Hello everyone, and welcome to this professional and non-professional US market analysis channel. I am market analyst Xiao.

Today, we're going to talk about a hot topic: the "Big Short" Michael Burry is shorting the entire AI sector. He specifically brought up a concept called depreciation.

We will do a very detailed breakdown of whether his concept itself, or his view itself, is valuable, and whether it's worth our reference, and whether the entire AI sector has a bubble. Coincidentally, after the market close on Monday, Nvidia's "prodigal son," CoreWeave, released its earnings report. We will use this company's perspective to trigger a very detailed breakdown of his view itself, okay?

Before we delve into these topics, let's first use FactSet data to summarize the third-quarter earnings report, okay? I won't waste any more time; let's get straight to it. Everyone, please like and leave comments.

First, let's look at Monday's asset price movements. The US dollar and US bond markets saw little volatility, but the three major indices experienced significant gains.

I mentioned in the morning, around 10 or 11 o'clock, that today's gains weren't that comfortable, and participation wasn't very high. However, after I made that statement, around 11:30 or 12:00, looking at the hourly candlestick chart, the market formed a lower shadow line and then rebounded all the way.

Looking at the final results, overall market participation on Monday was okay, showing a relatively clear rebound. If I remember correctly, the S&P 500's 20-day and 50-day moving averages both returned to 44% or 48% (you can check the specific data on my app). Nasdaq participation also increased. So, looking at the closing results, participation was okay and had increased.

Monday's gains were almost entirely from big tech: 2.8% for big tech, 2.6% for the chip sector, and 2.6% for the tech sector overall. The S&P Equal Weight Index, in contrast, rose 0.6%.

Three sectors experienced declines: consumer staples, utilities, and REITs, which are relatively defensive sectors.

The overall trend on Monday was somewhat "risk on," contrasting with last week's "risk off." Last week saw sector rotation, with tech performing well and defensive sectors weakening. This Monday, we saw defensive sectors weaken and funds flow into tech.

I made a position adjustment on Monday. What kind of adjustment it was, I hope everyone can open my APP to see, because all my position adjustments are shared for free on my APP.

Let's look at the companies that will release earnings this week. CoreWeave released earnings after the market close on Monday, which we'll discuss in detail shortly. On Tuesday before the market opens, New Relic will also release earnings, which we can discuss tomorrow if there's an opportunity.

Why did I sell this company? It's not that I'm not optimistic about their earnings report on Tuesday. I don't know whether they will rise or fall after the earnings report is released. The sale of Sea Limited is entirely because I shared with everyone two or three weeks ago that I would be "lowering beta" and taking out some positions, and Sea Limited is one of those companies. I actually made position adjustments last week, and today I just sold the remaining positions. It's not that I have a very strong judgment about this earnings report; I have no judgment. I don't know whether it will rise or fall after the earnings report is released. That's one point.

Then on Wednesday and Thursday, my remaining two holdings, which are basically the last two companies in all my positions, will release earnings reports. On Wednesday and Thursday before the market opens, Circle and Disney will release earnings.

I said on X today that I'm not that concerned about the earnings reports of these two companies themselves. Because for Circle, you can look at the circulation of USDC, and for USDC, we already saw a very high month-over-month growth rate in circulation when Coinbase released its earnings report last week, or the week before last.

And for Disney itself, because of Disney+ and ESPN (a part of ESPN's content has already been put into Disney+), the number of subscribers for Disney+ should show a relatively significant increase this quarter. The important thing is to look at the execution. So, I'm not that worried about the fundamentals of these two companies' earnings reports. Of course, I have no certainty whether they will rise or fall after the earnings report is released after hours.

On Holding, this company released its earnings report before the market opened on Wednesday. On Holding is actually one of the companies I've operated on most perfectly so far this year. I can only say "relatively perfect." I bought it at $28-29 last year, then reduced my position at $60 at the beginning of this year, and then made the final technical reduction at around $50 (I think it was $50). I said if it fell below that price, I would liquidate all my positions.

From now on, it has continued to fall by 30% from the price at which I liquidated. If you look at its fundamentals, its data, it actually hasn't gotten worse. I personally believe the market is following a logic, which is that these two companies, On Holding and Decker, had very good gains, and I believe it's because they took a lot of market share from Nike.

And for Nike, I think the logic the market is following now is: is it possible that Nike's market share will slowly come back? That is, Nike is slowly reclaiming the market share that was originally taken by the On Holding and Decker brands. I think the market might be pricing this in, which is why it might be relatively weaker recently, or the stock price trend might be weaker. Of course, another simpler explanation is "valuation kill," because the valuations of these two companies are not low.

Let's look at the earnings data. First, as of now, 91% of companies have already released their earnings reports. We can see that on September 30th, before the earnings reports were released, the market's expectation for the third-quarter EPS growth was only 7.9%. Now, this figure has reached 13.1%.

From a sector perspective, except for communication services, where the market originally expected 3% but the final result was -7.1%, all sectors performed better than market expectations. So this quarter was basically a very comprehensive "beat."

But the final results weren't that good. This quarter we saw a widespread "sell the news" phenomenon. With basically all companies (91% have released, 9% haven't) having released their earnings reports, the market's expectation for 2025 EPS growth has reached 11.6%. It was 11% last time it was updated, right? Right? After big tech companies released their earnings reports, it has reached 11.6%.

After the fourth-quarter earnings reports are released, when we look back at 2025 from 2026, I estimate that the final figure might settle around 13%, just as I predicted last time. The current market expectation for 2026 EPS growth is 14.3%, close to 14%. This number will be adjusted further. Towards the end of the year, we will then make a forecast for 2026.

During this period, or in the next month and a half or so, I will continue to execute the plan I formulated in early October: if the market and chip stocks reach a relatively high position, I will gradually reduce the beta in my positions along the way.

Where is "sell the news" reflected? For example, 166 companies had EPS growth results that were 0 to 5% better than market expectations, but these companies on average fell by 2.1%. Under normal circumstances, they should have risen, not fallen.

From another perspective, if it was better than expected, the average increase was 0.9%, but this quarter it was 0, with no increase. If you were worse than market expectations, the average decrease was 2.6%, but this quarter it was -5.1%. So, the market clearly punishes underperforming companies very severely, and even well-performing companies don't necessarily see gains.

Finally, let's talk about AI, or how the "Big Short" is shorting AI. One of the core concepts behind his view is "depreciation" in accounting principles.

Generally speaking, when we talk about capital expenditure (Capex) and operating expenditure (Opex), the underlying concept is the same: a company spends money. Operating expenditure is simply how much you spend each year, but capital expenditure has a trickier aspect.

This means that if you spend 100 yuan now, this 100 yuan will not immediately enter your expenses for this year, because it may be depreciated over 5 years, 2 years, or 15 years. So, should it be 2 years, 5 years, or 15 years? There are valid arguments from different perspectives, leading to disagreements in viewpoints.

Disagreements in viewpoints lead to disagreements in assumptions, causing some to believe these assumptions in the market are overly optimistic, while others might consider them overly pessimistic, or reasonable. No one would consider them overly pessimistic. If they are considered overly optimistic, like Michael Burry, he believes they are overly optimistic, then there is a bubble in the entire market.

For example, capital expenditure. If I spend 100 yuan now, and the GPU I buy now is worth nothing after 5 years, then the annual expense is 20 yuan, right? The depreciation you should record each year is 20 yuan. If it has no value after 4 years, then your depreciation should be 25 yuan.

What's the difference between 25 and 20? Of course there's a difference. If your depreciation this year is 20 yuan, it will increase your profit this year (because this is an expense). If your depreciation this year is 25 yuan, compared to the 20 yuan expense, it will reduce your overall profitability. If you depreciate over two years, your average annual depreciation is 50 yuan, which will significantly reduce your profit this year.

So we find that this depreciation assumption will significantly affect the profitability of companies like CoreWeave in their first two years. If you are relatively optimistic, and you assume, for example, like the chart Michael Burry shared today, assuming 5 or 6 years, then your profitability (margin) in the first two years will be very high, and the next few years might look relatively bad.

From another perspective, the data here is the GPU price on AWS. You can see that the price at which it can be sold each year has actually been falling. This precisely proves that assets like GPUs and CPUs are depreciating assets. Their value is highest in the first two years, so the price you can sell them for is higher, but after the next three years, their price will continue to fall, and the magnitude of the fall may be quite significant.

For example, here are two examples: one is the H100, and the other is the A100. The H100 was $3.29 in Q3 and Q4 of 2024, and this year it's $2.72, a year-over-year decrease of 17%. The A100 was $2.49 in the same period last year, and this year it's $1.02, a year-over-year decrease of 60%.

This precisely illustrates that for these assets (GPUs, CPUs) themselves, the longer you wait—you don't even need to wait 4 or 5 years, just one or two years—their price falls, indicating a very significant decline in the asset's inherent value.

Next, let's talk about CoreWeave. First, let's look at the data. The data itself is actually very good: revenue was 6% higher than market expectations, EPS was 77% better than market expectations, and adjusted operating income also exceeded market expectations by 22.3%.

Why did the stock fall after hours? It's because the delivery of one of their data centers has been delayed. Let's look at more details, as not all viewers are very familiar with this company.

First, let's look at their revenue for this quarter: $1.4 billion, a year-over-year growth of 134%, more than double. What is the most important data here? RPO, which is $50 billion. RPO refers to the business that companies cooperating or contracting with them must fulfill in the future.

It is precisely because the certainty of their future revenue is very high that these types of companies dare to invest so much capital expenditure now. Secondly, it is also precisely because the certainty of their future revenue is very high that they can evolve into their current financing method.

What kind of financing method? It's what's currently being discussed a lot in the market: they use the GPUs they bought from Nvidia, plus the orders they have with other companies, as collateral to borrow money. If you simply use GPUs as collateral to borrow money, you'll find that no one would want to lend you money for this deal, because you know that GPUs themselves are depreciating assets.

It's like if you use a car as collateral now, they would at most lend you 30% or 20% of the car's value, because everyone knows that the car itself is a depreciating asset. GPUs are the same. However, if they include the contract with Microsoft in this agreement, and include the future cash flow contract with Microsoft, then this collateral itself becomes very valuable.

Because who is Microsoft? Microsoft tells you, or Microsoft promises you this cash flow, and the probability of it being fulfilled is even higher than the cash flow promised by the US government. The US government's rating is not as high as Microsoft's; Microsoft is triple-A, and the US government is double-A, right?

So this type of lending is called DDTL. There's a lot of discussion in the market that this type of DDTL lending is very high-risk. I, on the other hand, don't think the risk is very high here. Because if you simply use GPUs as collateral, the risk is indeed very high. But if you also put the contract itself with Microsoft as collateral, then the credibility of this collateral itself becomes much higher.

Of course, they can also directly issue bonds in the secondary market for $1.75 billion, or directly raise capital, directly issue equity. So, all together, CoreWeave has raised $4 billion in financing so far. Is this $4 billion enough? Clearly not.

For the entire year 2025, their capital expenditure budget is $13 billion. Plus interest expenses (you have to pay interest when you borrow money), which already reached $1.23 billion in 2025. If you add it all up, the total capital expenditure plus interest expenses has already reached $14 billion.

Then, adding to that, they now expect 2026 to be more than double the 2025 figure. Think about it, let's do some math: 2025 totals $13 billion (I'm not even counting interest expenses), and next year, 2026, will be even higher. $13 billion plus $26 billion means that the total capital expenditure for these two years combined will be at least $40 billion.

Why do they dare to spend so much money? Because of the RPO we discussed earlier, which is for this profit, for this revenue (I keep saying profit, but it's their future revenue), the certainty is very high. The total RPO is $50 billion. So, with this future revenue, they dare to invest heavily now, spending $40 billion.

What does $26 billion a year mean? It's actually a very large amount. Because if you think about Amazon, one of the largest big tech companies, how much does it spend in a year? Over $100 billion, maybe $120 billion or $130 billion. So $26 billion is already one-fifth of Amazon's capital expenditure. Your scale is very different from Amazon's, and your revenue is very different from Amazon's, but your capital expenditure has already reached one-fifth of Amazon's. So we need to have some concept of the numbers behind it, and we'll know that these companies are indeed investing a very large amount of capital expenditure.

Where do the risks for these types of companies come from? Actually, their risk doesn't come from a potential decrease in AI demand. Because even if AI demand decreases, their revenue is not at risk. As discussed earlier, if AI demand decreases, let's take Microsoft as an example. If AI demand decreases, Microsoft will be impacted, and Microsoft's margin in their cloud business will be impacted. But Microsoft still has to pay CoreWeave the revenue they originally promised, or this sum of money must be paid.

Whether you use it or not, I will provide you with this part of the computing power. Whether you use this part of the computing power or not, or how much you use, you must pay me the money we originally promised. Otherwise, companies like CoreWeave would not be able to invest so much capital now. So, the point is, if demand decreases, companies like CoreWeave will not experience a very large impact on their revenue. Instead, Microsoft's margin will be affected, or AWS's margin will also be affected.

So what's the risk? Concentration risk. Concentration risk was also something an analyst asked them about in this earnings report. He said that in the future, because their current orders are very concentrated, with three major companies: Meta, Microsoft, and OpenAI. Is it possible that there will be more companies in the future?

Actually, I don't think this is a risk. Why? Let me think, if they went to many other small companies, even if this small company today promised you a $10 billion order over the next five years, would you dare to accept it? Because these small companies themselves don't have such healthy balance sheets, and their revenues might not be that healthy. So even if they throw a $10 billion order at you, its certainty isn't that high.

Conversely, the orders that these big tech companies, like Meta, can sign with you have high certainty. But these big companies come with a concentration risk. However, at the current stage, I don't think these concentration risks are that important. Instead, they become a guarantee for your future revenue, right? If AI demand slows down, these big tech companies guarantee your revenue, because big tech companies still have the ability to fulfill the contracts we originally signed.

The second risk is called execution deployment risk. This risk of execution itself was actually seen in this quarter's earnings report. The delivery time point for one of their data centers was delayed. This is actually a risk for them. For example, if I am Meta, right, and I signed an agreement with you saying you must deliver a certain amount of computing power to me before the end of this year, and you don't deliver, then I can exit this contract. I can completely sign a contract with someone else, understand? Or you pay me a penalty, etc., because you didn't fulfill the contract.

Of course, currently, because the demand for computing power is still very high, meaning supply is less than demand, Meta says, "I can wait." So, this is still the situation now, but it won't be the case in the future.

The last risk is technology risk. This risk is actually the one Michael Burry challenged earlier. What does it mean? It means that if Nvidia updates and iterates their GPUs in the future, it might make your previous generation of GPUs worthless. Of course, this premise might also be based on the overall market demand starting to slow down.

When market demand slows down, there won't be a situation where H100s or A100s are still in reserve. If market demand starts to slow down, everyone will definitely demand, and will definitely focus on the most advanced chips, and older, less advanced chips will no longer be used.

If no one uses the older, less advanced chips, it will force companies like CoreWeave to update and iterate. If they update and iterate to the new generation of Nvidia chips, it means that CoreWeave will need to invest more money in capital expenditure.

So, if demand slows down in the future, for companies like CoreWeave, it's not that their revenue will be affected; that's not the case. Rather, if demand slows down, it will make the older generation of chips worthless. Everyone will need to pursue the new generation of chips, pursue performance, and pursue efficiency.

In this situation, with the agreements signed between them and big tech companies like Microsoft, big tech companies have the right to compel, question, or negotiate with CoreWeave to update and iterate. Or, they might sign another contract where you, in this new contract, must provide me with the latest generation of Nvidia chips.

If that situation occurs, it would mean that a chip they originally thought would become worthless after 6 years would suddenly become worthless after 3 years. That would directly impact CoreWeave's margin. This is why the revenue risk for companies like CoreWeave is actually not significant, but where is its risk? Its risk is in the margin.

Its margin, it's very likely that in the first and second years you think you can make this money, but clearly in the next third, fourth, and fifth years, your margin is very likely to be lower than in the first and second years, even much, much lower, even to the point of losing money. If the GPU you are currently using itself has no value, then this is actually a point that Michael Burry challenged earlier.

Of course, one last thing I want to add is that I neither agree nor disagree with Michael Burry's point of view. I actually think his point is at least reasonable. Because I see some people who are very optimistic about the AI industry, for example, they think: "Look, in the past, H100s, A100s, including Google's TPUs, could be used for 5, 6, 7, 8 years." They made an assumption that TPUs could be used for 7 or 8 years in the past, so if these CoreWeave companies now assume that these chips can still last for 6 years in the future, I think this assumption is very dangerous.

Because you are treating a past assumption as a future assumption. Who can guarantee that in two or three years, this generation of Nvidia chips can still run for another four or five years? If this demand decreases, then your original assumption that it could still run for four or five years would no longer hold true, right? That's the future, that's entirely a judgment about the future. I'm just saying that past assumptions or past facts we've seen don't mean that these facts will continue in the future. I think it's a relatively dangerous assumption to make a linear extrapolation from historical results.

Finally, for our loyal viewers, Futu Moomoo's current promotion in the US market is very strong, and this promotion is only available to 8 KOLs in all of North America. So, if you go to other channels, you might not even see this promotion.

What's the promotion? You can get a $100 stock cash coupon by depositing any amount. On top of that, if you reach a certain deposit volume, you can get an additional $1,000 worth of free stock. So these two things are stackable. At the same time, idle funds can earn up to an annualized return of 8.1%.

In the Canadian market, by opening an account and depositing funds, you can receive a $30 CAD stock cash coupon, plus our channel's exclusive $50 CAD Amazon gift card, plus a commission-free card for two months up to $2,000 CAD, and idle funds can earn up to an annualized return of 6%.

So, for viewers interested in these promotions in Canada or the US, you can find your account opening links in the pinned comment. That's all for today's content, and we'll see you tomorrow.