저자: TomLee · 외국인
요약 보기
AI 관련 반도체주는 레버리지 청산(디레버리징) 국면이며 바닥 신호가 나타나고 있어 큰 반등이 임박했다고 주장. 마이크론 선행 PER 4.8배로 10년 평균 대비 저평가(약 33% 상승 여력), 이익 급증(345%)은 그대로이며 하락은 패닉 청산 때문이라고 설명. S&P500 DeMark 9-bottom 신호로 단기 추가 하락 가능성은 열어두되 큰 랠리 예상. 미국 마진부채 1.4조달러 및 한국 레버리지 ETF 청산이 원인이며 AI 자본지출(6000~7250억달러) 지속을 근거로 제시.
다듬은 글
Wall Street's top strategist and founder of Fundstrat, Tom Lee, clearly stated in this week's update video that AI has experienced a historic deleveraging over the past few weeks. But there are now signs indicating we are nearing the bottom, and according to him, a major rebound is just around the corner.
Let's take a look together. I think the theme over the past week is the big leverage unwind. And what I mean by that is we know these bottleneck stocks, like Micron which makes memory, is down 41%from its highs at 739. And the decline and sustained decline in memory and the bottleneck stocks and semis has actually caused Wall Street to ask funds heavily concentrated in AI to post additional collateral.
Now I've flagged for you before that margin debt has jumped sharply in the past 12 months to 1.4 trillion and historically this does signal a period where markets tend to have to consolidate that increased level of debt. But we also know a lot of this originated from Korea, and it has since spiraled significantly.
The Korea KOSPI index was up, but then began to decline starting in late June, and that spiraled in the last few days to be down 41%. The decline in the country index was triggered by this. Such a decline, even warned by financial authorities, was administered to retail investors because of these leveraged ETFs and the subsequent heavy losses suffered.
But here's the thing, valuations actually have come down a lot in semiconductors, paying 25 if you discount the earnings times, and you can see this here. By the way, memory stocks are down times and software stocks are 25 times but again, in memory, uh seven times is the average, but if you exclude Seagate you're actually closer to five times.
And if you look at software excluding Palantir you're in the teens. I mean the multiples have come down a lot. In fact , take a look at Micron here. Trades at 4.8 times forward earnings, which is a 1.7 turn discount to its 10-year average. So, it's gotten cheap. That's a 33%move just to get to its long-term average PE.
And it had a PE of almost 13 times just eight months ago. But as these semis and bottleneck stocks are correcting, keep in mind the Mag 7 and software, Ethereum and crypto have been recovering, something we've been pointing out, so just keep that in the back of your mind. And as for the semis , I do think we're near the end of the sell-off.
Let me explain. I'm showing you here the year-to-date chart for the S&P 500, and I'm indicating here the DeMark indicators. Um, the 13s, the top , these are called countdowns, and as you can see, we had 13 top in January and some in late May and June, and as you can see, the market consolidated for a bit, and then eventually fell.
Well then, it creates what are called buy countdowns, and the buy countdowns recently have been nine bottoms. As you can see, we had a nine bottom, uh, as the war escalated earlier this year, markets fell a little bit further, but then we had a 21%rally. Well, do you see we got another nine bottom today?
So, maybe we fall a few more points, but at the end of the day, I think a big rally is around the corner. So, let's look at Granny stock performance. Um, on the large-cap Grannies, the top five performers of the past week were Packaging Corp, AMD, Deere, Vistra, Northrop Grumman. The bottom five were Alphabet, Meta, American Express, Oracle, and Tesla.
And as for Small Granny, the top five performers were Riot, Weatherford International, Carpenter Technology, Mueller Industries, and Evergy. The bottom five performers were Valmont, MP Materials, UIPath, Zscaler, and Bloom Energy. As for the themes, style tilt remains the strongest, followed by easing financial conditions and the PMI recovery.
To understand Tom Lee's "big leverage unwind," we must first know what happened over the past year. In the first half of 2026, the AI investment race entered a near-frenzy state. Amazon announced 2026 capital expenditures of over $ 200 billion, Microsoft $ 190 billion, Google $ 90 billion, and Meta $ 115 to $ 135 billion.
Adding it all up, $ 600 to $ 725 billion was poured into AI data centers, GPUs, and high-bandwidth memory. Goldman Sachs even predicted that by 2027, the cumulative AI capital expenditures of these hyperscalers would exceed $ 1.15 trillion. What kind of scale is this? This is almost equivalent to using the entire GDP of Switzerland to build these data centers .
With such fierce demand, the tightest links in the supply chain naturally skyrocketed: NVIDIA's GPUs, Micron's HBM memory, and SK Hynix's storage chips. These are exactly the " bottleneck stocks" Tom Lee referred to. In the first half of the year alone, Micron hit an all-time high of $ 1,255. The Korean KOSPI index also soared to the sky following the semiconductor boom.
The question is, how much of these gains were built on borrowed money? The answer is, quite a lot. The Wall Street Journal exposed the bottom line in a report at the end of June. The US margin debt balance soared to an all-time high of $ 1.4 trillion in May 2024, a 54%year-over-year surge. This figure accounts for 6.2%of the M2 money supply.
When was the last time we reached this level? It was 6.3%during the 2000 dot-com bubble, just a 0.1 percentage point difference. South Korea is even more extreme; retail investors there are the world’s most enthusiastic users of leveraged ETFs, with 3x long semiconductor index products being bought until they hit liquidation.
Tom Lee mentioned in his video that much of the leverage risk stems from Korea. Then, starting in late June, the sentiment shifted. Some began questioning returns, and progress on advanced chips in China made the market nervous. Add to that the need for consolidation, and semiconductor stocks began to pull back.
Once the pullback started, the chain reaction began. Their description is very precise: as semiconductor and tech stocks continued to fall, Wall Street demanded that heavy-position funds meet margin calls, and those who couldn't keep up were forced to liquidate their positions. That selling pushed stock prices even lower, which in turn triggered even more margin calls.
This is a deleveraging spiral, or to put it more simply, a stampede. What happened over in South Korea? According to Goldman Sachs data, out of a total population of just over 50 million in South Korea, nearly one in every 30 people received a margin call. The KOSPI plummeted 41%from its high. South Korean financial authorities held an emergency meeting, regulators suspended the issuance of new leveraged ETFs, and the Finance Minister even came out to publicly apologize to retail investors.
So, please remember a key judgment: this is not a decline because AI is failing, but a decline because the leverage couldn't hold up. The fundamentals haven't deteriorated in the slightest; Micron just released its best earnings report in history, Nvidia's revenue is still growing by 85 %, and hyperscalers are still increasing their AI investments.
It’s not the companies that have changed, but the position structure. This distinction is extremely important because it directly determines how to view things after the sell-off. Tom Lee didn't just guess that the bottom is in ; he provided evidence from three dimensions. The first is that valuations have fallen to ridiculous levels.
Tom Lee's video provided a set of figures. The P/E ratio for the semiconductor sector as a whole is about 25x, but after excluding AMD and Intel, it is actually only in the teens . NVIDIA's forward P/E has fallen to 16x. Note that NVIDIA's revenue is still growing at 85%year-over-year. The storage sector's average is 7x, but after excluding it, it's only around 5x .
Micron is at a 4.8 x forward P/E. What is the concept of 4.8 x? Tom Lee provided a precise positioning, which is even lower than Micron's 10-year average. He says that a simple mean reversion would represent a 33%upside. Eight months ago, the P/E was 13x; now it's 4.8 x. That's nearly a threefold difference, not because profits plummeted—Micron's profits are actually still surging.
It's purely because the stock price was hammered by panic. The second point is that capital is rotating rather than fleeing. Many people actually overlook this, but I think this is the most ingenious part of Tom Lee's logic. He says that while semiconductors were being dumped, the " Magnificent Seven," other businesses, software stocks, and Ethereum were actually recovering.
What does this indicate? Capital hasn't fled the entire risk asset market; it is simply doing one thing: deleveraging. Who has the highest leverage? That would be semiconductors. What are Korean retail investors "all-in" on? It's leveraged semiconductor ETFs. Where is the U.S. margin debt concentrated?
It's also in AI-related stocks. So, it's not that AI is failing, but rather that the leverage was too heavy and is now being cleared. Once the clearing is done and the selling pressure disappears, these stocks will have huge upside elasticity . Because the fundamental support is still there, it's just temporarily obscured by panic and forced liquidations.
Just how ridiculous is Micron's 4.8 x P/E? Let's take a minute to get a sense of how outrageous this number is. Micron's FY2026 Q3 data shows revenue of $ 41.4 billion, a 345% year-over-year increase, net profit of $ 28.24 billion, and EPS of $ 25. Wall Street's consensus was around $ 20, so it beat expectations by 25%.
Gross margin is 85%, with Q4 revenue guidance of $ 50 billion and a gross margin. Let's make a comparison. NVIDIA's revenue is $ 81.6 billion, growing 85% year-over-year, with a forward P/E of about 16x. TSMC, the king of AI foundries, has a forward P/E of 18x. Texas Instruments has a growth rate far lower than Micron's, yet its P/E is over 20x.
Micron's growth is 345%, yet its P/E is only around 4.8 to 5.4 x. So why is the market giving it such a low valuation? The first reason is the cyclical nature of memory chips; many in the market have been taught that the boom in storage chips will eventually end. So, even if you’re making a fortune right now, they won’t give you a high valuation, fearing this time is the same—but the question is, is it really the same?
What drove previous memory cycles? Smartphone shipments, PC shipments, and server upgrades—these demands have ceilings and will saturate . But this time, AI capital expenditure is different: Microsoft is spending $ 190 billion a year, Amazon is spending $ 200 billion, and both are still ramping up. Demand for HBM for NVIDIA’s next-gen platforms will only grow larger, not smaller.
Micron’s own management stated on their earnings call that HBM4 has already secured a matching share with NVIDIA, and volume will continue to ramp up in the fourth quarter. The demand drivers this time are completely different. It’s not a cyclical spike, but a structural, step-like growth. If the market is still using the 2018 or 2022 templates to price Micron, it is making a cognitive bias error.
The second reason is the passive selling pressure from deleveraging. As mentioned earlier, many funds used margin to buy Micron, and after receiving margin calls, they are forced to sell—not because they don’t like the stock, but because they lack the margin. Passive selling doesn't look at valuations.
It only looks at how much of a position must be liquidated to meet the margin call requirements, right? This is why a company with profits soaring by 345% could see its stock price fall from $ 125 to $ 73. Because the price discovery mechanism has been distorted, but such distortions won't last forever .
They specifically mentioned South Korea, a case study very much worth digging into. The KOSPI index in South Korea is down 41%from its highs, which is an extremely rare decline for a national-level index. Goldman Sachs ' analysis report points out that the main driver of this plunge is not fundamentals, but mechanical leverage liquidation and a liquidity vacuum.
Simply put, the collapse of triple-leveraged ETFs triggered passive liquidation, pushing the market lower, which triggered more margin calls and forced retail selling, causing the market to drop further. This is a pure liquidity crisis, not a valuation crisis. Interestingly, the South Korean market has already started to show signs of a rebound.
Goldman Sachs pointed out that after the SPI plunged 8.95%in mid-July, it quickly rebounded. That is because once everyone forced to sell had been liquidated, those remaining were long-term investors or low-leverage accounts, and the selling pressure naturally disappeared. Historically, there are similar precedents.
The memory chip cycle bottom in late 2022 followed the same pattern: Micron fell from $ 98 to $ 49, a 50%drop, and the whole market was shouting that the storage cycle was over. As a result, it began to rebound in 2023, and by 2024, it had already surged several times over. Rebounds after stampedes end are often faster and stronger than expected, because those who needed to sell have already finished selling.
Finally, let's do some forward-looking thinking. If Tom Lee's judgment is correct and deleveraging is nearing its end, with the bottom basically confirmed, what might happen next? Space One: Valuation reversion to the mean. For Micron to return from 4.8 x to its 10-year average of about 6.4 x would represent a 33%gain.
From the current $ 87.4, that corresponds to about $ 116. Note that this just returns to a normal level without any premium; if it returns to the 13x seen before August, that would be a doubling in price. Space Two: Earnings revisions. Micron's guidance for the fourth quarter is $ 5 billion in revenue with an 86%gross margin, and Wall Street's consensus for 2027 is still being revised upward.
If we extrapolate based on current earnings trends, Micron's reasonable target price could be $ 1,500 or even over $ 2,500. Of course, this is an extremely optimistic scenario, but it shows that in a high-growth environment , the margin of safety at current prices is indeed significant. Space Three: Macroeconomic tailwinds.
In his July 9th CNBC interview, Tom Lee mentioned another point: inflation easing in the second half of 2025. Core PCE was 3.41%in May, and crude oil has already fallen from $ 114 in April to around $ 74. If inflation continues to fall, expectations for Federal Reserve rate cuts will heat up, which acts as a systemic positive for growth stocks and tech stocks as a whole.
Space Four: The certainty of AI capital expenditure. As we mentioned earlier, the big four tech giants are spending $ 600 billion to $ 725 billion in 2026, and NVIDIA’s supply commitment contracts have already piled up to $ 119 billion. NVIDIA CEO Jensen Huang stated in an SEC filing that AI infrastructure development is the largest infrastructure expansion in human history.
This money is contractually committed; it’s not just a vision. It will be spent: GPUs must be bought, HBM must be purchased, and data centers must be built. As one of the world's three major HBM suppliers, Micron is a definitive beneficiary. Moomoo, a brand under a Nasdaq-listed company, has become the top investment platform of choice for Chinese investors globally.
I’ve secured an exclusive limited-time offer: win up to $ 100 in stock, and earn up to 8.1%APY on idle cash. Check the link below if you’re interested. To wrap up today’s logic, the recent crash in the AI semiconductor sector is, in essence, a deleveraging event. US margin debt hit a record high of $ 1.4 trillion, 1.2 million accounts in South Korea faced margin calls, and $ 1.4 trillion in market cap evaporated.
Micron, hit the hardest, is at a 4.8 x forward P/E—a figure only seen during times of panic. My bullish thesis is based on the convergence of three factors: valuations are extremely cheap , technical indicators are flashing buy signals, and capital is rotating rather than fleeing. The conclusion is that deleveraging is nearing its end, and a major rebound may be just around the corner.
Of course, nearing the bottom doesn't mean a rally starts tomorrow; the market could still fluctuate for days or even weeks. But if you agree that this is a multi-year structural trend and that $ 725 billion in annual capital expenditure is no joke, this price window created by panic is definitely worth looking into.
For more in-depth videos, click below to join the channel membership. See you in the next one.
Comments 0