125. 与Altimeter合伙人Freda聊：下注OpenAI、Robinhood往事，美国资本坏小孩、算盘与泡沫

Today's topic is very clear. There are three main lines in the U.S. The

first line is definitely AI. The second line is re-industrialization. The third line is digitization of finance. It's the innovation of a financial industry. It

was first seen that OpenAI was a product company, not just a model company.

There's also a rumor that Norm Shazier, the founder of Character AI, changed his bug after returning to the US. Robinhood's trading model is basically bad. It's a mess. Stock trading itself has a very strong cycle. It's not a good business.

strong cycle. It's not a good business.

Hello, everyone. Welcome to Zhang Xiaojun's Business Conversation. I'm Xiaojun. This is a deep conversation program produced by the Language and World Workroom. We hope to explore the new world with you from here. In our annual review series, which is in the first two episodes of the series other than 2025, Zhu Xiaohu said, I think many

investors are deliberately leading. They just want to adjust the price. Maybe they will adjust the price for a couple of months. I think it will be a good technology for the next year.

而戴宇森则预测。

26年呢,我有一个关键词叫26年是year of R, 就是R是年。

这个英文字母代表着几个含义。

第一个R,我的含义呢,我觉得叫return,就是回报。

你觉得站在2025年的尾巴上,你有嗅到泡沫的味道没有?

人类历史上每次技术革命都带来了泡沫,几乎毫无例外。

不管是从当时的修运河,还是修铁路,还是修高速公路后来的互联网, We are going to talk about the second episode of this series, Gui Gu Vision

from the First Line. A month ago, in early November

Line. A month ago, in early November of 2025, Sam Altman hosted a podcast program hosted by the American foundation, Ultimater Capital. When the host asked OpenEye how to buy a $1.4 trillion-level

Ultimater Capital. When the host asked OpenEye how to buy a $1.4 trillion-level algorithm and infrastructure, Sam said, if you want to sell your shares, I can help you find a buyer. That's enough. Then, the AI version of the whole appeared to fluctuate. The discussion about whether AI exists bubbles further heated. Our guest today,

to fluctuate. The discussion about whether AI exists bubbles further heated. Our guest today, Freda Dunn, is from Altimeter Capital. She is a partner. Altimeter is a shareholder technology fund across the first and second-level markets. In the first-level market, the investment case is OpenEye, Anthropix, and Self-Driving. In the second-level market, the investment case is Invida,

Snowflake, Robinhood. In this episode, Frida will analyze the huge investment of

Snowflake, Robinhood. In this episode, Frida will analyze the huge investment of American star companies. She also talks about the new order of American capital from the perspective of a stock investor. The bad children in their eyes, the rebels, the founders of the four-dimensional and Nezha type, and the bubble. I look forward to our joint progress in 2025 with AI.

Hello, Freda. Let's say hello to our audience first and make a brief self-introduction. Hello,

everyone. I'm a technology investor based in Guigu. I'm in a fund called Altimeter Capital.

Freda, you are one of the few Chinese in the US E-Line Fund. Can you

introduce your fund and your founders and the background of your foundation? We

are a company based in Guigu, covering the first and second-tier markets. Our

founders are serial entrepreneurs. We are famous for our open AI, anthropic, and self-sufficiency companies. In the second episode, we are more likely to invest in non-competitive investments, including Meta in 2022 and 2023, Invitas, Robinhood this year, and many other big software companies such as Snowflake. One interesting thing that is different from

the rest is that we are rare in the U.S. and are willing to publicly research and think about the fund. Many people know that we are a podcast called BG2. We've had a lot of interviews with Sam Altman and Zhen Zenghuang in

BG2. We've had a lot of interviews with Sam Altman and Zhen Zenghuang in recent weeks. Our partners will also share their insights on Twitter and the sub-site.

recent weeks. Our partners will also share their insights on Twitter and the sub-site.

I'll be sharing about AI and some advanced technologies, including AI applications, Internet, hardware technology, and financial technology.

说起来我觉得很恍惚,因为21年,22年,我记得硅谷还是笼罩在财源的风波里,那到了25年,似乎整个的情绪和叙事都发生了非常大的变化。

你看MyTie又是在天价的挖人,那身为一线投资人,你觉得这几年你的真实体感有变化吗?

你们描述一下在每一年,你身处其中,你心里能够定义那一年的一些瞬间。

每年说几个重点词吧。

2020 is the year of the pandemic, the fall of the pandemic, and the market's boom. 2021 is the year of the market's continued boom. But you can actually see

boom. 2021 is the year of the market's continued boom. But you can actually see some signs of a trend that is like a tide of water. For example, in stocks, small and medium-sized companies or stocks that are not of such high quality, their peak is in February 2021. Then the big stocks, similar to micro-soft, started to fall in November. In 2022, the stock price fell by 35% due

to the surge in the market. The stock price of Meta fell by 70%. In 2023, when everyone was not reacting at all, AI started to become

70%. In 2023, when everyone was not reacting at all, AI started to become popular. In 2024, AI was the trend. It was actually the best year

popular. In 2024, AI was the trend. It was actually the best year for us to invest. In 2025, the tariffs were lifted. The Sino-US relations were still in the flow of AI. What do you think of these ups and downs in recent years?

其实一共只有六次是下跌幅度超过20%的。

但是在2020年之后,我们就经历了三次,所以这个频率是有明显变高的。

那你看23年之后,突然一下整个市场就被AI主导了。

回头来看,你觉得在当初有没有一些端倪,有可能更早的预见吗?

还是说还就是非常的突然?

我觉得其实是有非常明显的预兆的。 The market was in a dark place in 2022, but there was a very strange financing phenomenon. There was a company called Inflection AI, which was a large model company owned by Mustafa. At that time, it was considered a model company at least in the top five or top ten. It made $200 million in the A-level. And not just one company, but many

ten. It made $200 million in the A-level. And not just one company, but many companies, including Anthropic, Cohere, Stability, were all internal model companies. Each

company made hundreds of millions in the A-level. There are a few strange things about this. First, they are actually using VC money to make capital. Second, no one

about this. First, they are actually using VC money to make capital. Second, no one has ever given a million dollars to A-class. On average, they should have given ten million dollars. - - - - - - - -

million dollars. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - So sometimes, the logic of investing in the right time and opportunity is so simple.

That is to say, the money of VC is all rushed to the end.

Yes, you can say that. It also talks about the benefits of a crossover fund.

It's all done in one or two seasons. In the early 2023, There were 20-30 companies that were doing large-scale models. It was hard to tell which company would become the first to take over. But because our foundation is a crossover, we work in the same team. So I found it easy to use Invida to express our views on models and AI in the second-tier market. In the

past two years, most of the AI standards have been in the second-tier market, including cloud, electricity, AI benefits and losses. It's been two years since the beginning of the third year of the pandemic. Our blogger has also invited many AI founders to chat. What do you think the top US investors are doing today? What is their

chat. What do you think the top US investors are doing today? What is their overall mood? Today is actually very clear. There are three main lines in the

overall mood? Today is actually very clear. There are three main lines in the U.S. The first one is definitely AI. The second one is called reindustrialization.

U.S. The first one is definitely AI. The second one is called reindustrialization.

The third one is called Digitization of Finance. It's a financial innovation. The three main lines are very interesting because they have a relationship between a thousand and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred

and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and

four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four

hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred

and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and

four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four hundred and four

hundred and four hundred and four hundred and four hundred and four hundred and four hundred and The industrialization line includes a lot of things, including the return of the American local manufacturing industry, including the data center. It will take a long time to see if the robot can pull up the level of American and new market labor. I think everyone may not realize that in April, Trump made a

deal with the US government Japan and South Korea agreed to invest more than $10,000 billion in the US. This money will be used to invest in US infrastructure projects, including energy and data centers. This is the first line we just talked about, which is that AI has realized a wall. The most worrying thing for the market

is whether AI investment is a bubble and where the money comes from. I

think part of it is actually to consider this. Because we have seen that Japan has invested more than 100 billion in energy-related projects in the United States to power these data centers. The third one is called "Innovation of the Financial Industry" It's also very interesting. This year in July, the U.S. has a very important bill called the "Genius Act" It's a stable currency bill. Stable currency

payment is what we call "Cheaper, Faster, Better" If there's an agent of commerce next month, and the agent helps me pay for my stuff, I think the concept of "token" or "stable currency" will be re-mentioned. This is also the same as the first one we just talked about, AI forms a currency ring.

I'm curious about how investors in the US view the Chinese market. It's very lively, but after following it, I feel that

Chinese market. It's very lively, but after following it, I feel that the actual investment amount has not changed significantly in the past few years. But it

is true that a number of American investors have been visiting China since this year.

Everyone who visits has been very impressed by our EV and robot. I understand

that the most important thing for American investors is two things. One is the result of the Sino-US negotiations. The other is the activity of the Hong Kong stock IPO. I think the IPO has already come up from the perspective of quantity.

stock IPO. I think the IPO has already come up from the perspective of quantity.

But more of the companies that are in the consumer category are listed in the Hong Kong stock market. If there are more technology IPO after that, I believe some of the investment amounts in the US will return to the Chinese market.

You mentioned that you have invested in some big US companies and they are all heavy-duty. Can you also talk about these specific companies? For example, OpenEye, you

heavy-duty. Can you also talk about these specific companies? For example, OpenEye, you are its shareholder. When did you see such a trend and started to invest? What

was the situation like at that time? It was the earliest time to see that OpenEye was a product company, not just a model company. When was this?

When I first saw that OpenEye was a product company, not just a model company, I thought it was a 2 billion-dollar company. The market was very chaotic, and no one could tell which company

company. The market was very chaotic, and no one could tell which company was the real model company. This conclusion sounds very common now, but in 2023, everyone was very worried about the competition pattern of 2C. They felt that GPT had no annuality. At that time, the competition pattern was not clear, including perplexity, cloud, and grog. A good product is not only about the model, but also

about the user experience. If you think about GPT, it's not a very effective online business model like Meta. It's a one-to-one distribution, but it's its early release advantages and better user experience that make the product more sticky. OpenEye has completed a 5 trillion financing. We saw a media

sticky. OpenEye has completed a 5 trillion financing. We saw a media report that its income forecast in 2030 is 2 trillion. But now there is still a lot of money. Can you tell us what you think about its business model?

So if you only have one year of training

cost, 你第二年大概是能以二这个收入去收回上一年的训练成本。 But at the same

你第二年大概是能以二这个收入去收回上一年的训练成本。 But at the same time, you need to cut down on the cost of training for your new model.

So it's a very simple plus 2 minus 10, which becomes a negative 8. The

essence of the big model business model is a negative snowball. Every year will be more burning than the cash flow of the previous year. If one day the cash flow can turn over, then logically there are only two possibilities. Either your

second year's income is no longer twice the cost of training the previous year, but a bigger number. 要么就是你不再去烧一个新的十倍的模型,

只有这两个可能性。 The CEO of Anthropia, Dario, has also talked about this logic. He has also divided the latter, which is the "stop burning money"

this logic. He has also divided the latter, which is the "stop burning money" into two parts. He said that there are two possibilities to stop burning money. Either

you have a physical limitation, the model is too big, you can't move from a training angle, or you have to burn all the money in the world. The second

possibility is that the model itself is okay, but because the scaling law has slowed down, it is not worth burning another 10 times as big a model.

结束的时间点其实非常有暴力美学,公司的利润表会非常好看。

因为就像我们刚才举的最简单的例子,你加2但你不减10了嘛,所以就是一个非常简单的数学问题。

所以Skinny Law消失对于投资人来说还是个好事儿是吧,从算术上来看。

对,在那个时间点,投资人会看到他们最想看到的这个利润率。 OPI is indeed a magical business model. In the field of technology, there are very few capital-intensive businesses. But

it's not like it hasn't happened before. For example, Netflix. The Netflix cash flow was paid in the past few years. And it's also like a snowball. It's

paid more every year than the previous year. Because Netflix is also like that. You have to spend a lot of money to make a drama and content.

that. You have to spend a lot of money to make a drama and content.

And its content is also a discount. It's also a four-year discount. So if

you think about it, many places are quite similar to the data center. The most

miserable thing about Netflix on the cash flow is that it paid 3 billion in 2019. What's interesting is that in 2020, the cash flow suddenly skyrocketed to

in 2019. What's interesting is that in 2020, the cash flow suddenly skyrocketed to 2 billion. This rise is very alarming. Then you go to see why. It's

2 billion. This rise is very alarming. Then you go to see why. It's

because of the epidemic. So everyone can't come out and shoot movies and TV shows anymore. So you call it content cost or training cost, there is no

anymore. So you call it content cost or training cost, there is no more. So it is to say in the most straightforward words, the money you burn

more. So it is to say in the most straightforward words, the money you burn will only be real on the day you don't burn it. This is the essence of a model company. But if a content platform like Netflix doesn't invest in content, the advantage of the brand will disappear. You can't just not do content for Disney. Yes, I think you're right. But Netflix is not completely without content

Disney. Yes, I think you're right. But Netflix is not completely without content now, it's just that it's growing. Unlike its previous period when it was in the user count, it needed to expand its content so much. Now there

should be more than 300 million subscribers worldwide. They are all very young. All they

need to do is maintain their position. I think the same goes for model companies. I don't think they will completely stop training one day. But even

companies. I don't think they will completely stop training one day. But even

if the training cost doesn't increase several times as much as it does now, it's enough for the company's profit. OK, I really haven't heard of the difference between Netflix and OpenAI before. Do you think the structure of this industry will be similar to the current media in the future?

内容是没有排他性的,所以媒体本身竞争格局都比较分散。

你看YouTube, Netflix,像你刚才讲的迪斯尼,每个人可能都是个位数的一个市场份额,因为内容上来讲并不是一个完全商品化的内容。

Just like we have to chase different dramas on different platforms. And you would want a family to record all your content. I think GPT is more like a search pattern. The market is very concentrated. Google and Microsoft Bing are the two

search pattern. The market is very concentrated. Google and Microsoft Bing are the two search engines that occupy the entire market. Users choose what they like to use more.

For OpenAI, other than ChatGPT, what other income do you have? What is its income structure? The media reported that it has four income. The first one is the GPT,

structure? The media reported that it has four income. The first one is the GPT, which accounts for 70% of the total. It also includes the company's GPT.

The second one is the API. If you compare it with OpenAI, the API can meet half of the Anthropy's income. But in the long term, Anthropy is expected to increase faster, reaching five times or even more than the OpenAI API's income. The

third line is the agent, including the collaboration with the software. These are new products. Today, most of the discussions in the market are focused on GPT, but the

products. Today, most of the discussions in the market are focused on GPT, but the company predicts that the latter two lines will grow faster and have a bigger future. If you only look at GPT, you can see a 300 million subscription,

future. If you only look at GPT, you can see a 300 million subscription, because Netflix is such a value, and then there are 30 yuan a month, which is about 100 billion in volume. In addition, everyone is discussing 2C GPT, and seeing it as a purely consumer company. But ordinary investors may not see that the ratio between them and the company is similar. I think the

market underestimated its enterprise segment. The enterprise segment has millions of users. For

example, I am one of the users of the enterprise segment. I use

GBT to connect my company's email, Slack, and my company's internal system. This is

very easy to use and has a lot of imagination. Because the U.S. has a

large enterprise market. China doesn't have this concept.

很多人说OpenAI想做一个超级入口。

- Another reason is that the business itself is very resistant to this. Because

it hopes that users can come directly to its platform. For example, Amazon. Only

when I'm gone, can it recommend me to buy toothbrushes and toothbrushes to increase sales.

And Amazon has more than 7 billion in advertising revenue today. If you buy it from an agent, it's a problem that the advertising revenue can't be collected.

But you see, it's actually much better in the business.

I think OpenAI is going to start selling ads. Are there

ads. Are there any other ways to do that? Yes, he doesn't have any advertising income today, but the logic of advertising is so smooth.

He has more than 800 million active users today, and about 50% of them are paid users. He can tell users that if you are a free user,

paid users. He can tell users that if you are a free user, you must make money, so you will advertise. For example, I search for running shoes.

If I am a free user, I may see that ad. If I'm a paid user, GBT will guarantee me that its results are absolutely smooth. So you

will see many other small and medium brands. Advertising itself will go up very quickly.

Another interesting business model I think is cloud. Sam Allman also said that he is a business that is very natural extension is cloud this business. Because if you think about how Amazon AWS came about, it was actually built by itself, and then it found that it could not fully control its use of data centers.

Because the data center itself is a very important business model. So it's a bit of a random thing. Maybe when you don't need it so much, you rent it out.

-

-

Today, Gemini has a monthly revenue of $6.5 billion, and GBT has more than $10 billion. I think Google can play a role in the price war of OpenAI

billion. I think Google can play a role in the price war of OpenAI from several angles. First, absolute price. Because Google uses TPU to integrate, the profit rate of the same model should be much higher than that of OpenAI. Second, tied sales, which is a classic sales model. For example, now I am a YouTube user who makes 20 yuan a month. It's completely possible to say, "Add 10

yuan to my Gemini." As long as you tie it up, it's easy to generate higher profits. So recently, Gemini has had a certain impact on the OpenAI user market. I think the most reasonable way to do it is for Google to

market. I think the most reasonable way to do it is for Google to eat a medium-low market. For example, 20 yuan per month subscription level, and then leave the high-end market for a few hundred yuan or even a few thousand yuan a month for GPT. Which of these two will you choose?

I think they are all winners. But I think the whole market is a little bit on the side with the negative emotions of OpenAI and the positive emotions of Google. Because if you think about it, the threat that Google faces is not really solved. Today, Google's search ads are still the same. It's because

today, we still use GPT as a big Wikipedia. We use it to do research and ask why it's blue. Google doesn't change these questions. But

for example, insurance, law firm, these are all the biggest advertisers today. For

example, I buy a car insurance. I may Google it today and choose the top advertising position, which may be Geico. But this question is too suitable to ask GPT. The GPT will give me a comparison and I will choose the one that

GPT. The GPT will give me a comparison and I will choose the one that suits me best. In this, the GPT may already have my history, such as knowing that I drive badly, so it will customize the one that suits me best. When the GPT starts doing advertising, when it starts doing Argentic Commerce this

me best. When the GPT starts doing advertising, when it starts doing Argentic Commerce this month, I think this pattern will change. So, to put it in a bigger way, the annual revenue of advertising in the US is this much. Today,

Google and Meta are monopolizing the market. If OpenEye wants to sell ads, it has to start from these two companies. So in summary, the competition between Google and OpenAI does exist. But I think the threat that Google faces is also real. This is very interesting. When OpenAI and ChessBD first came out, they

also real. This is very interesting. When OpenAI and ChessBD first came out, they all said that this should replace Google. How did Google suddenly change the pattern?

我看巴菲特也买了。

对,主要就是模型能力上来了,以及大家发现它搜索广告目前还没有掉这个份额。

那模型能力这块也很有意思,我听说内部是找到了优秀研究员的正确打开方式,就是它把激励制度 are all tied to this model list. Whoever can brush up the list will get promoted. So it's very suitable for the student-boss culture of Google these years. And

get promoted. So it's very suitable for the student-boss culture of Google these years. And

there are also rumors in the world that Noam Chazir, the founder of Character AI, said that after he returned to Google, he changed a bug by himself. Since then,

Gemini's pre-training has been easy to use. When Gemini 3 came out, it also found that the pre-training did not hit the wall, so the whole trend is continuing to improve. That must be a bug worth billions of dollars. Yes,

this is a world of genius researchers. I think that Noam, as a researcher, is more suitable than a CEO of a 2C company. Do you think Google's search engine will change?

-

Google search was indeed a good way to suppress the search. This is

because I can always find the information I want in the top two or three articles. This convenience is the most important thing for people. No one will see

articles. This convenience is the most important thing for people. No one will see the results of the second or third page. But now you can use a model to search. This model itself is very hard to work with. It can

turn a few pages in a second until it finds the most suitable information. So

you use Bing or your own search engine, there may not be a big difference.

I think this is very big, because the search for this household name has completely changed. How do you see the management of OpenAI? Sam is not a technical

changed. How do you see the management of OpenAI? Sam is not a technical background, but he can be the number one. And the employees are very lucky to him. This is very interesting in itself. I talked to them and I

to him. This is very interesting in itself. I talked to them and I think the employees trust him very much. I also thank Sam for leading the employees to pay. Another thing I think you must give Sam credit is

to pay. Another thing I think you must give Sam credit is that he was the first to believe in AGI in this market.

This is what he led the market. Recently, we've all heard about the 1.4 trillion dollar number. In fact, he is very smart in his own way.

Because AI is very money-saving, he wants to tie all these lines together. But I

think he's too far off on the timeline. The numbers are too big. It's a

market scare. If he only directs one or two years of spending, the results may be much better. But I think it will take some time for the market to slowly digest. Do you think OpenEye has been the best investment in recent years? There is a concept in it. We see from media reports

recent years? There is a concept in it. We see from media reports that whether it is 3 billion or 5 trillion, it is said to be alone. But the investors get the share price of each stock at a time. In

alone. But the investors get the share price of each stock at a time. In

this, OpenEye has a share price of 3 billion to 5 trillion. Its

share price per share is not nearly 20 times, but 6 times. Because the market share is very serious. How do you see OpenEye's recent series of market operations? I think this is very reasonable. And recently, OpenEye and Microsoft have

operations? I think this is very reasonable. And recently, OpenEye and Microsoft have finally made it clear that this is very important. For example, the media reported that OpenEye may have a 10 billion market share IPO in 2027. It is

about 10 times the valuation of its income forecast. So as long as its income can do it, this valuation is reasonable. Let's be

honest, in Boston, there's a huge fund, like Fidelity, that will give out $1 billion to everyone. I know a lot of people in the media are saying that they're waiting for OpenAI to go public. I

don't think it's that bad. Because even if you see the last wave of Internet, like NetScape, which went public in 1995, it was indeed more and more bubble after that. But the breakdown is also five years later.

after that. But the breakdown is also five years later.

国内在这波AI浪潮里面, 经常会说一个词叫Club Deal, 俱乐部交易。

所以现在像欧凤海, 它不仅是在一级市场Club Deal, 它也要去二级市场Club Deal, 大家一起来存钱投。 对,不仅是大家一起存钱,

而是大家都非常愿意把钱放到这个AI的领军公司里面。 因为... you are the closest to the entire AI application. I feel that in the AI wave, companies are more concentrated.

Yes, because AI itself is a very money-saving business. So this money will only burn a few big companies. Speaking of OpenAI, let's talk about Anthropic. What

is the similarity between these two companies in your observation? First of all, the business is different. One is 2C, one is more like 2B. Anthropic's business is more than 80% of the revenue is 2B. From the perspective of user year, it's actually good. Both companies have a slight curve. At the beginning, the user's retention may

actually good. Both companies have a slight curve. At the beginning, the user's retention may go down a little bit, but then it will come back. And new users also have higher retention rates than old users at the same time. So there

is no problem in this regard. And the media should also report that the profit rate of Unprofit is growing very fast. In a year, he has achieved a profit of about $1 and a loss of $2. By the end of this year, he can achieve a profit similar to that of OpenEye. So he

has done a lot of work in this. It also gave me a big shock. He has a great control of his profit. Whether it's the model structure or

shock. He has a great control of his profit. Whether it's the model structure or the auto scaling, including optimization, he did it very quickly and very well. So now the overall model company, I don't think there's

very well. So now the overall model company, I don't think there's anything to worry about the profit margin. It will be 70% to 80% in the long term. From the perspective of growth, these two are in a state of

long term. From the perspective of growth, these two are in a state of chasing. From 100 million to 10 billion income, OPI has grown very fast, using

chasing. From 100 million to 10 billion income, OPI has grown very fast, using less than two seasons. Then I'm probably using four or five seasons. But 10 billion is the opposite. The annual income of Unproper is less than 10 months. The

annual income of the two companies that have increased in the past few months is actually similar in number. The product line of Unprofit is also very clear. One is to choose a few larger, more profitable trades, including

clear. One is to choose a few larger, more profitable trades, including coding and finance. Cloud Finance is a very useful product. It also does some trades related to life science and law. On the other hand, it tries to make the best model for agents, including its Cloud Skills this year, which is a mini

agent system. I'm still going back to the commercial model of the big model that

agent system. I'm still going back to the commercial model of the big model that I just mentioned. I don't think it makes any sense to leave the commercial model and look at the company's cash flow. The financial prediction of these two models is very interesting. Improper means that his total profit after 28

years is not increasing. So the second point we just talked about is that the cost of training is not increasing rapidly and the profit rate is rising.

And if you look at the "unpropry" part, it's actually assuming that the ROI of the model is increasing year by year. This corresponds to the first point we mentioned, which is that it's not just 1 yuan to recover 2 yuan, but 1 yuan to recover 5 yuan in the second year.

It's by using this to pull up its own profit. So it's not that "unpropry" spends less, but it assumes a higher return rate. Of course, this prediction itself is very interesting. It's worth thinking about why the two companies made two directions.

Do you think there are any progress in other model companies? For example, the SSI from Iliya and the Thinking Machine Lab from Miro. First of all, there is a very important thing. There are not only two or three companies now.

In recent months, there have been ten small new model companies financing. We

call them Neo Labs, like Neo Cloud. Many of them are from academic backgrounds.

The financing amount is very large, more than $1 billion. Actually, I didn't really think about it myself. I don't know if there will be a time point after that. Everyone will think that the big model will actually be a couple of

that. Everyone will think that the big model will actually be a couple of companies at the top. Then the rest will be to give up training and embrace open source models. Back to the few companies you mentioned, which are also familiar names. SSI from E-LIA is doing continuous learning. And Mira just

familiar names. SSI from E-LIA is doing continuous learning. And Mira just launched their first product called Thinker, which is to help users customize open source training models. I'm also thinking that if you invest in this, the open source market will actually grow in the BAT market. I see that these model companies are

all financing. How do investors give these companies a valuation? They will look

all financing. How do investors give these companies a valuation? They will look at their income forecast and give a ratio. But to be honest, I would compare it with OpenAI. For example, today's OpenAI is 500 billion. When other model companies come to invest, investors will mark the OpenAI worth more than 100%. So back

to the point, is the company's benchmark the most important in the market today or OpenAI? Yes, definitely. I think the company is the most important benchmark in

OpenAI? Yes, definitely. I think the company is the most important benchmark in the 1st and 2nd generation market. Speaking of OpenEye and Anthropic, many people asked me about Robinhood. This is a very important investment in the second-tier market. How did

about Robinhood. This is a very important investment in the second-tier market. How did

you understand this stock at that time? Can you tell us what you saw when you invested in Robinhood? Why did you make this judgment? Robinhood is a very interesting company. It is the number one stock with a record 500 in

interesting company. It is the number one stock with a record 500 in the year. It also lost a lot of shares this year. But if you really

the year. It also lost a lot of shares this year. But if you really understand this business, Robinhood's business model is basically very poor. It can be said that it is a bad hand. Because the stock exchange itself has a very strong cycle. It is not a good business. Many people have the same experience

cycle. It is not a good business. Many people have the same experience in the stock market. It's like a bull in a hole. and the bear is still lying still. His trading income is the exchange rate multiplied by the rate of commission. The exchange rate itself is directly related to the degree of market activity. But

commission. The exchange rate itself is directly related to the degree of market activity. But

there are a few points that he can control and do. The first

point is also the most easy to think of, which is diversification. It's not just about trading income. So you see Robinhood is very active in making banks, making predictions markets. 去做财富管理,去做国际市场。 他现在已经有超过11条线的业务,

markets. 去做财富管理,去做国际市场。 他现在已经有超过11条线的业务,

有超过一个亿的收入 而且这个还会继续增加。

那第二他能做的就是去抢占市场份额 就是市场的强周期性是你没有办法控制的 但是你可以通过去抢份额, 去平滑这个周期性。 Third, pricing.

但是你可以通过去抢份额, 去平滑这个周期性。 Third, pricing.

For example, Robinhood's Jiaming has grown from 10 points of commission in three years to 60 points, which means it controls its own fate. Of course, the cost is also very important, because investors care most about profit growth. So if your income fluctuates, but your cost can cooperate with the fluctuation of your income, you can still

endure it. The opposite is the META of 2022, which is the price of the

endure it. The opposite is the META of 2022, which is the price of the cash flow cut in half because the income fell and the cost rose.

So it's not the logic of buying a brokerage from a bull market.

很多人这么说,我觉得投资挺重要的就是要对自己诚实,知道自己赚的是什么钱,是Beta还是Alpha。

可能最简单的就是你把Coinbase的股价和比特币去做一个回归,你会发现从2022年开始,Coinbase相对于比特币是没有任何Alpha的。

So if you say you judge that there is a bull market with a encrypted cycle, then you might as well buy Bitcoin. This is the beta of the market we talked about. But if you do the same with Robinhood, you will find that whether it is the price of Bitcoin or the NASDAQ index, there is a very strong So it's a different path. How do you see

the company's growth? It's very obvious that the United States has a generation of accounts. Like the 60-year-old baby tide generation is using cash. 45-year-old age is using

accounts. Like the 60-year-old baby tide generation is using cash. 45-year-old age is using E-Trade. Robinhood users are average age is over 30. This is the account of the

E-Trade. Robinhood users are average age is over 30. This is the account of the new generation of middle-aged and young people in the United States. The company

model can be simply imagined as a loophole. As long as money can enter it, it can generate income by making a series of transactions. So the most important thing is to increase the average assets of each user. Today, on Robinhood, users have an average of 10,000 yuan of assets. Added financial or called IBKR is more than 150,000 yuan. Robinhood users are 34 years old today, which is very important

because 35 years old is a significant age gap in the US. People between

35 and 40 years old will have a three-fold increase in average wealth compared to the previous five years. This increase will reach its peak at the age of 55. So even if Robinhood doesn't do anything today, it will naturally increase assets. The beginning of the transaction account is only one of its steps. As these users enter the middle age, there will naturally be

financial needs. So you see, the income of over 60% of the family finance comes

financial needs. So you see, the income of over 60% of the family finance comes from wealth management. Then there will naturally be the need for storage. After the money is absorbed, the most important thing is to make the money roll up. Robinhood is

very good at this. It's a way to increase the activity of users. For

example, I have two credit cards with Robinhood and Coinbase. Obviously, Coinbase will have a lot of new activities and reward mechanisms, so I can keep going back to their app to see. And Coinbase, we may see it once a month. So in the big picture, Robinhood is the next new financial asset. You bought it this year, right? Did you buy it once or several

asset. You bought it this year, right? Did you buy it once or several times? Can you tell us about some of your specific operations? Yes, I bought

times? Can you tell us about some of your specific operations? Yes, I bought it many times at the beginning of the year. I was very impressed.

The stock price fell very sharply at that time, and the market was very poor.

But I think this business was underestimated. Coinbase had a market value of more than 100 billion, and Robinhood was worth 3 billion. Coinbase was falling in share price, and Robinhood was very good at execution, and it was stealing shares everywhere.

This story was not seen by the market. Everyone thought they were both almost beta. So at that time, I was under a lot of pressure, thinking that

almost beta. So at that time, I was under a lot of pressure, thinking that this year might be a bear market, but I still bought Robinhood. This is what everyone says, when others are afraid, you are greedy. This year, it has already increased nearly three times. How do you see this company in the future? Is it

not cheap anymore? Yes, it is not cheap anymore. But this is also the most likely to become a company called a war-like financial application. This company is quite interesting. It has always had a kind of The mission of equal rights

quite interesting. It has always had a kind of The mission of equal rights for ordinary people. Many ordinary people get their first IPO share through Robinhood. I

also look forward to their VC fund. Another asset like VC has more than $100,000 billion in the United States. There are many good companies such as OpenAI, SpaceX, and other institutions that we just mentioned. But ordinary people can't get in touch with them.

So if Robinhood can create a VC fund that has all the shares of big companies that we can all hear about, and then charge me to take over ARK, which is a company that is a woodcutter, and charge me 2% or 3%, anyone can buy it, then this demand will be huge. I don't think it's weird to do it to the top of one billion. Are their main

business in the US? Yes, but not only that. This year, they entered the European market very cleverly. They used the most popular concept called tokenization. There are

dozens of small countries in Europe, and each has its own supervision. If

I go to take a photo one by one, it may take a year or two. But tokenization can be spread all over Europe at once. The transaction fee

two. But tokenization can be spread all over Europe at once. The transaction fee will be much cheaper. 你怎么看待Robinhood这个团队?

这个团队非常不一样,它动作非常的快, 就是你看它一年的产品的产出量, 可能能是别人五到十年那么多,完全不夸张。 The founder's personality is also interesting.

可能能是别人五到十年那么多,完全不夸张。 The founder's personality is also interesting.

For example, Coinbase's founder is a good boy. On the first day of his startup, he emphasized that he must return to the company. So I think he can't cut down on his operating costs today. He may be able to return to the company. Robinhood's founder is called Vlad. He is like a bad boy. He is very

company. Robinhood's founder is called Vlad. He is like a bad boy. He is very impulsive and dare to do what he wants. His personality also makes him very energetic when communicating with the public. The customers like him very much. I have more contact with him, and he will ask me sincerely which companies he needs to learn from. Especially those companies that are bigger and longer than him. He will

from. Especially those companies that are bigger and longer than him. He will

ask which ones are better than him today. This learning ability and enthusiasm is very rare.

然后他这个人也很神奇,他也同时非常的aggressive,很强势,只要是他想拿下的市场和业务,他会非常激进地去抢,所以我自己也和他就是学习了很多。

现在硅谷的资本是更喜欢乖小孩还是坏小孩?

硅谷 Robinhood's speed is supported by how? This is not reported by the media. But

in fact, Robinhood almost redid the company in 2022. It

completely changed its organizational structure. It was very centralized before. Then it was changed to a structure called Single

centralized before. Then it was changed to a structure called Single GM, which means that each business unit has its own independent structure, including independent PM, developer, CFO, HR. So it is very hard to make each business unit go crazy to innovate and produce products. This kind of single-jam organization may be more

conducive to the emergence of different products. Speaking of retail, we have heard more and more about retail in the past two years. During the epidemic, Wall Street Bets was used, and many companies called it retail tickets. Do you think this is a sustainable phenomenon? I think retail is already a big deal today. You see, over

phenomenon? I think retail is already a big deal today. You see, over 80% of all期权 are operated by retail. Some fund managers hate this phenomenon. But

I think retail is a very important part of the market. In my own investment framework, I think the market is a very chaotic system. Everyone is doing what they think they can make money. But in the end, the price performance is a reasonable result. So it's very important to know and understand how to operate these important

result. So it's very important to know and understand how to operate these important market groups. There are actually a lot of retail investors on Twitter. And

market groups. There are actually a lot of retail investors on Twitter. And

then you will eventually find a lot of stocks that can catch up with one or two of these KOLs. They are like a tribe. For example,

some people call them Robinhood guys, some people call them Palantir guys. I am very familiar with a lot of these KOLs. I will also go to understand what kind of specific stocks retail investors like, and what their trading system is like. I

think this itself is quite interesting. And I dare to say, for example, Robinhood and Palantir, the degree of research on retail is very high. I saw a pure love for investment in them. So one whole stock idea is to find a stock that the retail investors like, but the institutions can also like.

This kind of stock is usually more dynamic and doesn't have so many zero-hug games. Speaking of this zero-hug game, in the past few years, the US

games. Speaking of this zero-hug game, in the past few years, the US stock has been software stocks. There are almost no retail investors in it, all hedge funds, no long-term money, just short-term games. So the stock price is full of pain at 30% up and down. You mentioned a few times the so-called new species, and good kids, bad kids. I saw on your website

that you are in charge of subversive technology partners. What does this subversion mean? How do you define it? This is a very broad range. When you grow

mean? How do you define it? This is a very broad range. When you grow up, in fact, technology investment is to find these companies that can continue to grow. And then these companies are usually subversive. For example, I just talked about a

grow. And then these companies are usually subversive. For example, I just talked about a lot of Robinhood, OpenAI, including Pinduoduo, TikTok, Even the company, Netflix, was considered unreasonable when it first came out. For example, Robinhood is a gold mine, and OpenWire is a money-burning company. Netflix completely changed the rental mode of

DVDs. They all have their own peculiarities, but they eventually tamed the market. What do

DVDs. They all have their own peculiarities, but they eventually tamed the market. What do

you think is the commonality of such a taming company? To sum up, there are a few things. First, huge time, market scale. For example, Robinhood is going to be a super app for the entire financial industry. Second, the

product has to be perfect. Whether you're Google, Netflix, or GPT, the best product doesn't need to be sold. It can maintain a very high growth rate in a very long time. Third, when we look back, we usually say, what's the difference between heaven

long time. Third, when we look back, we usually say, what's the difference between heaven and earth? You feel like you're destined to be at that time and appear

and earth? You feel like you're destined to be at that time and appear as a subversive. For example, with the expansion of smart phones, short videos, including TikTok, were born. With the end of the Mordor Law, the GPU promoted AI, and the blankness of the lower market created the "Pinduoduo" that was so popular back then. In fact, each of them has a sense of destiny.

今年看下来可能最有意思的就是prediction market,就是预测市场。

大家可能也听过叫Calci Polymarket。

我因为本来看Robinhood,所以最早发现这块量上的非常的猛。 Like the Kelshi prediction market, its latest trading volume has exceeded 6 billion. The quarter return ratio is over 100%. It's a very typical new wild growth. In this prediction market, you can

100%. It's a very typical new wild growth. In this prediction market, you can use one or two dollars to hold everything. You can hold the US election, who is the president. 你可以去押特斯拉的财报,是不是会超预期。

甚至还有人在那预测说Taylor Swift会不会怀孕。

是一个很百花齐放的一个状态。 它绝对不是一个小众市场。

你看Bloomberg或者Google Finance都已经介入了它的数据。 Back to what I said earlier, I think the emergence of this industry is full of heaven, earth, land and people. First

of all, today's American mainstream media has a very serious credit crisis. And because of the pandemic and tariffs in the past two years, the situation is very turbulent.

People want to know what the truth is. And it seems like there's always someone who knows something about this world. For example, Google's Gemini 3, which is when it can come out. On Polymarket, someone set it to November 18th at the earliest, which is exactly the day. And then the other thing is that the prediction market is legal in the United States. 而你到今天在加州德州很多地方

连体育博彩都是不允许的 那当然还有就是这届政府在监管上比较宽松 所以prediction market正好可以趁机把量做上去 所以可以说是预测市场拿了一手王渣的好牌 就看之后能做多大了 我们说了很多AI的应用啊 还没说到自动驾驶和机器人

今年对这两个赛道有没有什么新的想法 The biggest surprise this year is Waymo. Its execution

capacity and speed are far above that of other companies. It is now operating in five regions and will soon be able to reach more than ten cities.

I observed that the speed of its opening is also accelerating significantly. Sanfan has been used for almost five years, and Austin has been used for less than two years.

Guigu opened in half a year, and the cost is not high, which is about tens of millions of dollars to open a new city. It

has 2,500 cars on the road today, and the annual income is estimated to be about 80 million yuan. I think it will go up very, very fast in the coming year. Because of the collaboration with Korea Modern Car, there are already factory in Georgia in the US. You can see that this factory can produce 100,000 units. When this amount goes up, I think it will have a great

impact on the entire transportation industry. Today, the price of Waymo and Uber is similar, but when you produce, and then BOMCOS, which is the cost of materials, I think it will have a great impact on Uber. Today, Wemo has already made a profit in Sanfran. The biggest cost is the discount. If I assume that the total

price of his car is $170,000, according to the four-year discount, the other major costs, including his long-distance safety, transportation, and some insurance costs, I add them all up and calculate them, today is also profitable. I think Uber and Waymo are also partners.

Today, Waymo and everyone are partners, including Uber, rental companies, and fleet management companies. But I also think about what Waymo needs in essence. It needs people to clean, take care of, charge, and

essence. It needs people to clean, take care of, charge, and park. Uber is not a labor-intensive company. The best way to

park. Uber is not a labor-intensive company. The best way to operate this business is to hire a company or a professional team. Of course, Vivo needs to balance the demand for

team. Of course, Vivo needs to balance the demand for high-quality and low-cost cars. Uber is a good choice because it's a private company and can adjust the price to match the team. But

when Waymo is in need of low-end stocks, the team will be idle, which will affect the profits. How do you compare Tesla and Waymo? Which

one will be the long-term winner? One of them has hardware problems and the other one has software problems. Weibo is now a hardware problem, which is a business model problem. The hardware cost will definitely come down, but the question is how much it can come down. For example, if I take it apart, its car and sensor are both 70,000 to 80,000 yuan, and the chip in

the car is about 20,000 to 30,000 yuan. The cost is also needed to cut it down. And how to cut it should not be cut down to the price of 30,000 yuan. So the long-term fate of Weibo depends on whether Tesla can succeed. Tesla doesn't have any hardware issues. They can just

can succeed. Tesla doesn't have any hardware issues. They can just mass-produce whatever they want. But the problem is software. Iran's 10-year-old

bull said that they only need vision and no laser. But today, they haven't proven themselves on Robotaxi. Are there only Vimo and Tesla left in the US? Yes,

they are. They are getting bigger and bigger. The auto driving market is actually very large. Americans drive about 30 million miles a year. If you

assume that one mile is one dollar, then this is a market of 300 billion a year. One dollar is actually very attractive to users because it corresponds to the price of Uber, which is about three dollars a mile.

So as long as it's safe, the auto driving market does not have any demand problems. I've been told by GPT that if I want to get 10% of the market share in the top 10 cities in the US, how many Robotaxis do I need? It's estimated that it will cost about $10,000 to $20,000. It's actually a

need? It's estimated that it will cost about $10,000 to $20,000. It's actually a small number. Because today, there are probably more than 3 million Uber cars in

small number. Because today, there are probably more than 3 million Uber cars in the U.S. After the big-scale implementation of autonomous driving, I think many industries will

the U.S. After the big-scale implementation of autonomous driving, I think many industries will change. For example, transportation, even real estate, because people will live more scattered. So

change. For example, transportation, even real estate, because people will live more scattered. So

after 10 years of autonomous driving, is it finally going to happen? In

the U.S., will these two be the last to occupy the entire market? Today, autonomous

driving is still a learning model, probably a small model with a billion light-years.

Because you have to operate the model on the side, and latency is such a consideration. For example, this model sees the stop sign. It doesn't really understand that

consideration. For example, this model sees the stop sign. It doesn't really understand that stop means to stop. It simply imitates human driving. Humans usually stop in front of such a sign, so it stops too. But it's not a real way to understand the world. There's a bit of a dead-end, a lot of overfit. So you can see this kind of video on the Internet. If

of overfit. So you can see this kind of video on the Internet. If

you put the stop sign on a person,

-

图像、视频、编程。

那通用模型就是会比锤类的小模型效果更好。 So I'm curious and want to see how big companies do the automation and whether latency can be solved. What about the robot? How is the development in the US in the past few years? It's a

robot? How is the development in the US in the past few years? It's a

very popular track. I think a lot of VCs are closing their eyes. I

think the local motion industry is basically solved. The robot can run, can jump, and can do a lot of basic movements. But the manipulation, which is the upper body operation, is still a lot worse. I've been worried about the future of the internet for a few years. I think there's still no consensus on the path. For

example, on hardware, do we need humanoid or 4-bit? If it's humanoid, what does it look like? On data, some people only use real-world data, some people only use

look like? On data, some people only use real-world data, some people only use simulation, and some people use it mixed up. So there is no consensus. But

the internet has been doing well, but it hasn't been without progress. I think there have been some changes in the past few years.

比如说共识越来越收敛到觉得应该硬件软件需要一体化去做训练,而不是只做一个模型层。

也有更多的人同意就是住宅会是比工业更合适的一个落地场景。

- And the robot is really lacking in data. And there will be a lot of noise in it. This is like the auto driving 10 years ago. When

a model is too short of data, you give it any data, it will feel that this is very helpful. But this can't be pushed out of line, because this process will be slower than everyone thinks. And as an investor, I think it's hard to evaluate. Because the robot is not like a big language model.

At least there is a list. I can see the size of the model.

然后因为机器人今天的任务完成度都不高,所以我也没有办法去定性的去判断,所以很多时候除了说这个赛道真的很大以外,我觉得很难下手去下重注。

你觉得机器人下一步会怎么发展,或者叫所谓的机器人的GPT moment会是什么样的,大概会在什么时候出现?

Now I think the whole market is very scattered and very messy.

Yes, it is very messy. I think it's more optimistic to have a "GPT Moment" called "Ghost" in two or three years. And this definition means that in manipulation, the robot can do some common actions on the upper body. Including folding clothes, packing things, putting the dishwasher, these simplest actions are done first. And

it may not be very complete, but it will start from here. It's important

and interesting that some American robot companies sell their products to their clients, including ONE X. I think the market needs to solve the problem of auto driving first before solving the problem of robots. The hardware and data of robots are far from being in place. We all know what a car is like. There is no definite form of a robot. The data

is far from being the same. What do you think of the comparison between China and the United States? I have seen and experienced almost all the auto and robot companies in China and the United States. I think the development of autonomous driving in China will be very good. And there will be more players in it than in the United States. Because we have data. But many people don't

know that the data of these cars in the United States cannot be collected.

Because a lot of data is not connected. Even if it is connected, it only has one or two cameras in the car. So the data collected is useless.

China is very different. We have experienced the EV era. Everyone's technology has been updated.

There are also many cameras in the car. So there is no lack of data.

So I look forward to China's development in this area. The robot is China's ability to see the world. We just talked about some star companies in Guigu. I'm also

very curious. At the beginning, you mentioned that you are rarely a Chinese in the US. What do you think of Chinese researchers and investors in

the US. What do you think of Chinese researchers and investors in the US? I think they are braver.

the US? I think they are braver.

I think there's a certain logic

to it.

The best investment in level 2 is non-common sense. VC is common sense investment.

In the second stage, you are looking for high risk high return. For example, in 2022, Meta and this year's Google are all very profitable. The example of the reversal. VC is actually looking for consensus. Because you have to burn a

reversal. VC is actually looking for consensus. Because you have to burn a lot of money every round. In fact, you need the market to recognize your point of view and always support your chosen company with you. This process may take many years. So there is a clear difference between consensus and non-consensus.

这两种投资关注的东西有明显的不同吗? 说到底看的都是人、产品、市场这三点

那早期VC会看人看的更多 那之后越来越多的会看市场 But from experience, a very good founder, even if he needs to change a few times, he rarely really messes up. For example, Elon Musk has done all kinds of things, but he has never lost an investment. So he came out today to finance with just one paper, which says I have financed hundreds of times, but I

have not let an investor lose money. This is his confidence. In terms of people, do each fund have different titles or tastes? The aesthetic is a bit different. I'll pick a few interesting ones. There's a fund called Altos Ventures, a big

different. I'll pick a few interesting ones. There's a fund called Altos Ventures, a big shareholder of Roblox. They like this kind of "finger-like" talent. Finger-licking has been their passion for a lifetime. So they are more of a fan of David, the kind of founder who does what he's most passionate about in his life. They don't like

people who are looking for hot spots and call themselves "continuous entrepreneurs." I also like another VC called Ruby Capital. They like this kind of rebel. They are entrepreneurs who are as rebellious as the world. This kind of people usually have a lot of exposure. For example, Robinhood is their typical favorite team. There is also a VC called Hummingbird. This company likes me as a Nezha-type founder. Founders usually have a

lot of psychological appreciation. They have a little bit of "I win my fate, I don't have a day" to prove to the world that the world is wrong. For example, the company they are in is called Kravkin.

is wrong. For example, the company they are in is called Kravkin.

There are a few game companies that are more typical examples. Founders are

the kind of people who will never sell their company to get a line.

But overall, I think the difference between the funds is not too big. The

people who are good can see it clearly. In the past years, have the US funds made any obvious changes or directions in the investment strategy? One

is to focus on the stock market. There is also a basic idea of investment. You may have heard that many people say that investment is an egg that

investment. You may have heard that many people say that investment is an egg that cannot be put in a basket and that it is risky to scatter. But

the risk will not become small because you scatter. The market will fall together.

So buying dozens of warehouses is like buying yourself a big plate. I really

appreciate an investor in the US, Dr. Mark Miller. He said that the best investment is to put all the eggs in one basket. This is actually very counterintuitive.

But if you dare to put them all in one basket, you must have done a lot of research before you dare to make this judgment. So

the risk is much lower than you don't understand and you don't know how to do it. Because you only have one basket, you will spend all your energy to look at this basket and this egg to grow together.

Of course, if you want to generate a surplus profit, if you just want to get a better percentage of profit than the bank, then I think it's very good to buy S&P directly. The surplus profit can almost only be achieved by heavy warehouses and heavy warehouses. And if you do it for a long time, you will find that there are really not many good

companies in the world. I can study dozens of the best companies and do the transaction well, that's enough. I don't think it's necessary to look at everything. Why

do you think this change will happen? It's still a long time for everyone to see the annual return rate. Put the eggs in a basket, is there any successful cases in recent years? For example, Ribbit Capital. In 2021, Robinhood should have made some user-friendly

Capital. In 2021, Robinhood should have made some user-friendly moves. Then, Ribbit put in 5 billion within 24 hours. This should be more

moves. Then, Ribbit put in 5 billion within 24 hours. This should be more than 7 times. Then there is Carvana. There is a fund that went to the bottom when it was 10 yuan.

So everyone is looking for opportunities

like this.

Yes, but many funds have a large scale and structure that makes it impossible to do. Because when the fund is large, the organizational structure is impossible to be flexible.

do. Because when the fund is large, the organizational structure is impossible to be flexible.

Every partner has to do a project every year, but there are not so many good projects. So here is another counter-sensitivity, that is, the return rate of

good projects. So here is another counter-sensitivity, that is, the return rate of VC is usually not particularly high. We have tracked more than 2,000 VC return data.

Only 200 of them can achieve 3 times the amount of the return. 80

of them can achieve more than 5 times the amount of the return. Of

these 80, the largest fund is worth $500 million. And you have to think of 5x net return because VC is a 10-year investment period, which is about 16% to 17% IRR. If you have 3x return, then you have 10% return rate. So to give a specific example, if you are a

200 billion dollars of funds. If you want to give LP a five-fold net return, then you need to return more than $100 billion to LP. Suppose you

have 10% stake when you withdraw. I think this is already very generous, maybe less than 10%. That means you need a total withdrawal of $10,000 billion. This

is equivalent to the total value of the entire US IPO in the past five or six years. This is impossible to do. This return rate is much lower than we thought. It sounds like it needs to be smaller and more beautiful.

Yes, this summary is completely correct. I think in addition to consensus and non-consensus, the other difference between VC and second-tier is the volume. VC funds want to do well, but there is no way to discuss it. We just gave the example of the 200 billion US dollar-sized fund. But the volume of second-tier is infinite. So many times, when you see a track or an opportunity, I

infinite. So many times, when you see a track or an opportunity, I think it's much easier to express it in the second-tier market. And the

money won't be locked for 10 years. Like buying Inmeda in 2023, at that time, it's really hard to judge which one of the dozens of big-singer companies can succeed. But if you put your weight on Inmeda, you can achieve very high profits. You count all the stock market dilution we just talked about, Inmeda's

high profits. You count all the stock market dilution we just talked about, Inmeda's return is the highest among them. Has the first and second-tier companies in the United States all become a new trend? I think there are a few companies that have both funds in the first and second level. But it's really not that close to the same investment team. Yes, it's usually two teams. Yes, it's very

separate, almost like two funds. So you're a team that covers both? Because

the boss controls very little of our staff, we have less than 30 people in the entire fund. And the investment team is very few, so he hopes that everyone has a two-sided view. 这个会带来一些投资难点吗?

如果你同时投资一二季。 我觉得时间分配上肯定自己要做决定,

而且说实话也比较辛苦 但是确实两边会有很多我们叫synergies 就是因为你两边是都在看 所以会有像我们刚才讲的不管是英伟达呀 还是一些具体的投资案例的这么一个帮助。 我们能不能把硅谷火的赛道给听众过一下,

还是一些具体的投资案例的这么一个帮助。 我们能不能把硅谷火的赛道给听众过一下, 你是怎么看的?哪些是你比较看好的? 今年最火的肯定是边城啊,

你是怎么看的?哪些是你比较看好的? 今年最火的肯定是边城啊,

We have a public secret in investment. If a sector can create a annual income of 10 billion in a year, no matter whether you understand it or not, you must invest in this sector. Because this is the market telling you that there will be world-class companies in this sector. Such a sector is actually quite difficult to encounter. But look at the one in Biancheng. It has created

more than $5 billion in annual income in a year. This is an important point. In Biancheng, you can see that the annual income is more than $6

point. In Biancheng, you can see that the annual income is more than $6 billion. There are two in this one, Cloud Code and OpenEd Codex.

billion. There are two in this one, Cloud Code and OpenEd Codex.

都是在他们发布之后直线拉升,后来居上 邪律非常陡峭在抢占市场。

我自己非常看好这个市场 但是我觉得里面的竞争才刚刚开始。

你刚刚看这个Google和Windsurf 它把编程就算是免费了 那这个对Cursor意味着什么? 所以竞争格局现在是比较难看清楚的。 In addition to

那这个对Cursor意味着什么? 所以竞争格局现在是比较难看清楚的。 In addition to programming, there are also many other top-notch industries this year, including finance, law, healthcare, and corporate software. This is also very interesting, because the development of software in the

corporate software. This is also very interesting, because the development of software in the last round started with a common, parallel software, and then went to the top-notch ones.

But now, entrepreneurship feels like only a chance for the "train" because the "common" layer has been eaten by the "big model" In these "train" areas, I think programming is also the most suitable for model companies to do it themselves. It is very different from other "train" because it doesn't need you to specifically gather knowledge of specific areas. It goes to market very naturally, and doesn't need to

specific areas. It goes to market very naturally, and doesn't need to hire a team to sell. So model companies are very smooth and reasonable.

This is very different from finance and law. We've talked a lot about the video industry before. What do you think? I'm very optimistic about the video industry. First

of all, it's huge. Because the media industry in the US is a market with more than $800 billion. I think video is an important element of AI and human communication. It's the most widely used input form for the brain. And

the logic of the video industry has changed completely. Like what TikTok did before, it was to match. Let the creator produce a video and then find a way to match it to the user. The optimization is the function of matching itself.

But now, video is the first target that can be optimized. You can set target parameters and optimize it. For example, we can use the user participation as a target to optimize the video. Or optimize the ad click rate as a target. There are so many ways to do it. I think there will be

target. There are so many ways to do it. I think there will be millions of big companies in it. So Google, Meta, and Elon Musk all focus on this. You wrote a deep research report on "Agentic Commerce". Is this a hot

on this. You wrote a deep research report on "Agentic Commerce". Is this a hot topic? Yes, "Agentic Commerce" is the next direction of OpenAI. You may see it this

topic? Yes, "Agentic Commerce" is the next direction of OpenAI. You may see it this month. I think this has a lot of impact on the entire Internet industry, including

month. I think this has a lot of impact on the entire Internet industry, including advertising, e-commerce, e-payment, Agenda e-commerce is where the agent buys things for me. For example, I want to buy a mask, and the agent will help me

me. For example, I want to buy a mask, and the agent will help me recommend it. Then, they will choose the one that suits me the most, and

recommend it. Then, they will choose the one that suits me the most, and then help me pay. If you really think about it, I think this is a nightmare for most of the merchants. Because merchants really want their users to come directly to their website, so that they can sell more of their products to me. We call it cross-sell. And traffic itself is very expensive. For example, Booking.com,

to me. We call it cross-sell. And traffic itself is very expensive. For example, Booking.com, China's name is Xie Cheng, half of its trading volume is from Google. Many

people don't know that it's actually zero profit in this part. It's completely unprofitable because Google's advertising fees are very expensive. Long-term agents will also charge fees. So if traffic is more and more by agents to keep, it's not what any merchant wants to see. Then you look at Amazon, its annual

advertising revenue is close to 100 billion. Advertising income is also called eye income. It's

all users coming directly to its website to buy things. He will have a chance to sell ads. There will be a big variation in this. So I

think about it, I think the only benefit is some long-term businessmen. Because they

couldn't afford advertising fees before, they couldn't get a lot of traffic. Now

Agent can be very precise. He may have gone through dozens of pages of search results, and found the one that suits me best. And these merchants usually open stores on Shopee. The payment industry will also change a lot. Today, many people use the Paypal express payment button. Why do I use it? Because it is my wallet. It stores my credit card information. I don't have to input it

my wallet. It stores my credit card information. I don't have to input it myself. And Paypal provides convenience, so it will receive a fee. But in the age

myself. And Paypal provides convenience, so it will receive a fee. But in the age of agents, agents are not afraid of trouble. GPT is my electronic wallet. It has

all my information. Is it necessary to paypal this middleman? I think it's a serious question. So agentic commerce is not just a company or an app. If it really

question. So agentic commerce is not just a company or an app. If it really works, it will change the whole Internet pattern. The software industry was very popular in the US. How do you see the difference between software companies and AI companies?

the US. How do you see the difference between software companies and AI companies?

Software companies are like chicken breast. They taste very similar. Basically, you

can close your eyes and deal with it. Because these companies are 80% of the same interest rate. The net flow rate is more than 120%. So

in the AI era, many investors don't want to touch AI investment because they think the interest rate is too low. But the profit margin of AI applications is a counter-sensitivity. The more people use it, the lower your profit margin. Now, we don't

a counter-sensitivity. The more people use it, the lower your profit margin. Now, we don't have a unified opinion on how to give AI companies a high valuation.

I think it depends on the absolute profit margin. Although the profit margin is low, the contract margin of AI is usually higher. The profit margin of each user can be several times that of a traditional software company. This is like The Amazon cloud was also 50-60% of the profit. Many people didn't like it because they thought that there was no software to make money. But this market

later proved to be bigger than all software. We talked about many celebrity companies and key channels, and also talked about the secret of the stock investment circle. Now a

topic that is very popular in the market is whether the market is in a bubble. These AI companies have huge revenue. Where do you think their money comes

bubble. These AI companies have huge revenue. Where do you think their money comes from? It's a concept of saving. These big companies that already exist, like

from? It's a concept of saving. These big companies that already exist, like Google and Meta, still have some new directions. If it's saving, where does such a huge amount of AI investment mean? That's a great question. First of all, there's a concept called electronic income or electronic tax. This is a total of about 4 trillion in the United States.

- people's subscription income, including Netflix and YouTube, is

about 5 billion. So if you go and grab the Internet's storage, the amount is not that big, it's just a total income of 4 billion. So if

OpenAI comes in, and reports to the media that it will make a revenue of 2 billion, if it takes it, then Google, Meta, Amazon, and so on will give it the share. So you're right, if we spent so much money and burned a small share, it wouldn't be that meaningful. Will the revenue from advertising

increase because of AI? Many people ask this question. I disagree with this statement.

I think the revenue from advertising in the United States has been very stable in the past 20 years. It's just that the annual revenue is 5% or 6%.

There is no reason for the total amount of advertising to increase. All the American companies add up and spend an average of 3% of their revenue on advertising.

There is no reason for AI to spend more than 3%. I think it's good if it doesn't decrease. If the revenue of the model is not from your defined electronic income, where can it come

from? The US GDP is a scale of 300 million. And the labor cost

from? The US GDP is a scale of 300 million. And the labor cost is as high as 150 million. So just the customer service is a market worth 300 billion. If you tear a hole in this, it's easy to have billions of

300 billion. If you tear a hole in this, it's easy to have billions of income. Michael Dow and Jensen's logic is that AI can generate more than 100% of

income. Michael Dow and Jensen's logic is that AI can generate more than 100% of global productivity, which can generate more than 100 million USD of GDP growth. So

if you invest 30 to 40 million USD every year, and the construction of the algorithm is reasonable. So my thought is that at the beginning, especially when the model capability is not yet too advanced, the easiest way to eat will still be to return to the electronic income. So OpenAI will definitely grab it, and the competitive pattern of advertising will definitely deteriorate. But if you give a little time,

the model capability comes up, the real agent, or called AI for Science, can emerge, then we can create new values, and we can eat into the 1.5 billion labor market. This account is over. But there is a problem here. If you greatly

labor market. This account is over. But there is a problem here. If you greatly increase the labor force and cause the company to be in debt, who will consume it? Will it have a negative effect on the overall economy? I think it

it? Will it have a negative effect on the overall economy? I think it depends on the situation. The simplest way is to give an example. If I

have 100 people today, I can produce 100 products. If I use an AI, the efficiency will be higher. 80 people can produce 110 products. These products may be cheaper and easier to use. Then the real GDP will increase by 10%. The rest

of the 80 people's income will definitely be greatly improved. The company's profits will also be greatly improved. So it's actually a positive consumer cycle. It's true that it's quite difficult to get rid of these 20 people. I don't know if AI companies will pay AI taxes or lay off employees to help those affected. For

example, an African member of Ro Khanna has started to talk about this. So it's

not a minor issue. But I'm optimistic. If you look back at the increase in productivity in history, it's actually always followed by the layoffs.

然后当然也是花一段时间之后 社会会去吸收这个变化 重新就业 然后财富有一个重新的分配。 很多人关注AI投入产出比, 这里的回报你是怎么算的?

然后财富有一个重新的分配。 很多人关注AI投入产出比, 这里的回报你是怎么算的?

其实不会算得特别清楚 因为AI既能产生收入 同时也能节省成本。

But we can go through a big number to see the future, for example, 12 to 18 months of relatively clear AI income. The big

head in this is OpenAI, which can see 3 billion, and then Anthropic sees half, which is about 1.5 billion. There are two hammerheads in the market that are close to 2 billion. One is programming, and the other is video generation.

There are also a few 500 million-dollar types of AI, including audio, customer service, law, healthcare, and some listed companies. So in total, we can see about 7 billion in AI income. There are many numbers in this, but I think the most important thing is that the AI income that can be tracked is actually only OpenAI and Anthropic. The rest of the startup companies add up to the maximum of

hundreds of billions. Do you think this market is a bubble? Is there a bubble?

I think the bubble is too subjective. It may be empty. But there are two levels to think about this issue. First, are we in a bubble today? I don't think so. Because the AI application is actually very fast. GPT has

today? I don't think so. Because the AI application is actually very fast. GPT has

been in use for less than three years. The user's popularity is equivalent to the results of 10 years of Internet development. We have already had millions of AI income. - - - - - - - -

AI income. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The second level is to see if the income of AI has increased. It's

back to the big numbers we just went through, which is the investment and output ratio of AI. And the biggest income in this is on OpenAI and Anthropic. We have done a lot of re-evaluation discussions here. Because today we are also at the end of the year,

can you make some progress for the next year? Because many listeners are very interested in investment, such as the most important companies in US stocks. Of

course, we also do not make any investment suggestions. The market has changed significantly in the past two weeks. The company's requirement is that it can't only invest in AI, but also have real-time income from AI. Let's talk about the biggest companies in the US. Google is the top of the list today. Its model is

the US. Google is the top of the list today. Its model is SOTA, and its position is the best. It is also the most logical in recent years. Many people have said that. But next year, I will still be

recent years. Many people have said that. But next year, I will still be very concerned about the competition in advertising. Because one is the increase in open-air advertising, and the other is TikTok, which is almost forgotten by everyone. It has the same number of users in the United States and Instagram, and is often much higher than Instagram. But its income is only less than one-third of Instagram's. So in the coming

Instagram. But its income is only less than one-third of Instagram's. So in the coming year, I think the overall competition in advertising will increase. It's also a good thing that it will force model companies, including Google, to look for a new AI business model. Then maybe we can talk about Meta. Meta and

Tesla's two-vote setup is very similar. To be honest, the basics are not very good, but the most important thing to affect the stock price is their AI process. Meta will see when its model can reach the first tier. But overall, it

process. Meta will see when its model can reach the first tier. But overall, it feels like next year is not easy because its cash flow will definitely drop. Meta has a rule that says that since its 15th year of listing, its

drop. Meta has a rule that says that since its 15th year of listing, its stock price has been rising every year, but it only fell for two years, and the cash flow for those two years has been rising. So next year really depends on the process of AI. Then Tesla has only one thing important, whether or not to take the security guard out of the car completely. Everything else

is not important. If you add another one to Tesla, what kind of form will it use for XCI's investment, and how much stock market share will it cause. Then go through the cloud manufacturers in the United States, the increase in the next year will be very good. I think Google's GCP's season growth may see an increase of more than 50%. But overall, I don't really

like the changes in the competition framework of the cloud. We have three fish today.

These three companies can make up to 35% or 40% of the profit. So it's

a lot. But now the competitive situation has completely changed. As a new player in Oracle, when he takes a project, he only requires 10% of the profit.

We have many new clouds, like CoreWeave, and several model companies are starting to do cloud-based business. So, from 3 clouds to 10 clouds, although the market has grown, it will certainly affect the profit. Cloud-based business has two basic problems. One is that the GPU rent is too high, so each

company has to make their own self-made chips. Second, the customer's concentration is too high. For example, Azure's increase of 70% is from OpenAI.

The CEO of Microsoft admitted that his profit will not come from renting GPUs, but from some additional projects, including database storage and security. For the most popular games like Nvidia or chips, I will focus on

security. For the most popular games like Nvidia or chips, I will focus on whether the self-propelled chip can deliver. Especially now that Google's TPU is going to be sold to users outside. In a world of power shortage, the price difference between GPU and self-propelled chips does not seem that important. I will also think

about the model effect of Blackwell training in the second quarter of next year.

Apple's ticket is actually very confusing. It doesn't have AI, and the stock price is very good. The company's culture is also quite different. Other companies are constantly

very good. The company's culture is also quite different. Other companies are constantly trying to make mistakes, and only Apple is waiting for its product to be absolutely perfect before it can be released. So it is very user-oriented and product-oriented.

Next year, we should see Tim Cook leave and a new CEO come on stage. So recently, I can see Tim Cook buying coffee and eating at

on stage. So recently, I can see Tim Cook buying coffee and eating at Zygoo. I feel very relaxed.

Zygoo. I feel very relaxed.

除了你说这几个大票,明年整体投资机会你是怎么判断的? I think the big picture is the split of the economy next year. It will be a typical "reversed fund market" where there is an opportunity to do more and do less. Simply

put, it will be more AI benefits and less AI losses. The losses will include IT outsourcing, content production, and some new industries. In

addition to semiconductors, energy, and clouds, I think the surplus benefits will come from AI applications. and it seems that many of the traditional AI applications are in the industry. There is also a very interesting thing, that is, this year's third quarter is the first time that many companies with a standard of 500 give

specific AI-based data, and say that this data can be very clearly displayed on finance. This is a very fast speed, because in the previous round of the cloud,

finance. This is a very fast speed, because in the previous round of the cloud, companies with a standard of 500 probably spent 五年的时间才开始慢慢的做匀的这个迁移 那这个里面也可以举几个比较有意思的例子 比如说大家都熟悉的沃尔玛 它用了一年的时间

用AI去做供应链的需求预测 它的缺货率就降低了30% 连每年它卡车的行驶里程数都减少了3000万英里 所以这里可以直接看到利润的改善 AI has significantly improved the loan rate and audit rate in the financial industry. The processing time has been shortened by more than 80%, and the breach

industry. The processing time has been shortened by more than 80%, and the breach rate has been reduced by half. There are banks that have already announced that their net profit can be increased by 15% with AI. Then there is CH Robinson, the world's largest logistics company. It has half of its budget today, which is all

AI-responsible. In the past two years, the production rate has increased by 35%.

AI-responsible. In the past two years, the production rate has increased by 35%.

So to dig into the heart of the traditional industry and find a target for AI to benefit directly, I think it will be a very interesting direction next year. Is there any commoner who is most worth paying attention to today? There is

year. Is there any commoner who is most worth paying attention to today? There is

a concept that today's US stock market is the AI market. In the first half of this year, 90% of the US GDP increased from AI investment, so the whole market is held by AI. Today, the trend is no longer about the United States. The market is more concerned about the income of AI, not the

investment. Today, the biggest and most sustainable income is the

investment. Today, the biggest and most sustainable income is the income of OpenAI. So the whole market is focused on this company.

We can focus on the increase in users, the growth of income, and whether the U.S. government will participate in its infrastructure projects in some form.

U.S. government will participate in its infrastructure projects in some form.

These are the most important. As a question, should we worry about the U.S.

red line next year? At present, the U.S. consumption is still very healthy overall. and the financial system has also been down a lot. There will be some

overall. and the financial system has also been down a lot. There will be some economic stimulus in the first quarter of next year, including tax cuts and tax exemptions. So in general, the bottom-level people in the United States may be better

exemptions. So in general, the bottom-level people in the United States may be better off next year than this year, and their spending capabilities will be better.

But the most important thing here is that next year is the middle election year of the United States. The KPI of this government is very concerned about the performance of the stock market, so it's enough to know this.

今天的节目就是这样。

这里是商业访谈录,是一档由语言及世界工作室出品的深度访谈节目。

你可以到公众号关注我们的工作室获取更多的信息。

我们的公众号是语言及世界,languageisworld。

我们希望和你一起从这里探索新的世界。