Samsung: Kiwoom's KRW 390k vs KB's KRW 600k, By the Numbers — 'Level' and 'Growth Rate' Don't Conflict (2026-07-08)

🔗 Read together: Goldman token demand vs JPMorgan ASP peak-out — do the two calls really conflict? · Samsung Electronics: Korea’s AI-memory bellwether (deep dive) · The big-tech financing relay — AI CapEx hasn’t shrunk · CXMT IPO — separating HBM risk from client-DRAM risk

On the same morning, Kiwoom cut Samsung’s target price to KRW 390,000 and KB raised it to KRW 600,000. It looks like a head-on clash. But once you take both reports apart by the numbers, they do not refute each other. They are touching different parts of the same elephant. This post minimizes the “who’s right” verdict and focuses on decomposing what each report actually argued — its reasoning and its figures.

TL;DR

• Kiwoom is about the deceleration of the growth rate (dG/dt, the second derivative); KB is about the absolute level of earnings. They can coexist mathematically, and within the figures given, both are true. Framing them as “opposing reports” is a misspecification.
• It hinges on the provision adjustment. The incentive provision in Q2’s reported OP of 89.4tn totals about 17.6tn, but the genuinely one-off part is only the ~5tn retroactive top-up of Q1’s under-accrual (the rest recurs every quarter, including Q3). Normalize just that 5tn and Q3 QoQ is neither the headline +26% nor the +4.7% you get by stripping the whole provision, but ~+16–19%. The growth acceleration moderates (Kiwoom’s direction) but does not collapse (KB’s level holds).
• The market already answered. Right after a record quarter, the stock fell sharply and a circuit breaker tripped. What the market priced was not a demand collapse but a peak in the growth rate + extreme positioning.

1. What each report actually said

Facts first. The related coverage of both reports was verified, and the figures below are on a media/company-disclosure basis. That said, the full original 2026-07-08 PDF reports from both brokers (with detailed valuation tables) could not be obtained directly. That part is uncertain and is left in [Blocked] below.

One fact jumps out immediately. Their Q3 OP estimates are essentially the same — 112tn vs 110tn, and both match consensus (~111tn). So Kiwoom did not lower the “level” of earnings. What it lowered is the multiple applied in the target price. That defines the nature of this debate — it is a P/E debate, not an EPS debate.

What KB adds on top of the earnings level is a set of options: the scale of global AI investment (~USD 800bn this year → 1.1tn next → 1.5tn in 2028), a shortage into 1H 2028, 2027 HBM price negotiations, buyback/cancellation and special dividends, potential new foundry wins, and an ADR-listing review. Many of these are still event options or estimates, and should be separated from confirmed facts.

2. Level vs growth rate — it hinges on how you adjust for the provision

A stock trades the rate of change of earnings, not the level. So “record earnings (level)” and “a slowdown in the growth acceleration (growth rate)” can hold at the same time. But there is a common trap when you try to show this in numbers: how you adjust for the incentive provision.

On KB’s basis, Q2 adjusted (ex-incentive-provision) OP is 107tn and reported OP (with the provision) is 89.4tn. The difference of about 17.6tn is the total incentive provision booked in Q2.

Here is the key correction. You must not strip out that full 17.6tn as “one-off.” The incentive provision is tied to profit and is booked every quarter — Q3 results will carry a provision too. The genuinely one-off part of Q2 is the retroactive top-up: the incentive under-accrued in Q1 relative to its profit, booked belatedly in Q2. That top-up is estimated at about KRW 5tn, in proportion to Q1 operating profit (~57tn) (Q2’s 89.4tn ≈ +56% over Q1).

So when normalizing the base, you add back only that one-off ~5tn, not 17.6tn — leaving each quarter’s own normal provision in place.

• Q2 normalized OP ≈ 89.4tn + 5tn = ~94tn
• Q3 estimated OP ≈ 110–112tn (this too includes a normal provision)
• Normalized Q3 QoQ ≈ 112tn ÷ 94tn ≈ +18% (about +16% on 110tn) → ~+16–19%

Line up the three numbers and the illusion shows:

• Headline “Q3 +26%” — uses the provision-depressed Q2 report (89.4tn) as the denominator, overstating the acceleration.
• +4.7% if you remove the full 17.6tn — an asymmetric comparison that strips Q2’s provision but leaves Q3’s in, so it understates growth instead.
• ~+16–19% after normalizing only the one-off (~5tn) — this is closest to reality.

In short: the earnings level is a record (KB right). The growth acceleration does moderate from the extreme Q2 pace (Kiwoom’s direction right). But that moderation is not a Q3 cliff; it is a gradual curve — Q2 (surge) → Q3 (~+16–19%) → Q4 (124tn ÷ 112tn ≈ +10.7%). The two reports are not contradictory; they look at different derivatives (level vs rate of change) of the same earnings curve, and the growth-rate slowdown is far milder than a “strip the full provision” calculation implies.

3. The data the market printed on July 7

The debate was, in fact, settled by the market a day earlier. On July 7, when Samsung reported a record quarter (Q2 preliminary OP of 89.4tn), the stock fell sharply — over ~5% — to the KRW 300,000 area, and the year’s 6th circuit breaker tripped. The flow data in the related coverage explains the backdrop.

• Foreign ownership had fallen to a 17-year low, and heavy foreign profit-taking piled on.
• Retail margin debt was at an all-time high, so positioning was extremely one-sided.
• At the same time, geopolitical risk around the Strait of Hormuz (a US CENTCOM strike on Iran) spiked oil, and US semiconductor stocks fell in tandem.

The message is clear. The market has never priced an AI-demand collapse. A circuit breaker on record earnings shows that what the market traded was not the “level” of earnings but a peak in the growth acceleration plus the unwind of extreme, crowded positioning. If so, the rebuttal that “AI worries are just noise” targets something the market wasn’t worried about (a demand collapse).

4. Price data — ASP still rises, but the slope bends

The industry-data basis for Kiwoom’s case is the slowdown in memory contract-price growth. The essential distinction: a slower growth rate is not a price decline. On TrendForce’s basis, contract-price growth was framed as follows.

So prices still rise in Q3 — but the slope drops sharply, from the Q2 surge to the low-teens/mid-teens. Kiwoom’s line that “2H price growth is unlikely to beat expectations by a wide margin again” should be read not as cycle-end but as a narrowing of the beat. This data lines up directionally with the gradual growth-rate moderation in Section 2 (Q3 ~+16–19%).

5. Valuation is a P/E question, not an EPS question

The distance between the KRW 390k and 600k targets is not a difference in earnings estimates but in the applied multiple (P/E). Plugging the KRW 296,000 close and the verifiable 2026E EPS into a back-of-envelope (EPS: Kiwoom 43,429 / KB 44,379, approx. 44,000):

• Current 296,000 ÷ 2026E EPS ~44,000 ≈ 6.7x P/E
• Kiwoom target 390,000 ÷ 44,000 ≈ 8.9x
• KB target 600,000 ÷ 44,000 ≈ 13.6x
• KB target 600,000 ÷ 2027E EPS 58,361 ≈ 10.3x

The structure the numbers show: Samsung trades at about 6–7x on 2026 earnings. In absolute terms that is low. But memory stocks almost always look cheap when peak EPS is visible. The KRW 390k figure applies about 9x to 2026 earnings — close to a multiple grounded in currently verifiable results — while KRW 600k presumes a re-rating to a higher multiple that only holds if 2027 earnings durability is accepted. In other words, the bottleneck in this debate is not the earnings themselves but when the low P/E breaks (i.e., proof of earnings durability).

6. Why Q, not P, is the real axis — the bifurcation of memory demand

The debate is conducted mostly around price (P), but the real variable Kiwoom planted is volume (Q). The core of Kiwoom’s mechanism is: “component price hikes → set (PC/smartphone) price hikes → demand worry → a shift in OEM memory-buying.” Right now it is still at the price-resistance stage; the question is whether that turns into an actual cut in order volumes.

The essential split is that memory demand is not a single curve.

• Consumer memory (mobile/PC DRAM/NAND): elastic. When set prices rise, set demand falls; this is where volume wobbles first.
• AI memory (HBM, server DRAM, eSSD; hyperscaler demand): relatively inelastic. A separate demand curve tied to data-center CapEx.

This bifurcation matters because even if consumer volume is cut first, a rising mix of high-value AI product can offset it, so blended ASP and margins may actually improve. That means the consumer part of Kiwoom’s worry and the AI part of KB’s bull case can both be true, and the actual outcome is decided by their relative speeds. The real axis of the debate is not “do prices rise” but “does the inelasticity of hyperscaler HBM/server demand hold.”

7. So what to watch (data checkpoints)

More than a verdict, what matters is the data that would confirm or overturn the reasoning above. When the news drops, track only these numbers.

• ASP slope: does Q3 DRAM/NAND contract-price growth hold within TrendForce’s ranges (DRAM +13~18%, NAND +10~15%); any Q4 price-decline signal?
• Order volume (Q): does PC/smartphone OEM memory order volume move past price resistance into an actual cut?
• Hyperscaler CapEx: at the late-July big-tech earnings calls, is AI-CapEx guidance maintained, or an outright cut?
• HBM4 share: does Samsung’s HBM4/eSSD share actually expand (Nvidia volume share especially)?
• China supply: does CXMT/YMTC legacy and server-DRAM penetration erode the mix?
• One-offs / capital policy: Q2 final results by division and the exact incentive-provision size; formalization of buyback/cancellation, special dividend, ADR.

8. In sum

The nature of this debate is not a bull-vs-bear clash but a time-axis illusion. KB is right on the earnings level, Kiwoom is right on the growth rate, and because the market trades the growth rate, the short-term price responded to Kiwoom’s sentence first — the circuit breaker on record earnings is that data point. At the same time, if KB’s level basis (a shortage into 2028) is real, this drop is more likely a pullback inside the cycle than the top of it. In the end, what separates the next phase is not the target-price number but two observable data points: whether consumer price resistance turns into an order-volume cut, and whether the inelasticity of AI-server demand holds.

Evidence classification (Appendix)

[Fact]

• Kiwoom: target cut 430k→390k, maintain Buy, Q3 OP ~112tn (matches consensus ~111tn). (Hankyung report)
• KB: target raised 550k→600k, Q3 OP ~110tn, 2H OP 234tn (Q3 110 · Q4 124), Q2 adjusted OP 107tn vs reported 89.4tn. (Etoday report)
• Samsung Q2 preliminary OP 89.4tn is company-disclosed. Sharp drop right after the July 7 record print; year’s 6th circuit breaker. (related coverage)
• TrendForce: Q2 commodity DRAM +58~63% / NAND +70~75%; Q3 DRAM +13~18% / NAND +10~15%.
• Verifiable 2026E EPS: Kiwoom 43,429 / KB 44,379; prior close 296,000.

[Inference]

• The incentive provision in Q2 reported 89.4tn totals ≈17.6tn, but only ~5tn is genuinely one-off (a retroactive Q1 top-up, in proportion to Q1 OP ~57tn). Since a provision recurs in Q3 too, removing the full 17.6tn is an asymmetric comparison. Normalizing only the ~5tn one-off puts Q2 base ≈94tn and Q3 QoQ ≈ +16–19% (headline +26%; +4.7% if fully stripped), Q4 ≈ +10.7%.
• Growth acceleration moderates gradually: Q2 (surge) → Q3 (+16–19%) → Q4 (+10.7%). Kiwoom’s growth-slowdown case is directionally right but milder than a full-provision-strip illusion.
• The real axis is P/E, not EPS (now ~6.7x → 390k ≈ 8.9x, 600k ≈ 13.6x on 2026E / 10.3x on 2027E).
• Memory demand bifurcates into consumer (elastic) and AI (inelastic); a consumer volume cut may be offset by AI mix improving blended margins.
• The July 7 drop reads as a peak-in-growth-rate + extreme-positioning (foreign 17-yr low, retail margin ATH) unwind, not a demand collapse.

[Speculation]

• 2027 HBM price 2x YoY (KB negotiation view).
• Potential new foundry wins, ADR-listing review, special dividend/buyback (KB, unconfirmed event options).
• Hormuz escalation acting as a macro trigger amplifying the crowded-long unwind.

[Blocked]

• Full 2026-07-08 Kiwoom/KB original PDFs and detailed valuation tables.
• Samsung Q2 final results by division and the exact incentive-provision size.
• Whether Samsung officially pursues an ADR, special dividend, or new large foundry wins.
• The actual Nvidia HBM4 volume-share figure.

Sources: Samsung official disclosures, Hankyung and Etoday reports, TrendForce contract-price forecasts, verifiable broker reports. Items where the original PDF was not obtained are flagged [Blocked].

이 글은 리서치·정보 제공용이며 투자 조언이 아닙니다. 종목명은 분석을 위한 예시이며, 매수·매도 권유가 아닙니다. 본문의 주가·목표주가·이익 추정·EPS·PER은 보도 및 각 증권사·회사 발표 기준이고, 집계 기준·시점에 따라 값이 다를 수 있습니다. 데이터 기준일: 2026년 7월 8일 KST.

Disclaimer: For research and information purposes only. Not investment advice. Company and broker names are cited for analytical illustration. Prices, target prices, earnings estimates, EPS and P/E figures are based on media reports and company/broker disclosures and may vary by source and timing; unverified items are flagged as [Speculation]/[Blocked]. Data as of July 8, 2026 (KST).