How should an investor who owns 28 stocks in the US, Japan, South Korea, and Taiwan bet on the second half of AI?

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

Japanese stocks are the shovel sellers of the AI era, searching for the next batch of winners in the Asian supply chain.

In late June 2026, volatility in AI and semiconductor supply chain stocks continued to intensify. Micron was scheduled to release its earnings report on June 24; Kevin Warsh's stance on future interest rate management was hawkish; and the South Korean KOSPI index broke through 9,000 points intraday for the first time on June 18. Amidst this intertwined macroeconomic uncertainty and hardware frenzy, 168X invited Fiona ( @nft_hu ), an independent research-based investor with expertise spanning memory, MLCCs, optical interconnects, power semiconductors, and global equities.

Fiona initially engaged in cryptocurrency trading, but decisively exited the market after the massive liquidation last October, shifting her funds to AI, which she heavily utilizes. She approaches this from a "first principles" perspective: working backward from Silicon Infrastructure's viewpoint to determine the essential requirements for the successful development of AI data centers. Currently, her holdings span US, Japanese, Korean, and Taiwanese stocks, with storage and optical interconnects being her two largest holdings, resembling a small, globally diversified semiconductor ETF. In this conversation, she offered a clear assessment: the most trustworthy people in the market are those like Jensen Huang, who are at the forefront of AI and possess the deepest understanding of the supply chain; and storage will be a persistent bottleneck in this cycle, with prices, demand, and profits remaining constant, while she largely ignores short-term fluctuations. She also shared a personal insight: reducing trading frequency, as all swing trading ultimately results in "selling too early" on good, low-cost positions.

I. Research Framework: From Cryptocurrency Transactions to Using First Principles to Determine Silicon-Based Demand

Mr. Z: Professor Fiona's research spans many areas, including storage, MLCC, optical interconnects, and 800V. Could you please briefly introduce yourself and explain how your research framework came about?

Fiona: Hello everyone, I'm Fiona. First of all, thank you very much to 168X for inviting me. I have read your outline, and it seems that you have thoroughly reviewed my tweets. Some of them were ones that I had posted but had forgotten about, and I think you have found them all.

I actually started out trading cryptocurrencies. But as you all know, the cryptocurrency market is getting smaller and smaller. In fact, when meme coins were all the rage, I realized it wasn't my preferred style, because I lean more towards research-driven trading. Then, after that incredibly absurd liquidation on October 11th last year, I left cryptocurrencies. I was quite decisive; I cleared out all my positions right then and there, feeling I needed to find a place for that money.

As a fairly heavy AI user, though I might not be able to compare to some of my coding friends, I wondered if I should start researching the AI I use. Because by the time it reaches us, it's already a model, a product; the key is everything that drives it. So, it was actually quite by chance that I started looking into AI, initially absorbing information like a sponge because I knew absolutely nothing about it.

The sector that really impressed me was storage. I feel very lucky because storage was experiencing a significant price surge at the time, both in the spot and futures markets, which garnered a lot of media coverage. I started researching: why was storage prices rising? Was it a short-term or long-term surge? That's how I entered the storage field, delving deeper and learning more, and increasing my position size. It has definitely been the most profitable sector of all my investments this year.

Because AI involves many intricate frameworks, I prefer to analyze it using first principles: what do silicon-based companies need to develop successful AI data centers? I work backward from their perspective. This gradually formed my own investment framework. Of course, I'm also a heavy Twitter user, so I created a small bot to catch up on trending topics online as a reference. This is roughly my introduction and research framework.

II. Don't Sell Off Storage: Jensen Huang's Shift, Micron's Financial Report, South Korea's Revaluation, and HBM4E

Mr. Z: Speaking of storage, you previously predicted Micron's stock price would rise from 240 all the way to 1,000, which is a demonstration of the "diamond hand" and value investing. What are your thoughts on this current storage market trend?

Fiona: Here I'd like to briefly quote something Jensen Huang said before. It must have been in January of this year at CES: The bottleneck in AI has now shifted from what we used to call insufficient computing power or insufficient GPUs to insufficient context.

His words had a profound impact on me. I thought to myself: what made someone who sells GPUs say that the bottleneck has shifted to context? Context, which is what everyone later became familiar with as storage. We can see that the new bottleneck in AI development actually requires a larger storage capacity to support it. I wonder if people using AI have noticed that it becomes less intelligent the more they use it? This is because its long context can cause some disorder. So there are many different opinions in the market, with everyone saying which position is most important; but I believe the most trustworthy person is someone at the forefront of AI and who understands the supply chain best, and that's Huang Renxun, whom I trust the most. So when he was so optimistic about storage and believed that context would be the next driver of AI development, I studied this part even more deeply.

The reason for the DRAM shortage is relatively easy to understand: there's a "scissors gap," a mismatch between supply and demand. The speed of computing power development far exceeds that of HBM (High Bandwidth Memory), so no matter how much storage is expanded or stacked, it can't keep up with the speed of GPUs. Therefore, the more storage there is, the better.

HBM is essentially a type of DRAM, and its production squeezes out a significant amount of existing capacity. This is because there's no time to immediately set up a new factory; we all know that building a memory plant costs hundreds of billions of dollars and requires a lot of time. Therefore, the fastest way is to "convert production": directly shifting existing DRAM production capacity to HBM production. This is the fastest method, but it also puts pressure on the supply of DRAM itself. So, if you look at all the memory products on the market, you'll ultimately find various shortages.

I think there's another trend now, and I don't know if everyone has noticed: some people are starting to hype up NAND as the next storage gap. I think this is mainly driven by economic benefits. Although it has some disadvantages compared to HBM, it's much cheaper, costing perhaps only one-fiftieth of HBM's current price. Given such economic advantages, people will still be inclined to switch, creating new gaps and new demands. This includes figuring out how to connect the two and how to make a transition, and I think there will be new opportunities in all of this.

Mr. Z: Right now, the market's biggest focus is Micron's earnings report on June 24th, coupled with yesterday's Fed decision, which revealed Warsh's attitude towards future interest rate management. You probably also have your own storage holdings; how will you manage them during this period? Do you think next week might be very volatile?

Fiona: I understand. I probably wouldn't trade in this situation, because I realized later this year that almost all my trades were wrong. All the swings are just to sell off those good positions you bought at low cost.

So, as I said, I'm more likely to stick to first principles analysis. I think storage is the bottleneck; it's a continuous process—prices are continuous, demand is continuous, and profits are continuous. Therefore, I basically ignore short-term fluctuations. If there are fluctuations due to earnings reports or these past few days, I actually think the market will become more volatile, and I'd like to take advantage of that volatility to buy more storage—that would be one of my trading strategies.

Mr. Z: Huang Renxun's Asia trip at the end of May and beginning of June only included Taiwan and South Korea. Some Japanese newspapers are writing that Japan might miss this AI supercycle. Because, if I remember correctly, in the 80s and 90s, Sony was involved in memory chip manufacturing for a while, but later couldn't compete, so it became Taiwan and South Korea, with SK Hynix and Samsung leading the way. So Japan is lagging behind in this area. Speaking of South Korea, the Lee Jae-myung government introduced a new law requiring South Korean companies to destroy treasury stock, because chaebol families have long controlled companies with very few shares, and too much external liquidity isn't good for EPS. What are your thoughts on this wave of reforms in South Korea, and how will it be reflected in KOSPI's performance?

Fiona: Actually, this has already been confirmed. I think I wrote about it back in February, when South Korea passed the corporate reform law. It's just that back then, people weren't paying as much attention to the Korean stock market; it was around 6,000 points then, and today it just broke 9,000 points, right? So you see, the Korea Discount is gradually being leveled off. Of course, SK Hynix is also going to issue ADRs on the US stock market, so I think there might be another wave of excitement ahead. But the impact of this event has already occurred over the past four months.

Victor: So what are your thoughts on the current three giants of the storage industry: Micron, SK Hynix, and Samsung? Regarding the latest HBM4E, while Samsung has been the fastest in terms of speed and press releases, rumors suggest that SK Hynix is more likely to supply in large quantities. Today, SK Hynix also announced that it has begun delivering HBM4E samples to customers. What are your predictions for the future impact of HBM4E?

Fiona: I also tend to believe it's SK Hynix. Considering both its past product performance and stock price, I think it's definitely Hynix. Of course, Samsung is also very good, so you could buy both.

III. The fundamental differences between MLCCs and storage: Compatibility requirements vs. the more the better.

Mr. Z: You mentioned before that there are significant differences between storage and MLCCs. Storage models are relatively limited in scope, while MLCCs have physical limitations. Could you explain this in more detail?

Fiona: It's actually very simple. As I just mentioned, AI training models require more storage, and this demand is currently inexhaustible. So everyone is constantly piling up HBM and DRAM to make fuller use of computing power. The logic is very crude: the bigger the better, the more the better.

However, MLCCs (multilayer ceramic capacitors) are different. Their demand growth is "complementary," which is how I define it. For example, if I buy a Peak racket, it might give me two balls as a gift. This complementary quantity is fixed; it's not a case of the more the better. Therefore, the demand for MLCCs is quantifiable, and you can calculate its total volume as server racks are shipped. There are many different calculation methods now, but one that I've seen is relatively widely accepted, roughly a 3.x increase in demand.

Moreover, as you mentioned, there are many different types of MLCCs. This surge in demand and price increases actually occurred in a small segment of higher-end MLCCs: these have more complex manufacturing processes, may have hundreds of layers, and can only be produced by a few manufacturers. This very small segment received a flood of orders. Therefore, I think it's somewhat different from the "general price increase" seen in the memory industry.

Of course, when a concept becomes popular, you can make money by buying anything related to it. Here's an interesting example: a friend in my group said he couldn't buy Murata, which I recommended because it's a Japanese stock, so he bought a Chinese A-share company called Sinocera Materials, which performed well, and he was quite happy. However, I had done some research at the time, and this company actually had a relatively low correlation with high-end MLCCs. It's just that recently the market has been rising by sector, so anything related to MLCCs has benefited. Personally, I prefer to buy those that truly benefit, because I don't want to trade frequently.

IV. Japanese stocks are the "shovel sellers" of the AI era: Murata, Taiyo Yuden, Kioxia, and Ibiden

Victor: Many leading MLCC manufacturers are actually based in Japan, such as Murata and TDK, while Taiwan has Yageo. Japanese stocks are currently generally undervalued, and if the US starts raising interest rates after Warsh's Federal Reserve (FOMC) meeting, a stronger dollar and a weaker yen would actually benefit Japanese export companies. What's your view on Japanese stocks, and what's your current portfolio allocation?

Fiona: First of all, I really like Japanese stocks, maybe even more than I like US stocks, especially recently. I feel that the market is still quite overvalued, and the cheap valuations of Japanese stocks give me a greater sense of security, although this may be a misconception.

I've written two long articles about MLCCs. One is about Murata, which is undoubtedly number one; the second is about Taiyo Yuden, which I believe is the purest MLCC stock among Japanese stocks. Murata has a slight issue: its capacitor segment (which includes MLCCs) is too large. Although MLCCs benefit significantly, the improvement in overall revenue and profit margin isn't immediately apparent from one or two quarters' reports. Taiyo Yuden is different; it's relatively smaller, more focused, and more volatile. I remember its revenue increased by about 20% due to high-end MLCCs, but its operating profit increased by about 90%.

However, I currently only hold Murata shares. It's not that I don't want to buy Taiyo Yuden, but I haven't been able to get any of the orders I placed; it's just too aggressive.

Speaking of Japanese stocks, I'd like to mention that I think they've played a very good role as a "shovel seller" in the AI era. They may not have the best business models, but if you look closely, whether it's semiconductor manufacturing equipment, materials, chips, passive components, or packaging substrates, they have excellent companies and strong positions in every环节 (link/stage). That's why I've been buying quite a lot of Japanese stocks recently, and I'll continue to increase my holdings if there are good pullback opportunities.

Let me talk about my top three holdings right now. One is Kioxia (285A), which is also a memory concept stock; another is Murata Manufacturing Co., Ltd. (6981), which makes MLCCs; and the last one is I'm not entirely sure of its Chinese name, I just call it Ibiden (揖斐電, 4062), which makes packaging substrates. I wrote a post about it today, you can check it out if you're interested.

Victor: I saw your tweet about Ibiden today. You said you bought it and forgot about it, but then found out it kept going up. That's a pretty interesting position.

Fiona: Yes, because my husband buys my Japanese stocks for me, and I rarely look at that account. I just have him buy whatever looks good for me, so I can hold onto it more easily. So one day I checked and wondered why 4062 had risen so much? I had actually forgotten why I bought it, because I had only bought it once before. Later, I realized, wow, that's great! So I said we'd buy more if there was a pullback, and I was lucky enough to do so recently.

V. Power Semiconductors, 800V and Third-Generation Semiconductors: Why China Has Gained a Cost Advantage

Victor: You started listing companies related to 800V DC, power semiconductors, and compounds like gallium nitride and silicon carbide around May. Is power semiconductor a sector with the potential for explosive growth? Because when the overall semiconductor market crashed in early June, companies like Navitas, Wolfspeed, and ON Semiconductor experienced significant pullbacks. Could this be a good entry point?

Fiona: Let me share with you first that my current power semiconductor holdings are mostly still around cost price, not yet in the profit margin. I initially bought them for two reasons. First, I looked at the new 800V HVDC (high-voltage direct current) architecture, and among the sectors that benefit from it, power semiconductors are unavoidable, so it has fundamentals and demand-driven growth. Second, when I was looking at it, I felt that power semiconductor prices were relatively cheap, having not risen much. So frankly, we're somewhat "betting on the future": I hope it will rise, I believe it will rise, but it's not certain it will.

Victor: Regarding the power semiconductor sector in Japanese stocks, I remember that in March, it was reported that Rohm, a major Japanese semiconductor manufacturer, was in talks with Toshiba and Mitsubishi Electric to merge and form a joint venture to produce power semiconductors.

Fiona: I don't know that yet. I'll look into it when I get back. The power semiconductor stocks I'm currently investing in are mainly from the US and European markets, and I'm buying relatively more of those.

Victor: Actually, the power semiconductor field is quite broad. I personally view them all within the broader context of the "electrical revolution," roughly dividing them into three tiers: the first tier consists of companies more focused on power supplies and electrical equipment, such as Delta Electronics in Taiwan; the second tier comprises power semiconductor companies; and the third tier consists of companies that work on materials like silicon carbide and gallium nitride. However, these companies are generally very large, and a portion of their business is in EVs, automotive, and consumer electronics. Does that mean their AI content isn't as high?

Fiona: It's quite low. The current situation is because, as you mentioned, its scale is relatively large, so it's somewhat similar to MLCCs: if you say AI demand currently accounts for less than 10% of Murata's total revenue, then these two are similar. But the problem is that this part is growing very quickly, and the profit margin improvement will be relatively high. I think this is probably why their valuations will see a huge improvement due to this small window of opportunity. Also, there's the AI sentiment, which is also quite significant.

Then you just mentioned Delta Electronics. I actually think Delta Electronics is really good, it's just that its prices are quite high now. It will definitely benefit from the power architecture of AI data centers, and as far as I know, its complete system solutions have a price advantage. It's good and cheap, so everyone will definitely buy it.

Victor: Delta Electronics' stock price has risen about eightfold since May 2025, when Nvidia announced its plans for 800V HVDC at Computex. It's already one of the top three stocks by market capitalization in Taiwan, so this much growth is quite remarkable. People are now looking for other targets, like Lite-On Technology, etc.

Fiona: Yes, that's really exaggerated.

Victor: In the European stock market, are there any power semiconductor companies worth paying attention to? Could you introduce them to us?

Fiona: I think there are two European stocks worth considering. One is Infineon (IFX), the leading power semiconductor company in Europe. Its stock price wasn't great last year, largely due to pressure from the electric vehicle sector. However, I feel it's currently at a relatively good price with strong fundamentals. With electricity reforms and Europe's own data centers, Infineon should benefit significantly. Furthermore, Europe is pushing for its European Chips Act 2.0, which included Infineon in version 1.0, and should be supported in version 2.0 as well. Europe's AI sector is relatively weak, so I tend to favor companies with strengths in these areas. Another option is STMicroelectronics (STMicroelectronics), but it focuses more on power architecture, making its niche market narrower. Personally, I currently hold more Infineon shares.

Victor: We've also been keeping an eye on power semiconductor stocks in Taiwan recently, such as VIS (Vancouver Semiconductor). It's a subsidiary of TSMC, with TSMC holding a 19% stake. It mainly focuses on mature processes and has a joint venture with NXP in Singapore called VSMC. Earlier this year, it also obtained a gallium nitride (GaN) technology license from TSMC. However, it primarily focuses on foundry services, so we need to observe whether it can directly benefit from the NVIDIA ecosystem. Most of its business is still in the EV (electric vehicle) and traditional electronics industries. Fiona, are you looking at any other stocks?

Fiona: I've been reading quite a lot too. The most important materials in power semiconductors are these two new third-generation semiconductors: silicon carbide and gallium nitride. They can withstand higher voltages and higher temperatures, and have lower losses, so they've become the core materials, and I've been looking at the derivatives as well.

After reviewing the data, I discovered a rather interesting fact: almost all the raw material producers with the most price advantages are currently based in China. I previously invested in a US-listed company, Wolfspeed. It underwent a complete bankruptcy liquidation and restructuring last year. It was estimated that it invested around $6 billion or $10 billion to create a company capable of producing 6-inch silicon carbide wafers, but its current market capitalization is less than a third of that initial investment, making it a worthwhile investment. I was somewhat convinced by this logic, especially since it was manufactured in the US, so I bought some shares. However, after conducting deeper research, I discovered that although it possesses the world's only 6-inch silicon carbide manufacturing process, its costs are still significantly higher than those of Chinese silicon carbide. Gallium nitride faces a very similar issue. Therefore, I've recently been looking at a few related Chinese companies because their prices are incredibly competitive.

Furthermore, there's a Hong Kong-listed company called Innoscience, a leading manufacturer of gallium nitride (GaN) products. They won a lawsuit against Infineon's China operations, taking it all the way to the Supreme People's Court. This will directly hinder Infineon's sales of all GaN-related products in China. They've gone to such lengths, so I think we really should take a look at some Chinese companies.

Victor: I was talking to some friends in the semiconductor industry, and they have a thesis that it's best not to enter any sector where Chinese manufacturers can easily penetrate, because they can drive costs down to very low levels, leading to a fierce price war. For example, with Innoscience, I remember it was because gallium nitride requires certain raw materials (aluminum and gallium), and China has a relatively high production volume of these raw materials, giving it a cost advantage.

Fiona: Yes. I've looked at similar examples, including indium phosphide (InP). Why does China have an advantage? You'll find that, firstly, it controls the mines, and secondly, its electricity is cheap. So it can keep costs very low, almost controlling most of the production capacity or the best prices, and often both of these factors occur simultaneously, making it difficult for other companies to compete.

VI. Optical Interconnection and CPO Inflection Point: Don't get bogged down in terminology; the direction is what matters.

Victor: Let's talk about the optical sector. Recently, optical stocks in the A-share market have been rising sharply. You previously published an article ranking the optical technologies, with LITE and COHR in the first tier and MRVL and AVGO in the second tier. That article was published in mid-May, before Huang Renxun recommended MRVL at Computex in June. What's your view on the optical sector now?

Fiona: My thinking hasn't changed much. I believe optical interconnects are a direction that AI must develop in the future. At Computex recently, you could see that all the relevant manufacturers, regardless of what chips they make, are working on optical interconnect solutions.

Objectively speaking, why do we need optical interconnects? Because the distance and speed at which electrical signals can be transmitted through copper wires are already approaching physical limits. Transmission between GPUs and between server racks exceeds the capacity of copper wires, making optical interconnects the only solution. As for the specific method used—CPO (Co-packaged Optical), NPO (Near-packaged Optical), or pluggable—these are actually not important. The only difference lies in the length of the copper wires between the optical engine and the switch; some are closer, some farther, some are pluggable, and some are co-packaged. But the direction of optical interconnects is certain.

You asked me about the order of priority. I think when looking at optics now, the most important thing is to consider its positioning – whether it's truly crucial and irreplaceable. The second is its ability to realize production capacity. As far as I know, the lack of production capacity in optics is due to various reasons: some are raw materials, some are production capacity issues, and some are a combination of both. The shortage is very severe, with many orders piling up, but the number of manufacturers actually able to fulfill them is actually quite small.

So why do I like Lumentum and Coherent so much? Because they both possess the two qualities I mentioned. Furthermore, their original architectures were relatively lightweight, and their market capitalization wasn't too large, suggesting good potential for explosive growth. Of course, recently I've felt the market's focus isn't on optical interconnects, so it might take some time. If you haven't built a position yet, you can look for opportunities; I'm not in a rush myself. But over the next two to three years, I believe optical interconnects will be the biggest growth driver.

Mr. Z: Last Wednesday, Dylan Patel and his SemiAnalysis team released a report stating that they're basically not looking at CPO optical communications in the second half of 2026, but are looking ahead to 2027 or even 2028. I see that most experts online agree with this. If this AI supercycle continues, what do you think will be the inflection point or catalyst that propels the optical communications sector even higher?

Fiona: I saw that article, and it shocked me at the time because I've always read their research reports. I felt his tone was a bit too absolute, and the impact of this matter was somewhat exaggerated.

There might be some confusion regarding the terminology. For a while, you could see people using "CPO" instead of "optical interconnect." Even my husband, who wasn't very familiar with it, would sometimes say "CPO" to me, and I would correct him: it's not CPO, it's optical interconnect. CPO is just one solution for optical interconnect. Furthermore, earlier this year, if I remember correctly, NVIDIA stated they wouldn't use CPO because, as everyone knows, CPO is a co-packaged solution. Optical interconnect components aren't expensive, but when packaged with very expensive GPUs, they become almost unrepairable, making repair costs and difficulties extremely high. Therefore, until the technology is absolutely mature, I don't think anyone should, or would risk, adopting the CPO solution.

Furthermore, in terms of optical technology, I think the most aggressive and leading company right now is Google, not Nvidia. So if you want to see a turning point, I think you can look at how these two companies adopt their solutions: if one day people really start using CPO, then CPO should have reached a turning point where it can be used in large quantities.

But even if that day doesn't come, I don't think it matters. Because in reality, our current deployments should already include multiple optical interconnect solutions: NPO, pluggable, which are suitable for different scenarios—scale-up solutions for scale-up, and scale-out solutions for scale-out. So I don't think the absence of CPO is the end of the world; it's just one solution, and the technology isn't yet ready for large-scale application. We'll just wait for that day to come.

Mr. Z: That's quite interesting. I rarely hear anyone say that Google will be a leading indicator in the field of light. What is it about Google that's the reason?

Fiona: It probably started with its TPU 8t architecture. I think Google placed bandwidth and network interconnection on a higher level, and the design was more sophisticated. I wrote a tweet before about how Google also launched a new network architecture, Virgo network, which actually uses more light.

VII. Sovereign AI: The Removal of Fable 5 and Mythos 5 and the Positioning Game Among Various Countries

Mr. Z: The hottest news recently is that Anthropic's Fable 5 and Mythos 5 have been ordered to be removed from app stores by the US government. On the same day, DeepSeek, founded by Liang Wenfeng, announced that it had raised $7.4 billion at a valuation of $50 billion, with Liang Wenfeng personally investing $3 billion, and the rest from Tencent and CATL. Interestingly, Alibaba couldn't actually invest in DeepSeek. What are your thoughts on the current interference of national sovereignty in AI development?

Fiona: I think it's more related to the overall global landscape. We live in what appears to be a globalized era, but the relationships between people are not as harmonious as they were at the beginning of this century; there's more cooperation amidst competition.

As the next generation of productivity, AI's direction is quite clear, and every country now hopes to have its own best AI. AI is increasingly integrated with military operations; whoever has better AI can discover network vulnerabilities, breach enemy websites, and paralyze their traffic control systems. Therefore, AI has become a rather formidable force. I think we've already reached that day to some extent, or are just around the corner.

In terms of international politics, countries will inevitably keep their strongest assets hidden, somewhat like nuclear weapons—only they themselves can use them. Therefore, everyone wants to develop their negotiating power and strength in this era. China and the United States have a significant lead in model capabilities, but Europe doesn't. It has some models, but the gap is enormous. I remember seeing someone say, "How come anyone uses those European models?" The answer is that governments purchase them, somewhat like the early Chinese state-owned enterprises' "government guarantee" concept—no matter how poorly you perform, I'll bail you out.

Therefore, the direction of Europe, or rather, these types of countries, is to secure a crucial position in the entire AI production chain: if I can't secure a position in the model itself, then I can secure a position in other things. Just like Huang Renxun said there are five layers of a cake, they'll secure whichever layer they can. Europe should also be quite realistic about this and will find some narrower niches for itself.

8. How to manage 28 stocks: concentrated holdings and style shift from crypto to US stocks

Fiona: It's quite interesting. A couple of days ago, someone asked me if I owned more than 20 stocks, so I counted them carefully and found I had 28. But then I realized many of them were duplicates. For example, I buy a storage ETF, and I also buy shares from three storage companies, and I also bought Kioxia. So if I divide them into sectors, they're actually quite few, maybe just four sectors.

Victor: You now have 28 stocks. How will you manage each position? How will you categorize them by sector or by country's stock market? What are the main themes you are particularly optimistic about for the second half of the year?

Fiona: Let me clarify a concept here: 28 stocks sounds like a lot, but actually, a significant portion might consist of just 6 stocks that account for more than 50% of the portfolio, making it a relatively concentrated position. Some are "observation positions," like Robinhood, which I've been discussing these past few days. It's not part of the main theme, but because I felt its price had reached a certain level and might rebound, I bought some. These are included in the 28, but in terms of weighting, they are negligible.

My two largest holdings are currently in the storage and optical interconnect sectors. I've also been adding some materials stocks recently, such as Ibiden, which I just mentioned. I'll also be buying some non-AI stocks, like Robinhood, which I added to the day before yesterday. Because my current portfolio is quite vulnerable—90% of it is in semiconductor stocks—if AI expectations don't meet expectations, I'll likely experience significant drawdowns. Therefore, I'm considering adding some fintech or other sectors to my portfolio.

Victor: I've been following Fiona's tweets for a while now. You were very skilled at trading in Crypto, especially "trend chasing," able to catch major trends in cryptocurrencies, get on board when volume spikes, and hold for a very long time. Now that you're trading in US stocks, how has your trading style changed?

Fiona: I think there are two different points.

One change I made relatively early was that I dared to buy on the left side of the price drop. In the crypto world, I'm generally a right-side trader because the crypto market is very risky; you can't trade on the left side. Trading on the left side is practically helping the project team take over their losses. Often, if you buy at a 99% drop, it might not even rise again. But the US stock market is different: if it's a good company, it will rebound, and you can buy on the dip. So, I went against my three-year habit of "only trading on the right side" in the crypto world and did some left-side trading, and the results were quite good.

Another thing I've recently started to correct is reducing my trading frequency. I used to like swing trading, thinking that if a stock seemed to be stalling, I'd sell and wait for the next trend to start before buying back in. I'm very good at this in the crypto market, but I've failed consistently in the stock market this year because selling at any point always resulted in missing out. When the demand for a trend hasn't disappeared, if you buy a good stock, you just need to hold it; the fluctuations are short-term disturbances, making trading difficult. So, in the US stock market, I'm now trying to trade as little as possible, just holding and buying more when a good opportunity arises. I will also make some sales, for example, to cover living expenses or to rebalance my portfolio. For instance, storage has now reached about 60% of my portfolio, which is a bit too high, so I'll sell some. But I'm trying to minimize other trading activities now.

9. Robinhood, Hyperliquid, and Crypto: The Demand for Transactions Remains Unchanged; Genuine Buybacks are the Key

Victor: Speaking of Robinhood, qinbafrank, whom I interviewed a few days ago, also mentioned it in a recent article. What are your current positive points regarding Robinhood's narrative? Their crypto revenue is decreasing, so are you more optimistic about the growth in the prediction market driven by the World Cup?

Fiona: I think cryptocurrency is now a negative rather than a positive factor for it, which is why it fell so badly after its April earnings report came out: the earnings report wasn't actually that bad, but the cryptocurrency part was so terrible that it made the whole earnings report look bad, and the market gave it a very merciless drop.

My entry point was when the Trump administration announced a policy related to Robinhood that allowed newborns and children to open investment accounts. Trump himself also had a Robinhood position, and I started building a watch position at that point. In addition, I noticed two positive things: first, as you mentioned, event trading, especially the significant increase from the World Cup, which is a great seasonal driver; second, there were some good insider buying, which is probably the first time since I started observing it. These signals combined, along with its candlestick chart, made Robinhood a stock I felt might break out, and also a way to diversify my semiconductor investments.

Victor: What's your overall view on Crypto? You bought Hyperliquid (HYPE) in March and April. Is that similar to your logic with Robinhood?

Fiona: Yes, very similar. They both believe the demand for trading will remain strong. I think Hyperliquid and Robinhood have a very similar reason for their success: they de-cryptofied and embraced US stocks very well.

Imagine, back when I left the crypto world, much of my money was still held in stablecoins like USDT and USDC. If I wanted to trade US stocks, I could transfer the money to a fiat currency account and then to a US stock account, but there would be losses along the way, and more importantly, there would be tax issues. Therefore, a cryptocurrency platform that offered mainstream US stock trading would be an excellent choice. I bought Hyperliquid in March because I noticed its excellent stock picks. It listed all the storage stocks I loved at the time, and soon after, it started listing optical communication stocks—basically, it covered all the good sectors. The trading experience is also excellent, and its buybacks are genuine. Many crypto projects use buybacks as a pretext to dump shares, but Hyperliquid is genuinely buying, which I think is even more important.

Mr. Z: I've been studying Hyperliquid's business model very carefully these past few months, and I'm truly impressed. Founder Jeff Yan comes from a high-frequency trading background, so his trading engine is incredibly solid, but he completely ignores marketing—not the typical approach of a project team. The builder code is also innovative; you can build your own exchange on its HIP-3 protocol, similar to TradeXYZ's strategy. I'm curious, have you completely left the crypto world, or are you just temporarily staying away, waiting for BTC, ETF inflows, or DAT to improve before returning?

Fiona: It's probably going to be difficult. I think you've already enjoyed that kind of life in the US stock market, so it's hard for you to go back to the crypto world. And the US stock market has one great thing: some people can make money in the crypto world, but not someone like me who relies on investment research to drive trading. I use very publicly available information. Relatively speaking, the opportunities offered by the US stock market are much greater and better.

10. Advice for the audience: Immerse yourself in the research and take responsibility for your decisions.

Mr. Z: Finally, I would like to ask Fiona to encourage our listeners, whether in terms of investment or how to cope with the AI era?

Fiona: I think "sharing" might be a more appropriate word. It's also the most useful point I've found along the way.

I used to rely heavily on what others said, like "you should buy this stock based on insider information" or "this stock will have some positive news." I would use a lot of other people's opinions to drive my trading. But my current style is: I'd rather miss out on these insider opportunities (besides, many times insider information is a trap, not an opportunity at all). I prefer to do things that I believe are valuable, especially those with long-term value.

Because I believe that thematic investing, especially in sectors like AI that truly offer generational opportunities that will change our lives, offers so many possibilities. Therefore, you need to immerse yourself in research and find a sector you enjoy to study thoroughly. This way, when others are promoting or discussing something to you, you'll better understand what they're talking about, and you'll know where you and they have similarities and differences, instead of blindly following the crowd in your investments.

Of course, in a very good market, you can make money by making random investments, and this has been happening recently. But if you want to achieve long-term success and a win rate greater than 60%, you still need to have your own judgment, and I think that's extremely important.

Mr. Z & Victor: Thank you so much, Fiona, for spending an hour talking to us. And thank you to every listener who's listened all the way here. If you enjoyed this episode, please subscribe to 168X on Weibo and YouTube, and share the show with more friends interested in macroeconomics, AI, and hard technology. Happy Dragon Boat Festival to everyone, and see you next time!