Samsung Electro-Mechanics' KRW 1.5T Silicon-Capacitor Win — A Reclassification, Not a Single Order. But KRW 967K Already Prices Most of It In

Thu, 21 May 2026 00:00:00 +0000

📚 Samsung Electro-Mechanics series Part 1: SEMCO — ‘The Invisible Infrastructure of AI Silicon’ (MLCC / FC-BGA / Optics deep dive, May 15) Part 2: SEMCO Hybrid Challenger — Between Murata and Ibiden, What KRW 1.01M Implies for 2027E OP (May 17) Mirae KRW 1.3M target: The valuation-frame shift (May 7) First re-rating call: Reclassification into an AI-infrastructure components name (Apr 21)

TL;DR

The headline is positive. SEMCO just proved an actual KRW 1.557T order — not a concept — inside the AI-packaging value chain.
The real point is reclassification. From “MLCC + camera + substrate” to “power-integrity component company sitting inside the AI-package.”
The market priced it immediately. May 20 close KRW 1,061,000 (+7.50%); May 21 pre-market KRW 1,109,000 (+4.52%). April 20 → May 20 = +56% in a month.
But at KRW 967K the stock is already expensive. 2026E PER ~60x. New buys: Wait. Holders: Hold. Re-entry on a pullback to KRW 850–900K or a consensus 2027E EPS lift to KRW 28–30K.
Next checkpoints: named customer, application position (Top / Land / Embedded), 2027 revenue-recognition cadence, yield / CAPEX, Si-Cap margin, and the second / third design wins.

1. The contract in numbers

1.1 The basics

Item	Value	Read
Contract value (KRW)	KRW 1,557.0 billion	Very large by SEMCO new-business standards
Contract value (USD)	~USD 1,035.4 million	At ~1,503.8 KRW/USD
Term	Jan 1, 2027 – Dec 31, 2028	Recognition mostly in 2027–2028
Implied annual revenue	~KRW 778.5 billion / year	Contract ÷ 2 years
vs FY25 revenue	~13.8% total / ~6.9% annual	Per disclosure
Implied quarterly	~KRW 194.6 billion / quarter	Even-recognition assumption

Check: KRW 1,557.0B ÷ 2 = KRW 778.5B. KRW 778.5B ÷ KRW 11,314.5B (FY25 revenue) ≈ 6.9%.

1.2 Versus the global market

The global silicon-capacitor market is estimated at ~USD 2.3 billion in 2026. The contract’s annualized USD 517.7M equates to roughly 22.5% of that figure. Caveat: definitions differ across research houses and recognition is split across 2027–2028, so treat this as a sizing check, not as a literal market-share claim.

1.3 OP sensitivity — margin is not disclosed

Si-Cap margin is not public. The table below is a sensitivity, not a forecast.

Assumed OPM	Annual OP contribution	After-tax NI (24% tax)	EPS contribution (~74.7M shares)
15%	~KRW 116.8B	~KRW 88.7B	~+KRW 1,188
20%	~KRW 155.7B	~KRW 118.3B	~+KRW 1,584
25%	~KRW 194.6B	~KRW 147.9B	~+KRW 1,980
30%	~KRW 233.6B	~KRW 177.5B	~+KRW 2,376
40%	~KRW 311.4B	~KRW 236.7B	~+KRW 3,169

At 30% OPM, this single contract adds ~KRW 2,400 to EPS — about a 10% lift to Hana Securities’ standing 2027E EPS of KRW 22,874. Actual margin depends on early-stage yields, depreciation, customer-price terms, wafer-process cost, and test cost.

2. Price action — May 20 and May 21

2.1 May 20 regular session

Date	Open	Low	High	Close	Chg	Volume
May 19, 2026	1,031,000	957,000	1,040,000	987,000	-4.27%	879K
May 20, 2026	968,000	924,000	1,106,000	1,061,000	+7.50%	1.91M

The disclosure landed around 2 PM. The stock immediately reversed from the morning’s weakness, hit KRW 1,080,000 (+9.42%) intraday, and closed +7.50%. Volume = ~2.17x the prior session.

2.2 May 21 pre-market (NXT)

Time	Price	Change
May 21, 08:23 NXT	KRW 1,109,000	+4.52% vs prior close
vs May 19 close	KRW 1,109,000	+12.36%

Pre-market high was KRW 1,139,000 — a test of the 52-week high (KRW 1,133,000).

2.3 Flows — institutions absorbing foreign profit-taking

May 20 flows: foreigners -111,248 shares / institutions +82,167 shares. That is the pattern of “institutional re-rating absorbs foreign profit-taking” — not a one-day thematic pop, especially when combined with the +56.0% one-month run from April 20 to May 20.

3. Why this is a reclassification event

3.1 The bottleneck Si-Cap solves — power integrity

AI accelerators pull current very fast across hundreds of billions of parallel operations. If voltage swings, the chip mis-computes, throttles down, and heats up. The basic equation is simple.

Voltage swing ≈ L × di/dt + I × ESR L = ESL inductance, ESR = equivalent series resistance, di/dt = instantaneous current change

AI chips have very high di/dt, so the only fix is to put a storage capacitor as close to the chip as possible and minimize ESL / ESR. A silicon capacitor is thinned to ≤100 µm and placed Top-Side, Land-Side, or Embedded into the substrate. That is what relieves the bottleneck.

If a board-level MLCC is a “battery a few rooms away,” a Si-Cap is “the emergency reservoir taped to the chip’s leg.”

3.2 Aligned with the TSMC direction

TSMC describes integrating deep-trench capacitors into the silicon interposer for CoWoS-S, and integrating eDTC beneath the SoC for CoWoS-L for power management. The direction of advanced packaging is exactly “signal routing + HBM linkage + power-stabilization components, all integrated.” SEMCO is now in one of those layers.

3.3 Why this matters specifically for SEMCO — only when bundled with FC-BGA

The real value is not Si-Cap as a stand-alone line. It is becoming an integrated supplier of FC-BGA + MLCC + Si-Cap.

MLCC only: faces Murata head-on.
FC-BGA only: faces Ibiden / Shinko head-on.
MLCC + FC-BGA + Si-Cap as a bundle: the only player globally with all three.

From the customer’s perspective, “design my substrate and my power-stabilization parts together” is dramatically more attractive — and once designed in, it’s hard to swap out. That lock-in is the real moat.

3.4 High qualification barrier — sticky once in

Si-Cap reliability is tied to overall AI-package reliability. A defect is not a part-level issue; it becomes a GPU / ASIC package-level reliability issue. So qualification is long, and design-in is hard to reverse. Reporting notes the field has been concentrated among a small number of qualified suppliers precisely for this reason.

4. Customer reading — separate fact, inference, and unknown

[Fact] The counterparty is not disclosed. Whether it is NVIDIA, Marvell, Broadcom, Amazon, Microsoft, or Google cannot be confirmed today.

[Fact] Earlier reporting indicated SEMCO supplies silicon capacitors into Marvell’s AI-accelerator multi-die packaging platform, with the Marvell program reportedly entering mass production in 1Q 2025.

[Inferred] Given the deal size and the 2027–2028 supply window, the end use is almost certainly AI-server GPU / ASIC or large data-center silicon, not mobile APs. Three reasons:

Deal size is too large for a mobile-only program
Si-Cap’s technical need is biggest in AI / HPC packages
SEMCO has publicly framed high-end semiconductor packaging and AI-server use as the target end markets

[Unknown] It would still be premature to call this “an NVIDIA contract” or any specific hyperscaler. Public documents do not distinguish whether the immediate counterparty is a fabless designer, a hyperscaler, or a packaging-house intermediary.

5. What the current price prices in — and what it doesn’t

5.1 Where we are

Item	Value
May 20 close	KRW 1,061,000
May 21 pre-market	KRW 1,109,000
Market cap	~KRW 72.2 trillion
Shares out	74,693,696
2026E revenue	~KRW 13.36 trillion
2026E OP	~KRW 1.57 trillion
2026E EPS	~KRW 16,315
2026E PER	~60x

This is not a cheap stock. The disclosure is not a “buy because it’s cheap” story — it is a “the high multiple now has more earnings visibility behind it” story.

5.2 Multiples are stretched; the swing variable is the EPS-revision pace

SK Securities lifted its target from KRW 1.10M → KRW 1.50M, citing FC-BGA undersupply, larger AI-package area, ASP gains, MLCC improvement, and Si-Cap / glass-substrate differentiation. 2027E OP at KRW 2.353T.
Hana Securities (April) set KRW 810K = 2027E EPS KRW 22,874 × global FC-BGA peer 35.4x.

To cleanly justify the KRW 970K–1.10M zone, 2027E EPS needs to reach the KRW 28–30K band. That requires all three of: this Si-Cap contract recognizing on schedule, FC-BGA price hikes, and the MLCC up-cycle continuing.

5.3 Bands and calls

Price band	Read	Call
850K–900K	Implies 2027E EPS 24–26K at 35x	High buying interest
950K–1,050K	Si-Cap deal starts being priced in	Conditional buy on dips
1,061K (now)	After a one-month +56%	Wait / Hold
1,100K–1,130K	52-week-high test	Chasing inefficient
Above 1,130K	Without a fresh catalyst, short-term overheat	Trim or take profit

6. Trading call — split for holders and new entrants

6.1 Existing holders

Call: Hold.

Fundamentals are intact and stronger. The KRW 1.5T contract reduces the de-rating risk one notch. But additions should follow at least one of:

2Q or 3Q earnings showing FC-BGA / MLCC ASP gains
A new Si-Cap customer disclosure
Qualitative or quantitative evidence of Si-Cap OPM at 25–30%+
Consensus 2027E EPS lifted above KRW 28,000

6.2 New buyers

Call: Wait.

Reason: a stock that is +56% in a month and is now stacking a KRW 1.5T positive on top is a poor entry point. Two cleaner entries:

Price: a pullback into the KRW 850K–900K zone. At 2027E EPS KRW 24–26K, that band offers a 35x entry.
Earnings: consensus 2027E EPS lifted to KRW 28K+ AND price held under KRW 1.0M. That combination earns the AI-package premium honestly.

6.3 Thesis-kill conditions

Re-evaluate the call on any of:

Amended disclosure reducing contract size or duration
2027 Si-Cap recognition slipping into 2028
Confirmed Si-Cap OPM under 20%
FC-BGA or MLCC pricing-cycle rollover
Single-customer concentration above 70% with no visibility on 2028+ renewal
Yield problems following large CAPEX

7. Six- to twelve-month checklist

Application position — Top Side / Land Side / Embedded? Embedded carries the largest FC-BGA bundling value.
Customer profile — GPU fabless / AI-ASIC fabless / hyperscaler / packaging house?
Revenue-recognition cadence — even in 2027, or skewed into 2028?
Margin — above the legacy MLCC / package-substrate blended margin?
CAPEX and yield — do mass-production yields hold without distorting operating leverage?
Second / third customer — does another AI design-win land?
FC-BGA bundling — is Si-Cap a stand-alone supply, or part of a substrate + power-integrity solution?
Consensus 2027E EPS path — does it move from KRW 22,874 toward KRW 28,000+?

8. How this links to other posts

SEMCO — ‘Invisible Infrastructure of AI Silicon’ (May 15) — established the three-division framing (Component / Package Solution / Optics). This article adds a fourth product line to the Component division.
SEMCO Hybrid Challenger (May 17) — the “what KRW 1.01M implies for 2027E OP KRW 2.7T” framing. The KRW 1.5T Si-Cap contract contributes roughly KRW 100–200B of incremental OP at base-case margin assumptions.
Mirae KRW 1.3M target — valuation-frame shift (May 7) — the “if SEMCO is reclassified as a global AI-component peer, KRW 1.3M is reachable” framing. This article is the first hard evidence of that reclassification.
First re-rating call — reclassification into an AI-infrastructure components name (Apr 21) — the bet placed one month ago; +56% over the month and a KRW 1.5T contract today is the validation.

9. One-line read for non-specialists

AI silicon is no longer just a “make the chip smaller” race. GPUs and ASICs draw so much electricity, so quickly, that the part that delivers stable power inches from the chip has become a bottleneck. Silicon capacitors fill that role.

A regular capacitor is a backup battery in another room. A silicon capacitor is the emergency reservoir taped to the chip’s leg. AI chips that suddenly want more current need that reservoir, or voltage shakes, performance drops, errors rise, and heat climbs.

For SEMCO, that means a category change. Historically it was an MLCC + camera + substrate company. This KRW 1.5T contract says SEMCO is now inside the AI-package itself. If it stacks with more customers and FC-BGA bundling, the multiple moves from “parts maker” toward “AI-package core component supplier.”

But the stock is already trading at KRW 967K–1.11M. It did not just become a good company — it is already priced like one. Chasing is inefficient; pullbacks are the entry.

10. The one-line bottom line

This disclosure reclassifies SEMCO from “smartphone parts maker” toward “AI-packaging core-component platform.” The direction is clearly right. But the price already reflects part of that shift. From here, what matters is not the headline quality — it is the pace of the consensus EPS revisions.

This is not a stock that just became a good company. It is a stock already priced like a good company. Chasing is inefficient; pullbacks are the opportunity.

This article is research and commentary only and is not investment advice. The KRW 1,557.0B contract disclosure (May 20, 2026; term Jan 1, 2027 – Dec 31, 2028) is per the company’s official filing. The USD 1,035.4M figure at FX 1,503.8 KRW/USD is per the disclosure attachment. FY2025 revenue KRW 11,314.5B and OP KRW 913.3B are per the company’s official release. The May 20 close of KRW 1,061,000, May 19 close of KRW 987,000, and May 21 pre-market KRW 1,109,000 are per KRX, Naver Finance, and Investing.com. 2026E revenue / OP / EPS / PER reflect Naver Finance / FnGuide consensus and are subject to change. SK Securities’ KRW 1.5M target, Hana Securities’ KRW 810K and 2027E EPS KRW 22,874 are per each broker’s report. Marvell-related reporting (June 19, 2025 via Yonhap and others) is not evidence identifying the counterparty in this contract. The ~USD 2.3B silicon-capacitor market figure (2026) references Mordor Intelligence. The OPM sensitivity table (15–40%) is the author’s assumption; actual margins are not disclosed. Band-by-band reads, entry conditions, and thesis-kill conditions are the author’s views and are not guaranteed. Global macro variables (US rates, oil, FX, VIX) can independently move the stock. The analysis may be wrong. Data cut-off: May 21, 2026 08:30 KST.

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.

Si-Cap on Korea Invest Insights