Author: Victor (@vcmktasa) · Mr. Z (@168MrZ)

Japanese Stocks Are the "Pickaxe Sellers" of the AI Era: Finding the Next Winners in the Asian Supply Chain

In late June 2026, volatility in AI and semiconductor supply chain stocks continued to intensify. Micron is about to report earnings on June 24; Kevin Warsh made hawkish remarks on future interest rate management; South Korea's KOSPI broke through the 9,000-point mark intraday for the first time on June 18. At this moment where macro variables and hardware frenzy intertwine, 168X invited Fiona (@nft_hu), an independent research-driven investor whose coverage spans memory, MLCCs, optical interconnects, power semiconductors, and global equities.

Fiona initially traded cryptocurrencies but decisively exited after the massive liquidation last October, redirecting her capital into the AI tools she heavily uses. She approached it from a "first principles" perspective: reasoning backward from a silicon-based standpoint to determine what an AI data center truly needs to develop well. Her current portfolio spans US, Japanese, Korean, and Taiwanese stocks, with memory and optical interconnects as her top two positions, resembling a globally allocated mini semiconductor ETF. In this conversation, she offers a very distinct judgment: The most trustworthy person in the market should be someone like Jensen Huang, who is at the forefront of AI and understands the supply chain best; memory will be a persistent bottleneck in this cycle, with sustained prices, demand, and profits, so she basically ignores short-term fluctuations. She also shares a hard-earned insight: reduce trading frequency; almost all swing trades end up "selling off" your good, low-cost positions too early.

I. Research Framework: From Crypto Trading to Reverse-Engineering Silicon Demand with First Principles

Mr. Z: Fiona, you cover many sectors, including memory, MLCCs, optical interconnects, and 800V. Could you start with a brief introduction and share how you developed this research framework?

Fiona: Hello everyone, I'm Fiona. First, thank you to 168X for the invitation. I've seen your outline, and it feels like you've really dug through my tweets. You probably found some even I forgot I posted.

I actually started with cryptocurrency trading. But as everyone knows, the crypto market has been shrinking. By the time meme coins were rampant, I realized it wasn't really my style, as I lean more toward research-driven trading. Then, after that incredibly absurd major liquidation on October 11 last year, I left crypto. I was quite decisive, cleared all my positions, and felt I needed to find a new home for that capital.

Since I'm a moderately heavy AI user myself—though perhaps not comparable to friends who do coding—I thought, maybe I should study the AI I use? Because by the time it reaches us, it's already a model, a product; what truly matters is everything driving it behind the scenes. So it was a somewhat serendipitous opportunity that led me to look into AI. Initially, I was like a sponge absorbing water frantically, because I knew nothing about it.

The sector that really captivated me was memory. I feel very fortunate that at the time, memory was experiencing a significant price surge, both in spot and contract markets, so there was a lot of media coverage. I started researching: Why are memory prices rising? Is this a short-term spike or a super-cycle? That's how I entered the memory field, delving deeper and deeper, and my position grew larger and larger. It was indeed the most profitable sector across all my investments this year.

Because AI involves so many fragmented frameworks, I prefer using first principles to analyze: from a silicon-based perspective, what exactly is needed to develop an AI data center well? I reason backward from its standpoint. This gradually shaped my own investment framework. Of course, I'm also a heavy Twitter user, so I built a small bot to capture trending topics online as a reference and confirmation. That's roughly my introduction and research framework.

II. Don't Sell Off Your Memory: Jensen Huang's Pivot, Micron's Earnings, Korea's Re-rating, and HBM4E

Mr. Z: Regarding memory, you previously tracked Micron from 240 all the way to 1,000, which is a showcase of diamond hands and value investing. How do you view this current memory rally?

Fiona: Here, I'll slightly paraphrase a previous statement by Jensen Huang. I think it was this January at CES, he said: the bottleneck in AI has shifted from what we previously called compute—insufficient computing power, not enough GPU chips—to a lack of context.

That statement deeply influenced me. I thought at the time: What makes someone who sells GPUs say the bottleneck has shifted to context? Context is what later became widely known as memory. We can see that AI's new development bottleneck actually requires larger memory to support it. I wonder if you've noticed AI getting dumber the more you use it? It's because its long context can cause it to become a bit disordered. So people in the market have varied opinions; everyone says their own position is the most important. But I feel that the most trustworthy person should be the one at the forefront of AI who understands the supply chain best, and Jensen Huang is definitely the person I trust most. So when he is this bullish on memory, believing context will be the next driver of AI development, I delved deeper into this area.

If we talk about why DRAM is so scarce, it's also relatively easy to understand: there's a "scissors gap," a supply-demand mismatch. The current pace of computing power development far outstrips that of HBM (High Bandwidth Memory). So no matter how much memory is expanded or stacked, it can't keep up with the speed of current GPUs, meaning the more you stack, the better.

And HBM is essentially a type of DRAM; its production squeezes out a tremendous amount of capacity. Because no one has the time to immediately build a new fab—we all know a memory fab not only costs tens of billions of dollars but also requires a lot of time. So the fastest way is "production conversion": directly converting existing DRAM production capacity to manufacture HBM. This is the quickest method, but it further squeezes the supply of DRAM itself. So if you look at all types of memory on the market, you'll ultimately find various kinds of shortages.

I think there's another trend now, I'm not sure if you've noticed: some are starting to hype NAND as the next memory bottleneck. I think this is primarily driven by economic efficiency. Although it has some drawbacks compared to HBM, it's very cheap, potentially costing only one-fiftieth of current HBM. For something this economically efficient, people will still consider switching, so this part will generate new gaps and new demand, including how to connect the two and how to perform the conversion. I think there will be new opportunities here.

Mr. Z: The entire market's focus right now is on Micron's earnings next week on June 24, plus yesterday's Fed decision and the attention on Warsh's stance on future rate management. You likely hold memory positions yourself. How do you manage them during this period? Do you think next week could be very volatile?

Fiona: Understood. In this situation, I probably won't trade much, because I've found this year that my trades have basically all been wrong. All swing trades serve only to sell off your good, low-cost positions too early.

So as I said, I'm more likely to stick firmly to my first-principles analysis. I believe memory will be this bottleneck, and this is persistent: prices are persistent, demand is persistent, and profits are persistent. So I basically ignore these short-term disturbances. If volatility does arise due to earnings or these few days, I actually think the market should become more volatile, and I would hope to use that volatility to buy more memory. That would be my trading direction.

Mr. Z: Jensen Huang's Asia trip in late May and early June only included Taiwan and South Korea. Some Japanese newspapers wrote that Japan might miss this AI supercycle. Looking at memory fabs, if I recall correctly, Sony was in the game during the 80s and 90s but couldn't compete later, leaving it to Taiwan and Korea, with Korea led by SK Hynix and Samsung, so Japan lags in this area. Speaking of Korea, the Lee Jae-myung government introduced new legislation requiring Korean companies to cancel treasury shares, as chaebol families have long controlled companies with very few shares, and too many shares outstanding isn't good for EPS. How do you view this wave of reform in Korea and how it reflects on KOSPI's performance?

Fiona: This has actually already been validated. I think I wrote about it back in February, when Korea passed that corporate reform act. It's just that attention on Korean stocks wasn't that high back then; the index was around 6,000 points, I believe. Didn't it just break 9,000 today? So you can see, the Korea Discount is gradually being eliminated. Of course, SK Hynix is also planning to issue ADRs in the US, and I think there might be another wave of euphoria waiting ahead. But the impact of this event has already unfolded over the past four months.

Victor: What's your view on the current memory Big Three—Micron, SK Hynix, and Samsung? Regarding the latest generation HBM4E, although Samsung seems fastest in terms of speed and press releases, rumors suggest SK Hynix is more likely to supply in large volumes. Today, SK Hynix also announced it has started delivering HBM4E samples to customers. How do you see the impact of HBM4E going forward?

Fiona: I'm also inclined to believe it will be SK Hynix. Whether judging by its past product track record or stock price performance, I think it should be Hynix. Of course, Samsung is also very good, so one could buy both.

III. The Fundamental Difference Between MLCCs and Memory: Complementary Demand vs. The More, The Better

Mr. Z: You previously mentioned that memory and MLCCs are fundamentally quite different. Memory has relatively singular models, while MLCCs have physical limits. Could you elaborate on that?

Fiona: It's actually very simple. As I just mentioned, AI model training requires more memory, and this part currently seems perpetually in short supply. So everyone just keeps stacking HBM and DRAM to utilize computing power more fully. The logic is very straightforward: the bigger, the better; the more, the better.

But MLCCs (Multilayer Ceramic Capacitors) are different. Their demand growth is "complementary"—that's how I define it. For example, if I buy a pickleball paddle, it might come with two balls. That complementary quantity is fixed; it's not a case of the more, the better. So the demand for MLCCs is quantitative and can be calculated based on server rack shipments. There are many different calculation methods now; one I've seen that seems relatively credible suggests a demand increase of roughly a bit over 3 times.

Moreover, as you mentioned, there are many types of MLCCs. This demand explosion and price increase are actually concentrated in a small portion of relatively high-end MLCCs: those with difficult manufacturing processes, possibly hundreds of layers, that only a few manufacturers can produce. It's this very small segment that has received frantic orders. So I think it's somewhat different from the "broad-based rally" in memory.

Of course, when a concept emerges, buying anything tangentially related can make money. Here's a fun example: a friend in my group said he couldn't buy the Murata stock I recommended because it's a Japanese stock, so he bought a Chinese A-share company called Sinocera Materials, which also rose nicely, and he was quite happy. But I had researched it at the time, and this company's correlation with high-end MLCCs is actually quite low. It's just that recently the market is rallying by sector, so anything related to MLCCs benefits. However, I personally prefer to buy the true beneficiaries, because I don't want to trade frequently.

IV. Japanese Stocks Are the "Pickaxe Sellers" of the AI Era: Murata, Taiyo Yuden, Kioxia, and Ibiden

Victor: Many MLCC leaders are actually in Japan, like Murata and TDK, with Yageo in Taiwan. Japanese stocks are generally undervalued right now, and if the US potentially starts raising rates after Warsh's FOMC, a stronger dollar and weaker yen would actually benefit Japanese export companies even more. How do you view Japanese stocks, and what is your current allocation like?

Fiona: First, I really like Japanese stocks, perhaps even more than US stocks, especially recently. I feel the market is still considerably frothy, and the cheap valuations of Japanese stocks give me a greater sense of security—though of course, this could be a mistaken notion.

I've written two long articles on MLCCs. One was on the number one player, Murata, which is the undisputed leader. The second was on Taiyo Yuden, which I consider the purest MLCC play among Japanese stocks. Because Murata has a minor issue: its capacitor segment (which includes MLCCs) is too large. Although MLCCs benefit significantly, the improvement reflected in overall revenue and profit margins isn't that obvious from one or two quarterly reports. Taiyo Yuden is different; it's relatively smaller, purer, and has higher elasticity.I remember its high-end MLCC-driven revenue increased by about 20%, but operating profit should have increased by 90%.

But I currently only hold Murata. It's not that I don't want to buy Taiyo Yuden, but my limit order just hasn't been filled—it's been too strong.

Speaking of Japanese stocks, I think Japanese equities have played the role of a "shovel seller" well in the AI era. It may not have outstanding models, but if you go through the chain—whether it's semiconductor manufacturing equipment, materials, chips, or passive components, packaging substrates—at every link, there are excellent companies with a strong position. So I've been buying quite a lot of Japanese stocks recently, and I will continue to add on good pullback opportunities.

I can share my current top three holdings. One is Kioxia (285A), which is also a memory play; one is Murata Manufacturing (6981), that's MLCC; and another one whose Chinese name I'm not entirely sure of, I just call it Ibiden (4062), which makes packaging substrates. I wrote a post about it today; those interested can go check it out.

Victor: I saw your post about Ibiden today. You mentioned you bought it, forgot about it, and later found it had been rising all along. That position is quite interesting.

Fiona: Yes, because my husband buys my Japanese stocks for me, and I rarely check that account. When I see something good, I just ask him to buy it, which makes it easier to hold. So one day I looked and thought, how did this 4062 go up so much? I had actually forgotten the reason I bought it, because I only bought it once. Later, when I traced back, I realized, oh, great. So I said if there's a pullback, we'll add a bit more, and we were lucky to add some recently.

5. Power Semiconductors, 800V, and Third-Generation Semiconductors: Why China Has Mastered the Cost Advantage

Victor: Around May, you started posting about 800V DC, power semiconductors, and compound semiconductor companies like GaN and SiC. Is the power semiconductor sector poised for a breakout? Because when the overall semiconductor market sold off in early June, names like Navitas, Wolfspeed, and ON Semiconductor all saw significant pullbacks. Could this be an entry opportunity?

Fiona: To share with everyone first, my power semiconductor positions are basically still around my cost basis and haven't entered profitable territory yet. There were two initial reasons I bought them. First, I looked at the new 800V HVDC architecture, and among the sectors it benefits, power semiconductors are an unavoidable one, so there are fundamentals and demand drivers. Second, when I looked, I felt power semiconductor prices were relatively cheap and hadn't risen much. So frankly, we are somewhat "betting on the future": I hope it rises, I think it will rise, but it might not necessarily rise.

Victor: Regarding Japanese power semiconductor stocks, I remember back in March, major Japanese semiconductor firm Rohm was reportedly in talks with Toshiba and Mitsubishi Electric about integration, to band together and make power semiconductors.

Fiona: I wasn't aware of that; I'll go back and research it. The power semiconductors I'm buying now are mainly focused on US and European stocks, where I've bought relatively more.

Victor: Actually, the power semiconductor space is quite broad. Personally, I view them all within the context of the entire "electricity revolution," roughly divided into three layers: the first layer is more power supply and power equipment companies, like Taiwan's Delta Electronics; the second layer is power semiconductor companies; the third layer is companies making materials like SiC and GaN. It's just that these companies themselves are very large in scale, with part of their business still in EV, automotive, and consumer electronics. So, is their AI content not that high?

Fiona: It's still quite low. The current situation is also because, as you mentioned, their scale is relatively large, so I think it's somewhat similar to MLCCs: if you say AI demand currently accounts for Murata's total revenue volume, it might be less than 10%. Both are similar in this regard. But the issue is that this segment is growing very fast, and the margin improvement will be relatively high. I think this might be why their valuations could see huge improvement through this small window, plus there's the AI sentiment, which is hard to escape.

And regarding Delta Electronics you just mentioned, I actually think Delta is really good too, it's just that the price is quite high now. It will definitely benefit from the power architecture in AI data centers, and as far as I know, its integrated solutions have a price advantage—good and cheap, so people will definitely buy.

Victor: Delta Electronics started rallying from Computex in May 2025, when Nvidia announced it would adopt 800V HVDC, and has risen about 8x since then. It was already one of Taiwan's top three stocks by market cap, so rallying that much more is quite夸张. People are now also looking for other targets, like Lite-On Technology and others.

Fiona: Yes, it's really夸张.

Victor: In European stocks, are there any power semiconductor companies worth paying attention to that you can introduce to us?

Fiona: I think there are two in European stocks worth a look. One is Infineon (IFX), which is the leader in European power semiconductors, though its stock price wasn't very strong last year, largely suppressed by the EV/automotive side. But my feeling is that it's at a relatively good price now, fundamentals are decent, and when the electricity revolution comes, plus Europe having its own data centers, Infineon should benefit quite a bit. Moreover, Europe is now pushing version 2.0 of its European Chips Act; Infineon was included in 1.0, and it should be supported in 2.0 as well barring surprises. Europe is weaker in AI, so I would lean more towards betting on these companies with advantages. There's also STMicroelectronics, but it leans more towards the power architecture side, a bit narrower. Personally, I'm currently buying more Infineon.

Victor: Recently, we've also been looking at Taiwan-listed power semiconductor related targets, such as Vanguard International Semiconductor. It's TSMC's "own son," with TSMC holding a 19% stake, mainly doing mature nodes. It formed a joint venture VSMC in Singapore with NXP from the Netherlands, and early this year also obtained GaN technology licensing from TSMC. However, it's mainly a foundry, so we need to observe whether it can directly benefit within the Nvidia ecosystem; most of its business is still in EV and traditional electronics. Fiona, are you looking at any other targets?

Fiona: I've looked at quite a few. The most important thing in power semiconductors is the two new third-generation semiconductor materials: SiC and GaN. They can withstand higher voltages, higher temperatures, and have lower losses, making them core, so I've been looking at derivatives.

After looking, I discovered a rather interesting fact: the raw material producers with the best price advantage right now are almost all in China. I previously bought a US stock, Wolfspeed. It went through a whole round of bankruptcy liquidation and restructuring last year. People assessed that it spent something like 6 billion or 10 billion to build a company capable of making 6-inch SiC, but its current market cap is less than one-third of what was invested then, so the thesis was to buy it. I was somewhat convinced by that logic at the time, and it's US domestic manufacturing, so I bought some. But later, after deeper background research, I found that although it possesses the world's only process solution for 6-inch SiC, its cost is still significantly higher than Chinese SiC. GaN has a very similar problem. So I've also recently looked at a few related Chinese companies, because their pricing is just too competitive.

And for instance, there's a Hong Kong-listed company called Innoscience, which is a GaN leader. It directly won a lawsuit against Infineon's China business, all the way to the Supreme People's Court, which will directly hinder Infineon's sales of all GaN-related products in China going forward. They've already reached this level, so I think we really should look at some Chinese companies.

Victor: I've talked to friends in the semiconductor industry, and they have a thesis: any sector where Chinese manufacturers can break in, it's best to avoid, because they can drive costs down extremely low, turning it into a fierce price war. Like Innoscience, I remember it's because GaN requires certain raw materials (aluminum, gallium), and China's raw material output is relatively high, giving it an inherent cost advantage.

Fiona: Right. I've looked at similar things, including Indium Phosphide (InP). Why does China have an advantage? You'll find that, first, it controls the mines, and second, its electricity is cheap, so it can push costs down very low, almost controlling the majority of production capacity, or offering the best prices, and often both simultaneously, making it quite difficult for other companies.

6. Optical Interconnects and the CPO Inflection Point: Don't Get Hung Up on Terminology, the Direction is Certain

Victor: Let's talk about the optical sector. Recently, A-share optical stocks have also surged strongly. You previously wrote a post ranking the optical field, with LITE and COHR in the first tier, and MRVL, AVGO, etc., in the second tier. That was posted in mid-May, before Jensen Huang's Computex shout-out for MRVL in June. How do you view the optical sector now?

Fiona: My thoughts haven't really changed much. I think optical interconnects are a direction AI must develop next. We could also see at Computex recently that all related manufacturers, no matter what chips they make, are working on an optical interconnect-related solution.

Objectively speaking, why optical interconnects? Because the distance and speed of electrical signal transmission in copper wires are approaching physical limits. For transmission between GPUs and between racks, it has exceeded what copper wires can handle, so optics is the only solution. As for which method to use: CPO (Co-Packaged Optics), NPO (Near-Packaged Optics), pluggable—these actually don't matter. The difference is just the length of copper wire between the optical engine and the switch; some are near, some far, some pluggable, some co-packaged. But the direction of optical interconnects is certain.

You asked about my ranking. I think looking at optics now, first, you need to see its positioning—whether it's that critical and irreplaceable; this is the most important point. Second, look at its capacity realization capability. As far as I know, there's a capacity shortage in optics, some due to raw materials, some due to production capacity, some a combination of both. The shortage is very severe, with many orders accumulated, but not many manufacturers can actually deliver.

So why do I like Lumentum and Coherent so much? Because these two companies possess both of the qualities I mentioned. And their original structure is relatively light, with market caps not too large, so the potential and multiples for a future breakout are better. Of course, recently you can feel the market's wind isn't really in optics, so it might take some time. If you haven't built a position yet, you can look for opportunities; I'm not in a hurry myself. But over a two to three-year timeframe, I see optical interconnects as the biggest growth point.

Mr. Z: Last Wednesday, Dylan Patel and his team at SemiAnalysis published a report saying that CPO optical communication is basically not on the radar for the second half of 2026, looking directly to 2027 or even 2028. I see most analysts online agree. If this wave of AI supercycle continues, what do you think the inflection point or catalyst will be to make the optical communication sector run up again?

Fiona: I saw that article; it was a shock to me at the time too, because I always read their research reports. I think his tone was a bit too absolute, and the impact of this matter has been somewhat magnified.

People might be confusing the terminology. For a while, if you looked at the market, everyone wasn't saying optical interconnects, but rather "CPO." Even my husband, who doesn't quite understand, would tell me about that CPO thing, and then I'd correct him: it's not CPO, it's optical interconnects; CPO is just one solution for optical interconnects. And earlier this year, if I remember correctly, Nvidia had already said they wouldn't use CPO, because everyone knows CPO is co-packaged. The optical interconnect components aren't expensive, but packaging them together with very expensive GPUs makes them almost unrepairable; the repair cost and difficulty are simply too high. So before the technology is absolutely mature, I don't think anyone should, or would, risk adopting the CPO solution.

And in optics, I think the most aggressive and leading player right now should be Google, not Nvidia. So if you want to see an inflection point, I think you can watch these two companies' adoption of solutions: if one day someone really starts using CPO, then CPO should have reached an inflection point for mass adoption.

But even if that day hasn't come, I think it's fine. Because in reality, our current deployments should naturally involve multiple optical interconnect solutions coexisting: NPO, pluggable, each suited for different scenarios—scale-up uses scale-up solutions, scale-out uses scale-out solutions. So I don't think the absence of CPO is the end of the world; it's just one of the solutions, and the technology isn't yet at a stage for large-scale application. We just wait for that day to come.

Mr. Z: That's quite interesting. I rarely hear people say Google would be a leading indicator in optics. What characteristic of Google makes that so?

Fiona: I think it starts with its TPU 8t architecture. I feel that Google has placed bandwidth and network interconnection on a higher dimension, with a more ingenious design approach. I previously wrote a tweet about it — Google simultaneously launched a new network architecture called the Virgo network, which actually uses more light.

VII. Sovereign AI: The Delisting of Fable 5 and Mythos 5, and the Jockeying Among Nations

Mr. Z: The hottest news recently is that Anthropic's Fable 5 and Mythos 5 were ordered to be taken down by the U.S. government. On the same day, China's DeepSeek, led by Liang Wenfeng, just released news that it raised $7.4 billion at a $50 billion valuation, with Liang Wenfeng personally contributing $3 billion, and the rest coming from Tencent and CATL. Interestingly, Alibaba actually couldn't invest in DeepSeek at all. What do you think about the current intervention of sovereign states in AI?

Fiona: I think this is quite related to the overall global landscape. We are in an era that appears globalized, but the relationships between everyone are no longer as harmonious as they were at the beginning of this century; it's more about cooperation amidst competition.

AI, as the next-generation productivity driver, has a fairly clear direction, and now every country hopes to have its own best AI. And AI is increasingly integrated with military operations. Whoever has better AI can discover network vulnerabilities, breach the other side's websites, and paralyze their traffic command systems, so AI has become a rather terrifying form of power. I feel that, to some extent, we have already reached this point, or are just on the verge of it.

So from the perspective of international politics, each country will definitely hide its strongest part, somewhat like nuclear weapons — only I can use it. Therefore, everyone hopes to develop their own rights and power to negotiate in this era. Both China and the U.S. are far ahead in model capabilities, but Europe does not have this. It has some models, but the gap is very, very large. I remember seeing someone say, who would even use those European models? The answer is that the government will procure them, somewhat like the concept of "government backstop" for early Chinese state-owned enterprises — no matter how poorly you do, I'll cover for you.

So the direction for Europe, or countries of this type, is to try to secure a relatively key position in the entire AI production chain: I can no longer squeeze into the model layer, so I can try to secure other things. Just like Jensen Huang said, there are five layers to the cake; you just try to get into whichever layer you can. Europe probably has a fairly realistic view of this now and will find some narrower positioning for itself.

VIII. How to Manage 28 Stocks: Concentrated Positions and the Style Transition from Crypto to U.S. Equities

Fiona: It's quite funny — just a couple of days ago, someone asked me if I held more than 20 stocks. I counted carefully and found I have 28. But later I realized many are overlapping. For example, I might buy a storage ETF, and also buy the three major storage companies, plus Kioxia. So if you break it down by sector, it's actually quite few, probably just about four sectors.

Victor: You now have as many as 28 stocks. How do you manage each position? How do you divide them by sector or by country's stock market? And what main themes are you particularly bullish on for the second half of the year?

Fiona: Let me clarify a concept here: 28 sounds like a lot, but in reality, probably 6 relatively large stocks account for over 50% of the position, so it's a fairly concentrated allocation. It's just that some are "observation positions." For example, Robinhood, which I've been discussing recently, is very non-core. But because I think its price has reached a certain level and might rebound, I bought some. That gets counted among the 28, but in terms of weighting, it's negligible.

My top two holdings right now are, of course, the storage sector and the optical interconnect sector. Recently, I've also been adding some materials sector stocks, like Ibiden, which we just mentioned. I'll also buy a few non-AI stocks, like Robinhood, which I added to the day before yesterday. Because my current portfolio is actually quite fragile — 90% are semiconductor stocks. If AI expectations aren't met, I should experience very severe drawdowns, so I want to appropriately add some tech-finance or other sectors for allocation.

Victor: I've been following Teacher Fiona's tweets for quite a while. You were very adept at trading in Crypto before, very skilled at "chasing the trend," able to catch the major trends of coins and get in where volume started to pick up, and you could hold for a very long time. Now that you've come to U.S. equities, what differences are there in your trading style?

Fiona: I think there are two different points.

One is a shift I made relatively early: I dared to buy the dip on the left side. In crypto, I was more of a right-side trader, because crypto is very dangerous — you can't trade the left side. Left-side trading is almost like helping the project team offload their bags. Many times you buy at a price that has dropped 99%, and it might never go up again. But U.S. equities are different: if it's a good company, it will rebound, and you can buy the dip. So I went against the "right-side only" habit I developed over three years in crypto and made some left-side trades, and the results were quite good.

Another point, which I only started to correct recently, is reducing my trading frequency. Because I still somewhat like to swing trade. I feel like if it seems unable to rise further, I want to sell and wait for the next trend to start before buying again. I was very good at this in crypto, but in the stock market this year, I've been failing continuously, because selling at any point ultimately meant selling too early and missing out. When the demand for a trend hasn't disappeared, with a good stock, you just need to hold on. This volatility is short-term noise, and it's very hard to trade. So in U.S. equities, I now try to trade as little as possible. I just plan to hold, and buy a bit more when there's a good opportunity. I will also do some selling — for example, to take out some money for living expenses, or to rebalance the portfolio. Like now, storage has risen to maybe 60% of my holdings, which is really a bit too high, so I'll sell some appropriately. But aside from that, I try to minimize trading as much as possible.

IX. Robinhood, Hyperliquid, and Crypto: Trading Demand Never Dies, Real Buybacks Are Key

Victor: Speaking of Robinhood, Teacher qinbafrank, whom we interviewed a few days ago, also mentioned it in a recent post. What narrative are you bullish on for HOOD right now? Their crypto business revenue proportion is decreasing, so are you more bullish on the incremental growth from prediction markets driven by the World Cup?

Fiona: I think cryptocurrency is currently a detractor for it, not a positive. That's also why it fell so badly after its April earnings report: the earnings weren't actually that bad, but the cryptocurrency part was just terrible, making the whole report look ugly, and the market punished it with a merciless plunge.

My entry point was when the Trump administration announced a policy related to Robinhood, allowing newborns and children to open investment accounts, and Trump himself has related Robinhood positions. I started building an observation position at that point. Additionally, I noticed two relatively good things: one is the event trading you mentioned, especially the huge incremental volume from the World Cup, which is a great seasonal driver; the second is some relatively good insider buying, which seems to be the first time since records have been kept. With these signals layered together, combined with its candlestick chart, Robinhood is a stock I think might break out, and it also helps diversify my semiconductor investments a bit.

Victor: So what's your overall view on Crypto? You bought Hyperliquid (HYPE) in March and April — is the logic similar to Robinhood?

Fiona: Yes, very similar. Both are based on the view that trading demand will definitely remain very strong. I think Hyperliquid and Robinhood share a very similar reason for success: it de-crypto-fied itself and embraced U.S. equities very well.

Think about it: people like me who left the crypto space at that time still held a lot of money in stablecoins like USDT and USDC. When I wanted to trade U.S. equities, of course, I could transfer the money to a fiat account and then remit it to a U.S. equities account, but there's a capital loss in between, and the bigger issue is tax problems. So if there was a crypto platform that could offer mainstream U.S. equity trading, it would be a very good choice. I bought Hyperliquid in March because I noticed the targets it selected were excellent. It listed all my favorite storage stocks at the time, and soon started listing optical communications stocks — essentially, it listed all the good sectors. Not to mention its trading experience, and its buybacks are real buybacks. Many crypto projects' buybacks are fake buybacks and real dumping, but it has genuinely been buying continuously, which I think is even more important.

Mr. Z: I've been very seriously studying Hyperliquid's business model over these past few months, and I'm truly impressed. Founder Jeff Yan comes from a high-frequency trading background, so the trading engine is built very solidly, but he completely doesn't care about marketing — not the typical project team operation. And builder code is also an innovation; you can build your own exchange on its HIP-3 protocol, like TradeXYZ's approach. I'm quite curious, Teacher, have you completely left the crypto space? Or are you just not touching it for now, waiting for BTC, ETF fund inflows, or conditions like DAT to improve before coming back?

Fiona: It should be quite difficult to go back. I think once you've enjoyed such a life in U.S. equities, it's very hard to return to crypto. And there's another great thing about U.S. equities: in crypto, some people can make money, but not people like me who are driven by research-driven trading. I use very public information, and relatively speaking, the opportunities U.S. equities offer are much larger and much better.

X. Advice for the Audience: Immerse Yourself in Research and Take Responsibility for Your Own Decisions

Mr. Z: Finally, I'd like to ask Teacher Fiona to offer some encouragement to our listeners, whether regarding investing or how to cope in the AI era.

Fiona: I think "sharing" might be a more appropriate word. This is also what I've found most useful along the way.

I used to read a lot of what others said — others telling you that you should buy this insider information play, that this will have some positive catalyst. I used to drive my trades with a lot of other people's input. But my style now is: I'd rather miss out on these insider opportunities (and often, insider info is a trap, not an opportunity at all). I'm more inclined to do what I believe is valuable, especially things with long-term value.

Because I think thematic investing, especially in something like AI, which is a generational opportunity that will truly change our generation, contains so many opportunities. So you need to immerse yourself in research, find a sector you like, and study it thoroughly. That way, when someone pitches you something or discusses something, you'll better understand what they're talking about, know where you align and where you differ, rather than just blindly following the crowd in your investments.

Of course, in a very good market, you can make money even with haphazard investing, and this has been happening a lot recently. But if you want to achieve long-term victory, with a win rate greater than 60%, you still need to have your own judgment. I think this is extremely, extremely important.

Mr. Z & Victor: Thank you very much, Teacher Fiona, for spending an hour talking with us about so much, and thanks to every listener who has stayed with us to the end. If you enjoyed this episode, feel free to follow 168X on X and YouTube, and share the show with more friends interested in macro, AI, and hard tech. We also wish everyone a healthy Dragon Boat Festival. See you in the next episode.