Hosts: Mr. Z ( @168MrZ ), Victor ( @vcmktasa )

Guests: 3X Long Labubu ( @labubu_trader ), AI Industry Excavator ( @QihongF44102 )

Crowded trading, deleveraging, and supply bottlenecks mean that semiconductors remain the only AI play in the public market.

In early June 2026, the AI ​​stock market, which had risen for two consecutive months, suddenly plummeted, with the S&P 500 and NDX indexes testing key moving averages. Multiple variables triggered a simultaneous collapse: Federal Reserve Chairman Kevin Warsh's first FOMC meeting, concerns about interest rate hikes driven by Middle Eastern oil prices, the highly anticipated Micron earnings report, and worries about potential capital flight from SpaceX's IPO. A report from research firm SemiAnalysis questioning the CPO mass production timeline further devastated the optical communications sector. Simultaneously, memory and MLCC manufacturers raised prices, resulting in a simultaneous sell-off and a supercycle.

This issue of 168X features two highly recognizable AI stock and semiconductor investors on both English and Chinese Twitter. One is 3X Long Labubu ( @labubu_trader , Teacher Huang), a Silicon Valley AI engineer known for his expertise in market structure, leverage, and options fund flows; the other is AI Industry Excavator ( @QihongF44102 ), a seasoned AI professional who has worked his way up from computer vision and search recommendation to agentic system development, adept at using first principles and supply and demand cycles to uncover the real bottlenecks that are "exploding" behind AI infrastructure. Both share a clear consensus: before CapEx ceases operations, or before OpenAI and Anthropic go public, semiconductors are practically the only AI play in the public market. The question isn't whether to buy the dip, but when to do so.

I. Silicon Valley Engineers and AI Practitioners: One Analyzes Market Trends and Leverage, the Other Uses First Principles to Identify Bottlenecks

Mr. Z: First, could you two teachers please briefly introduce yourselves? Labubu, how did you approach the AI ​​and semiconductor market from the perspectives of trading and technical analysis?

Labubu: Hello everyone. My day job is as an engineer in Silicon Valley, and the industry I work in is also related to AI. I've done some AI-related work, like inference and training, but I've never worked in the semiconductor industry. So when I first entered this market, I mainly focused on the fundamentals: as someone working in the field, I usually use my spare time to research these things. For example, I'm currently using GPUs and ASICs, so what exactly are they? In addition, I've communicated with many friends in the industry chain, such as those in excavator manufacturing, and many friends in Asia (Taiwan, Hong Kong, and mainland China) to understand industry information.

After a few years, I discovered something: even if something has solid fundamentals, it will still exhibit fluctuations in terms of capital flows and technical indicators. Our advantage as retail investors lies mainly in our flexibility; we can leverage these fluctuations to make short-term profits. So, I started as a fundamental investor, gradually learned how to do short-term trading, and now I prefer a strategy that combines fundamental and technical analysis.

Mr. Z: What about the excavator teacher? You also explored supply chain opportunities such as MLCC and storage from the perspective of AI technology and industry chain.

Excavator: I'm also a practitioner in the AI ​​industry, with a similar professional background to Labubu, but he's more focused on the underlying layers, while I'm more application-oriented. I studied computer vision in school, and after starting work, I've been working on search and recommendation systems. When Agentic AI emerged, I immediately switched to this area, and now I'm involved in the design of Agentic systems, specifically the design of intelligent agents like Codex and Claude Code.

How did I discover this industry chain? It's very closely related to my work. In the early days, everyone was focused on training, so you naturally paid attention to NVIDIA, optical modules, and CPUs, because this area had the biggest limitations on GPUs and interconnects. Gradually, with the emergence of DeepSeek and the rise of inference, the focus shifted from training to inference. You saw Cursor, Claude Code, Codex, and especially AI Agents, which developed very quickly. I saw a statistic: the volume of inference was expanding rapidly. What changes did this rapid expansion bring? The biggest changes were in the demand for CPUs and memory, starting from August or September of last year. So, a major reason why I heavily invested in storage and NAND Flash starting last August or September is this.

The recent changes are: almost everything in the supply chain is rising in price. You'll gradually see even analog circuits starting to increase, and power semiconductors related to power supplies are also rising. Then, reports have surfaced that MLCCs and passive components (capacitors, inductors, resistors) are all increasing in price. The passive component market isn't actually that big; if you use Codex or Claude Code, you can basically see that the leading companies are just a few. Comparing this cycle to the one in 2017-2018, you can easily see that this cycle is clearly larger; it's just a matter of timing.

This is a key point I want to emphasize today: when everything is rising in price, CSPs' (cloud service providers) annual CapEx (capital expenditure) figures are increasing, but because of the price increases, their "actual" CapEx may already be declining, which is a risk. So how do we deal with this? One approach is to be cautious about semiconductor-related stocks at this time. Another consideration is that when everything is rising in price, other parts of the industry chain will feel the pressure, and CSPs will feel cash flow pressure. The US and Chinese governments may then require everyone to expand production, because only by expanding production can upstream prices come down, allowing CSPs to make the most of their resources. Therefore, if there is an adjustment in the semiconductor sector, I think the more important focus will shift to semiconductor equipment and materials.

II. How to interpret the June pullback: Bull Trap, early selling, and "false risk vs. real risk"

Victor: First, I'd like to ask Labubu, I've been following your posts for a long time. Around Monday, you mentioned that if the S&P 500 fell below 7,400, it would lead to a further decline to 7,200-7,260. On Tuesday, the market did indeed open higher to lure in more buyers, then dropped to a low of 7,237, before recovering at the close with a large candlestick. Yesterday's CPI figures met expectations. What's your view on the market now?

Labubu: Here's how I see it. Actually, I started saying around May 23rd, at the end of May, that there might be a big bull trap in the next two weeks, catching the bulls in, and then there would be a significant pullback. But I didn't expect it to happen so quickly.

Let me first explain why I made this judgment at the time. June had many events: Geographically, the Strait of Hormuz wasn't fully open, while many refineries in Europe and Asia had already reached their inventory limits; this strait supplies roughly 10-20 percent of global crude oil (I don't remember the exact figure), and if it remained closed when inventories bottomed out in June, it would lead to a surge in oil prices. Macroeconomically, the oil price shock would push up CPI inflation, triggering fears of inflation; if inflation continued to rise, the Federal Reserve would interrupt its rate-cutting pace, and market traders were already pricing in a rate hike before the end of the year. All of these factors were a significant drag on risky assets like the stock market.

Meanwhile, there's Kevin Warsh's FOMC debut next week. Because he's rarely appeared in public before, there's a lot of speculation about him: he might be a very hawkish figure, only concerned with inflation, constantly thinking about quantitative tightening, and rumored to be different from his predecessor, less communicative and less transparent. The market is particularly sensitive to the Fed chair's "debut": the last chair's debut (in 2018) caused significant losses for the market, so everyone is especially wary of Warsh this time. There's also the SpaceX IPO, one of the largest potential listings in the US stock market in recent years; will it impact liquidity? All these risks are present.

Based on our past experience, when there is uncertainty, the market always rushes to react to the event, even if it turns out to be a non-event. It might turn out that SpaceX didn't withdraw liquidity, Walsh wasn't as hawkish as expected, or the CPI wasn't as hot as anticipated, but the market will still hedge against risk due to this uncertainty. As retail investors, we need to be even more proactive than institutional investors. This is why we said at the end of May that we might conduct a large-scale reduction in holdings and lower leverage in the first week of June to mitigate risk.

In fact, as everyone has seen, there was a significant downward revision starting last Wednesday and Friday, and the market is still in a correction phase today. Actually, today's CPI data is very good: last month's core CPI month-over-month increase was 0.4%, this time it's 0.2%, which actually suggests that the oil price shock may be one-off and may have already passed its peak. So, in my opinion, this CPI data is very good. But even so, the market is still hesitant. Today it rebounded sharply, but then it was pushed down again because Trump actually went to war with Iran. This shows that the bulls are quite hesitant and will wait until everything is clear, especially the FOMC, before seeing a clearer path.

However, I personally remain quite bullish. We've risen for eight or nine weeks since the lows of March and April, forming a very steep trend. Falling into a dip in the middle doesn't mean anything, because we've fallen into similar dips before: April 19, 2024; early August 2024; October 10, 2025—similar events, but it was still within a bullish trend. It's just that due to sentiment or crowding of funds, especially the leveraged crowding in the semiconductor sector, we'll experience a rapid sell-off. So the question isn't whether to buy the dip, but when to buy the dip. I personally don't use much leverage, and I'm currently considering which entry point would be best: for example, if there's a bottoming signal around the 50-day moving average, I might try buying first, and then add to my position when it recovers some key moving averages or support levels.

III. Moving Averages are Just Consensus: Technical Trading Framework and the "Scenario + Probability" Betting Method

Victor: You previously posted a risk checklist for June, listing events you considered "false risks" and "real risks," such as the SpaceX IPO draining liquidity, Walsh potentially using QT (quantitative tightening), and the escalation of conflicts in the Middle East. You also mentioned that the real risk is "if these funds believe in the false risks and sell," the market will still fall.

Labubu: Yes, let's clarify this: the key point isn't whether he believes it's a risk, but rather that he thinks it's uncertain. What's interesting about the market is that it can price in both good and bad news; but if it's uncertain, something that can go either way, that's the situation the market dislikes most.

Are many people, especially those on Wall Street, less intelligent than me? Impossible. Do they have less information than me? Also impossible. But they see it as uncertainty and won't take the risk. So, when timing the market, you shouldn't be thinking about "your" thoughts, but rather "his" thoughts. Personally, I don't see these as risks. I was an early investor in SpaceX, and when I talked to them, I felt there were no such risks; but the market perceives them as risky.

So this is a bit of a split personality: you're an average investor, but you're also trying to imagine yourself as a fund manager in the market, and then trying to figure out what a less intelligent person would choose? And that choice is usually what most people would choose. As retail investors, we can only follow the flow. Therefore, at times of concentrated risk, we should appropriately reduce our positions or add some hedges to protect our profits. I think everyone here should have made a fortune this year; those who can protect their profits during pullbacks will actually earn even more later.

Victor: I'm curious how Labubu developed this trading framework? I've noticed that you frequently use indicators like the 21-day and 50-day moving averages, and you also look at option fund flows.

Labubu: Essentially, these moving averages are all based on consensus. For example, the 21-day moving average can be understood as the average holding cost of all traders over the past 21 days. I like using it, not because these indicators themselves are magical, but because everyone uses them. When everyone uses the same type of indicator, it becomes a consensus. This is why we see amazing phenomena: prices precisely reverse at the 21-day or 50-day moving average, or once they break below the 50-day moving average, they continue downwards without looking back. It's not a golden standard; it's just a market consensus, a statistical analysis based on past experience. For example, in the smooth, flat upward trends of the past 10 years, you'll find that the market has never closed below the 21-day moving average.

So the tools I use are the same as everyone else's – the more common, the better. Avoid anything fancy; they simply help me find entry and exit points: moving averages, gaps, highs and lows, and some key levels. For example, why is 7,330 important? Because around May 19th, there was a lot of trading activity in that area, making it a generally accepted fair price. As a technical trader, you need to memorize these levels. How do I train this? By trading futures, because futures are purely about price points. An ES point is worth 50 dollars, and a NQ point is worth 20 dollars, which gives you a more lasting impression.

I learned that the market inexplicably surged at the opening on Tuesday, reaching Monday's high of 7,480. This didn't make sense to me, so I liquidated all my leveraged positions. And sure enough, it didn't make sense either, and the market immediately fell on Tuesday. This is just a general consensus: in a market with high sentiment, it would easily break through highs, hold, and continue upward; but this wasn't the case. When people reached that high point, they panicked and fled, thinking, "I've finally broken even," and rushed to get out.

Victor: Labubu is truly exceptional at grasping market timing. Instead of predicting the market's ultimate direction, she analyzes various scenarios and makes the best choice in the present moment. How do you train yourself psychologically to do this?

Labubu: Anyone who tells you they can precisely time the market is a fraud, because nobody can do it. What we can do is: anticipate many scenarios, estimate the probability of each, and then decide on betting and position sizing based on those probabilities. This time, I discussed this with some friends (like the well-known @qinbafrank on Twitter), and we felt we must be faster than the market because we've suffered too many losses in the past: every time we waited until the event actually happened, it was just a relief rally after all the bad news had been priced in. So it's not that I'm particularly good, but that we've suffered too many losses like this before; anyone who has survived in this market and experienced many similar events probably does better than me.

Simply making scenario predictions won't guide your betting decisions, so you need to estimate the probability for each case. What about black swan events, like the US and China dropping nuclear missiles tomorrow? Is it possible? It's possible, but the probability is too low; we won't price it. We only prepare contingency plans for the most likely scenarios, and then decide on position size, which assets to buy, and what hedges to take based on the probability. For example, right now, Wednesday's price might be the final low of this pullback, but there's also a possibility of a major event later that could push it down to the 50-day moving average. I might want to buy some things, like Lumentum (LITE) or AAOI, and I think the price is good, but I'll control my position size because I know there's another path.

IV. What to watch next: FOMC, Micron's financial reports, and the deleveraging risk of "crowded trades"

Victor: At this point in time (Taipei time, June 11, daytime), what events or indicators will you continue to monitor to assess the market?

Labubu: The most important event in the short term is next week's FOMC meeting. There's another event that many people think is unimportant, but it's actually very important: Micron (MU)'s earnings report.

Why is it important? Because everyone now knows that storage is something "must buy." Ask any hedge fund, and almost every one will have some storage in their portfolio—Micron, SanDisk (SNDK), WDC, and so on. Ask any emerging market fund, and they'll almost certainly have Samsung or SK Hynix, or both, otherwise their performance won't beat the beta benchmark. And their positions are all very large. This is very similar to 2024, when everyone was buying Nvidia, with very heavy and concentrated positions, especially in South Korea, where the average leverage was two to five times. So, if Micron's financial statements had even the slightest flaw, it would lead to a significant deleveraging.

Even if Micron's financial report is perfect and exceeds sell-side expectations, it will inevitably run into problems later due to some seemingly insignificant events. For example, the recent SemiAnalysis report, and Nvidia's Vera Rubin's announcement last week to reduce pre-installed storage. Actually, that's not a problem; Nvidia is simply saying that when providing solutions and servers, they won't pre-install much storage themselves, but customers will install it themselves. It has no impact on the storage narrative or fundamentals. However, in a time of high leverage and concentrated positions, any nonsensical or absurd event is highly likely to lead to a large-scale deleveraging. For example, Micron's financial report, and the upcoming IPOs of Chinese companies like Changxin Memory and Yangtze Memory Technologies, could all become negative catalysts.

In the public market, AI's play is semiconductors, and semiconductors' play is memory. These things may accompany us throughout this year and even next year, with large-scale deleveraging and re-leveraging, because everyone knows this is in short supply. Memory has transformed from a cyclical stock into something with a potentially long cycle, and people may re-expand its valuation multiples. When everyone agrees that something is good, it becomes very dangerous. So, in my view, the real danger lies in Micron's financial report, or the listings of Changxin Memory and Yangtze Memory Technologies; as for the FOMC, OPEX (option expiration), CPI, PPI, and even Trump and Iran, they are actually not so bad. The market may have overreacted, and when the events are finalized, it will be an opportunity to release risks and create a relief rally.

V. The Controversy Regarding SemiAnalysis's CPO: Optical communication isn't that fast; CSPs prioritize stability.

Mr. Z: You just mentioned SemiAnalysis' report on CPO and optical communications two days ago, which triggered a sharp drop in the sector. Let me summarize their key points for everyone: CPO is on the right track, placing the optical engine next to the chip and shortening the electrical signal distance to reduce power consumption, but mass production reliability and maintainability are not yet mature, and it will not completely replace existing pluggable solutions in the short term; the real battleground is scale-up, and Nvidia's current push for CPO is more like "training" for the next-generation Feynman architecture, with large-scale deployment waiting until after Feynman; the real beneficiaries are players like TSMC that can integrate silicon photonics, EIC, and PIC, while the real alpha lies in the underlying bottlenecks such as external laser, optical engine, FAU, and fiber coupling. The conclusion is that CPO will not be realized until 2027 or 2028. Mr. Z, what do you think of this report?

Excavator: I think it's fine. It's an objective reality; CPO (Consumer-on-Producer) just isn't that fast. First of all, for CSPs, the current focus is still on scale-out, and stability is paramount for CSPs. If Nvidia wants to do this, it essentially has to sell to CSPs, and the CSPs decide. Building a data center is so expensive; they won't take the risk until yield rates are well resolved. I think it's reasonable that it can't be scaled up. There's been a lot of hype surrounding CPO lately, which I find unreasonable because, from what I understand, it just isn't that fast. That's the first point.

Secondly, the current supply chain capacity is insufficient. Whether it's CPO or silicon photonics modules, they all require CW (continuous wave) laser sources, and there's a shortage of all CW sources. Where are we going to find enough resources to do this? There simply aren't enough.

Mr. Z: What did the Vera Rubin, Rubin Ultra, or Feynman era at Nvidia really mean for the CPO? In which generation will the CPO truly become a core component?

Excavator: I'm not sure, mainly because of the maturity of the technology. Just because Nvidia plans to use a CPO at Feynman doesn't mean they'll actually use it. Look at GB200 and GB300, they keep getting delayed because they need their supply chain partners to help them solve problems. If they can't solve them, there will be further delays. So, frankly, I can't give a judgment right now.

VI. How to Choose Optical Communication Targets: AAOI, Marvell, and the Judgment that "Optics are Too Expensive"

Mr. Z: The market has been discussing several optical communication stocks lately, such as Lumentum (LITE), AAOI, Marvell, etc. How would you differentiate these companies, and who is the real bottleneck?

Excavator: First of all, I'm genuinely not optimistic about AAOI. Many people on Twitter know that I was one of the earliest bloggers to look into AAOI and did very detailed research. It has three bases: Houston, Taiwan, and Ningbo in mainland China, with the main module production base actually in Ningbo. However, based on my research, the progress there is extremely slow. The figures Dr. Lin Zhixiang (AAOI founder) touted publicly, based on my actual research, will most likely only deliver one-third, or even less, of the promised figures, yet the stock price has already reached the level the CEO boasted about. Therefore, I am very pessimistic about this company. Of course, it's a small company, and when sentiment is good, it's easy for the stock to rise just because of a CEO interview; it's not easy to short it, but I just don't have a positive outlook.

Marvell actually has three main business segments. First, there's DSP (Digital Signal Processor), because there's a severe shortage of modules in the entire optical communication field, and DSPs are also in high demand. There's been talk that its DSP market share has been taken over by Broadcom and MaxLinear, but the overall industry pie is actually growing, so DSPs are still a strong sector. Second, it also makes PCIe switches and related components. Third, a significant part of its business involves collaborations with CSPs to develop ASICs (Integrated ASICs), something Marvell has been talking about for years but hadn't yet implemented, but now it's starting to see results. These factors combined, along with Jensen Huang's endorsement of Marvell, have led to its recent strong performance. However, it's quite difficult to get in at this level; I think it depends on its actual delivery.

Mr. Z: So you basically agree with the direction of SemiAnalysis, and that the CPO should indeed be extended?

Excavator: Yes, the CPO is indeed going to be delayed. Actually, it's like this: Back in May, I clearly stated I was "bullish on Lumentum," but I didn't say it on Twitter. In private conversations with Labubu and others, I explicitly stated I was "not bullish on Lumentum." Why? Because in May, North American companies like Lumentum were already priced at 30x PE for 2027, even considering the possibility of supply chain constraints and the actual EPS not reaching that level. It's not that I wasn't bullish on it in the long term, but I felt it was already overpriced at that time. It's a valuation issue.

VII. MLCC Supercycle: Why is this round bigger than 2017-2018?

Mr. Z: You posted some tweets about MLCCs in May, saying that this MLCC cycle is likely to be larger than the 2017-2018 cycle because the entire demand logic has changed. If I remember correctly, the price increase of passive components in 2017 and 2018 was due to the generational upgrade of smartphones, coupled with the fact that the sector was mainly dominated by major Japanese and Korean manufacturers and supply was restrained, which caused the stock prices of many passive component manufacturers to surge, like Yageo in Taiwan. That cycle was what everyone called the "Yageo 2018 turmoil." Why will this AI super cycle cause a change in the entire valuation logic of MLCCs?

Excavator: First, the entire AI industry chain has already seen price increases across all segments; this is the first piece of background. Second, before the price surge, passive components had relatively low price-to-book ratios (PB), with companies like Taiyo Yuden having very low PB ratios and low valuations. Third, this wave of price increases mainly focused on high-end MLCCs, which are basically controlled by Murata, Taiyo Yuden, Yageo, and Samsung Electro-Mechanics. These companies essentially have a monopoly, accounting for 80-90% of the global market share. The demand for these components in AI servers is several times higher: not 30% or 50%, but 5 to 7 times the demand. Under such demand, the price elasticity is extremely high.

Mr. Z: So, what valuation method should we use for the MLCC passive component sector? Because the valuation differences among giants like Murata and Taiyo Yuden are quite significant.

Excavator: I think we should look at the price-to-book ratio (PB) first. Just like when the memory market was first hyped, we looked at the PB ratio initially. Only when the market realizes that it is a long-term cycle can we use the price-to-earnings ratio (PE) to estimate it.

Labubu: I agree. Initially, this could only be done using PB (Price-to-Stock) valuation. Because under the logic of shortages, factories with low PB valuations indicate they had sufficient CapEx (Cap-to-Exchange) available beforehand. In a shortage environment, their production capacity might be even larger, allowing them to benefit more from price increases and thus have a higher relative value. That's why we also bought Yageo. I heard you Taiwanese companies are very optimistic about Yageo?

Victor: Yes, everyone was going crazy. Actually, at the beginning of the year, I had a friend in the semiconductor business who was talking to his company's purchasing department, and the purchasing department mentioned that Yageo was in very short supply and that the price would rise, so some people started to position themselves in January and February; however, Yageo's surge didn't actually begin until April, May, or June. In the past two days, the Taiwan stock market has generally corrected, but Yageo and passive components are among the few sectors that have hit new highs again and are showing great strength.

Excavators: The logic remains the same: the entire industry chain is experiencing a shortage, and this is a "new gap." The old gaps have already been driven up by speculation, but the price increase for passive components is just beginning. The first wave was driven by expectations; next, it will be driven by EPS delivery (actual profit realization), leading to another wave of price increases, which will likely see the largest increases. Real price increases, such as those announced this week or last week, would have resulted in significant price hikes for companies like Taiyo Yuden and Murata if the recent market conditions hadn't been so favorable.

Mr. Z: So this isn't the tail end of the trend?

Excavators: This isn't the end of the story. The real price surge, as I understand it, has only just begun.

Labubu: Let me interject. I've actually been communicating a lot with some friends, and we actively positioned ourselves in this area from the end of last year to January and February of this year. But as you just mentioned, the initial surge at the beginning of the year was pretty much nonexistent and quickly subsided. This sector really started to pick up recently, probably around May: when some sell-side investment banks started covering this, for example, Morgan Stanley said that the usage of MLCCs in Rubin might increase by more than 180% compared to GB300. It wasn't until April or May that we started to see significant trading opportunities and real funds coming in to buy.

I want to say this: even if you have insights that are ahead of the market, you may still need to wait until many people know about it, see a technical breakout, and a significant upward slope before joining in. Like when we bought Vishay (VSH) and FengHua High-Tech in January and February, we lost everything in just a few days. This wasn't because your investment thesis was flawed, but because the timing was wrong. As a retail investor, you probably need to wait for large institutional investors to enter, for everyone to reach a consensus and hype the issue before you start investing to profit; otherwise, you'll be stuck there for three months.

VIII. Memory is Back: The IPOs of Changxin Memory and Yangtze Memory Technologies, and the Five-Year Valuations of Samsung, SK Hynix, and Micron.

Victor: Speaking of storage, the giants like Samsung, SK Hynix, Micron, and SanDisk all have different configurations. Some use American brands like Micron and SanDisk, while others use Korean storage. Labubu also mentioned paying special attention to Chinese companies like Changxin Memory and Yangtze Memory Technologies. What are your thoughts on the competition among these giants and the future market outlook?

Excavator: First, how do I view the current position of storage? From a technical perspective, looking at the weekly and monthly charts, the deviation rate and slope are too high. I think it may need to fluctuate for two months because the fundamentals are still very solid, and the decline won't be much, but fluctuation is normal.

Secondly, what's your view on the IPOs of Changxin Memory and Yangtze Memory Technologies? I don't think their listings will have a significant impact on the overall storage landscape. First, their technology is still relatively underdeveloped; second, storage technology can't currently penetrate the Chinese market, while China has its own AI, a population of over a billion, its own models, and its own market. Just meeting China's own supply is difficult enough, not to mention consumer electronics, which are currently too expensive to buy. Therefore, I estimate that Changxin and Yangtze will have difficulty meeting the needs of the Chinese market, so there's no need to worry too much about that.

Another point: If memory prices fall, and then Changxin Memory and Yangtze Memory go public, because there are very few truly profitable semiconductor companies in mainland China, Changxin and Yangtze may be the only one or two that can make big profits. Therefore, their valuations could be driven very high, perhaps 20 or 30 times earnings. Based on next year's profit of 200 billion RMB, a 20x valuation would be 4 trillion RMB, and a 30x valuation would be 6 trillion RMB. The IPOs could generate 2 to 3 trillion RMB. If sentiment is boosted, I think it could actually be beneficial to the entire memory sector, rather than detrimental.

As for Micron, SanDisk, SK Hynix, and Samsung, from a long-term perspective, say five years, it currently appears that storage is in short supply and there won't be an oversupply, because demand is growing too rapidly. However, one point to consider is that by next year, Samsung and SK Hynix's profits might exceed the combined profits of all semiconductor companies. This is a risk. I don't know how things will develop then. If Trump says, "You Koreans are making the lion's share of the profits in AI," will he tweet urging them to expand production? That's a risk.

If we set aside politics, I'm very bullish. Looking at next year, both SK Hynix and Samsung have P/E ratios of less than 5. Samsung currently has a market capitalization of about $1 trillion. If its net profit reaches $400 billion by 2028, that's equivalent to a P/E ratio of only 2.5. Since there won't be oversupply for the next 5 years, these companies theoretically deserve a valuation of at least 10 or 15 times earnings. Of course, SK Hynix is ​​similar, and Micron is similar as well. So my own strategy is: I will continue to buy when prices fall, but at the same time, I will use a short-term put option as protection.

Continuing from what I just said: because storage is in severe shortage and currently accounts for a very large proportion of the entire CapEx, capacity expansion is inevitable and will be very rapid. However, even at a rapid pace, it won't be able to keep up with the demand. Therefore, I think the same applies to equipment; equipment and materials are definitely a very promising area.

9. Trump's National Team for Technology: Government Equity Investment is Essentially a Form of "Transfer Payment"

Victor: Recently, Trump said he wanted to invest in some AI companies, talk to tech giants, and try to form a "national tech team." What's your opinion on government involvement, and how will it affect the stock prices of companies in AI, semiconductors, and memory?

Excavator: I think Trump did something very smart. Why? This wave of AI was actually achieved at the expense of other industries because it's essentially all about taking away a share of the labor market.

I've been saying "CapEx replaces OPEX (operating expenses)," which is actually eroding the workforce in other industries. Without intervention, you'll gradually see all industries shrinking, except for AI, which is developing rapidly, inevitably leading to unrest. How can the government resolve this unrest? The only solution is transfer payments (in economics, this refers to the government transferring fiscal revenue, such as taxes, to a specific group without compensation, like subsidies or social welfare, without receiving goods or services in return): collect an AI tax and then transfer payments to other industries.

Trump's decision to allow the government to take a stake in the company is essentially a form of transfer payment. Therefore, I think this is a very smart and correct approach, and other countries may gradually follow suit. For example, South Korea has proposed redistributing the profits of Samsung and SK Hynix by imposing higher taxes on them and then redistributing the money to the people; China may also do something similar. If this is done, the public reaction will be less intense.

10. Equipment and Materials: This round of price increases even extends to equipment.

Victor: You just mentioned focusing on equipment and materials. What are your thoughts on semiconductor equipment manufacturers like ASML, Tokyo Electron (TEL), Applied Materials (AMAT), and Lam Research? Are their current valuations too high, or are they still worth investing in for the long term?

Excavators: I think this is still a worthwhile area to invest in. Equipment is typically cyclical, and prices usually don't rise, but you'll find that in this cycle, even equipment prices are increasing, which is a very important point. Plus, with the current semiconductor shortage, its growth rate will definitely be very fast, so it's certainly not at a very high level yet; it's at least in the middle.

Another point: equipment prices don't rise as much as materials, so my own feeling is that equipment suppliers or material suppliers may be more flexible, because these things can increase in price.

Victor: What aspects of materials do you pay particular attention to?

Excavator: I haven't studied it in great depth yet. I've been quite busy with work lately. I'll study it more carefully later and share my findings when I have the chance, or I might post them on my Twitter.

11. TSMC vs. Intel: Competitive Advantages, Geopolitics, and "Going All-In on 2330"

Mr. Z: Speaking of price increases, we have to talk about TSMC. They held their shareholders' meeting a few days ago, and C.C. Wei made several memorable statements, repeatedly urging everyone to buy their company's stock. TSMC is such a large company with a market capitalization of $2.1 trillion, so why hasn't its stock price fluctuated as much in the long run? From my own observation, firstly, it only raises its prices to customers by 10% or 15% at a time, not by four or five times at once; secondly, the more recent pressure is actually geopolitical: Trump is building a national technology team, handing over wafer foundry tasks to Intel's Chen Liwu, and TSMC has already begun prioritizing AMD chips over Intel. One is price increases, the other is geopolitics—what are your thoughts on this?

Excavator: My personal understanding is that geopolitics certainly has an impact, but TSMC's position is actually quite difficult to shake. As for price increases, everyone is raising prices now, so why can SK Hynix have a 90% profit margin while TSMC only has a 50-60% profit margin? I think TSMC's price increase is reasonable.

Labubu: I think TSMC is a relatively farsighted company, and it's quite restrained in raising prices. It hasn't deliberately pushed its margins extremely high, unlike Samsung and SK Hynix, which I see as rather greedy. TSMC probably wants to maintain the ecosystem: get the margins it's entitled to, but not squeeze the downstream too much. So, the geopolitical factors I just mentioned, plus its restraint on margins, are two very important reasons. Also, it is indeed very large. Purely from a valuation perspective, isn't it cheap? It is very cheap, but it can't rise much further because the amount of money needed to drive it up would be enormous.

Mr. Z: What about Intel's strategy? It's pushing EMIB, which is basically competing with TSMC's CoWoS technology. Coupled with Trump's foundry strategy, how will it impact TSMC?

Labubu: I think Trump's actions are more of a "necessary evil." I don't want to discuss politics, but I feel he's thinking about this: if Taiwan "abandons" its supply chain in the future, how can the US guarantee the security of its supply chain with TSMC? So, of course, he needs to replicate Intel. But in terms of packaging, achieving the same quality and scale as TSMC is still a very long way off. In my opinion, the current hype surrounding Intel is purely speculative. I'm not an expert in this area, but this is my feeling. However, I know that people at TSMC actually see EMIB and similar technologies as a significant threat.

Mr. Z: I heard from my engineer friend in Hsinchu that he basically said "go long on 2330 (TSMC's stock code) before 2030".

Labubu: Haha, going all in until 2030.

Victor: What's your take on the geopolitical power struggle between TSMC and AMD, as well as between the US supply chain and the Taiwanese supply chain?

Labubu: That's beyond my understanding; I can't answer that from a supply chain fundamentals perspective. But from a trading and speculation standpoint, I'll be actively monitoring both (AMD and Intel) and looking for good entry points. Because the trend is very positive for both AMD and Intel, I'll use this to my advantage. But if you ask me how I specifically view the supply chain fundamentals, I can't answer that.

12. The CPU Market and ARM/Qualcomm: Follow the Trends, Don't Bet on What You Think Are Insider Tips

Victor: In the CPU sector right now, ARM and AMD have seen significant price increases recently. Apple, AWS, Nvidia, and Qualcomm are all using ARM-based CPUs. What are your thoughts on ARM as a company, and the future development of the entire CPU industry? If the correction ends, which sectors of the AI ​​semiconductor industry will have the most explosive growth potential in the next phase?

Labubu: Let me start with ARM and QCOM (Qualcomm). I've talked about ARM stock on Twitter before, and I made quite a bit of money from it. The logic behind this is simple: AI is driving this surge in CPU demand, which you can confirm through many financial reports, such as Lisa Su's comments on the GPU to CPU ratio in AMD's financial report. We previously invested in ARM because it hadn't risen much before, there were many themes to speculate on, and it was also a good long position target for SoftBank.

QCOM is a company that's been hyped for its custom server-side CPUs. It used to have a very strong team called Nuvia, which publicly claimed to have secured numerous orders from major manufacturers, leading many to believe it might release a server-side ARM CPU this summer. However, from what I understand, almost the entire core team has left to start their own startups. These people originally came from Apple. Therefore, QCOM might be purely a speculative stock, and I'm very bearish on its fundamentals. ARM, on the other hand, is very promising. Besides the licensing aspect, it might start making its own CPUs later, perhaps in collaboration with Samsung. So, in the long run, QCOM is a good speculative stock, but its fundamentals are very bearish; ARM has good fundamentals, but the problem is that it's too expensive now, no longer the $100 thing we used to work with.

If you ask what to do after the pullback ends, I think you should first look at which stocks have room to rise relative to your target price. For example, Lumentum (LITE), I think it will only reach $1,200 at most next year, so I might wait until it drops to around $800, thinking that a 50% upside is better. This completely depends on your own valuation based on fundamentals. For example, Micron, if you use the 15x valuation that the "excavator expert" mentioned, that's absolutely outrageous; over $2,000 sounds a bit exaggerated. I might think it's worth $1,000, $1,200, or $1,500. You need to have your own valuation target price, and then, through this pullback, find which stocks have significant upside relative to their target price.

Secondly, it's still important to look at market momentum. Some things you think are good, like QCOM. Based on information you read online, you might think its fundamentals are worth $300, which seems great, but the market momentum might show otherwise. It's possible the information you have is false or outdated, and the market knows things you don't. For example, I knew a long time ago that the entire Nuvia team at QCOM had left, and I don't know what talent the remaining members possess. Therefore, what retail investors can do is: first, assess the upside based on the current price and fundamental valuation; second, observe its dynamic momentum to see if the market will continue to pump it.

As a retail investor, unless you're an industry chain expert, it's difficult to know these details. All you can do is follow the trend. The information you get on Twitter and Facebook groups is likely secondhand and outdated. Unless you have connections like the "excavator expert" who are truly involved in the industry chain and can verify things yourself, retail investors can only follow the trend. For example, you can add what you know to a watchlist, check it daily, and use indicators to measure its momentum and relative strength. For instance, I recently noticed some cleanroom-related stocks starting to surge, which could be a trend.

Thirteen, Momentum Trading in Practice: How to Set Target Prices and How to Take Profits – "Knowing when to Sell is the Key to Success".

Victor: It sounds like you're more of a right-side trader, someone who follows the trend. How do you identify when a sector is about to break out? And how do you set target prices, stop-loss orders, and manage long-term and short-term positions?

Labubu: Target prices calculated using fundamentals are usually long-term. For example, I might think a certain stock could reach $1,200 next year, but the path from the current $800, $700, or $600 to $1,200 could be very winding, with highs and lows, making the process quite painful. So, in this case, I usually look at technical analysis to see when to enter and exit the market, using things I just mentioned: gaps, moving averages, key support and resistance levels, to observe which entry point is better.

When deciding where to exit, I might set a trading stop or look at indicators like ATR (Average True Range) to see if the stock has risen too sharply and deviated from its average. There are also some top patterns: when it's at a relatively high position, the candlesticks get bigger and longer, and the volume increases significantly. This often indicates a buy climax, meaning the divergence between those exiting and entering the market is widening—a time when everyone is calling each other idiots. This is often a warning sign. You can clearly see this in Samsung Electro-Mechanics on May 27th, 28th, and 29th: the volume increased, the candlesticks lengthened, and the price moved further away from the 5-day moving average. That was actually a good selling point.

Victor: The SPX also saw a large volume spike on May 29th, followed by two more days of gains, before the recent pullback began.

Labubu: Yes. So, ATR multiples, combined with observation of daily volume and price relationships, are excellent profit-taking methods. As for stop-loss, such as key support levels, gaps, or moving averages, that's relatively simple. Actually, for most people, the hardest thing is often profit-taking, not stop-loss; "knowing when to sell is the true master." Having experienced this bull market, many have witnessed stocks that skyrocketed, like Samsung, SK Hynix, SanDisk, Samsung Electro-Mechanics, Taiyo Yuden, and even ARM at the beginning of the year—all rather volatile stocks. Therefore, high-volume tops and ATR deviations are methods I often use to take profits, ensuring that I both preserve profits and avoid selling too early.

14. Advice for the audience: How to live in the Year of the Horse (2016), and the signals behind the price reduction of models.

Victor: We've been talking for almost two hours. My final question is, in Chinese metaphysics, 2026 is the Year of the Horse (丙午), a year of great upheaval and significant change. We've seen the market fluctuate wildly this year, with sharp rises and falls. At this point in time, and in the second half of 2026, what advice would you give investors on how to follow the trend and minimize risks?

Labubu: I think starting from the market movements of the last two days, the second half of the year might see a very choppy market, the kind where it goes up for a few days and down for a few days, which can be frustrating and uncomfortable, but the overall market is in an uptrend. So until we see a clear signal that CapEx has stopped or decreased, especially for those trading semiconductors, I think it's crucial to strengthen your ability to withstand pullbacks during this process. Some of the things SemiAnalysis says are half true and half false, half right and half wrong, and they're quite "disgusting." Do you think they have an ulterior motive? I think they do; there might be someone behind them trying to acquire shares at a lower price.

Therefore, in this process, I think we should reduce the use of leverage, but hold onto our targets and set stop-loss and take-profit orders. Sectors like storage, devices, and optics—although the analyst just complained a lot—I still think optics is a very good sector, as are MLCCs. I still believe we can make a lot of money from these. Before CapEx stops or decreases, or before OpenAI and Anthropic go public, semiconductors are the only AI play in the public market.

Victor: Do you think the revenue growth rate of model manufacturers is a potential risk? Recently, there has been a discussion about whether models are "becoming less intelligent".

Labubu: I don't think so. On the contrary, I think it's a good thing. The "lower intelligence" reflects insufficient computing power, but look at today's news, such as OpenAI wanting to lower prices to compete with Anthropic and reduce token prices. If I spend 100 yuan here but run a "lower intelligence" model, I will definitely switch to another one; and in fact, the migration cost from Claude Code to OpenAI's Codex is not that high. Even if you have written a lot of skills, the migration can be done quickly with the help of AI. So in the end, it will be like the competition between cloud service providers in the past, the unit price of tokens will decrease, which is actually good for downstream applications, allowing the whole AI story to continue and applications to expand significantly.

Competition among model vendors is intensifying. If Anthropic is still charging 200 yuan for a relatively weak model, I can get a very intelligent one from OpenAI for 100 yuan, so I'll definitely go there. Therefore, their overall revenue will still grow exponentially, but the unit price of tokens will definitely decrease. Conversely, this also means that the upstream semiconductor industry's strategy of "constantly relying on price increases" will one day be broken very quickly, and this price reduction will likely be transmitted to the upstream of the industry chain faster than people imagine. This is the same as what the excavator expert said.

Victor: One last, easier question. Why is your ID "3X Long Labubu" (3x Long Labubu)? Why are you so bullish on Labubu?

Labubu: Because my daughter loves Labubu. Before, when Labubu was harder to find, I tried my best to collect them for her, spending a lot of time trying. Now I don't need to. Mainly because my daughter likes them, I buy them whenever they release a new one.

Mr. Z & Victor: Thank you so much to 3X Long Labubu and AI Industry Analyzer for spending nearly two hours today sharing a wealth of the latest market insights, covering technical, data, fundamental, and macroeconomic perspectives. Thank you also to every listener who has listened all the way here. If you enjoyed this episode, please follow both instructors, as well as 168X's X on WeChat and YouTube, and share the program with more friends interested in AI, semiconductors, and trading. See you next time!