The most important question facing investors in Samsung Electronics and SK hynix may no longer be which company has the lower P/E ratio.
It is whether today's extraordinary earnings represent the peak of another memory cycle — or the beginning of a structurally higher level of profitability.
That distinction changes almost everything.
If the current earnings surge is primarily the result of temporary memory shortages and extreme pricing, today's apparently low valuation multiples could prove deceptive.
But if AI, HBM and increasingly differentiated memory products have raised the sustainable return on capital of Korean memory producers, the old valuation framework may also be obsolete.
As discussed in Part II of this series, the memory industry has always created a paradox for investors: earnings often look strongest — and P/E ratios lowest — near the top of the cycle. https://www.geomarketsignal.com/2026/08/part-ii-memory-cycle-is-not-dead-why-pe.html
This time, however, there is another layer.
AI may be raising the floor of the cycle even if it does not eliminate the cycle itself.
The challenge is separating that structural improvement from today's shortage-driven profits.
And the best place to start is not valuation.
It is supply.
Record Revenue Does Not Mean Record Physical Demand
Korea's semiconductor export numbers are extraordinary.
In the first half of 2026, semiconductor exports reached $192.4 billion, up 162.6% year-on-year. In June alone, semiconductor exports increased 199.5% to $44.82 billion.
At first glance, those numbers could suggest an unprecedented explosion in semiconductor demand.
But Korea's Ministry of Trade, Industry and Resources explicitly identified higher memory contract prices as a major reason for the increase.
That distinction matters.
Export value is essentially:
Volume × Price × Product Mix
A 160% increase in export value does not mean Korea shipped 160% more memory bits.
In fact, available industry data suggest that physical bit growth has been dramatically smaller than revenue growth.https://english.motir.go.kr/eng/article/EATCLdfa319ada/2677/view
Micron provides one of the clearest demonstrations.
In its fiscal third quarter of 2026, Micron's DRAM revenue increased 67% quarter-on-quarter.
But DRAM bit shipments increased only in the low-single-digit percentage range.
The dominant variable was price: DRAM average selling prices increased in the low-60% range.
NAND showed the same pattern.
Revenue rose 99% quarter-on-quarter, while bit shipments increased only in the mid-single-digit percentage range. NAND ASP increased in the mid-80% range.
Revenue exploded. Bits did not. Price did most of the work.
This does not mean physical demand is weak. Samsung itself reported record DRAM and NAND bit sales in the second quarter.
It means something more precise:
Bit growth has been nowhere near large enough to explain the magnitude of the revenue and profit explosion.
Pricing and product mix have done much of the heavy lifting. https://s25.q4cdn.com/621799436/files/doc_financials/2026/q3/Micron_Q3_26_Earnings_Deck.pdf
Samsung and SK Hynix Are Printing Extraordinary Profits
The resulting profitability is remarkable.
Samsung Electronics reported second-quarter 2026 consolidated revenue of KRW 171.5 trillion and operating profit of KRW 89.5 trillion.
Its Device Solutions division generated KRW 127.5 trillion in revenue and KRW 89.2 trillion in operating profit.
Memory revenue alone reached KRW 120.8 trillion.
Samsung also reported that its company-wide ROE reached 56% in the quarter, compared with 5% in the same quarter a year earlier.
Samsung's own presentation described the environment clearly: record memory earnings came amid limited capacity, an industry-wide upward price trend and continued supply constraints.https://images.samsung.com/is/content/samsung/assets/global/ir/docs/2026_2Q_conference_eng.pdf
SK hynix's numbers are equally striking.
Second-quarter revenue reached KRW 79.32 trillion, while operating profit reached KRW 60.54 trillion.
That implies an operating margin of approximately 76%.
A year earlier, SK hynix had generated KRW 22.23 trillion in quarterly revenue and KRW 9.21 trillion in operating profit.
The company attributed the improvement to strong AI demand, higher-value products and significant price increases across both DRAM and NAND.
It also began mass shipments of HBM4 during the quarter.https://news.skhynix.com/en/q2-2026-business-results/
There is little doubt that AI is producing real economic value for the Korean memory industry.
But investors should ask a harder question:
How much of this profitability survives when memory prices normalize?
Has the Memory Supply Glut Already Begun?
This may be the most important question in the entire valuation debate.
The current evidence says:
Not yet.
Samsung expects accelerating demand for server DRAM, enterprise SSDs and HBM to keep the memory market undersupplied despite weaker conditions in some PC and mobile markets.
The company also says supply constraints should continue despite efforts to raise production.
SK hynix uses similarly explicit language.
It describes the current environment as one in which customer demand exceeds supply capabilities, while accelerating production at its M15X facility and preparing additional capacity.
Micron goes even further.
It expects industry DRAM bit shipments to increase in the low-to-mid 20% range in calendar 2026, yet still expects DRAM and NAND supply-demand conditions to remain tight beyond 2027.
Micron also says HBM growth is putting additional pressure on non-HBM supply because newer generations consume increasing manufacturing resources.https://s25.q4cdn.com/621799436/files/doc_financials/2026/q3/Micron_Q3_26_Earnings_Deck.pdf
The current market therefore looks less like an oversupply cycle and more like a shortage economy.
But that does not mean investors can ignore supply.
It means the supply risk has moved one step further into the future.
The Current Shortage Did Not Appear by Accident
To understand today's extraordinary prices, investors need to go back to the collapse of 2022 and 2023.
After memory demand deteriorated, the major suppliers reacted aggressively.
Micron announced in November 2022 that it would reduce DRAM and NAND wafer starts by approximately 20% and expected its calendar-2023 DRAM bit supply to decline year-on-year.https://investors.micron.com/news/press-release/2022/Micron-Announces-Further-Actions-to-Address-Market-Conditions-11-16-2022/default.aspx
SK hynix announced that it would reduce 2023 investment by more than 50% year-on-year while gradually reducing production of relatively less profitable products.https://news.skhynix.com/en/sk-hynix-reports-third-quarter-2022-results/
Samsung later said it was reducing inventory through production adjustments rather than aggressive sales expansion.
In the third quarter of 2023, Samsung's memory bit growth came in below guidance while DRAM and NAND ASPs increased from the previous quarter.https://news.samsung.com/global/samsung-electronics-announces-third-quarter-2023-results
The point is not that the three companies acted together.
There is no need to make that claim.
The economically relevant fact is that after the 2022–23 collapse, the industry's major producers independently adopted unusually disciplined supply and capital-allocation strategies.
The sequence was powerful:
Production reductions
→ Slower bit growth
→ Inventory normalization
→ AI demand shock
→ HBM capacity absorption
→ Memory shortage
→ Extraordinary pricing
That sequence helps explain why memory prices have risen so dramatically.
AI did not hit an industry operating at unlimited capacity.
It hit an industry that had just spent years repairing a massive supply-demand imbalance.
HBM Makes This Cycle Different
This is where the current cycle begins to diverge from previous ones.
HBM is not simply another higher-priced DRAM product.
It requires advanced DRAM dies, stacking, packaging, thermal management, customer qualification and close integration with AI accelerator roadmaps.
And as HBM volumes rise, they consume manufacturing resources that could otherwise support conventional memory.
Micron explicitly says that the increasing HBM trade ratio puts additional pressure on non-HBM supply.
Samsung is expanding HBM4 production while continuing to focus capacity on high-value server products.
SK hynix has begun HBM4 mass shipments and is increasingly entering multi-year supply arrangements with major customers. As of its second-quarter announcement, the company said it had finalized long-term agreements with around ten customers.https://news.skhynix.com/en/q2-2026-business-results/
This matters because memory may be moving away from a purely spot-driven commodity model.
Not completely. But partially.
The more capacity becomes tied to technically differentiated products, customer qualification and long-term supply agreements, the less meaningful it becomes to value every dollar of memory earnings as if it came from undifferentiated commodity DRAM.
That does not eliminate cyclicality.
It stratifies it.
But the Next Supply Response Has Already Started
There is an uncomfortable feature of every semiconductor boom.
High prices eventually create their own competition.
Record profitability produces record incentives to build capacity.
And that process is now underway.
SK hynix announced in August 2026 that it would invest approximately KRW 54 trillion in new Yongin and Cheongju facilities.
KRW 35.2 trillion is planned for Yongin Y2 and KRW 19.1 trillion for Cheongju M17.
M17's first cleanroom is targeted for December 2028, while the first Y2 cleanroom is targeted for June 2029.
The company has also been accelerating near-term production through M15X and its first Yongin facility.
Micron is expanding as well.
Its first new Idaho DRAM fab is scheduled for initial wafer output in mid-2027, while another Idaho fab is planned for late 2028.
Additional Taiwan and Singapore capacity is also scheduled to contribute from 2027.
This does not mean oversupply is imminent. Building semiconductor capacity takes years. But it does mean the industry's cycle mechanism remains alive.
Today's shortage is producing tomorrow's capacity.
That leads to perhaps the most important distinction for investors: The evidence does not show that a memory supply glut has already begun. It shows that the supply-expansion cycle has begun.
Those are very different things.
The Valuation Risk Is Tomorrow's Bit Growth
Investors often wait for visible inventory problems before declaring that a memory cycle has turned. That may be too late. The stock market normally anticipates changes in earnings.
A more useful warning signal may therefore be the relationship between ASP growth and bit growth.
If physical supply begins accelerating while ASP increases slow, the economic balance is beginning to change even before inventories visibly pile up.
The sequence investors should worry about is:
Bit supply accelerates
→ ASP growth slows
→ Pricing peaks
→ Earnings revisions peak
→ Inventory begins accumulating
→ Operating leverage reverses
By the time the final stage becomes obvious, equity markets may already have repriced the stocks.
That is why the question is not simply: Is there an oversupply today?
The better question is: How far away are we from the point where tomorrow's bit growth catches up with today's prices?
This Changes How We Should Read Today's P/E
Record earnings make P/E ratios fall mechanically.
That can create the illusion that a stock is becoming cheaper even as its cyclical risk increases.
Suppose a memory producer's share price rises 50%, but earnings double. Its P/E falls.
The stock looks statistically cheaper. But if a significant portion of those earnings comes from temporary shortage pricing, the denominator is doing most of the work.
This is precisely why conventional P/E analysis has repeatedly failed investors near memory-cycle peaks.
The current cycle requires another adjustment.
We need to separate today's earnings into two conceptual layers:
Structural AI earnings and Cyclical shortage earnings.
The first deserves a potentially higher valuation. The second does not.
P/B Becomes More Important — But Only Through ROE
This brings us back to P/B.
A memory stock does not deserve a higher P/B merely because AI has become fashionable.
Higher book multiples are justified only if the business can sustainably generate higher returns on that book value.
A simplified justified P/B relationship is:
P/B = (ROE − g) / (r − g)
where ROE is sustainable return on equity, g is long-term growth and r is the cost of equity.
For illustration, assume a 10% cost of equity and 3% long-term growth rate.
These are GeoMarketSignal valuation assumptions, not forecasts.
The relationship looks like this:
10% sustainable ROE → approximately 1.00x justified P/B
15% sustainable ROE → approximately 1.71x
20% sustainable ROE → approximately 2.43x
25% sustainable ROE → approximately 3.14x
30% sustainable ROE → approximately 3.86x
35% sustainable ROE → approximately 4.57x
40% sustainable ROE → approximately 5.29x
The implication is powerful.
AI does not have to eliminate the memory cycle for Korean memory stocks to deserve higher P/B multiples. It only has to raise their through-cycle ROE.
That is a much lower hurdle — and a much more defensible investment thesis.
Samsung and SK Hynix Should Not Have the Same Equilibrium
This is where the two companies diverge.
SK hynix is the more direct expression of the AI-memory thesis.
Its 2025 HBM revenue more than doubled year-on-year, and the company generated annual revenue of KRW 97.15 trillion and operating profit of KRW 47.21 trillion before profitability accelerated even further in 2026.
Its HBM exposure, AI-server DRAM positioning, increasingly long-term customer relationships and HBM4 execution make a larger portion of its earnings potentially structural.
But that advantage cuts both ways.
SK hynix is also far more directly exposed to memory pricing. If AI-memory demand disappoints or conventional DRAM prices collapse, there are fewer unrelated businesses to absorb the shock.
Samsung Electronics has the opposite structure. Memory has become the dominant driver of current profits, but Samsung is still a diversified technology company spanning smartphones, displays, foundry, system semiconductors, consumer electronics and other businesses.
That diversification can dilute the impact of a memory rerating. But it can also reduce downside dependence on a single memory cycle.
Samsung therefore presents a different valuation question.
Its potential rerating depends not only on memory prices but on whether its HBM4 ramp, advanced-node execution and broader semiconductor ecosystem can raise the sustainable profitability of the entire company.
This is why applying the same P/B multiple to Samsung and SK hynix makes little sense.
Their sustainable ROE structures are different.
A Better Way to Think About the Two Stocks
Rather than asking which stock currently has the lower P/E or P/B, investors should think in terms of three earnings regimes.
Peak-Price Regime
Shortage + extreme ASP + strong AI mix
The key valuation lesson: Do not annualize today's earnings blindly.
Normalized AI Era
ASP normalizes, but the AI/HBM premium survives.
This should be the core valuation case.
True Oversupply
Bit supply exceeds demand and ASP falls sharply.
This is the downside stress case.
The second regime is the one that matters most.
If commodity DRAM prices normalize but HBM economics, advanced server memory, packaging differentiation and long-term customer relationships remain intact, profitability does not necessarily have to return to the old cycle averages.
That is the central argument for a new valuation equilibrium.
AI does not need to prevent memory prices from falling.
It only needs to leave profitability structurally higher after they fall.
That is a fundamentally different proposition.
What Would Prove the Thesis Wrong?
A new valuation regime should never be treated as an article of faith.
It should be tested continuously.
If industry bit growth accelerates materially faster than end-demand growth, ASP increases reverse, customer inventories start rising, HBM price premiums compress, capacity additions arrive faster than expected or Chinese suppliers capture significantly more conventional-memory demand, the normalized-ROE thesis must be revised downward.
Conversely, if HBM and advanced server memory retain strong pricing, long-term supply agreements become more widespread, AI memory continues consuming disproportionate manufacturing resources and conventional-memory supply remains constrained, then historical P/B ranges may increasingly lose relevance.
The market will ultimately decide this question through ROE.
The New P/E–P/B Equilibrium
The memory cycle is not dead. The extraordinary 2026 earnings figures themselves may contain significant cyclical scarcity profits.
Korea's export data and Micron's bit-shipment data strongly suggest that prices have contributed far more to the current revenue explosion than physical volume alone.
Meanwhile, the major memory producers are already responding to today's profitability with new investment.
That is classic cyclicality.
But something else has changed.
HBM, advanced server DRAM, sophisticated packaging and deeper customer integration are creating a layer of memory whose economics increasingly differ from traditional commodity DRAM.
The investment question is therefore no longer whether the old memory cycle survives.
The question is: How much of each company's earnings remains trapped inside that old cycle?
For Samsung Electronics, the answer depends on whether its renewed memory competitiveness can raise the sustainable ROE of a much broader company.
For SK hynix, the answer depends on how much of today's extraordinary AI-memory profitability survives once conventional memory prices normalize.
That is why the two companies may deserve different valuation equilibria.
And it is why neither P/E nor P/B should be read in isolation.
The real variable connecting them is sustainable ROE.
The next memory downturn will therefore be more than another test of semiconductor demand.
It will be the first real test of whether AI has permanently raised the profitability floor of Korean memory.
If it has, the valuation regime has changed.
If it has not, today's low P/E ratios may prove to be exactly what they have often been at the top of previous memory cycles:
A warning disguised as cheapness.
The AI Memory Cycle & Korean Semiconductors
This article is Part IV of the GeoMarketSignal AI Memory Cycle series.
← Previous: Part III — AI, China and the New Korean Semiconductor Order