When is the right time to buy - SK Hynix and Samsung Electronics Response Report
Where to buy — After passing Tuesday's forced selling, from SK Hynix 1.68~1.79 million · Samsung 217~232 thousand Target prices remain unchanged (SK Hynix 2.6 million +41% · Samsung 330 thousand +29%)
Read as text
ACTION REPORT · 2026.07.18
When is the
time to buy
SK hynix · Samsung Electronics Response Report
000660 · KOSPI | 005930 · KOSPI
The stocks are verified; all that remains are the entry point and timing
After Tuesday's margin call liquidations pass: Hynix 1.68M~1.79M · Samsung 217K~232K — Target prices remain unchanged (Hynix 2.6M +41% · Samsung 330K +29%)
HBM4 Delay CXMT Listing NAND Super-cycle Margin Call Red Zone Kimi K3 Priced-in
Written on Saturday, July 18, 2026 · Target Audience: After Monday, July 20
Price Basis: Korea 7/16(Thu) Close · US/Japan 7/17(Fri) Close (Korea market closed on 7/17)
Confirming NAND Super-cycle · Real Cause of Sell-off · Kimi K3 Verification · Reading Friday's US Market · Supply/Demand Dashboard · Target Prices · Practical Roadmap
Onion (Macro/Liquidity) · Acorn (Sector/Value) · Peanut (Fundamentals/Risk) · Carrot (Charts/Supply & Demand/Sentiment) · Goguma (Synthesis)
Executive Summary
Conclusion and Action First — Detailed Rationale Follows
Action Summary for the Next Trading Day
| Item | Conclusion |
|---|---|
| 7/20(Mon)~7/21(Tue) | New Buys Prohibited. Period where margin calls from the 7/16 crash are executed at the 7/21 market open — Do not get trampled by forced liquidations |
| Exceptions | If a sharp drop on Monday triggers entry into 10% allocation if entering 1,762,000~1,790,000, 20% allocation if entering 1,678,000~1,730,000Conditional execution only (levels where the risk-reward ratio justifies the exception) |
| Weight of Buying | Hynix 1,678,000~1,790,000, Samsung 217,000~232,000 — Allocate capital toward the downside, not the upside |
| Stop-Loss Structure | Close below 1,678,000 → Reduce half / 1,600,000 → Trim 1/3 / 1,400,000 → Full Stop-loss |
| Cash Rule | Do not exhaust more than 30% of cash until SK hynix 2Q earnings on 7/29 |
| Priority | SK hynix 1st · Samsung 2nd (approx. 60:40). Remove 'Conditional' tag from Samsung |
| Target Price | Hynix Base 2,600,000 / Samsung Base 330,000 — Maintain. Only Bear case lowered (see Changelog below) |
Bottom Line. What the market sold last week was ① Hynix's HBM4 schedule ② CXMT shaking the DRAM oligopoly ③ Valuation multiples of NAND 'stocks' that had surged dozens of times These three things. One thing must be clearly distinguished here — the NAND industry cycle has not collapsed. NAND contract prices are in a super-cycle, rising over 70% per quarter, and supply shortages will continue until 2027~28. What collapsed was not the industry cycle, but the multiples of stocks whose prices had already front-run that boom several times over. Therefore, there is no reason to lower target prices, but the timing of purchase should be delayed and the entry price range should be lowered — the stocks to buy are correct; we are just changing the entry points.
Changelog
What Has Changed — History of Judgment Revisions
Additional investigation confirmed that six of the team's previous judgments were different from facts or inaccurate. As these items directly affect investment decisions, they are explicitly recorded.
Fact Correction (Reversed by Investigation)
| Item | Previous Judgment | Measured/Verified Result | Impact |
|---|---|---|---|
| NAND Nature | "NAND liquidation = Direct negative for Samsung" | Factual error. NAND contract price 2Q26 +70~75% QoQ, Samsung 2026 NAND price outlook +91% YoY. Samsung's NAND profit contribution **20~25%**is higher than SK hynix (17~21%), and the profit pool is 1.5~1.6x — NAND is a positive pillar | Samsung Re-rating |
| Divergence | "Only NAND is weak, DRAM/HBM are bottoming out" | Full-scale decline on a 1-month trend. Micron -16.8%, SanDisk -32.0%, Seagate -23.6%, WDC -29.9% — The difference is in the degree of damage, not the direction. However, differentiation was observed in Friday's candle (Chapter 4) | Remove Overconfidence |
| ADR Interpretation | "ADR +1.13% = Bottoming out" | Cherry-picking. Previous session **-13.69%**was missing, and Friday's candle closed at 38.8% of the range from the high = Pullback after an attempted rebound | Revised gap scenario |
| Supply and Demand Dashboard | "Forced liquidation 1.8%, Credit 34.7T — 1st liquidation exhausted" | Digit error. Measured forced liquidation ratio 10.2%(Exceeded Team Red threshold of 9%), Credit loans 36.6T(Above warning line of 35T), Deposits 139.7T → 107.1T (-23.3%) | Postponed purchase |
| Cause of the crash | "NAND-driven memory fear" | It was elsewhere. The trigger for Hynix's -15% on 7/13 was HBM4 shipment delay report, and for -11.5% on 7/16, it was CXMT listing and a data center delay report | Replacement of monitoring targets |
| Bain exit | "Liquidation of an 8-year fund lifecycle" | 44% → February block deal → June 14% → 0% complete liquidation. Timing was a few days after the all-time high — not a trim, but a full exit | Heightened vigilance |
Changes in judgment and rules
| Item | Previous | Revised | Reason |
|---|---|---|---|
| Hynix Bear | 1,650,000 | 1,450,000 Downrated | The previous cushion (-10.4%) was smaller than the single-day drop (-11.5%) on 7/16, so it wasn't a valid scenario |
| Samsung Bear | 235,000 | 215,000 Downrated | 235,000 lacked support due to the volume profile 0.00%. 61.8% retracement + volume zone is 217,000~222,000 |
| Base, Bull, Super Bull | Hynix 2.6M/3.15M/3.7M Samsung 330K/380K/450K | All maintained Maintain | NAND upward adjustment (+6-7%) and CXMT competition profit discount (-4-5%) arithmetically offset → Net effect +1.4-2.1% |
| Hynix 1st Buy | 20% at 1.75M~1.84M | Reduced to 10% + deferred to after 7/22 Changed | Current price is already within the band, and 1,825,000~1,975,000 is the largest volume wall for 2026(2.2~2.6% per section) |
| Hynix Core Weight | Core 30% at 1.69M~1.73M | 40% at 1,678,000~1,730,000 Downrated | Measured risk-reward ratio — 3.1:1 for 1,842,000 entry vs 7.4:1 for 1,678,000 entry11.8:1 |
| Samsung Buy Level | 240K 30% / 235K 40% | 247,000 20% / 228~232,000 30% / 217~222,000 50% Changed | 230,000~250,000 is a volume vacuum (0.00%) — the previous plan allocated 70% of funds to a zone with no support basis |
| Samsung Hold Rule | "Upgrade to Hold if NAND spot prices drop" | Discarded → "Hold upon confirmation of CXMT server DRAM certification/hyperscaler adoption" Replaced | Since NAND is a positive factor for Samsung, the trigger was monitoring the wrong area |
| Stop Loss | Full exit if closing price is 1,600,000 for 2 days | Two-tiered: Reduce half if 1,678,000 is breached / trim 1/3 at 1,600,000 / full exit at 1,400,000 Redesigned | In an environment with 8.38% daily volatility (133% annualized), a single -13% stop-loss line gets caught in noise |
| Liquidity Rules | "Do not exhaust 50% before 7/27" | "Do not exhaust 30% before 7/29 earnings" Strengthened | The real gateway is Q2 earnings on 7/29 (HBM4 confirmation) |
| HBF Reflection | Used as basis for Super Bull | Cannot be reflected in target price, only as a narrative downgrade | Primary source demand will around 2030 materialize, lack of TAM estimates, non-exclusive, competitors leading, and the double-edged nature of HBM ASP cannibalization |
| 7/27 Risk | Kimi K3 open-weight = biggest variable | Weight downrated — the sell-off already occurred on Friday. Instead, the CXMT listing on the same day is a bigger variable Replaced | Confirmed that Kimi K3 was the trigger for Friday's US sell-off |
| Priority | Hynix 1st / Samsung 2nd (after NAND confirmation) | Hynix 1st / Samsung parallel 2nd, weight gap 70:30 → 60:40 Adjusted | Removed Samsung's conditional tag as it was based on a factual error. However, it's not a reversal since CXMT exposure is Samsung 51% > Hynix 39% |
First Page
NAND is in a super-boom — what was liquidated was the multiple, not the industry condition
Under the headline 'Memory Winter,' NAND prices were rising by over 70% per quarter. Interpreting NAND as a negative factor is the most common misreading of this phase, and resolving this contradiction is the starting point for judgment.
Physical Data — NAND is in a Super-Boom
| Indicator | Measured Value | Meaning |
|---|---|---|
| NAND Contract Price Q1 | +55~60% QoQ | Cumulative increase of over 100% in the first half. There is no evidence of price decline anywhere. |
| NAND Contract Price Q2 | +70~75% QoQ | |
| NOR·SLC Cumulative H1 | Over +100% | |
| Samsung 2026 NAND Price Outlook | +91% YoY (Upward revision from +50%) | Profit estimates are actually rising |
| SK Hynix Q2 NAND ASP | +52% | Average Selling Price Surging |
| 2026 Demand vs. Supply | Demand +20~22% vs. Supply Bit +15~17% | Structural Shortage |
| Shortage Expected to Persist | Until 2027~2028 | Difficult to resolve until new fabs operate |
| Shift in Demand Structure | AI Enterprise SSDs surpass smartphone consumption for the first time this year | Sluggish smartphone demand (-13%) offset by high-value reallocation |
Then why did Kioxia and SanDisk collapse?
The answer lies not in the industry conditions, but in Stock Price and Leverage. Kioxia skyrocketed +5,400% from its 2025 low, briefly becoming the #1 market cap on the Tokyo market, and its margin buying balance hit a record high of approximately 870 billion yen. Three factors overlapped here — forced liquidation in Japan due to margin calls, controlling shareholder Bain Capital selling its entire stake (44%→0%, realizing approx. $15 billion), and losing a patent lawsuit. Even with a -52% drop, it is still up +2,500% from its low. SanDisk is also -30% from its peak but remains +700% YTD.
In summary, what collapsed was 'the multiples and debt of pure NAND stocks that had risen dozens of times,' not the NAND market conditions. However, Bain's complete exit cannot be taken lightly — the fact that an entity that looked into NAND supply-demand and orders from the board for 8 years cashed out 100% immediately after an all-time high should be read as a signal regarding the cycle's position, beyond just a story about multiples.
Samsung Re-rating. Measuring NAND profit contribution, Samsung is 20~25%, SK Hynix is 17~21%, and the absolute profit pool for Samsung is 1.5~1.6 times larger. In other words, Samsung is the biggest beneficiary of the NAND super-cycle. The previous judgment that "Samsung is riskier due to NAND exposure" was a factual error and is thus discarded. However, since NAND price sensitivity is nearly identical for both companies (per +10%p ASP, company-wide profit +3.1~3.4%), NAND is not a variable that determines the priority between the two stocks.
Second Page
What the market is really selling — HBM4 Delay and CXMT
If the memory stocks collapsed despite the NAND super-cycle, the reason for the sell-off lies elsewhere. Retracing the triggers of the crash day by day, two names remain.
July 13 -15% — The trigger was HBM4
The record daily drop for SK Hynix was caused by a report from a brokerage firm. The gist was twofold — Q2 operating profit would miss consensus by about 8%, and HBM4 shipments are being delayed compared to the plan. HBM4 is the core engine of SK Hynix's profit, and if the customer qualification schedule is pushed back, the 2026 HBM ASP estimates will drop entirely. In short, the market sold the heart of SK Hynix, not NAND.
July 16 -11.5% — The trigger was CXMT
China's ChangXin Memory Technologies (CXMT) raised $8.55 billion (approx. 57.9 billion yuan) through its listing on the Shanghai STAR Market. The allocation of the raised funds includes 7.5 billion yuan for line modification, 18.0 billion yuan for DRAM technology upgrades, 9.0 billion yuan for R&D, and HBM3 development is included here. More importantly, there's the market share — CXMT's DRAM market share jumped from 4.7% to 7.6% in just one quarter (Samsung 38.6% / SK Hynix 28.8% / Micron 22.4%). This was accompanied by reports that Apple has begun evaluating CXMT DRAM for devices intended for the Chinese market.
CXMT DRAM Market Share 4.7% → 7.6%
Rise in a single quarter · HBM3 development included in $8.55 billion IPO funds — A direct challenge to the 3-company oligopoly
No matter how much NAND prices rise, this vector is not offset. What CXMT is targeting is the profit margin of the DRAM oligopoly, which is the main body of profit for both companies. The team's estimated competitive profit exposure is Samsung 51%, SK Hynix 39%(DRAM portion of operating profit × non-HBM portion). High Bandwidth Memory (HBM) remains outside CXMT's reach due to high technical barriers, and Hynix's 56.4% HBM market share acts as a defensive wall — this is the core reason for placing Hynix as Priority 1.
**Offsetting Calculations — Why we are not raising the target price.**If we generously assume NAND average selling price (ASP) exceeds the model by 20 percentage points, Samsung's operating profit increases by +6.9% and Hynix's by +6.3%. However, applying even a 10% discount to profits competing with CXMT results in Samsung -5.1% and Hynix -3.9%.The net effect is Samsung +1.4% and Hynix +2.1%— This is not a range that warrants moving target prices. The NAND tailwinds and CXMT headwinds almost exactly offset each other arithmetically.
Page Three
Kimi K3 — Is the era of inference an ally to memory?
The ultra-large model released by China's Moonshot AI was the psychological trigger for this week's semiconductor sell-off. At the same time, it is the most interesting empirical evidence for the future of memory demand — both are true.
The Reality of the Model
Kimi K3 is a Mixture-of-Experts (MoE) model with 2.8 trillion parameters, making it the largest open-source model ever. It ranked in the top 3 on global leaderboards, and its price is the highest in Chinese AI history at $3 for input and $15 for output per 1 million tokens. The open-weight release is scheduled for July 27.
Evidence of Increasing Memory Demand
This model uses only 16 out of 896 experts for calculation per token — with active parameters at about 50 billion, the computation is light. However, all 2.8 trillion weights must reside in high-bandwidth memory because it is unknown when they will be called. Based on compression (FP8), it requires approximately 2.8TB and 24–32 latest accelerators, and the developer recommends a cluster of at least 64 accelerators. The sparsity rate of 1.8% is the most extreme 'light computation, heavy memory' design ever. The direction that the era of inference is one where memory capacity and bandwidth are the bottlenecks is proven at this point.
Evidence of Decreasing (Opposite Side)
However, the core technology of this model (Hybrid Linear Attention) is precisely an invention to use less memory— it reduces temporary memory by 75% and increases throughput by 6x when processing long contexts. The 1 million token context was achieved not by using more memory, but by bypassing it to use less. Output tokens are also 21% more efficient than the predecessor. The price is also closer to a positioning strategy immediately after entering the top tier rather than evidence of cost (proven by the discounted price of $0.3 upon cache hit). The Team's judgment is **"Total demand increases, but memory intensity per unit decreases — both are true."**it is.
**Important Reinterpretation of Timing.**The Team was originally wary of a 'misinterpreted sell-off' based on 'Cheaper AI → Reduced Capex' when the open weights are released on July 27. However, that sell-off already occurred on Friday, July 17 — Foreign media pointed to the debut of this model and the resulting question of 'Has AI value exceeded fundamentals?' as the direct trigger for Friday's semiconductor sell-off. In other words, while the risk on July 27 has not completely disappeared, it has been partially priced in, and the CXMT listing scheduled for the same day has emerged as a bigger variable.
Chey Tae-won's Remarks — What to Take and What to Discard
In the same week, the SK Chairman said, "Memory will continue to be needed; it will trend upward over time," and "Even if we double production capacity within 5 years, all customers say it is insufficient." The Team's weighting is 60 for Demand Visibility vs. 40 for Stock Price Defense. What to take is the commitment involving capital — the plan to double wafer capacity and the specific figure of "customers requesting a 5–6x expansion" are remarks that can only be made by someone looking at the long-term supply agreement order books. What to discard is the rhetoric of "don't buy and sell frequently"; the informational value of an owner's remarks during a sharp decline is zero due to clear conflicts of interest. And a footnote — Historical memory downcycles have always begun near the declaration of 'doubling capacity'. This commitment is being added to the oversupply watchlist for 2028-29.
Chapter 4
Reading Friday's US Market — Analyzing the 'Quality' of Recovery through Candles
US semiconductor stocks fell on Friday but closed off their intraday lows. Whether this reversal is genuine or not is told by where the closing price landed within the day's range.
| Ticker/Index | Change | Low | High | Closing Position (0=Low, 100=High) | Reading |
|---|---|---|---|---|---|
| SOXX (Semiconductor ETF) | -1.64% | 498.54 | 533.19 | 67.2% | Bottom support formed |
| SMH (Semiconductor ETF) | -2.18% | 536.81 | 567.41 | 64.4% | Inflow of buying pressure |
| Philadelphia Semiconductor (SOX) | -1.63% | 11,194.60 | 11,939.44 | 64.3% | +4.3% recovery from low |
| Nasdaq | -1.40% | 25,250.63 | 25,703.01 | 59.6% | Recovery |
| NVIDIA | -2.21% | 197.97 | 206.65 | 55.8% | Neutral |
| Micron | -0.50% | 804.00 | 903.96 | 45.0% | Reached +12% then pulled back |
| SK Hynix ADR | +1.13% | 145.57 | 167.37 | 38.8% | Failed rebound attempt |
| SanDisk | -3.99% | 1,325.03 | 1,501.20 | 16.9% | Still closed near low |
Here is how to read it. Indices and ETFs closed at the 60-67% level of their upper range — this means buying pressure actually entered at the intraday lows, and foreign media reported that "the sector recovered from lows as investors engaged in dip-buying." However, individual memory stocks are at 38-45%. Hynix ADRs soared nearly 15% from low to high but gave back more than half — a typical failed rally. And only pure NAND (SanDisk) closed near its low at 16.9%.
Do not mistake the price correction of NAND stocks as a signal for the industry cycle. While SanDisk was the only one in the table to close near its low, the trigger for this stock's recent decline was not earnings or industry deterioration, but the initiation of a 'Hold' rating by a research firm (citing valuation burdens). At the same time, the market consensus remains Strong Buy (14-18 Buy · 0-1 Sell), and the 12-month average target price is about 50% higher than the current price. Notably, Goldman Sachs doubled its EPS estimates even while lowering the multiple from 22x to 20x — a combination that perfectly illustrates the nature of this correction: multiples are cooling while earnings are growing. We reaffirm that the stock price decline of NAND companies and the NAND industry outlook are moving in opposite directions.
Conclusion of Chapter 4. The feeling that "the decline is stopping" is true at the index level — buying pressure was confirmed at the bottom, which will act as a psychological buffer for Monday's opening in Korea. However, it is still incomplete at the individual memory stock level. Candles that attempted a rebound but pulled back represent 'searching for a bottom' rather than 'confirming a bottom.' Therefore, we cannot move up the buying timing based on this recovery; instead, we are lowering the probability of a massive gap-down scenario.
Chapter 5
Supply and Demand Dashboard — The Red Light is On
When current figures are applied to the threshold of the market inspection harness created by the team last month, the dashboard is red, not green. This is the most direct reason for delaying the buying timing.
| Indicator | Actual Value | Team Threshold | Judgment |
|---|---|---|---|
| Forced Liquidation Ratio vs. Unpaid Balance | 10.2% (Jan 1.0% → Jun 5.1% → 7/9) | 7% Yellow / 9% Red | Exceeding Red |
| Margin Loan Balance | 36.6336 Trillion | Warning at 35 Trillion+ | Entered Warning Zone |
| Customer Deposits | 107.1 Trillion (Compared to 139.7 Trillion on 6/4)-23.3%) | — | Decrease in Dry Powder |
| VKOSPI (Volatility Index) | 87.14 | 30 Hedge / 45 Extreme | Approx. 2x Extreme |
| Hynix 20-day Moving Average Disparity | 77.5% | Below 85% = Oversold | Downside Overheating (Rebound Potential) |
| Hynix Daily Volatility | 8.38% (Annualized approx. 133%) | — | 1σ Band ±8.4% |
The Timetable for Forced Liquidation — Why Tuesday is Dangerous
Forced liquidation moves mechanically. Collateral shortage occurs → Additional payment by next business day → Forced disposal at the market open on the following business day. Applying this schedule to this week results in the following:
7/16 (Thu) Crash → 7/17 Market Closed → 7/20 (Mon) Payment Deadline → 7/21 (Tue) Forced Liquidation at Market Open
Even if Monday seems quiet, forced volume will hit at Tuesday's open
What the Volume Profile Tells Us — Current Price is 'In the Middle of a Wall'
Breaking down 2026 trading volume by price level, Hynix's 1,825,000~1,975,000 range is the largest volume wall of the year(2.17–2.61% per 25,000 KRW interval). Conversely, below 1,800,000, the density drops to less than half at 0.5–1.06% per interval. Buying near the current price means purchasing right in the middle of the largest volume of trapped positions of the year, and you will have to break through a wall of break-even sell orders every time it rebounds.
The risk-reward ratio confirms this numerically (based on Stop Loss 1,600,000 / Target 2,600,000).
| Entry Price | To Stop Loss | To Target | Risk:Reward (R:R) |
|---|---|---|---|
| 1,842,000 (Current Price) | -13.1% | +41.2% | 3.1 : 1 |
| 1,817,000 | -11.9% | +43.1% | 3.6 : 1 |
| 1,762,000 | -9.2% | +47.6% | 5.2 : 1 |
| 1,678,000 | -4.6% | +54.9% | 11.8 : 1 |
Samsung Electronics has the same issue. A breakdown of the volume profile shows that 230,000–250,000 is a vacuum zone with 0.00% volume — it was passed through during two rapid surges (+8.5%, +10.1%) in late May and early June, leaving no memory of supply and demand. In other words, no support levels exist in this range. The actual levels are 247,000(where the 50% retracement of 247,350 and the July 14 low of 247,000 align with a 0.14% margin of error) and 217,000~222,000(an overlap of the 61.8% retracement at 217,342 and the 215,000–225,000 volume profile).
Chapter 6
Four Perspectives
Onion · Macro/Liquidity
"NAND contract prices are jumping by over 70% per quarter, and enterprise SSDs have overtaken smartphone consumption for the first time in history. The demand structure has fundamentally improved, centering on IT infrastructure. The rule of 'hold if NAND weakens' should be discarded — NAND isn't weakening; it's at the heart of the boom. Now that 80% of margin liquidations have been absorbed, even a small inflow of buying can lead to a strong rebound in this supply-demand vacuum."
Acorn · Sector/Value
"NAND profits are being revised upward, but the multiple the market will assign to those profits is decreasing. A +70-75% quarterly increase is both a super-cycle and a signal for a peak multiple discount. Translating Samsung's NAND profit upward revisions into after-tax discounted multiples results in 6,000–18,000 KRW per share — raising the target price to 360,000–380,000 is the error of capitalizing the peak of NAND prices as normalized profit."
Peanut · Funda/Risk
"What the market is selling isn't NAND. The trigger for Hynix's -15% was the HBM4 delay, and the trigger for the -11.5% was CXMT. Bain Capital fully liquidated its 44% stake to 0% — the entity that observed supply and demand from the board for 8 years cashed out 100% immediately after an all-time high. Furthermore, in memory, a low forward PER is not a sign of a floor, but a signature of a ceiling. In the past, Micron peaked at a forward PER of 4–5x and then dropped -56%."
Carrot · Chart/Flow/Sentiment
"Don't just look at the +1.13% ADR. There was a -13.69% before it, and Friday's candle was a failed rally that gave back more than half its gains from the high. And the problem isn't the target price, but the center of gravity of capital — the current primary buying zone is in the middle of the year's largest volume wall, and Samsung's 240k·235k levels are vacuum zones with 0.00% volume weight. We need to lower the ladder. To 1.68–1.76M for Hynix, and 2.17–2.22M for Samsung."
Synthesis of the four perspectives. 'NAND = Negative for Samsung' was discarded by a 4-0 unanimous vote, and maintaining the Base target price also saw no disagreement (offsetting NAND upgrades with CXMT discounts). What differed was the speed of execution — Onion was for immediate buying, while the other three favored delaying the timing and lowering the price levels. The team adopted the latter. The rationale is not a 'hunch' but three empirical measurements: a 10.2% margin call rate (Red Zone), the Tuesday market-open forced liquidation schedule, and the fact that the current price sits in the middle of the year's largest volume wall.
Chapter 7
Setting the Price — Team's Proprietary Target Prices
We maintain the Base and above, and only lower the Bear. This is because the previous Bear was 'daily volatility' rather than a 'scenario.'
SK Hynix (000660) — 1,842,000 KRW · 1st Priority Buy
| Scenario | Target Price | Rationale |
|---|---|---|
| Bear | 1,450,000 KRW -21% Downward Revision | HBM4 re-delay + CXMT price competition intensifies. At -51% from the peak, this is the lower end of past memory downcycle drops (50~80%). The previous Bear scenario (1,650,000) failed to function as a scenario because its cushion was -10.4%, which is smaller than the single-day drop on 7/16 (-11.5%). |
| Base | 2,600,000 KRW +41% Maintain | Forced selling exhausted + HBM4 schedule normalization. Forward PER 6.79x → 8x normalization. 2.4% margin of error from Fibonacci extension 0.618 (2,540,060). 31.1% of 2026 trading volume must pass through this level. |
| Bull | 3,150,000 KRW +71% Maintain | Loan short covering ignited + forward PER in the 9x range + confirmation of 2027 HBM volume in Q2 earnings |
| Super Bull | 3,700,000 KRW+ +101% Maintain · Rationale replaced | CXMT yield failure + HBM4 normalization(Previous HBF rationale deleted). Margin of error from Fibonacci extension 1.618 (3,710,060)0.3% |
Tail scenario (coordinates for sizing): 1,150,000 if CXMT price war and HBM4 re-delay occur simultaneously. This is the worst-case coordinate, not the Base case.
Samsung Electronics (005930) — 255,000 KRW · Parallel 2nd priority
| Scenario | Target Price | Rationale |
|---|---|---|
| Bear | 215,000 KRW -16% Lowered | CXMT's server DRAM penetration + non-memory sluggishness. Overlap of 61.8% retracement (217,342) and volume profile (215,000~225,000). The previous 235,000 had 0.00% volume weight, providing no support rationale. |
| Base | 330,000 KRW +29% Maintain | NAND super-boom profit (+6.9%) offset by CXMT competition profit discount (-5.1%) → Net +1.4%. 1.2% margin of error from Fibonacci extension 0.618 (325,928) |
| Bull | 380,000 KRW +49% Maintain | Foundry break-even visibility + expansion of HBM customer certification (some institutions forecast HBM market share reversal in 2027) |
| Super Bull | 450,000 KRW+ +76% Maintain | Major 2nm orders + foundry turning a full profit + net asset multiple re-rating. 0.7% margin of error from Fibonacci extension 1.618 (453,078) |
Samsung's advantage: NAND profit pool is 1.5~1.6x that of Hynix, and NAND is out of CXMT's reach. No burden of capital increase at the peak. Hynix's advantage: Exposure to CXMT competition profit is 12 percentage points lower, and HBM market share acts as a barrier. Therefore, the 1st priority is Hynix, with a 60:40 split.
**Prerequisites for the target price ladder.**This ladder is calculated on the premise of defending the 50% retracement line (Hynix 1,817,000 / Samsung 247,350). If this line breaks, the reference point drops to 61.8% and the Base is recalculated — Hynix approx. 2,434,600 (-6.4%), Samsung approx. 314,500 (-4.7%). While there is no reason to lower the target price now, these two numbers are the recalculation triggersis clearly stated.
Eighth Page
Actual Buying Roadmap
This page is the core of this report. One conclusion — lower the ladder and delay the timing until after Tuesday.
SK Hynix — Based on 100 units of reserve funds
| Timing/Price Range | Allocation | Conditions and Rationale |
|---|---|---|
| 7/20 (Mon) ~ 7/21 (Tue) | 0 | **New buying prohibited.**Margin calls from the 7/16 crash will be executed at the market open on 7/21. The Team Hannes rule is to avoid being trampled by forced liquidity. |
| Above 1,850,000 | 0 | **Chasing prohibited.**1,825,000~1,975,000 is the largest volume wall for 2026. Even if it rebounds, it is a volume digestion zone. |
| 1,790,000~1,840,000 | 10 | After 7/22, Only when margin calls are confirmed to have stabilizedExecute scouting position (reduced from previous 20) |
| 1,762,000~1,790,000 | 30 | Risk-reward ratio 5.2:1. Primary defense zone overlapping with the ADR Friday low conversion line (1,761,800).Exceptional pre-execution of 10 allowed if a sharp drop occurs on Monday |
| 1,678,000~1,730,000 | 40 | **Core buying.**Risk-reward ratio 11.8:1. The 7/14 low (1,678,000) is a structural pivot created by the year's highest volume reversal day.Pre-execution of 20 allowed if reached on Monday |
| 1,545,000~1,600,000 | 20 | 61.8% retracement (1,540,900) zone. Forced liquidation overshooting spot — will not stay long |
**Stop-loss structure (2-stage).**In an environment with daily volatility of 8.38%, a single stop-loss line is prone to noise. Therefore, we divide it into three stages.
- Closing price breaks below 1,678,000 → Current holdings reduced by half (Loss of structural pivot)
- Closing price breaks below 1,600,000 for 2 consecutive days → Additional 1/3 trim (Not a full stop-loss — this zone is a volatility spot, not a fundamental value impairment)
- Break below 1,400,000 → Full stop-loss (Bear scenario confirmed)
- Note: Intraday momentary breaks are considered liquidation overshooting and not used as triggers — must be based on the closing price.
Samsung Electronics — Based on 100 units of reserve capital
| Price Range | Allocation | Conditions and Rationale |
|---|---|---|
| 250,000~255,000 (Current price range) | 0 | Risk-reward ratio of 1.5:1 is not viable. Lacks support rationale as 230,000~250,000 is a supply vacuum zone. |
| Defense confirmed at 247,000 | 20 | Primary defense line where the 50% retracement (247,350) and the 7/14 low (247,000) align with a 0.14% margin of error. |
| 228,000~232,000 | 30 | Accumulation in the zone below the 120-day moving average (approx. 237,950) |
| 217,000~222,000 | 50 | Core buy. Overlap of 61.8% retracement (217,342) + supply zone (215,000~225,000). Risk-reward ratio of 7.3:1. |
Replacement of suspension rules. The existing "suspend buying if NAND spot prices drop" is discarded (NAND is a positive factor for Samsung, so the trigger was monitoring the wrong place). The replacement rule is "Upgrade to buy suspension if CXMT's server DRAM passes certification or adoption by major cloud providers is confirmed".
Common Rules
- Prohibited from exhausting more than 30% of total capital before SK Hynix Q2 earnings on 7/29 (Strengthened from the previous "50% before 7/27")
- On days when circuit breakers or sidecars are triggered, defer new trades to the next day
- Priority Hynix : Samsung = 60 : 40
Post-market opening 30-minute diagnostic checklist
| Indicator | Checklist Item | Interpretation |
|---|---|---|
| Defense of 1,817,000 | Is the low between 09:00~09:30 above this line? | 50% retracement line. A double defense line overlapping with the 7/16 low (1,821,000). If broken, the next confirmation line is 1,761,800 |
| Foreigner program direction | Shift to net buying | 7/16 Foreigners -970B, Institutions -880B, Retail +1.8T. Retail net buying is margin leverage, so it is not a defense but fuel for future forced liquidations |
| Sidecar · Circuit Breaker | Re-triggering status, 09:10 trading volume | If triggered, defer new trades for the day |
Action branches by confirmation result
| Condition | Action |
|---|---|
| 1,817,000 defended + Foreigners shift to net buying | Still Buy suspension (Wait for Tuesday's forced liquidation confirmation). However, if it drops below 1,762,000, execute 10 units. |
| 1,817,000 defended + Foreigners continue net selling | Full standby. Set limit orders only at 1,762,000 / 1,678,000 |
| 1,817,000 broken + Sidecar re-triggered | Standby. Execute 20 units only when approaching 1,678,000 |
| Closing price breaks below 1,678,000 | Halt new buys, reduce current holdings by half. Re-evaluate at 1,545,000 |
| Recovery of 1,975,000 (Breakout of supply wall) | Do not chase. Wait for 1,900,000 pullback, then execute 20 units |
Chapter Nine
Event Calendar and Confirmation Indicators
| Timing | Event | Response Guidelines |
|---|---|---|
| 7/21 (Tue) | Execution of forced liquidations at market open from 7/16 crash | No new buys for the day. Lows created by forced volume should be after Wednesday confirmed before buying |
| 7/22~7/24 | Check for forced liquidation stabilization | Maintain daily range of 50B = Exhausted / Re-break above 100B = Remaining liquidation. If stabilized, execute first 10 units. |
| 7/27 (Mon) | CXMT Shanghai Listing (Key Variable) | Monitor the expansion roadmap and HBM3 progress announcements using funds raised after listing. First data to gauge the reality of DRAM oligopoly impairment. |
| 7/27 (Mon) | Kimi K3 Open Weights Release | Sell-off was partially priced in on 7/17. If GPU rental rates and hosting demand actually move after release, the 'Inference = Memory' logic upgrades from news to data. |
| 7/29 (Wed) | SK Hynix Q2 Earnings (The Biggest Hurdle) | Checklist Items — ① Does it confirm the warning of Q2 operating profit missing consensus by 8%? ② Status of HBM4 shipment schedule normalization ③ Comments on 2027 HBM volume and pricing. Do not exhaust more than 30% of cash reserves until this announcement. |
| Late July to early August | Earnings results of the Big 4 global cloud providers | Re-evaluate demand assumptions if even one AI CapEx guidance is revised downward |
| Late July | Samsung Electronics finalized earnings results | In the breakdown by business division, NAND profit contribution and the extent of reduction in Foundry losses |
| 8/5 · 8/19 | Implementation of leveraged product regulations | Panic selling triggered by implementation is an opportunity for confirmation, not fear |
| Occasional | NAND/DRAM spot prices (weekly) | A slowdown in the rising trend of NAND contract prices slowdown + weakening of enterprise SSD orders simultaneously Only when observed is it a hold signal (ignore simple spot price fluctuations) |
Hypothesis verification index — Is 'Era of Inference = Era of Memory' true?
- GPU rental rates and hosting supply announcements following the release of Open-Weights — If self-hosting demand is real, hourly rental rates for high-spec accelerators rise and hosting providers line up. If quiet, this channel remains purely conceptual.
- Direction of NAND/DRAM contract prices — Currently both are rising (NAND +70~75%, DRAM +58~63% QoQ). Regardless of stock price adjustments, if prices continue to rise, this decline is confirmed as multiple liquidation. Conversely, if the contract price growth rate drops to single digits, that is the peak scenario turning point — you must look at contract prices, not the stock price.
- Q2 earnings comments on 2027 HBM volume + cloud CapEx guidance — Whether the claim "doubling capacity is still not enough" is confirmed by numbers
Things to be most wary of. ①HBM4 schedule — This is the heart of SK Hynix's profits; if delays are reconfirmed, the base scenario itself is shaken. ②CXMT — The speed at which IPO funds actually lead to capacity expansion and HBM3. ③Tuesday's margin call liquidations — Mechanical volume unique to the domestic market. ④ And the trap of "it's cheap because forward PER is low" — remember that in memory, a low forward multiple may be the shadow of a profit peak, not the bottom.
Closing thoughts. The greatest gain for the team in this investigation was not the target price, but the recalibration of 'what to monitor.' The first misreading was interpreting NAND as a negative factor — NAND contract prices are rising over 70% per quarter, shortages will persist until 2027-28, and the biggest beneficiary is actually Samsung, which has a larger profit pool. It was wrong to delay Samsung out of concern for NAND prices, and conversely, it was wrong to try to raise target prices based on the NAND boom (as it is offset by the CXMT discount). The real variables are SK Hynix's HBM4 schedule and China's CXMT DRAM penetration, layered with domestic mechanical liquidation schedules. Therefore, this week's discipline is simple — Do not buy on Monday and Tuesday. Lower the ladder. Spend 70% of cash reserves after seeing the July 29 earnings results. Low prices created by fear always come, but to catch them, you must have bullets left until then.
Appendix
References
NAND Price/Supply and Demand
- TrendForce — 2Q26 Memory Contract Price Outlook (NAND +70~75% QoQ, DRAM +58~63%): trendforce.com (20260331-12995) / 3Q26 Price Trends: trendforce.cn (20260703-13133) / NAND Industry Analysis 2Q26: trendforce.com (RP260603AI)
- IDC — Global Memory Shortage Crisis and Impact on Smartphones/PCs: idc.com
- NAND Shortage 2026 Guide (Demand +20~22% vs Supply +15~17%): unibetter-ic.com / Memory Market Outlook: ampheo.com
- Forbes — 2026 Digital Storage and Memory Outlook: forbes.com / NAND Research Crisis Update: nand-research.com
- Samsung/SK Hynix NAND Profit Contribution Estimates: Synthesized from Samsung IR, EBN, dealsite, Shinhan Investment & Securities, and Mirae Asset reports (refer to table in text)
Kioxia, SanDisk, Bain
- Bloomberg — Kioxia market cap halved ($185B wiped out): bloomberg.com / Nikkei Asia — AI Deleverage: asia.nikkei.com
- Bloomberg — Bain Capital Exits Kioxia Entirely (Realizing approx. $15B): bloomberg.com (bain exits) / Japan Times: japantimes.co.jp / Investing.com: investing.com
- Newspim — Kioxia Margin Buying Hits Record 870 Billion Yen, Facing Margin Calls: newspim.com / Hankyung — Precursor to Winter or Excessive Fear: hankyung.com
- SanDisk Plunge Analysis: INDmoney · TradingKey · Yahoo Finance
- SanDisk Consensus & Target Price (Strong Buy, 12-Month Average $2,041.88): TipRanks · MarketBeat · Benzinga / Goldman Sachs Raises Target Price (Multiple 22→20x, EPS Upgraded 2x): TheStreet
CXMT, HBM4, and Causes of the Plunge
- Etoday — CXMT Challenges Memory Big 3 with $8.5 Billion IPO: etoday.co.kr (2604509) / Seoul Economic Daily — China's CXMT, World's 4th Largest DRAM Maker: sedaily.com (20066523)
- Hankook Ilbo — Semiconductor Duo Plunges Despite Successful SK Hynix ADR Offering (HBM4 Delay Report): hankookilbo.com / Seoul Shinmun — 1.8 Million Hynix & 250,000 Samsung in Just Three Days: seoul.co.kr
- Invezz — Memory Sell-off Triggered by TSMC Earnings: invezz.com / WSJ — TSMC Raises Capex: wsj.com
Kimi K3 and Friday's US Market
- Motley Fool — 7/17 Closing "Semiconductor Plunge Deepens": fool.com / Yahoo Finance — "Chip Stocks Close with Rebound from Lows": finance.yahoo.com / TheStreet: thestreet.com
- VentureBeat — Kimi K3 Unveils 2.8 Trillion Parameters: venturebeat.com / MarkTechPost — Architecture Details: marktechpost.com / Artificial Analysis Leaderboard: artificialanalysis.ai
- KDA Paper (75% KV Cache Reduction, 6x Faster Decoding): arXiv 2510.26692 / Serving Hardware Requirements: TECHi · ModemGuides
- Interpreting the 'DeepSeek Flashback': Fortune · Axios
Chey Tae-won, HBF, and Domestic Supply/Demand
- Money Today — "Just Keep Holding On": mt.co.kr / Newspim — "Upward Trend Likely Over Time": newspim.com / Financial News — HBM Demand is Exponential: fnnews.com / Tom's Hardware — Capacity to Double Within 5 Years: tomshardware.com
- HBF: Tom's Hardware (Standardization Cooperation) · IEEE Spectrum · SanDisk Official Blog · SemiEngineering
- Ajou Business Daily — 41% of KOSPI Margin Buying Concentrated on Samsung, Hynix, and Square: ajunews.com / Etoday — Margin Debt at 38 Trillion, Forced Liquidation at 10.2%: etoday.co.kr / Segye Ilbo — Forced Liquidation Statistics & Concerns of a Second Wave: segye.com
- Financial News — Constitution Day Holiday and US Semiconductor Crash (ADR -13.69%): fnnews.com / Segye Ilbo — KOSPI Plunges Over 6% (VKOSPI 87.14): segye.com / GraniteShares — ADR Ratio 1:10: graniteshares.com
- FSC — Improvement Plans for Leverage Products (Effective 8/5, 8/19): fsc.go.kr
- Price, Volume Profile, Disparity, Fibonacci, Volatility, Candle Structure: Measured directly by the team using Naver Finance API (000660, 005930, KOSPI Daily) and yfinance (SKHY, MU, SNDK, STX, WDC, SMH, SOXX, ^SOX, ^N225, 285A.T, EWY, KRW=X)
This report is a response analysis material synthesizing real-time re-investigation and cross-discussion by the Goguma Team (Onion, Acorn, Peanut, Carrot, Goguma).
The presented target prices and buy levels are independent estimation scenarios of the Goguma Team, not securities firm consensus, and do not constitute an investment recommendation. Investment decisions and responsibilities lie with the investor.
Price basis: Korea July 16, 2026 (Thu) closing price · US/Japan July 17 (Fri) closing price · Response target date: From July 20 (Mon) onwards · SK Hynix (000660) · Samsung Electronics (005930)
Comments 0
to comment Login