市场概述2026年2月10日

-

Then there is CPI. The most important thing is the southern payroll on Friday. This is the data from last week. I think the data from last week

Friday. This is the data from last week. I think the data from last week gave a clear indication of this week's data. The data from last week was not good. ADP is the data for the private sector.

-

It's not easy to find a job. But the company hasn't started to get a large-scale resource. In fact, this is also related to the progress of AI now. It's

large-scale resource. In fact, this is also related to the progress of AI now. It's

not that people will hire a lot of people. Even if my business is expanding, then I'm trying to say that when using AI, I probably didn't cut them off. It's like Kaushen. Kaushen recently used Anthopia to do

off. It's like Kaushen. Kaushen recently used Anthopia to do account opening. I think I said it last year. I expected it to

account opening. I think I said it last year. I expected it to be open AI. But the first one used was Kaushen using Anthopia. It's

like using them to do customer opening and KYC, which is the background work. The background salary for Kaushen is about 200,000

work. The background salary for Kaushen is about 200,000 dollars. The technical company is 200,000 per year. The VP is

dollars. The technical company is 200,000 per year. The VP is 300,000 per year. These people, including the internal control, won't be fired because the company still needs them. But when the business increases, they won't increase.

This is a big trend in employment now. It may be like this in the future. That's miserable. It's going to be a slump. So I've always felt

future. That's miserable. It's going to be a slump. So I've always felt that the risk of a slump in the United States is greater than inflation.

通胀其实没有那么大的压力 主要去年的话说有通胀 是因为特朗普加了关税 然后如果把这个影响在今年去掉的话 那通胀上压力没有很大 但是同时的话 上周的数据你看ISM This shows that the business index is not

bad. In other words, it is still the "playbook" that we talked about earlier this

bad. In other words, it is still the "playbook" that we talked about earlier this year. The state of the economy is not bad. The state of the

year. The state of the economy is not bad. The state of the economy is not bad. The basic economy is not bad. The growth of the economy is not bad. 然后通胀底,这个是一个我觉得是接下去一段时间的playbook

那这个对于下周的数据我觉得有可能能放配入会很差的 就是周五大家小心能放,能放的数据不会特别好 通胀数据不会很高,然后能放数据变差 有可能市场开始price in

Bad data is good data. Last week, the Washington people thought that QT was bad. Now, because this has not completely passed the mark, it is

bad. Now, because this has not completely passed the mark, it is possible that they will press in that bad data is good data. But it

will also quickly turn into bad data is bad data. This probability is very high. But these are all the trends of the short term. I think that

very high. But these are all the trends of the short term. I think that the US or the US economy, or the US economy, There are many people who talk about stiflation, because history will always be in our minds.

There are many things that can't be destroyed. The mind is always like this. I

don't think it's a stifling. I think it's entering a new state. If

stifling is called stiflation, we call it deflationary boom. What is

it called? Deflationary boom is called deflationary boom. I think deflationary boom is China is

deflationary.

I'll talk about the market and then talk about the model. Last

week, the Japanese parliament election was the first to get a pressure. There are several implementations for high-tech companies to achieve their pressure-saving advantage. First, it's easy for Japanese people to implement all their policies. Their pressure-saving advantage is now more than

what we expected. They got 2/3 of the votes. They got 2/3 of the votes in the next month. They got 316 votes. First, Kataichi has achieved pressure-saving advantage in LDP. Kataichi has achieved pressure-saving voice power within LDP. So when he is implementing any policy, he can just go and rest. He can rest. When he

is done, he can rest. Then, most of the time, I think, what kind of policy will he adopt? I think Gao Shan and his colleagues also think so.

When he is implementing a policy, he must be bold. If he is bold enough to get the advantage, he won't adopt conservative methods. Most of the time, it's probably on the budget.

risk

We actually talked about this last week. When will the character be affected? The year rate vol is rising. This is the decade. But in the

affected? The year rate vol is rising. This is the decade. But in the short term, Under a non-responsible financial policy, there won't be a particularly responsible currency policy. The market is starting to set prices.

Although Japan has inflation, your short-term monetary policy won't be particularly efficient in controlling inflation. So the 10-year interest rate has gone up. So inflationary expectations

controlling inflation. So the 10-year interest rate has gone up. So inflationary expectations may also go up. So Japan's stock price rose by 5% yesterday. So

the carry trade will continue. So the key to the coverage trade is the dollar/yen rate. The dollar/yen rate has been rising in the past

dollar/yen rate. The dollar/yen rate has been rising in the past week because of the LDP rise. But there were some NYDs yesterday.

I don't expect that because we want to be stable before the election, so I saw the rebound of US stocks last Friday, including the rebound of Monday, it is very related to the carry trade. So this is roughly the global situation.

-

-

Shot

- Long shot.

I bought a lot of stocks similar to Google. In the middle, for example, the president of this government is very aggressive, very romantic. He has his own system. There

are quite a lot of companies with Google systems. These companies buy these companies a basket Then the colleagues control openai Because last year's Gemini model is better than openai So I think Gemini Google model is going to rule Unite everyone Then openai was kicked out So the control will be openai And openai may be bankrupt They think openai is not good enough So the control is

more fierce in it is nivillion Then the control of nivillion is completely unreasonable Because anyone wants to use its algorithm They will say why And then, for example, Oracle. Oracle is also not reasonable. Oracle just said that the agreement it

Oracle. Oracle is also not reasonable. Oracle just said that the agreement it signed with OpenAI is not that its card can't be sold without OpenAI. This is

actually a pressure from the computing industry. And then, Microsoft, and then, for example, Covive, right? They are doing these things. And then,

another obvious investment is to do more than three.

uh I actually used it to make an international chess game.

This is a cheat. I used an international chess game. But it took me a few minutes. So it really made

chess game. But it took me a few minutes. So it really made it easier to make a game. I'm not a big fan of making games. will replace a lot of the programming work. So people will think that

games. will replace a lot of the programming work. So people will think that my SaaS applications will be slowly replaced by them. So this is a market of billions. After the SaaS applications are replaced,

billions. After the SaaS applications are replaced,

foreign

It's not that powerful. But it's like this. First, you

train a model. I think it's a combination of resources. Now,

in the large language model, you can get a better model by investing more resources. This is what Gemini proved last year. The problem with the scaling law was that I didn't

year. The problem with the scaling law was that I didn't have a scaling law. So I didn't calculate the power of the model. But I

forgot who it was. So I had a few technical problems. After I got back to Google, Google restored the scaling law. But Google has a lot of power. It has a lot of TPU. How did it get TPU?

Because it can get the Kovos line from the TSMC. It got the Kovos line from the TSMC Corecom and MTK are its younger brothers.

Including Samsung, the entire Android phone system is its own. So you can think of the last... Originally, it had a large client, one Apple and one Google. Google

the last... Originally, it had a large client, one Apple and one Google. Google

is its big client. Through buying so many products now, it got some cooperation. It

got a lot of Covers' production. So it has a lot of TPU.

and we used this to build a very good model called General. This is the general model. I think the quality of this general model is really good.

general model. I think the quality of this general model is really good.

And the second one, it uses the TPU and GPU's good difference.

That is to say, we haven't seen it yet. Yesterday, I reported the

uh

uh So the GPU model, the NVIDIA 7200 Link, hasn't been released yet. No model

for training is available. The GPU model for the GPU model hasn't been released yet.

So you can think that TPU's update period is about a year and a half. But it really hasn't been updated that fast. The GPU update is

half. But it really hasn't been updated that fast. The GPU update is about a year. So now, at this time, looking back, Google's model can get the upper hand in the use of the model. I think it will be difficult in the next year. It will launch new models, but it won't make

you feel that the same model is better than the previous model. I think

this is a big problem. The second is the resources invested by the model. Google has an advantage in the same model. The strategy of

model. Google has an advantage in the same model. The strategy of these three companies is different. The model is the number of computing resources you invested. Google got the computing resources. They also got a lot of computing resources, especially OpenAI, which got a lot of GPUs.

uh

OpenAI is GPT-5.3. Compared to Gemini, your common model may not be much different, or Gemini may be better than it. After Gemini

was released, it followed up with 5.3. But 5.3 was forced to be released later. It actually had some problems during 5.2 training.

later. It actually had some problems during 5.2 training.

Then, after 5.3, it improved. But because Gemini was faster, it didn't finish the Meta Training completely. It just sent 5.3.

But after the Meta Training, the 5.3 model, I sent 5.2. After the Meta Training, the 5.3 model, it was about finance, about some special, what we call the vertical field, OpenAI training is very good. I

will talk about the software and cloud awareness later. Its

model is now equivalent to the legal training for the vertical field. They are now training lawyers and finance. The background of

field. They are now training lawyers and finance. The background of training finance is open and controlled.

What does that mean?

is that the original AI is a more convenient to use, I can chat with it, I think it's very fun. Now AI has entered a state that can be used in practice and replace artificial.

This is different. What is the difference between inflation point and token? The difference is that each token is used by someone else. And whether the token can be used continuously. In these two companies, especially in OpenAI,

continuously. In these two companies, especially in OpenAI, In January, it was 1 billion. Yesterday, it sent a text to the three-digit company saying that we had returned to 10% in a month. But in fact, the growth of OpenAI in the 2B business is not only 10%. Last year, it

was 91 billion. This year, it increased by 10 billion in January. It's

basically doubled. This year, the total income of OpenAI is about 5 billion. And then his 500-billion income is what we call Infra. Infra is

5 billion. And then his 500-billion income is what we call Infra. Infra is

equivalent to me interacting with people. I sold the token to Gao Sheng and then to two insurance companies to let them do all their work, including sales. And then I can get 90 to 95% of the profits from these jobs.

sales. And then I can get 90 to 95% of the profits from these jobs.

This is Both of these companies are doing it. What is the difference between these two companies? OpenAI is doing all vertical fields. I know

that order book has reached 11 billion this time. If this time the financing is 15 billion, it can spend a lot of money when it is doing a few particularly profitable vertical fields. You think it is impossible for a lawyer to do it, because it doesn't have the knowledge of this industry.

He can spend $500 million on a team and give you all the content of all the lawyers. It's all about you. All the data is marked.

How difficult is this? He spent 35 million on the financial investment bank's data.

The data of the investment bank IBD is all marked for you. This is done.

Then you think the lawyer's case I spent 35 million is not enough. I spent

3.5 billion. So he actually has a huge amount of cash. And then when we think it's more difficult to do the training, he can invest a lot of it, including computing resources, plus human resources to eat this industry. That

is, in the future, ten people will have to kill 10% of the market business, for example, the lawyer industry. Look at the market business of the market. What

about Claudie? His resources will be a little less than his. Because the

business round just ended. It's 3.5 billion. The money he invested is a little less. There wasn't that much money before. This

doesn't depend on the money. What is the last resource? It's the

computing resources. So they invested 70% in mid-training.

The computing resources are probably the most open-ended. So

he focuses more on

- -

The difference between these two companies is that they will immediately enter the physical market. Why are you making software? Because their

performance growth is up to $100 billion by the end of this year. And the

margin of $100 billion is 90%. That is, people will think that software business has been changed. Software business was originally a pharmaceutical business. That is, I now make

been changed. Software business was originally a pharmaceutical business. That is, I now make a patent, and then I say I apply for a new drug, I finish a patent, and then I can copy this patent without cost and then sell it.

Then follow the software to the cloud after the cloud is lying in the shadow of me. Microsoft and Amazon are lying in the shadow because I always want

of me. Microsoft and Amazon are lying in the shadow because I always want to be in the cloud. You have to put all this stuff in the cloud. Then I sell software and then I sell software subscription and software

cloud. Then I sell software and then I sell software subscription and software service. Then after I finish it, I can sell it to you at a

service. Then after I finish it, I can sell it to you at a very low cost, almost no cost, and then I can also charge the fee.

The company can charge a lot of money. So the SaaS industry can give 30 times or even 40 times the EV. This is the entire software industry before. But it won't be in the future. The future is the factory.

before. But it won't be in the future. The future is the factory.

You can think of the software industry as the entire software industry, whether it is OpenAI, Claudia, or even Gemini. I don't think Gemini can do it. Or the new software companies will do this. They are all selling tokens. That is, every time I interact with him, he has costs. Then Claudia let me make this thing. Maybe

the token I spent might be 10 dollars. It is equivalent to producing tokens from the factory and then selling them. After selling them, it may go to the company to sell them. It replaces the human resources. That

is, how can I produce more tokens and sell them? In this demand, we see that it is absolutely not enough.

um

So Oracle, even if it can't do its own CDS, it has sold its stock

and sold other businesses, it will also buy this card. Because for them, this is playing a big game. I bought a card, I bought land. If you go to Taipei today, you won't buy the land. Then you will be done if you go up. Because this is limited. This card is extremely short. Now you

up. Because this is limited. This card is extremely short. Now you

will see that they sell this. sell tokens into a factory mode. If it

becomes a factory mode, the original software value will not be maintained. So I think everyone's work is very, very concentrated in many industries. Finally, let's talk about Glock. I relatively prefer Glock to this company's model. Many people prefer Meta for this

Glock. I relatively prefer Glock to this company's model. Many people prefer Meta for this reason. But I think the Meta team may have some problems. But I prefer Glock.

reason. But I think the Meta team may have some problems. But I prefer Glock.

The problem is that he can't get a lot of computing power. His card

is not enough. He has a very strong card-supporting ability. He did

something called "I quickly deployed the ability of the large model product." That is, I can immediately I don't know if I'm good at building models or good at matrining. But how do all models form a algorithmic group? It has the ability to build a algorithmic

group with the highest efficiency. So he did this kind of thing.

I first set up the algorithm. Then I said how to do the calculation. So

I use Glock. Glock may not need Microsoft. I don't need Microsoft's cloud. Then I

don't need Azure. I just go to the big sister myself. Then the cost is lowest and the efficiency is highest. which means I have the strongest organization capability, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my

hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I

have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most

troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my

hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I

have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the most troops in my hands, and I have the troops in my hands, and I have the troops in But the problem is, I think AI is now, like what Jason said, AI now turns software as a tool. I made it a tool, to software use tools. So

software use tools, and then AI turns something fun, something advanced, into something usable.

Then tokens from everyone not having to use it to me having to use it to replace the productivity. In this process, a lot of revenue will be generated.

Maybe it's just these two. I think these two will be running now. Then Glock

may launch a product. I think it's because its computing group is very capable. And

then if it has the ability to do product, Elon Musk is also very strong.

But I prefer to look at it. The only problem is that it can't get the card or the algorithm of the English answer. So this is the layout.

In this layout, the core version is the one who buys the land. The

land owner is the English answer. Why is the advantage of the English answer so great? In the end of 2024, because the card release was

great? In the end of 2024, because the card release was slow, they were going to make a group of 72 cards. They probably made a big machine. When they made the server, it was more like a red card. There

big machine. When they made the server, it was more like a red card. There

were a lot of problems with making it. There were a lot of problems with the night cold, the connection, and other problems. So all the tuition that was paid at that time became the threshold for today. I said

this at that time. Because every pit you step on today is the future threshold.

Because the city needs a lot of money to adjust this thing. And then

the first thing you do, including the red sea, including Dell, everyone will cooperate with you to overcome this thing. Everyone spends money. But when AMD becomes the second to do it, no one is willing to do it for you.

Because I have already made a deal with Nvidia. Why should I do it with AMD again? So its current card will not sell GPU cards. It is now selling the solution solution solution that only sells the big model solution solution. Then there are countless accessories in it. Then there are countless accessories.

solution. Then there are countless accessories in it. Then there are countless accessories.

Then why do you think HBM will be out of stock? HVM is out of stock because of the old yellow. Because I sell big machines, I am very sure that I can sell it. And I am out of stock now. If

you lack of cards, you need 3 million cards. This year, you need 4 million cards. This year, you need 4 million cards. Because it doesn't sell cards, it only

cards. This year, you need 4 million cards. Because it doesn't sell cards, it only sells solutions. You have to sell this thing together. So

sells solutions. You have to sell this thing together. So

for example, Helix, Samsung or MU, for them, at least I have three of them, who is the most reliable for my HPM? It must be the most reliable for NVIDIA. NVIDIA locks the output power, causing others to not even get the output power. This is the

market. It's like in 2008, you couldn't make a phone. You could make

market. It's like in 2008, you couldn't make a phone. You could make one in 2012, because there was no shortage at that time. In 2008, if you make a smartphone, all the good suppliers are Apple suppliers. That is, their production is primarily given to Apple. Today, on the semiconductor, no matter if

it's a light bulb, memory, including everything, They have the production, the first one is for Inglaterra, because they bought it. When the production is not so lacking, these products can only be spit out. Now it is impossible to spit it out.

You see the model war began this year, there are two more to join.

One is Meta, one is Microsoft. They must add this thing. Why? Because I

talked a few more words about the moon. I thought I might not have QAN, I will talk more about it today. Why do they want to join? Because

if they don't join, the original Tangying's voice mode has also been changed. The

voice was originally Tangying, because I can get a very high margin. But you

think as a company, I originally need to put all my internal data, for example, the Pallantier business will also be affected. Do you think Pallantier is Tangying? He

took the defense order, it may not be Tangying. Because Palantir, including Databricks, used to help companies clean up data, mark it, and then go do AI. Then I used voice to deploy my company and then do this.

do AI. Then I used voice to deploy my company and then do this.

But Cloudy and OpenAI are now directly outputting the results to you.

That is to say, who is purchasing the voice? They are purchasing. Can their

purchasing power be the same as the company's own purchasing power? That's

definitely different. So under this big red line, Every company, even the ones that are doing well in AI, Snowflake, Databricks, are really strong. You can't be sure if Palantir is the last winner. But it's possible. Palantir is because AI is in the Ministry of Defense. The order of the Ministry of Defense will definitely

increase in the index level because the proportion is too small. But in the end, you don't have this certainty that it will win. But when you look back, the whole industry is now unable to be replaced. Who is it? It's Nvidia. And

then there's TSMC. TSMC said yesterday that its Covost production is said to be a positive increase in the market share. Isn't it possible to produce more?

And then Invida will sell more. All the production is now very much in the back. You see a lot of shortcomings. It's not that the shortage will impact the impact of the U.S. dollar. It's not that the more the shortage is like, for example, when Apple just came out, the more the shortage of this thing is, the more difficult it is for competitors

to pick up or compete at this time. It's very, very likely that these four models will be the last question. The question of Gemini is that I discussed it with the team a few days ago. I think this is very reasonable.

It's a problem of its organizers. Google is not a very entrepreneurial company. They don't have these two companies that they want to

company. They don't have these two companies that they want to make money from. Of course, it was originally a lab. OpenAI is

a lab. Many people think it shouldn't be so profit-oriented, so they left.

But Gemini is indeed a very lab company. Its team is in London, DeepSick. DeepMind is in London. Gemini's team It's very

DeepSick. DeepMind is in London. Gemini's team It's very decisive how this model is going to go. He really wants to pursue a better, more comprehensive model. But in the vertical field, will he invest?

I think the deep mind of this group is very good at speaking. They

are not good at commercialization. Google itself is not good at it. They are

good at search engine commercialization. But they are not good at product commercialization. They are not interested in it. In fact, Google is not interested in

commercialization. They are not interested in it. In fact, Google is not interested in this. So if you are in this position, I think the risk will be quite

this. So if you are in this position, I think the risk will be quite high. There are several reasons. I think Codex is

high. There are several reasons. I think Codex is the first one. After Codex is released, there will be a common model, a common machine model, a common machine model with a B-card training. You

will see that there will be people who will talk about TPU in the large scale model for the scale of Inverna. And then the second one you will see that this year every month they are all reporting performance. It's the open air and the performance of the month on month will be very, very fast.

It means that the order will be fast and the third one is that OpenAI will close the 1,000 to 1,500 billion. I guess it's 1,500 billion. So now everyone says that Oracle can't fill the order of OpenAI because it may go bankrupt.

Because OpenAI will burn $2 billion by 2023. OK, it has 1,500 billion this year. I know that 1100 billion has already broken out of

year. I know that 1100 billion has already broken out of the Outer Bulk. So this will impact the price of this presentation. But I'm not sure about this presentation. After all, software has dropped too much. Because

now software can't give a price. You can't give 30 times EV or 15 times EV or a few times EV. If they can't give, they will call it long semi. You know, in the semi index, ETFs are ETFs. Which one

is the biggest? It's MU. It's the deposits. So you see why MU didn't rise yesterday. Because the software rebounded. The software rebounded by

yesterday. Because the software rebounded. The software rebounded by 3%. Then the semi index was moved down. This is two market positions. If

3%. Then the semi index was moved down. This is two market positions. If

you want to make a deal, you need to pay close attention to it. If the trend changes, the whole position will be N1. It's

to it. If the trend changes, the whole position will be N1. It's

similar to the effect of buying AMB and short Intel. So this is something you need to pay attention to. And finally, I want to talk about - SaaS will suffer the first wave of damage, the most severe damage, and the

most severe damage on the stock price. But it's not about eating SaaS. For

example, if I use CodeWork to create on Excel, I will not cancel Excel. I don't think that I will cancel SaaS so quickly. SaaS is very good.

Excel. I don't think that I will cancel SaaS so quickly. SaaS is very good.

But I think you basically have to look at who to replace the model. It can be replaced in three directions. The first one is people.

model. It can be replaced in three directions. The first one is people.

It takes away the work of people. The second one is probably replaced by software. It's SaaS service, right? The third one is replaced by a competitive company. The

software. It's SaaS service, right? The third one is replaced by a competitive company. The

fourth one may be improved efficiency. It's in these four directions. The competitive company says that I used it in a high degree, but Morgan didn't use it. Then Morgan's

business will give it to me. Then the improvement efficiency is the whole industry. I do more business and let GDP improve.

industry. I do more business and let GDP improve.

- It's not that fast. It actually increases the cost of human resources by more than the cost of the land. But you have to improve competitiveness.

In fact, when AI, the new technology, first introduced into the market, it had commercial effect. It had commercial effect. It was fun since last year. It didn't have commercial

effect. It had commercial effect. It was fun since last year. It didn't have commercial effect. It has commercial effect since this year. It has commercial effect. New

effect. It has commercial effect since this year. It has commercial effect. New

things must be killed first. Then new production power will be produced. I killed the old, and I killed the human. And the world has 7 billion, an average of $25,000 a year. He has to replace 30% of the middle. What he really did was the work of the human. So I will talk about employment data. I will talk about it. Back to what I just said about

data. I will talk about it. Back to what I just said about the basic view of the American economy. I think it's called the contraction of the economy. The contraction of the economy is the contraction of the economy.

the economy. The contraction of the economy is the contraction of the economy.

Will it be completely transparent? It's possible. It's a whole industry. In 1848, the productivity increased too much, and no one bought it. It's a whole industry. This is possible.

But it's definitely not now. Now it's transparent. So I think who is the real food? It's human work. So will the short software turn back? I think

this percentage will definitely turn back. But it doesn't have to be now. Because even

though it has fallen a lot, it doesn't dare to give a story now.

Then we will slowly realize that these two companies will soon be profitable companies. For

example, how much money does he burn to do training? And everyone will know that the factory is a process that needs to be constantly relayed and then constantly producing tokens, constantly relaying my factory, and then constantly producing tokens and producing tokens. It will make the original cloud and the original software

operation and management mode, the production and business relationship, completely change.

So this is what we will be observing later. I won't do the Q&A today because I have a long talk today. Thank you for your time. That's it.