AI Server on Korea Invest Insights

Samsung Electro-Mechanics (009150) Analysis: Mirae Asset's KRW 1.3M Target and the MLCC/FC-BGA Re-Rating Frame

Thu, 07 May 2026 18:10:00 +0900

Related reading: AI PCB and Substrate Hub, Samsung Electro-Mechanics AI Infrastructure Re-Rating, AI PCB and Substrate Thesis, and Korea AI PCB Ecosystem: 10 Companies.

Mirae Asset Securities raised its Samsung Electro-Mechanics target price from KRW 530,000 to KRW 1.3 million. The number looks dramatic, but the mechanism is simple: move the valuation year from 2026 to 2028 and apply a 37x PER, which Mirae Asset links to the early phase of the 2017 MLCC shortage cycle.

That makes this less a normal earnings revision and more a valuation-frame revision. The core debate is whether Samsung Electro-Mechanics should still be read as a cyclical electronics component company, or whether AI server MLCC and FC-BGA demand has moved it into an AI-infrastructure bottleneck category.

TL;DR

Mirae Asset’s target math is 2028F EPS of KRW 34,764 multiplied by 37x, or KRW 1,286,268, rounded to KRW 1.3 million.
The 37x multiple is the key assumption. It is not a normal-cycle multiple; it is a peak shortage-cycle multiple drawn from the 2017 MLCC shortage setup.
At the May 7, 2026 close of KRW 917,000, Samsung Electro-Mechanics already trades at 26.4x that 2028F EPS. That is above a normal upcycle frame of roughly 25x and close to the 2018 shortage-peak frame of 28x.
Shortage signals are visible: some MLCC lead times are reportedly 20-24 weeks versus a normal 10 weeks, 2Q26 MLCC utilization is discussed around 95%, and inventory coverage is cited around 4 weeks versus a normal 6 weeks.
FC-BGA is the more structural leg. Mirae Asset lifted its FC-BGA ASP assumption by 13%, and package-solution operating profit is forecast to compound strongly through 2030.
The stock is no longer undiscovered. Hana, NH, Meritz, iM, KB and Mirae Asset have all moved into the KRW 1.0-1.3 million target range. The remaining question is how much of the shortage multiple is already in the price.

1. What KRW 1.3 Million Actually Means

The target-price formula is straightforward:

Target price = 2028F EPS x target PER
 = KRW 34,764 x 37.0
 = KRW 1,286,268
 = roughly KRW 1.3 million

The arithmetic is not the thesis. The thesis sits inside two choices:

Input	Mirae Asset’s choice	Why it matters
Earnings year	2028F EPS	Pulls forward the Vietnam FC-BGA capacity contribution and the AI server mix shift
Multiple	37x PER	Imports the early 2017 MLCC shortage multiple into today’s AI MLCC/FC-BGA cycle

Moving the earnings base from 2026 to 2028 is already a large change. Mirae Asset’s EPS path, as reported by local coverage, implies roughly:

Year	EPS	Growth
2025	KRW 9,099	n/a
2026F	KRW 16,914	+86%
2027F	KRW 24,555	+45%
2028F	KRW 34,764	+42%

Using 2028 EPS rather than 2026 EPS almost doubles the earnings denominator. Applying 37x to that denominator turns an earnings upgrade into a re-rating call.

The question for investors is therefore not, “Is KRW 1.3 million possible?” A better question is, “What has to be true for a 37x multiple on 2028 earnings to become acceptable?”

2. The 37x Multiple Is the Debate

Mirae Asset’s reported historical PER frame can be summarized this way:

Frame	PER	Interpretation
10-year average	15x	Normal long-term electronics-component frame
+1 standard deviation	19x	Stronger but still ordinary upcycle
+2 standard deviations	25x	Normal-cycle upper band
2021 upcycle peak	22x	COVID recovery and component upcycle
2018 shortage peak	28x	MLCC shortage peak
2017 shortage early phase	37x	Market paying up before shortage earnings fully appeared

The 37x multiple is not a neutral number. It says Samsung Electro-Mechanics is entering a shortage cycle similar enough to 2017 that the market may again pay a peak shortage multiple before all earnings have been printed.

That may be right. It may also be too aggressive. A 37x multiple on a company whose 10-year average is closer to 15x requires a very specific regime: tight supply, visible pricing power, customer prepayments or volume commitments, and margin expansion that survives more than one quarter.

This is why the article should be read as a conditional re-rating framework, not as a mechanical target-price endorsement.

3. Where the Current Price Already Sits

Samsung Electro-Mechanics closed at KRW 917,000 on May 7, 2026 in the local Research OS market snapshot. Against the same 2028F EPS of KRW 34,764, that close implies:

Current implied PER = 917,000 / 34,764 = 26.4x

The scenario table looks like this:

Scenario	PER	Implied price	Distance from KRW 917,000
Bear case, EPS -20% and 19x	19x	about KRW 528,000	-42%
2021-style upcycle	22x	about KRW 765,000	-17%
Normal upper-band upcycle	25x	about KRW 869,000	-5%
Current close	26.4x	KRW 917,000	0%
2018 shortage peak	28x	about KRW 973,000	+6%
Mirae Asset frame	37x	about KRW 1,286,000	+40%

This is the most important practical point. The stock no longer prices Samsung Electro-Mechanics as a simple normal-cycle component recovery. At KRW 917,000, it is already above the normal upper-band frame and close to the 2018 shortage-peak frame.

From here, further re-rating depends less on “earnings are improving” and more on “the market accepts a 37x shortage multiple.”

4. Are the Shortage Signals Real?

The shortage indicators cited in the local report flow are meaningful:

Indicator	Normal level	Reported current level	Read-through
Some MLCC lead times	about 10 weeks	20-24 weeks	Lead time has roughly doubled
2Q26 MLCC utilization	not specified	around 95%	Close to full capacity
Inventory coverage	about 6 weeks	about 4 weeks	Inventory is being drawn down

These are valid shortage signals. Lead times stretch, utilization rises, and inventory falls. That combination is what gives price increases a path.

But the wording matters. The lead-time commentary refers to some global MLCC products, not necessarily every MLCC category. AI server MLCCs can be tight while commodity IT MLCCs remain less tight. The difference matters because the 2017 shortage cycle was broad. The 2026 cycle may be more AI-server-specific.

Samsung Electro-Mechanics’ own 1Q26 announcement supports the direction of demand. The company reported consolidated 1Q26 revenue of KRW 3.2091 trillion and operating profit of KRW 280.6 billion, with AI server, power and network demand driving growth. It also said demand for high-value-added FCBGA substrates for AI, servers and networks should remain strong in 2Q.

That is enough to say the AI-infrastructure leg is real. It is not enough to assume the whole MLCC market has already become 2017 again.

5. MLCC Price Hikes Are the Swing Factor

Mirae Asset’s upside logic depends heavily on AI server MLCC pricing. The reported scenario analysis frames the operating-profit impact roughly as:

AI server MLCC price increase	2026F OP uplift	2027F OP uplift	2026F component OPM	2027F component OPM
Base, no price hike	n/a	n/a	15.2%	16.3%
+15%	KRW 100.8B	KRW 184.5B	16.8%	18.9%
+20%	KRW 134.4B	KRW 245.9B	17.4%	19.8%
+25%	KRW 168.0B	KRW 307.4B	17.9%	20.7%

This is why 2Q26 matters. If AI server MLCC ASPs begin to move visibly, the shortage multiple becomes easier to defend. If utilization is high but pricing is slower, the 37x frame becomes more fragile.

The market has already paid for part of the pricing story. It now needs evidence that the price story is not just a sell-side assumption.

6. FC-BGA Is the Structural Leg

MLCC is the shortage signal. FC-BGA is the structural AI substrate signal.

Mirae Asset reportedly raised its FC-BGA ASP assumption by 13%. The logic is that raw materials remain tight, major suppliers are sold out, and customers are willing to support capacity through prepayments, volume discussions and pricing negotiations.

That last phrase should be handled carefully. “Customer support” is different from fully disclosed, legally binding purchase guarantees. Publicly available data does not provide enough detail to verify the size, duration or enforceability of those arrangements. The conservative interpretation is that customer behavior is supportive, but not all of the capacity expansion should be treated as risk-free.

Still, the direction is important for the broader Korea AI PCB thesis. If Samsung Electro-Mechanics can lift FC-BGA ASP assumptions by 13%, that is a read-through for the rest of the Korean substrate ecosystem:

Korean node	Read-through
Daeduck Electronics	FC-BGA and MLB exposure benefit if pricing remains firm
Korea Circuit	FC-BGA option becomes more valuable if high-end substrate demand persists
Isu Petasys	AI network MLB demand remains a parallel bottleneck
Doosan Electronic BG	High-end CCL pricing power is supported by downstream substrate tightness
Kolon Industries and Pamicell	Low-loss material demand becomes more durable if CCL tightness persists

This is why the report matters beyond Samsung Electro-Mechanics. It is one of the clearest large-cap confirmations that AI substrates are moving from a narrative into price, capex and margin numbers.

7. 1Q26 Was the First Accounting Checkpoint

Samsung Electro-Mechanics’ official 1Q26 numbers:

Item	1Q26
Revenue	KRW 3.2091T
Operating profit	KRW 280.6B
Revenue growth	+17% YoY
Operating-profit growth	+40% YoY

Local coverage also notes a KRW 71.4B one-off retirement-benefit cost tied to changes in wage-base accounting. Excluding that item, underlying operating profit is discussed around KRW 351.4B, which is materially stronger than the headline number.

The headline result already validates demand. The adjusted result validates the margin argument more clearly. The next checkpoint is whether 2Q26 shows:

Checkpoint	Why it matters
AI server MLCC price realization	Confirms that shortage can become earnings, not just lead time
FC-BGA ASP and margin	Tests the +13% ASP assumption
Package-solution OPM	Shows whether the business is moving toward mid-to-high-teens margins
Customer commitment details	Distinguishes demand-backed capex from ordinary capex risk

8. The Stock Is No Longer Undiscovered

Mirae Asset’s KRW 1.3 million target is the high-end call, but the rest of the street has also moved up. Reported target prices include roughly:

Broker	Target price
Hana Securities	KRW 1.0M
NH Investment & Securities	KRW 1.0M
Meritz Securities	KRW 1.02M
iM Securities	KRW 1.1M
KB Securities	KRW 1.1M
Mirae Asset Securities	KRW 1.3M

That matters. Samsung Electro-Mechanics is not a hidden second-derivative AI name anymore. It has been discovered by the market, and the valuation already reflects part of that discovery.

The May 7 flow also shows a divided tape: foreign investors were net sellers by roughly KRW 70.3B while institutions were net buyers by roughly KRW 47.2B in the local Research OS snapshot. That does not break the thesis, but it argues against treating the trade as a clean one-way crowding story.

9. What Could Break the 37x Frame?

The upside frame is clear. The failure paths are also clear.

Risk	Why it matters
MLCC tightness stays limited to a narrow product set	A narrow shortage deserves a lower multiple than a broad 2017-style shortage
AI server MLCC price increases lag expectations	Lead times do not automatically become margin
FC-BGA ASP lift is offset by depreciation or material costs	Capex-backed growth can dilute margins if pricing does not keep up
Customer prepayment and volume-support details remain vague	The market may discount the capex story if contract quality is unclear
Consensus becomes one-sided	With most brokers already constructive, disappointment can travel quickly
2028 EPS proves too far forward	Pulling valuation two years ahead increases duration and forecast risk

The most important distinction is between a real shortage and a priced-in shortage. Both can be true at the same time. The company can be in a strong AI component cycle while the stock already discounts a large part of it.

Final Note

Mirae Asset’s KRW 1.3 million target for Samsung Electro-Mechanics is not just an earnings upgrade. It is a claim about market regime. The claim is that Samsung Electro-Mechanics has moved from a mobile/electronics-component cycle into an AI-infrastructure bottleneck cycle, and that the market can value this cycle with a 2017-style MLCC shortage multiple.

The evidence is not imaginary. Samsung Electro-Mechanics reported a strong 1Q26, AI server and network demand is visible, FC-BGA pricing assumptions have moved up, and MLCC lead-time, utilization and inventory data point to tightness in high-end products.

But the price has already moved into a demanding frame. At KRW 917,000, the stock is around 26.4x Mirae Asset’s 2028F EPS, already above a normal upper-band upcycle multiple and close to a 2018 shortage-peak frame. The remaining re-rating requires the market to accept something closer to 37x.

For the broader Korea AI PCB thesis, this report is useful even if one does not accept the full 37x multiple. It confirms that FC-BGA, MLCC, CCL and low-loss materials are being discussed in pricing, capacity and margin terms, not only as a theme. That strengthens the logic behind the AI PCB hub, Daeduck, Korea Circuit, Doosan Electronic BG, Kolon Industries and Pamicell as linked parts of the same system bottleneck.

The next evidence window is 2Q26: AI server MLCC price realization, FC-BGA ASP, package-solution margins and more concrete customer-commitment language. Until then, the cleanest interpretation is this: the business has improved, the shortage signals are real, and the stock now requires a peak-shortage multiple to unlock the final leg of Mirae Asset’s target-price math.

Source notes: Samsung Electro-Mechanics’ official 1Q26 announcement reported KRW 3.2091T revenue and KRW 280.6B operating profit, with AI server, power and network demand supporting growth. MoneyToday/Daum coverage of Mirae Asset’s May 6 report cited the KRW 1.3M target, 13% FC-BGA ASP assumption lift, 2028 valuation-year shift, 20-24 week MLCC lead times and 95% utilization / 4-week inventory framework. Local Research OS market data confirms the May 7, 2026 close at KRW 917,000.

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 Electro-Mechanics: Why the Stock Doubled in 90 Days

Tue, 21 Apr 2026 10:00:00 +0900

AI substrate map: Samsung Electro-Mechanics is the premium anchor inside the AI PCB and Substrate Hub. For the broader system-bottleneck framing behind FC-BGA and MLCC, read the AI PCB and Substrate Thesis.

The 90-day double wasn’t theme-driven. Three structural shifts locked in simultaneously — pricing power transfer, customer-mix inflection, and a secured next-gen option. That’s what got re-rated.

TL;DR

Why the stock moved: (1) FC-BGA entered Nvidia’s Vera Rubin ecosystem via NV Switch and then as first vendor on Groq3 LPU (Q2 2026 mass production) — a confirmed slot in the AI-server package-substrate value chain; (2) MLCC pricing cycle turned as AI-server content exploded (~600,000 MLCCs per VR200 NVL72) and Murata / Taiyo Yuden led hikes; (3) the KRW 1.8T Vietnam expansion and Sumitomo glass-substrate JV made the 2027+ growth axis visible.
Investor framing: A parts-and-substrate leverage play on a Jevons-paradox AI infrastructure cycle. Not a cyclical recovery — a structural mix shift plus pricing-power handover. Event-driven (1Q26 print, Murata price call, Groq3 ramp) overlapping with quality-compounder optics.
Next action: watch 1Q26 results late-April → component and package OPM trajectory → formal Murata price-hike date in 2Q as the confirmation event. Hold, or scale in on pullbacks.

1. Header

Item	Detail
Company	Samsung Electro-Mechanics
Ticker	009150.KS (KOSPI)
HQ	Suwon, South Korea
Business model	Materials (MLCC) + OSAT/Packaging (FC-BGA) composite, plus optical solutions (cameras)
Reference date	2026-04-21
Primary sources	SEMCO 4Q25 results and conference call, sell-side reports (Shinhan, Mirae Asset, Meritz, DB, KB, Samsung), trade press (The Elec, ETNews, ZDNet Korea), CEO Jang Duk-hyun’s AGM and CES 2026 remarks, Bloomberg

2. Executive Snapshot

Core products: (1) high-capacity, high-temperature MLCC for AI servers, (2) FC-BGA for AI accelerators (NV Switch, Groq3 LPU, AMD MI series).
Numbers: 2025 revenue KRW 11.3T (record); operating profit KRW 913.3B (+24.3% YoY). 2026E consensus OP: KRW 1.1–1.28T. Component OPM structurally rebounding.
Moat summary: Global #2 in MLCC (duopoly with Murata). Korea’s first server-grade FC-BGA volume producer (since 2022) and one of only a handful of high-end server-substrate capable names globally. In-house know-how in fine patterning, large-area and high-layer-count substrates, low-loss materials.
Trend vector: Strong tailwind. AI servers need ~20× the MLCC content of smartphones (≈1,000 per phone vs. ~600,000 per AI server). FC-BGA large-area, high-layer trend is consuming supply capacity.
Quantum-jump triggers: (a) A 10% MLCC ASP hike carries ~KRW 600B operating-profit leverage; (b) glass-substrate mass production entering 2027; (c) graduation to Nvidia Tier-1 supplier.
Investor stance: Technology leadership + event-driven + valuation re-rating — all overlapping. Short-term overheated zone, but the structural shift is still mid-play.

3. Non-specialist read

Samsung Electro-Mechanics is one of the very few companies that supplies both the skeleton (FC-BGA substrate) and the blood-pressure regulator (MLCC) of an AI chip. If Nvidia’s GPU is the brain, the AI server cannot function without the substrate that carries it and the MLCCs that feed stable power to it.

Until recently SEMCO was seen as a Samsung proxy — PC CPU substrates plus flagship-phone MLCC meant earnings tracked Galaxy volumes. Between 2024 and 2026 the customer mix shifted structurally: PC to AI server/data center, smartphone to AI server and automotive, and single-customer dependence to diversification across Nvidia, AMD, Tesla, Broadcom, Apple, and AWS. At the same time supply is not keeping up with demand — CEO Jang himself said demand exceeds capacity by 50%+ — so pricing power is migrating from buyers to SEMCO. The stock is pricing that transition.

4. Revenue Mix, Margins, Growth

4-A. Segment revenue (FY2025, KRW 100M)

Segment	Key products	2025 rev	Mix	YoY
Components	MLCC, power inductors, chip resistors	~46,000	~41%	+low teens
Package Solutions	FC-BGA, BGA, FC-CSP	~22,000	~19%	Server FC-BGA +17%
Optical Solutions	Camera modules, comms modules	~37,000	~33%	+9%
Other / automotive	—	~8,000	~7%	—
Total		113,145	100%	+9.9%

[Inferred] Annual segment figures are cumulative from quarterly filings. 4Q25 standalone components revenue of KRW 1.32T was a single-quarter record.

4-B. Customer concentration (disclosable range)

Metric	Value	Note
Samsung Electronics revenue dependence	<30% (2025)	[Fact] Structural decline: 60%+ historically → mid-40s in 2024 → sub-30% in 2025
Customer diversification	Nvidia, AMD, Tesla, Broadcom, Apple, AWS — “Big 6” secured	[Fact] DB Securities report
Customer-level revenue	Undisclosed	NDA-bound. Big-6 lock-in is the core intangible asset in FC-BGA

4-C. Margins and growth

	2023	2024	2025	2026E	Read
Revenue (KRW T)	~9.4	~10.3	11.31	12.3–12.8E	3Y CAGR ~10%
Operating profit (KRW 100M)	~6,300	~7,343	9,133	11,000–12,800E	[Inferred] sell-side range
OPM	~6.7%	~7.1%	8.1%	9–10%E	Structural lift from ASP + mix
Component OPM	~13%	~14%	~16%	18–20%E	MLCC ASP the driver
Package Solutions OP (KRW 100M)	~1,355	~2,000	3,861E (Mirae)	5,046E	FC-BGA +10% ASP embedded

Data gap: Customer-level revenue fully undisclosed. Within-segment splits of AI/server vs. IT vs. automotive are qualitative commentary only.

5. Origin Technology and Technical Moat

Core IP stack

Area	Core tech	Difficulty / specificity
MLCC	Ultra-thin multilayer (hundreds of layers), BaTiO₃ particle grading and purity, 1kV+ high-voltage parts	Material and process know-how plus mass-production yield
FC-BGA	Fine-line patterning, large area (140×140mm+), 20+ layers, low-loss CCL materials	[Fact] Very small global group of high-end server-substrate volume producers
Glass substrate	In-house glass core (Sumitomo JV), TGV pitch, CTE control	[Fact] Warpage reduced -30% at room temperature and -90% at high temperature vs. organic
Camera modules	Folded-zoom actuators, hybrid lenses, automotive-grade modules	—

Moat mechanics

Performance: On the capacitance/voltage curve MLCC is a Murata-SEMCO duopoly. For 1kV+ AI-server-grade parts there is effectively no alternative outside these two.
Yield / cost: Large server FC-BGA (4× area, 2× layers) has catastrophic yield if warpage control fails. Accumulated R&D plus dedicated Busan and Vietnam lines (~KRW 1.9T cumulative CAPEX plus another KRW 1.8T in Vietnam) create a cost-and-time barrier to followers.
Qualification: Big-6 supplier references (Nvidia, AMD, Tesla, Broadcom, Apple, AWS) form an infinite loop for new entrants — no qualification without references, no references without qualification. [Inferred] DB Securities cites this as the basis for “visible steady growth.”
Ecosystem lock-in: The glass-substrate JV (with Sumitomo and Dongwoo Fine-Chem) internalizes the glass core, a potential cost-advantage lever vs. SKC’s Absolics. [Fact] SEMCO holds majority of the JV.

Hard-to-copy vs. copyable

	Details
Moat	① High-voltage, high-capacity MLCC for AI servers (decades of materials-process compounding); ② Server FC-BGA large-area, high-layer yield (equipment + experience compound barrier); ③ Big-Tech reference book.
Catch-up possible	① PC-grade FC-BGA (Taiwan Unimicron, Japan Ibiden competitive); ② commodity IT MLCC (Taiwan Yageo, Korean Samwha); ③ camera modules (LG Innotek, Sunny Optical).

6. Customer Status

FC-BGA customer matrix

Customer	Status	Product / use	Source	Confidence
AMD	Mass production	HPC server CPU/GPU (presumed MI series)	[Fact] Official supply announcement, July 2024	High
Nvidia (NV Switch)	Mass production (ramped 2H25)	Intra-server GPU interconnect switch	[Fact] Supply-chain entry disclosed Feb 2026	High
Nvidia (Groq3 LPU)	First vendor, mass production in 2Q26	Inference accelerator for Vera Rubin platform	[Fact] ZDNet Korea exclusive, Apr 8 2026	High
Tesla (AI6)	In progress, adoption likely	Tesla in-house AI chip	[Inferred] Industry observation	Medium
Broadcom	In progress (glass substrate samples)	Custom ASIC	[Fact] Glass-substrate sample shipment reported	Medium
Apple	In progress (under discussion)	Next-gen silicon + glass substrate	[Fact] Apple-glass substrate talks reported June 2025	Medium
AWS	In progress	Graviton / Trainium and other in-house silicon	[Inferred] Included in “Big 6” mentions	Medium
Samsung Foundry	Mass production	Samsung SoC substrates	[Fact] Captive	High

MLCC customers (many-to-many by nature)

AI server: Nvidia VR200 NVL72 — ~600,000 MLCCs per server, 30%+ more than GB300. [Fact] Samsung Securities (Lee Jong-wook)
Automotive: ADAS (3,000+ per car), xEV (10,000–15,000 per car)
Smartphone: Galaxy flagship, Chinese OEMs

Silicon capacitor (new)

Marvell supply started June 2025. [Fact]
Purpose: high-frequency decoupling for CPU/AP, extendable to AI servers.

7. Trend Impact Matrix

Trend	Impact	Reason
AI server CAPEX expansion (hyperscalers)	Strong tailwind	Tens of thousands of MLCCs per AI server, FC-BGA mandatory. Hyperscaler custom ASICs (Groq, Trainium, MAIA) diversify accelerator-substrate demand.
AI accelerator large-area, high-layer trend	Strong tailwind	[Fact] Nvidia Rubin Ultra carries 16 HBM stacks → substrate enters 140×140mm+ → CCL limits push glass-substrate adoption. SEMCO’s server FC-BGA benefits from 4× area, 2× layer-count demand.
Chiplets / Advanced packaging	Tailwind	2.1D, co-package, silicon interposer and glass interposer alternatives are all in the SEMCO portfolio. For CoWoS-class flows, OSAT primes (TSMC, ASE) capture the lead.
MLCC pricing cycle	Strong tailwind	[Fact] Taiyo Yuden May hike notice delivered, Murata price review formalized, AI-server MLCC lead times 20+ weeks. Oligopoly → chain reaction.
Automotive / ADAS / EV	Tailwind	Rising MLCC count per vehicle; SEMCO share gains in the automotive grade.
Humanoid / Physical AI	Neutral → tailwind (long-term)	CEO Jang at CES 2026 flagged actuator investments. Materializes post-2027 as a camera + MLCC + FC-BGA triangle.
Export controls / geopolitics	Neutral	US China-semi restrictions hurt some customers (Chinese phones) but are diluted by SEMCO’s AI-server and Big-Tech mix shift. [Inferred] Chinese MLCC entrants remain a mid/long-term low-to-mid-end risk.
Glass-substrate transition	Strong tailwind (from 2027)	SEMCO’s JV aims for ecosystem leadership. [Unclear] Whether Absolics / Ibiden / DNP will win the final mass-production race is not yet decidable.
Mobile cycle softness	Mild headwind	Optical solutions growth flattish; folded-zoom and hybrid lenses defend ASP.
Raw-material prices (CCL, nickel, copper)	Neutral	Price hikes exceed input inflation, so real margins improve (Mirae Asset).

8. Price-Driver Dissection

This section separates already-priced-in triggers from expected triggers. The ~100% move between February and April 2026 (KRW 400k → KRW 680k+) reflects simultaneous progress on the following.

Trigger 1. Nvidia supply-chain Tier-1 entry (event-driven, priced in and progressing)


Definition	NV Switch → Groq3 LPU first-vendor status broadens the substrate footprint inside Nvidia’s AI platform.
Preconditions	Server FC-BGA mass-production record + AMD precedent + yield validation
Confirmed markers	(i) Feb 2026 NV Switch supply confirmed; (ii) Apr 2026 Groq3 LPU first vendor, 2Q26 mass production; (iii) Tesla AI6 likely to adopt
Expected effect	Package Solutions OP: KRW ~200B (2025) → KRW ~386B (2026) → KRW ~505B (2027) per Mirae Asset
Risk	(a) Nvidia GPU shipment slip; (b) Unimicron / Ibiden share defense; (c) Groq3 real volume below plan

Trigger 2. MLCC pricing cycle (event-driven, partly priced in)


Definition	Murata / Taiyo Yuden hikes transmitted through to SEMCO
Preconditions	(i) Lead time 20+ weeks [Fact]; (ii) SEMCO utilization 90–95% full [Fact]; (iii) Murata flagship MLCC demand 2× capacity
Leading indicators	Taiyo Yuden May 1 hike notice sent / Murata late-April earnings may formalize hike / 2Q26 double-digit hike plausible (The Elec)
Expected effect	[Inferred] Mirae: AI-server MLCC 15–25% hike = incremental OP in the hundreds of billions KRW. Sensitivity: 10% MLCC ASP = +KRW 600B annual OP
Risk	(a) Murata reversal; (b) China / Taiwan low-end pressure; (c) raw-material surge diluting realized margin

Trigger 3. FC-BGA capacity + full-utilization pricing power (progressing)


Definition	KRW 1.8T additional Vietnam investment + full utilization in 2H26 + 10% ASP increase
Preconditions	Demand 50%+ above capacity to persist
Leading indicators	[Fact] Apr 14 2026 Vietnam investment-registration certificate issued / [Fact] CEO Jang: “full utilization expected in 2H26” / Prismark: FC-BGA 2025–2030 CAGR 15%+
Expected effect	Large-area, high-layer transition lifts ASP, capacity-constraint dynamics drive price and volume simultaneously
Risk	(a) Demand air-pocket post-capacity-add; (b) LG Innotek server-substrate entry (2027); (c) raw-material bottleneck

Trigger 4. Glass substrate commercialization (long-term, partial pre-ride)


Definition	Mass production from 2027 + Big-Tech sample qualification, constraining SKC Absolics
Preconditions	(i) Sumitomo JV completion by 1H26 [Fact, Jang call]; (ii) glass-core in-sourcing; (iii) sample qualification
Leading indicators	Development org moved into Package Solutions (Feb 2026) / AMD, Broadcom sample shipment [Fact] / Apple, Intel, Tesla talks in progress / Intel reported to be considering exit from glass-substrate business
Expected effect	[Inferred] If Nvidia Rubin Ultra (2027) onward adopts glass interposers, SEMCO-SKC becomes a global supplier duopoly candidate
Risk	[Unclear] (a) Brittleness / micro-defect reliability unresolved (per The Elec); (b) Intel / AMD glass-substrate adoption delay; (c) SKC Absolics first-mover production lead

Trigger 5. Humanoid / Physical AI (exploratory, optionality)

CEO Jang at CES 2026: “optimize camera, MLCC, FC-BGA for robotics.” Actuator investments under review.
2025–2026 cash-flow impact immaterial. Option-value bucket. Partial embedded premium from the long-run narrative. [Inferred]

9. Risks & Watchlist

Technology risks

Glass-substrate reliability: Brittleness, micro-defect, seware phenomenon unresolved. Production delay would compress the valuation premium.
FC-BGA large-area yield: Warpage and yield on 140×140mm+ can regress. Absolute yield gap vs. Ibiden not publicly resolvable [Unclear].

Business / customer risks

Rising Nvidia-platform dependence: NV Switch + Groq3 concentrate single-customer exposure. Vera Rubin shipment slippage is a direct hit.
Chinese MLCC pursuit + capex load: Vietnam KRW 1.8T + Sejong glass substrate capex stacked → FCF pressure. If end-market peaks, depreciation becomes a headwind.

Macro / regulatory risks

US-China semi trade escalation: Direct hit to SEMCO is limited, but if Nvidia China-export restrictions tighten, GPU shipment adjustments translate indirectly to FC-BGA volumes.

Next-quarter / half checkpoints (decision-changing data)

#	Checkpoint	Timing	Meaning
1	SEMCO 1Q26 results (Components / Package OPM)	Late Apr 2026	First quarter to confirm ASP + mix flow through numerically
2	Murata results + formal price hike	Late Apr 2026	Formal MLCC cycle confirmation (gate for #2 SEMCO chain hike)
3	Groq3 LPU 2Q mass-production entry	May–Jun 2026	Nvidia first-vendor revenue contribution begins
4	Glass-substrate JV finalization	1H26	Next-gen axis commercialization progress
5	Vietnam capex spend + 2027 ramp plan	2H26	Supply bottleneck resolution → durability of ASP rise

10. Sources

Disclosures / IR: SEMCO 4Q25 results and full conference-call transcript (The Elec)
Official releases: CEO Jang CES 2026 remarks, AGM remarks, Apr 14 2026 Vietnam investment (Bloomberg)
Korean sell-side: Mirae Asset (Park Jun-seo), Shinhan (Oh Kang-ho), Meritz, DB (Cho Hyun-ji), Samsung (Lee Jong-wook), KB, Daishin (Park Kang-ho)
Industry press: The Elec, ZDNet Korea, ETNews, Newspim, Seoul Shinmun, Finance Post

Conflicting or undisclosed data

Customer-level revenue: fully undisclosed; only the Big-6 lock-in is confirmable.
Glass-substrate production timing: CEO Jang’s language has shifted from “2026–2027” (2024) → “2027–2028” (2025) → “2027” (2026). [Inferred] Realistic base case is 2H27+.
MLCC hike magnitude: Industry sources talk “double-digit within 2Q” while some reports cite a 10–20% range; final figure hinges on Murata announcement.
Early rally context: KB 460k target → Meritz / Shinhan 700k target in four months, a +50%+ target-price revision wave. The consensus itself is being re-formed dynamically.

Bottom line

The three-month double is not simply an “AI-tailwind beneficiary” tag. It reflects three structural confirmations landing together:

Pricing-power transition: SEMCO from price-taker to price-setter (MLCC oligopoly + FC-BGA ≥50% supply shortfall).
Customer-mix quality step-up: Samsung dependence from 60%+ to diversified Big-6 (Nvidia, AMD, Tesla, Broadcom, Apple, AWS), and specifically a Tier-1 slot inside the Nvidia platform as a reference asset.
Next-gen option secured: glass-substrate JV positions SEMCO alongside SKC Absolics as the only Korean supply options.

Caveat: at ~2.8× PBR and ~18× PER (on 2025), valuation is not yet at cyclical-peak extremes, but 2026E OP of KRW 1.1–1.3T is already in consensus. If earnings surprise underwhelms, the drawdown can widen. 1Q26 results and the pace of ASP pass-through are the key pivot.

From a Jevons-paradox lens, AI-inference efficiency gains (TurboQuant, DeepSeek, etc.) are more likely to redirect into aggregate server-deployment growth than to depress MLCC / FC-BGA demand. At the part-and-substrate layer the mid-term demand curve should stay upward-sloping — a leverage vector that works opposite to the software-efficiency theme.