Zhu Xiaohu's 19 Years: It's hard to place a bet, and it's even harder to know when to let go

"I don't get it."

Zhu Xiaohu recently said these three words again.

He exited the humanoid robot track a little earlier. When he looked back, the valuation of Unitree had already rushed to 400 billion yuan. This industry that is still searching for a business model has become one of the most popular sectors in the capital market.

This is a judgment he talked about in a recent interview with Tencent News *Deep Web*. Reading through the whole interview, you will find that "I don't get it" is not only his attitude towards humanoid robots.

In the past 19 years, he has invested in Didi, Ele.me and Xiaohongshu, but also missed Douyin. Three years ago, he said he would not invest in large models. Now, he judges that the story of foundational large models may only have the last year left.

If you only look at these results, it is easy to regard him as someone who always misses the trend. However, what Zhu Xiaohu focuses on has hardly changed: demand, efficiency, supply chain, customer acquisition and cash flow.

From PC Internet to mobile Internet, and then to AI, 19 years have passed, and the trends have come and gone round after round.

When the trend is at its hottest, he often stands outside the crowd. But every time the market starts to calculate returns, people will re-understand him.

I. Trends Can Be Missed

Zhu Xiaohu is now talking about missed opportunities very calmly.

"The market will educate you." This is a sentence he said when looking back on those star projects.

What investors are most likely to be remembered for is often what they have successfully invested in. But there are always some uninvested projects that cannot be avoided for Zhu Xiaohu, and Douyin is one of them.

He admitted that Zhang Yiming is not the type of founder he is good at judging. For ordinary founders, after chatting for more than ten minutes, many points can be seen clearly. Zhang Yiming talks very little, and it is difficult to form a judgment in ten minutes. Zhu Xiaohu is better at judging entrepreneurs with direct expression and clear ideas.

This preference also draws the boundary of his investment. Some missed opportunities can be reviewed, while others are difficult to change. He has accepted this point.

Lashou.com is the most expensive lesson he has ever learned. The general direction was not wrong. Later, Meituan made group buying a huge business, and today Douyin is still doing the same thing. The problem of Lashou.com lies in execution and strategic decision-making.

That failure made him re-recognize two things: how to cooperate with large manufacturers, and how much risk the founding team may bring.

Later, he divided the risks of early-stage investment into three categories: market, technology and team. He can take at most one type of risk at a time, and he is the most unwilling to take team risks.

"Is this person honest, willing to tell you the truth; does he understand and is he good at what he wants to do."

It doesn't matter so much whether the resume is impressive or not. Whether the person is reliable is the hardest part to remedy in early-stage investment.

Zhu Xiaohu also said in the interview that venture capitalists are actually the group of people who fear risks the most.

Only when the project has a sufficient margin of safety can he sleep well at night. This sentence summarizes his investment style more accurately than the word "prudence". He does not reject risks, but is unwilling to take a risk that he cannot explain.

No matter how hot the market is or how grand the story is, he has to pass his own test first. Such choices will certainly lead to missed opportunities.

"It is unrealistic to expect to seize all opportunities."

Humanoid robot is the latest example. Zhu Xiaohu began to be alert to the valuation of this industry very early. He admitted that the bubble blew much larger than he expected.

In his opinion, the 400 billion yuan valuation of Unitree is too exaggerated. There are nearly 200 companies in the industry, but the real commercial orders are limited, and many demands still come from government and scientific research projects.

Investors see a highly demonstrative future, but Zhu Xiaohu did not chase in again. He said he could not understand this track anymore, and GSR Ventures no longer focuses on it.

For an investor, publicly admitting that he doesn't understand is not a smart move, but this is his boundary. He allows himself to miss opportunities, and allows himself to not understand.

II. All Stories Eventually Need to Be Counted for Returns

Zhu Xiaohu doesn't like projects that make him unable to sleep at night.

Calculating returns is not only his investment method, but also his way to control risks.

When others discuss how much imagination space a technology can open up, he often asks two questions first: with such a large investment, how long will it take to earn the money back? Where will the profit come from?

When talking about AI, he is still asking these two questions.

Last year, when Internet and technology companies announced increases in capital expenditure, their stock prices often rose accordingly. This year, the market reaction has changed. Investment continues to increase, cash flow is tightening, and stock prices are under pressure instead. On the other hand, software companies that were once thought to be replaced by AI are seeing their revenue and profits improving.

The market began to ask: When will the money spent on AI be reflected in the income statement and profit statement?

These two days, the capital market has just given a very direct signal. Zhipu AI just announced a financing of about 5 billion US dollars, and MiniMax is still talking about model capabilities and commercialization, but on September 14, the stock prices of the two companies plummeted together. On September 12, OpenAI confirmed that it will not go public this year.

The popularity of AI has not disappeared, but the capital market has started a new evaluation. What the model can achieve is no longer the only question. How much money model companies can make in the end needs to be calculated more and more clearly.

This is exactly the account that Zhu Xiaohu has been calculating.

He does not think there is a bubble in the entire AI industry. After the computing power centers are built, their products are still easy to sell, the demand for infrastructure such as storage and memory remains strong, and AI applications have just entered the stage of large-scale commercialization. The bubble mostly appears in the valuation of individual companies.

His judgment method is very straightforward: first look at the PS (Price-to-Sales ratio), then see if the revenue can be sustained, and finally return to the PE (Price-to-Earnings ratio).

AI Coding is one of the first scenarios to generate revenue right now. Zhu Xiaohu calculated another account in the interview: the narrow Coding market may only be two to three trillion US dollars, which is difficult to support the annual one trillion US dollars of capital expenditure alone.

The market still needs to find the next large scenario after Coding.

Zhu Xiaohu has never been very optimistic about the long-term barriers of model companies. In his judgment, models will eventually become more and more similar to public utilities such as water and electricity.

Cutting-edge models can still obtain high gross profit by virtue of performance and scarce computing power at present. With the continuous catch-up of open-source models and the gradual popularization of computing power, it is difficult for API prices to stay at today's level for a long time.

"In the long run, the gross profit margin of APIs as public utilities like water and electricity will be 10% to 20%."

Looking back, the reason why he did not invest in large models in the past was that he focused on the long-term profit structure. Recently, he made the timetable more specific. This year may be the last window for foundational large models, and the capital market will need new stories next year.

This perspective sometimes seems too early and easy to miss the opportunities in the process, but it has always been very stable.

The accounts of the application layer have become clear instead. When the foundational model is good enough and the calling price continues to drop, application companies finally have the opportunity to figure out their business models.

Now when he looks at AI application projects, the most direct indicator is the growth rate, and a monthly growth of more than 20% is just the starting point. He said that this round of AI entrepreneurship basically has no thresholds or barriers.

This judgment sounds a bit harsh. With the rapid spread of technology, the differences between startups will eventually fall on execution and growth curves.

WorkBuddy follows the same logic. What Zhu Xiaohu cares about is not the model parameters.

In his opinion, the model can be replaced, but the organizational relationship, knowledge base and historical communication data accumulated by the enterprise cannot be replaced. When he looks at AI applications, he will eventually return to whether the customer relationship can be precipitated as an asset.

GSR Ventures itself has also purchased a WorkBuddy account to analyze the monthly financial statements submitted by invested enterprises. The analysis of cash flow, profit and revenue that used to take several days to complete can now be generated in a few minutes, and indicators and charts can be adjusted at any time.

This kind of efficiency improvement generated by real use is closer to the commercial value that Zhu Xiaohu cares about than model parameters. The indicator he set for the WorkBuddy ecosystem is also very specific: can there be more than 100 partners each earning more than 100 million yuan?

The grand ecosystem has finally been turned into an income statement by him.

III. The First Place May Not Stay in the End

Zhu Xiaohu has another judgment on AI:

"It is more important to survive longer."

This sentence comes from his observation of multiple technology cycles. The earliest entrants have to bear the costs of immature technology, uncertain market and a large number of trial and error. After the direction is verified, followers can follow the proven path and catch up quickly at a lower cost.

AI has accelerated this process again. Training a cutting-edge model for an American company may cost 500 million to 1 billion US dollars. Chinese companies only spend a small part of that cost, and have the opportunity to catch up after a few months. The faster the model capabilities spread, the shorter the validity period of the first-mover advantage.

Zhu Xiaohu no longer believes in the myth of being the first.

He has seen too much competition in the technology industry after completing the process from 0 to 1.

Solar energy, panels and electric vehicles have all attracted a lot of capital to enter. After the technology matures, price competition follows. The industry expands production round after round, and enterprises exit in batches. The companies that finally stay are not necessarily the ones that took the lead at the earliest.

This set of industrial rules may also appear in the humanoid robot sector. Zhu Xiaohu pays special attention to supply chain, efficiency, channels and brands. After the technology begins to converge, industrial giants such as Huawei, Xiaomi and BYD can quickly catch up with the help of mature manufacturing systems. The fact that startups run out first does not mean that they will definitely stay in the end.

There is a detail in the interview that can well illustrate his choice. Humanoid robots can appear on the Spring Festival Gala, and can also complete a set of beautiful movements in the center of the exhibition booth. It is intuitive, lively and very suitable for display.

GSR Ventures has invested in a robot company that cleans ships on the seabed. The machine goes into the water to clean the hull, photovoltaic piles, and can also assist the customs to check smuggled goods hidden under the ship. Its commercialization has already started, but the audience can hardly see it working from the shore.

Zhu Xiaohu chose the project that is not easy to be noticed.

Ships need to be cleaned, ports need to be desilted, and customs also need safer inspection methods. These demands have long existed, and the machine only needs to prove that it is more efficient than labor.

Compared with the high valuation of humanoid robots, this kind of industrial automation project is much cheaper and easier to calculate the revenue clearly.

In the past two years, some AI companies released products at the earliest time and got attention first, but then their popularity gradually faded. Other companies did not start early, but quickly generated revenue after the model cost dropped.

The faster the technology spreads, the less important the starting line is.

The late-mover advantage Zhu Xiaohu talks about is not just entering the market a little later. Latecomers do not need to bear the first round of trial and error for the entire industry, and have the opportunity to re-compete at a lower cost.

IV. Finding the Constants in Changes

What really puzzles Zhu Xiaohu about AI is the speed of change. He said, "The speed of mobile Internet is about twice that of PC Internet, and the AI era is three times that of mobile Internet."

As someone who has experienced 19 years of technology investment, he still admits that he is confused. Faced with such a speed, the solution he gives is very traditional: meet people, meet people, and must meet people on the front line.

Entrepreneurs face users, orders and cash flow every day. They are the first to feel the temperature of the market and know where the real demand lies. Zhu Xiaohu still understands the latest technology by constantly meeting people, and then corrects his judgment from the rapidly changing information.

"We need to see those constant factors from such rapid changes."

This sentence is more important than any of his judgments on the AI track. Large models may move toward public utilities, office software will be reorganized, humanoid robots will still experience reshuffling, and new entrances to AI applications will continue to emerge.

The life cycle of technology is getting shorter and shorter. If you keep chasing the technology itself, it is easy to fall into the next round of narrative. But people still need to solve problems, enterprises still need to make money, and products still need users. Startups cannot avoid customer acquisition, and industries cannot do without efficiency.

AI has accelerated the spread of technology, and these old problems have become important again.

Zhu Xiaohu is currently focusing on AI applications, industrial automation and intergenerational consumption at the same time. The three fields seem far apart, but he uses the same standard: the demand already exists, the technology can significantly improve experience or efficiency, the commercialization path can be verified, and the valuation also leaves a margin of safety.

AI companion toys are an example. When talking about this project, he mentioned his childhood in the interview. In the 1980s, virtual pet toys were popular all over the world. The product experience at that time was very simple. He couldn't afford it himself, but he remembered that the whole world was buying them.

Many years later, he invested in an AI companion toy company. Two months after the product entered Japan, it sold for 80 million US dollars.

What attracted him was not the concept of "AI toy", but a demand that had been proven decades ago, which was redeveloped with new technology. The experience has improved, and payment has followed.

The same is true for consumer investment. When he looks at outdoor jackets, he focuses on young people's working methods and dressing habits. When he first looked at Ele.me in the early years, he saw that young people's lives were changing. At that time, there were also companies selling semi-finished dishes at the subway entrance, waiting for young people to buy them home and cook by themselves. A few years later, the post-90s generation got used to ordering takeout.

Technology creates new opportunities. The really lasting opportunities are often hidden in the changes of people's behavior.

Zhu Xiaohu calls his investment method "15 degrees off the mainstream". 15 degrees does not mean leaving the mainstream. It just allows him to stand next to the hottest place and see what else has not been noticed.

This is the case with AI companion toys, as well as the changes in young people's consumption and industrial automation. He doesn't need to find an industry that is always right. He only needs to find a demand that already exists, and wait for technology to redevelop it.

19 years of investment career did not make him find an industry that will never change. He just became more and more clear about what is worth waiting for and what money he can choose not to earn.

Words Beyond the Page:

The real cycle is not to predict the next trend, but to know that the trend will eventually pass.

After 19 years, what Zhu Xiaohu left behind is more like a set of investment models: knowing what is worth betting on, and also knowing what can be let go.

The latter, obviously, is more difficult.

The market always likes an investor who stands on the trend. But Zhu Xiaohu probably does not plan to be such a person.