美股投资young

美股科技投资者 ai狂热者分享每日快讯. all in ai

299Following

673followers

Feed

This Wednesday could be the most important trading day for AI tech stocks in October! $META and $LRCX will release their earnings reports after the U.S. market closes on October 21. One represents the demand side of AI capital expenditure, and the other represents the supply side of semiconductor manufacturing equipment. Both companies reporting on the same day could directly influence the market's outlook on the AI industry for the coming quarters. $NVDA $MU $SNDK $SOXX $QQQ Why do I think these two earnings reports are so important? Because the market is currently experiencing a very contradictory situation. Semiconductor leaders like TSMC and ASML continue to show strong operational performance, yet the entire semiconductor sector has been plunging continuously. Last week, the Nasdaq dropped 4.5% cumulatively, and $SOXX closed lower for nine consecutive trading days. The market is no longer satisfied with companies telling investors that AI demand is strong; it is now demanding these companies prove how much profit the hundreds of billions of dollars in capital expenditure can actually generate. Let's first look at $META. According to the company's previously announced guidance, full-year capital expenditure for 2026 is expected to reach $130 billion to $145 billion, including financing lease principal payments. What does this mean? It means a company is investing over a hundred billion dollars annually to build AI infrastructure, and the economic returns on these investments may take many years to gradually materialize. More noteworthy is that Meta's operating cash flow in Q2 reached $31.86 billion, but free cash flow was only $784 million. Meanwhile, the company's Q2 revenue grew 28% to $60.8 billion, ad impressions increased 14% year-over-year, and average ad prices rose 12%. This data is very interesting because it simultaneously indicates two things: Meta's core advertising business still has strong profitability, but AI infrastructure construction is consuming a large amount of cash. So for this earnings report, what I care about most is not whether Meta's revenue can exceed market expectations, but how management explains future capital expenditure and investment returns. If AI can further improve ad recommendation efficiency, enhance commercial conversion, and drive the development of new paid products, then the huge investments may gradually gain profit support. But if revenue growth starts to slow while capital expenditure continues to rise rapidly, concerns about free cash flow could deepen. Now let's look at $LRCX. Lam Research is a key global supplier of semiconductor manufacturing equipment, especially strong in etching and thin film deposition technologies. In the last fiscal quarter, the company’s revenue reached $6.72 billion, a 15.1% quarter-over-quarter increase, and it provided guidance of about $8.1 billion in revenue for the current quarter. In other words, at least based on previous operational data, semiconductor equipment demand remains strong. But the market now needs to know whether this growth can continue into 2027. Will capital expenditure on DRAM, NAND, and advanced logic processes continue to expand? Have customers in South Korea and Taiwan adjusted their equipment purchasing plans? How much impact will demand from the Chinese market and export restrictions have? These questions are important reference points for $MU, Samsung, SK Hynix, and the entire semiconductor equipment sector. However, it is crucial to distinguish between equipment orders and end AI demand. Even if $LRCX continues to deliver strong earnings, it does not mean all AI data center projects will be completed smoothly, nor does it prove that every cloud computing company’s investment will yield reasonable returns. There is a clear time lag between semiconductor equipment orders, chip manufacturer shipments, and end customers generating profits through AI. I believe the most cautionary scenario this week is that companies report good earnings but stock prices continue to fall. This would mean the market might be actively lowering valuations across the entire AI industry chain, not just worrying about a single quarter’s performance falling short of expectations. Conversely, if earnings and future guidance remain strong and semiconductor stocks stop breaking down further, the market’s re-pricing of AI growth expectations may gradually approach a more reasonable stage. But there is still a key distinction here: improving fundamentals does not mean stock prices will immediately bottom out. Last week’s semiconductor sell-off involved not only controversies over AI capital expenditure but also interest rates, financing costs, institutional positioning, and overall risk appetite. Even if $META and $LRCX both beat expectations, it may not solve all problems in a single day. I have always been a long-term AI bull and still believe global computing power, advanced processes, storage, and semiconductor equipment have long-term growth potential. But after this round of continuous declines in October, I pay more attention to the relationship between corporate profit growth and stock valuation. A correct long-term industry direction does not mean every stock is worth buying at any price, nor does it mean investors can ignore short-term liquidity risks. Therefore, this Wednesday, I will focus on Meta’s advertising business growth, capital expenditure guidance, and free cash flow, as well as LRCX’s outlook on next quarter’s revenue and demand for DRAM and NAND equipment. After the earnings release, we will see whether the market chooses to reward companies that beat expectations or continues to use good news to sell off previously crowded tech holdings. If both companies deliver strong earnings but $SOXX still cannot stabilize, I will remain cautious. If fundamentals and price structure start to improve in sync, I will reevaluate opportunities to increase positions. I have always believed that the true long-term bull market for the AI industry cannot rely solely on continuously announcing bigger investment plans to sustain itself. The future market will definitely pay more attention to three questions: who truly generates revenue, who truly earns profits, and who can sustainably produce free cash flow. This Wednesday’s earnings reports may be an important start for the market to answer these questions anew.

Tomorrow is October 19th, the 39th anniversary of the 1987 Black Monday in the US stock market. On this day 39 years ago, the Dow Jones Index plunged 22.6% in a single day, setting the worst single-day percentage drop record in US stock market history. Now, the global semiconductor sector is experiencing continuous sell-offs, with South Korean, Japanese, and US tech stocks under pressure one after another. Will history repeat itself? I believe what truly deserves study is not the date, but the mechanisms that caused the market crash back then. $SPY $QQQ $SOXX $NVDA $MU On Friday, October 16, 1987, the Dow Jones fell 4.6%. The following Monday, the Asian markets led a sharp sell-off, and after the US market opened, sell orders kept pouring in, resulting in the Dow Jones dropping 508 points in a single day, a 22.6% decline. The entire market fell into a liquidity crisis, and many investors couldn’t execute trades at their expected prices. More notably, the 1987 crash was not caused by US companies losing more than 20% of their earnings in one day, but by the market’s internal trading mechanisms amplifying the existing selling pressure. At that time, a strategy called Portfolio Insurance was very popular on Wall Street. Its core idea was to use futures and other instruments to dynamically adjust stock risk exposure, aiming to reduce losses during market declines. But this strategy had a fatal flaw: as prices fell further, the model required selling more stocks or stock index futures. The result was that market declines triggered selling, which pushed prices down further, triggering even more selling. What was supposed to help investors control risk became one of the key factors accelerating the market’s decline when used simultaneously by many institutions. This issue remains relevant today. The financial markets in 2026 are certainly not the same as in 1987, but today there are volatility targeting strategies, CTA trend-following, leveraged funds, ETFs, and complex options hedging systems. When the market suddenly enters a high volatility state, some strategies may passively reduce stock exposure, and institutional risk departments may require traders to actively cut positions. These actions individually make sense, but if large amounts of capital reduce risk simultaneously, it can create a similar negative feedback loop. Especially when buying interest is insufficient, price declines don’t necessarily mean a sudden equivalent deterioration in corporate value; it may simply mean the market temporarily lacks funds willing to absorb the sell orders. However, I do not support the conclusion that Wall Street is currently replaying 1987. Today, the US stock market has circuit breakers, and trading systems, clearing mechanisms, and central bank tools to address liquidity crises have changed significantly. The 1987 portfolio insurance is not exactly the same as today’s options Gamma, CTA, and volatility control strategies, and without real position data, it’s inappropriate to directly conclude that large-scale forced liquidations are occurring. But we should indeed be cautious about the current crowded positions in global AI assets. Last week, the Nasdaq fell cumulatively by 4.5%, $SOXX closed lower for nine consecutive trading days, and South Korean memory, Japanese semiconductor equipment, and US tech stocks have all been hit. The market is reassessing the sustainability of AI capital expenditures, and some data center companies relying on external financing face greater credit pressure. What I think is most worth watching in this adjustment is not just whether $NVDA, $MU, or $SNDK’s future earnings have changed, but whether investors holding these stocks still have sufficient risk tolerance. If institutions continue to reduce positions due to rising volatility, even if fundamentals have not fundamentally deteriorated long-term, short-term stock prices may still face pressure. There is another detail many overlook about 1987: although the crash was extremely severe, it did not directly trigger a US economic recession. The Federal Reserve subsequently took measures to support financial system liquidity, and the Dow Jones recovered more than half of Black Monday’s point loss in the next two trading days. Within two years, the US stock market surpassed its previous highs again. This illustrates two points. First, extreme short-term stock market declines do not necessarily mean the long-term economic fundamentals have collapsed. Second, market recovery does not guarantee that investors who used high leverage during the crash will survive to see the recovery. I have always been a long-term AI bull and still tend to believe this October’s correction is a deep pullback within a bull market. But if history teaches us anything, it’s that being right about the long-term direction does not guarantee you can withstand short-term liquidity shocks. On Monday, I will focus on whether the Asia-Pacific market continues to transmit selling pressure to the US, whether $SOXX can end its continuous breakdown, and whether there is noticeable improvement in capital inflows to VIX, credit markets, and tech stocks. If the market suddenly rebounds sharply, I will not assume the risk is resolved based on just one day’s rise. Black Monday 39 years ago taught us that the truly dangerous moment is often not when everyone starts to turn bearish, but when a large number of investors suddenly realize they need to sell at the same time. I do not believe a stock market crash is destined to happen on October 19, 2026, nor do I think we should panic sell just because of a historical anniversary. But when global AI assets are continuously falling and liquidity in the tech sector is under pressure, controlling position sizes and holding cash is definitely not excessive. Because the harshest truth of the market is that stocks may eventually reach new all-time highs, but an investor forced to liquidate during a crash may never get the chance to participate on that day.

U.S. Treasury yields have started to drop sharply, so why are semiconductors still plunging consecutively? This might be one of the most alarming market signals since October. $SOXX $QQQ $NVDA $MU $SNDK $SPY Last week, the yield on the U.S. 10-year Treasury note fell from about 5.24% to around 5.10%, a weekly drop of about 14 basis points, marking one of the largest weekly declines since February. Meanwhile, the Nasdaq fell 4.5%, the Philadelphia Semiconductor Index dropped about 11%, and $SOXX recorded a rare nine-day losing streak not seen in nearly 20 years. This is quite unusual. According to traditional valuation logic, a decline in Treasury yields means less discounting pressure on future cash flows, which theoretically should support high-valuation growth stocks. Especially for AI-related companies like $NVDA and $AVGO, where the market has very high expectations for earnings growth over many years, interest rate changes usually have a significant impact on their valuations. So why did tech stocks fall even harder this time when Treasury yields dropped? Because a decline in Treasury yields does not necessarily mean that market risk has been eliminated. Stock performance is determined not only by the risk-free rate but also by future earnings expectations, equity risk premiums, financing conditions, and market liquidity. If the yield decline is due to easing inflation, sustained economic growth, and improved Federal Reserve policy expectations, it could indeed be favorable for tech stocks. But if the yield drop is because investors are starting to worry about economic growth and corporate credit risk, leading them to buy U.S. Treasuries as a safe haven, the situation is entirely different. I believe the recent market is reflecting the latter risk. Capital is beginning to reassess AI infrastructure financing models, and some data center companies that heavily rely on debt financing are facing tighter credit pricing. At the same time, Treasuries are being bought, while tech stocks continue to be sold off. There is also a frequently overlooked issue: a decline in U.S. Treasury yields does not mean that all companies’ financing costs will fall in tandem. A company’s actual borrowing cost depends not only on Treasury yields but also on its own credit spread. If the 10-year Treasury yield falls by 0.15 percentage points but the company’s credit spread widens by 1 percentage point, its real financing cost could actually rise significantly. This is especially important for companies relying on external financing to build AI data centers. These projects require substantial upfront capital, while future revenues may take years to materialize. If investors demand higher risk compensation, what once seemed like a reasonable investment return could be severely squeezed. What’s more troublesome is that this financing pressure could propagate along the entire AI industry chain. Rising financing costs for data center operators may slow future expansion; slower expansion plans could affect procurement expectations for servers, GPUs, storage, and networking equipment; declining procurement expectations could further compress valuations of upstream semiconductor companies. Note, I am talking about potential transmission risks, not that there is currently evidence of a comprehensive collapse in AI industry orders. But stock trading is often about future changes, not confirmed revenues today. This is why companies like TSMC and ASML report strong operational data, yet semiconductor stock prices continue to fall. Another set of data worth noting: last week, while the Nasdaq fell 4.5%, the Dow Jones rose 0.7%, and about 70% of S&P 500 components actually recorded gains. This shows the market is not indiscriminately selling all stocks. Sectors like healthcare and consumer staples showed relative strength, while the previously most crowded AI tech trades are under greater pressure. Therefore, I do not think a simple observation of falling Treasury yields should immediately prompt a full-on bottom-fishing in semiconductors. What I really want to see is three conditions improving simultaneously: Treasury yield declines no longer mainly driven by risk-off sentiment, credit risk for AI-related companies easing, and semiconductor companies’ earnings expectations and price structures stabilizing again. If oil prices fall, inflation pressures ease, Federal Reserve policy expectations improve, and large tech companies’ earnings reports prove AI investments can still generate reasonable returns, then these combined changes could provide a more solid foundation for the next sustained rally. I have always been a long-term AI bull and still tend to believe the October correction is a deep pullback within a bull market. But I won’t assume the tech stock correction is over just because Treasury yields have fallen from their highs. Especially now, some institutions are still reducing tech risk exposure, semiconductors continue to break down through support levels, and the market has not yet formed a sufficiently stable buying base. Even if there is a sharp short-term rebound, it may face profit-taking and selling pressure from trapped capital again. So my strategy remains to control overly high positions, keep cash on hand, and wait for further improvements in price structure and macro conditions. I think the most important thing to understand now is that rising Treasury yields and rising tech stocks are not always linked. When the market buys stocks because of economic improvement versus buying Treasuries out of risk concerns, the underlying capital logic is completely different. What Wall Street needs to solve now may no longer be just the problem of high interest rates, but how much risk investors are willing to take on for AI’s growth over the next few years. Whether the long-term bull market can continue ultimately depends on earnings realization, financing environment, and market risk appetite—not just how much the 10-year Treasury yield has fallen alone.

South Korean memory stocks finally stopped falling on Friday, but U.S. memory stocks continued to plunge that same night. $MU dropped 5.6% in a single day, and $SNDK plummeted 6.85%. Now the global memory market is showing a very dangerous phenomenon: Asia has just rebounded, while the U.S. has started a new round of sell-offs. $SOXX $SKHY $STX $WDC Let's first look at the market over the past few days. On October 15, South Korea's KOSPI plunged 7.57%, Samsung Electronics fell 9.48%, and SK Hynix dropped 11.06%. The next day, the Korean market finally rebounded, with KOSPI rising 2.05%, Samsung rebounding 2.62%, but SK Hynix only up 0.79%. However, by the U.S. market close on Friday, Micron and SanDisk were again heavily sold off. Even more dramatically, Yahoo Finance's statistics on October 16 showed that the Roundhill Memory ETF $DRAM has retraced more than 20% from its September rebound high, re-entering a technical bear market. Since October 6, the combined market value lost by Samsung, Micron, SK Hynix, and SanDisk has exceeded $700 billion. In just over ten trading days, the world's most important memory companies have simultaneously undergone such severe repricing, which is clearly more than an ordinary sector adjustment. But I believe two questions must be clarified now: Why are memory stocks falling more severely than many chip companies? Is this decline due to a fundamental deterioration in the memory industry, or is it simply the unwinding of previously overcrowded long positions? First, the fundamentals. $MU, SK Hynix, and Samsung are all affected by DRAM and HBM demand, but their business structures are not identical. SK Hynix has strong competitiveness in the HBM market, Micron is continuously expanding its high-bandwidth memory business, and Samsung has a broader semiconductor and end-product portfolio. Meanwhile, $SNDK mainly operates NAND flash memory, which is fundamentally different from HBM. $STX and $WDC primarily involve mechanical hard drive businesses. They all belong to the broad data storage industry chain, but just because their stock prices fall together doesn't mean all companies share the exact same profit logic. This is why I think the recent market sell-off in memory stocks may involve a certain degree of indiscriminate risk-off. Currently, the long-term demand for high-performance memory and large-capacity storage in AI data centers has not suddenly disappeared. Some analysts in South Korea even believe this round of memory stock declines has clearly exceeded changes in corporate earnings expectations. On the other hand, the memory industry itself is a typical cyclical sector. Rising DRAM and NAND prices quickly improve profit margins, but once the market starts worrying about future supply expansion, price cycle reversals, or a slowdown in AI capital expenditure growth, valuations can also contract rapidly. Especially for stocks that have risen significantly before, the market has already priced in years of future earnings growth. Even if this year's performance remains strong, as soon as investors start doubting earnings growth in 2027, stock prices may face huge pressure. Another repeatedly discussed issue is the technological progress of China's ChangXin Memory. But it is necessary to distinguish the technical and commercialization differences between ordinary DRAM and HBM. Even if Chinese manufacturers make progress in some DRAM products, it does not mean they can immediately replace the world's advanced HBM supply, nor can it directly explain $SNDK's NAND business. Therefore, I do not agree with immediately assuming a fundamental change in the entire industry's long-term competitive landscape every time a new memory news item appears. Looking ahead to Monday, my main focus is whether the Korean market can truly stop falling. Because the rebound of Samsung and SK Hynix on Friday was far from recovering the previous day's losses, while Micron and SanDisk experienced a new round of declines afterward. If Korean memory stocks open sharply lower again on Monday and continue to be sold off by foreign investors during the session, the risk release of global memory assets may still not be over. But if the Korean market can absorb the U.S. losses from Friday, and Samsung and SK Hynix do not hit new lows and even start to see sustained buying, then the U.S. memory sector may also gain some emotional support. However, Asia stabilizing first does not guarantee U.S. stocks will rise the same day. I have long been optimistic about the memory industry, especially the demand for HBM, DRAM, and enterprise storage driven by AI infrastructure. But I have also repeatedly emphasized that being bullish on the industry cycle does not mean one must endure the entire valuation cycle's decline. After the severe adjustments in March and July, I clearly understand how fast memory stocks can rise once they rebound, but precisely because of this, I do not believe October will necessarily repeat the previous rallies. Now I prefer to study the actual price changes of DRAM and NAND, corporate earnings expectations, inventory levels, and whether global memory stocks are beginning to stabilize synchronously, rather than guessing the bottom daily based on the latest news. If the market continues to break down, it is still necessary to reduce positions when appropriate. If a truly worthwhile opportunity arises later, having cash on hand allows rebuilding positions. I still tend to believe this is a deep pullback within the AI bull market. But I will not assume there is no risk now just because I believe in memory demand growth over the next few years. For long-term investors, the most worthwhile thing to wait for is not a sudden 10% rise one day, but a reasonable relationship reestablished between stock prices and corporate earnings expectations. Global memory stocks may eventually re-enter an uptrend, but until the market proves that selling pressure has truly eased, there is no need to rush to bet the principal on the bottom.

$LRCX has declined for 10 consecutive trading days, with a cumulative drawdown of about 22%, $KLAC fell about 17% during the same period, and $AMAT also dropped about 22% within 9 trading days. It is difficult to simply explain such a level of sell-off among leading advanced semiconductor equipment companies as normal profit-taking. On October 21, $LRCX is about to release its new quarterly earnings report, which I believe could be an important basis for judging whether the semiconductor equipment sector can stabilize. $SOXX $QQQ Let's first look at the fundamentals. In the last quarter, $LRCX's revenue reached $6.722 billion, a quarter-on-quarter increase of 15.1%, with a Non-GAAP gross margin of 52% and an operating margin of 38.4%. The company previously guided this quarter's revenue to $8.1 billion, with a fluctuation range of $400 million, indicating that management originally expected the business to continue growing significantly. From these data, at least up to the last quarter, demand for semiconductor manufacturing equipment has not deteriorated comprehensively. $LRCX's competitive advantages focus on etching, deposition, and wafer manufacturing-related technologies. The more complex the advanced process, the more precise the transistor structure, and the higher the requirements for high-precision etching and deposition equipment in the manufacturing process. Especially with the increase in 3D NAND layers, upgrades in DRAM manufacturing technology, and continuous development of advanced packaging, there are opportunities to drive equipment demand growth. This is why I am long-term bullish on companies like $LRCX, $KLAC, and $ASML, which have high technological barriers. But now Wall Street is worried about 2027. If the financing environment for AI data centers continues to deteriorate and downstream customers slow capital expenditures, chip manufacturers may delay some expansion projects. Equipment companies are upstream in the capital expenditure transmission chain; once future order growth expectations are downgraded, even if the current quarter's performance is good, stock prices may continue to be under pressure. Therefore, this earnings report needs to focus on three key questions: whether this quarter's revenue can reach near the previous guidance midpoint of $8.1 billion, whether the gross margin can be maintained around 52%, and whether management's judgment on wafer manufacturing equipment demand in 2027 has changed. Particular attention should be paid to whether investment plans for DRAM, NAND, and advanced processes have been delayed, as well as the impact of demand in the Chinese market and export restrictions. Simply exceeding EPS expectations is no longer enough to dispel market concerns. Personally, I believe that if $LRCX maintains strong growth in this earnings report and management continues to release positive demand signals for 2027, then the valuation compression caused by the previous consecutive sharp declines may see a phased recovery. If companies like $KLAC and $AMAT also stabilize simultaneously, the credibility of the entire equipment sector reestablishing an upward structure will also increase. Conversely, if the earnings report exceeds expectations but the 2027 growth guidance weakens, or the market continues to fall sharply after the positive announcement, then this round of adjustment may not be over yet. I still believe that advanced semiconductor manufacturing equipment is one of the most worthwhile long-term holdings in the AI industry chain, especially optimistic about the technological barriers of $ASML, $KLAC, and $LRCX. But no matter how good the industry outlook is, valuation and profit cycles must be considered. The earnings report on October 21 may provide important clues, but a single report cannot confirm the entire sector's reversal alone. Only when earnings expectations stabilize, stock prices stop hitting new lows, and capital flows back in, will there be more sufficient reasons to judge that the semiconductor equipment sector is emerging from this extreme adjustment.

I'm actually a bit worried about a sudden surge in the US stock market on Monday. $SOXX $QQQ $NVDA $MU $SNDK It's not because I don't want the market to rise, but after experiencing such extreme declines consecutively, once a strong rebound occurs, many people might completely abandon risk management. $SOXX has fallen for 9 consecutive trading days, setting a rare record of continuous decline in nearly 20 years. The Nasdaq dropped 4.5% last week, with semiconductors, storage, optical communications, and software stocks all facing sell-offs. The price structure of the entire tech sector has clearly deteriorated, but the market has also accumulated a considerable degree of short-term oversold pressure. If on Monday South Korea, Japan, and Taiwan continue to be under pressure, but the US stock market suddenly opens low and then rallies, even with $SOXX rebounding 3% to 5% in a single day, I wouldn't find it strange at all. After continuous declines, short sellers taking profits, institutions readjusting risk exposure, and short-term funds chasing rebounds could all drive the market to rise quickly. Another factor that cannot be ignored is options hedging. When market makers are net negative Gamma, price increases may trigger trend-following buy hedges, further amplifying gains. But actual Gamma exposure cannot be fully confirmed just by public holdings, so not every sharp rise should be attributed to market makers being forced to buy. The problem is that a sharp rise and a true bottom are two different things. Having experienced the adjustments in March and July, I am very clear on how fast prices can rebound at the end of extreme market conditions. But precisely because of this, I won't simply assume that October will repeat the patterns of the previous two times. The market environment is different; interest rates, oil prices, credit risk, and AI capital expenditure expectations have all changed. We especially need to be wary of a scenario where the US stock market opens sharply lower on Monday, then quickly rallies, with stocks like $NVDA, $MU, and $SNDK collectively rebounding, and immediately on X there are many voices about a bull market return, short squeeze, and golden pit confirmation. Then the rebound fails to continue the next day, previous support levels become new resistance, and prices fall below the lows again. This is not uncommon in a downtrend. Because funds previously trapped may use the rebound to reduce positions, while genuinely willing new buyers may not have returned yet. So on Monday, I will focus on three questions. First, is the rebound driven only by a few heavyweight stocks pulling the index, or are sectors like semiconductors, software, and cloud computing all starting to strengthen synchronously? If $QQQ rises but most tech stocks remain weak, the foundation of the rebound is questionable. Second, can $SOXX regain and hold the key areas it previously broke below? Once prior support is lost, it often becomes resistance that must be retaken during a rebound. Only sustained recovery, rather than intraday rallies followed by pullbacks, is worth attention. Third, can the rebound gain confirmation in subsequent trading days? A 5% rise on Monday is certainly exciting, but if Tuesday and Wednesday see consecutive declines, the market still hasn't escaped its prior weak structure. I have always been a long-term AI bull and still tend to believe October is a deep pullback within a bull market. But being long-term bullish doesn't mean every rebound is worth chasing, nor does it mean one should bet on a market bottom with short-term options and leverage. I've been repeatedly emphasizing reducing positions recently because the risk-reward structure has changed. For those with heavy positions, if the rebound lacks sustained buying, I am more inclined to use the rise to reduce risk rather than buy back all previously reduced positions just because of a big green candle. For those with ample cash, there's no need to worry about missing the first day of a market rebound. If a new upward trend truly forms, it usually won't give investors only one trading day to act. What I least want to see now is people panic-selling during continuous declines, then chasing the rebound full throttle out of fear of missing out. The experiences in March and July tell us that pullbacks in a bull market can be very fierce, and rebounds can be very rapid. But historical experience is not a guarantee that October will necessarily rise. I still believe the foundation for AI's long-term growth hasn't disappeared, and I believe quality semiconductor companies have opportunities to rise again in the future. But right now, I'd rather miss out on a technical rebound than take on unbearable position risk trying to guess the bottom again. If there really is a surge on Monday, I'll be happy, but I won't change my risk management principles just because of one day's rise. What would truly make me increase my position again is not a big green candle, but the market starting to consistently prove that prior selling pressure is easing, new buying can absorb selling, and the downtrend is changing.

What is currently most frightening for the global semiconductor industry may not be negative news from any single company, but that South Korea, Japan, Taiwan, and the United States are all enduring the same round of capital sell-offs. South Korea falls, then the U.S. falls; after the U.S. falls, South Korea has to digest new selling pressure again. The entire AI industry chain seems to be forming a continuously transmitted negative feedback loop. $SOXX $QQQ $NVDA $MU $SNDK $TSM Look at how extreme last week's market was. On October 15, South Korea's KOSPI plummeted 7.57%, with SK Hynix and Samsung facing fierce sell-offs, and Japan's semiconductor sector was also under pressure. On Friday, the South Korean market finally rebounded 2.05%, but U.S. tech stocks continued to decline, with $SOXX plunging about 4.1% in a single day, marking a rare nine-day losing streak not seen in nearly 20 years. Asia barely caught a breath before the U.S. passed the pressure back. Why do I believe this round of declines deserves high vigilance? Because although the global AI industry chain is distributed across different countries, the underlying profit expectations and capital risks are highly correlated. South Korea's SK Hynix and Samsung supply memory chips, Taiwan's TSMC handles advanced wafer manufacturing, Japan's Tokyo Electron and others provide semiconductor equipment, and the U.S.'s $NVDA, $AMD, and $AVGO occupy the core segments of chip design and computing power demand. On the surface, these companies are in different markets, but in reality, many institutions hold the same AI growth logic. When the market starts doubting whether AI capital expenditures will continue in the coming years, institutions may not just sell one stock but might simultaneously reduce risk exposure across the entire industry chain. Here lies the problem. A decline in U.S. chip stocks may affect Asian investors' expectations for memory, equipment, and foundry sectors; concentrated sell-offs in Asian markets may further impact U.S. pre-market risk appetite. If combined with ETF redemptions, leveraged liquidations, foreign capital withdrawals, and exchange rate fluctuations, what would normally be a routine valuation adjustment could be amplified into a cross-market continuous decline. But I must emphasize, there is currently insufficient evidence to prove that all market sell-offs are passive liquidations, nor should every decline be attributed to malicious institutional short-selling. What we can observe is high price correlation, declining risk appetite, and some institutions reducing previously crowded AI positions. More interestingly, corporate fundamentals have not deteriorated to the same extent within just two weeks. TSMC and ASML recently still delivered strong operational data, and the long-term demand for advanced processes, HBM, and AI infrastructure has not suddenly vanished. But the market is now trading not just this year's earnings but the growth rate over the next few years and how much investors are willing to pay for that growth in valuation. This is also why I have always believed that trading based on news is easy to fall into misconceptions. During the March crash, there were many negative factors; similarly, in July's crash, there were also many negatives; now in October, various explanations are emerging again. News can explain why the market panics but rarely tells you when that panic will end. Ultimately, what determines whether a rebound can sustain is capital support, liquidity, valuation, and earnings expectations. I have always been a long-term AI bull and tend to believe that this still belongs to a deep correction within a bull market. But I will not assume that short-term further declines are impossible just because I am optimistic about AI long-term. Especially now that the global semiconductor industry has formed a highly synchronized weak structure, any market rebound may be affected by continued pressure in other markets. What I most want to see on Monday is not South Korea suddenly surging 5%, nor $SOXX opening with a 4% jump. I hope to see South Korean memory stocks stop hitting new lows, Taiwan and Japan's semiconductor sectors gradually stabilize, and U.S. tech stocks show sustained buying rather than every rebound being swallowed by new selling pressure. If only the U.S. rebounds while Asia continues to plunge; or if Asia just stabilizes and the U.S. breaks down again, then the global AI asset de-risking process may still not be over. So my strategy remains unchanged. If your position is too heavy, actively reduce risk; be especially cautious with short-term options and leverage; those with ample cash should patiently wait for truly competitive companies to fall to reasonable valuations. I believe the long-term AI bull market still has a foundation to continue, but what the market needs now is not more people shouting "buy," but real capital willing to keep buying. When will the global semiconductor industry end this round of continuous sell-offs? I think the most important signal to watch is when Asia and the U.S. no longer transmit new selling pressure to each other but begin to form synchronized and sustained capital support. Until then, don't mistake a sharp rebound for the entire market being safe.

If the Asia-Pacific markets collectively plunge again on Monday, can the U.S. stock market still hold up? I believe the next 24 hours could be one of the most important trading windows since October. $SOXX $QQQ $NVDA $MU $TSM $SNDK Last week, the global semiconductor sector showed extremely volatile performance. On October 15, South Korea's KOSPI plunged about 7.6%, Japan's Nikkei index dropped nearly 5%, and major memory giants like Samsung and SK Hynix faced intense sell-offs. Although the Asia-Pacific markets rebounded somewhat on Friday, U.S. tech stocks continued to decline, with the Nasdaq falling 4.5% for the week, and $SOXX recording a rare nine-day losing streak not seen in nearly 20 years. This means that when Asian markets open on Monday, they will need to digest the selling pressure from the U.S. on Friday again. But I don't think it’s as simple as South Korea falling means the U.S. will definitely fall, or the U.S. falling means South Korea will definitely fall afterward. Currently, the global AI industry chain holdings are highly correlated. South Korean memory, Japanese equipment, Taiwanese foundry, and U.S. chip companies are all affected by the same round of AI risk appetite changes. If institutions collectively reduce risk exposure, it can easily create cross-market negative feedback. But if Asia-Pacific continues to fall on Monday, we must distinguish whether new risks have emerged or if it’s just a catch-up drop from the U.S. sell-off on Friday. I mainly consider three scenarios. First, Asia-Pacific continues to plunge, U.S. futures weaken simultaneously, and $SOXX breaks down further after the open. This would mean the global AI sell-off pressure may not have fully released yet. If the VIX rises in tandem and tech market breadth continues to deteriorate, I would not rush to bottom-fish. Second, Asia-Pacific plunges, but U.S. stocks open lower then rally strongly, even rebounding 3% to 5%. This is entirely possible. After consecutive extreme drops, short sellers taking profits, short-term funds entering, and options hedging adjustments can all drive a rapid market rebound. However, I believe this kind of rebound requires even more sober judgment. Technical rebounds during a downtrend can sometimes be more intense than rallies in a normal bull market. If the rebound lacks sufficient volume and market breadth support, and previous support levels remain unrecovered, a big rally does not necessarily mean the correction is over. Third, Asia-Pacific stabilizes first, Samsung, SK Hynix, and TSMC-related assets stop hitting new lows, and U.S. semiconductor stocks begin to see broad buying. This is the scenario I most want to see. Especially if $SOXX can regain key levels and the rebound is no longer quickly swallowed by selling pressure, the market may gradually form a more reliable bottom structure. Why have I been emphasizing reducing positions recently? Because the issue now is not just how much semiconductors have fallen, but that the entire tech sector’s price structure is deteriorating. $SOXX keeps breaking down, software stocks are also catching down moves, and rebounds from the previous day are often fully erased the next day. It’s not that there is no bottom-fishing capital, but these funds have not yet shown strong and sustained buying power. Moreover, this round of correction has already shown clear signs of existing long holders reducing risk. Data previously disclosed by Goldman Sachs Prime Brokerage indicates that hedge funds’ selling pressure on tech stocks has persisted recently. But this does not mean shorts are not active, nor can we judge all institutional real positions solely by candlestick and options data. I have always believed the short-term risk to watch most closely is concentrated selling pressure after rebounds. Especially in an environment of negative Gamma risk, leveraged liquidations, and insufficient liquidity, price fluctuations can be amplified both up and down. I have been a long-term AI bull and still tend to believe October is a deep pullback within a bull market, not a sudden failure of the entire AI industry logic. But being bullish does not mean you must be fully invested, nor does it mean every plunge should be met with immediate buying. The market will not guarantee you won’t lose 30% or more in the short term just because you are right about the long-term direction. So on Monday, I will focus on three questions: whether selling pressure in Asia-Pacific further spreads, whether U.S. stocks can show sustained buying independent of Asia, and whether $SOXX’s rebound can hold through the close and be confirmed in the following trading days. If Asia falls, then the U.S. falls, then the U.S. rebounds but Asia falls again, the global AI asset de-risking process may not be over yet. The market has already fallen extremely far, so a sharp short-term rebound is not surprising. But I won’t assume all problems are solved just because of one big green candle. I still believe the foundation for a long-term AI bull market continuation exists and that this plunge may ultimately create long-term investment opportunities in quality companies. But the golden bottom may not be today; what’s truly worth waiting for is simultaneous improvement in valuation, capital, and price structure. Until the trend improves, reduce positions when needed, keep cash when needed. Short-term risk control and long-term AI optimism are not contradictory. The bull market can continue, but the premise is that you have enough capital to participate in the next real uptrend.

$100 trillion! Wall Street may be re-evaluating the largest infrastructure investment plan in AI history. But I believe the market has previously overlooked a very practical issue: how many of the data centers announced for construction in the coming years will actually be completed? $NVDA $AVGO $CRWV $VRT $MSFT $QQQ On October 18, the Financial Times published a major investigation pointing out that the expansion of AI infrastructure in the U.S. is encountering increasing real-world obstacles. First, let's clarify a concept: the $100 trillion is not already signed orders, nor is it money already paid by tech giants, but an estimate by research institutions of the AI infrastructure investment demand in the U.S. over the next several years. A study from Columbia Business School estimates that by 2032, AI construction in the U.S. may require investments exceeding $100 trillion, equivalent to about 3.6% of the U.S. GDP annually. How exaggerated is this scale? According to this study's estimate, the economic scale of AI infrastructure construction relative to GDP may even surpass the historical expansions of railroads, electrification, and internet infrastructure in the U.S. But massive investment plans and actual construction completion are completely different matters. There is also a very noteworthy news today. According to British media reports, $MSFT has withdrawn from a data center collaboration with AI cloud service provider Nscale in Loughton, Essex County, UK. The report points out that the project has been delayed multiple times, and the local power grid even expects that adequate power supply will not be available until the 2030s. Microsoft has not publicly confirmed the specific reasons for withdrawal, while Nscale stated it will continue to advance the project and seek alternative power supply solutions. But this incident itself already illustrates a problem. You can announce today the construction of a multi-billion-dollar data center and even sign long-term computing power contracts in advance, but you cannot rely on a business plan to make the local power grid have sufficient power capacity the next day. This is the risk of AI infrastructure construction that is most easily overlooked. In recent years, the market has mainly valued around GPUs, HBM, advanced packaging, optical modules, and server orders. But a large AI data center from planning to actual production must at least solve issues such as land, approvals, power access, equipment supply, construction costs, and project financing. Any delay in any of these links may affect the speed of fulfilling the entire investment plan. More importantly, these links are not independent of each other. If power supply is delayed by a year, the data center may not be able to go into production as originally planned. If the data center cannot go into production, the customer's computing power rental income may be delayed in recognition. Revenue delays, but financing costs may continue to accumulate. If the project still relies on debt financing, the originally reasonable investment return rate may significantly decline. Ultimately, a long-term contract that looks very attractive may not be able to convert into free cash flow within the expected timeframe. This is also why the AI data center financing market has recently started to show pressure. Reuters reported on October 15 that AI infrastructure-related companies such as AirTrunk, Firmus, and SB Energy have delayed or withdrawn their IPO or financing plans. Some investors have begun to more strictly scrutinize valuations, leverage, contract quality, and project execution risks. But this does not mean AI investment has completely stalled. During the same period, some data center mergers and acquisitions and debt financing activities remain active. I believe the market is shifting from simply believing in growth scale to studying whether this growth can truly be realized. So what does this mean for semiconductors? The impact may be more complex than many imagine. For upstream chip suppliers like $NVDA and $AVGO, continuous data center expansion can bring huge demand for computing and networking equipment. For power and cooling infrastructure suppliers like $VRT, data center construction also means long-term business opportunities. But if some projects are delayed and financing conditions tighten, customers may reschedule equipment procurement. Long-term industry demand has not disappeared, but that does not mean every quarter's orders can grow according to the most optimistic forecasts. Especially when stock prices have already priced in demand years ahead, even if actual orders only slow in growth, it may trigger significant valuation compression. This is exactly why I think the October tech stock adjustment deserves serious study. I have always been a long-term AI bull and have not changed my judgment on advanced processes, storage, and AI infrastructure long-term development. But I will not take every announced multi-billion-dollar investment plan as guaranteed future revenue. Going forward, I pay more attention to actual project start rates, power access progress, secured financing, irrevocable customer contracts, and how much cash flow projects can generate after production. For the market, the real risk may not be that AI is unused, but that the entire industry expanded too quickly before, and some projects' construction speed and commercial returns cannot keep up with capital market expectations. Both situations can have equally significant impacts on stock valuations. Therefore, I still believe that we cannot ignore short-term valuation and financing risks just because the AI industry looks promising long-term. When it's time to reduce positions, reduce them; cash and risk tolerance must be preserved. If in the coming years, those companies that truly complete construction, maintain high computing power utilization, and continuously generate free cash flow begin to stand out, I still believe the AI industry will present very valuable long-term investment opportunities. But the market may no longer be willing to pay a high premium for every construction plan that has not yet materialized. What the AI industry needs to prove in the future is not just how many data centers can be built, but whether these data centers can be built on time, put into use, and truly make money.