Yesterday was a red light, today is a blue light again... The stock market is truly fluctuating wildly. They say the semiconductor supercycle will continue, and backed by this, the economic growth rate has been significantly revised upward to 3.0%... But why are semiconductor stocks not what they used to be, and instead "stalling"? I was curious about how a semiconductor expert, rather than an economic expert, views the current situation.

SBS YouTube's "Discovery of Knowledge" met with Professor Kwon Seok-jun of the Department of Chemical Engineering at Sungkyunkwan University. We discussed various semiconductor-related topics, ranging from the government's "Three Mega Projects" to Chinese semiconductors, which pose the greatest threat to South Korea's semiconductor industry. Among these, we have compiled only the content related to stock prices. Professor Kwon Seok-jun's perspective—that a hegemony battle between "tech" and "finance" is currently taking place in the AI semiconductor market—is highly intriguing. We also recommend watching the interview in video format. This interview was conducted on July 13.

Q. There are talks that the "semiconductor supercycle" has passed its peak. What is your assessment, Professor?

Professor Kwon Seok-jun: I do not believe the supercycle has peaked. I expect the supercycle to continue for a few more years. However, I think signals for a psychological peak, to some extent, are starting to appear. Therefore, when we talk about a peak, we need to distinguish between a fundamental peak, a demand peak, and a psychological peak. Currently, there is no clear distinction among these peaks, and they appear somewhat mixed.

In times like this, I believe we need to look cool-headedly at the data. Usually, when analyzing this, we look at the gap between supply and demand. However, rather than simply looking at the absolute scale of the difference, we calculate the ratio of supply to demand, assuming a supply of 100. We call this ratio the "sufficiency ratio." Currently, we see the sufficiency ratio hitting its lowest point in the second and third quarters of 2026, at around -8.2% to -8.5%. While this will likely ease gradually heading into the third or fourth quarter, it will still remain negative.

Since the ratio is expected to turn positive around the first half of 2028, the current phenomenon of skyrocketing memory prices caused by supply shortages will not suddenly die down. However, separate from memory manufacturers improving their profitability or posting record earnings, investor sentiment is less focused on how much memory these companies sell and more on how long their biggest buyers—namely, U.S. hyperscalers—can continue purchasing memory. This anxiety is being reflected preemptively, which is likely why we are seeing repeated market corrections recently.

Q. So we can understand it as currently heading toward the peak.

Professor Kwon Seok-jun: If we look at the current trend, a peak will eventually come. Common sense dictates that DRAM prices skyrocketing four to five times, as we have seen over the past six months, cannot repeat forever. Since the curve of memory suppliers like Samsung Electronics, SK Hynix, and Micron expanding their existing fabs and even building entirely new fabs to supply the market is within a somewhat predictable range, when we discuss when the peak will arrive, we can predict within a reliable range when supply and demand will reach equilibrium, rather than predicting a peak in terms of price.

Reporter Jeong Yu-mi: So you are saying that the sufficiency ratio could turn positive around the first half of 2028.

Professor Kwon Seok-jun: Yes, but that is only when looking at a very simple curve. The memory semiconductor market we are seeing today is different from the dynamics shown by the commodity memory semiconductor market in the past. What we can note here is that while demand is exploding, suppliers can also artificially generate demand. This might be hard to understand, as some might think supply is supply and demand is demand.

In fact, when AI semiconductors, particularly AI data centers, create such massive volatility in the market, one thing to consider is the dynamics between HBM (High Bandwidth Memory) and DRAM. For instance, if people think, "If AI data centers need much more HBM, we can just make more HBM," the reality is that making more HBM reduces the supply of DRAM. This is because making HBM requires vertically stacking DRAM. For example, making one HBM wafer requires three to six DRAM wafers. Therefore, producing more HBM and increasing HBM supply signals can mean a reduction in DRAM supply from a broader market perspective. To prepare for a decrease in DRAM supply, companies that need to purchase it in bulk will suddenly pull forward their DRAM demand. This can be seen as a complex dynamic emerging as the share of memory semiconductors, especially AI-specific memory semiconductors, grows—something we rarely saw in the past.

Q. It seems people are paying more attention to the "semiconductor peak-out" narrative because of stock prices. Samsung Electronics and SK Hynix have fallen significantly from their peaks, and even as we record this, stock prices are dropping. They say stock prices preemptively reflect future value, so looking at this market, people think we have almost reached the peak, or that we are at the peak now and there is nowhere to go but down.

Professor Kwon Seok-jun: There will certainly be a point where supply and demand match, and much of what we call preemptive reflection—which leads investor sentiment by about two to three quarters—is reflected here. One of the reasons stock prices have dropped significantly recently is not due to individual stocks, but rather, in my view, because the weight of ETFs has grown too much, especially with many leveraging multiple-tracking products, which has introduced a lot of psychological factors.

However, when discussing whether we have reached a peak or not, I believe the capital expenditure (CAPEX) that AI data centers—or more fundamentally, hyperscalers—can continuously generate is far more important than the peak of memory manufacturers. We need to look at this from a more macroeconomic perspective.

This is not unique to specific AI semiconductor industries like AI data centers or AI hyperscaler semiconductor companies. Historically, whenever an industry innovates and a new market is created, there has always been a power game between the tech camp and the traditional financial camp. Financial institutions have continuously tried to demonstrate their power over stock prices by writing reports based on valuations and stock price guidance, taking short and long positions, and liquidating positions.

On the other hand, the tech industry aims to accumulate enough profits so they can raise capital, create new markets, and shift the power of wealth back to the tech side without going through the financial filter of Wall Street. I believe the battlefield for this hegemony struggle, which we have witnessed for about a century, is now shifting to AI semiconductors. Therefore, if you think about who is shouting "peak-out," it is closer to the financial side rather than the tech side.

In fact, it is not inherently bad for these financial institutions to issue such outlooks (regarding the semiconductor peak-out). It is understandable, as they need to protect investor interests and provide at least a minimum level of objective guidance. However, this is a market that humanity has never experienced before, and the memory semiconductors required by AI data centers exhibit dynamics entirely different from traditional B2B memory. In a situation where past references cannot be sufficiently extrapolated, making bold forecasts about the market or its peak carries the risk of reaching somewhat hasty conclusions.

Q. A recent statement by Morgan Stanley also caught people's attention: "Underweight semiconductor stocks." If we ask you, a semiconductor expert, about this, what would you say?

Professor Kwon Seok-jun: Since I am not a stock or economic expert, it would be best to take my words as just a reference opinion. For instance, just two years ago in 2024, Morgan Stanley claimed that the "HBM market is overheated" and "is a bubble." I recall Morgan Stanley estimating that a stock price of around 120,000 won was sufficient for Hynix. Back then, those who took positions believing that Hynix's stock price would indeed fall and that the HBM market would not grow much must be feeling very regretful now. So, in truth, Morgan Stanley can be wrong sometimes and right at other times.

I view Morgan Stanley's recent advice to adjust stock weightings in AI-related industries, including semiconductors, as a plausible safety measure. However, I find it hard to interpret this as a call to liquidate all current positions. I think it is more appropriate to understand it as: "Since it will be difficult for investors to maintain a strong psychological stance in this environment, those who lack confidence should rather reduce their exposure." Nevertheless, because I believe there is a high probability that memory demand—specifically the demand triggered by AI data centers—will remain robust, those who have held positions for a long time do not necessarily need to adjust their portfolios now. If they must make adjustments, I believe it is right to look at other related bottlenecks.

Reporter Jeong Yu-mi: Could you explain more about looking at the bottlenecks?

Professor Kwon Seok-jun: What is being discussed in the so-called "Three Mega Projects" are semiconductor mega fabs, AI data centers, and physical AI. However, all three of these must run on top of a basic, massive infrastructure. Naturally, as many people know, electricity and water can be very important infrastructure elements, and power infrastructure will ultimately become the bottleneck in this speed race.

For example, various power semiconductors and power supply units (PSUs) used in generators or substations cannot be manufactured instantly just because we want to make them suddenly. It will be difficult to match the timing. Instead, demand in these areas is bound to explode, and those facing bottlenecks will ultimately hold the power to control supply. Since a second cycle, similar to the memory supercycle we are currently witnessing, could also emerge in this infrastructure, for those who wish to diversify their portfolios and prepare for market volatility risks, adjusting portfolios toward related industries that are not completely independent but have somewhat separated bottlenecks to prepare for potential future risks would be a good direction.

Reporter Jeong Yu-mi: Hearing you speak, it sounds like you would be very good at stock investing.

Professor Kwon Seok-jun: No, I am not that good at it myself.

Q. You mentioned that the hegemony struggle between the financial and tech sectors has always existed. Could you explain that further?

Professor Kwon Seok-jun: In both the AI industry and the financial sector, AI refers to AI plus semiconductors, while the financial sector refers to Wall Street. One interesting thing we should observe here is that this is turning into a game of who gets to determine the discount rate for Strong AI, a technology that has not yet been realized. Since it is in the future, we must apply a discount rate to bring its value to the present. There are already formulas for discount rates. There are textbook formulas that determine how much weight to assign and how to price these options. And most of these formulas were created by Wall Street financiers and economists.

However, the tech industry's view is that it will be difficult for the financial sector to fully evaluate the actual impact of what value AGI (Artificial General Intelligence) will actually hold. For example, if AGI, or Strong AI, truly emerges, the tech industry believes it will not just transform the IT industry but overturn and transform every industry humanity has known. If so, most of the profit cycles that can currently be calculated on accounting ledgers can be viewed as markets that will be formed in the future. This leads to the argument that they can change the discount rate to a concept they desire.

But Wall Street is sending signals saying, "AGI might be possible, but you don't get to decide how much value it actually holds; we, who know the industries because we have been reading them, will decide. And if you want to borrow money for this, you have to come to us. If you don't play by our power game regarding this valuation, we will call you a bubble." This power game is not just someone else's problem; it is becoming one of the core arguments for our country's memory semiconductor industry and especially our "Mega Projects." Thus, it is by no means something we can just sit back and watch.

Q. I was actually going to ask if we should keep the possibility of an AI bubble in mind, but hearing you speak, should we view the term "AI bubble" itself as ultimately originating from Wall Street?

Professor Kwon Seok-jun: Of course, it is also risky to conclude that this is solely a biased perspective from Wall Street. Those who know technology well also constantly warn about an AI bubble. I have written such pieces a few times myself. If an AI bubble were to occur, looking at it from a narrow perspective, I believe there is a bubble in the valuations of certain companies, particularly hyperscalers, and OpenAI, though it is not listed yet.

For those who talk about a bubble from a broader perspective, it has nothing to do with Wall Street. They question whether AI, while admittedly important, is truly a game-changing innovation capable of causing such a tectonic shift, whether it will really change all industries, and whether there is a roadmap within visible reach for the concept of AGI to be realized right now.

For people like me who look further ahead in semiconductors, if semiconductors themselves hit technical limits or run into physical limitations around the late 2030s or early 2040s, what will we do when computation is needed? Will we pay a much higher price to use semiconductors that are far more expensive and energy-consuming, or will something new emerge to overcome this? In a situation where no one can predict such circumstances, continuing to say that this industry will grow and that it is sustainable can also trigger warnings of risks or crises regarding a bubble.

Therefore, the side focusing solely on valuation can be seen as the bubble crisis theory coming from the financial sector. On the other hand, arguments that the sustainability of the technology itself or the industrial impact of AGI is overblown can be viewed as a bubble crisis theory from a more general perspective.

Q. In this hegemony struggle, since the field of AI is on a completely different dimension compared to past technological advancements, won't the tech side inevitably hold the hegemony?

Professor Kwon Seok-jun: "Inevitably hold" is a limiting expression that scientists are very reluctant to use. In fact, while that is a possibility, Wall Street has dominated the financial market for over a century, and the logic of capital procurement, capital expansion, and capital recovery... writing numbers on the accounting ledger is still entirely decided by Wall Street.

Furthermore, it is not as if these tech companies can avoid issuing shares. Even if they have secured operating cash flow, their "barns," metaphorically speaking, are currently running dry. Even the mighty Google is now issuing corporate bonds and utilizing SPVs (Special Purpose Vehicles). In the case of OpenAI, they are even bringing in capital at a high interest rate of 17.5% by putting up revenue sharing as collateral. Therefore, if they continue to execute hundreds of billions of dollars in CAPEX annually, the one quietly smiling in the background will be Wall Street.

In a way, they are pushing their homework down the road. In other words, Wall Street is just waiting for the day to grade that homework. They will want to inspect, on an annual and quarterly basis, whether this is actually reflected on the accounting ledgers, whether this valuation is correct, whether the cash flow is actually being generated, and whether the "physical AI" you keep talking about is truly creating new profits, lowering costs, or saving labor costs in the real world and manufacturing—beyond just robotics and mobility—to build an economy that can remain stable while creating a new supply chain. Tech companies, on the other hand, will likely try to hold out by arguing that this scale is near-infinite and difficult for Wall Street to measure; since the numerator of the scale has grown so large, no matter how much grading of the denominator is done, we cannot view this market itself as a "bubble."