Silicon-Capacitor Showdown: AP Memory (6531.TW) vs Murata (6981.T) vs SEMCO (009150.KS)
As of 2026-05-29, intraday. Prices verified against Google Finance, Investing.com, TradingView, and Yahoo before any valuation math.
The one-paragraph answer
These three are not the same trade, and the market treated them as one on May 29. A wave of brokerage demand-revision notes on AI-server MLCC and silicon wafers lit a sector-wide melt-up: AP Memory +8.7%, Murata +12.7%, SEMCO +15.7%, all in a single session, all at or within a few percent of 52-week highs. Strip out the shared momentum and you get three genuinely different propositions. AP Memory is the only pure-play on the actual silicon-capacitor thesis and the only one not in a technical blow-off. Murata is the best business in passives, full stop, but it has roughly doubled off its 50-day average in two weeks and now trades above every prior cycle-peak multiple it has ever printed. SEMCO is the weakest of the three on margins, cash flow, and valuation, and it is the most extended. The right ranking is by risk-adjusted entry quality today, which is AP Memory > Murata > SEMCO, with the loud caveat that none of the three is a buy at these prices. The discipline is to know which one you want when the parabola breaks.
A critical framing note that governs everything below: the consensus price targets these stocks are being measured against are days to weeks stale and are actively being revised upward right now. Read every "X% above mean PT" line as "ahead of pre-revision consensus," not as a fundamental ceiling.
1. At-a-Glance Snapshot
| Metric | AP Memory (6531.TW) | Murata (6981.T) | SEMCO (009150.KS) |
|---|---|---|---|
| Company | AP Memory Technology | Murata Manufacturing | 009150/009150 |
| Sector / Industry | Fabless specialty memory + merchant silicon caps | #1 global MLCC / passives | #2 MLCC + #2/#3 FC-BGA substrate |
| Price (local) | TWD 1,190 | JPY 9,625 | KRW 2,127,000 |
| Change on day | +8.7% | +12.7% | +15.7% |
| Market cap | TWD 194B (~$6.2B) | JPY 17.5T (~$110B) | KRW 161T (~$107B) |
| Enterprise value | ~$5.5B | ~$93.8B | ~$92.7B |
| Revenue (TTM, USD) | ~$217M | ~$11.5B | ~$7.8B |
| 52-week range (local) | 268 – 1,200 | 2,018 – 10,040 | 120,200 – 2,192,000 |
| Distance below 52w high | -0.8% (at high) | -4.1% | -3.0% |
| YTD performance | +139% | +156% | +585% |
| 52-week performance | +290% | +295% | +1,391% |
| Dividend yield | 0.64% | 0.82% | 0.13% |
| Beta | 1.67 | 1.10 | 1.99 |
| RSI (14-day) | 64.3 | 87.2 | 92.7 |
The snapshot already tells most of the story. Murata and SEMCO are now near-equal in size (~$110B vs ~$107B) despite SEMCO carrying lower revenue and roughly half the margins. AP Memory is one-seventeenth their size, which is exactly why its AI optionality is leveraged rather than diluted: a silicon-cap win that would be a rounding error inside Murata can re-rate AP Memory's entire income statement. Note the RSI column: SEMCO and Murata are in textbook blow-off territory (above 90 and 87), while AP Memory at 64 is the only one not technically overextended. The market is treating all three as the same momentum vehicle. They are not.
2. Business Model Comparison
AP Memory is a fabless IC designer with two legs. The base is low-power PSRAM (IoTRAM, ApSRAM, Xccela) for wearables, IoT, displays, and modems, roughly 55-65% of revenue, mature and cash-generative. The thesis leg is stack silicon capacitors (S-SiCap) and silicon-capacitor interposers, an estimated 15-25% of revenue, that sit inside TSMC CoWoS and Intel EMIB packages to deliver clean power to AI accelerator dies. Wafers are outsourced to Powerchip, so capex runs under 1% of revenue and the model is genuinely asset-light. The segment split is not officially disclosed; treat the percentages as analyst inference.
Murata is the TSMC of passives: the world's #1 MLCC maker at roughly 34% global share, vertically integrated from in-house barium-titanate powder synthesis through finished module, with a 50-year process-IP head start that dominates the high-margin tiers (high-cap, ultra-miniature 008004, auto-grade AEC-Q200). MLCCs are about 45% of revenue but over 60% of operating profit; the rest is RF/communication modules (where a JPY 48.9B SAW-filter goodwill impairment recently landed), inductors and sensors, and a marginal battery business. A net-new 48V AI-server power-module line is the only major new product in five years.
SEMCO is Samsung Group's captive components arm and functionally two AI picks-and-shovels stories stapled to a legacy camera-module business. Component Solution (MLCCs, ~44% of revenue, ~70% of segment profit) is the #2 global MLCC franchise. Package Solution (FC-BGA substrates, ~23% of revenue, growing +45% YoY) is the genuine breakout, with 2026 capacity 100% booked by hyperscalers. The drag is Optical & Communication (~33% of revenue), low-margin captive Samsung Galaxy camera modules that suppress blended margin by an estimated 4-6 points. A pure-play SEMCO would show ~12-15% operating margin instead of the reported 8.7%.
| Dimension | AP Memory | Murata | SEMCO |
|---|---|---|---|
| Revenue model | Fabless design + IP | Vertically integrated mfg | Captive + merchant mfg |
| Moat | Qualification + capacity allocation | Process IP + scale + vertical integration | Scale + Tier-1 qual + captive demand |
| Moat strength | Strong-but-narrow (S-SiCap) | Wide and durable | Mid, durable but ceilinged |
| Customer concentration | Moderate, bifurcating | Moderate (Apple ~20%) | Concentrated (Samsung 30-40%) |
| Pricing posture | Niche price-maker | Leader (15-35% hikes) | Follower (considering 5-10%) |
| Silicon-cap exposure | 15-25% of rev (purest) | <1% of rev (incumbent #1 by volume) | ~0% today ($1B deal, no shipments) |
The pricing row matters most for the AI-MLCC thesis specifically. Murata announced 15-35% hikes effective April 1 and the market is following its lead; SEMCO is still only "considering" 5-10%. That gap is the cleanest single proof that Murata's pricing power is real and SEMCO's is derivative.
Takeaway: Murata has the most durable business and the widest runway you do not have to bet a single thesis on. AP Memory has the largest relative runway because of its size and purity. SEMCO's business model carries a structural margin drag (Optics) and a structural governance drag (captive Samsung) that the other two do not.
3. Silicon-Capacitor Scorecard (the specific lens you asked for)
This is the question that started the thread, so it gets its own section. Silicon capacitors are IC-package decoupling devices (EMIB, CoWoS, I-Cube) that work above ~100 MHz where MLCC physics fails. The merchant market is a four-supplier setup: Murata (#1 by volume), SEMCO (emerging), AP Memory (#4 merchant but first-qualified into Intel's EMIB socket), and TSMC's in-house trench caps.
- AP Memory is the only investable pure-play. S-SiCap is 15-25% of its revenue, it holds first-qualified-supplier status for Intel EMIB with Q2 2026 shipments starting, and Gen4 (3.8 uF/mm², +50% density) ramps through 2026. If you want exposure to the silicon-capacitor story specifically, this is the vehicle by a wide margin.
- Murata is the incumbent #1 by volume but it is immaterial to the P&L. Murata runs a defensive silicon-cap line, characterized in the deep-dive as research-stage, "not yet a commercial-scale threat, could be a 5-year story." You do not buy Murata for silicon caps; you buy it for MLCCs and get the silicon-cap incumbency as a free option.
- SEMCO is narrative, not yet numbers. Its $1B silicon-cap deal (disclosed May 21, explicitly framed as challenging the Murata-TSMC duopoly) has zero shipment history and undisclosed ramp timing. At full ramp against ~$9.8B revenue it would be ~10%, but current contribution is essentially zero. Do not let the $1B headline anchor any current valuation.
For silicon-cap purity: AP Memory >> SEMCO > Murata. For silicon-cap incumbency and volume: Murata > SEMCO > AP Memory. They are different questions.
4. Financial Health, Side by Side
| Metric | AP Memory | Murata | SEMCO |
|---|---|---|---|
| Revenue (TTM) | TWD 6.79B (~$217M) | JPY 1,831B (~$11.5B) | KRW 11,785B (~$7.8B) |
| Revenue growth (YoY) | +115% | +11.8% | +17.2% |
| Gross margin | 46.4% | 42.3% | 20.6% |
| Operating margin | 28.0% | 28.1% (reported) | 8.75% (and declining) |
| Net margin | 23.6% | 12.8% | 6.97% |
| FCF (TTM) | TWD 2.07B (~49% margin) | JPY 115B | KRW -168B (negative) |
| Net cash / (debt) | +TWD 11.5B (~$368M) | +JPY 598B (~$3.75B) | +KRW 527B (~$350M, ~0.3% of cap) |
| ROE | 13.3% | 8.83% | 8.81% |
| Capex intensity | <1% of rev | ~11-12% of rev | high, leveraging up |
Three honest distortions to flag, because the headline numbers mislead in all three cases:
- AP Memory's trailing 155x P/E and 23.6% net margin are both distorted by a one-time Q2 2025 charge (NT$684M EBIT loss, likely an inventory write-down or qualification NRE). The clean operating margin is 25-28% and gross margin is structurally 46-51% because Powerchip carries the capex. FCF margin near 49% is real. This is a genuinely high-quality asset-light profile.
- Murata's reported operating margin overstates the through-cycle reality. Organic revenue CAGR is only ~3%; the entire thesis is margin recovery (13.1% trough FY24, to ~18-19% FY26, to 22%+ on the FY27 guide) plus AI mix, not top-line compounding. Its trailing 75x P/E is flattered by the SAW-impairment base. Use the ~40x forward, not the trailing.
- SEMCO's operating margin is moving the wrong way for the thesis. It declined sequentially 10.1% to 9.7% to 8.7% across the last three quarters even as revenue accelerated. The thesis requires an H2-2026 step-up that has not yet appeared. TTM free cash flow is negative, the company is leveraging up to fund Vietnam ABF capacity, and net cash is a trivial 0.3% of market cap. The one real bull data point: Q1 incremental gross margin was ~28% and incremental EBIT ~18%, so new revenue is structurally higher quality than the reported blended figure.
Financial-health ranking: AP Memory > Murata > SEMCO. AP Memory has net cash worth 6% of its cap, no debt, and 49% FCF margins. Murata is a fortress balance sheet with structurally thin FCF (high capex). SEMCO has negative free cash flow and is adding debt; it is not distressed, but it is the weakest of the three and the gap is real.
5. Growth
AP Memory's reported +115% revenue growth is the highest, but its forward number is the one that matters and the one that is heroic. Consensus FY26 EPS of NT$25.80 against FY25's NT$7.67 implies earnings rise 3.3x. A silicon-cap segment that is 15-25% of revenue cannot triple total earnings by itself even if it triples; the forward number quietly assumes net margin expands to ~47-50% and the whole base grows at once. The 46x forward multiple is "cheap" only if you underwrite that multi-leg compounding. Believe it and the stock is the cheapest of the three on growth-adjusted terms; doubt it and the forward multiple is illusory.
Murata's growth is the most reliable but the least exciting: a cyclical ~3% organic top line where the AI-server MLCC content story (15-25k MLCCs per GB200/GB300 baseboard, Murata's own forecast of 3.3x server-MLCC volume by FY2030) shows up as mix and margin rather than headline growth. It failed the "credible 8-12% sustained revenue growth" screen in the deep-dive for exactly this reason.
SEMCO has the most visible near-term growth driver in FC-BGA (+45% YoY, 2026 capacity sold out), and that is genuine. The problem is that growth is not yet converting to margin or cash, which is what the 77x forward multiple is paying for.
Growth ranking on forward runway: AP Memory > SEMCO > Murata. On quality of growth: Murata > AP Memory > SEMCO. AP Memory has the highest ceiling and the least proof; Murata has the most proof and the lowest ceiling.
6. Valuation
| Multiple | AP Memory | Murata | SEMCO |
|---|---|---|---|
| P/E (trailing) | 155x (distorted) | 75x (impairment-flattered) | ~195x (derived) |
| P/E (forward) | 46.1x | 40.6x | 76.9x |
| EV/EBITDA | 92.7x | 29.3x | 73.9x |
| EV/Revenue | 25.6x | 8.16x | 11.8x |
| P/S | 28.5x | 9.57x | 13.6x |
| P/B | 17.0x | 6.44x | n/a |
| Price vs mean PT | +11% (2 analysts) | +76% (17 analysts) | +55% (27 analysts) |
| Price vs prior-cycle multiple | new thesis, no 5y anchor | above every prior peak | ~2.5x historical peak P/E |
Two things to internalize. First, SEMCO's forward 76.9x is itself the optimistic read: annualizing its current Q2 consensus EPS (KRW 4,120) puts the multiple closer to ~129x, so the 77x already bakes in the H2 margin step-up the thesis requires. On half of Murata's operating margin, with negative free cash flow, that is the richest valuation in the group by a wide margin. The vault's DCF triangulation pegged SEMCO fair value at KRW 480-550K; the live price is roughly four times that. Second, Murata now trades above every cycle-peak multiple it has printed in the last decade (forward 40.6x versus a 22x five-year average and a ~32x FY22 peak). The "full but defensible" read from nine days ago does not survive at 9,625; on Murata's own DCF back-solve it is now full and hard to defend.
AP Memory's 46x forward is the lowest of the three and the only one resting on a structural rather than cyclical story, but it is hostage to the heroic earnings ramp described above, and its trailing multiples (92x EV/EBITDA, 28x P/S) are the most extreme in the group on TTM numbers.
Valuation ranking, cheapest to most expensive on a growth-adjusted basis: AP Memory > Murata > SEMCO. AP Memory only if you underwrite the ramp; SEMCO is expensive on any honest reading.
7. Quality, Capital Allocation, and Risk
On capital allocation, all three avoid the cardinal sin: none is a serial diluter. AP Memory's share count has crept under 1% in three years, it is 36.6% founder-owned, but it pays out 91% of earnings and has articulated no buyback (grade B; the knock is that it returns nearly everything rather than self-funding the AI ramp). Murata is the model operator: countercyclical buybacks executed in the FY24-25 trough, no equity issuance in a decade, cleanest governance, the one ding being ~$3.75B of under-deployed net cash (grade B+). SEMCO timed its buybacks well at trough prices but converts capex to revenue at a mediocre 0.40x, is leveraging up, and management owns under 1% personally; the principal beneficiary of this rally is the Samsung Electronics parent and Korean retail, not insiders (grade B-).
On risk, the ranking inverts the naive intuition, and this is the single most important judgment in the showdown:
- Murata is not the "safe anchor" its quality implies; at 9,625 it may be the most dangerous entry of the three. A high-quality cyclical bought at peak-of-peak multiples after doubling off its 50-day average in two weeks (RSI 87) falls just as hard as anything else when the parabola breaks. Quality protects the business, not the entry price. Its vault checklist explicitly refused to buy above 7,400 for insufficient margin of safety. It is at 9,625.
- AP Memory carries the most idiosyncratic risk (2-analyst coverage, undisclosed segment data, hyperscaler design concentration in 2027-28, founder is both Chair and CEO with no named CFO) but the least technical risk: RSI 64, lowest forward multiple, net cash cushion.
- SEMCO carries the most valuation and mean-reversion risk: beta 1.99, RSI 92.7, +585% YTD, negative FCF, a chaebol related-party discount that is compressing exactly when it should not be. It has the highest beta to the shared tail risk (a 2027 AI-capex digestion), because its valuation premium must unwind on top of any earnings cut.
8. Composite Scorecard
| Dimension | Weight | AP Memory | Murata | SEMCO |
|---|---|---|---|---|
| Business Quality | 20% | 3.5 | 5.0 | 3.0 |
| Financial Health | 15% | 4.5 | 4.5 | 2.5 |
| Growth | 20% | 4.5 | 3.0 | 4.0 |
| Valuation | 20% | 2.5 | 2.0 | 1.0 |
| Quality & Capital Allocation | 10% | 3.5 | 4.5 | 2.5 |
| Risk (inverted) | 10% | 2.5 | 3.0 | 1.5 |
| Technical Timing | 5% | 3.5 | 1.5 | 1.0 |
| Weighted score | 100% | 3.55 | 3.50 | 2.43 |
AP Memory and Murata are a statistical tie (3.55 vs 3.50); they win on opposite axes. AP Memory takes growth, financial health, valuation, and timing. Murata takes business quality and capital allocation by a wide enough margin to nearly close the gap. SEMCO trails clearly on five of seven dimensions.
9. Final Verdict
| Rank | Ticker | Score | Verdict | One-line rationale |
|---|---|---|---|---|
| 1 | 6531.TW AP Memory | 3.55 | PASS at 1,190 / re-engage first | Purest silicon-cap play, least extended, lowest forward multiple, but heroic forward earnings and thin disclosure |
| 2 | 6981.T Murata | 3.50 | PASS at 9,625 (BUY thesis intact, entry stale) | Best business in passives, worst entry timing; the quality anchor for a pullback, not a chase |
| 3 | 009150.KS SEMCO | 2.43 | HARD PASS / pair-trade short | 77x forward on half Murata's margin, negative FCF, most parabolic; long-Murata/short-SEMCO remains the cleanest expression |
If you can only buy one
None today. Every one of the three has blown through its own entry discipline in a single-session melt-up, and buying any of them at Friday's close is buying the most extreme up-day in their history. But if you are pre-committing capital to deploy on the next meaningful pullback, the answer splits by what you actually want. For thesis-pure silicon-capacitor exposure, AP Memory is the only real answer and the one to re-engage first, on the WATCH zone of NT$650-800. For a quality anchor you intend to hold through the cycle, Murata is the better business and worth owning below roughly JPY 7,000-7,400 where the margin of safety returns. You would not own SEMCO over either unless you are running it short against Murata.
If you want diversification across the theme
The cleanest two-name combination is AP Memory (thesis-pure, small, asymmetric) plus Murata (quality, scale, pricing leader). They are the same secular tailwind expressed at opposite ends of the risk spectrum, and they do not overlap the way Murata and SEMCO do (both large-cap MLCC, one clearly better). Adding SEMCO to that pair buys you a worse version of the Murata exposure plus a chaebol discount; the only role SEMCO plays in a book here is the short leg.
Cross-references and caveats
- Stale-anchor risk is the dominant caveat. Every "above mean PT" figure compares a live blow-off tick to consensus that has not refreshed since the May 29 broker upgrades. PTs are moving up underneath the price. Re-check before acting.
- Prior vault verdicts (this MLCC project): AP Memory PASS at NT$1,080 (May 28); Murata BUY scale-in at JPY 6,733 (May 20, now stale, ladder unreachable); SEMCO Hard Pass at KRW 1.34-1.49M (May 23), trim trigger KRW 1.7M already breached. All three verdicts are reinforced, not weakened, by the live data.
- SemiAnalysis cross-check: no recent dedicated SA coverage of any of the three on the silicon-cap or MLCC-pricing angle. SA's 2022 work flagged SAW-filter erosion at Murata (BAW substitution), which has since materialized as the FY26 goodwill impairment. SA is one input, not authoritative; no contradiction to flag.
- Data note: the AP Memory share count was corrected to ~162.85M (an extracted draft carried a 5x-wrong figure); all per-share math here uses the verified base. FX used: TWD 0.03189, JPY 0.0062809, KRW 0.00066535. Murata cap shown on yfinance share basis (~JPY 17.5T); Google Finance shows ~JPY 18.9T (~$118.6B) on a higher share count.
Sources: yfinance live pull 2026-05-29; vault profiles for all three tickers; Google Finance, Investing.com, TradingView, Yahoo for price verification; Bloomberg/Digitimes/Investing.com for the MLCC price-hike cascade and May 29 sector catalyst.
Topics
- [[ai-infrastructure
- optical-components
- supply-chain-security
- green-finance