In those days of innocence, we openly pitched one name, Samsung Electronics, to Warren Buffett and a year later explored why Elon Musk might be looking at it. A company written off just a few quarters ago now sits on a plausible trajectory toward being one of the most profitable enterprises on earth by 2027. The journey has involved underappreciated changes in LLM architectures that made its products more desired than the hallowed processors, a supply gap that few foresaw, and remarkable business decisions. 

For a change, the company’s eightfold profit increase is well noted. Until now, we were the only quirky ones trying to highlight the enormity of these companies’ (including SK Hynix) absolute achievements; thankfully, now there is an industry trying to come up with new ways to showcase the same using more creative yardsticks.

And of course, the stocks trade at around 5 times forward earnings, despite EPS projections for the year ahead in the same vicinity as their stock prices were at some point last year. All-time high profits. All-time low multiples. And Samsung Electronics rose barely 2% on the day of the announcement.

We are glad about at least one thing. Nobody is now arguing that because one of the two Korean memory makers is doing well, the other must be doomed, that they are locked in a suicide pact for market share, and possibly none may survive either's success! That particular obsession produced several essays from us last year, which we obliged with some enthusiasm. The time to move on has arrived. There are newer and stranger oddities to examine.

Standard equity analysis lacks a clear framework for what is happening here. One has to reach into behavioral finance, and even there, the familiar tools require adaptation. What we appear to be witnessing is not anchoring or loss aversion in their textbook forms. It appears to be something newer, something the field has gestured at without fully naming. We will attempt four names. We promise our readers: Samsung and SK Hynix are the occasion for the inquiry into behavioral traits, not the subject of another note on them.

I. The Safety of the Eventual

There exists a class of statements that functions as the preferred instrument of sophisticated caution in technology markets. They are always correct. They cannot be falsified in any near-term window. And their time-invariance sounds, in each deployment, precisely like insight.

Scaling laws will hit limits. AI's transformative effects will turn out to have been overstated relative to the capital deployed. AI will disappoint relative to expectations. What goes up in semiconductors comes down. None of these is false. They describe real dynamics operating on unspecified timelines. Their rhetorical power derives from that unspecified timeline, which is also their entire analytical uselessness. They satisfy what behavioral scientists call closure bias: the deep need to resolve uncertainty with a tidy intellectual bow on a horizon long enough to guarantee eventual vindication.

In data-rich environments, sophisticated actors absorb extraordinary numerical outputs without constructing a coherent story about what those numbers imply. The numbers are filed as statistical anomalies, not structural signals. This is not stupidity. It is a different optimization function.

Kahneman documented how System 1 substitutes easy questions for hard ones. The easy question, will scaling laws eventually hit limits, is emotionally irresistible because it guarantees eventual vindication. The hard question, whether the current supply structure of high bandwidth memory justifies a sub-five times forward earnings multiple for the leading producers, is ignored. The speaker checks their sobriety status. The market prices the present.

The behavioral pattern has a long history in technology. GPU economics were confidently predicted to commoditize; the forecast became correct roughly a decade after it ceased to matter. Throughout the 2010s, a credentialed community argued that photovoltaic cost curves were approaching physical limits, that further reductions were implausible, and that the energy transition would disappoint. The cost curves did not consult these forecasts. They kept falling. The forecast was correct about the existence of limits. It was wrong about everything that determined practical relevance.

GPU scaling was said to slam into power walls by 2024. Cluster-level optimizations, liquid cooling, and software efficiency shifted the constraint while analysts treated power budgets as fixed physics. High-quality data was projected to be exhausted by 2025. Synthetic generation and self-play kept models advancing on reasoning benchmarks long after peak data predictions. One comes across these statements in all innovation segments, and perhaps in everywhere else. Each such prediction will be true someday. There is no talk of today.

A person who refuses to take a new job because economic cycles eventually turn is not wrong. They are simply applying a timeless truth to a decision that requires timing. The rhetorical panic about AI profitability found a quiet answer for now, and may be almost to the last digit. Those looking for US$600bn in profits may find the figure being achieved if they add the profits of Asia’s three semiconductor makers from Suwon, Icheon, and Hsinchu to those of NVIDIA and a few others. The skeptics will have to look at segments whose profits they may not consider as consideration-worthy, and unfortunately, we do not have a solution for that.

The prophecy will surely adapt. Those who forecast cyclical reversal when memory profits were modest now face different arithmetic. The new formulation is that growth rates cannot be sustained at such elevated levels. Today’s announcement is tomorrow’s denominator. The high base will soon be the new tool in the arsenal. The inevitable deceleration, and perhaps even a fall, from extraordinary to merely excellent, will be recast as vindication of the original caution. The statement remains true. The pricing peculiarity persists.

II. What Happens to a Company When It Suddenly Has Everything

There is a question markets have not seriously posed about the memory complex, and it is not about earnings sustainability. It is about identity.

A person who suddenly acquires extraordinary wealth does not remain the same person. Mental accounts are recalibrated. Ambitions reconstitute around new possibilities. Risk tolerance shifts. The person at the end of the transformation is categorically different from the person at the beginning. The same is true of a corporation. When annual earnings begin to exceed what the entire enterprise was worth in the best of times, the mind still files the company under its original category. This is Substrate Blindness: the cognitive gap around the value of the layer that enables the hero product but is not the hero product itself.

Amazon's retailer label survived well into the AWS era, Nvidia's gaming identity masked the CUDA monopoly, and Microsoft, Meta, and Google each passed through the same category lag while extraordinary cash flows rewrote their optionality. The stories are known. The mechanism repeats anyway.

Reliance began as a textiles business, synthetic fabrics being hardly frontier technology, yet the cash flows funded petrochemicals, refining, telecom, retail, and new energy, with the analytical framework appropriate for each prior identity producing systematic error at every transition. The global shipping container was an invisible steel box for fifty years. Its standardization rewired global trade while the mind filed it as low-value infrastructure. The port operators who controlled it became among the most profitable entities on earth. The retailers who depended on it got the attention. The substrate was invisible until it was not.

Memory is the substrate of the AI economy. It is treated with the same invisibility.

Thaler described how people assign cash flows to mental categories that resist revision. The category labeled Korean memory manufacturer was calibrated on a business model that no longer exists in its original form. Applying that category's discount rate to the current business produces the valuation we observe. A company generating a hundred billion dollars a year is not a larger version of the company that generated ten billion. It is a different entity with a different feasible set, different bargaining power, and different strategic optionality. The cash is not a reward for past execution. It is fuel for future transformation.

This is Denominator Decay. The denominator feels stale, a rounding error of history rather than evidence of step function change. A ten percent earnings increase registers. A three hundred percent increase registers, barely. A shift so large it inverts the numerator-denominator relationship is cognitively illegible. The mind still files the company under ingredient. The earnings have already become the meal.

Samsung and SK Hynix are funding HBM4 capacity, advanced packaging facilities, foundry programs, and custom silicon development without leverage, and they are soon going to embark on much more. The skeptics will see this as the new way to destroy value. In more likelihood, these giants will integrate and expand in ways to increase the strength of their moats. We will see. 

III. The Peculiar Physics of Momentum and Value Occupying the Same Stock

What was, not long ago, the quiet observation of a small number of investors who found the memory valuation divergence remarkable has become market noise. Every analyst now has a comparison. The profits of the Korean memory complex are being measured against the GDP of countries, the earnings of entire indices, and the combined revenues of industry sectors. These comparisons, and the shares’ rise, are grabbing global headlines. 

The retail investor has arrived. Not through analysis of HBM yield curves, but through Korean ETFs listed in non-Korean markets, through leveraged instruments tracking Korean technology indices, through the momentum that attaches to any story large enough to escape its original audience. The stocks’ multiples have not changed as their surge can barely keep pace with those in profits, but the counters at their single-digit multiples have become the momentum trade accessible to anyone with a brokerage account, with meme-stock-like volatility in any short term.

This is the Proxy Perception Trap. The market reduces a complex, evolving entity to a simplified symbol for broader trends: AI capex sentiment, Korean tech exposure, global growth proxy. Idiosyncratic achievements are subsumed by symbolic utility. The stock is no longer priced as a company. It is used as an instrument reflecting the market mood on the day.

Sophisticated observers may simultaneously believe a structural transformation is underway and treat results as merely cyclical, holding both models without resolving the tension. This produces accurate forecasts paired with muted conviction and no change in position sizing in light of the additional risk brought in by the presence of momentum investors. 

So we have the unprecedented case of the same asset that is simultaneously a deeply undervalued fundamental position and one of the most actively traded momentum instruments in global markets. Momentum and meme stocks, as the practice goes, are assumed to trade at high double digit multiple of their sales. 

We see this in real life: a person's public reputation can become a proxy for a broader cultural movement while their private conduct remains entirely unchanged. In many minds, memory stocks are now tools to trade sentiments. Their stock-specific factors become a casualty, just like on the day of Samsung’s results. Even if the sentiments were better, the stock might have gone up somewhat more, but nowhere near the extent of the surprise that might have justified. Globally, technology has countless examples of stocks that move by high double digits in a session on a small earnings miss or a beat. The Korean giants have an excess liquidity discount: they simply cannot suddenly jump 30 or 50% on almost any type of stock-specific news.

The behavioral failure this produces is circular. The stock's volatility is recruited as evidence of uncertainty. The uncertainty justifies a discount. The discount produces a low multiple. The low multiple also creates a starting point or historic average that locks in the arguments for tomorrow’s share price on incremental newsflow and catalyst analysis.

The fundamental question, what multiple does a structurally constrained oligopolist in AI critical infrastructure deserve, is never cleanly answered because the price action always seems to be saying something that demands immediate attention.

The optical names trade at fifty to seventy times earnings on the same AI demand narratives. The memory without which no AI accelerator functions trades at a fraction. Its starting point has become a liability in its move towards a rating that is easily attainable for everyone else.

In concrete terms, the analyst who has watched the stock rise sets a target twenty to thirty percent above current levels, which is simultaneously bold by momentum standards and trivial by fundamental ones. Nobody attempts to price the business at a rational multiple on a comparative or more fundamental DCF basis. Nobody attempts to price a premium for the criticality of the product in the AI supply chain, the companies’ innovation abilities, or their management’s extraordinary business decisions. The momentum behavior is observed. The fundamental reality is acknowledged. The synthesis is avoided. Rather, curious justifications are proffered and become acceptable answers to justify the perpetuation of today’s weirdness.

IV. The Story That Nobody Told

When a story lacks readily available narratives and well-regarded narrators, it suffers a peculiar disinterest that has nothing to do with the underlying facts. This is the Boring Supremacy Blindspot: entities that actually constrain or enable the frontier are cognitively filed as commodities or infrastructure, with no one credible to bring about a change. One only has to look at the excitement around far slower-growing, far more “commoditizable” optical subsegments, and dozens of companies in the space, to realize how the most profitable layer of the AI stack is also the least discussed beyond its financial numbers.

The more seamlessly integrated and foundational a technology becomes, the less its individual economic significance is automatically recognized. Indispensable elements become cognitively invisible. The brilliant engineer who cannot articulate the implications of their work is consistently passed over for a competent manager who can. The competence gap matters less than the communication gap. Markets operate on the same principle.

In the early years of TSMC's ascent, Morris Chang's sustained narration turned a radical departure from the integrated device manufacturer paradigm into an understood strategic asset. The foundry model needed its narrator. It had one. The valuation reflects, in part, decades of accumulated interpretive work. Indian equity markets offer the same observation in a different key: multiple companies have commanded premiums because their promoters and proponents delivered visions with conviction and continuity over long periods. This week, we had the Amazon leader who showed the power of annual letters. The stories do not always come from business leaders. Analysts and market veterans with a reputation can play a huge role, although never through the analysis of PE charts or next quarter EPS forecasts. When a segment lacks voices that matter, and few can compete with Korean memory makers on this dimension as well, a lot may never be picked up by the market.

The analyst community must share the blame until it gathers the conviction to move beyond the most rudimentary financial analysis. 

When Nvidia announces a new architecture, an entire ecosystem activates: bandwidth per watt breakdowns, developer community commentary, and platform economics frameworks. Memory coverage produces the same recycled diagrams on HBM3E yields. Stacking twelve or sixteen layers of DRAM using through-silicon via technology, achieving commercial yields on a product whose base die now requires foundry-level logic processes, doing this at the volume demanded by accelerator manufacturers who have no near-term alternative supplier: this is, by any measure, a remarkable manufacturing achievement. The time when the narrator community around the segment begins to produce the first readable dozens of page reports on these companies’ achievements or synthesis on why no one else globally could match what they do for long periods, like the case is for TSMC or NVIDIA, is the time when they will find the need to discard the current Price-to-book charts going back two decades.

In narratives at least, the Korean address carries a permanent location-based risk premium even when earnings are world-class. The numbers are global. The discount remains local.

The Familiarity Discount operates in the same direction. Markets undervalue consistent excellence from geographically distant or historically less glamorous sources. The same OLED panel manufactured in Korea is valued differently by markets depending on whose brand name appears on the product. The substrate is identical. The mental wrapper changes everything.

The memory complex, even decades later, is at the stage TSMC was before the language existed to describe what it was building. The numbers are speaking. The framework has not yet been built. 

What the Professor Will Say

Imagine a future MBA class. A professor opens with a simple slide. One quarter's operating profit exceeded the prior year's entire result. The valuation at which the stock traded that week. The aggregate profits of the Korean memory complex set against the earnings of several national equity markets. The students are asked to justify the combination.

They will try. Armed with better tools than today's analysts, they will deploy sophisticated arguments that will likely include the same conclusions: cyclicality, geopolitical risk, governance discount, structural limitations of component manufacturing. Some of these arguments will be refinements of arguments being made today. Some will be new ones developed with the benefit of having watched what happened next.

Then the professor shows the second slide. The same stocks three years later.

Today's prices may be the peaks. We may be expecting stocks to move in a specific direction, but we are fully aware that the exact opposite may happen. The behavioral tendencies described in this essay cut in all directions. The same mechanisms that suppress multiples today can persist, and with adverse newsflows, lead to undershoots in the other direction. The point of this note is simple: as much as we must accept today’s reality as given, it is futile to look for the reasons behind its existence through any rational, efficient market lens.

The behavioral arguments can provide a better explanatory cover, perhaps, but no perfect guide on what may happen tomorrow. That said, they could do a better job than arbitrarily slapped, let’s say, six times PE multiple on where these stocks may go. 

About that set of extreme numbers that surround the two companies, one of their biggest utility is the conversation points they provide to writers like us today and professors looking for case studies tomorrow.