Understanding AI Doesn't Mean Understanding the Market! How to Allocate in the Second Half with Hardware and Software? From Big Tech, Intel to Military and Healthcare

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

Guest: Investment TALK Jun (@TJ_Research)

Position Equals Opinion: Hardware Should Be Reduced but Not Cleared, No Need to Be Pessimistic in the Second Half

By July 2026, AI trading is at a delicate juncture. Over the past year, the market has rewarded Nvidia, storage, and semiconductors, while the big tech companies that have invested heavily in CapEx have faced repeated skepticism. Now, with significant deleveraging in semiconductors and storage, news of Meta selling computing power has shocked the market, while at the same time, Anthropic is reportedly accelerating its IPO process. Is the market pricing a bubble, or is it pricing a shift in the storyline?

In this episode of 168X, we invite Investment TALK Jun (@TJ_Research), a macro researcher and US stock investor with 170,000 YouTube subscribers, who has been tracking AI implementation, big tech earnings, and Federal Reserve policies. In this nearly hour-long conversation, he shares a poignant insight: understanding AI does not mean understanding the market; understanding the market does not guarantee a rise. He breaks down the complete logic from Dylan Patel's interview to Amazon, explaining why Meta selling computing power actually proves a shortage of computing power, why Oracle was wrongly punished, why he only bets on Intel in hardware, and how to allocate a "dual repair" investment portfolio in the current high emotional climate.

1. Understanding AI Does Not Mean Understanding the Market: The Logic Line from Anthropic to Amazon

Mr. Z: Today, 168X is very honored to invite Investment TALK Jun. TALK Jun has been researching tokens, the chase for open-source models, the ebb and flow of major LLMs, and the US stock market. Given the current market situation, is the market slowly shifting its focus to companies that can truly convert AI into revenue, cash flow, and productivity? You recently said, "Understanding AI does not mean understanding the market; understanding the market does not guarantee a rise." Is this your biggest realization in the past year?

Investment TALK Jun: First of all, thank you for the invitation. That statement is not my biggest feeling, but rather a reflection: I see some people choosing the right targets and direction, but the price ultimately still falls or does not rise, which is not pleasant. I’ll take one of my holdings this year as an example, which is Amazon.

This logic line goes back to last summer, August and September. At that time, I listened to a very popular episode of SemiAnalysis, an interview with Dylan Patel, where he shared his views on two major model companies, with very clear opinions: the market was very fond of OpenAI, but had low understanding of Anthropic. However, Anthropic has a very clear monetization path; they are focused on the North American software engineer market, a market with $20 trillion in annual labor costs, monetizing directly to B2B, with a very clear product rollout roadmap. He was less favorable towards OpenAI because OpenAI wanted to do everything.

It was later validated: OpenAI wanted to get into hardware, and their C-end ambitions for agentic commerce still lack direction; recently there was news about their collaboration or testing with Walmart, but the conversion was not that good. Step by step, it became clear that OpenAI's implementation direction was not clear, merely piling up computing power. Only when Anthropic excelled in coding did OpenAI start to catch up, and we saw Codex emerge, performing very well, as it was actually following in Anthropic's footsteps.

After listening, I thought: following this logic line is actually very simple. First, do you agree with Dylan Patel's view? I do. At that time, Anthropic's ARR was already ten times a year, but the scale was still small, unlike the latest data of $40 billion or $50 billion. Following the logic, the most direct question is: who provides Anthropic with chips and computing power? The answer is Amazon.

At that time, another logic line the market was running was that big tech's early capital expenditures had no ROI. Since the market questioned whether big tech's investments had returns, it made no sense that companies like Amazon, which invested in companies and sold chips and provided computing power, had such good implementation and revenue growth of more than ten times a year, and the market would not like Amazon.

The market's answer was: in the early days of the year, Amazon performed very well, but then there was a significant pullback. When I expressed that view, Amazon's YTD was just like the S&P 500, and the recent rebound has only slightly improved.

From establishing a viewpoint, deducing to investment targets, combining the market's logic line for this target at that time, and judging that this logic line might reverse, looking back now, I see no major issues. At that time, no one was talking about Anthropic's implementation, and I even thought I had some edge data and information, but that does not mean it would ultimately reflect in stock returns. Although as of today's close, Amazon's returns are still okay, they have not performed as well as hardware companies.

Even if the logic line is deduced very well, when you measure the returns of a target over a period, it does not mean that what you originally thought will reflect during that period. From an investment perspective, whether the target you choose will be liked by the market in the next six months or even a year, and whether the storyline will fall on it, at least in value investing, is truly uncontrollable.

2. Three Types of Timing and Anthropic IPO: The Market Will Not Demystify, Only Linearly Extrapolate

Mr. Z: So basically timing is everything? Market timing is very important, and business timing is also very important. How should investors solve this problem? The second question is, Anthropic is accelerating its IPO process and is expected to go public as early as October this year. I worry this might be a moment of demystification for the industry: after going public, many things will come to light, and if revenue and ARR growth do not meet expectations? After all, since the end of last month, semiconductors have still been in a downturn. As more information about Anthropic is disclosed, will the entire industry's charisma decline significantly?

Investment TALK Jun: I do not mean to express that timing is everything because timing is layered. As investors, we can grasp and judge the timing of the industry and the timing of the company's fundamentals; what we cannot grasp is the timing of the market trading logic line, which is the most difficult layer. Often, what investors should do is wait for this timing to arrive, rather than trying to predict when it will come.

As for demystification, I personally do not worry that the market will demystify Anthropic. When the market heard that Anthropic was moving towards profitability, my judgment at that time was that they were preparing for this year's IPO. Because once the data is ready, whether for roadshows or the S-1 filing, the market loves to do one thing: linear extrapolation: last Q4 was still losing money, Q1 was close to breaking even, Q2 was very profitable, and now the market is talking about Q3 where even stock incentives can make a profit, then the market will immediately linearly extrapolate that Q4 2026 will be even more profitable, and 2027 will be very profitable, with EPS growth two to three times or even higher: because if revenue growth is still four to five times, EPS growth could be six to seven times. Therefore, if a large company can show such EPS growth, I actually believe it will not demystify, but rather the market will find it very exaggerated.

So today, with the news that Anthropic has started preparing for its IPO, I think it makes sense. In the past two quarters this year, they focused on landing price increases, not hesitating to buy computing power and sell tokens at very high short-term rental prices to boost revenue first and then solve profitability issues. By the time they go public, the financial reports will look very good.

3. Anthropic's Strongest Point: $20 Trillion Software Engineer Market with the Wealthiest Clients

Mr. Z: What do you think is the strongest point that allows Anthropic to do this? Is it model capability, API pricing, or as you mentioned, their focus on enterprise sales?

Investment TALK Jun: API pricing is a result of supply and demand. Depending on which version of the model, its gross profit may reach 70% to 80%, which is already at the gross profit level of standard software companies. To reach such a high level is certainly determined by market supply and demand. But why is the market willing to give them this price? Ultimately, it comes down to their model.

This goes back to their early focus on coding implementation. Why coding? Because that is the best and largest application market they see. This is their smartest point: first, the market itself is huge, with $20 trillion in labor costs, and this market is almost unique to North America; no other region has such a large scale. Why is it unique to North America? Because this $20 trillion expenditure mostly comes from tech companies, which spend so much money on software engineers every year. During the pandemic, software engineer salaries and demand rose in sync, creating a certain labor market bubble: Meta laid off employees, and Google expanded its workforce by 30% in 2021, a significant portion of which was for software engineers, and the entry threshold is very low, leading to what I believe is a salary bubble.

Anthropic just focused on this market. The size of the market is one part, and the other part is who your customers are. Their smart point is that their customers are those willing to spend money and are wealthy. Even if the market is large, if customers are not spending much on operational expenses, you cannot quickly open up the market.

So this is a combination of favorable timing, location, and people, and it must be based on their clear early landing goals: clear goals lead to training models in that direction. As for why tech companies are willing to pay? It must be because they are solving problems they want to solve, addressing pain points, whether it is cost reduction or efficiency improvement.

The entire market will definitely continue to grow. It now has an ARR of $50 billion, compared to a $20 trillion market, which is only 2% to 2.5%. So although open-source models are coming in and there is competition in coding, this market is large enough to absorb open-source, OpenAI, and Anthropic, and I believe its revenue growth is still to be expected.

4. How Long Can 70-80% Gross Profit Last: Depends on How Much Tech Companies Spend

Mr. Z: But how long can its gross profit maintain at 70-80%? I'm quite curious about this.

Investment TALK Jun: I can't answer that question because it depends on how much these tech companies are willing to spend on OPEX (operating expenses) and on burning tokens, as well as their growth rates.

The entire market is still expanding: assuming a $2 trillion market can replace 500 billion, then from the current 50 billion, plus OpenAI, it adds up to 60 billion, 70 billion, and there is still about 7 times the space to reach 500 billion, which is the space these tech companies are willing to spend. It will only start to hurt their margins when tech companies are unwilling to spend money, and as the market share of other competitors increases.

So it depends on the speed at which tech companies spend money, as well as the speed at which other companies land, iterate products, and catch up. It's hard for me to judge how long this can last. But you see, he is basically monetizing as much as he can now, so he is willing to pay a high price to acquire computing power, earning as much as possible because he is currently the leader. The changes in this market are too great; we can only continue to observe.

5. Meta Selling Computing Power is Not a Bubble Signal: A Win-Win Situation Due to Insufficient Computing Power

Mr. Z: Is the overall market now focusing on both "spending money" and "earning money" at the same time? There is a keen interest in how much you are willing to spend on AI in CapEx, while Wall Street is also starting to seriously look at: how well can you monetize this money, and is it gradually reflected in revenue? Because the news about Meta selling computing power on July 2 shocked the market and severely impacted some NeoCloud companies. What is your take on this, TALK?

Investment TALK: The market started paying attention to this last year. I mentioned the example of Amazon earlier; in October and November last year, the market began to question whether the capital expenditures of big tech had returns, and this skepticism has continued to this day. It's just that after this period of adjustment, the market has become slightly more reasonable.

A few weeks ago, I mentioned that one unreasonable point in the market is the discussion about whether Micron can break free from being a cyclical stock, whether its price-to-earnings ratio can trade at 15 times or 20 times, which I think does not make sense. The market is simultaneously thinking about whether these revenue-generating hardware companies can break free from cycles while doubting the operational expenditures and ROI of big tech. These two things cannot coexist. If big tech's investments do not yield returns, they cannot maintain high-growth capital expenditures; if they do not maintain high-growth capital expenditures, how can all other hardware companies escape the cyclical theory? All premises must be based on the fact that big tech's investments have returns.

This skepticism started last year and continues to this day; these big tech companies need to slowly prove themselves. The earnings reports of big tech are about to come out, and I believe almost all of them will be very impressive for the market, at least these cloud providers will have no issues. Because Anthropic's ability to acquire Meta's computing power at a high price precisely proves that there is currently a shortage of computing power in the market. Earnings reports themselves are a rearview mirror, so I don't think there will be any issues with the investments and returns of these Hyperscalers in the second quarter.

As for whether Meta selling computing power harms NeoCloud, I think it is also part of the trading storyline. But NeoCloud is also divided: Nebius and CoreWeave, CoreWeave is actually quite weak, while Nebius is the market favorite, considered a popular stock, liked by retail investors and some small institutions. However, apart from the different financing methods, from RPO, from orders, and from landing, it does not justify the valuation difference between the two companies. After the declines of the past week or two, I think it has become relatively reasonable because Nebius has also seen a significant pullback.

Meta's selling of computing power is also related to its own fundamentals. The market was not very fond of Meta for a while, but my understanding of Zuckerberg is that he has a strong ability to correct mistakes; he is not afraid to admit errors and knows when he is wrong and when to pivot. The market's doubts about Meta in AI have been questioned for three consecutive quarters; to some extent, Meta's stock price has been stagnant for the past year and a half, with the lowest PE among all big tech companies. The market has actually been punishing it for the lack of returns on capital expenditures. They have indeed wasted a lot of money: from developing open-source models to closed-source models, from laying out a large amount of computing power to now starting to sell computing power.

However, my view on selling computing power is that it is a win-win situation. First, I don't know if they made an offer proactively; the news did not discuss this, but I estimate it should be Anthropic because currently, only Anthropic can acquire at such a high, top price due to their excellent landing. Meta's choice to sell computing power, including Google's choice to sell computing power and xAI's choice to sell computing power to Anthropic, is largely not because they want to monetize, but because Anthropic's offer is too good to refuse. And the reason Anthropic can make such a good offer is that their landing has been excellent.

To some extent, it is because AI's landing has been so good that these two companies are considering selling computing power. Building a data center incurs costs; previously, costs were lower, and now they are higher. But no matter how low the cost is, for example: if your cost is $10, now you are asked to sell it for $12 or $13, earning a gross margin of 20% or 30%, would you be willing to do it? Similar to the current model of NeoCloud, they are unwilling to do so because their original layout of computing power was for training models, and training models was for better monetization in the future, which would yield higher returns. But now, if a company gives you an offer: I am willing to acquire at $20 for your $10 cost, you can immediately monetize with a 100% return, would you do it?

For these two companies, the original layout of computing power was somewhat excessive, and internal training does not require that much computing power, so this situation naturally comes to fruition. Therefore, from the outcome perspective, this is a win-win process.

This does not mean that the market has excess computing power. If there were really excess computing power, Anthropic would not be able to acquire computing power or buy tokens at such a high premium. This is a decision made after aggregating all AI demand, not just a company selling computing power that indicates a bubble or excess in computing power. On the contrary, it is precisely because there is insufficient computing power that this situation arises.

6. Computing Power is Like Oil: NeoCloud Breaks Even in Three Years, Building Models You Never Know

Mr. Z: You previously compared computing power to oil. A company can, as you just mentioned, train models with computing power and then sell APIs, or directly lease computing power to generate cash flow immediately. What are the differences in return cycles and return rates between these two models?

Investment TALK: Simply building data centers and then selling computing power has a quick return cycle, basically a three-year return cycle, breaking even in three years, and starting to make money in the fourth year. But it also depends on how high the short-term rental prices are; if short-term rental prices continue to rise, the return cycle will be shorter.

This is NeoCloud's operating model: their order margins are generally set between 25% and 35%, around 30%, meaning that for a $100 billion order, there is basically a $30 billion profit. Moreover, returns come every year; this year they lay out and go live, and next year they start generating revenue and profit, entering the return cycle.

Big models, you never know. For example, Meta's closed-source model has been around for almost nine months now, from the first generation Muse Spark to the current second generation and 1.1 generation, the layout has been quite long. Moreover, transitioning from open-source to closed-source, taking Meta as an example, they have only just launched the API model now, and this layout has taken more than two years. These two years were just training, and they have only just landed now, without entering very meaningful, large-scale revenue, and have not scaled to the level of Anthropic, so the return cycle will definitely be longer.

However, if you can monetize through your own API, like Anthropic achieving margins of 70% to 80%, removing all operating expenses, the final operating margin may reach 50% to 60%, then this return is much higher than that of NeoCloud.

7. Why NeoCloud Appears in North America but Not in Asia: The Landing Market Determines Computing Power Demand

Mr. Z: I am quite curious; NeoCloud is basically a North American thing, and the companies often mentioned in the market are all North American names. So why is there no NeoCloud in Asia, like Korea or Taiwan? Taiwan actually has a NeoCloud called GMI Cloud, which is a partner of NVIDIA, and as far as I know, they have ordered quite a few GPUs from NVIDIA and even plan to build an AI factory in Taiwan. So why is there not much of a NeoCloud in Asia?

Investment TALK: I may not be able to provide a very in-depth answer to this question. In terms of layout, if I remember correctly, CoreWeave is only in North America, while Nebius is global, including Europe. From the perspective of business distribution, Nebius is more diversified, which has its pros and cons; I think being diversified is relatively better.

First of all, why did companies like NeoCloud emerge? A large part is because when they were first established, these large cloud providers, Hyperscalers, like Microsoft, often were not willing to take on this part of the business. I have talked to Microsoft many times, and for them, these infrastructure-based projects have returns similar to utility projects, which they are not very willing to do. The biggest difference between Hyperscalers and NeoCloud is that Hyperscalers earn more from value-added services on top of the infrastructure, making money from other services on that basis.

So when the market's demand for computing power is very high, and large cloud providers are hesitant to invest in CapEx initially, these NeoCloud companies naturally see the opportunity to step in because if you don't do it, someone else will. A 20% to 30% profit margin is acceptable for NeoCloud.

Another difference is that building data centers incurs significant upfront costs, so NeoCloud needs to finance; each company's financing methods are different. For big tech, while there are indeed many orders in the market, they absolutely cannot swallow it all themselves, as that would severely harm their finances and cash flow. Now everyone is discussing big tech: you have thrown away all your free cash flow; what will you do next? Will you rely on debt? Some have already started to take on debt. Therefore, they cannot use their current balance sheets to take on all the computing power and data center construction, which has led to the emergence of NeoCloud.

These NeoCloud companies each have their own methods of financing, but initially, someone needs to invest a sum of money, combined with their own financing, to slowly build up through returns; otherwise, they would have to take on very high leverage, which means starting out with leverage to do this. Once the scale is large, this has already exceeded the capabilities of big tech; they are unwilling to take it on because they have to be responsible for their investments, and investment returns are a very important indicator for many big tech companies when deciding whether to engage in this area.

Regarding why NeoCloud is not prevalent in Asia, I believe a significant part of it stems from the underlying demand and practical implementation. Last time, I mentioned that Anthropic is very smart because they have a strong market presence; North America has the largest software engineering market in the world. I remember a comment that left a deep impression on me, stating that there are many software markets in mainland China, but the misconception was that I was referring to the salaries paid. The amount of money these tech companies spend on software engineers each year is indeed very high. The market is substantial, and Anthropic's choice to establish itself in this market has gradually led to a growing demand for computing power, opening up future possibilities.

In fact, I am somewhat concerned about other implementations because they may not be as smooth as Coding is now: to take on Coding, the market needs to be large, and your buyers must also have the necessary balance sheets.

Returning to computing power: the foundational implementation is very good, and the market for it is also very strong, leading to a high demand for computing power; North America lacks sufficient computing power, and big tech cannot accommodate it, hence the emergence of the NeoCloud market. In contrast, in mainland China, many tech companies are subsidizing Tokens. Why subsidize Tokens? Because there is no software market for you to disrupt. The entire software market in mainland China is completely different from that in North America, and the spending is not as high; there is no very clear and large TAM (Total Addressable Market) there, so there is no significant demand for computing power, and there is even a need to subsidize computing power.

The differences in these markets, I believe, ultimately come down to how large the foundational implementation market is, which determines how extensive the upper-level computing power layout will be.

Section 8: Oracle's Misjudgment: The Storyline of Negative Correlation Between Hardware and Software, and the Gap with RPO and Valuation

Victor: I am particularly curious about a company like Oracle. Hyperscalers and NeoCloud currently have quite similar financing and spending models; they first lock in some contracts, possibly with a 25% to 30% return rate, and making a $100 billion order could earn $30 billion, which can be realized in two or three years. However, to do this now, you must first finance and leverage, somewhat like stockpiling and taking on orders. So why has Oracle been hit so hard? Before the earnings report in early June, everyone expected great results, and the report was indeed good, but the market did not anticipate that they would need to undergo another round of financing, which led to the drop. Moreover, Oracle is quite unique as it possesses attributes of Hyperscalers, NeoCloud, and software. In the recent context of a significant drop and rebound in IGV, it still fell by 49%, nearly halving. Is the market somewhat forcing them to start paying attention to CapEx ROI, similar to Meta: you can no longer finance recklessly and must also consider investor returns?

Investment TALK: I think Oracle is simply not favored by the market. When the market does not favor a company in the short term, it often relates to trends, logic, and storytelling, as short-term market influences are not closely tied to fundamentals.

Oracle's challenge is that this period coincides with a pullback in NeoCloud, with targets like Nebius also experiencing a downturn. I have been closely monitoring Oracle, and for several days, I have seen software stocks decline, including Oracle, because the narrative during this time is that software and hardware have a negative correlation: when hardware rises, software falls, and when hardware falls, software rises. Oracle, as you mentioned, has attributes similar to NeoCloud and also possesses software characteristics, so when software falls, it falls too; when software rises and NeoCloud falls, it still declines, leading to low sentiment.

However, looking back at your last question: is the market forcing them to slow down capital expenditures? I think it is possible; why not? If they are indeed pressured to slow down, they can certainly take a pause because their RPO (Remaining Performance Obligations) is substantial; I remember it being over 600 billion. I calculated that their RPO orders are more than ten times that of Nebius, yet their valuation is about six or seven times lower, while both have similar margins.

Moreover, Oracle's business in software cannot be disrupted by AI; the database segment cannot be disrupted by AI. As for the SaaS aspect, software is divided; each company's software is different. But the market does not care; as long as you are in software and within the IGV basket, it all gets sold off together. This indicates that there is a strong storyline behind the market, with momentum driving it; there is no way around it. This storyline will need to wait for a future point in time when one or several companies can bring it to light to break this loop; otherwise, there will always be someone trading on this logic, and those quant traders will continue to trade on this logic.

From a fundamental perspective, I believe Oracle has been significantly misjudged, at least relatively speaking. Just like the market favors hardware companies and does not favor Hyperscalers, I think it doesn't make sense: you cannot trade hardware as a cyclical recovery while believing that Hyperscalers have no ROI. Similarly, you may favor Nebius and not Oracle, but the disparity in their RPO and valuations is too great, which does not make sense. And precisely because of this lack of sense, opportunities arise.

Whether it is big tech or Oracle, if they slightly slow down capital expenditures, I believe the issues can be resolved immediately. Because their revenue is set for explosive growth next year, the investments made this year and last year in CapEx will yield rapid returns from NeoCloud, and we will see results by next year. Additionally, Oracle has adopted financing methods similar to CoreWeave and Nebius: prepayments, and even clients bringing chips into the deal, which are all ways to reduce early leverage.

Why are others willing to do this? Why are they willing to provide prepayments and bring chips into the deal? It is simply because the market's demand for computing power is too great; otherwise, no one would want to engage in this. Essentially, you are transferring the leverage responsibility to the clients, and the clients are willing to bear this leverage because they feel it is necessary to position themselves in computing power. Therefore, this leverage itself exists in NeoCloud and Oracle.

Thus, I believe the market's recent sell-off of Oracle is somewhat unreasonable. In the past few days, I have been contemplating whether I need to position myself in Oracle.

Section 9: Positioning Equals Perspective: No One in Mag 7 Holds Position, Hardware Should Reduce Holdings but Not Liquidate

Victor: In this wave of rebound for Mag 7, how do you view the growth and decline among these companies? I see you mainly focusing on both hardware and software, while also allocating Intel, Amazon, Apple, Microsoft, and other big tech; how do you view these companies? Additionally, there are companies like Tesla in Mag 7 that are not favored this year; what are your thoughts on the upcoming market?

Investment TALK: Each big tech company is different; they cannot all be viewed together because each has its own storyline.

Speaking of this, I recall the first question I was pondering: what are the things that have had correct logical judgments this year but have not risen? Looking back, it might just be big tech. This year, big tech has performed relatively poorly; YTD, Mag 7 has not even outperformed the S&P 500, and after this wave of rebound, it has only returned about 5%, while the S&P has performed better than Mag 7.

I reduced my holdings in Tesla at the beginning of the year because I believed this year would be an investment year for Tesla, and the market might not favor it. However, this logical line also aligns with big tech: this year is also an investment year for big tech, although it is not the first year; they started investing last year, but the scale of investment this year is larger than last year. To some extent, this thesis should also apply to big tech, so if we reflect, I think this aspect could be improved. This point is also contradictory in Amazon: I previously extended from Anthropic to Amazon, but Amazon is also the one writing the checks, and I did not anticipate how the market feels about those who write checks.

As for the second half of the year, I won't discuss hardware for now because the market sentiment is too high. Recently, chip valuations and the overall SOX valuation have reached relatively high levels, around 29 times, nearly 30 times, so a pullback is very normal. After this period of pullback, we may now be at a relatively neutral position. High sentiment and high valuations make a pullback quite normal.

The market cannot continuously trade the entire hardware and chip sector because at a certain point, when positions are full, there are no additional buyers. Many people do not understand: Micron's earnings report was excellent, and Samsung's fundamentals are also strong; why did prices fall? This is normal because pricing was too high previously. To put it bluntly, the market was too full of positions, and without additional buyers, a drop is inevitable; no matter how good the earnings report is, it will still fall. However, through continuous turnover, fundamentally sound companies will eventually find support because once valuations return to a reasonable range, new buyers will come in.

From this perspective, we need to ask: which stocks are not positioned in the market? My view of the market has always been that fundamentals are the top layer of all judgment logic, but whether they can emerge often depends on whether the market has positions in them and whether it is crowded, which will then lead to a catalyst. I have always referred to crowding as a setup: fundamentals plus positioning create a setup, a condition; but having a condition does not mean prices will necessarily rise; you still need a catalyst.

Currently, there is little positioning, and the market even dislikes stocks like Mag 7 and software. The market currently does not favor software, even believing that PE is not worth as high as before, and there has been little rebound; Microsoft has led the way but performed quite mediocre, so it is very certain that there is no positioning in software. Mag 7 has performed very poorly, and it can also be inferred that there will not be many positions because everyone is chasing Nvidia, both institutions and retail investors.

Returning to fundamentals: software needs to be selective; you must determine whether a company's software will be disrupted by AI. If there are no issues with fundamentals, sentiment, or valuations, the next step is to wait for a catalyst, for the market's logic to return. However, the market will not return without reason; it will definitely need a catalyst.

Thus, my approach to balancing hardware and software is as follows: I do not believe hardware should be completely liquidated because from an investment perspective, unless valuations are extremely full and you believe fundamentals will significantly worsen, there is no need to liquidate; fundamentals still exist, and as long as valuations are restored to reasonable levels, funds will return. Hardware should still be retained, but it was indeed time to reduce holdings recently.

Section 10: Big Tech Holds the Initiative in Writing Checks: The Prisoner's Dilemma of Capital Expenditures and FOMO

Investment TALK: Big tech, as the side of capital expenditures, actually holds the initiative. However, because all big tech companies are experiencing FOMO, it leads to mutual harm: when they all FOMO, storage prices rise, and everyone pays the price for storage.

Big Tech has taken control of the spending power. If they slightly reduce their tone on capital expenditures, it would actually be beneficial for them, and significantly so. Of course, this requires a consensus, which could turn into a prisoner's dilemma: you need others to cooperate. If you slow down your capital expenditure growth and others do not, you will worry about falling behind in the future. As long as everyone reaches a certain consensus and avoids FOMO, because FOMO leads to no good results. The FOMO of Big Tech in capital expenditures is no different from retail investors' FOMO in hardware; FOMO drives prices up in the short term, but ultimately, it will not yield good results.

If Big Tech avoids FOMO, it would be a positive for them: the growth rate of costs for each plan will not be so fast; at the same time, the market will once again believe that after you gain control over spending, the return cycle will still yield returns, and each Big Tech's EPS growth will be very good. Naturally, funds will return, sentiment will improve, and stock prices will rise again. Moreover, inflation in hardware may also ease. Therefore, I believe Big Tech holds the initiative.

Additionally, many Big Tech valuations are not expensive now, so I think they should be included in the portfolio.

11. Apple, Tesla, and the Software Sector: Each Big Tech Must Be Viewed Separately

Investment TALK: The software sector is subjective and varies from person to person. Many believe that some software companies have indeed been damaged, and some have been. Therefore, software needs to be selected carefully. If we talk about sectors, the only software company currently performing well is Cyber Security, which is also the market consensus; however, this consensus at one point reached a very high valuation and will also experience a correction like hardware, which has nothing to do with fundamentals.

So my layout for the second half of the year is: continue to hold hardware, just with lower positions than before; Big Tech's fundamentals are sound, they hold the check-writing side, valuations are not expensive, and YTD performance has been worse than many sectors. Many are worried about a market correction, but at this position, I feel very comfortable holding Big Tech, with no concerns at all.

More importantly, the reason Big Tech has its current status, I think Meta is a great example: the management of these Big Tech companies has the ability to correct mistakes, and this ability often appears after shareholders' interests are harmed in the short term. Just like Meta in 2022, and during this time, Zuckerberg's pivot and transformation, I believe it is very friendly to shareholders. So when people ask me if I worry about a market correction, I am not worried at all because if I hold Big Tech, I am completely unconcerned about the overall market correction.

Of course, Big Tech is not the same. For example, Apple has a relatively high valuation, but I am very optimistic about its AI implementation because I believe any implementation on the consumer side will likely concentrate on Apple, Google, or social media like Meta. So each company still needs to be viewed separately.

As for Tesla, the judgment made at the beginning of the year was correct, and I think it is still okay now. However, I believe this year is a time to accumulate positions in Tesla; I might wait for better opportunities to gradually increase my position. Because autonomous driving is a matter of time, and at this point, I tend to think we need to wait for them to launch the next generation of chips to truly solve it completely. This also requires hardware cooperation: what can be done algorithmically, with Tesla's current hardware conditions, may have already reached its limit. So after the next generation of hardware updates, there may be an "aha moment" for autonomous driving, which will suddenly explode, but it requires time.

12. Hardware Focused on Intel: CPU Recovery, Edge AI Implementation, and the Second Arm of Wafer Foundry

Victor: I would also like to ask, why is your hardware exposure mainly focused on Intel? Additionally, your current investment portfolio also includes targets from other sectors, such as ISRG, which is a surgical company, and GE, which is in aerospace transportation. In a stage where AI has reached a certain level and emphasizes diversified allocation, why choose these companies and sectors?

Investment TALK: First, let’s talk about Intel. This year, the vast majority of my investment portfolio returns came from Intel because my entry and accumulation points were quite good. When I bought Intel, I saw it as a good turnaround in adversity, plus the benefits from this wave of AI.

Now, looking forward: first, in the CPU segment, the market has finally realized that this is not just a simple GPU market; CPUs also have a future, especially with the future implementation of AI Agents. In fact, the discussion around AI Agents has decreased significantly over the past six months; it cooled down a lot after "Little Lobster" (OpenClaw). Why did it cool down? I believe it is because the API prices are too high. With high API prices, you cannot avoid discussing AI Agents because AI Agents burn tokens 24/7: humans can only focus on one thing, while AI Agents can focus on ten things directly, leading to an exponential increase in token consumption compared to humans. Therefore, the price of tokens must come down.

Yesterday, there was news about the implementation of edge AI, which I believe is a significant positive for AI, including hardware from companies like Apple and Google; if edge AI is implemented, it will also greatly benefit the consumer side and the implementation of AI there.

Back to Intel. It has two businesses: product and foundry. The product segment is not worth discussing; last quarter, the poor CPUs turned into treasures because of the high demand for CPUs. The previously mentioned edge implementation is also a direction that Chen Liwu has focused on since he took over: he believes that GPUs can no longer compete or catch up with NVIDIA or AMD, so he is very clear about targeting the implementation of edge AI.

The implementation of edge AI was already discussed by AMD's Su two years ago regarding AI PCs. But at that time, discussing AI PCs seemed like a very abstract concept: what is an AI PC? If I have a PC, what can AI do? But just like the news yesterday, if we can compress open-source models to run on the edge without needing to adjust any APIs, it suddenly materializes the implementation of edge AI, and you can see the entire path of implementation. At this point, when discussing AI PCs, you will find that it makes sense.

Next, I expect many consumer hardware manufacturers to experience a refreshment: your laptops and desktops must keep up with hardware to run AI; if the hardware does not keep up, the capabilities of edge AI will not be strong. As hardware specifications rise, the models you can run will become more powerful, and the things you can do will increase. So you will find that the AI PC discussed two years ago now makes sense. This is also the area that Chen Liwu believes Intel can pursue in the AI market, which is now slowly becoming visible on the product side.

As for foundry: aside from TSMC building a factory in Arizona, TSMC's overall geopolitical risk has indeed been significantly mitigated due to its U.S. factory; however, Intel is still a local foundry and the only one. The U.S. government itself also invests in Intel and supports it, and other companies will inevitably place orders with Intel to diversify risks between it and TSMC.

Moreover, it is not just about diversifying risks. When Intel's yield improves, other companies can use Intel to negotiate prices with TSMC because you have another option. So you see, when Intel's yield starts to improve, more companies are coming out to say they are considering future collaborations, including Apple and the cooperation between Intel and the old Ma's Terafab, etc. Its foundry business, as long as it can improve yields, will have a market, and these customers are willing to allocate some business to Intel.

Including packaging, Intel's packaging is not much worse than TSMC's, so it is also starting to be discussed, and advanced packaging can also become an independent business. Therefore, Intel has many arms. Recently, the market has been studying which links in the entire AI supply chain are short, but for me, the most direct approach is: there is no need to spend too much time, just understand all of Intel's businesses clearly, how big the market for each arm is, whether this market is expanding, and whether the fundamentals are improving. Confirming these is enough. This also explains why I am willing to hold such a large position in Intel while only allocating to Intel in hardware. Another consideration is valuation; from a valuation perspective, whether it is reasonable.

13. GE and the Medical Sector: Each Target Plays a Different Role in the Portfolio

Investment TALK: I will speak more quickly about the other targets mentioned earlier.

GE is a very common layout; it is more focused on military: half military and half aerospace, which is aviation transportation. Military is not a monopoly, but its market is its market; it has sufficient competitiveness and basically has absolute pricing power. Moreover, this is a secular market, essentially a service: although it sells hardware, it actually makes money through services and recurring services, so it is a very good business.

This part is independent and not closely related to AI. I believe that after the recent war between the U.S. and Iran, Europe will rethink defense spending and layout: whether to rely entirely on the U.S. or support local enterprises. From these perspectives, military is a very long cycle that does not require too many adjustments; only when valuations are high do some adjustments, otherwise, this sector itself is a relatively long cycle.

The medical sector is because my position in the medical sector is not large, and it is almost unrelated to AI. Right now, all the market's focus and heat are on AI. I am not saying to leave AI, nor am I saying to leave hardware; whether it is Big Tech, hardware, or Intel, they are all benefiting from AI and are meant to enjoy AI, so it is impossible not to be optimistic about AI. However, when sentiment is so high, diversifying into relatively conservative targets and into the medical sector can help spread the overall portfolio risk.

Since we are talking about the investment portfolio, each target plays a different role in the portfolio, and we cannot pursue extreme returns for every target. When discussing the investment portfolio, there must be high-risk targets and low-risk targets, and high-risk targets need to match what, as well as how strong the overall portfolio's ability to withstand risks is if certain risks arise in the future. This is what I consider more, which is why I have increased my position in the medical sector and may continue to add more.

14. If you're in AI, pivot to Crypto? Only Stablecoins Have Really Broken Out

Mr. Z: Do you all remember one of the hosts of the All-In Podcast, who said back in August or September 2023, "if you're in Crypto, then pivot to AI"? Now, it's Jeff Park from the crypto world saying, "if you're in AI, then pivot to Crypto." Is it true that Crypto has hit rock bottom and can't get any worse? Is it a good time to pivot to Crypto?

Investment TALK: I think the market needs to move away from this binary thinking. I don't believe you necessarily have to leave AI to invest in Crypto right now. However, if you currently have no exposure to Crypto, it might be a good idea to allocate some resources there.

After going through several rounds, the market has slowly realized that most Crypto projects are not worth investing in. Ultimately, only the top two or three projects may survive, and I am being very generous here.

Stablecoins are the only thing I see as viable at this point. I consider myself a two-time victim in the crypto space; when I first entered, I was fed the concept that "Ethereum will disrupt traditional finance." Now I see nothing—absolutely nothing. It was only when Circle went public that I truly saw a Crypto project that could potentially disrupt traditional finance. This is why I hold Circle: I genuinely see that Crypto can break out of its niche, and stablecoins can achieve that.

Another point is that the U.S. government is willing to support stablecoins because it needs the expansion of the dollar: the government's interests can be expanded globally through stablecoins, increasing the adoption of the dollar and its reach. This is essentially incentive alignment, where all interests are aligned. This is the first time I feel that it could potentially break out. But whether it can break out depends on how well the product is developed and whether companies and institutions choose to adopt it; we can only wait and see.

Mr. Z: You just mentioned the U.S. government. I understand they want more buyers in U.S. Treasuries, so they are encouraging the development of stablecoin legislation like CLARITY. However, from my observation, since Trump launched TRUMP and MELANIA (meme coins) last January, a lot of external liquidity has been drained from the crypto space, and there are no other marginal buyers. Many recent bills have been proposed but haven't materialized; it's quite questionable whether the U.S. government is really establishing a so-called Bitcoin reserve.

Investment TALK: Trump is like a rat in the soup. I don't know if you've been following the recent progress of CLARITY; I have been. CLARITY is currently stuck on something they call ethics: the issue of U.S. presidents and congressmen investing themselves. The entire left is concerned about conflicts of interest.

This conflict of interest is specifically aimed at Trump because his family office has invested heavily in Crypto while he himself promotes it. Their goal is not really related to CLARITY; they just don't want Trump to monetize his position right now, targeting him specifically, while the left is blocking it. However, CLARITY is meant to serve the entire stablecoin sector and define a framework for it. So, while they are pushing a policy that is beneficial for the industry, some legislators have personal grievances that are holding Trump back.

I saw the latest news that Trump is supposed to have some discussions with senators tomorrow. I don't know if he is willing to make compromises; both the left and the right are reluctant to budge on this matter. But ultimately, it is a matter of exchanging interests: if there are truly good developments for the industry that align with the overall interests of the U.S., at some point, I believe it can still move forward, although there have been some setbacks in the legislative process recently.

Outlook for the Second Half of the Year: Not Harder Than the First Half, Positioning Equals Perspective

Mr. Z: Finally, do you have any words for the audience? In this currently perilous and uncertain market, how should everyone operate?

Investment TALK: I actually don't think the market is very perilous. When we first communicated, I saw a question about how to view the next 6 to 12 months. I am not overly pessimistic about the second half of the year. If you look at the index's growth, especially QQQ, it has only risen about 7% to 8% from its peak at the end of October last year, maybe 10% at most; in eight or nine months, it has only increased by 10%. Is market sentiment really that high? Not at all. But we have seen so many AI implementations this year; the fundamentals of AI have improved significantly compared to last year and are much clearer.

The market has been heavily influenced by the hardware sector, chips, and optical communications, as the focus has been on those areas. The pullback and the rise have been centered there, leading to the perception that the market is entering a bear phase. But if you take a step back, the recent performance of QQQ has actually been a high-level consolidation; nothing has happened.

So, from a difficulty perspective, the second half of the year may not be harder than the first half, as the first half had a war event with very high uncertainty, and many people were extremely panicked at that time. The second half will still have risks; there will always be storylines and risks in the stock market: they may come from macroeconomic factors, the Federal Reserve, or the listings of two tech companies or two model companies, including SpaceX's listing: its market impact has not yet fully materialized, as the lock-up period has not completely ended; we have just completed the first phase.

But I believe that having a balanced position, not overly concentrated in hardware, while allocating some resources to big tech, is the way to go. The fundamentals of hardware are still being implemented, and in this situation, having a balanced position is sufficient. I don't think there is a significant risk that warrants a drastic reduction in positions. If you have already made money in the first half, just balance your positions. I still have a positive outlook on AI implementations.

Mr. Z: Or let me ask it this way: Where is the Alpha?

Investment TALK: I always say that positioning equals perspective. My allocation is in big tech, and I have a significant position there. The balancing method is to add some in the healthcare sector. AI cannot move in a straight line; if AI pulls back, the healthcare sector will not be affected at all. The GE we talked about earlier will also not be affected. So from the perspective of sector and position allocation, diversify some of the AI exposure, but you can still maintain a relatively large position in AI-related assets because the overall EPS growth is still coming from AI hardware. So I believe that positioning equals perspective.

Mr. Z & Victor: Thank you very much to Investment TALK for joining us at 168X for nearly an hour of wonderful sharing today, and thanks to every listener who has tuned in. If you enjoyed this episode, please follow 168X on X and YouTube, and share the program with more friends interested in AI, U.S. stocks, and technology. We'll see you in the next episode.

Disclaimer: This content is provided for general branding and informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online events, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets or to use any services. Crypto assets are highly volatile and may result in loss. WEEX services and online events may not be available in all regions and are subject to applicable laws, regulations, and eligibility requirements. You are responsible for ensuring that your use of WEEX services complies with local laws and for carefully assessing the risks before participating in any crypto-related activities.