中國的AI要把美國卷死了嗎？

Hello everyone, I'm Terry Today I'm going to talk about the most popular topic of the weekend The deep-seek AI, the new company in China, shocked the entire US stock market The US stock market fell significantly yesterday, and even the cryptocurrency couldn't escape Because the current currency trend is basically to follow the Nasdaq

Let's talk about what DeepSeq is and what it can do to make the entire US tech industry so responsive. And what impact does this event have on our investment? First of all, DeepSeq and OpenAI are similar companies. They are all tech companies that develop large language models, LOM. We use ChatGPT every day, which is made by OpenAI. The artificial intelligence model behind this program is just a interface.

The real back end to answer these questions are the models that have been trained through big data. In the United States, not only OpenAI is doing this, but also Meta's Lama and Google's Gemini. These small and medium-sized companies are training their own models. The strongest model in the competition of these large-scale models has always been OpenAI. In fact, it is still OpenAI's model called O1. After all, OpenAI is the founder of this large-scale model.

The first to launch the chat GPT and shocked the entire technology industry. Only then will these companies slowly rise up like Anthropic, Perplexity, XAI. But the top three models to take over the big circle model ranking have been developed and trained by American companies. Like GPT-4, O1, Gemini, Cloud 3.5, Sonic.

Until recently, a large-scale model developed by China's new innovation is not developed by big companies like Ali Tencent It is a small company called DeepSeek that developed a large-scale model called R1 Basically, its reasoning logic ability and various indicators have been able to match the current strongest model of OpenAI, R1 Not only that, DeepSeek also opened the whole R1 model It means that any company can download this R1 model now Then run on your own server and develop and optimize

It can also ensure the safety of data. It won't be a model developed by a Chinese company and then have data lag. Like me, I can actually download this R1 model and run it on my computer. It's just that my computer may not run, but anyone can do this. In comparison, OpenAI's R1 is a completely commercialized model. It should be said that it is the strongest model in the United States at present. If companies want to use this R1 model, they must use their API. Or you can also use their services in the Microsoft cloud.

Anyway, it's very expensive. Like our company, the current cost of these LLM APIs is tens of thousands of dollars a month. Of course, Deepseek also provides this API service. You can see its price is compared to OpenAI O1. The price of the input from the storage is 50 times the price of the O1 of OpenAI. And the output is 20 times, almost 30 times the price difference.

The Chinese price war is not a big news But this time it's the most powerful and proud of the United States The top of the pyramid OpenAI model If the performance of R1 is much worse today, it's not a big deal But today this R1 can basically match O1

and some performance is better than O1. I think it's impossible to shock the entire US stock market by simply playing the price war. This time it caused a collective collapse of the chip stock, plus many technology companies could not be relieved. The main reason is that DeepSeq said they spent less than $6 million, and then completed the development of the large-unit model in two months.

Compared to that, OpenAI's total loss in 2024 was about $6 billion. Their income was about $3.7 billion. In 2024, they spent about $80-90 billion on operating costs. Of course, the actual money spent on developing the O1 model may be just a small part. I don't know how much exactly, they didn't say. But I asked ChatGPT.

He said he estimated it might be about 70-80 million dollars Even if this number is good, it is also more than ten times the development of R1 And the investment of the three major technology giants of the United States, Google, Microsoft and Meta on AI In 2024, it is estimated to reach more than 150 billion dollars This is the reason why the chip stock has collapsed Because it made investors and Wall Street start to doubt whether these money invested in AI really needs so much

Or you can do it like China, spend a few million dollars to get a model done. If that's the case, these high-end chips like NVIDIA's Blackwell may not be necessary at all. Why can Deepsea build this model at a price of less than $6 million? The main reason is that they claim that they only use NVIDIA H800 chips with 2000 chips. What is H800 chip?

The H800 chip is designed for NVIDIA's export to China Because the US has sanctions on China's chip exports So high performance chips like NVIDIA's H100 cannot be exported directly to China So NVIDIA specially designed the H800 to meet the specifications of the chip export to China What is the difference between the H800 and the H100 in performance? You can imagine the H800 as a weak version of the H100, which is about half the performance

So they claimed that they only used 2000 pieces of H800 chips to complete the construction of the entire model. After this news came out, of course, it blew up the entire tech circle of the department. Everyone is discussing this thing all weekend. Deepseek's mobile app also rushed to the first download line of App Store. Investors began to doubt that these big companies spend tens of billions of dollars a year. The capital invested in AI for hundreds of billions of dollars, is it really necessary?

Or is it that low-end chips can achieve the training we need? It's just that we've been blindly throwing money on the GPU and not optimizing our algorithm If that's the case, then those chip developers like Nvidia AMD, downstream suppliers like TSMC, equipment factories like ASML, are they overestimating? Because later big companies may start to review their budget for investing in AI infrastructure If you can optimize the algorithm to achieve this training cost reduction

It is not necessary to buy high-end chips like Blackwell. So it caused the US stock to fall 17% in the supply of AI chips yesterday. AMD fell 6% in the supply of NVIDIA chips. Taichi Electronics fell 13% in the supply of ASML chips. We already know the whole story. What should we do next? First of all, I think this is a very important wake-up call.

Let's first ignore whether DeepSeq really only cost $6 million and completed the entire model development. Let's assume this is true. I think this is the first time a Chinese company has completely counter-examined the reality of technology products. Why would I say this? In the past, Chinese technology companies have been adding hardware and then fighting the price war. Let's take the simplest example, like Apple's mobile phone and other brands' mobile phones. Apple is the one who is strong in the ecosystem and their software.

One of the things that these US companies are doing is constantly optimizing software. So you don't have to use very high performance mobile hardware. In fact, Tesla's electric car is also a similar reason. They are constantly optimizing the motor and power management. So you don't need to plug in the battery without a brain to reach a very high battery life. This is the innovation that Chinese technology has been lacking compared to the United States.

This DeepSick incident proves that Chinese tech companies don't need to buy high performance GPUs without thinking. It can reach the same performance as the products developed by these tech companies in the US for billions of dollars. So back to the most important question, should I sell my NVIDIA stocks?

I don't know how to do it myself Of course you have to do whatever you want This is not an investment suggestion I'm still a little bit longer when he fell to 120 yesterday I think this will really affect the needs of future AI chips Big companies will also start reviewing their own budget for AI infrastructure But this will not affect the development of AI On the contrary, this is a bigger progress for the entire AI market

We are still very early in the development of AI. Real world AI, real world AI, autonomous driving, and robots have not yet begun. In the future, basically everything we do will involve AI. Just like now, basically everything can be done on the Internet or on mobile apps. I think NVIDIA, AMD, and TSMC will definitely encounter a time of chip attack. But not now, at least in the next year or two, I still can't see this situation.

I don't think an LOM paper will directly affect the needs of the entire GPU industry chain. Moreover, many people doubt that DeepSync may use more than 2,000 H800 chips. Like the founder of Scale AI, CEO Alex, said in the news that DeepSync actually has more than 50,000 NVIDIA H100 chips. And Elon also replied on X, of course it is so obvious.

I think it is possible because H100 cannot be legally sold to Chinese companies. And these chips are also likely to be purchased by some middlemen. These are all possible and not difficult to do. But no matter what the fact is, NVIDIA's high-end chips must have its needs. And DeepSeq also mentioned in their paper that they are looking forward to NVIDIA's next generation chip Blackwell to solve some of the problems they are currently facing in FP8 advanced training.

I think the fall yesterday was just a overreaction of the market. For long-term investors, it is a good opportunity to increase. Of course, let's not be blind to O-Ink. We still have to pay attention to the guidance they gave after the next month's financial report came out. And the next few weeks is the financial week of the fourth quarter of the technology company's financial report. You can also learn from the earning cost of these big companies. How much do these big companies and these technology companies plan to invest in AI? And then evaluate whether NVIDIA's chip has this demand.

I will update the result of this earning system in the video So if you don't want to miss it, you must subscribe to my channel I also added Tesla in addition to NVIDIA In fact, I don't understand what Tesla was doing yesterday I think this is a good thing for Tesla Because Tesla is also a NVIDIA customer They also have a chip for the development of the push So in theory, if you can achieve the result of training in a lower cost way

For Tesla, it's definitely a very powerful thing to reduce the cost of FSD training. This is the same as Meta, Google, Microsoft. I don't even know what they're talking about. I think this is another representation of the financial market. Wall Street's financial resources for AI. If the company can spend less money in the future to do the same thing, it will definitely save costs.

It should be a good thing for these tech companies. I guess the reason is that the market will slowly be occupied by these Chinese LLM. Like DeepSick, it is the top of the App Store in the day. But you have to think about what these companies like Google and Meta are making money on. They make money by selling ads, but LLM doesn't make money. What about Microsoft? What does Microsoft make money on? They make money by selling software and cloud services. LLM's business is very high now, so they don't make much money.

And later LLM will be cheaper and cheaper Because the cost of training is lower Then the competition will be more intense Then the threshold will be lower and lower Just like now There are 4G networks and 5G networks everywhere Wi-Fi is free, it's the same So I didn't hesitate to add Tesla yesterday In fact, I had a mistake yesterday

I had a $385 cash order at the time. It was early in the US winter. But I forgot to choose "Premarket Trading" at the time. What does this mean? On the IB, you can actually open an unconventional time to trade. If I had chosen this function at the time, I would have bought this $385 cash order.

Because it rose a little bit at the beginning So I put a deal on 395 again Finally bought it, but in this case Need to support non-standard real estate trading like IB I will put the IB link in the information section below the video, you can go and see Finally, if you have American SSN Don't miss the benefit of Moomoo's New Year Special for new users And this is only for Chinese users and Chinese viewers Now new users use my registered link to register Moomoo in the United States

Then the deposit is full of 6,888, you can have 8.1% cash interest Plus free 30 stocks for you Plus a stock of Nvidia for you This is the most generous time I've seen Mumu send new users benefits Of course, this is only limited to American users. New users in Canada and Australia also have corresponding benefits I will put the link in the information column below the video Don't miss Mumu's red envelope to new users Okay, that's it for today's video

I will post all the deals I made, whether it's buying stocks, or investing in currency, or selling something. All the deals will be posted on IG's live stream. If you are interested, you can follow my IG. Okay, I wish you all a happy Lunar New Year. See you next time.