Transcription
So, Wall Street is in panic mode once again. The NASDAQ futures, before the market opens, are down 3%. We've got semiconductor stocks like Nvidia and ASML down 7 to 9%. We've got big tech companies like Microsoft and Meta down 3, 4, 5%. Why? Because of news that China's new Deep Seek AI model has overtaken all the US models at a fraction of the cost.
So, what's the implication for us big tech, and what should we do about it? Let me break it down in this video.
[Music]
This news started three weeks ago in late December when a Chinese AI company called Deep Seek said, "Hey, we developed a new open-source large language model, or those AI chatbots, that is better than ChatGPT, better than Google B." And you know what? You guys spend billions of dollars to create your large language models. For example, Google spends about $31 billion a year in capex. OpenAI's ChatGPT spends about $5 billion a year. Anthropic's Claude spends about a billion a year. And you know what? We did everything better than you guys at $5.6 million, right? A fraction of the cost.
The most surprising thing was that, based on a third-party comparison, their Deep Seek model outperformed Meta's Llama 3.1, OpenAI's GPT-4, and Anthropic's Claude 3.5 in accuracy, ranging from complex problem-solving to math and coding, which you can see from this chart over here. Again, Deep Seek is outperforming all of the US large language models.
And the best part is that Deep Seek said, "You know what? We did it with the low-end Nvidia chips." As you guys know, the US government has banned Nvidia from selling their high-end chips to China. Their latest high-end chips are the H100s, so China can only get the lower-end chips that are half as powerful, called the H800s.
So, Deep Seek is saying, "You know what? We used the lousiest Nvidia chips at only $5.6 million, and in two months, we managed to outperform all your large language models that cost billions and billions of dollars with the latest Nvidia chips."
This has caused many US tech companies to freak out because it's like, how can you now justify spending billions on Nvidia chips and all the latest technology when the Chinese company is spending only $5 million and can overtake all of you?
It's kind of like the US tech companies, like Meta, OpenAI, and Microsoft, spending millions of dollars buying a McLaren car, whereas the Chinese driver spends $30,000 on a Toyota, and on the racetrack, the Toyota is able to go faster than the McLaren.
This is causing a lot of panic on Wall Street because there are a few implications.
Number one, if this is really true, that means China is providing open-source free large language models to the rest of the world. Why would people pay money to use OpenAI's ChatGPT? Why would people pay money to use Google's Advanced Gemini systems? They can get it free from China, right? So that's one concern.
The second concern would be, okay, if this is true, then all your hyperscalers like Amazon, Google, and Meta, spending billions of dollars on AMD and Nvidia chips, maybe they don't need to. Maybe they can spend 80% less and buy the lowest-end chips or buy fewer chips and get the same performance.
So that's causing people to freak out, and again, Nvidia is crashing, AMD is crashing, all these stocks are crashing.
So how concerned should we be?
The first thing, as an investor, is don't panic. Okay, look at the facts and think rationally. The first question to ask is, are these claims true? How likely is it that a Chinese company could again spend only $5.6 million and use the lower-end Nvidia chips and overtake the US tech companies that are spending billions of dollars?
By the way, Google spends $50 billion a year. I misspoke; I said $30 billion. Google is spending $50 billion a year in capex using the most advanced Nvidia chips.
So how likely is that Chinese company? In other words, how likely is a Toyota able to outperform a McLaren? That's the question. And honestly, I'm not sure, but I’ve got to be a bit skeptical and question how true this is.
So what's interesting is that there was this article that came out, and they interviewed another Chinese AI company CEO. This is what he said. His name is Alexander Wang, and he said this: "According to Wang, when it comes to Chinese accessing Nvidia's most advanced GPUs, are they really accessing it, yes or no?"
He said, "You know, the Chinese labs have more H100s than what people think." So he's kind of like letting the cat out of the bag, right? He added that his understanding is that Deep Seek has about 50,000 H100 Nvidia chips.
Now, why isn't Deep Seek admitting to this? For a very simple reason: if they did, they'll be in deep trouble. Remember that in 2022, the Biden Administration banned Nvidia from selling their superior chips to China.
So the H100s are not allowed to be sold to China. Nvidia had to specially develop a lower-end chip, the H800, which can be sold to China. So again, Deep Seek is claiming they're using the H800, which, by the way, are now banned as well.
So the US has now banned all these things, right? No H100, no H800, right? They banned it in October of 2023.
So what Wang is saying is that, you know, China and this company, they can't admit they've got H100s. If they did, Nvidia may be in trouble for maybe slyly selling it to them. Sometimes it's not Nvidia's fault; sometimes there are companies in other countries that set up shell companies in other countries that are not exposed to the export ban.
So they buy the chips from Nvidia and then resell them to China. So China is getting it through a loophole, right? And of course, they can't admit it. If they admit it, then that middleman company is going to get in big trouble.
So they're going to say, "No, we don't have the H100s; we have the H800s." Okay, so that's what Wang is saying. He said they can't talk about it, obviously, because it is against the export controls that the US has put in place.
He also thinks that they have more chips than what other people expect. If what this guy Wang says is true, then Deep Seek actually has 50,000. I'm not sure why he's so specific, but he claims that Deep Seek has 50,000 H100 GPUs.
Now, if that is true, then if you take the average cost of one H100 GPU and multiply it, basically Deep Seek actually spent between $1.25 billion to $1.75 billion just on Nvidia chips, which is not too far away from what Anthropic spent in the US.
Am I saying that Deep Seek is bullshitting everyone in order to create panic in the US markets? I don't know, right? It's a possibility; it's a theory, but I'm not 100% sure, obviously, because I don't know.
Okay, but let's give the Chinese the benefit of the doubt. Let's say it is true that they didn't use the H100s, that they're using the H800s. Could they have pulled it off? Well, it's possible as well.
So one possibility is that instead of developing their own proprietary AI models using the most advanced chips, Deep Seek basically relied on widely available open-source technology, which is available from OpenAI already, and iterating on existing technology, tweaking available data sets, and leveraging existing models.
So in other words, basically Deep Seek kind of copied the output of ChatGPT to create their LLM systems.
Okay, now what's interesting is that a couple of days ago, if you logged into Deep Seek and you typed this question, "What model are you?" Guess what answer they gave? This is from Deep Seek, by the way. They said, "I'm an AI language model called ChatGPT developed by OpenAI. Specifically, I'm based on the GPT-4 architecture."
Holy cow! So Deep Seek doesn't even believe that they are Deep Seek. Deep Seek has got an identity crisis. Deep Seek thinks that it's ChatGPT.
So it sounds like they're kind of copying wholesale ChatGPT by somehow replicating ChatGPT's output. I don't know how the hell that happens, right? I'm not a tech engineer; I'm not a techy guy.
Sounds like someone I know, right? You know, called CB, right? Except the difference is that Deep Seek is supposed to be superior to what they see, but that CB is totally inferior. Okay, that's the only difference.
So anyway, once again, let's give the benefit of the doubt. Let's assume that the Chinese company really uses the lower-end chips, they only spend $5 million, and they can replicate this and they can scale this.
Let's say, okay? And by the way, the other thing that's freaking out the market is now Deep Seek is number one on the Apple App Store, and they say, "Oh, everyone's going to Deep Seek; no one's going to use ChatGPT anymore. Deep Seek is going to take over."
So what are the implications? First, let's look at the implications for Deep Seek's direct competitors, which would be companies that have developed their own large language models, which is basically OpenAI, Anthropic, Meta AI, and of course, Alphabet's Google's Gemini.
I always believe that in anything, there's always the negatives and the positives. So let's begin with the negatives.
So if this is true and Deep Seek now offers a free open-source large language model that is superior to ChatGPT and Gemini, what's going to happen?
The first thing that could happen would be this could definitely lower the revenue of these companies if they are forced to lower the prices for their LLM subscriptions. Their large language model subscriptions. When you subscribe to Gemini Advanced or ChatGPT-4, you pay money to access. But why would you do it when you can get it for free with Deep Seek, right?
So they could be forced to lower their prices for API access, their enterprise solutions, or cloud computing services that leverage AI.
Now, among all these companies, which would be the most impacted? The most impacted would obviously be OpenAI and Anthropic. Why? Because the entire business model, their main revenue comes from subscription of its AI large language model. So they'll be the biggest to be hit.
Now, as far as Meta and Google are concerned, which are the stocks I own, I think it could be more positive than negative.
The first reason it could be positive is because, again, Meta is projected to spend over $60 billion in capex in developing their AI models in 2025, and Google is projected to spend $50 billion.
Now, imagine if it is true that Deep Seek can achieve so much with a fraction of the cost. Then isn't it good news? That means that Meta could potentially not spend $60 billion; Meta could spend maybe $6 million, right? If what Deep Seek says is true, Google, instead of spending $50 billion, may spend $5 million.
So if Meta could save $60 billion and Google could save $50 billion, achieving what they want to achieve in AI, then that would lead to huge cost savings, huge capex savings, and increase their profit margins, increase their earnings per share, and be great for the share price.
The second reason I'm not too concerned about Google and Meta specifically is because their main business model, their main revenue generator, is not subscription from their large language models.
For example, Google does charge a subscription for their Advanced Gemini, but again, it is very negligible, the revenue that they get, right? As far as Google is concerned, where do they make most of their money? From advertising, right? Advertising is their main revenue generator.
And by the way, Google also owns Waymo, which is the market leader in autonomous driving software, which is going to be another huge source of income. And of course, Google owns YouTube, which is another big source of revenue.
Meta, same thing. Meta primarily makes their money through advertising, not through AI subscriptions per se.
So these two companies, Google and Meta, primarily monetize AI through advertising, through better user targeting, recommendation engines, and ad platforms.
But of course, we can think logically and rationally, but the market tends to react to this news. The market in the short term tends to overreact to bad news and overreact to good news.
So as an investor, once we know that we've got a high-quality company that we want to buy, we know the intrinsic value, we use these short-term panics and fears to add shares if we don't already have a full position.
So as I'm speaking right now, before the market opens, you can see that the NASDAQ is already down 3.44% in terms of the futures. You can see a big drop right here. Well, not really that big, but it's a significant pullback because of this news.
Now, if we look at, for example, Google or Alphabet, all right? So you can see that chart right here. Now, if we look at the pre-market data, so pre-market, it's down all the way to 193. So it's down, I think, about 4-5%.
If the market opens there, 193, is that 193? Yeah, 193. So that would be somewhere about there. So Google could open all the way down here, dropping from the previous day.
Now again, if you look at it from the grand scheme of things, is it a big drop? Well, not yet, right? So again, it's all about intrinsic value. My intrinsic value for Google is $26.
So with that drop to 193, yeah, it's undervalued, but it's not undervalued enough. All right? So as an investor, I like to buy great companies when there's a bigger margin of safety, when the share price is much lower below the intrinsic value, so that I get a discount.
What I do every month is that I will update my buy levels for each stock. So for Google, for example, my buy levels where I would add more shares if I didn't have a full position would be at 177, 167, 154, and 147.
These are my four buy levels determined by technical support levels on a daily, weekly, and monthly time frame.
So in other words, for me to get interested in adding more Google, it's going to drop a lot more. All right? It's going to drop to... Now, I already have a lot of Google personally, right?
So for me, I'll only be tempted to buy only if it gets to like 154, 147, and maybe I'll be tempted to buy, right? But right now, it's not that big of a discount to get me excited just yet.
How about Meta? I also have a pretty big position in Meta, which I bought at much lower levels. And Meta, by the way, is now overvalued.
Okay, my intrinsic value for Meta is $549. So right now, it's $652. But based on the pre-market data, let's see, it's down to 620 based on the pre-market data.
So if it opens at 620, let's see where that brings us. If it opens at 620, it would be somewhere there. Still not cheap enough, right? Just a drop there, no big deal, right?
So again, it's still overpriced, right? You know, for Meta, for me to want to add more, Meta is going to go below my $549 intrinsic value, where it's going to drop a lot lower.
Come on, you can do better than that, right? So my buy levels for Meta are 541, 496, 453, and 414.
And the reason I always have three to four buy levels is because we never buy at once. We always average in our position.
Yeah, and again, I already have a big position in Meta, so I'll only be tempted to buy only at the third or fourth buy level. But if I did not own any Meta shares at all and I was building a new position, then I would start adding at the first support level.
I would start nibbling a bit at 541 and then buying more at 496 or lower.
So let's take a look at now the implications and the impact on semiconductor stocks, specifically Nvidia, which now designs the bulk of the high-end advanced accelerators, the advanced AI GPUs, and of course the related semiconductor companies like ASML or TSM or Broadcom.
How does it affect if Deep Seek's claims are true?
Now, if Deep Seek's claims are true that you can do more with a lot fewer chips, then could their demand for advanced Nvidia chips be affected? Yes, there could be a short-term demand hit.
Now, how much demand hit is very hard to speculate. It's very hard to quantify. Could it be a 10, 20, 30, 50% drop in orders? I don't know; it's really hard to speculate, right?
So like I said, if Deep Seek's model truly requires fewer specialized AI chips to achieve superior performance, the major cloud providers, again your Amazon, Meta, Microsoft, may cut down on their purchase of the high-end Nvidia chips or may slow down the orders of the AI accelerators.
And could that hit the share price in the short term? Absolutely. But again, by how much is very hard to quantify.
So at this point in time, am I going to change or reduce the intrinsic value of my semiconductor stocks? Not yet, because again, it's very speculative at this moment.
Again, I don't know whether the claims are true, and if they are true, does it mean that the hyperscalers will cut all their orders and not buy Nvidia chips anymore, which they've ordered? Again, we don't know, right? It's hard to quantify.
So for now, what's my game plan? My game plan is to hold my Nvidia shares. I've got about a 4% allocation in my portfolio. I'm going to hold the shares. I'm not going to panic and sell my Nvidia or ASML. I'm just going to hold it, right?
But when it comes to adding more shares, again, as always, I never buy and sell based on speculation or prediction. I look at the intrinsic value, and I only buy if the price goes below the intrinsic value by a certain discount, so I've got a margin of safety.
So I'm going to maintain my valuation but give myself a bigger margin of safety.
Let's take a look at where Nvidia is right now in terms of the charts and the intrinsic value. Again, the market has not opened yet, so I can only look at pre-market data. When the market opens, the price could be a lot lower or higher; I don't know, right?
So Nvidia, currently my valuation based on the current free cash flow and their current projections is about $130. The market closed at $142 on Friday last week.
So based on the pre-market data, I think Nvidia is down like 9% right now. Let me just double-check. Okay, just go to trade here, and yep, so the opening bid is about 128, 129, which is about a 9% drop, right?
So where does that land on the chart? So that would land at 128, right? That would land somewhere here, right? 128. Let me just double-check that again by looking at the pre-market data.
Yeah, 128 is the pre-market data. Okay, so if it opens at, let's see, 128, that would be below my intrinsic value of 130.
But again, would it be cheap enough for me to add more shares? No, because I don't really have a margin of safety. Because what if their orders get slowed down because the hyperscalers cut their orders? Then the valuation may readjust downwards.
Okay, so as you can see, as always, I've got my four buy levels at 121, 111, 90, 75. And again, it's not even at my first buy level. 128 is not even at my first buy level.
Okay, and if I didn't have any Nvidia shares at 121, I may nibble a bit, right? But for me, I already have a 4% position, so I probably only want to buy below $100. Below $100 gives me at least a bit of a margin of safety.
Okay, so how about other companies like ASML? So ASML already got hit, actually, right? Because of the cyclical slowdown in their orders from Samsung and TSM.
So they've dropped quite a bit, but they've kind of rebounded, and now with this ban, news is going to drop again. Pre-market data puts it at 651, so that's, I think, a 12% drop.
651, so that would bring it to somewhere here, almost back down to the previous lows, right? Back down to previous lows.
So my valuation for ASML is $797. I'm not changing the valuation, and that would actually bring it down near the fourth support level, which would actually be pretty attractive for ASML.
Aren't I concerned that if Deep Seek's claims happen to be true, wouldn't that really affect the business of the semiconductor stocks? And if so, why am I holding? Why am I not selling it away?
Well, because again, I'm a long-term investor when it comes to these stocks. I'm investing for the very long run.
And while there could be, again, I use the word could be, right? Well, there could be a short-term demand hit if the claims are true, but I believe that long-term, the demand will grow. The demand will make up for it.
Okay, over time. Why? Because as large language models become more cost-effective based on Deep Seek's claims, then there could be a wider adoption of AI across industries that previously found it too expensive.
So right now, it's only the very rich companies, the hyperscalers, the Meta, the Google, the Amazon that can afford to build their own proprietary AI models.
But if Deep Seek has proven that you can do it a lot cheaper with less advanced chips and with lower budgets, then this could create a wider adoption of AI from startup companies, from academic labs, from smaller enterprises.
So this would result in a larger overall user base that will increase the aggregate demand for AI-related hardware, which means you still need a lot more hardware as we grow in this AI revolution.
And this would offset a lower demand per model. At the same time, even if one solution, which is LLMs, is more efficient, there'll be a rise among the hyperscalers to deploy more advanced or specialized models that could keep the demand for advanced chips high.
So like it or not, we are in an AI race, and I believe that your hyperscalers, again like Amazon and Meta, they will still want the best of the best chips in order to gain an edge.
And if you think about it the other way, now that China has proven that it's catching up with the US, if you are the US companies, will you be more frightened? Yes, like, "Damn, the Chinese are catching up."
So all the more, you may not want to cut back on your capex. All the more, you may say, "I want to spend more and get the best chips and the most advanced chips so that I can take it even further," right?
So it may swing the other way as well.
So if you ask me, am I 100% convinced that this news will lower the demand for Nvidia chips? I'm not convinced.
All right? I'm not saying it's not going to happen, but I'm not 100% convinced.
So because of that, like I said, I'm going to maintain my position. I'm not going to panic and sell. But at the same time, before I add more shares, I'm going to make sure that I've got a good discount, and I'm looking at it from an investment long-term perspective.
Now, if you happen to be a short-term trader, not a long-term investor, then of course, it's a different story. If you're a short-term trader, then you purely trade based on the price action.
You ignore the news, look at the price action. If Nvidia drops and you notice a double bottom pattern or a slingshot pattern, which I teach in my course, or Alson teaches his downtrend reversal pattern, then you take the trade, right?
Take the trade, and you put your stop loss, put your profit targets, and you could play the rebound. And you never know, the rebound could come faster than you think.
So I do hope this has given you a more comprehensive perspective of what's happening in this situation, and you can make the best decision for yourself.
So thank you for watching, and as always, may the markets be with you.
If you want to catch my latest videos, click on the subscribe button right now. Click on the bell so you get instant notifications once I upload my latest video.
If you want to check out my online courses, go on to PPROF and learn how to trade financial markets and create an income from all around the world.
If you want to join my live Wealth Academy program, go on to WealthAcademyGlobal.com and find out more about how you can learn investing and trading live online.
This is Adam Koo, and may the markets be with you.