All conditioned things are impermanent. When you see that all things are not-self, you see the Tathagata.

As a die-hard AI enthusiast, I know that any hint of pessimism amid a bull run will inevitably be met with ridicule and scorn.

However, I wanted to write this piece not only for introspection and reflection on my own actions and trades over the past two and a half years but also on the madness of this market.

Fortunately, as a hobbyist writer, engaging in trades has allowed me to make many new friends.

I am grateful for the mutual appreciation and inspiration. Over the past 2 years, I transitioned from fully divesting BTC in early April at age 24 to AVGO and NVDA, pre-election into AAOI, PLTR, TSLA, INTC, post-trade war into INTC, GOOG, increased holdings in INTC and AAOI at the end of 25, exited before the Chinese New Year, then fully loaded on AMD and NOK (recently also made moves on ORCL and LITE).

Although I experienced losses in speculative plays like SUP and MRVL, I have always believed that AI's demand for tokens would benefit the semiconductor industry. Every dog has its day, and I have achieved substantial gains. Many of my friends have also seen returns of over 10x.

However, I now realize that our stock-picking prowess and foresight were not the only factors behind our gains; it was also due to the overall market momentum. Reflecting on my actions in the last week of March, when I went all-in before the weekend, I see that my profits were not solely because I chose AMD correctly but because I correctly assessed the market's fund flows at that time.

Therefore, at this juncture, it is crucial to remain calm and contemplate how the market will proceed.

Drunk people do not know they are drunk, just as those lost in dreams do not realize it was all just a dream. This process can last a long time. We may continue to be wild, but we must maintain a sober awareness at a higher level of consciousness, always aware of the imminent risk of being awakened.

One of the four beasts saying, COME AND SEE – Revelation 6:1

Unlike many on Wall Street, I am a true believer in AI.

For the first time in 30 years, the AI model has replaced serial computing with parallel computing, directly transforming semiconductor advancements into productivity. (Previously, technological progress merely served as a platform for software users and creators to add value.)

The result of AI is a token demand surge at an exponential rate, while semiconductor supply can only increase linearly.

In addition, the entire semiconductor industry did not expand production capacity due to "faith" during the golden time of the past few years. After the demand truly started to surge this year, the entire industry chain faced shortages.

This has led to this bull market.

However, the lifeblood of the bull market is that the market has priced in:

The model's income acceleration, which can soon justify Hyperscalers' Capex

It's important to note that anything that shakes the lifeblood expectations will plunge the market into panic.

1. Impregnability: one of the horsemen is indeed invulnerable at the micro, company level

From examining model architecture, from architecture to demand, and observing the semiconductor landscape at a micro level, every company is nearly flawless. The more you scrutinize each one, the more you delve into understanding, the more confident you become—this is a typical low PE bubble.

Rule of Thumb: When you have looked at dozens, even hundreds of companies, and each one's demand, earnings are truly solid, you must question whether there is a problem in this market.

The low PE bubble is still a bubble, but it is different from the one in 2000. A high PE is an expectation bubble; the market will continuously set higher profit expectations. Once the core company's expectations are shattered, the overall expectations immediately adjust, and it's like a domino effect—pieces fall one after the other. Just like last year's IGV, suddenly the market no longer gave EV/EBITDA valuation.

However, the low PE bubble does not have this problem. Earnings growth will give you enough confidence to keep buying more after a slight dip. After all, who can resist single-digit PE in storage, with price and earnings rising every year, 80% gross margin. In this process, you buy a little on small dips and buy heavily on big dips.

As long as it doesn't hit the "fatal point," the market will remain very resilient.

Continuous positive feedback, like a beautiful dream that makes people reluctant to wake up, unwilling to believe it's a dream. (Those who have invested in P2P know the feeling I'm talking about)

Asymmetric feedback: A clear experience over the past year is that the market's feedback on macro negative risks is significantly weaker than on positive news. At this point, it's not that there are no "barbarians," but the fortress of companies' profitability is impregnable. Regardless of the Iran crisis, oil prices, inflation, interest rates, or the Fed, everything is kept at bay.

On the contrary, once the gate is broken, all barbarians will come back to kill, making up for the lack of previous falls in one fell swoop.

2. Deregulation and Macro Liquidity

The entire financial system has undergone a long period of regulation and deleveraging since the 2008 financial tsunami. In order to compensate for systemic liquidity losses, the Fed launched a continuous round of QE. At that time, banks had an average leverage of x30~x40, which later decreased to x15~20. Last year, the Fed reduced the eSLR requirement, effectively allowing banks to expand their balance sheets by $4-5 trillion in theory.

In fact, after April, interbank liquidity flooded, exacerbating the speculative frenzy of risk assets and leverage entry into the market at low PEs.

Liquidity is the key to everything, and abundant liquidity conquers all. If the $120 billion per month during the epidemic was a trickle, now it is a deluge. With no shift from past balanced QE to QT, direct laissez-faire regulation of banks has led to a massive influx of liquidity.

However, liquidity is like dopamine, not a molecule of happiness, but of anticipation.

Once let loose, now stuck in inflation, the Fed actually has no room to cut interest rates in the short term. Even after a crisis occurs, it is difficult for a Fed that only has interest rate tools to engage in QE, making it a paper tiger in the face of a possible crisis. On the other side of short-term liquidity is the fact that leverage in the semiconductor sector, especially in storage, is already too high.

3. Rise of Heroes: He Went Forth Conquering, and to Conquer

Perhaps because TSMC led the way and did not believe in AI, the semiconductor industry as a whole has not responded to the large-scale expansion of AI demand in the past 2 years.

The consequence is that when the entire industry is faced with a flood of silicon-based demand, it suddenly finds itself at a loss. In the past, the semiconductor industry chain was mainly designed and priced for consumer electronics. Now, not only are orders suddenly full, but continuous architectural iterations are also required.

NVDA has progressed from a GPU fabless model to a rack solution, and is moving towards a token factory. Each process requires different technologies and suppliers. None of the suppliers have undergone such mass production design. It's like forcing a tractor onto a high-speed train and letting it race at 200 miles per hour; every part will buzz. Every segment of the semiconductor industry is facing shortages, insufficient production capacity, and not enough increased capacity.

Associated with this is an issue: in the past, in a seller's market, only TSMC, a "gatekeeper," controlled the supply, margins, and prices. The industry that once benefitted from the original soup now has meat to eat, with demand-side expectations for prices and capacity, leading to a benign development.

Today, the sudden five hundred million Inspector "Reno" lost control, allowing every "thug" to raise prices and demand protection money. The real outcome is not that the semiconductor field becomes more powerful, but rather spirals out of control: this loss of control will affect the cost calculation for each GW, impacting the lifeline's expectations. It will cast a shadow over capex and revenue model convergence.

4. Wire Dance

The AI revenue models are overly optimistic.

In retrospect, the market was convinced that pricing in the revenue from Anthropic + OpenAI + Gemini would rapidly increase and make Capex reasonable. This unwavering faith in the lifeline cannot falter in the slightest. Any slight shake-up would lead to violent fluctuations in the secondary market.

So, in the past, when GPT 5.0's models fell short of expectations, why didn't the semiconductor industry experience a significant slump as the market questioned the scaling law?

Firstly, at that time, the semiconductor industry had not blossomed as comprehensively as it has today, nor did it have such high leverage; secondly, at that time, hyperscalers had more cash flow redundancy, and as long as they indicated their belief, providing money for capex support would easily help them through the tough times.

After all, once they had money and continued capex, with visible high certainty of NVDA returns for 2 years, no one could short NVDA against the wind.

Later, there was also a situation where GPT was equated by Gemini, leading to Orcl holding a substantial backlog of orders being shorted. However, the good buddy of the big boss, Old Larry, dared to say F you to Wall Street, continuing to leverage equity financing. Soon, Opus emerged, making people realize that the AGI era had arrived.

However, this year's capex has already reached 770 billion, and next year it will be 1 trillion. Wall Street doesn't care how much total revenue Hyperscalers have (just like ORCL at the time: OAI Anthropic was all funded by their investments); they must see the sustained growth of Anthropic and OAI, which is the core that keeps the entire chain spinning.

At the same time, the major players in this game have always played the role of AI guardians. But their free cash flow has turned negative. "The parents have aged, exhausted their strength," and the road ahead must rely on themselves. Moreover, in a market full of expectations, with an unwavering focus on this path, there is no margin for error on the "mainline."

(You can recall the NV rack liquid cooling issue, the switch CPO yield issue – these small problems can be solved with time, but the market does not provide any margin for error.)

The Key to Success Lies in the Anthropic Principle: To surpass is to make mistakes on a grand scale. To rely on a single pillar is to weaken the foundation.

The issue of supply delays and even technical bottlenecks is a deep-seated problem; any disturbance at the model's edge is the biggest concern.

The current universal issue with the three major models: insufficient computing power, hence the loss of intelligence.

Through my extensive use of model programming and trading program deployment, and the interaction of multiple AWS services, I have found that the actual capability of Opus4.8 is far inferior to the Chinese model kimi. Although GPT can still barely serve its purpose, it is also gradually losing its intelligence. The market tends to consider AI with traditional software thinking, where system development costs are high, but usage costs are very low.

By testing usability now, we can ensure stable system quality in the future. However, AI is like a factory; model outputs incur costs. If too many people eat at the restaurant, the kitchen can't keep up, and naturally, the quality deteriorates.

The second market's traditional mindset: the demand for index tokens should be constant, unaffected by quality changes. In reality, although I have used many more tokens due to the model's stupidity, I suspect that the continuous decline in token quality and the strange scissor effect of increased token consumption can persist.

Many companies indeed now have KPI assessments for token usage, and many inertial subscription users are accelerating their usage. But if the computing power bottleneck cannot be resolved, I am very concerned that Anthropic's growth curve will flatten. Moreover, the computing power issue is not something that can be solved overnight or by a new model.

However, when the market begins to realize this problem, investors will question:

1. If your model can't beat distillation, why spend so much money on training?

2. Has the model been commoditized?

As the "dream" of the model that carries the entire AI era comes into question, whether the entire revenue logic can sustain itself in the future will be doubted. All semiconductor companies that are valued with PE today will similarly be questioned—Are you cyclical?

Some may say that if the lack of intelligence is due to insufficient computing power, then investment in semiconductors should be increased to solve the problem.

That's right! But what about the money?

If we exclude the funding that OAI and Anthropic received from cloud giants, how much net cash flow do they have to continue to support this investment? This calculation is very simple for Wall Street. Of course, there is the possibility that they will be like orcl back then, even if they have to dilute their equity, they will continue to invest to get through this challenge.

By then, the market's reaction may far exceed last year's. Assuming AMZN Meta faces a CDS surge, negative cash flow, whether he can go all-in like old Larry is uncertain.

Some might argue whether OAI revenue growth can offset investor concerns. In such a highly anticipated scenario, being replaced due to "being useless" is hardly reassuring (in a battle where Guan Yu was killed, you think there's still a Zhang Fei? Useless)

At the same time, we must also be aware. This Low PE bubble is also very resilient, nearly invincible until its vital expectations are directly hit. However, once its fatal flaw is exposed, it will collapse in an instant.

Throughout my career, I have witnessed several instances of paper shredding gold rush, ups and downs, none of the frenzies ended well;

Nor has anyone dreaming within it understood how to "lose" in this race. The most recent case was the DeFi summer familiar to the crypto community.

Uniswap swept tens of billions USD in a very short time, back then thinking that replacing some traditional finance was just a matter of time, a matter of quantity, never thought, nor understood "how to lose." However, if a drunk person understands how to lose, he won't lose.

Whatever the Anthropic model, whatever issues it faces, it will not shake the fact that AI is humanity's third industrial revolution. The AI process will not be affected by any crisis, rolling over the bodies of revelers and will continue until it replaces billions of intellectual laborers.

Final comment on the Low PE bubble: Don't talk about going short, even stepping on air, the pain index is very high.

Looking back at the funds that shorted subprime in 2007, how many of them survived the final frenzy?

In an environment where being on the sidelines during a bull market will become a laughingstock, I will need a lot of Aura power and focused strength to both bearish and bullish simultaneously.

Forgive me for reducing my cultivation time on Twitter, preaching:

You can drink, but not get drunk;

You can tell stories, but should not believe them;

You can dance on the party table, but your eyes must be fixed on the DJ. If he changes the music and runs, you have to run with him.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia