investing in large established companies, but investing in universities" since 2021. As a result, 70% of Xinglian's projects come from Tsinghua University.

Now the market has gradually reached a consensus: AI is similar to mathematics, the top 1000 practitioners contribute 95% of the industry's total output. This strategic shift two and a half years ahead of the market allowed Zhang Mingchen to capture opportunities from top domestic researchers, integrate deeper into Tsinghua's entrepreneurship ecosystem, and access information that only circulates in small inner circles. Xinglian has cooperated with 9 AI-related departments of Tsinghua University. "When some researchers just join the research group, their startup projects are already referred to us."

How to judge a professor's commercialization potential? He says we need to look for three traits: the courage of a tiger, the agility of a monkey, and the execution of a bear. How to form top-tier cognition that can support VC monetization? "70% technology, 30% intuition". In Zhang Mingchen's view, when you see enough projects, make enough decisions and get enough feedback, you will form "investment intuition", just like training an AI model.

In the mobile internet era, Zhang Mingchen was one of the first employees of Google China, then joined Redwood China as a partner, and invested in Pony.ai, Opay, Qutoutiao and other projects. How did Zhang Mingchen get to know Tang Jie through Lou Tiancheng? Why does Tsinghua have such a high startup success rate? Will only investing in the "top performers" screened by the education system make you miss outliers? What's the actual feeling of new US dollar LPs after visiting Chinese large model manufacturers?

We talked with Zhang Mingchen about these questions, and below is the edited conversation between Waves and Zhang Mingchen:

Conservative Entry, Aggressive Follow-on Investment

Waves: Is Zhipu AI the first project invested by Xinglian Capital's first fund?

Zhang Mingchen: Yes, we led the round in 2021, then invested in four consecutive rounds afterwards. 40% of the capital in the first fund went to Zhipu AI. We also invested in other projects such as MetaX and Horizon Robotics, and kept increasing our stakes in them later.

Waves: How did you first get in touch with Zhipu AI and make the investment decision?

Zhang Mingchen: When I was at Redwood China, I invested in Pony.ai. Lou Tiancheng, CTO of Pony.ai, was once known as "the top programmer in China". He graduated from Tsinghua's Department of Computer Science in 2004 and was the first student of the renowned Yao Class, many people at Tsinghua call him "Master Lou". One day he told me, "You must meet Tang Jie, one of the best professors in computer science for computer vision, he is extremely outstanding. Professor Tang won the first ACM Best Paper in the history of the department. Now more people win this award, but before 2020, such achievements were extremely rare.

Waves: Was Zhipu AI already starting its business at that time?

Zhang Mingchen: Not yet. But he was already leading a team to work on projects, Yang Zhilin, Liu Zhiyuan and others were all in the same group. At that time, they were at the Beijing Academy of Artificial Intelligence, and developed the first large model in China's whole industry, called "Wudaomodel". Why is it named Wudao? Because the model was trained in Wudaokou.

This team is actually the Whampoa Military Academy of China's large model industry. Many core technical backbones of domestic large model manufacturers came out of this team. They mastered the skills of model training, spread out, and formed the overall landscape of China's large model industry.

Zhipu AI started its business with knowledge graph, that's why it's named Zhipu. The decision to work on large models was to some extent discussed and finalized by Tang Jie and me. Of course the final decision was made by Zhipu's own team, but we played an important role. The first model trained by Zhipu's system was funded by Xinglian's investment.

Waves: Did you ever consider investing in Kimi?

Zhang Mingchen: Of course. But three or four years ago, there was only a limited amount of capital willing to bet on large models at the early stage. The capital supply side was a zero-sum game, so the resources were tight, so we focused our bets on Zhipu AI. Now everyone sees the value, these projects have become "national treasure-level projects" chased by a flood of capital.

Waves: Zhipu AI's stock price has fluctuated sharply after its listing. Will LPs keep asking you about this?

Zhang Mingchen: Yes they do. After listing, all information is public, its market value exceeded 1 trillion yuan at the peak, which naturally attracts a lot of attention. But secondary market fluctuations are very normal, it is not only determined by fundamentals, but also affected by many factors such as macro money supply and sector rotation, just like a spring, fluctuating up and down around the fundamental value.

We pay more attention to Zhipu AI's commercial fundamentals. At the mid-term performance meeting, they disclosed new data, after calculation, the revenue in a single month of August was equivalent to the sum of the previous 6 months, and the growth rate is extremely fast.

Waves: It is not rare to invest in a star project. But continuous follow-on investment really tests the courage of investors. For Jiliu, how many rounds did you invest in to become its largest external shareholder?

Zhang Mingchen: About five or six rounds. When Jiliu first came to me, they only had strong technology, no customers. After our discussion, I not only provided investment, but also helped them get in touch with Zhipu AI, coordinated a batch of GPUs to support their technical verification. So I not only gave them the first term sheet, but also helped them get their first commercial contract.

Why does Tsinghua have a high startup success rate? First of all, their technology is definitely excellent, and secondly, Tsinghua people really unite and help each other when encountering difficulties.

Waves: Among the rounds you invested in Jiliu, which ones were the timely help when they were in trouble, and which rounds did you have to fight for share allocation?

Zhang Mingchen: I remember very clearly, when the valuation was 1.5 billion yuan, an investor who had already agreed to invest dropped out, the company called me, I said no problem, we will lead the round. That was last year, in less than a year, Jiliu's current valuation has reached 15 billion yuan.

Waves: Where does your confidence to continuously increase positions come from?

Zhang Mingchen: We are very close to the enterprises, we get along with the founding teams like colleagues, we can call each other anytime when there is something. So we have a very clear picture of the company's existing orders, future pipelines, potential order demands and so on. External investors may not have such a comprehensive understanding of the business, especially for AI enterprises, the whole situation may change completely in three months.

We are brothers in the same trench, working together to expand the business. So when he encounters difficulties, he will come to me, and when there are good opportunities, he will also think of me. For example, when Jiliu has spare computing cards to rent out, we can also know at the earliest time which of our invested enterprises need computing power.

Waves: Will you hesitate to follow-on invest after the valuation has increased several times?

Zhang Mingchen: Investment should look forward. If you invest at today's price, can you still get 10 times return on investment afterwards? If yes, today's price is a good price. Judging only by the absolute valuation of 10 billion or 100 billion yuan, you cannot tell whether it is expensive or cheap. We know Jiliu's business well enough, so we did not hesitate.

Waves: So heavy position and additional investment are very common at Xinglian?

Zhang Mingchen: We have made multiple follow-on investments in projects such as Zhipu AI, Shengshu, Mianwu and Jiliu.

Waves: Many projects require heavy support and companionship, can you manage them all?

Zhang Mingchen: As more enterprises are invested, some of them have grown up. It's a bit like having many children at home, the eldest takes care of the second, the second takes care of the third, and they operate mutually.

Waves: Will you describe your investment style as "radical"?

Zhang Mingchen: I'm quite conservative. We don't bet on hypothesis, we won't pay for unproven assumptions. For a complex task, if there are 8 problems to be solved, I won't enter the market when only 2 are solved and 6 are left; I will enter at the timing when the academic community has solved 6 problems and only 2 hardest ones are left, provide capital, and let them solve the hardest nuts.

Waves: Will you discuss philosophy of technology? For example, how AI will change organizations and society?

Zhang Mingchen: We don't need philosophy, we only do quantitative discussion, philosophy is useless.

We directly ask researchers: how many steps have you completed for this task, is there any revenue, can it be fully implemented. Before, many people were frantically discussing a new type of chip, many people did qualitative discussions: whether it can be used to run models. I directly asked several model companies: how is your usage experience? They said the performance degradation is very serious. OK, the discussion is over.

When we have formal meetings, we are very rigorous, we talk about technology and practical problems. Philosophy is something we can talk about in our spare time if we want.

Digging for Treasures in Tsinghua

Waves: From the outside world, Xinglian Capital has several labels, such as Zhipu AI CVC, Tsinghua-affiliated, AI native, how do you define this fund yourself?

Zhang Mingchen: We define ourselves as a Researcher Fund. In the US AI industry, the best researchers are in large companies, AI investment funds look for talents around several large companies; China's top researchers are in several top universities, the logic is different.

Waves: So 70% of your projects come from Tsinghua. Is the label "Zhipu AI CVC" the one you most want to clarify now?

Zhang Mingchen: Not really. Zhipu AI is indeed one of our most important partners. But there is a misreading in the market that we are the investment department under Zhipu AI, which is not true. Some projects thought there would be conflict of interest and were unwilling to talk to us. We will explain: Xinglian is a market-based fund, most of our capital is raised through market-based channels, Zhipu AI is not our major LP.

Waves: The projects you invested in before, such as Zhipu AI, Mianwu, Jiliu, are all founded by Tsinghua entrepreneurs. What's your actual feeling of intensively investing in Tsinghua projects?

Zhang Mingchen: Investing in Tsinghua projects is relatively simple.

For projects led by Tsinghua professors, we don't need to do special technical due diligence. Being admitted to Tsinghua already means they have passed strict screening; there are 150 students in one grade of Tsinghua's Department of Computer Science, only one or two of them can stay on campus as faculty. After staying on campus, they still face the "up or out" assessment. Those who can get tenure at Tsinghua are all "world-class talents" in their respective fields. So we mainly judge whether he has enough potential in business.

Waves: Is there any methodology for judging commercial potential? For example, how to judge whether a young PI raises financing to build a commercial company, or just to buy computing cards for scientific research?

Zhang Mingchen: We say we need to "look for three traits", the courage of a tiger, the agility of a monkey, and the execution of a bear.

The courage of a tiger means you need to have leadership, not only do your academic work well, but also can unite more outstanding researchers to start a business. The agility of a monkey means you are smart enough, with strong vision, insight and taste. The execution of a bear means you have strong implementation and landing capabilities. We have also seen many people in the industry who only talk but don't act, who are more like strategic advisors, and such people are not suitable to be the No.1 leader of a company.

The ideal situation is definitely having all three traits. After seeing enough cases, you will form a muscle memory. After a few contacts, we can perceive which trait the founder is particularly prominent in, and what his strengths are.

Waves: Now is the era of technology investment. If the whole industry's peers have done peer review and screening, why did many investment institutions fail to invest in Tsinghua-affiliated large model manufacturers at the early stage?

Zhang Mingchen: First, AI technology is really new. Many established TMT investment institutions do not have sufficient technical stack reserves. Second, whether the institution's cognition is deep enough. Historically, some investors who have invested in multiple Tsinghua projects missed several Tsinghua large model projects and important infrastructure companies, still because they did not understand the "model" track. They