Korea Semiconductors on Korea Invest Insights

The US-China Split in Agentic Inference Infrastructure and the SRAM Opportunity

Thu, 09 Jul 2026 17:20:00 +0900

Related reading: NVIDIA Vera Rubin inference stack / AI token value and memory value capture / AI data center power bottleneck map

TL;DR

The United States and China are both scaling AI inference, but they are solving different constraints. The US bottleneck is power, rack density, HBM, advanced packaging, and token-per-watt efficiency. China’s bottleneck is access to leading-edge logic and HBM, so it is more likely to push Ascend-based clusters, optical mesh, parallel scaling, and domestic memory hierarchy workarounds.

For Korean listed equities, the investable landing zone is not China optical modules by default. The cleaner landing is the US-style AI factory bottleneck: HBM4/HBM4E, power equipment, HBM packaging equipment, and a less appreciated Samsung Electronics option around SRAM/LPU foundry, AI storage, and memory hierarchy.

The business winner and the best stock setup are not the same thing. SK hynix has the cleanest HBM4 leadership but is already a crowded winner. HD Hyundai Electric has high purity to US data center power bottlenecks, but order momentum is widely recognized. Hanmi Semiconductor is a real HBM tool bottleneck but needs repeat orders and customer diversification. The most asymmetric candidate is Samsung Electronics, if HBM4E, SRAM/LPU foundry, and AI storage start to show up in numbers.

1. What Is Being Verified

This note starts from YS-VC’s article “Power or Chips: Why US and Chinese AI Inference Stacks Are Diverging”. The core argument is directionally right: inference scaling is no longer a generic GPU story, and the US and China face different binding constraints.

The adjustment is in the Korean equity landing. A China optical-mesh story does not automatically translate into Korean optical component revenue. China’s AI infrastructure supply chain is strategically closed and increasingly domestic. Korean equity exposure is more direct through components sold into the US-style AI infrastructure bottleneck.

Claim	Assessment	Investor read-through
US and China AI inference stacks are diverging	Mostly true	The US optimizes power and rack efficiency; China works around logic and HBM limits
The US stack is moving toward GPU/HBM/SRAM rack-scale inference	True	Vera Rubin, LPX, HBM4, and SRAM/LPU become one system
China is using Ascend, optical mesh, and memory hierarchy workarounds	Partly true	Directionally plausible, but CloudMatrix performance and reliability need independent evidence
US export controls limit China’s NVIDIA data-center compute access	True	China cannot count on steady access to top-end GPUs and HBM
HBM is still a key control point	True	BIS has explicitly treated HBM as important for AI training and inference at scale
Korean opportunity equals China optical modules	Too aggressive	The cleaner Korean landing is HBM, power, packaging, and Samsung foundry optionality

2. Why The Stacks Are Diverging

Agentic AI increases token volume. Coding agents, research agents, voice agents, and enterprise workflow agents do not just answer once. They read long context, call tools, interpret tool output, and generate new tokens repeatedly. That makes prefill, decode, KV cache, storage, network, and power all matter.

The bottleneck differs by country.

Area	United States	China
Binding constraint	Power, rack density, transformers, HBM4, advanced packaging	Leading-edge logic, HBM access, export controls, software optimization
Architecture direction	Vera Rubin, LPX/LPU, HBM4, high-power AI racks	Huawei Ascend, optical mesh, parallel scaling, domestic memory hierarchy
Strength	Access to the best GPUs, HBM, and packaging ecosystem	Faster power buildout, state-led infrastructure, domestic cloud demand
Weakness	Time-to-power, grid interconnection, transformer lead times	No EUV-based leading logic, HBM limits, closed ecosystem
Korean equity landing	HBM, power equipment, HBM tools, foundry optionality	Limited unless direct suppliers are verified

The IEA expects data-center electricity demand to reach about 945 TWh by 2030. ([IEA][1]) Power is no longer background infrastructure. It is becoming an upper bound on AI capacity.

BloombergNEF data cited by Energy Connects says China may add more than 3.4 TW of generation capacity over five years, almost six times the US. The same article says data centers could account for 38% of US power demand growth from 2024 to 2030, versus 6% in China. ([Energy Connects][2]) This is why the US is forced to care about token throughput per MW, while China can lean more heavily on parallel infrastructure buildout.

3. The US Stack: HBM Plus SRAM/LPU

The key US signal is NVIDIA’s Vera Rubin and LPX architecture. NVIDIA describes LPX as an inference accelerator for Vera Rubin. It combines Rubin GPUs using HBM with Groq 3 LPX using SRAM-based LPUs. ([NVIDIA LPX][3])

Metric	NVIDIA LPX disclosure
Agentic token load	Up to 15x more tokens
Vera Rubin NVL72 + LPX	Up to 35x higher throughput per MW
SRAM per LPU accelerator	500MB
SRAM bandwidth per LPU accelerator	150TB/s
Scale-up bandwidth per LPU accelerator	2.5TB/s
LPX rack	256 LPU chips
SRAM per LPX rack	128GB
DDR5 per LPX rack	12TB
SRAM bandwidth per LPX rack	40PB/s

This is not an HBM replacement story. HBM remains central to the Rubin GPU. SRAM/LPU complements HBM by handling latency-sensitive decode work closer to compute.

Inference stage	Simple description	Main bottleneck
Prefill	Read the prompt and documents first	GPU compute, HBM bandwidth
Decode	Generate answers token by token	Latency, SRAM, LPU
KV cache	Reuse previous context	HBM, DDR, SSD, cache hierarchy
Serving	Handle many users reliably	Rack network, scheduler, power

This creates an underappreciated Samsung Electronics angle. Samsung is not only an HBM catch-up candidate. At GTC 2026, Samsung highlighted HBM4 mass production, HBM4E, SOCAMM2 mass production, PCIe 6.0 SSDs, and broader AI computing solutions. ([Samsung Newsroom][4]) If inference moves toward HBM plus SRAM/LPU plus storage hierarchy, Samsung’s combined memory, foundry, packaging, and storage position becomes more relevant.

4. China: Optical Mesh Is A Competitive Signal, Not A Clean Korean Equity Trade

China’s workaround is logical. If access to the latest GPUs and HBM is constrained, the system must rely on more chips, better interconnect, optical mesh, and domestic software optimization. But that is not the same as a direct Korean revenue bridge.

China’s AI supply chain is strategically local. US export controls also push the system further inward. CloudMatrix-style performance, failure rates, and total cost of ownership still need stronger independent evidence. Therefore, China optical mesh is an important competitive variable, but it is not the best first Korean equity landing.

5. Korean Listed Equity Map

Layer	Bottleneck	Korean names	View
HBM4/HBM4E	Vera Rubin and AI server memory	SK hynix, Samsung Electronics	Structural beneficiaries. SK hynix leads, Samsung is the catch-up option
SRAM/LPU foundry	Low-latency decode accelerators	Samsung Electronics	Still not fully visible in revenue, but the option is underappreciated
AI storage and KV cache	Agent memory, eSSD, PCIe 6.0	Samsung Electronics, FADU	Extends the memory hierarchy below HBM
HBM packaging tools	TC bonders, advanced packaging	Hanmi Semiconductor	Real bottleneck, but customer concentration and valuation matter
Power equipment	Transformers, switchgear, grid connection	HD Hyundai Electric, LS ELECTRIC, Hyosung Heavy Industries	Direct US AI data-center power exposure
China optical mesh	Optical modules for Chinese clusters	Limited direct Korean exposure	Avoid unless supplier evidence appears

6. Power Equipment Is The Front-End Bottleneck

In the US, power is a binding constraint. Data centers need grid connection, transformers, switchgear, and high-voltage equipment before GPUs can be monetized.

HD Hyundai Electric is one of the clearest Korean examples. It raised its 2026 order target by 23%, from $4.222 billion to $5.185 billion, citing North American power infrastructure, 765kV transformers, North American data centers, and onshore generator demand for data centers. ([Seoul Economic Daily][5])

Factor	Investment read
P	Power-equipment pricing, high-voltage transformer margins, North America premium
Q	Data centers, transmission projects, 765kV demand, replacement cycle
C	Raw materials, capacity, delivery timing, quality costs
What to watch	Book-to-bill, North America orders, margins, 2027 capacity expansion
Failure mode	AI capex delay, grid permitting delay, local sourcing pressure, margin compression

The company quality and the entry price should still be separated. The theme is visible now. Better setups come when AI capex fears pressure the stock but backlog and pricing remain intact.

7. HBM4/HBM4E: More Tied To The US Stack Than China

BIS has treated HBM as important for AI training and inference at scale. ([BIS][6]) If China’s access is restricted, allied AI infrastructure becomes more dependent on Korean HBM.

SK hynix is the clearest leader. Yonhap reported that SK hynix won more than two-thirds of NVIDIA’s HBM orders for a next-generation AI platform. ([Yonhap][7]) But the investment issue is crowding. The market already recognizes SK hynix as the HBM winner.

Samsung is different. The market has long treated it as an HBM laggard. If HBM4E, SOCAMM2, PCIe 6.0 SSD, foundry, and packaging are repriced as one AI memory-hierarchy platform, the discount can narrow.

Name	Strength	Risk	Current view
SK hynix	HBM4 leadership, NVIDIA allocation, pure HBM exposure	Crowding, high expectations, 2027 supply risk	Wait
Samsung Electronics	HBM4E catch-up, foundry, storage, packaging option	HBM qualification delays, foundry losses, customer trust	Conditional Buy candidate

8. Samsung’s SRAM/LPU Foundry Option

The non-consensus point is Samsung. If the inference stack moves from GPU-only to HBM plus SRAM/LPU plus cache storage, Samsung is no longer just a memory catch-up story. It is also a foundry and memory-hierarchy option.

The thesis strengthens if at least two of the following become visible:

Condition	Evidence to watch
HBM4/HBM4E customer qualification and shipment	Earnings, guidance, HBM mix
LPU or AI ASIC foundry revenue or utilization improvement	Foundry loss reduction, HPC customers, AI accelerator production
AI storage and memory-hierarchy revenue	SOCAMM2, PCIe 6.0 SSD, eSSD, KV-cache demand

It weakens if HBM4 qualification is delayed by two or more quarters, foundry losses do not narrow, or LPU/AI ASIC revenue does not accrue to Samsung in a material way.

9. Hanmi Semiconductor: Real Bottleneck, Price Discipline Needed

HBM4 increases packaging complexity. Hanmi Semiconductor is a real TC bonder bottleneck. The Elec reported that Hanmi received its first HBM4 TC bonder order from SK hynix. ([The Elec][8])

The thesis is sound, but equipment stocks have order-gap risk. Follow-on SK hynix orders, official Samsung or Micron orders, and margin durability matter more than the broad HBM narrative.

10. Practical View

Priority	Exposure	View	Why
1	Samsung Electronics	Conditional Buy candidate	HBM4E, SRAM/LPU foundry, and AI storage options may not be fully priced
2	HD Hyundai Electric	Wait	Direct US data-center power exposure, but order momentum is visible
3	SK hynix	Wait	Best HBM4 business position, but crowded winner
4	Hanmi Semiconductor	Watchlist	HBM tool bottleneck, but needs repeat orders and customer diversification
5	China optical-mesh beneficiaries	Avoid	Direct Korean supplier evidence is weak

Conclusion

The Korean opportunity is not “China optical mesh.” It is the US AI factory bottleneck: power, HBM, SRAM/LPU, advanced packaging, and storage hierarchy.

SK hynix may have the strongest business position. HD Hyundai Electric may have the cleanest power-equipment purity. But the most asymmetric rerating candidate is Samsung Electronics if the market stops viewing it only as an HBM laggard and starts pricing HBM4E, SRAM/LPU foundry, and AI storage together.

Sources

Samsung and SK Hynix 2028E Profit Valuation: Cheap-Looking Numbers and Cycle Tests

Thu, 09 Jul 2026 14:20:00 +0900

Related context: this is a follow-up to AI token value and memory value added, Samsung 2Q26 signals and noise, late-July Big Tech earnings and the memory thesis, and Micron FY3Q26 AI memory review. Relevant hubs are the Exclusive Analysis Hub and the AI HBM Hub.

TL;DR

Samsung Electronics and SK Hynix look very cheap on 2028E net income. If we use MarketScreener aggregate estimates, 2028E net income is about KRW 439.8tn for Samsung and KRW 346.6tn for SK Hynix. Applying a 9 to 12x P/E gives fair market caps of KRW 3,958tn to 5,277tn for Samsung and KRW 3,119tn to 4,159tn for SK Hynix.

Against current market caps, that implies roughly +120 to +193% for Samsung and +104 to +172% for SK Hynix. The market is valuing them at only about 4.1x and 4.4x 2028E P/E, respectively.

But the bottleneck is not the multiple. It is the 2028E profit number. If 2028 earnings are normalized, both stocks are cheap. If 2028 is the cycle peak, a 4 to 5x P/E may not be cheap.

Item	Samsung Electronics	SK Hynix
MarketScreener 2028E net income	KRW 439.8tn	KRW 346.6tn
9 to 12x fair market cap	KRW 3,958tn to 5,277tn	KRW 3,119tn to 4,159tn
Current implied 2028E P/E	About 4.1x	About 4.4x
Scenario-weighted 2028E net income	About KRW 272tn	About KRW 208tn
Scenario-weighted 9 to 12x fair market cap	KRW 2,444tn to 3,258tn	KRW 1,876tn to 2,502tn
Scenario-weighted upside	About +36 to +81%	About +23 to +64%

The stronger relative setup is Samsung. SK Hynix has higher-quality HBM earnings, but more of that quality is already embedded in the price. Samsung has HBM catch-up, commodity DRAM/NAND leverage and a lower implied 2028E P/E.

1. Current Coordinates

Samsung’s net income belongs economically to both common and preferred shares, so the right denominator is total equity value, not common-share market cap alone. Depending on the public screen, Samsung market cap appears around KRW 1,753tn, while a common-plus-preferred equity value framework is closer to KRW 1,800tn. SK Hynix is around KRW 1,530tn.

Item	Samsung Electronics	SK Hynix
Reference share price	KRW 274,000 to 274,500	KRW 2,151,000 to 2,160,000
Analysis market cap	About KRW 1,800tn	About KRW 1,530tn
2026E OP reference	KRW 371tn to 381tn	KRW 280tn to 291tn
2027E OP reference	KRW 500tn to 574tn	KRW 379tn to 432tn

Samsung guided to 2Q26 revenue of KRW 171tn and operating profit of KRW 89.4tn. SK Hynix reported 1Q26 revenue of KRW 52.576tn, operating profit of KRW 37.610tn and net income of KRW 40.346tn, with operating margin of 72%.

Using net income as roughly 75% of operating profit, current prices imply 2026E P/E near 6 to 7x and 2027E P/E near 4 to 5x. The market is not treating these companies as secular growth stocks. It is treating them as memory companies at or near peak earnings.

2. The Point-Estimate Valuation

MarketScreener aggregate estimates put 2028E net income at KRW 439.781tn for Samsung and KRW 346.562tn for SK Hynix. The same dataset puts 2028E EBIT at KRW 543.635tn and KRW 423.866tn, respectively.

KRW tn	2028E net income	9x fair cap	10x fair cap	12x fair cap	Current cap	Upside
Samsung Electronics	439.8	3,958	4,398	5,277	About 1,800	+120% / +144% / +193%
SK Hynix	346.6	3,119	3,466	4,159	About 1,530	+104% / +127% / +172%

This simple calculation says both stocks are cheap. It also says Samsung is slightly more asymmetric because its implied 2028E P/E is lower.

The danger is that 2028E is still far away. Memory cycles have repeatedly produced large peak profits that later collapsed. Therefore, the real question is whether 2028E is a normalized profit base or a peak.

3. How to Estimate 2028E

Extending 2027E mechanically is not enough. A top-down “AI capex keeps rising, so memory profit rises” approach is also too thin. Memory profits are often set by supply, not demand. The 2028 supply base is already being shaped by 2026 to 2027 capex.

The better process is:

Step	What to do	Why it matters
1	Split profits into contracted HBM, non-contracted HBM/server DRAM, and spot DRAM/NAND	Each bucket has a different earnings floor
2	Track supply instead of forecasting it abstractly	2027 capex determines 2028 bit supply
3	Segment demand into AI servers, general servers and consumer devices	Price elasticity differs sharply
4	Force memory-cycle base rates into the model	Peak-year earnings historically fall hard
5	Use scenario distributions and reverse implied earnings	Current price already embeds a view

The single most useful future disclosure would be the 2028 HBM long-term agreement coverage ratio. If a large share of 2028 shipment is already contracted, the 2028 profit floor is much higher than in historical cycles.

The main supply variables are 2027 capex guidance, equipment orders from ASML/AMAT/TEL, Hynix’s Yongin and Cheongju investments, Samsung’s HBM and packaging catch-up spending, CXMT and YMTC expansion, and whether HBM bottlenecks remain in stacking and packaging rather than wafers alone.

Demand must also be segmented. AI server demand is relatively inelastic. General server demand is cyclical. PC and smartphone memory demand is price-sensitive and can already show demand destruction when memory prices rise too quickly.

Historical base rates are harsh.

Cycle	Operating profit change
Samsung consolidated 2018 to 2019	-53%
Samsung DS 2018 to 2019	-69%
SK Hynix 2018 to 2019	-87%
2023 memory downcycle	Loss-making episodes

This cycle deserves an upward adjustment because of HBM contracts, oligopoly structure, wafer cannibalization by HBM and low AI-server price elasticity. But it also deserves caution because of Chinese supply, consumer demand destruction, policy-driven capacity expansion and the statistical extremity of 70%+ operating margins.

4. Scenario Valuation

Scenario	Probability	Samsung 2028E NI	9x	12x	Upside	SK Hynix 2028E NI	9x	12x	Upside
S1 shortage persists	25%	KRW 420tn	3,780	5,040	+110 to +180%	KRW 315tn	2,835	3,780	+85 to +147%
S2 soft landing	45%	KRW 280tn	2,520	3,360	+40 to +87%	KRW 215tn	1,935	2,580	+26 to +69%
S3 hard landing	30%	KRW 135tn	1,215	1,620	-32 to -10%	KRW 110tn	990	1,320	-35 to -14%
Probability-weighted	-	KRW 272tn	2,444	3,258	+36 to +81%	KRW 208tn	1,876	2,502	+23 to +64%

MarketScreener’s estimate sits closer to S1. The probability-weighted model blends S2 and S3 risks. These two approaches do not conflict. One shows how cheap the stocks are if consensus is right. The other shows how much risk discount should be applied to that consensus.

5. What the Current Price Already Embeds

Reverse the current market caps by a 9 to 12x P/E and we get the 2028E net income implied by today’s prices.

Item	Current cap	Implied 2028E NI at 9x	Implied 2028E NI at 12x
Samsung Electronics	About KRW 1,800tn	About KRW 200tn	About KRW 150tn
SK Hynix	About KRW 1,530tn	About KRW 170tn	About KRW 127tn

Against rough 2027E net income estimates, current prices already imply a meaningful drop.

Item	Rough 2027E NI	Implied 2028E NI	Implied decline
Samsung Electronics	About KRW 400tn	KRW 150tn to 200tn	-50 to -63%
SK Hynix	About KRW 310tn	KRW 127tn to 170tn	-45 to -59%

This is the key reframing. The current bet is not necessarily that 2028 earnings will be spectacular. It may simply be that the decline from 2027 will be shallower than the 45 to 63% decline already embedded in prices.

6. Samsung Versus SK Hynix

Company	Core thesis	Risk
Samsung Electronics	Lower implied 2028E P/E, HBM catch-up, commodity DRAM/NAND leverage	HBM4 delay, foundry/system LSI losses, weaker reported-to-core profit conversion
SK Hynix	Highest-quality HBM profit pool, customer reference, strong margin	More optimism embedded, ADR/event flow, need to prove 2028 LTA coverage

Samsung is not the HBM leader, but that is partly the point. Expectations are lower. If HBM4 qualification, yield and allocation improve, the stock gets both earnings revision and multiple repair. Samsung also has more leverage to conventional DRAM and NAND pricing.

SK Hynix has the cleaner HBM exposure and better current profit quality. But the market already gives it more credit for structural HBM change. The stock needs clearer evidence that 2028 profit is contracted and not just a peak-cycle extrapolation.

7. Practical Stance

Stock	Stance	Reason	What to confirm
Samsung Electronics	Hold-biased, pullback candidate	More asymmetric 2028E valuation and catch-up optionality	July 30 segment data, HBM4 progress, NAND/DRAM pricing
SK Hynix	Watchlist, no event-chasing	Best HBM quality, but more expectation embedded	ADR absorption, 2Q call LTA coverage, HBM4 premium

For Samsung, selling aggressively here requires assigning a very high probability to a 2028 hard landing. The data does not yet justify that. For SK Hynix, the company quality is clear, but ADR, new-share absorption and 2028 contract visibility matter before chasing.

8. Invalidation

The thesis weakens if two or more of the following happen.

Invalidation signal	Meaning
2027 capex across the three major memory firms rises more than 50% YoY	Supply discipline breaks
CXMT qualifies high-end server DDR5 with major customers	Commodity DRAM ceiling falls
Chinese suppliers make visible HBM progress	Premium durability weakens
Big Tech capex guidance slows sharply	AI server demand downside
No disclosure or evidence of 2028 HBM LTA coverage	Peak-profit discount remains
YMTC NAND ramp accelerates	NAND pricing risk

Final View

Samsung and SK Hynix both look cheap on 2028E earnings. But cheap-looking is not the same as cheap.

Using MarketScreener point estimates, both have roughly double-like potential. Using a supply, contract and base-rate scenario model, the upside narrows to +36 to +81% for Samsung and +23 to +64% for SK Hynix. That is still attractive, but the margin of safety is smaller.

The key is 2028E. It cannot be built from demand stories alone. It depends on HBM LTA coverage, 2027 capex discipline, DRAM/NAND pricing, Chinese supply and Big Tech capex.

The relative preference is Samsung. SK Hynix has better HBM earnings, but requires more belief. Samsung has catch-up optionality, broader DRAM/NAND leverage and a lower implied 2028E P/E. For both names, the next checks are late-July Big Tech capex, Samsung’s July 30 detailed results, SK Hynix ADR absorption and the 2Q call.

Evidence and Limits

Classification	Content
Fact	Samsung 2Q26 preliminary revenue KRW 171tn and OP KRW 89.4tn. SK Hynix 1Q26 revenue KRW 52.576tn, OP KRW 37.610tn and net income KRW 40.346tn. MarketScreener 2028E net income of KRW 439.781tn for Samsung and KRW 346.562tn for SK Hynix
Inference	Net income approximated at 75% of OP. Current prices imply 2028E net income of KRW 150-200tn for Samsung and KRW 127-170tn for SK Hynix
Speculation	Scenario probabilities and 2028E net income assumptions: S1 25%, S2 45%, S3 30%
Blocked	2028 HBM LTA coverage, customer allocations, actual contract prices and detailed analyst-by-analyst 2028E assumptions are not fully available from public data

Sources

Samsung Electronics recorded the world's largest quarterly operating profit. Signals and noise poured in today.

Wed, 08 Jul 2026 22:35:00 +0900

Context This is a follow-up to the Samsung 2Q26 preview, Micron FY3Q26 review, NVIDIA earnings analogue work, late-July Big Tech earnings-call scenario map, and Big Tech financing relay note. Related hubs are the Exclusive Analysis Hub and the AI HBM Hub.

TL;DR

Samsung Electronics presented provisional sales of 171 trillion won and operating profit of 89.4 trillion won for the second quarter of 2026. Based on Samsung Electronics’ official newsroom, this is a very large number compared to the operating profit of 57.23 trillion won in the previous quarter and 4.68 trillion won in the same period last year. Looking at memory prices, AI server demand, hyperscaler investment, and HBM4 ramp-up together, the quality of Samsung Electronics’ investment argument has actually become stronger.

However, the stock price moved in the opposite direction. Based on local DB, Samsung Electronics closed at 296,000 won on July 7 and 277,500 won on July 8. On July 7, foreigners net sold about 1.82 trillion won, and on July 8, they sold about 874.1 billion won. Individuals made net purchases of approximately 2.32 trillion won and 584.7 billion won, respectively. Although good performance was achieved, foreign and programmatic supply and demand saw peak-out, interest rates, exchange rates, and overcrowding of positions before profit levels.

The conclusion of this article is as follows.

Item	Judgment
Samsung Electronics Investment Argument	enforce. DRAM/NAND prices, AI servers, and hyperscaler CapEx are all favorable.
Short term timing	damage. Despite good performance, there was selling pressure for two days in a row.
supply and demand	danger. Foreign selling and private absorption are conflicting, and the Korean market is weak.
Sell decision	no. The current data is closer to digesting expectations and supply and demand pressures than to damaging the thesis.
Additional purchase decision	Not yet. You must check ASML, TSMC, Samsung Electronics IR, foreign supply and demand, and exchange rates and interest rates.
action	Strong holding, freeze on additional purchases, make a decision after confirming the next gateway.

To sum it up in one line, the information released today is not “Samsung Electronics has worsened,” but is closer to “the market has already had too many good numbers, and now demands the next number and the quality of supply and demand.”

1. Why should we separate signal and noise?

Today’s data is not read in only one direction. Numbers are good. However, stock prices and supply and demand are bad. Memory price comments are strong. However, spot trading takes a breather on some items. Hyperscaler CapEx continues to grow. However, long-term interest rates in the United States are also high.

On days like this, looking at only one side is wrong. “It’s the world’s largest quarterly operating profit, so you must buy it” is also hasty, and “the cycle is over because it fell behind despite good performance” is also hasty.

Therefore, this article divides the data into four layers.

floor	Content	Role in Investment Decisions
Official Performance	Samsung Electronics 2Q26 provisional sales and operating profit	Check profit level
industrial signals	DRAM·NAND price, AI server, hyperscaler CapEx	Check the quality of the thesis
market noise	Sluggish spot trading, Morgan Stanley peak logic, ASML·TSMC high bar	Check short-term timing
Supply/Demand/Macro	Foreign selling, exchange rate, interest rate, Korean market breadth	Position Sizing

The key is to separate the profit direction and the stock price direction. Samsung Electronics’ main business profit direction has improved. However, the drivers of short-term stock prices were not profits, but supply and demand and the discount rate.

2. Officially confirmed values: Sales 171 trillion won, operating profit 89.4 trillion won

[Fact] Samsung Electronics announced its second quarter provisional results through its official newsroom on July 7, 2026. Consolidated sales are approximately 171 trillion won and operating profit is approximately 89.4 trillion won. According to Korean disclosure regulations, the company presented a median value rather than a range, and explained that the actual estimated range was KRW 170 trillion to KRW 172 trillion in sales and KRW 89.3 trillion to KRW 89.5 trillion in operating profit. Source: Samsung Electronics Official Newsroom

Item	2Q26 provisional figures	1Q26	2Q25	Compared to previous quarter	Year-on-year comparison
Sales	171.0 trillion won	133.87 trillion won	74.57 trillion won	+27.7%	+129.3%
operating profit	89.4 trillion won	57.23 trillion won	4.68 trillion won	+56.2%	+1,810.3%

This number is extremely unusual for a single quarter’s operating profit. The expression “world’s largest” may be subject to consideration depending on the comparison target and accounting standards, but it is clear that it is an overwhelmingly large number, at least compared to the normal quarterly profits of global mega-caps.

What is important is not the operating profit itself of 89.4 trillion won, but the quality within it. The interpretation presented in the attached research is as follows.

Item	Interpretation
reported operating profit	89.4 trillion won
Adjustment of one-time provisions such as DS performance bonus	The logic that profits from main business should be considered higher
Core OP estimation	Possibility of over 100 trillion won excluding provisions
DRAM ASP	+47% compared to previous quarter
NAND ASP	+66% compared to previous quarter
HBM4	Maintain ramp-up expectations

[Inference] Just looking at the reported OP of KRW 89.4 trillion is strong, but if DRAM and NAND prices rose significantly at the same time and there was an effect of provisioning, the strength of the main business may be stronger than the reported number. This point is the most important signal of Samsung Electronics’ thesis.

3. Stock price and supply and demand: Foreigners sold after good numbers

Despite good performance, the stock price fell. The prices and supply and demand based on local DB are as follows.

date	closing price	Daily fluctuations	Foreign net purchase estimates	Estimated institutional net purchase	Individual net purchase estimate
2026-07-03	309,500 won		-KRW 387.2 billion	+1.3063 trillion won	-KRW 918.4 billion
2026-07-06	318,000 won	+2.75%	-640.9 billion won	+5.8 billion won	+596.4 billion won
2026-07-07	296,000 won	-6.92%	-KRW 1.8207 trillion	-KRW 547.5 billion	+2.3232 trillion won
2026-07-08	277,500 won	-6.25%	-KRW 874.1 billion	+282.7 billion won	+584.7 billion won

The source is the local Kiwoom-based prices_daily and investor_flow_daily tables. On July 8, the foreign ownership rate was recorded at 46.58%.

The message of this table is clear. The market did not believe that Samsung Electronics’ profits had weakened, but that it was a structure in which foreigners reduced their risk and individuals accepted it in stocks that had already risen. On July 7, foreigners and institutions sold together, and individuals bought significantly. Institutions received some on July 8, but foreign selling continued.

Therefore, today’s judgment is “performance is good, but supply and demand are bad.” It is not a basis for selling, but a basis for freezing additional purchases.

4. Four axes of Samsung Electronics thesis

The argument for investing in Samsung Electronics can be organized into four axes.

axis	Content	Current judgment
AI server DRAM·HBM	Demand for high-bandwidth and high-capacity memory for AI servers and GPU/ASIC systems	Strengthening
eSSD·NAND mix	AI server storage and enterprise SSD demand	Strengthening
Hyperscaler CapEx	Long-term investment in cloud and AI data centers	Strengthening
Foundry turnaround	Edge logic·packaging·customer recovery options	Not yet an option

Among these, the key ones are numbers 1 to 3. Foundry is a valuation upside option. For the stock price to hold up now, the foundry does not need to improve immediately. Investments in memory, eSSD, and hyperscaler alone explain the strength of profits. However, additional rerating requires reducing the foundry deficit and confirming HBM4 and HBM4E customers.

5. Signal 1: Legacy DRAM and NAND prices are stronger than expected

The most important bullish signal in the attached research is the memory price trend summarized in the comments of Meritz Securities Commissioner Kim Seon-woo. Since the full text of the original data was not directly compared, this article classifies it as “market comment-based interpretation.” Still, numbers and direction are important in interpreting Samsung Electronics’ investment argument.

The key point is that seasonal demand for the second half of the year began to arrive in July, and while B2B server demand was absorbing supply, B2C smartphone and PC customers also began to feel pressure to secure supplies.

Legacy DRAM

Item	Figures and Content
Nanya 3Q26 DRAM price proposal	DDR4·LPDDR4 +50~90% QoQ
Consumer DRAM price	Above $3/Gb, upward acceleration
supply expansion	Judgment that meaningful supply expansion will be difficult before 2027
Cause	Decrease in investment between 2023 and 2025 after loss in 2023, lack of cleanroom

Legacy NAND

Item	Figures and Content
Windbond·MXIC SLC NAND proposal	3Q26 +70~100% QoQ
price range	$1.9~2.0/Gb
Demand changes	SLC NAND orders increase due to MLC NAND shortage
Interpretation	Price pressure spreads not only to high-performance AI NAND but also to legacy NAND

Supply Structure

Item	Content
DRAM total supply growth rate in 2027	Expected 20~25%
Server demand growth rate next year	50~70% expected
Bottleneck causes	Low-end/legacy shortage occurs due to concentration on high-end memory

[Inference] If this data is correct, Samsung Electronics’ bull case is wider than SK Hynix’s HBM pure exposure. Samsung Electronics has a discount factor for being behind in HBM, but it also receives price increases for general DRAM, server DRAM, eSSD, NAND, and legacy products. In other words, “the frame of disparaging Samsung Electronics based solely on HBM” is weakened.

6. Signal 2: It is difficult to say that AI CapEx has decreased yet

Hyperscaler investment also supports Samsung Electronics’ thesis. According to numbers shared by Goldman Sachs charts, hyperscaler CapEx grows as follows:

Year	Hyperscaler CapEx
2024	$227 billion
2025	$406 billion
2026E	$740 billion
2027E	$996 billion

The forecast for 2027 is about 4.4 times that of 2024. This number is also in line with the bond and capital increase flows covered in Big Tech Financing Relay. Amazon, Alphabet, and Meta are even using corporate bonds and stock issuances to invest in AI infrastructure at a level that is difficult to cover with internal cash flow alone.

What is important for Samsung Electronics is that “AI CapEx does not end at GPU.” Investing in an AI data center is not just about purchasing GPUs. Server DRAM, HBM, eSSD, NAND, substrate, power, network, and cooling will also increase.

AI CapEx Items	Connection with Samsung Electronics
GPU·ASIC server	HBM, server DRAM
data center storage device	eSSD, NAND
High-density server proliferation	High capacity DDR5, LPDDR series, storage
Rack-level AI factory	Increased memory payload
Long-term investment cycle	Strengthening visibility in 2027 rather than short-term performance in 2026

[Inference] If hyperscaler CapEx increases by 2027, it is difficult to conclude that Samsung Electronics’ memory profits will immediately decline from the peak in one quarter. The problem is not whether there is no demand, but how much the market has already reflected that demand in the price.

7. Signal 3: AI server supply chain is not yet broken

Taiwan AI server supply chain data included in the attached research is also important.

Enterprise	Latest Signals	Interpretation
Wistron	June sales +10.9% MoM, +53.8% YoY	GB300 NVL72, B300 shipment increases
Wiwynn	June sales +32.5% MoM, +29.8% YoY	AMD Helios enters, prepares for mass shipment of 3Q ASIC AI servers
Unimicron	ABF·HDI substrate revenue record high	ABF expansion CapEx increase mentioned

This data does not fit the claim that “demand for AI servers has declined.” If server OEMs, ODMs, and substrate companies still show strong shipments and sales, memory demand is unlikely to disappear easily.

물론 이 데이터만으로 삼성전자 실적을 바로 산출할 수는 없다. But the direction is clear. AI 서버 공급망은 아직 멈추지 않았다. 오히려 GPU, ASIC, 서버, 기판, 메모리, storage가 함께 움직이는 구조다.

8. 소음 1: TrendForce spot 데이터는 약하지만 붕괴는 아니다

There are also opposing signs. TrendForce spot market comments were summarized as an atmosphere where trading was quiet and buyers were waiting.

Item	spot change
DDR5 16Gb	+0.50%
DDR5 eTT	0.00%
DDR4 16Gb 3200	+3.42%
DDR4 eTT	-1.51%
DDR4 8Gb 3200	+2.88%

Looking at this table, it is excessive to say, “Memory prices have fallen.” 대부분은 상승 또는 보합이고, 일부 eTT 품목만 하락했다. 더 정확한 표현은 “거래는 한산하고, spot 시장은 가격 발견 중”이다.

The important distinction is spot and contract.

Category	meaning	Relevance to Samsung Electronics thesis
Spot	Short-term transaction prices in distribution and module markets	Quickly reflects sentiment and short-term inventory
Contract	Large customer long-term contract price	Server DRAM, HBM, eSSD more important to profits

삼성전자 thesis의 핵심은 spot이 아니다. 핵심은 contract 가격, server DRAM, HBM, eSSD 주문, hyperscaler 수요다. 진짜 위험 신호는 spot 거래 부진 자체가 아니라 다음 네 가지가 함께 나오는 경우다.

Contract price cut
Slowdown in server DRAM orders
HBM·eSSD order slowdown
주요 고객의 inventory correction 코멘트

Currently, all four have not been confirmed. Therefore, it is correct to view TrendForce data not as bear confirmation but as a signal of anticipation digestion and price discovery.

9. Noise 2: Morgan Stanley peak logic is only partially correct

Morgan Stanley-type peak thesis should not be ignored either. The key point is that the speed of improvement in memory indicators is approaching its peak, and inventory normalization and price increases may slow down. AI trade 안에서는 반도체보다 Alphabet, Amazon 같은 AI cloud 쪽을 선호한다는 관점도 제시됐다.

This claim is partly correct. This is especially important for short-term stock prices.

Morgan Stanley-style warning	Areas to acknowledge
Improvement speed peak possibility	Stock prices are more sensitive to slope than level.
Price increases slowing	3Q 이후 상승률이 낮아지면 멀티플 압박 가능
Semiconductor overcrowding position	Sell-the-news is possible even with good performance
AI cloud preferred	Possibility of rotation towards consumers rather than semiconductors

However, extending this to the logic of structural selling is weak. There is too much opposing data.

Opposite logic	Evidence
Continued expansion of AI investment	Hyperscaler CapEx expected to increase by 2027
Prolonged supply shortage	Legacy shortage intensifies due to transition to cleanroom and high value-added products
Server demand growth	Server demand expected to increase by 50~70% next year
Domestic securities firms continue to buy	Target prices maintained or raised, including KB 600,000 won, Daol 585,000 won, NH 530,000 won, and Hana/iM 480,000 won

[Inference] The Morgan Stanley peak thesis is valid as a warning, “Don’t chase now.” However, the basis for “discarding Samsung Electronics’ thesis” is insufficient.

10. Noise 3: ASML and TSMC are high bar events

The next gateway is ASML and TSMC. ASML’s official financial calendar displays Q2 2026 financial results on July 15, 2026. TSMC’s official IR page announces the 2Q26 earnings conference on July 16, 2026. Source: ASML financial calendar, TSMC 2Q26 quarterly results

The problem is that even if good performance is achieved, it is impossible to know whether the stock price will rise. Samsung Electronics’ 2Q26 provisional results are already a preview. If the high bar is high, even good numbers seem lacking.

Here’s what to look for in ASML and TSMC:

Event	checkpoint	Connection with Samsung Electronics
ASML 2Q26	bookings, EUV·High-NA demand, Chinese regulatory comments	Confirmation of semiconductor capex expectations and equipment demand
TSMC 2Q26	AI/HPC revenue, CoWoS·advanced packaging, CapEx guidance	Identifying AI accelerator and HBM packaging bottlenecks
Market reaction	SOXX, NVDA, MU, TSM, ASML stock price reaction	Check if it is sold or relieved with good performance

As Mizuho commented, the fact that Samsung Electronics’ stock price fell despite its good preliminary results shows the possibility of a global IT momentum shift. So ASML and TSMC are events that see “how the market digests good numbers” rather than “confirming the numbers.”

11. Noise 4: Press multiple for interest rates and exchange rates

Even if Samsung Electronics’ thesis is good, the stock price may be depressed if the discount rate increases. The interest rate environment based on the attached research is as follows.

Item	level
US 10-year note	About 4.58%
US 30-year note	About 5.07%
Japanese 10-year note	About 2.88%
Japanese 40-year note	About 4.01%

These numbers do not undermine the demand thesis. However, PER can be lowered. A typical section where stock prices fall despite strong profits is “the section where EPS rises but the multiple falls faster.”

The same goes for exchange rates. A weak won may be beneficial to Samsung Electronics’ performance. However, from a foreigner’s perspective, it increases foreign exchange losses, foreign exchange hedging costs, and the risk of Korea’s share. The foreign ownership rate of 46.58% and continuous selling on July 8 are signals that this part needs to be checked.

12. Noise 5: Foreign outflow and leverage piping

The attached research includes comments on foreign outflow and leverage piping based on Bloomberg and Telegram. The figures presented here, such as foreign outflow in the Korean market since the beginning of the year -$100.16 billion, foreign securities outflow in March -$36.55 billion, and USD/KRW 1,554 won, need to be compared with official statistics by collection range. Therefore, in this article, only directional signals are used.

Still, the message is important.

Observation	Interpretation
Foreigners reduced Korea	Rather than avoiding Korea, it may be a memory peak-out bet.
Domestic institutions and individuals absorbed the rebound	It is a domestic liquidity absorption market rather than a foreign-led rally.
Single stock leverage/short gamma structure	Forced selling and hedging flows during declines can exaggerate prices.
Exchange rate rise	It is a barrier to additional purchases by foreigners.

[Inference] The short-term plunge is more likely to be the result of memory peak-out betting, liquidation of leverage positions, and reduction of foreign risks, rather than the collapse of Samsung Electronics’ main business. However, we cannot conclude that this is not a structural risk. We need to check whether foreign spot sales stop.

13. Korean market performance is also not good

This is not just a problem for Samsung Electronics. As of 06:10 on July 8, the local KR integrated screener classified the Korean market as NEUTRAL.

Item	figures
Proportion of stocks above the 50-day line	15.6%
Proportion of stocks above the 200-day line	21.5%
Passed items	17
New Passed	5
Breakaway	7
Breakout	0

Newly passed stocks were APR, Handsome, GS Retail, Hana Financial Group, and KT&G. Stocks that left included TSE, YG-1, and Samsung Electronics.

13 passed the quality compounder screen, and at the top were Gigabis, SK Hynix, VM, AP, and PSK. Only three companies, APR, Seegene, and Kolmar Korea, passed the smart money + quality condition.

The meaning of this data is simple. Market breadth is weak. Even if Samsung Electronics’ thesis is good, it is not widely accepted by the entire Korean market. In this environment, additional purchases should not be made by reporting only profit data. We need to see whether supply and demand and market width return.

14. Then what should I do?

The current conclusion is “strong holding, freeze on additional purchases.”

action	Judgment
holding	possible. The thesis is intact.
Additional Purchase	Still waiting. Need to confirm supply/demand, ASML, TSMC, IR
Sell	It’s not just data now. The profit direction actually improved.
Short term trading	After confirming the slowdown in foreign selling and the global semiconductor response from July 15th to 16th

There are four additional purchase conditions:

Foreign spot sales of Samsung Electronics are slowing down, and programmatic selling pressure is decreasing.
ASML bookings and TSMC AI/HPC·CoWoS·CapEx comments do not indicate a slowdown in AI investment.
DS core profit, DRAM·NAND contract price, HBM4·HBM4E customer progress, and eSSD orders are confirmed in Samsung Electronics IR.
The sluggish spot market does not spread to contract price cuts or a slowdown in server DRAM orders.

The conditions for invalidation are as follows:

Invalidation Conditions	meaning
Contract DRAM/NAND price decline	price thesis undermined
Server DRAM·HBM·eSSD order slowdown	AI memory demand thesis undermined
ASML·TSMC strongly talks about CapEx slowdown	Hyperscaler investment thesis weakened
Foreign selling and won weakening intensify simultaneously	Pressure to reduce Korea’s proportion
Market breadth further worsens	Even if Samsung Electronics is good, supply and demand absorption is weakened

15. Conclusion

Samsung Electronics’ 2Q26 provisional performance is not a number that weakens the thesis. Rather, it is the opposite. Operating profit of KRW 89.4 trillion, DRAM ASP +47%, NAND ASP +66%, soaring legacy memory prices, hyperscaler CapEx continuing until 2027, and AI server supply chain sales all strengthen the case for investing in Samsung Electronics.

However, today’s stock prices and supply and demand have damaged the timing. Foreigners sold, and individuals received. Interest rates are high, exchange rates are a burden, and high bar events such as ASML and TSMC remain. Market width is also weak.

So the conclusion is clear.

Samsung Electronics has not become worse. However, today’s market called for “the next good number and supply and demand to believe in that number” rather than “a good number.” Now is not the time to abandon the thesis, but rather to protect the position, stop making additional purchases, and check for the next gateway.

Basis classification

Category	Content
Fact	Samsung Electronics’ official 2Q26 provisional sales of 171 trillion won and operating profit of 89.4 trillion won. Comparative figures for 1Q26 and 2Q25. ASML July 15th, TSMC July 16th earnings schedule. Samsung Electronics price, supply and demand, and foreign ownership rate in local DB.
Inference	Samsung Electronics’ thesis was strengthened, but the short-term timing was judged to be damaged. TrendForce spot data is interpreted as price discovery, not bear confirmation. Morgan Stanley’s peak thesis is a catch-up warning and not a basis for structural selling.
Speculation	A scenario in which hyperscaler CapEx increases to approximately $1 trillion by 2027 and can support Samsung Electronics’ memory demand for a longer period of time.
Blocked	Comparison of Bloomberg foreign outflow figures with official statistics, confirmation of the full text report by Meritz·TrendForce, 2Q26 actual profit by Samsung Electronics DS division, and volume and margin by HBM4·HBM4E customer.

Evidence Ledger

Item	Source
Samsung Electronics 2Q26 provisional performance	Samsung Electronics Official Newsroom
ASML 2Q26 Schedule	ASML financial calendar
TSMC 2Q26 Schedule	TSMC 2Q26 quarterly results
Samsung Electronics price/supply	Local Research DB: `prices_daily`, `investor_flow_daily`, `foreign_ownership_daily`, as of 2026-07-08
Memory price·AI server·foreign outflow comments	Attached research packet. Uncollated items in the full text are separated from the main text as market comments or directional signals
Related prior analysis	Korea Invest Insights’ Samsung Electronics 2Q26 preview, Micron FY3Q26, Big Tech earnings call, Big Tech financing, NVIDIA similar quarter analysis

Late-July Big Tech Earnings Calls and the Memory Thesis Scenario Map

Tue, 07 Jul 2026 20:59:50 +0900

Context This is a follow-up to Micron FY3Q26 earnings, Samsung Electronics 2Q26 preview, Who pays for 2027 semiconductor consensus?, NVIDIA’s elasticity as a memory-cycle guide and the 1H26 AI infrastructure bottleneck review. Related hubs: Exclusive Analysis and AI HBM Hub.

TL;DR

The late-July earnings calls from Alphabet, Microsoft, Meta and Amazon are the next stress test for the memory thesis. My probability map is: strengthening 50%, unchanged 35%, weakening 15%. Within the weakening bucket, true structural damage is only around 4%.

The key question is not whether Big Tech cuts capex. The four companies are still constrained by compute, power and component availability. The real questions are narrower: whether 2027 capex falls meaningfully short of the market’s roughly $1 trillion expectation, and whether memory pricing plus memory’s share of server BOM stay intact.

The execution rule is simple: do not chase before the calls. Wait for the checklist. Korean memory stocks have already risen sharply over the last year, so the upside from good news can be smaller than the downside from a bad surprise.

Samsung’s July 7 Update

Samsung Electronics released its 2Q26 preliminary result on July 7. It guided to revenue of about KRW 171 trillion and operating profit of about KRW 89.4 trillion. The official ranges were KRW 170-172 trillion for revenue and KRW 89.3-89.5 trillion for operating profit. Source: Samsung Electronics 2Q26 earnings guidance.

That is broadly in line with the earlier reported OP range of KRW 87-92 trillion. It does not settle the debate. The important event is the July 30 Samsung earnings call, where investors need to separate reported OP from core semiconductor profit, check 3Q DRAM and NAND pricing tone, and listen for HBM4/HBM4E customer qualification and allocation comments.

What the Event Judges

Local timing	Event	Status
July 7	Samsung Electronics 2Q26 preliminary result	Officially released
Around July 22-23 US	Alphabet 2Q26 call	Aggregator estimate, official IR date still needs confirmation
Around July 28 US	Microsoft FY4Q26 call	Aggregator estimate, official IR page still needs confirmation
Late July	SK Hynix 2Q26 result	Aggregators conflict, official page needs confirmation
Around July 29 US	Meta 2Q26 call	Aggregator estimate
Around July 30 US	Amazon 2Q26 call	Aggregator estimate
August	SK Hynix Nasdaq ADR listing	Pending SEC process

The headline question is not “will Big Tech cut spending?” The real question is whether 2027 spending growth is still steep enough to support the memory earnings path.

Big Tech Capex Baseline

Company	2025 actual	2026 plan	Growth	2027 commentary	Demand evidence
Alphabet	$91.4B	$180-190B	About 2x	CFO said 2027 spending should rise materially above 2026	Cloud backlog of $462B, roughly doubled sequentially
Microsoft	About $118B, inferred	About $190B	+61%	Double-digit revenue and operating income growth, but no explicit capex figure	RPO $627B, +99% YoY
Meta	$72B	$125-145B	+87.5% at midpoint	No number, flexible language	Contractual commitments rose $107B in 1Q
Amazon	$125B	About $200B	+60%	No number	AWS backlog $364B, Anthropic contract separate
Total	About $410B	About $710B	+73%	Market expectation above $1T, +41%

Alphabet is the strongest setup. It raised 2026 capex guidance, said 2027 would be meaningfully higher, and later announced an $80B capital raise to expand AI infrastructure and compute. Microsoft is the key swing factor: RPO is huge, but sequential RPO growth needs to reaccelerate. Meta is two-sided: it raised guidance because of component and memory costs, but it also kept flexibility language for 2027. Amazon has the largest absolute capex plan, but its free cash flow pressure makes its tone worth watching.

Memory Price Facts

Quarter	DRAM	NAND	Interpretation
1Q26	+55-60%	+33-38%	Server DRAM was even stronger
2Q26	+58-63%	+70-75%	Peak rate of increase
3Q26 forecast	+13-18%	+10-15%	Still rising, but at a slower pace
4Q26 reference	PC DRAM +3-8%	Not specified	A consumer-side reference point

Source: TrendForce 3Q26 price outlook.

The 3Q deceleration is not a demand collapse. It reflects base effects, consumer affordability limits and a rising share of long-term contracts. The more important point is that server memory remains tight. If Big Tech capex grows more slowly while memory takes a larger share of server BOM, memory revenue can still grow faster than aggregate capex.

Micron SCA Matters

Micron disclosed 16 strategic customer agreements in its FY3Q26 materials. The contracts run from 2026 through 2030, include take-or-pay structures, and cover about 20% of DRAM volume and roughly one-third of NAND volume. Minimum contracted revenue is around $100B, and customer deposits plus commitments are around $22B. Source: Micron prepared remarks.

This is the most important underappreciated change in the cycle. Big Tech is not merely accepting higher memory prices. It is signing multi-year, non-cancelable supply assurance contracts. That lowers the historical cyclicality discount for memory earnings. It does not eliminate risk, because price ceilings may cap upside on contracted volume, but it changes the downside profile.

For SK Hynix, the key question is whether its long-term contracts have similar floors without the same upside caps. That detail is not officially confirmed. It should be the first question on the SK Hynix 2Q call.

Scenario Map

Scenario	Probability	Meaning for memory stocks
Strengthening	50%	Earnings revision cycle extends. Upside depends on how explicit 2027 capex becomes.
Neutral	35%	2026 plans remain intact, but 2027 language stays cautious. Short-term headline noise likely.
Weakening	15%	Split into narrative weakening and structural damage.
Narrative weakening	About 11%	Stocks correct, but price and contract data stay healthy. Could become an entry window.
Structural damage	About 4%	Thesis reset. Examples: 2026 capex cut, memory-content redesign, digestion language from multiple companies.

Call-by-Call Update Rules

Sequence	Result	Strengthening probability update
Alphabet	Plan maintained, 2027 reaffirmed, backlog up	50 → 55
Alphabet	2027 language retreats or backlog stalls	50 → below 40
Microsoft	$190B plan maintained, supply tightness, 2027 growth tone, RPO growth resumes	55 → 62
Microsoft	Normalization or single-digit growth tone	55 → 45
Meta	Plan maintained, commitments rise, cloud monetization becomes official	62 → 66
Meta	April-style flexibility language returns with numbers	62 → 50
Amazon	$200B maintained, memory price absorbed	66 → 68-70
Amazon	Amazon-only pacing comments	66 → 60, not enough to kill the thesis

If two or more companies cut plans or use digestion language, the table no longer applies. That becomes a thesis reset exercise.

Korean Memory Checklist

This section is a public research checklist, not a trading instruction. The late-July Big Tech calls matter here because they help test the Korean memory thesis.

Name	Role	What to verify
Samsung Electronics	Korean memory core with DRAM and NAND volume leverage	DS core profit, 3Q memory price assumptions and HBM4 customer qualification.
SK Hynix	HBM purity and long-term contract structure	HBM shipment, ASP durability, contract pricing structure and HBM4 allocation.

Phrases To Listen For

Direction	Phrases
Positive	supply cannot keep up with demand / customer commitments rising / backlog rising / component or memory price pressure absorbed by budgets / 2027 capex expansion
Negative	digestion phase / lower utilization / investment pacing / optimization / delayed customer commitments / easing component price pressure / memory supply improving earlier than expected / memory content redesign

Kill Conditions

One item triggers reassessment. Two or more start the thesis-reset process.

Any of the four companies lowers 2026 capex plans.
Microsoft RPO or Alphabet Cloud backlog turns negative sequentially.
Companies use “digestion” or “pacing purchases to power availability” language.
A concrete server redesign reduces memory content per server because of price.
TrendForce cuts 3Q memory price growth back to single digits while Samsung DS core profit disappoints.

The fourth item is the real kill condition. Supply-driven forced reductions are different from price-driven voluntary memory-content cuts.

Evidence Classification

Facts: Samsung’s 2Q26 preliminary result, Samsung’s July 30 call schedule, Alphabet’s 2026 capex range and AI infrastructure capital raise, Microsoft’s $190B capex run-rate and $627B RPO, Meta’s $125-145B plan and $107B commitment increase, Amazon’s roughly $200B plan and AWS backlog, TrendForce’s 3Q price outlook, Micron’s FY3Q26 SCA disclosures.

Inferences: the 50/35/15 scenario map, the idea that +73% to +41% capex deceleration is already embedded, the view that Meta’s cloud report is more monetization than capex-cut signal, and the view that long-term contracts reduce memory cyclicality.

Speculation: SK Hynix contract upside without price ceilings, possible Microsoft financing, the exact timing of a 2027 digestion phase, and Meta’s cloud economics.

Blocked: verified 2027 consensus multiples for Samsung and SK Hynix, Marvell’s exact share in Trainium 3 and 4, SK Hynix contract details, company-level 2027 capex consensus distributions, memory supplier allocation by hyperscaler and some official late-July earnings dates.

The conclusion is practical: the odds favor a stronger memory thesis, but price already reflects part of that. The right way to trade the event is to update probabilities as evidence arrives, not to make a blind prediction before the calls.

H1 2026 Review: AI Infrastructure Bottlenecks Broadened, but the Market Did Not

Tue, 30 Jun 2026 22:35:00 +0900

Related context This is a half-year follow-up to Samsung 2Q26 preview, Micron FY3Q26 earnings, the NVIDIA elasticity framework for Samsung and SK Hynix, how hard it was to beat the KOSPI benchmark, the KOSPI 60-day disparity framework, the Real Money Flow framework, and the AI supercycle midgame note.

TL;DR

H1 2026 looked like a broad risk-on period from the index level. Under the surface, it was much narrower. Korea was a KOSPI megacap crowding market. The U.S. was an AI infrastructure market where leadership moved from megacap AI platforms into memory, storage, servers, semiconductor equipment and power. Neither market was a broad “everything went up” market.

The headline Korean index was extraordinary. KOSPI rose from 4,214.17 on December 30, 2025 to 8,476.48 on June 30, 2026, up 101.1%. KOSDAQ fell from 925.47 to 916.18, down 1.0%. Across roughly 2,700 Korean stocks, 856 rose and 1,788 fell. The median stock return was -12.6%. The index doubled, but the average stock was in a weak market.

The U.S. story was also more specific than the index suggests. The S&P 500 gained 9.2% and Nasdaq gained 11.1%, but the real story was AI infrastructure bottlenecks. SanDisk, Micron, Western Digital, Dell, Marvell, ARM, Applied Materials and Lam Research were much stronger than NVIDIA. The market was buying the places where the AI factory actually bottlenecks.

The H2 strategy is not simply “find cheap stocks.” The better rule is to buy proven bottlenecks with real revenue and contracts on pullbacks. Option-like equipment, gaming and biotech event names should be sized smaller. Continuous-revenue bottlenecks in memory, eSSD, MLCC, substrates, power, shipbuilding, defense and large biotech platforms deserve more attention.

One-Line Conclusion

H1 was risk-on, but it was not broad. Korea was dominated by Samsung Electronics and SK Hynix crowding, while the U.S. rewarded the physical bottlenecks beneath AI compute. H2 is about separating recurring-revenue bottlenecks from option-like stories and leveraged flow risk.

0. Data Basis and Reconciliation

The data basis is:

Area	Basis
Korea indices and stocks	June 30, 2026 close
U.S. indices and stocks	June 29, 2026 close
Korea flows	KR local DB, Kiwoom amount basis
U.S. stock returns	US local DB and yfinance adjusted close
Secondary inputs	Naver Finance, local screeners and session flow analysis

Some figures differ slightly by dataset and reference date. For example, the final KOSPI return is +101.1% using the June 30 close, while some local basket snapshots show +96.7%, and the June 23 interim review showed +90.4%. U.S. AI stock returns also differ depending on adjusted-close treatment and vendor handling.

This post reconciles those differences as follows.

Data type	How it is used
June 30 Korea close	Main H1 performance benchmark
June 29 U.S. close	Main U.S. H1 performance benchmark
June 23 interim snapshot	Supporting evidence for breadth and buyer structure
Adjusted-close versus local DB differences	Used directionally when conclusions are unchanged
Less-verified single figures	Treated as H2 watch items, not as core proof

The point is not false precision. The conclusion is stable: indices were strong, breadth was weak, and AI infrastructure returns moved from GPUs into memory, storage, servers, equipment and power.

1. Index Performance: KOSPI Doubled, KOSDAQ Fell

Market	Basis	H1 return
KOSPI	4,214.17 on 2025-12-30 to 8,476.48 on 2026-06-30	+101.1%
KOSDAQ	925.47 to 916.18	-1.0%
S&P 500	2025-12-31 to 2026-06-29	+9.2%
Nasdaq	2025-12-31 to 2026-06-29	+11.1%
Russell 2000	2025-12-31 to 2026-06-29	+21.3%
SMH semiconductor ETF	2025-12-31 to 2026-06-29	+75.5%
SOXX semiconductor ETF	2025-12-31 to 2026-06-29	+104.2%

At the index level, Korea was spectacular. KOSPI doubled in six months. But KOSDAQ fell. Within the same Korean market, large-cap AI infrastructure and growth stocks lived in different worlds.

The U.S. index story also understates the internal rotation. The S&P 500 and Nasdaq were up around low double digits, while semiconductor ETFs were up far more. SOXX gained 104.2%, close to the KOSPI’s gain, showing how concentrated the global AI hardware trade became.

2. Korea: The Index Doubled, the Median Stock Fell

Item	Value
Observed Korean stocks	roughly 2,700
H1 advancers	856
H1 decliners	1,788
Median stock return	-12.6%
KOSPI H1 return	+101.1%
KOSDAQ H1 return	-1.0%

This is an unusual combination. The index was in a historic bull market, while the median stock was weak. That is why the lived investor experience was so different from the index headline.

The main H1 leaders were:

Stock	H1 return
Samsung Electro-Mechanics	+756%
Jusung Engineering	+626%
SK Square	+361%
SK Hynix	+307%
Jeju Semiconductor	+294%
Daeduck Electronics	+226%
Simmtech	+201%
Samsung Electronics	+179%

The Korean market’s structure was:

Samsung Electronics and SK Hynix
→ Samsung Electro-Mechanics, MLCC and FC-BGA
→ substrates, PCB and semi equipment
→ power equipment, industrials and some financials

KOSDAQ, many biotech names, platforms and parts of secondary batteries lagged. Korea was strong, but most Korean stocks were not.

3. Korea Was Not a Foreign-Buying Market

YTD flow data on the local Kiwoom amount basis:

Investor group	YTD cumulative
Foreigners	-KRW 142T
Institutions	+KRW 45T
Retail	+KRW 77.7T
Financial investment	+KRW 75.4T
Investment trusts	-KRW 1.7T
Private funds	-KRW 3.8T
Pension funds and others	-KRW 7.0T

The key point is that Korea was not a simple foreign-buying market. Foreigners sold Samsung Electronics and SK Hynix heavily. Domestic retail, financial investment accounts and institutions absorbed the flow.

Stock	YTD return	Foreign cumulative flow
Samsung Electronics	around +160%	-KRW 72.6T
SK Hynix	around +291%	-KRW 57.1T
SK Square	around +333%	-KRW 7.6T
Samsung Electronics preferred	around +125%	-KRW 3.2T

Foreigners sold the core megacaps while those stocks rallied. That is not a normal foreign-led rally. It was a domestic absorption market.

Foreigners did not sell everything. They rotated into select bottlenecks and global-fund-friendly assets such as Samsung Electro-Mechanics, Doosan Enerbility, Samsung SDI, FADU, Hanwha Ocean and Sanil Electric. The right description is not “foreigners bought Korea.” It is “foreigners sold the largest memory megacaps, domestic liquidity absorbed them, and foreigners reallocated into select second-layer bottlenecks.”

4. The Hidden Buyer Was Financial Investment, Not Long-Only Real Money

Institutional buying was not the same as long-only accumulation. The strongest institutional subcategory was financial investment, not investment trusts, private funds or pension funds.

Flow category	Interpretation
Financial investment	ETF, derivatives, program and rebalance-linked flow
Investment trusts	More likely to represent long-only fund demand, but weak in H1
Private funds	Mixed thematic, long-short and event-driven flow, but weak cumulatively
Pension funds and others	Long-term real money, but net selling in H1
Retail	Key absorber of megacap AI infrastructure flow and leverage

This matters. Real money can sustain a trend. Financial-investment and program flow can be powerful, but it is more vulnerable to futures expiry, ETF caps, derivative positioning and leverage mechanics.

H1 Korea was both a fundamental rerating market and a derivatives/ETF market. Confusing the two leads to repeated mistakes.

5. KOSDAQ Lacked Buyers, Not Stories

KOSDAQ had plenty of stories: biotech, robotics, CPO, SOCAMM, semiconductor equipment, gaming and AI software. The index still lagged because the buyer base was weak.

The June 23 interim snapshot was already clear:

Index	YTD	From high
KOSPI	+90.4%	-10.0%
KOSDAQ	-5.7%	-27.3%

The final June 30 structure was the same: KOSPI +101.1%, KOSDAQ -1.0%.

KOSDAQ struggled because:

Retail capital was tied up absorbing Samsung Electronics, SK Hynix and large AI infrastructure names.
Institutional real money did not buy KOSDAQ broadly.
Foreign buying was narrow, focused on names such as FADU, Jeju Semiconductor and Simmtech.
Rates and FX pressure weighed on biotech and unprofitable growth multiples.

The condition for KOSDAQ recovery is:

KOSDAQ recovery =
retail capital returning
+ institutional real-money buying
+ Samsung Electronics and SK Hynix moving sideways rather than falling
+ KOSDAQ earnings revisions stabilizing

If Samsung and SK Hynix crash, money does not automatically rotate into KOSDAQ. Risk-off comes first. The best environment for KOSDAQ is not megacap collapse. It is megacap consolidation.

6. The U.S. Market Was Not Just NVIDIA

The real U.S. signal was not “AI went up.” It was:

The physical bottlenecks after the GPU outperformed the GPU leader.

Representative H1 return ranges:

Stock	H1 return range
SanDisk	+645-764%
Micron	+263-301%
Western Digital	+247-279%
Intel	+235-257%
Dell	+224-232%
Marvell	+211-227%
ARM	+200-214%
Applied Materials	+171%
AMD	+152%
Lam Research	+141%

Big Tech was mixed:

Stock	H1 return
Google	+13%
Amazon	+4%
Apple	+4%
NVIDIA	+3-5%
Broadcom	+7%
Meta	-15%
Microsoft	-22% to -23%
Oracle	-25%
Palantir	-31%

The market bought the points where the AI factory bottlenecks:

GPU
→ HBM and DRAM
→ eSSD and storage
→ servers and racks
→ optical and interconnect
→ semiconductor equipment
→ power and cooling

This is directly relevant to Korea. HBM maps to Samsung and SK Hynix. eSSD and NAND map to Samsung, FADU and Jeju Semiconductor. MLCC and FC-BGA map to Samsung Electro-Mechanics and substrate names. Power maps to Doosan Enerbility and electrical equipment. Equipment maps to HPSP, TES, VM and Jusung.

7. Macro Was Not a Pure Tailwind

Asset or indicator	H1 move
Dollar index	+3.1%
USD/KRW	+7.9%
U.S. 10-year yield	4.16% to 4.39%
WTI oil	+24.1%
Gold	-6.8%
Bitcoin	-33.2%

This was not a clean multiple-expansion macro environment. The dollar rose, the won weakened, U.S. yields rose and oil rose. Gold and Bitcoin fell.

Stocks rose anyway because the AI infrastructure earnings cycle overwhelmed macro pressure. That is important for H2. If rates, the dollar and oil all rise together again, the most crowded bottleneck stocks can suffer multiple compression even if demand remains strong.

The realistic risk is:

AI demand stays strong
+ earnings remain good
+ rates, dollar and oil remain firm
+ valuation and leveraged flows crack first

That is the same stock-elasticity problem discussed in the NVIDIA/Samsung/SK Hynix framework.

8. Non-Consensus Truth 1: Leadership Moved From Compute to Memory

The first non-consensus point is U.S. AI leadership. NVIDIA is still the center of AI. But incremental stock performance moved into Micron, SanDisk, Western Digital, Dell, Marvell and ARM.

That does not mean compute no longer matters. It means the marginal return moved down the stack.

Question	Meaning
Is this company a cost item in AI capex?	It can be pressured when customers optimize costs.
Is this company a bottleneck item in AI capex?	It can reprice if customers need supply.
Is this company only a narrative?	It is vulnerable during catalyst gaps.

Cost items get pressured. Bottlenecks rerate. Narratives require timing discipline.

9. Non-Consensus Truth 2: The Real Top Came Before the Index High

Breadth weakened before the index did. The June index high was driven by Samsung Electronics, SK Hynix and a small set of AI infrastructure names. KOSDAQ and the median stock had already weakened earlier.

Weak median stock
+ weak KOSDAQ
+ strong KOSPI memory megacaps
= index high and weak lived market at the same time

For H2, the key indicators are breadth, ADR, KOSDAQ/KOSPI relative strength, megacap memory trading-value share, and KOSDAQ retail/institutional joint buying.

10. Non-Consensus Truth 3: Multiples Moved Before Fundamentals Fully Arrived

Samsung, SK Hynix, Micron and Samsung Electro-Mechanics all had strong fundamentals. But stock returns moved faster than near-term earnings. The market changed the multiple.

Old frame:
memory = commodity cycle = peak-earnings discount

H1 frame:
memory = AI infrastructure bottleneck = supply shortage plus contract premium

The H2 risk is not just bad earnings. It is good earnings with lower multiples if the market starts treating memory as commodity again. Investors should watch EPS revision speed, beat magnitude and FY2027 visibility, not just the next quarterly beat.

11. Non-Consensus Truth 4: Leverage Plumbing Amplified Tail Risk

Korean memory megacaps were affected not only by fundamentals, but also by leverage plumbing: 2x and 3x Samsung/SK Hynix ETFs listed offshore, domestic leverage and inverse products, futures and program flow.

Simplified upside loop:

stock up
→ leveraged ETF rebalancing buy
→ futures and program support
→ more megacap upside
→ more momentum capital

Downside loop:

stock down
→ leveraged ETF rebalancing sell
→ futures and program pressure
→ retail margin and forced-selling risk
→ larger drawdown

Korea can therefore move more violently than Micron even when the fundamental direction is similar.

12. Non-Consensus Truth 5: Supply Moved From Strong Hands to Weaker Hands

The important question is not only who sold. It is who bought.

H1 saw foreigners and some institutional accounts reduce exposure in memory megacaps while retail and financial-investment accounts absorbed the flow.

Positive read	Negative read
Domestic liquidity is strong	Shares may have moved into weaker leveraged hands
KOSPI held despite foreign selling	Forced-selling risk rises if margin unwinds
Large-cap buyer base widened	It may also represent distribution from better-informed accounts

H2 monitoring should combine foreign cash flow, retail margin, deposits, ETF balances, and program flow.

13. Non-Consensus Truth 6: Foreigners Bought Globally Explainable Korean Assets

Foreigners did not buy every Korean theme. The stocks they bought were globally explainable.

Stock	YTD foreign net buying
Samsung Electro-Mechanics	+KRW 2.28T
Doosan Enerbility	+KRW 1.71T
Celltrion	+KRW 1.52T
Samsung SDI	+KRW 1.33T
Doosan	+KRW 1.11T
APR	+KRW 1.08T
Hyundai Rotem	+KRW 0.85T
LG Energy Solution	+KRW 0.85T
FADU	+KRW 0.80T
Hanwha Ocean	+KRW 0.76T

Common features:

power, nuclear, shipbuilding and defense
global quality industrials
large battery names
large biotech platforms
direct AI infrastructure bottlenecks

For H2, foreign long-only capital is more likely to buy assets it can explain in an investment committee than obscure small-cap stories.

14. Non-Consensus Truth 7: Revenue Pattern Beat Company Quality

In H1, the quality of the revenue pattern mattered more than abstract company quality.

Weak revenue-pattern names:

Type	Weakness
Equipment	Order gaps can create earnings gaps.
Gaming	Revenue visibility breaks around launch cycles.
Biotech event stocks	Trial and licensing gaps create narrative gaps.
Some robotics and AI software	Expectations move before revenue.

Strong revenue-pattern names:

Type	Strength
Memory	Pricing and volume improved together.
eSSD and storage	AI server storage bottleneck rerating.
MLCC and substrates	Server and package bottlenecks.
Power equipment	Data-center power demand plus policy capital.
Shipbuilding	Long backlog visibility.
Defense	Export backlog and government budgets.
Large biotech platforms	Contracts, production and platform value are explainable.

The H2 rule is simple:

Size option-like names smaller. Size recurring-revenue bottlenecks larger.

15. Bonus Hypothesis: Korea Became a Leveraged U.S. AI Proxy

Korea’s market identity changed in H1. Historically, Samsung and SK Hynix were proxies for EM, China and global IT inventory cycles. In H1 2026, the linkage to U.S. AI capex became much stronger.

Session analysis suggested a stronger relationship between U.S. prior-day AI/semi moves and Korea next-day memory megacaps, including an interim correlation estimate of 0.42. That figure needs formal revalidation, but the price action is consistent with the direction.

The issue is diversification. If Korea behaves like a leveraged U.S. AI capex proxy, domestic buffers weaken when the global AI factor cools.

Another watch item is capex. Session analysis classified large SK Hynix capex commitment as a peak-confidence checkpoint. The exact amount and scope should be rechecked against filings before it becomes a core argument. But historically, peak capex announcements often arrive when demand looks strongest.

Key H2 capex questions:

Question	Meaning
Are customer contracts supporting capex?	Lowers supply-glut risk.
Does capex impair FCF?	Earnings can look good while cash flow weakens.
Does supply arrive near demand peak?	Raises 2027 cycle risk.
Is capex HBM and AI-memory focused?	Must be separated from commodity DRAM oversupply risk.

16. H2 Checklist

Checkpoint	Why it matters
Samsung 2Q26 Core OP	Core profit power matters more than reported headline OP.
SK Hynix 2Q and 3Q HBM allocation	Determines whether the crowded winner can reaccelerate.
Micron FY4Q26 guide delivery	Tests whether U.S. AI memory premium can persist.
DRAM and NAND pricing	Confirms margin defense beyond HBM.
KOSDAQ retail and institutional buying	Needed for KOSDAQ recovery.
Financial-investment and program flow	Real plumbing behind megacap volatility.
USD/KRW, U.S. 10-year yields and oil	Put a ceiling on AI bottleneck multiples.
AI capex contracts and FCF	Shows whether demand is paid for by cash flow, not just narrative.

17. H2 Strategy

Area	Strategy
AI infrastructure core	Keep exposure, but buy pullbacks rather than chasing.
Memory megacaps	Samsung is the unreflected-alpha candidate. SK Hynix is the reacceleration candidate after pullbacks.
HBM lower-stack bottlenecks	Buy only where orders, margins and customer qualification are visible.
KOSDAQ	Do not buy the whole index first. Start with semiconductor equipment, test and profitable growth names.
Option-like equipment, gaming and biotech	Smaller sizing due to catalyst-gap risk.
Global quality cyclicals	Continue to monitor shipbuilding, defense, power, nuclear and large biotech platforms.

The H2 strategy in one sentence:

Keep AI infrastructure core exposure, reduce option-like positions, and shift toward recurring-revenue bottlenecks and Korean quality cyclicals that global capital can explain.

18. Final View

The crowd may remember H1 2026 as the best Korea/semiconductor supercycle year. That is not wrong. KOSPI doubled and memory, substrates, equipment and power leaders exploded.

But the more important story is:

leadership became extremely narrow
multiples moved before fundamentals fully arrived
foreigners sold the core megacaps while domestic liquidity absorbed them
financial-investment and ETF/program flows amplified the market
KOSDAQ lacked buyers, not stories
U.S. AI leadership moved from GPU into memory, storage, servers and power

The H2 question is not only “how much more upside is left?” The better questions are:

Does stock elasticity remain positive even after strong earnings?
Can domestic liquidity keep absorbing foreign selling?
Do KOSDAQ buyers return?
Do AI bottlenecks convert from narrative into revenue and contracts?

H1 was risk-on, but it was not broad. It was a compressed leadership market. H2 requires more precision: real bottlenecks over cheap stocks, recurring revenue over one-off catalysts, and Core OP plus FCF over headline earnings.

Data Note

Korea data: KR local DB, Kiwoom investor flow data and Naver Finance support as of the June 30, 2026 close.
U.S. data: US local DB and yfinance adjusted close as of the June 29, 2026 close.
Interim snapshots: June 23 KOSPI/KOSDAQ relative performance and session flow analysis.
U.S. June 30 regular-session results are not included.
Less-verified capex single figures are treated as H2 monitoring items rather than core proof.

Disclaimer: For research and information purposes only. Not investment advice. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.

Samsung 2Q26 Preview: Micron’s Surprise Has Been Erased, Core OP Is the Real Test

Mon, 29 Jun 2026 14:35:00 +0900

Related context This is a follow-up to Micron FY3Q26 earnings, the NVIDIA elasticity framework for Samsung and SK Hynix, Samsung-Hynix-Micron parity, Samsung shareholder return and DS bonus buyback flow, and Samsung versus SK Hynix in HBM4E 12-high.

TL;DR

The number that matters in Samsung Electronics’ 2Q26 preview is not simply reported operating profit. The market needs to separate reported OP from Core OP, defined as reported OP plus the catch-up accrual for performance-based compensation.

At current prices, Micron’s FY3Q26 surprise has almost disappeared from Korean memory equities. Micron retained an 8.0% gain versus its pre-earnings price as of June 26. Samsung Electronics is down 5.9% versus the June 24 close based on June 29 intraday data, while SK Hynix is down 0.5%. The June 25 reaction was real, but most of it was erased by June 26 and June 29.

Samsung’s 2Q26 DRAM pricing is not the downgrade factor. Conventional DRAM contract prices are estimated to be up 58-63% QoQ, while mobile LPDDR is up 70-83% QoQ. A blended Samsung DRAM ASP assumption in the high-50% QoQ range is therefore not aggressive.

The reasonable range for 2Q26 reported OP is KRW 87-92 trillion, with a midpoint of KRW 89.5 trillion. Adding back KRW 15-17 trillion of catch-up bonus accrual implies Core OP of roughly KRW 104-106 trillion. In other words, the correct reading is not “earnings fell from the KRW 100 trillion expectation to KRW 89 trillion.” It is closer to “the memory core business reached more than KRW 100 trillion of profit power, but Samsung is recognizing excess-profit sharing costs in 2Q.”

The rating is Watchlist / Conditional Buy. Conditions are clear: reported OP should be at least KRW 87 trillion, Core OP should be at least KRW 102 trillion, the downgrade should be explained by accruals rather than DRAM/NAND price or shipment weakness, and 3Q26 OP durability around KRW 100 trillion should remain credible.

Main Point

Micron’s surprise was reflected in Samsung and SK Hynix for one day on June 25, but it has mostly been erased. Samsung is back to a price that barely embeds the Micron read-through. The real test is not headline 2Q OP, but Core OP above KRW 100 trillion and the durability of 3Q memory earnings.

1. Price reaction: is Micron’s surprise still in Korea memory?

Micron reported FY3Q26 results after the U.S. close on June 24, 2026. For Korea, the clean baseline is the June 24 close.

Stock	Jun 24 close	Jun 25 close	Jun 26 close	Jun 29 intraday	Current vs Jun 24
Samsung Electronics	KRW 340,500	KRW 358,500	KRW 339,500	KRW 320,500	-5.9%
SK Hynix	KRW 2,580,000	KRW 2,917,000	KRW 2,673,000	KRW 2,568,000	-0.5%
Micron	USD 1,048.51	USD 1,213.56	USD 1,132.33	latest DB value Jun 26	+8.0%

Source: KII local daily price database for Korea and U.S. stocks; June 29 intraday Korea prices from Naver Finance API at 13:19 KST.

SK Hynix reacted most directly on June 25, rising 13.1%. But it has since nearly returned to its pre-event level. Samsung reacted 5.3% on June 25, then fell below the pre-event price. Micron still held an 8.0% gain as of June 26.

The conclusion is simple: at current prices, Korean memory has not retained Micron’s earnings read-through. It was reflected for one day, then erased by macro pressure, KOSPI overheat adjustment and company-specific uncertainty. The unreflected alpha is more visible in Samsung. SK Hynix is more of a pullback-and-reacceleration candidate.

2. What Micron confirmed

Micron FY3Q26 was a strong benchmark for the memory cycle. Revenue was USD 41.456 billion, non-GAAP EPS was USD 25.11, non-GAAP gross margin was 84.9%, and FY4Q26 guidance called for USD 50.0 billion revenue and USD 31.00 non-GAAP EPS. These are official company numbers. Micron FY3Q26 earnings release

The key read-throughs are:

Micron signal	Meaning for Samsung and SK Hynix
USD 41.456B revenue and 84.9% non-GAAP GM	Memory pricing, mix and shortage-driven operating leverage are very strong.
FY4Q guidance of USD 50B revenue and USD 31 EPS	Pricing and shipments are unlikely to collapse immediately after 2Q.
Data center revenue above USD 25B	AI memory demand is broader than HBM alone.
HBM4 and HBM4E roadmap	2027 allocation and customer qualification remain the next battleground.
Strategic Customer Agreements	Long-term contracts may reduce the old memory cyclicality discount.

This matters because Samsung now has to prove the same point with its own numbers: whether core memory profitability is already in the KRW 100 trillion range.

3. Reported OP versus Core OP

The working formula is:

Reported OP = accounting operating profit
Core OP = Reported OP + catch-up accrual for performance-based compensation

Base case:

Item	Base case
Reported OP	KRW 89.5T
Catch-up bonus accrual	KRW 15-17T
Core OP	KRW 104-106T

The important point is that the expected OP cut is not driven by weaker DRAM or NAND assumptions. In the research input, Kiwoom’s 2Q26 OP estimate fell from KRW 100T in May to KRW 89T on June 29. But over the same period, DRAM ASP assumptions moved from +55% QoQ to +58% QoQ, and NAND from +72% QoQ to +75% QoQ. The downgrade is therefore better explained by bonus accruals, non-memory weakness, DX cost pressure, and continued Foundry/S.LSI losses.

Item	May estimate	Late-June estimate	Meaning
2Q26 OP	KRW 100T	KRW 89T	Headline lower
DRAM ASP	+55% QoQ	+58% QoQ	Price assumption stronger
NAND ASP	+72% QoQ	+75% QoQ	Price assumption stronger
Downgrade driver	not applicable	Bonus, non-memory, DX, Foundry/S.LSI	Not a memory pricing problem

4. Official 1Q26 base

Samsung’s 1Q26 official results provide the base:

Item	1Q26
Company revenue	KRW 133.9T
Company operating profit	KRW 57.2T
DS revenue	KRW 81.7T
DS operating profit	KRW 53.7T

Source: Samsung Electronics 1Q26 results.

DS operating profit of KRW 53.7T is the key base for the 2Q bonus accrual discussion. If 1Q did not fully accrue the new profit-sharing structure, 2Q can carry a catch-up cost.

5. DRAM pricing is supportive

Product group	Estimated 2Q26 price trend	Interpretation
Conventional DRAM	+58-63% QoQ	Kiwoom’s +58% assumption is near the lower end of the market range.
PC DDR5	+43-48% QoQ	PC demand is not strong, but pricing still rose.
PC DDR4	+35-40% QoQ	Lower than DDR5, but still strong.
Mobile LPDDR4X	+70-75% QoQ	Mobile DRAM helps pull blended ASP higher.
Mobile LPDDR5X	+78-83% QoQ	Positive for Samsung’s blended DRAM ASP.
Consumer DRAM	+45-50% QoQ	Customer price resistance exists, but pricing is still up.

The core conclusion is that a high-50% QoQ blended DRAM ASP assumption is reasonable. If reported OP comes in near KRW 89T, it should not be interpreted as DRAM price weakness unless Samsung’s own disclosures say otherwise.

6. Bonus accrual treatment

The existence of DS performance-based compensation is not new. The unsettled question is scale, timing and disclosure quality.

Analytical purpose	How to treat bonus accrual
2Q earnings quality	Add it back to estimate Core OP
3Q run-rate	Exclude 1Q catch-up, include normal ongoing bonus expense
FY26 EPS	Include it
FY27 valuation	Include it if recurring
Shareholder return and FCF	Include it
Short-term event trade	Watch both reported OP and Core OP

The two mistakes to avoid are: treating all bonus expense as a one-off, or treating all of it as structural deterioration. The catch-up portion matters for 2Q interpretation. The recurring profit-sharing structure matters for FY26/FY27 EPS and valuation.

7. Sell-side OP distribution

Source	2Q26 OP estimate	Interpretation
BNK	KRW 77.7T	Most conservative bonus retroactive accrual assumption
Samsung Securities	KRW 81.3T	DRAM ASP +46%, NAND ASP +65%, DS bonus reflected
Mirae Asset	KRW 84.1T	Strong pre-accrual cost scenario near KRW 35T
FnGuide consensus	KRW 86.2T	Early-June average, may not fully reflect latest prices and accruals
Kiwoom Jun 29	KRW 89.0T	Higher DRAM/NAND price assumptions, but bonus/non-memory costs reflected
KB	KRW 90.0-90.2T	Memory price up more than 50%, DS OP in the high-KRW 80T range
Daishin	KRW 91.0T	Memory price strength with partial cost reflection
Hana	KRW 92.0T	Stronger pricing and shipment assumptions

Investment framing:

Reported OP range	Interpretation
Below KRW 84T	Bear case. Costs and non-memory drag are very large.
KRW 87-92T	Base cluster. This is the realistic range.
Above KRW 94T	Bull case. Costs are limited or price/shipments are even stronger.

8. Scenario model

Item	Bear	Base	Bull
DRAM blended ASP	+50%	+56-58%	+62-63%
DRAM bit growth	+4%	+5%	+6%
FX effect	+1.0%	+1.5%	+2.0%
DRAM revenue	KRW 86-88T	KRW 91-93T	KRW 95-97T
DRAM OP margin	71-73%	73-75%	76-78%
DRAM OP	KRW 62-64T	KRW 67-70T	KRW 72-75T
NAND OP	KRW 18-20T	KRW 20-22T	KRW 22-24T
Foundry/S.LSI OP	-KRW 1.5T to -KRW 1.0T	-KRW 1.0T to 0	0 to KRW 0.5T
Display/DX/Harman combined	-KRW 0.5T to KRW 1.0T	0 to KRW 1.0T	KRW 1-2T
Reported OP	KRW 82-86T	KRW 87-92T	KRW 94-98T

Base midpoint:

Item	Midpoint
Revenue	about KRW 181T
Reported OP	about KRW 89.5T
OPM	about 49.4%
Core OP	KRW 104-106T

9. Sensitivities

Variable	OP sensitivity
DRAM ASP +1ppt	+KRW 0.4-0.5T
DRAM bit growth +1ppt	+KRW 0.4-0.5T
NAND ASP +1ppt	+KRW 0.1-0.2T
Bonus/accrual +KRW 1T	-KRW 1.0T
DX cost +KRW 1T	-KRW 1.0T
Foundry/S.LSI loss +KRW 1T	-KRW 1.0T

The OP downgrade from KRW 100T to KRW 89T despite stronger DRAM/NAND price assumptions implies roughly KRW 12-13T of cost, accrual and non-memory drag.

10. Reported OP and Core OP matrix

Reported OP	Bonus accrual	Core OP	Interpretation	Action
KRW 92T or higher	about KRW 15T	KRW 107T or higher	Clear beat	Buy possible
KRW 89-92T	about KRW 15T	KRW 104-107T	In-line positive	Buy weakness
KRW 87-89T	about KRW 15T	KRW 102-104T	Consensus area	Staged entry possible
KRW 84-87T	about KRW 15T	KRW 99-102T	Weak miss	Needs explanation
KRW 84-87T	KRW 25-35T	KRW 109-122T	Headline miss, but neutral to positive if full-year pre-accrual	Verify cost nature
KRW 80-84T	about KRW 15T	KRW 95-99T	Cannot be explained by bonus alone	Wait or reduce
Below KRW 80T	any	not meaningful	Thesis damage likely	Avoid

11. Catalysts

Event	Timing	Importance	Watch item
Samsung 2Q26 preliminary results	likely early July 2026	Very high	Reported OP at least KRW 87T
Samsung 2Q26 final results and call	likely late July 2026	Very high	Bonus accrual, DS core profit, 3Q outlook
3Q26 DRAM/NAND contract prices	July-August 2026	Very high	Flat to up QoQ
HBM4E, SOCAMM2, eSSD customer comments	2H26	High	Evidence of share gain
Bonus-share buyback disclosure	TBD	Medium-high	Distinguish compensation buyback from cancellation buyback
Micron FY4Q26 results	likely late September 2026	High	Whether USD 50B revenue and USD 31 EPS are delivered

12. Valuation

The Kiwoom June 29 estimates summarized in the research input are:

Item	2026F	2027F
Revenue	KRW 723.6T	KRW 820.1T
Operating profit	KRW 372.9T	KRW 421.3T
EPS	KRW 43,429	KRW 48,177
PER	7.8x	7.0x
PBR	3.5x	2.3x
ROE	53.9%	39.6%

A KRW 430,000 target price implies 9.9x 2026F EPS and 8.9x 2027F EPS. This is valid only if 3Q26 OP remains above KRW 100T, 2027 EPS stays near KRW 48,000, bonus and stock compensation are already embedded in EPS, HBM4/eSSD contributions are visible, and DRAM/NAND prices do not fall sharply.

13. Red Team and invalidation

Risk	What would invalidate the thesis
Macro failure	Rates and credit spreads slow AI capex, hyperscaler capex guidance falls, AI server orders are delayed.
Micro failure	Samsung does not gain HBM4/eSSD share, HBM4 qualification is delayed, Foundry/S.LSI losses do not shrink.
Accounting failure	Reported OP falls below KRW 84T without enough bonus/accrual explanation, or Core OP falls below KRW 100T.
Capital allocation failure	Compensation buybacks are mistaken for shareholder return, while recurring profit-sharing erodes FY27 EPS and FCF.

Hard invalidation conditions:

Condition	Meaning
Reported OP below KRW 84T and not explained by at least KRW 25T of pre-accrual	More than timing noise
Core OP below KRW 100T	Core memory profit power weaker than expected
3Q26 OP outlook below KRW 100T	Peak-out risk
3Q DRAM/NAND prices fall QoQ	Pricing thesis damaged
No HBM4/eSSD share gains	2027 EPS and multiple expansion limited
Foundry/S.LSI loss persists or widens	Samsung discount persists

Final view

At current prices, Micron’s surprise is almost gone from Samsung and SK Hynix. Samsung has returned to a price that barely embeds the Micron read-through. SK Hynix has almost returned to pre-event levels after a strong June 25 reaction.

Samsung’s opportunity is now specific: if 2Q preliminary results show that the headline OP decline is mostly bonus-accrual timing, while Core OP is above KRW 100T and 3Q durability remains intact, the stock can reprice.

The relative conclusion is straightforward:

Stock	Interpretation
Samsung Electronics	Unreflected alpha if Core OP and 3Q durability are confirmed
SK Hynix	Pullback and reacceleration alpha, not a “cheap” stock
Micron	Cleanest proof of the AI memory cycle, but already repriced

One sentence: Micron’s earnings were reflected in Korean memory names for one day, but most of that effect has been erased. For Samsung, the next battle is not headline 2Q OP but Core OP above KRW 100 trillion and 3Q durability.

How To Express The Samsung Electronics Catch-Up Trade: Preferred Share Discount And Samsung C&T NAV Gap

Mon, 22 Jun 2026 23:10:00 +0900

Samsung Electronics common has already re-rated sharply on HBM, memory-cycle recovery, and free-cash-flow return expectations. The question is how to express the same thesis more efficiently.

TL;DR

Samsung Electronics preferred is the cleaner catch-up expression. It lagged the common by 37.8 percentage points YTD and traded at a 36.6% discount on 22 June 2026.
The discount is extreme: 36.6% versus a 19.4% three-year average, 18.1% median, +3.11 z-score, and 99.1st percentile.
Samsung C&T is not a one-for-one Samsung Electronics proxy. It is a NAV and affiliate-dividend story.
ETF cap spillover is secondary. A simple redistribution assumption implies only about KRW 24.6bn into Samsung Electronics preferred and KRW 46.0bn into Samsung C&T.

Key Data

Stock	2 Jan Close	22 Jun Close	YTD	Versus Samsung Electronics
Samsung Electronics	KRW 128,500	KRW 353,500	+175.1%	Baseline
Samsung Electronics preferred	KRW 94,400	KRW 224,000	+137.3%	-37.8pp
Samsung C&T	KRW 245,000	KRW 520,000	+112.2%	-62.9pp

Metric	Samsung Electronics	Samsung Electronics Preferred	Samsung C&T
20D return	+18.0%	+19.3%	+24.3%
60D return	+86.3%	+67.0%	+84.7%
Correlation with Samsung Electronics	1.000	0.949	0.779
Beta to Samsung Electronics	1.000	0.912	0.848

Preferred Discount

Metric	Value
Preferred/common ratio	63.37%
Current discount	36.63%
Three-year average discount	19.43%
Three-year median discount	18.14%
Current z-score	+3.11
Three-year percentile	99.1%

Preferred/Common Ratio	Implied Preferred Price	Upside
Current 63.4%	KRW 224,000	Baseline
70.0%	KRW 247,450	+10.5%
75.0%	KRW 265,125	+18.4%
80.0%	KRW 282,800	+26.3%
85.0%	KRW 300,475	+34.1%

Samsung C&T NAV

Samsung C&T is calculated to hold 298,818,100 Samsung Electronics shares. At KRW 353,500, that stake is worth about KRW 105.6tn. Samsung C&T’s estimated market cap is about KRW 84.3tn.

Metric	Value
Samsung Electronics stake value	About KRW 105.6tn
Samsung C&T estimated market cap	About KRW 84.3tn
YTD increase in Samsung Electronics stake value	About KRW 67.2tn
YTD increase in Samsung C&T market cap	About KRW 44.6tn
Reflection rate	About 66.3%

Reflection Rate	Implied Samsung C&T Price
50%	KRW 452,000
60%	KRW 494,000
70%	KRW 535,000
80%	KRW 577,000
100%	KRW 660,000

Flow Check

Stock	1D Foreign	1D Institution	1D Real Money	5D Foreign	5D Institution	5D Real Money
Samsung Electronics	-4,418	-1,164	+617	-11,499	+7,449	+2,568
Samsung Electronics preferred	+149	+389	+122	-270	-142	-512
Samsung C&T	-627	+401	+387	-936	+450	+597

Action Framework

Stock	Stance	Confirmation
Samsung Electronics preferred	Conditional Buy	Discount remains around 34-37%, Samsung Electronics trend intact, five-day foreign/institutional flow turns positive
Samsung C&T	Watchlist / Pullback Buy	Holds KRW 500,000, foreign/program selling slows, or breaks above KRW 535,000
Samsung Electronics common	Core Hold / cautious add	Foreign selling slows, program selling eases, 20-day moving average holds

The key conclusion is that the next Samsung Electronics catch-up alpha is more likely to come from preferred-share discount normalization than from ETF cap spillover.

Sources: Samsung Electronics shareholder return, Samsung C&T shareholder return, KCMI cap background, Korea Herald on KOSPI200 cap abolishment, KII internal snapshots as of 22 June 2026.

Disclaimer: For research and information purposes only. Not investment advice.

Samsung Electronics Is Not a Dividend Stock: FCF Return and the DS Bonus Buyback Flow

Mon, 22 Jun 2026 17:45:00 +0900

Related reads: Samsung DS bonus treasury-share math · Samsung HBM4E 12-high sample · Jensen Huang’s HBM4 three-vendor comment · Samsung-Hynix-Micron parity · AI HBM hub

TL;DR

Samsung Electronics’ official 2024-2026 shareholder-return policy is annual regular dividends of KRW 9.8 trillion plus 50% of free cash flow over the three-year period.
On KII’s local consensus database, FY2026E net income is about KRW 300.2 trillion. Even if only 50% of net income converts into FCF, the 50% FCF return pool is about KRW 75.1 trillion, or 3.34% of combined common and preferred market capitalization.
The DS performance-bonus treasury-share purchase should not be counted as pure shareholder return because the shares are not cancelled. But it can still create roughly KRW 20-22 trillion of company purchase demand, around 1% of common market cap.
Samsung is therefore not a high-dividend stock. The better description is an AI memory mega-cap with an FCF return option and a separate DS bonus-related buyback flow.
The key checkpoints are FY2026 FCF conversion, board-level additional return decisions, DS bonus buyback approval, and HBM4/HBM4E customer qualification and shipments.

1. Why This Matters

The common mistake is to analyze Samsung only through dividend yield.

That screen is not flattering. Samsung’s regular dividend commitment is KRW 9.8 trillion per year. Against roughly KRW 2,246.4 trillion of combined common and preferred market value as of June 22, 2026, that is only about 0.44%.

But the current policy is not just a fixed dividend. Samsung says it will return 50% of FCF for FY2024-2026. When AI memory earnings expand sharply, the absolute size of that FCF-linked return option changes.

The DS bonus treasury-share issue is a separate layer. It is not a cancellation buyback, so it should not be added to shareholder-return yield. Yet it can still matter as company purchase demand and possible free-float absorption.

Bucket	Shareholder return?	What investors should track
Regular dividend	Yes	KRW 9.8tn annually, stable but low headline yield
Special dividend	Yes	Possible if surplus FCF is meaningful
Buyback and cancellation	Yes	Direct EPS/BPS accretion
DS bonus treasury-share purchase	Conservatively no	Not cancelled, but creates purchase flow and partial lockup

2. Verified Baseline

Samsung’s official IR page states the FY2024-2026 return policy: annual regular dividend of KRW 9.8 trillion and total shareholder return of 50% of FCF. It also says the company may consider early capital return beyond regular dividends if meaningful surplus resources are expected at year-end. The 1Q26 dividend was KRW 372 per common and preferred share, with total payout of KRW 2.45 trillion. Source: Samsung IR.

Samsung’s 1Q26 earnings presentation shows consolidated revenue of KRW 133.9 trillion and operating profit of KRW 57.2 trillion. DS revenue was KRW 81.7 trillion and DS operating profit was KRW 53.7 trillion, about 93.9% of consolidated operating profit. Samsung also cited high-value-added AI product demand, higher ASPs due to supply shortages, and mass product sales of HBM4 and SOCAMM2 for NVIDIA’s Vera Rubin platform. Source: Samsung 1Q26 earnings presentation.

KII local consensus and Naver Finance data as of June 22, 2026:

Item	Value
Common share price	KRW 353,500
Common market cap	KRW 2,066.7tn
Preferred share price	KRW 224,000
Preferred market cap	KRW 179.7tn
Combined market cap	KRW 2,246.4tn
FY2026E revenue	KRW 700.5tn
FY2026E operating profit	KRW 363.2tn
FY2026E net income	KRW 300.2tn
FY2026E EPS	KRW 44,426
FY2026E PER	7.97x
FY2026E ROE	52.64%

The critical number is not just 7.97x PER. It is KRW 300.2 trillion of expected net income. At this profit scale, the same FCF-return policy becomes a much larger option.

3. FCF Return Scenarios

The formula is simple:

FCF = FY2026E net income × FCF / net income ratio
Total shareholder-return pool = FCF × 50%
Return yield = pool / combined common and preferred market cap

Using FY2026E net income of KRW 300.2 trillion and combined market cap of KRW 2,246.4 trillion:

FCF / net income	Estimated FCF	50% FCF return pool	% of combined market cap
50%	KRW 150.1tn	KRW 75.1tn	3.34%
60%	KRW 180.1tn	KRW 90.1tn	4.01%
75%	KRW 225.2tn	KRW 112.6tn	5.01%
100%	KRW 300.2tn	KRW 150.1tn	6.68%

Even the conservative 50% FCF conversion case produces a KRW 75.1 trillion return pool. That is very different from the 0.44% headline regular-dividend yield.

This does not mean the full amount becomes incremental dividend or cancellation. The regular dividend is already part of the return pool, and Samsung may prioritize capex, strategic investment, M&A, or cash buffers. Still, the absolute option value of the 50% FCF policy rises sharply when AI memory earnings scale up.

Yonhap reported that Samsung and labor representatives agreed to introduce a DS special management performance bonus funded by 10.5% of selected business performance. The bonus would be paid after tax entirely in treasury shares, with one-third immediately saleable, one-third locked for one year, and one-third locked for two years. Source: Yonhap.

MoneyToday estimated that Samsung could purchase about KRW 20.85 trillion of shares for this bonus, using FY2026 operating-profit consensus, a DS profit share of around 94%, a 10.5% bonus pool, and a 40% tax assumption. Source: MoneyToday.

Updating that math with KII’s June 22 consensus:

Estimated DS OP = FY2026E consolidated OP × 1Q26 DS OP share
DS bonus pool = estimated DS OP × 10.5%
After-tax share grant amount = bonus pool × (1 - tax assumption)

Item	Value
FY2026E consolidated OP	KRW 363.2tn
1Q26 DS OP share	93.9%
Estimated DS OP	KRW 341.0tn
Bonus pool rate	10.5%
Bonus pool	KRW 35.8tn
Tax assumption	40%
After-tax share purchase amount	KRW 21.5tn
% of common market cap	1.04%
% of combined market cap	0.96%

Scenario table:

DS OP assumption	10.5% pool	After-tax 60% grant	% of common market cap
KRW 300tn	KRW 31.5tn	KRW 18.9tn	0.91%
KRW 331tn	KRW 34.8tn	KRW 20.9tn	1.01%
KRW 341tn	KRW 35.8tn	KRW 21.5tn	1.04%
KRW 360tn	KRW 37.8tn	KRW 22.7tn	1.10%

For Samsung’s scale, 1% of market cap may look small. But a single-stock company purchase flow of KRW 20-22 trillion is difficult to ignore, especially if two-thirds of granted shares are locked for one to two years.

5. Do Not Double Count

The right way to combine the pieces:

Item	Size	% of combined market cap	Interpretation
Regular dividend	KRW 9.8tn annually	0.44%	Stable, low headline yield
FCF 50% return pool, assuming FCF/NI 50%	KRW 75.1tn	3.34%	Official return-policy potential, includes regular dividend
DS bonus after-tax treasury-share flow	KRW 20-22tn	about 0.9-1.0%	Not cancellation, but potential company purchase demand

The regular dividend is part of the FCF return pool. The DS bonus treasury-share purchase should be kept outside pure shareholder-return math. But it should still be tracked as a market-flow variable.

6. What the Market May Miss

The skeptical view is familiar: Samsung has lower HBM purity than SK Hynix, a more complex business mix, low headline dividend yield, and a foundry discount.

The variant view is also becoming clearer: if FY2026 net income is around KRW 300 trillion, the 50% FCF-return policy becomes much larger in absolute terms. HBM4E progress could narrow the HBM discount. The DS bonus stock flow can add company purchase demand. And stock-based compensation partially aligns DS employees with shareholders.

The Samsung thesis is not simply “cheaper than SK Hynix.” It is:

AI memory profit scale rises
 -> absolute value of the 50% FCF return policy rises
 -> special return / cancellation / buyback options improve
 -> DS bonus treasury-share purchase flow adds market demand
 -> HBM catch-up and shareholder-return optionality can be priced together

7. Samsung vs SK Hynix

SK Hynix has clearer pure HBM exposure and deserves a leadership premium. Samsung is different. It has lower HBM purity but much larger earnings scale, an official FCF-return policy, and a separate DS bonus-related purchase flow.

Item	Samsung Electronics	SK Hynix
Main thesis	HBM catch-up, DRAM/NAND leverage, FCF return, buyback flow	HBM leadership and high-margin AI memory exposure
Strength	Earnings scale, cash flow, return policy, broad optionality	Customer lock-in, HBM share, pure AI memory premium
Weakness	Complex mix, HBM catch-up still needs proof, foundry discount	High expectations, possible dilution of exclusivity premium
Rerating trigger	FCF conversion, additional returns, HBM4E qualification, DS buyback approval	HBM4 allocation defense, ASP discipline, long-term customer contracts

8. Decision Framework

Positive checkpoints:

Checkpoint	Meaning
FY2026E OP and net income revisions continue upward	Larger potential return pool
FCF conversion of 50% or more becomes visible	Confirms actual power of the return policy
Board announces additional buyback/cancellation or special return	Direct rerating trigger
Board approves DS bonus treasury-share purchase and schedule	Flow trigger
HBM4/HBM4E qualification and shipments expand	HBM discount narrows
DS margin stays high	Sustains both profit and bonus pool

Risks:

Risk	Meaning
HBM4/HBM4E qualification delay	Samsung’s HBM discount persists
FCF materially below net income	50% FCF return pool shrinks
Capex, M&A, strategic investment, or cash buffers take priority	Additional return delayed
DS bonus buyback is mistaken for cancellation	Market overcounts the effect
Employee selling after lockup	Future overhang
AI capex slowdown, higher rates, stronger dollar	Memory multiple pressure

Conclusion

Samsung Electronics is not a high-dividend stock. That remains true.

But the 2026 setup requires a different lens. AI memory earnings have lifted expected net income to roughly KRW 300 trillion. Samsung has an official policy to return 50% of FCF. Separately, the DS bonus structure may create KRW 20-22 trillion of treasury-share purchase flow.

The cleanest sentence is this:

The DS bonus treasury-share purchase is not cancellation-based shareholder return. But it can still operate as company purchase demand, while Samsung’s official 50% FCF return policy becomes more valuable as AI memory earnings scale.

The next checks are not headlines. They are 2Q26 FCF, DS operating profit, HBM4E qualification and shipments, board-level return decisions, and the actual DS bonus buyback schedule.

Sources And Coverage

Samsung shareholder-return policy: Samsung IR
Samsung 1Q26 results and DS data: Samsung 1Q26 earnings presentation
DS special bonus agreement reporting: Yonhap
DS bonus treasury-share purchase estimate: MoneyToday
Price and consensus: KII local consensus database and Naver Finance, collected June 22, 2026