Text by Victor ( @vcmktasa ) and Mr. Z ( @168MrZ )

Guest: Frank ( @qinbafrank )

In early June 2026, the Taipei COMPUTEX conference was in full swing, with Jensen Huang proclaiming that Marvell would be the next company to break the trillion-dollar market capitalization mark, causing a collective surge in the optical interconnect sector. At the same time, the macro environment was far from calm: the Strait of Hormuz had been closed for over 100 days, oil prices remained stuck around $90, and the market held its breath awaiting the May CPI data to be released on June 13th; meanwhile, SpaceX was about to IPO with a valuation of approximately $1.75 trillion, and Anthropic had also secretly filed documents, indicating a reshuffling of the liquidity landscape for the second half of the year.

At this moment, amidst the hardware boom and macroeconomic uncertainties, 168X invited Frank ( @qinbafrank ), one of the few investors who can simultaneously navigate macroeconomics, US tech stocks, AI supply chains, and cryptocurrencies, and who has long tracked global capital flows. Frank's "top-down, cross-market" framework, honed through his experience in mobile internet, entrepreneurship, and VC, provided a remarkably comprehensive assessment in this over two-hour conversation: he doesn't believe there's a bubble in AI, arguing that the entire commercialization process has only just passed the monetization inflection point and is halfway through; what truly warrants vigilance is the tail risk of a "reset" of macroeconomic logic. He began by discussing Nvidia's three major debates, dissecting the penetration rate dividend, the capital expenditure war, optical interconnects and CPOs, Nokia and edge computing, and the three major funding logics of scarcity/upgrade/long-term, before moving on to the liquidity shock of SpaceX's IPO and the future of Bitcoin.

I. Research Framework: A Cross-Market Investment Path from Mobile Internet and Startups to Venture Capital

Mr. Z: Frank is one of the few top investors who can simultaneously demonstrate profound knowledge of macroeconomics, US stock technology, AI supply chains, cryptocurrencies, and global capital flows on Twitter. Frank, could you please introduce yourself and explain how this market research framework came about?

Frank: Z just mentioned top investors, but I really don't qualify. I'm far from it; I'm just an ordinary investor. There's an old Chinese saying, "A full bucket makes no sound, but a half-full bucket rattles," which describes me perfectly. I know a little about everything, but I'm probably not an expert in anything in depth.

I primarily invest in US stocks and cryptocurrencies, focusing on the secondary market. I also share my thoughts on macroeconomics, industry trends, and individual stock analysis on X, all based on my own reflections. My career can be divided into two phases: Over a decade ago, I worked in China's mobile internet industry for seven or eight years, as a product manager. I also started my own business and raised several rounds of funding, but it didn't succeed. Later, I transitioned to venture capital. I entered the crypto space in late 2017 and early 2018, continuing my approach from primary market VC, investing in projects initially called ICOs, later private placements. Around 2021 or 2022, due to increasingly stringent regulations in China, I stopped primary market operations and shifted to secondary market investments, essentially becoming an individual investor in recent years.

This framework is actually consistent with my previous work. My first job after graduation was in the internet industry, coinciding with the global transition from PC to mobile internet. I worked in major domestic companies, started my own business, and also worked in venture capital. My experience in venture capital essentially laid a foundation for me: you need to approach things from an industry perspective, analyzing the state of an industry's development, its trends, the existing benefits, the opportunities within those benefits, the companies operating within those opportunities, and how to assess an early-stage company's entry point, the pain points it solves, and even whether the founders are truly effective. This naturally allowed me to understand the strengths and weaknesses of various business types and business models in the internet and technology industries.

I opened my first US stock account around 2013, and I've been trading US stocks longer than crypto. But to be honest, in the early days I purely benefited from the industry's growth; I didn't really understand the market. My only thought at the time was: this industry has a clear trend, it's growing rapidly, so just jump in blindly. What really made me put in the effort to understand the market were the several major crashes I experienced after entering the crypto market: the outbreak of the US-China trade war in the second half of 2018, the Fed's last round of interest rate hikes, and the US stock market crashed in the fourth quarter, with Bitcoin falling from $6,000 to $3,000; then came the black swan event of the COVID-19 pandemic in March 2020, with the US stock market experiencing four circuit breakers in three weeks, a drop unprecedented in its speed, followed by the Fed's emergency interest rate cuts and announcement of unlimited quantitative easing, after which the US stock market went on a relentless upward climb.

That wave truly prompted me to think: why did this happen? So around 2020, I read almost every book I could find on the market about the history of the Federal Reserve, monetary economics, central banks, Modern Monetary Theory, and asset allocation. Books like Bernanke's *The Courage to Act* and his later co-authored *Firefighting* with Paulson and Geithner, which discussed the actions the Fed took during the 2007-2009 financial crisis, how it negotiated with Congress, and how it coordinated with the Treasury; Paulson's *On the Edge*, Geithner's *Stress Tests*, and even earlier, Volcker's *Changes in Fortune*. Essentially, I wanted to understand its logic and core principles, and over those one or two years, I gradually formed my framework for understanding macroeconomics.

So, whether I'm looking at RWA (Real World Assets) or AI, it all boils down to the same approach: treat it as a technology industry, analyze where its business is headed, what paths are blocked, what paths are possible, and how its business model should change. My past work, career, and investment experiences have gradually shaped my style of understanding industries and macroeconomics in a more comprehensive, top-down manner.

II. Three Major AI Debates: How are the Three Major Doubts – "Trend, Computing Power, and Profitability" – Verified Using Nvidia's Stock Price?

Mr. Z: The current AI craze in the US stock market, as well as emerging markets like Taiwan and South Korea, is somewhat reminiscent of the Fed's quantitative easing and the liquidity boom in cryptocurrencies in 2020. However, the reality is that the Fed hasn't cut interest rates, and may even raise them this year due to the war. What's your view on the current market, and how far has the money in the AI ​​sector progressed?

Frank: My own view is that while things seem similar to 2020, they are actually different. Looking back now, 2020 and 2021 marked the end of a golden age. The main driving forces at that time were zero interest rates and unlimited quantitative easing, with the Fed purchasing two to three hundred billion dollars of Treasury bonds almost every month. Before the pandemic, the Fed's balance sheet was about $3 trillion, and by the time it began shrinking in 2022, it had reached $8.9 trillion, meaning that from the second quarter of 2020 to the second quarter of 2022, it expanded by about four to five trillion dollars, which is a huge scale. In addition, the mobile internet had been developing for ten years and was at its most mature, leading to explosive growth in the performance of giants. So, both performance and valuations increased, and with the massive influx of liquidity, all assets soared, not to mention cryptocurrencies.

But today is different; we live in an era of "limited easing." Back in July or August 2024, I wrote a tweet predicting that the Federal Reserve would likely implement limited easing for the next two to three years: interest rates would likely be between 3% and 3.5%, and they might not engage in QE (quantitative easing), but would consider some small-scale balance sheet expansion. That's exactly what's happening now: interest rates are between 3.5% and 3.75%, there's no QE, and after the balance sheet reduction neared its end at the end of last year, a liquidity crisis occurred, prompting them to start purchasing short-term debt for small-scale balance sheet expansion. Moreover, the scale has been declining month by month for the past two months, from about $40 billion per month in March and April to only about $10 billion per month this month. In this environment, assets without fundamentals are struggling.

What makes AI different is that it's a truly significant trend that everyone recognizes. Last year, I wrote a tweet titled "Three Major Debates About AI," and you can see this clearly in Nvidia's stock price chart, which shows three distinct phases.

The first part, from the end of 2022 to the end of 2023, focused on the debate in the market about whether AI was truly a trend. ChatGPT 3.5 was released at the end of 2022, causing market excitement and a surge in prices. However, at that time, the global semiconductor industry had not yet recovered, corporate procurement and capital expenditures by technology companies had not picked up, and earnings had not yet materialized; it was purely speculation based on expectations. Therefore, Nvidia's performance essentially stagnated in the second half of 2023, fluctuating at high levels for about six months.

The second phase, from early 2024 to the first quarter of 2025, was a debate about whether AI needed so much computing power. I believe the turning point was the Davos Forum in January 2024, where Sam Altman spoke and subsequently promoted GPT-4 and Sora (video models), which ignited a frenzy. More importantly, Nvidia's performance began to improve significantly from the end of 2023, with enterprises purchasing GPUs at an increasingly rapid pace. Then, in January 2025, DeepSeek released a large model, reportedly with extremely low training costs, shaking the US tech industry. Marc Andreessen, the founder of a16z, tweeted his praise, causing market panic. Many felt that such a large computing power wouldn't be needed, and coupled with the tariff war at the beginning of the year, Nvidia's stock price plummeted.

Incidentally, I think Sam Altman is incredibly visionary. In early 2024 at Davos, he proposed using $7 trillion to build data centers and increase computing power. At the time, not only did outsiders think it was a pipe dream, but even Jensen Huang and TSMC CEO C.C. Wei thought it impossible. But by 2025, everyone gradually saw the logic, especially tech companies, because with more and more model parameters and richer data, more computing power was needed.

The third section, from Q4 2025 to Q2 2026, delves into the debate of "whether such massive capital expenditures will actually generate profits." I wrote a long article in February titled "What Does This Capital Expenditure War Mean?" At that time, I had already noticed some signs: in early January, Google, Anthropic, and Alibaba all launched their own agents, and Meta attempted to acquire Manus (although this was later rejected by the Chinese government). Throughout February, various agent applications became wildly popular, and a large number of people were burning through tokens. The trend was visible, but it still needed data verification.

In that long article, I predicted that the Q1 earnings season in April would be crucial, and I tended to believe that major cloud vendors (Microsoft, Amazon, and Google) would deliver exceptionally strong earnings, pushing the market from "skepticism" to "validation." This is precisely why the entire market has risen so rapidly since April: everyone is finally seeing the data. One reason is the massive surge in these three cloud businesses, as their revenue comes from B2B enterprises and B2C users consuming tokens; another is the rapid growth of Anthropic's annualized revenue (ARR). I remember it was around $30 billion in March, $40 billion in April, and $45 billion in May. That's a month-over-month growth of around 10% to 20%.

I even wrote a tweet back then, saying that this was an inflection point for the monetization of AI: before, everyone was worried that such a large investment would not be profitable, but now the evidence proves that it can be profitable and the growth rate is accelerating, so doubts have turned into verification, and everyone is naturally more confident to rush in.

III. Commercialization Inflection Point and Capital Expenditure War: Why Penetration Dividend Determines This Bull Market

Frank: So where exactly is AI developing? I have a framework. The first thing to look at is penetration rate. The tech industry essentially benefits from penetration rate; 10% is a turning point. Passing 10% has two meanings: First, it means the technology is truly useful, otherwise 10% of people wouldn't be using it; second, once it exceeds 10%, the spread becomes extremely rapid. Back when China's mobile internet started in 2010, smartphone penetration was less than 10%, but by 2017 and 2018 it had reached 60% or 70%.

In the AI ​​sector, Goldman Sachs' Q3 report last year indicated that the penetration rate of AI procurement by US enterprises was approximately 9.7%, close to 10%; a report released in March or April this year showed it to be around 18%. Normally, the penetration rate from 10% to 40-50% represents a period of rapid growth. The penetration rate represents the starting point for commercialization: the larger the user base, the more paying users there will be at the same payment rate. Moreover, as technology iterates, the payment rate also increases, with users progressing from the free version to the Pro version, and then to the Max version, resulting in higher and higher ARPU (average revenue per user).

Secondly, consider commercialization. The overall trend is still upward, as Huang Renxun said, demand is growing in a "parabolic" manner. In this situation, large tech companies are more confident in their capital expenditures; they won't contract or reverse course, but rather accelerate them. Look at Google's announcement a few days ago of an $80 billion funding round: $10 billion from a private placement by Berkshire Hathaway, and $70 billion from a secondary market offering, of which $30 billion was issued through underwriters and $40 billion was issued directly on the secondary market at ATM (market price). Essentially, on the one hand, its free cash flow has indeed decreased significantly, from two or three hundred billion dollars last year to about $10 billion now; but its operating cash flow is stable at four or five hundred billion dollars, and it has nearly $126 billion in cash on hand. It's simply taking advantage of the market's enthusiasm to raise funds through cost-free equity financing, which is a commercially savvy decision. On the other hand, it also means that they are still increasing their investment.

However, one point remains: such a parabolic rise in any asset is unsustainable and will always be followed by corrections. The three rounds of questioning AI has faced in the past two years (whether it's a trend, whether it requires computing power, and whether it can generate profits), coupled with some macroeconomic influences, will inevitably lead to minor or medium-level corrections, resulting in an overall wave-like pattern. My own view is that we are currently in the middle of the larger phase, just beginning to see an inflection point in monetization and accelerating, but there are still some macroeconomic risks ahead.

Mr. Z: I personally look at the US Liquidity Index, which is basically the Federal Reserve's balance sheet minus the Treasury's TGA (General Account) and ON RRP (Overnight Reverse Repo). The last bull market peaked at over 7 trillion in 2021, and it has since shrunk to around 6.5 trillion. With no increase in liquidity, why haven't I seen a cooling down in the AI, CPU, or storage sectors, and why is the US stock market still so buoyant? How long can this phenomenon last? And what indicators should we use to analyze it?

Frank: My data is based on a slightly different metric than yours. I'm looking at the Federal Reserve's balance sheet itself, which peaked at around 8.9 trillion when it began shrinking in March 2022, bottomed out at 6.5 trillion, and is currently around 6.7 trillion. Adding TGA and reverse repos to calculate liquidity is fine. The Fed's liquidity is roughly equal to the balance sheet minus the TGA account, then minus reverse repos, which includes cash in circulation plus bank reserves.

Let's get back to the core. First, we need to define it: the current US stock market isn't a full-blown bull market; it's a "structured" market. Unlike 2020 and 2021, when even altcoins and every sector rose, this wave in the cryptocurrency market has only seen BTC, SOL, and BNB reach new highs, while most altcoins haven't, or have even declined. The same applies to the US stock market. In the past two years, the main drivers have been AI and related semiconductor supply chains, along with strong performances in defense, military, and resource-related sectors connected to geopolitical conflicts. Financials are doing okay, consumer staples are decent, but consumer discretionary is rather weak.

For a stock to rise, it either needs to have positive expectations or strong performance. Positive expectations mean that it's riding the wave of AI, and the market believes it will benefit from AI in the future; strong performance means that it's actually making money.

IV. The Era of Limited Easing: Liquidity, Interest Rates, and the Stock Selection Logic of a "Structured Bull Market"

Frank: Let's talk about capital behavior first. In this limitedly loose environment, once the market has risen too much across the board, a correction is inevitable. There are many types of capital in the market: there are long-term funds, as well as funds that trade short-term or focus on swing trading and trend following. After making a profit of 100-200%, some funds will take profits and exit, and some may even short the market because they think it's too expensive. Therefore, after a short-term surge, prices tend to have priced in too many expectations, and coupled with a less-than-ideal macroeconomic environment, a correction is likely to occur.

Last year, I roughly summarized the patterns of correction levels in the US stock market over the past 20 years, using the Nasdaq as a benchmark; I'll elaborate on that later. The key point is: to determine whether a correction is large, medium, or small, the crucial factor is whether the growth rate of AI commercialization has slowed down. As long as the annualized revenue of large model manufacturers is still growing and cloud business continues to exceed expectations, the overall business logic hasn't been reversed. In that case, even if there's a short-term surge, expectations are priced in, and funds withdraw due to perceived high prices, resulting in a small to medium-sized market decline and individual stock drops of 20-30%, the market will rebound once a new catalyst emerges (like the earnings season in April this year, coupled with rapid annualized revenue growth).

Let me elaborate a bit more, because many people are worried about comparing this wave to the dot-com bubble of 2000. I think there are similarities and differences. The similarity is that prices have indeed risen significantly, with a parabolic upward trend in the last two months. This week, however, hasn't been particularly positive: when Jensen Huang announced plans for AI PCs, funds poured into the PC and CPU supply chains; when he announced the AI ​​Factory (AI computing power factory), funds again flooded into liquid cooling, high-voltage electricity, power distribution, and power-related sectors; yesterday he said Marvell would be a trillion-dollar company, and Marvell's stock price rose 30% that day and is still rising today, its market capitalization jumping from $200 billion to $60 billion – it's a bit frenzied.

The difference lies in penetration rate and the maturity of business models. In 1999, the internet penetration rate in the US was only slightly over 30%, and globally only slightly over 10%. The number of tech-savvy people was very small, and it wasn't until 2008 or 2009 that it reached 75% in the US. Moreover, at that time, the entire internet lacked a clear business model. Companies truly engaged in internet business (Amazon, Google) weren't making money; instead, companies like Cisco that sold products were profitable. It wasn't until 2002 to 2006 that they found the four major business models: advertising, e-commerce, value-added services, and games. Mobile internet is different. Starting in 2010, it covered in less than ten years what the internet took nearly two or three decades to accomplish.

Today, AI faces the infrastructure of four to five billion smartphones globally, with most people using platforms like Twitter, WeChat, Douyin, TikTok, Instagram, and WhatsApp. Information spreads extremely quickly, and the penetration rate of a revolutionary new technology can potentially match that of the mobile internet in just three to five years, with relatively mature business models. This is the biggest difference. It determines that while parabolic growth is also unsustainable, the magnitude and pace of this correction will be much smaller than in 2000 due to the improved infrastructure, higher penetration rate, and faster commercialization. In 2000, there were no viable business models and low penetration rates, so the bubble burst and the market collapsed, taking a long time to recover. At that time, the macroeconomy also faced 9/11 and the Great Recession. Market corrections are essentially measured in two dimensions: time and space. Fundamentals determine the magnitude of the subsequent correction.

Victor: Just yesterday, I saw that Herman Jin ( @ShanghaoJin ) , whom we interviewed before, also posted an article warning of risks . One key point was: if the revenue growth of large model manufacturers is not as expected, will it affect the entire narrative of computing power demand and cause market panic and decline? But if that happens, could it be like the DeepSeek moment, a good entry point, with a second phase to be taken after the AI ​​semiconductor correction?

Frank: The point made by the teacher you mentioned is actually consistent with what I've been saying. The surge in April and May had two core drivers: first, cloud vendors validated that large capital expenditures led to unexpected growth in their cloud businesses, and cloud monetization comes from token consumption by both B2B and B2C clients; second, Anthropic's annualized revenue grew rapidly. These two points are the foundation supporting the entire business logic: AI commercialization has reached an inflection point and is growing rapidly.

If major model vendors underperform expectations in the future, it means the very foundation of the market narrative is flawed. This is because Microsoft, Google, and Amazon source much of their computing power from these major vendors, who are further upstream and closer to the very beginning of commercialization. In this scenario, there will be at least a medium-level adjustment; all logic will need to be reset—not a complete overhaul, but a moderate reset. Then everyone will have to wait: new evidence is needed to prove a return to rapid growth, with scale and growth exceeding expectations, before confidence can be restored. Therefore, whether it's a good time to enter the market depends on whether the commercialization logic has been truly reversed.

V. Divergence among software stocks: Which will be replaced by AI, and which will be strengthened?

Victor: This year has been particularly challenging for stock picking skills, as the logic behind each stock is different. Software stocks, for example, experienced a significant sell-off at the beginning of the year. Recently, Snowflake reported strong earnings, and cloud computing companies like Oracle, which borrowed heavily last year to secure a bunch of RPO contracts, are nearing their harvest period. ServiceNow has also seen substantial gains. However, after the news of Google's $80 billion funding round this week, Microsoft, Google, and IGV (the software ETF) have experienced a slight pullback. What are your thoughts on the future prospects of software stocks?

Frank: Let's start with the software. Earlier this year, I discussed the logic behind Microsoft, IGV, security SaaS, and vertical SaaS. I agree with Huang Renxun's point: as the number of agents increases, they need to call upon more and more tools and assistants to help them; they don't necessarily do everything themselves. But the key is to look at differentiation.

The entire software needs to be differentiated. In summary: any general-purpose software that can complete public tasks without third-party tools is at risk and will likely be replaced. If the AI ​​itself can do what you're doing, why would I need you?

However, three categories are actually valuable, and even strengthened:

The first category is vertical software with deep know-how in a specific field. Large models are trained on public data, but much of the experience and data from these vertical industries resides in companies' private clouds and databases—not publicly available. AI training cannot access or access this data, creating a barrier to entry. Look at database companies like MongoDB; as AI develops, the need for vector databases increases, leading to rapid growth. Similarly, unlisted data container companies like DataStax, which focus more on AI-native technologies, are also in high demand.

The second category is companies that integrate software and hardware. For example, Cloudflare isn't purely software; it has hardware. No matter how advanced your AI is, you can't build CDN data centers in over a hundred cities globally. The more agents there are, the greater the demand for CDN, because users can't tolerate even the slightest latency. It can replace purely technical aspects, but replacing physical ones is still difficult, so this type of logic is hard and resilient. I published an article about security-related SaaS around mid-February, and looking back now, it was published right at the bottom of the market crash: because the more agents there are, the greater the security problems, and these types actually benefit.

The third category consists of software deeply rooted in specific industries (manufacturing, chemical, pharmaceutical). The experience data and parameter tuning are developed and kept secret by each company, making them inaccessible to large AI models. In the future, some companies may even train smaller models internally. Therefore, the logic is: general-purpose software is dangerous, while vertically focused and deeply integrated software with proprietary know-how and private data benefits and is strengthened—it's unavoidable. The same applies to integrated hardware and software software.

VI. Valuation Slash, Performance Slash, and Logic Slash: Three Scenarios for Individual Stock Corrections

Frank: Back to market assessment. After a broad-based rally, there will be a correction. For individual stocks, the logic behind the correction is essentially one of three things: valuation correction, earnings correction, or fundamental correction.

Valuation correction occurs when a stock has risen too much. Originally, the valuation was 20 times earnings, and people bought it because they felt it was a good deal. However, after a few months, it rose to 30 or 40 times earnings. Even though the company is still a good company and its performance is still growing, it's already a bit expensive, and many funds are unwilling to enter and start taking profits. Taking advantage of some shocks, the valuation is reduced from 30 or 40 times earnings back to 25 or 30 times earnings. At this point, funds feel it's a good deal again and can enter because the business is still growing.

The term "killing performance" no longer refers to killing losses, but rather killing "growth falling short of expectations." Previously, it meant a company turning from profit to loss; now, the market expects 50% growth, and if you only deliver 48% or 49%, then sorry, I'm not satisfied. The first-quarter financial report is a prime example: Microsoft's cloud business had market expectations of 39% growth, but actually grew by 38%. Although the CFO kept emphasizing that it was due to insufficient capacity—that if capacity could easily reach 40%—that this coincided with the height of the third wave of debates, causing a market downturn.

The most fatal flaw is the "killing logic": the entire narrative or strategic positioning upon which a company's existence rests is destroyed. Two examples illustrate this. In 2022, Meta experienced a wave of "killing logic," its market capitalization plummeting from over $1 trillion at its 2021 peak to just over $200 billion. This was due to its massive transformation into a metaverse-like ecosystem, the introduction of huge capital expenditures, and ultimately, massive losses. The market realized this approach didn't work, and the entire business logic was disproven. Another example is in the optical interconnect sector: many companies' core logic was "entering Nvidia's supply chain and being purchased by Nvidia," so everyone sought out these targets. But if Nvidia decides you're no longer viable and removes you from the supply chain, then that company's strategic position and competitive landscape within the entire supply chain are completely overturned and rebuilt—that's the "killing logic."

VII. Optical Interconnects and CPO: How Connectivity Will Become the Next Main Theme of AI

Victor: The optical interconnect sector has been very hot this week. Yesterday at COMPUTEX, during Marvell's presentation, Jensen Huang went on stage and directly endorsed Marvell, saying it would be the next trillion-dollar company. The entire optical interconnect sector, after a correction in the previous weeks, has fully recovered, with Marvell rising 30-40%. I was there too, and CEO Matt Murphy's presentation was excellent. What's your view on the optical interconnect sector? Because its revenue performance still needs time to materialize, people are currently more like buying into expectations.

Frank: I also summarized the key points of Murphy's speech yesterday and learned a lot. First, let's talk about the difference in stages: For example, Nvidia's growth in 2023 was driven by expectations, and its actual performance only started to improve in 2024; but the storage industry is the opposite, where performance improves first and then valuation improves. Storage manufacturers have had explosive performance from the third and fourth quarters of last year to the first quarter of this year, but their valuations have been getting lower and lower because people don't believe in the sustainability of their performance and have been wary of the previous cyclical cycles; recently, companies like Micron have started to surge again because many investment banks want to revalue them from cyclical stocks to growth stocks, and their valuations are slowly rising.

Optical and storage are two completely different industry sectors. The global storage market is highly concentrated, with only a handful of companies: Micron and SanDisk in the US, Samsung and SK Hynix in South Korea, and Changxin (CXMT) and Yangtze Memory (YMTC) in China. However, the entire optical and data center interconnect market is much more complex, encompassing both copper and optical components. Optical modules, NPOs, and CPOs have a long supply chain: from indium phosphide (InP) to optical chips, lasers, photoresistors, and more. Currently, the market size is relatively small. The combined investment in optical modules and CPOs is estimated at around $20 billion globally by 2025, with a more optimistic estimate of $90 billion to $100 billion by 2029. In contrast, the storage market revenue last year was already close to $200 billion to $300 billion, with a conservative estimate of $600 billion by 2029 and an optimistic estimate of $1 trillion by 2030. One factor is revenue size, and the other is market concentration. Guang is a relatively small and fragmented sector, but that doesn't mean Guang is bad. It can still reach a scale of hundreds of billions within two years and is growing rapidly.

Murphy's core point is that connectivity will become increasingly important in data centers. As Jensen Huang also mentioned, the computing model in the agent era was distributed: you break down a computational problem into many parts and distribute them to different areas of the data center, making data transmission a more critical point. Murphy also described an ultimate form called the " distanceless data center ": in the future, computing and storage will be "pooled" and decoupled, not necessarily placed together. Storage is storage, computing is computing, and CPUs are CPUs, relying on optical fiber as the connection—maximum efficiency, virtually zero latency—essentially becoming a pluggable data center. This will indeed make connectivity very important.

However, there is also differentiation within the optical interconnect industry. Some companies are experiencing strong performance, such as Innolight, a domestic manufacturer of optical modules. Innolight was indeed the top-performing stock in the A-share market last year, with its share price surging ten to twenty times and its market capitalization approaching one trillion yuan, because it is a major supplier of optical modules. Marvell's overall performance in optical switches is also very solid and growing rapidly. But it should be noted that optical modules are a very mature technology and have been mass-produced and verified, while CPO (Co-Packaged Optics) is still a very new concept.

I previously read and analyzed Bernstein's May article, "The War for Data Center Connectivity." Its main point was that cloud vendors wouldn't dare deploy CPOs on a large scale until at least 2026 or the first half of 2027. There are two reasons: first, Nvidia's CPO optical switches were only recently released; second, while previous optical modules were pluggable—if they failed, you could simply unplug and replace them—CPOs have the optical chips soldered next to the GPU. If a problem occurs, the entire system needs to be returned to the factory, potentially requiring replacement. Therefore, stability requirements are very high, necessitating repeated testing. This year's core technologies are still 1.6T and even 3.2T optical modules, as well as LPOs (Linear Pluggable Optics), NPOs (Near-Package Optics), and better testing equipment and materials (like ABF substrates), because the certainty is relatively high. CPOs are still undergoing gradual validation.

Therefore, in the stock market, some companies' performance rises first, while others rise based on expectations and valuations (the price-to-sales ratio has increased but they may not be profitable yet). The further upstream you go, the more likely some companies with highly hyped themes will see their stock prices rise dozens of times in an instant, but the real realization of their gains will take time.

However, the general direction is undoubtedly correct. Within this general direction, what are the certainties? Packaging, optical engines, light sources, testing, and system platforms are all essential regardless of whether you're doing LPO, NPO, or CPO; these are guaranteed. Then we need to observe NVIDIA's switches, which only started shipping this year, focusing on yield, reliability, and failure rate. High shipment volumes and good reliability will improve market expectations for commercialization in the next two years. Next year, vendors like Lumentum (LITE) will begin large-scale shipments of CPO laser light sources; we'll see how strong these shipment guidelines are. The most crucial indicator to observe is when cloud vendors begin large-scale procurement of CPO switches and CPO interconnection between server racks.

There are also different scenarios: connections within server racks and in encapsulated systems are something NVIDIA excels at; connections within data centers and between server racks are the strength of third-party CPO switch manufacturers.

8. Nokia, Edge Computing, and Physical AI: Extending from the Cloud to the Physical World

Victor: In your previous post, you mentioned that Nokia is positioned between the optical (optical welding) and edge computing (Edge AI) sectors, and that it's part of the NVIDIA ecosystem. These past few days, Qualcomm's keynote at COMPUTEX emphasized Edge AI, and today NXP is also emphasizing both Edge AI and Physical AI. Jensen Huang himself has also been talking about Physical AI. What are your thoughts on Nokia, a company that simultaneously operates in both optical and edge computing?

Frank: Nokia's performance in the optical, and actually more accurately, the entire connectivity (data center interconnect) field, has begun to materialize. It started with mobile phones, then moved into communication equipment, and made many acquisitions in recent years, such as the acquisition of Infinera last year, which strengthened its capabilities in optical transmission and data center connectivity. It has two strengths: first, it has the R&D capability for coherent optical DSP chips and also makes its own pluggable optical modules; second, it has a strong foundation in backbone optical transmission systems and fiber optics. Therefore, 1.6T is already in mass production, and 2.4T and 3.2T are already being tested.

Its Q1 financial report is very clear: the entire optical connectivity business grew by more than 20% year-on-year, AI data center customers (AI & Cloud) grew by 49% year-on-year, and new orders in the quarter amounted to about 1 billion euros, and the forecast was also raised. So it is a proven system-level supplier, not just a component manufacturer.

As for edge computing (which only accounts for about 8% of its overall business structure), I think it's more like a "valuation option." The core catalyst is Nvidia, which is now the driving force. Nvidia made a significant change in its Q1 earnings report: previously, the report only mentioned "data center," but this year it separated edge computing from data center computing as a separate entity. Separating a segment that only accounts for 8% of the total business is itself a signal. It's actually shaping a narrative: I'm not just selling data center chips; I'm an AI factory, a computing power engineer, and I'm building a full-stack operating system, extending not only to the cloud but also to the edge, to the endpoints, and to the physical world.

Here, we need to clarify a few concepts. Edge computing is the opposite of cloud computing: previously, all data was sent back to cloud data centers for unified computation. Edge computing, on the other hand, allows computation to occur close to where the data is generated, such as factory equipment, 5G base stations, hospitals, stores, and even cars—wherever data is generated, it is processed locally. On-device AI has a smaller scope than edge computing: it refers to AI running on mobile phones, PCs, wearables, and in-vehicle terminals, mostly running small models. Physical AI, as I understand it, is an advanced form of edge computing: it involves AI entering the real world and being linked to agents. Essentially, it's an agent online and a bot (physical agent) offline, helping humans perceive, understand, plan autonomously, and execute autonomously, enabling machines to see, judge, and act.

Therefore, Nokia's potential in the edge computing field is related to Nvidia's AI-RAN initiative: Nvidia aims to transform base stations into miniature AI data centers, while Nokia, which already builds base stations for telecom operators, has wireless integration capabilities, and strong operator relationships, can provide edge computing power to each base station. Furthermore, base stations need to be interconnected, and Nokia itself has solutions for high-speed interconnection of computing power. However, all of this is still relatively early in the development process.

I wrote about BlackBerry in early May, and it's a very typical example. On May 7th, I saw a report in the Wall Street Journal, and there were actually signs in April: BlackBerry was collaborating deeply with Nvidia, integrating its QNX endpoint security software into Nvidia's autonomous driving system platform. Many car manufacturers using Nvidia's autonomous driving systems will use it. In the future, QNX won't just be used in cars; industrial equipment, medical equipment, and robots will also use it. The logic is: first, it's tied to a hardware leader; second, it's positioned as the underlying security infrastructure for future edge computing and physical AI. Another very important point is that it's a typical example of a "turnaround." Before April of this year, BlackBerry's stock price had fallen for ten years, from about $80 billion to $2-3 billion. The decline had already priced in all the negative factors, so any slight improvement in performance or expectations afterwards changed the market's perception.

I've been tracking it since I wrote that tweet on May 7th, and it has risen significantly, more than doubling in value, from around 5 yuan to 11 yuan. The core reasons are: first, the demonstration effect of Nvidia, coupled with Nvidia's push for edge computing, naturally created expectations; second, the previous sharp drop prompted some to try and buy the dip; and third, several key events, including obtaining Class D certification from the Federal Risk and Authorization Management Program (FedRAMP), the highest level of certification, reportedly making it the only critical event management provider to meet this standard, highly valuable, and used by many federal governments; and its buyback program of approximately 5% of its outstanding shares in mid-May, indicating that the market feels confident and supported by buying at the bottom, hence its recent strong performance.

IX. Scarcity, Upgrading, and Long-Term: Deconstructing the Three Logics of AI Fund Rotation

Frank: Overall, the current market capital rotation, in my view, boils down to three core logics.

The first reason is the "scarcity logic." Early buyers of Nvidia saw increased expectations in 2023 and improved performance in 2024, leading to a frenzy of GPU purchases. The B-series (Blackwell) used even more HBM (High Bandwidth Memory), and storage itself was already in short supply: the global storage downturn in 2022 and 2023 resulted in a brutal capacity clearing, many manufacturers going bankrupt, and the remaining few closing numerous factories. Capacity was already insufficient, and then suddenly there was such strong demand. The demand initially came from HBM; the B-series sold well, increasing HBM demand, prompting Samsung and SK Hynix to squeeze DRAM and GDDR capacity, pushing it down layer by layer. Now, with more and more model parameters, longer contexts, and richer multimodalities, storage demand is also increasing. Unlike the internet era where search only returned a link ranking, every interaction with a large model and the invocation of an agent requires calculating all relevant data and storing even more data. From DRAM to NAND, to SSDs, to hot storage and cold storage hard drives, capacity is being squeezed at every level, essentially resulting in a shortage.

Starting in April, the anticipated CPU shortage also materialized. Previously, CPUs were considered unimportant in training and inference, but they are crucial in the agency era because tasks like agent scheduling and orchestration require CPUs, which GPUs cannot handle. The CPU-to-GPU ratio used to be roughly 1:8, then 1:4, 1:2, and recently 1:1. Looking at NVIDIA's latest Vera Rubin NVL72 rack, with 72 GPUs and 36 Vera CPUs per rack, the ratio is already 1:2, and it's expected to reach 1:1 in the future, meaning the demand for CPUs will continue to increase. The major CPU manufacturers are Intel and AMD. Intel started raising CPU prices in March of this year, marking the first wave (due to CPU shortages). The second wave is expected to be driven by Apple potentially utilizing Intel's advanced manufacturing process. In April, I outlined the "AI bottleneck environment entering the next stage," which focused on CPUs. I was optimistic about Intel and AMD, and specifically emphasized that the ARM architecture would benefit the most because its power consumption and multi-architecture mode would benefit more from the Agenda era. Moreover, Microsoft, Nvidia, Amazon, and Google all use the ARM architecture in their self-developed chips.

The second is the "upgrade logic." In the optical world, the upgrade logic is as follows: Previously, optical modules were considered sufficient, but now it's recognized that photoelectric conversion efficiency is still low and power consumption is high. The best approach is to solder optical chips next to the GPU. From optical modules to LPO, NPO, and CPO, the entire optical interconnect is constantly upgrading, requiring new packaging technologies. Co-packaging is very difficult and is only just beginning mass production. Another upgrade is in data center power distribution networks, moving from 48V/54V to the currently discussed 800V HVDC (high-voltage direct current): higher voltage, lower current, and lower energy consumption and losses. This technology actually spilled over from the 800V high-voltage fast charging industry chain of China's new energy vehicles. Many power semiconductors originated this way, representing an upgrade for data center power distribution, power management systems, and power switches. The third upgrade is advanced packaging, which is essentially the same as the "Tao (τ) Law" proposed by Huawei: when chips reach their physical limits, the improvement of each generation is not that dramatic (10%, 20%), and it is not cost-effective. Therefore, everyone is turning to 3D stacking, better materials (glass substrates, ceramics), better processes and more precise equipment. What Huang Renxun talked about in Taipei and what we are already promoting is this advanced packaging. He is essentially pushing the entire industry chain forward.

The third is the "long-term logic." This refers to edge computing and Physical AI, which are somewhat like the application layer. It's about transitioning to the real physical world: allowing devices and machines to replace humans in perception, decision-making, and action, performing calculations on the edge and hardware based on small models, instead of all going back to the data center. This goes further ahead, and we'll see robots and autonomous driving in the future. We might have to wait for Tesla's Optimus to be mass-produced, or for robots like Figure in the US. The key is the adoption rate brought about by mass production and cost reduction.

10. SpaceX IPO and the Liquidity Black Hole: The Impact of the Three Giants' IPOs

Mr. Z: SpaceX is going public this month. Will it create a liquidity black hole? How will it affect the tracking of passive funds (ETFs) and the overall market liquidity? After all, this year is very special. Anthropic has also secretly filed for an IPO, and there may be an IPO of OpenAI at the end of the year, but it is currently blocked by Musk's lawsuit.

Frank: It will definitely have an impact. It's going public around June 12th, with a market capitalization of about $1.75 trillion, aiming to raise $75 billion. The IPO allocation is roughly 30% for retail investors, 5% for employees, and 65% for institutions, which means institutions will receive nearly $50 billion. Some institutions that want to participate in the SpaceX IPO either don't have enough cash on hand or have to cut some of their positions to free up cash for subscriptions. That's a key point.

However, the impact might not be that significant. From a purely financial perspective, it might be a minor adjustment in the short term: investors will reduce some weaker or alternative positions, but those with strong fundamentals will remain. The real impact lies in its unique share release rules. Normally, all shares of newly listed US stocks are released after 180 days of lock-up. SpaceX is different: 20% is released after the Q2 earnings report, followed by a fixed release of 7% on days 70, 90, 105, 120, and 135 (totaling approximately 35%), and another fixed release of 28% after the Q3 earnings report. The remaining shares are released after 180 days. The first release also has a performance condition: if the stock price is more than 30% higher than the IPO price for 5 out of the 10 trading days before the Q2 earnings report, an additional 10% will be released on the first release, bringing the total to 30%.

Furthermore, Nasdaq changed its rules, allowing companies to be included in the index after 15 days of listing. However, the index weighting is based on free float, not total market capitalization. For example, if a company has a market capitalization of $1.8 trillion and free float increases from $75 billion to $80 billion, and this weighting is multiplied by a factor, its index weight will increase significantly, impacting the buying interest of many passive index ETFs. The purpose of this design is to allow more passive funds to absorb and hedge against future selling pressure. However, a problem does exist: if the company's size and absolute value are too large, the amount of shares released from lock-up later will be substantial.

With Anthropic having already filed for its IPO and OpenAI's IPO also expected to happen this year, I think there's a high probability that we'll see a correction in the second half of the year due to funding issues. Moreover, the market's recent strong performance is due to strong fundamentals; we've overlooked macroeconomics: the Strait of Hormuz has been closed for over 100 days, oil prices are still above $90, and the US-Iran talks are still unresolved. Next week, on June 13th, the May CPI will be released. The key focus will be on the broad CPI and core CPI: in the previous two months, inflation was mainly driven by energy (oil prices, jet fuel) and hadn't yet spread to the service sector; if both the broad and core CPIs rise, it means inflation is spreading from energy, which could panic the market. Coupled with expectations that the Fed won't cut interest rates, this will affect sentiment.

However, within my framework: if it's just a matter of liquidity or a significantly higher-than-expected May CPI, the resulting adjustment should be minor, as the fundamentals are sound. If this is compounded by persistently high inflation, the IPOs of the three major tech giants, and geopolitical uncertainties, it could lead to a minor to medium-level adjustment. But as long as the revenue growth of large-scale AI manufacturers and cloud businesses doesn't slow down, it won't be a crash like the one in 2000. US stock market corrections are swift and decisive, but once a reversal is confirmed, it rises rapidly and without hesitation. Only when the fundamentals (AI commercialization growth rate) truly falter will a more significant reversal signal be needed after the adjustment.

11. Major, Medium, and Minor Level Adjustments: Only a "Reset of Macroeconomic Logic" Constitutes a True Collapse

Frank: Last February or March, I summarized the patterns of correction levels in the US stock market over the past 20 years, using the Nasdaq as a benchmark. I have always advised everyone to look at the Nasdaq when analyzing the overall market, because it represents the purest form of technology stocks; the S&P 500 has 12 sectors, including defensive sectors.

Major fluctuations are above 25%, occurring roughly four to five times: a 50% drop in 2008, 25%-30% in Q4 2018, 30%-40% in March 2020, 33%-35% in 2022, and approximately 28% last year. Medium-sized fluctuations are around 15%, occurring once every one to two years. Minor fluctuations are in the single digits , like the 7-8% drop last November, which was a liquidity shock at the end of balance sheet reduction (not a crisis, shocks and crises are different) coupled with the nascent doubts about the massive capital expenditures in AI. This year, the Nasdaq has actually reached a medium-sized fluctuation, around 13%-14%, not quite reaching 15%.

Minor adjustments typically occur when the market has risen too much and there is an inherent urge to reduce valuations. This is compounded by macroeconomic disturbances, such as liquidity shocks, soaring inflation leading to increased expectations of interest rate hikes and the disappearance of expectations of interest rate cuts. As a result, people begin to seek safe havens and reduce valuations.

Mid-level shocks are always accompanied by major macroeconomic events. For example: the US-Iran war this year (higher oil prices, inflation, and the Middle East being an upstream source for many semiconductor materials and gases); the US stock market fell by about 15% from August to October 2023 due to a rebound in inflation and the 10-year US Treasury yield soaring to 5%; the unwinding of yen carry trades in July and August 2024 was due to two factors: Japan's unexpected interest rate hike and the weaker-than-expected non-farm payrolls in July, leading to strong recessionary expectations, a flash crash on August 5th, Bitcoin falling to $50,000, and the Nasdaq nearly triggering a circuit breaker and falling 6% in pre-market trading. Later that evening, the services ISM index exceeded expectations, alleviating recessionary concerns, and the market recovered.

A major event that could cause the market to drop by more than 25% must be due to a "reset of the entire macroeconomic logic." What does this mean? Last year's tariff war: Previous tariffs were between the US and China, starting in 2018, but last year the US launched its first trade war against its allies (the Five Eyes alliance, NATO, and NAFTA). This challenged the decades-old free trade system and order established after World War II. The market worried about the collapse of the US's credit foundation and the arrival of a Great Depression, hence the significant impact. 2022 saw the highest inflation in nearly 40 years (reaching 9.1% at one point), coupled with the fastest pace of interest rate hikes in nearly 40 years (a 0.25% increase in March, a 0.5% increase in May, followed by two consecutive 0.75% increases in June and July). The market is discussing a resurgence of the stagflation of the 1970s. March 2020 saw the first large-scale global pandemic in a century (since the 1918 Spanish flu). In late February, cases suddenly appeared in New York, Paris, Italy, and London, with some fatalities. Public opinion was talking about the Black Death and the 1918 flu. In the face of this panic over the strain on medical resources, people were on the verge of death, so what was the point of talking about the capital market? Thus, the most severe sell-off occurred, with the Nasdaq falling by more than 30% and nearly 40% within a month.

The trade war between the US and China in the fourth quarter of 2018 marked its first outbreak, transforming a previously harmonious relationship into a bitter conflict between the two major powers. As for the 2008 financial crisis, it goes without saying that there are subtle details: the US housing bubble actually burst in the second and third quarters of 2007, with a large portion of mortgage-backed financial institutions already bankrupt. However, the US stock market didn't fall at that time; it continued to rise until September because the Federal Reserve began cutting interest rates to provide bailouts. The real confirmation came in September 2007 when Northern Rock, Citigroup, BNP Paribas, and Credit Suisse reported huge losses in subprime funds. Only then did the market realize that this wasn't just a housing bubble but had spread to the entire financial system, and that defaults at the bottom would trigger a chain reaction of losses. Many people believe that the collapse of Lehman Brothers marked the beginning of the crisis. In fact, after Lehman's collapse in September 2008 and AIG's collapse in mid-September, no major companies went bankrupt. In October, the US Congress passed an unprecedented bailout bill and approved QE, marking the first time quantitative easing (before that, the Federal Reserve mainly relied on interest rate adjustments, and QE was not a major tool). At that time, the financial crisis was actually nearing its end.

Therefore, every major correction is triggered by a reset of our past macroeconomic logic. For the US stock market to experience a major correction now, it would either be due to a reset of logic (AI growth and commercialization falling short of expectations) or a complete collapse or breakdown of the US-led order. However, we cannot see this at present, since US-listed companies are still among the most globalized and innovative, which is their strongest advantage.

12. Bitcoin and the New Crypto Normal: Differentiation, the Combination of Capital Paradigms and AI

Victor: Bitcoin has recently experienced a significant sell-off, and it is often highly sensitive to macro liquidity and frequently serves as a leading indicator for the US stock market. Given that liquidity is being absorbed by US stock market AI, coupled with Saylor's potential selling of cryptocurrencies, what are your thoughts on the future development of Bitcoin and the entire cryptocurrency sector, as well as the "new round of asset migration"? Recently, Binance also listed US stocks, and major exchanges have all listed US stocks.

Frank: I wrote a tweet in July 2024 about my understanding of altcoins. At that time, my view was that "the cryptocurrency market has entered a new normal," and there were several points of logic behind it.

First, with the approval of ETFs, investors and funds in the cryptocurrency market are becoming more sophisticated. A more mature market is inevitably more efficient and differentiated. What is an inefficient market? It's a market where everything rises and falls together—everything soars when prices rise, and everything falls when prices fall—a characteristic of early, unregulated markets. The US stock market wasn't mature from the start either. The Buttonwood Agreement of 1792 was the precursor to Wall Street, the Federal Reserve was established in 1913, and the SEC in 1933. For the first century or so, it relied heavily on industry self-regulation and various market manipulation schemes. In mature markets, funds focus on two things: certainty and growth potential. Certainty refers to whether the asset will still be around in three or five years; growth potential refers to high annualized growth and development that overcomes all other issues. Funds will revolve around these two types of assets. The same applies to the cryptocurrency market. Bitcoin has grown from a few thousand dollars to tens of thousands, then to $100,000, creating increasingly large funds. Larger funds tend to have more conservative risk appetites, making it difficult to encourage them to invest in altcoins.

Secondly, the ICO and DeFi eras of the past were essentially about speculating on expectations. DeFi still had fundamentals, but GameFi, SocialFi, and NFTs basically have no fundamentals left. They lack certainty and ultimately have no growth potential, with weak innovation and value creation.

Third, 2021 and 2022 were two years of explosive growth for Crypto VC. Many VCs emerged managing hundreds of millions, billions, or even tens of billions of dollars (like a16z, Paradigm, and Multicoin). VC investment focuses on early-stage size and low failure rates: a fund with tens of millions of dollars might invest $1 million per project, investing in 50 projects, requiring reviewing hundreds or thousands of projects; but a fund with $1 billion or $2 billion might invest $10 million per project, reviewing tens of thousands of projects over several years. Therefore, larger VC funds naturally drive up valuations in the primary market. In 2022 and 2023, some crypto projects in the primary market were valued at billions of dollars, and upon unlocking, a few percent of the circulating supply could easily reach hundreds of millions or billions of dollars; in contrast, projects like Solana in 2020 and 2021 were only worth tens of millions of dollars at their IPOs. Larger scale requires more capital, but current monetary policy is limited and cannot support this. The core issue is still insufficient fundamentals, lack of innovation, and a lack of further creation. If the fundamentals and value creation are truly sufficient, then, like AI and the US stock market, those with certainty, trends, expectations, and performance will still rise.

In June 2024, I conducted a statistical analysis: Of the approximately 4,000 companies listed on the US stock market, the top 100 globally accounted for 91%-92% of the total market capitalization. The 3,000 smallest companies, with a combined market capitalization of only 6%-7%, averaged a few billion US dollars and had a daily trading volume of only two to three million US dollars. At that time, these companies probably wouldn't even rank in the top 100 of the cryptocurrency market. Therefore, the US stock market is inherently a highly differentiated market, and this becomes even more pronounced as it matures. As the cryptocurrency market enters a new normal, it will also become increasingly differentiated, with only a very small percentage of assets attracting significant investment.

Last August, I wrote a thesis with the core argument that crypto is essentially a "capital paradigm," not a "productivity model." AI, mobile internet, and the internet are productive forces, capable of revolutionary innovation and enormous value creation; the capital paradigm itself rarely creates new value. Mobile internet has a concept of "10x efficiency improvement": users use you because you're 10 times more efficient than the old system, naturally leading to large-scale adoption. So where has the crypto world achieved 10x efficiency improvement over the years? The answer lies in the issuance, trading, and circulation of assets. Issuing assets is extremely easy, trading is available 24/7, and global liquidity is unlimited—this is where its 10x efficiency improvement and capital paradigm lie.

This is also a key reason why the SEC wants to move the entire US financial system onto the blockchain. At the first meeting of the Tokenization and Cryptoworking Group last year, SEC Chairman Paul Atkins clearly stated that the US financial system should be moved from off-chain to on-chain. He believes that blockchain can reshape securities issuance and some previously unseen activities, citing two examples: first, automating dividend payments through smart contracts (in the future, if company revenue is held in stablecoins and stocks in token form, dividends can be automatically paid); second, allowing illiquid assets to reach a wider audience, improve trading depth, and become globally accessible.

So, in that deduction, I said: if blockchain is merely a capital paradigm, then the previously held beliefs about Web3 (games, social media, e-commerce) are invalid. It can only bring about a major explosion if it's closely integrated with reality and has real assets. The second direction is integration with AI. I discussed this around 2023 or 2024: the native inhabitants of the mobile internet are DAU (Daily Active Users, which are people), while the native inhabitants of the future virtual world may be bots or agents—agents online, bots offline. How will bots interact and incentivize each other to accomplish a task? It will most likely require smart contracts and cryptocurrencies, which is what everyone is currently hyping as agent payments and Atkins' agent finance. If a large amount of assets are on-chain in the future, this will be an even better integration method.

In the long run, crypto has real value in two directions: First, as a capital paradigm, it must be combined with real businesses and real assets; otherwise, the token economy is just empty talk without positive value injection. Second, it uses crypto to incentivize other bots, agents, and humans to complete tasks, becoming the primary medium of interaction and value transfer. The benefit here is that if there are enough on-chain assets, DeFi (decentralized finance) will experience a significant boom. DeFi is essentially finance and requires high-quality underlying collateral assets, but since 2021, the TVL (total value locked) of DeFi has not changed much. The core issue is that there are too few on-chain underlying assets that can serve as high-quality collateral. To put it bluntly, it's mainly Bitcoin. Ethereum used to be acceptable, but now five or six out of ten people probably don't consider it a high-quality asset. With too few underlying assets, lending and contracts cannot scale. Only when a large number of underlying assets (such as bringing assets like Apple onto the blockchain) appear can the scale of on-chain lending and contracts become very large, and many new applications can be derived. Of course, this point is somewhat pessimistic, as it may degenerate into a purely capital paradigm. In addition, I also think that stablecoins are very likely to become the settlement layer for global payments in the future and replace SWIFT.

Returning to the future of Bitcoin: Crypto is indeed in a rather uncertain and exploitative state right now, but Bitcoin has its unique and proven value. Looking at a four-year cycle, this year is its downward cycle, and it's possible that it will encounter a relatively good bottom range sometime in the second half of the year. I think anything below 70,000 is considered a bottom range now, although we don't know if the absolute bottom will be 60,000, 50,000, or even lower. But for trading, the core issue is: when it truly reaches its lowest point, you become more panicked, thinking it will go even lower. The lowest point is always discovered in hindsight. How many people actually bought at the bottom when it was around 15,000 in 2022? Most people started buying slowly from below 20,000, or only started chasing the price after it rose. Therefore, it might be a time to enter a cost-effective range and gradually build positions; we just don't know exactly where the lowest point will be.

Thirteen, advice for the audience: Master offense and defense, and delve into the fundamentals.

Mr. Z & Victor: Finally, does Mr. Frank have any words of advice for our audience on how to follow the market trend and maximize profits while minimizing risks?

Frank: I think the core is still to try. First, in the financial world, you must be good at both offense and defense, and clearly understand what offense and defense mean. Because some things in the market are always unpredictable, we may talk a good game now, but we may be proven wrong later. Since we can't predict it, we should be good at defense while attacking: it's okay to retreat a little when market sentiment is very high and leave yourself some room, but when the market is really pessimistic enough and it's time to act, you still have to act, otherwise you may miss opportunities.

Secondly, with stocks, you need to delve into the company's fundamentals. You need to examine its competitive environment within the industry chain, whether its competitive advantages are irreplaceable, its competitive landscape, and its strengths. Once you have a general understanding, you can determine when it's overvalued and when it's entered a value-for-money range. When it's in a value-for-money range, you should buy gradually; but when it's absolutely overvalued or when there's excessive enthusiasm, you might need to reduce your holdings to leave yourself some room for maneuver. I think this is very important.