Two adjacent passive bottlenecks. Four silicon cap names ranked, two listed LIC plays profiled head-to-head. The thesis is real; entry prices are not.
There are three distinct capacitor classes in play around the AI infrastructure buildout. MLCC is the one everyone is already talking about, with the four-supplier oligopoly thesis and the 30,000-units-per-AI-server demand step-change. This primer covers the other two: silicon capacitors, which sit inside the IC package substrate and decouple at frequencies MLCC cannot reach, and supercapacitors / Lithium-Ion Capacitors (LIC), which sit at the rack or facility level and handle bulk energy buffering for power-quality and backup applications.
The two have nothing in common physically. They show up here together because the same investor question keeps recurring: “What is the non-MLCC way to play the same AI capex cycle?” The honest answer is that each class has a different industry structure, a different stock list, and a different fair value gap right now. Silicon capacitor exposure is concentrated in two listed names with disclosed order books and supply tightness; supercap exposure is harder to access cleanly via listed equity because the only meaningful pure-play comp is private.
A silicon capacitor is a chip-form passive fabricated on a silicon wafer using semiconductor lithography. Trenches are etched into silicon to maximize surface area, coated with a high-permittivity dielectric (typically aluminum oxide or hafnium oxide deposited via atomic layer deposition), then metallized. The result looks and behaves more like a tiny silicon die than a traditional ceramic capacitor.
Specs against MLCC, the comparison that matters: silicon capacitors are ultra-thin (80-100 micrometers vs 200-1,000 for MLCC), have ultra-low parasitic inductance (below 0.1 nH vs 0.3-1 nH for MLCC), and stay effective at frequencies above 1 GHz where MLCC rolls off near 100 MHz. They lose on absolute cost ($0.10-1.00 each vs $0.001-0.10 for MLCC) and maximum capacitance (mostly under 100 nF vs up to 220 microfarad in a 1206-size MLCC). They do not replace MLCC in volume applications. They complement MLCC in three specific places where MLCC physics break down: ultra-high frequency rails, ultra-thin packaging where vertical stack height is constrained, and embedded-in-substrate use cases where the capacitor must integrate directly into the silicon package rather than sit on the PCB.
AI accelerator packaging is moving to 2.5D advanced packaging technologies that integrate multiple silicon dies (compute, HBM, sometimes specialty IP blocks) onto a common substrate or interposer. Intel calls it EMIB (Embedded Multi-die Interconnect Bridge), TSMC calls it CoWoS-S, Samsung calls it I-Cube. Each die in the package has its own power delivery network, and the high-frequency decoupling must happen physically as close to the silicon as the package geometry allows. MLCC on the PCB is too far for high-frequency rails. Even MLCC on the package top is too far for the fastest transients on modern AI chips that swing voltage at picosecond rates.
Silicon capacitors solve this by sitting embedded directly in the EMIB bridge or CoWoS interposer, or as ultra-thin chips bonded onto the back of the silicon die. The ABF (Ajinomoto Build-up Film) substrate that wraps the entire AI processor package can include embedded silicon capacitor layers as part of the substrate manufacturing process. Once the design choice is made at the architectural stage, the silicon capacitor is locked in for the life of the platform — there is no straightforward substitution path back to MLCC without redesigning the substrate.
Three specific design wins drive the current demand step-change. Per Jukan’s reporting on May 28, Google plans to launch its next-generation v8e TPU in the second half of 2027, and v8e adopts an EMIB substrate with embedded silicon capacitors. Per the same source, Intel is planning large-scale adoption of silicon capacitors in EMIB packaging starting in 2027 — this covers Intel Foundry external customers (Microsoft Maia, Amazon Trainium) plus Intel’s own Granite Rapids and Clearwater Forest products. Per Jukan via industry sources, Amazon is currently applying EMIB to its accelerator stack, which pulls in additional silicon capacitor content as Intel ramps. These three programs together explain why Zephyr called silicon capacitors a “PSA for Bottleneck bois” on May 24, naming Murata, SEMCO, and AP Memory as the highest-exposure names and predicting the segment “will likely pop off like MLCC players.”
Silicon capacitor production requires semiconductor fab capacity, not ceramic kilns. The qualified supplier set is short and barriers to entry are higher than for MLCC because the process is more aligned with logic foundry manufacturing than with traditional passive component fabrication.
Murata Manufacturing (6981.T) acquired IPDiA in 2016 specifically for the silicon capacitor IP. Production is concentrated at IPDiA’s Caen, France facility. The silicon capacitor business is reportedly small relative to Murata’s ¥1.83T consolidated revenue (likely under 5%, though Murata does not disclose this line separately), but high-margin and growing on AI server traction. Murata management acknowledged the silicon capacitor program on the May 27 sell-side meeting: TSMC, SEMCO, and Murata “all have silicon capacitor technology” and Murata is “developing MLCCs with high connection reliability and will be able to handle either approach.” This is Murata hedging against silicon capacitor substitution while participating in it directly. The positioning is consistent: Murata plays the same oligopoly game across multiple capacitor classes, and the same qualification regime, multi-year design cycles, and customer relationships translate across.
Samsung Electro-Mechanics (009150.KS) is the merchant supplier with the largest disclosed order book. Per Citi, SEMCO silicon capacitor revenue is projected at W609bn in FY27E and W905bn in FY28E, anchored by a disclosed W1.6 trillion 2-year supply order covering January 2027 through December 2028 from a “major global company” — almost certainly Google for the v8e TPU based on Jukan’s EMIB reporting. SEMCO already supplies silicon capacitors to Marvell per The Elec (May 20). The capacity expansion is real and constraint-driven: per Jukan via Zephyr’s quote, SEMCO decided to demolish the parking lot around its existing plant and build a new fab there because ABF substrate demand from customers has been extremely strong and there was no greenfield space available.
The bull-case framing that the initial sum-of-segments analysis missed: per Jukan’s May 28 X post (46.5K views), SEMCO’s competitive moat in the Intel EMIB silicon cap supply chain is not just being a silicon cap supplier — it is being the only company globally that operates both a silicon capacitor business and a substrate business. Within Intel’s EMIB bill of materials, the substrate is the highest value-added component, and Intel cannot make its own silicon capacitors and must source them externally. Delivering both the substrate and the silicon capacitor to Intel EMIB on a turnkey basis maximizes SEMCO’s bargaining power in a way that Murata (via Ibiden substrate partnership) and AP Memory (via TSMC substrate partnership) cannot match. Jukan: “the emergence of a new player like SEMCO should by no means be underestimated.” This turnkey-bundle premium modestly strengthens the bull case — fair value range bumps from KRW 1.25M (aggressive) to roughly KRW 1.4-1.5M, still implying 20-25% downside from current KRW 1.85M.
The triple stacking of MLCC + FC-BGA substrate + silicon capacitor exposure makes SEMCO the cleanest single-stock expression of the AI passive bottleneck thesis.
AP Memory Technology (6531.TW) is actually the highest-concentration silicon cap pure-play of the four. The initial framing as “wildcard via TSMC packaging adjacency” was wrong. AP Memory sells two explicitly disclosed silicon capacitor product lines: S-SiCap IPD (discrete integrated passive devices, Gen3 in production, Gen4 ramping 2026) and silicon capacitor interposer IPC (four-reticle mass production from end-Q3 2025). Both are described as distinct lines on the company’s product page and in PR Newswire releases.
The commercial signal is hard: per Digitimes May 21, AP Memory IPD entered Intel’s EMIB silicon capacitor supply chain as the first qualified supplier, with small-volume shipments commencing Q2 2026. This is the direct hook into the Google v8e TPU / Microsoft Maia / Amazon Trainium programs through Intel Foundry. The TSMC adjacency is also real (IPC into CoWoS-S, IPD into CoWoS-L) but independent of and additive to the Intel win.
Silicon capacitor is estimated at 15-25% of FY2025 revenue of NT$5.67B (unofficial, reconstructed because segment is not separately disclosed). That’s the highest revenue concentration of any merchant silicon cap supplier — vs Murata under 5%, SEMCO low single digits scaling to ~6% by FY28E on the W1.6T order, and TSMC opaque inside the foundry package. The Collyer Bridge subscriber chat (May 28) ranks: AP Memory > SEMCO > Murata for pure-play silicon cap concentration, validating both the Zephyr PSA and the profile finding.
But the stock has run. Spot NT$1,080 is +365% in 13 months and +160% in just 8 weeks. 142x trailing P/E. 42x forward P/E. Only two sell-side analysts covering, mean PT NT$1,075 = exactly at spot (consensus has caught up but not gone further). Morgan Stanley raised PT to NT$777 on April 19; the stock now trades 39% above MS’s price target. Q1 2026 results delivered revenue NT$2.10B (+115% YoY) and EPS NT$4.00 (+27% beat vs consensus NT$3.14), with 49% FCF margin and NT$7B net cash. Quality of the underlying business is real. Quality of the entry price is not. Verdict: PASS at NT$1,080 / WATCH NT$650-800 / BUY NT$400-550. The next gate is the Q2 2026 print around July 30 — consensus NT$3.14 EPS looks stale vs Q1 NT$4.00 actual, but a Q2 miss would catalyze 30-40% downside on positioning alone.
TSMC (2330.TW) itself offers Integrated Passive Devices (IPD) on advanced nodes as part of its foundry package. TSMC does not break out silicon capacitor revenue separately and likely never will, but the existence of an internal TSMC silicon capacitor business is structurally important because it caps the merchant market upside: hyperscalers that go TSMC for the compute die and packaging may get silicon capacitors bundled rather than sourcing externally from SEMCO or Murata.
The earlier MLCC peer swarm framing put SEMCO fair value at KRW 400-600K. That range did not credit silicon capacitor as a distinct business line. Reworking with the Citi-disclosed silicon cap revenue path:
A moderate base case — silicon cap revenue W905bn FY28E at 30% NI margin and 35x multiple (bottleneck premium) — adds W9.5T to gross SOTP. Combined with MLCC (W27T at 25x), FC-BGA (W19.8T at 30x with a TSMC discount), camera modules (W1.4T at 12x), and net cash plus other holdings (W3-5T), gross SOTP lands at W60-63T. Implied price: KRW 825-870K per share, against current KRW 1,849K. Still 53-55% downside.
An aggressive bull case — silicon cap scales to W2T by FY29-30 at 35% NI margin and 40x multiple, FC-BGA gets outright TSMC multiples (not a discount) on AI substrate scarcity, MLCC commands a 30x premium for Tier-1 quality at cycle peak — lands gross SOTP at roughly W91T. Implied price: KRW 1,250K, still 32% downside.
For current price to be defensible, you need to believe at least one of: silicon capacitor revenue path materially exceeds Citi’s projection (e.g., Intel EMIB program adds another W1T+ in FY28-29); FC-BGA deserves outright TSMC multiples, not a discount; or MLCC commands a 40x+ premium to peers. Each leg is defensible individually. Stacking all three simultaneously is heroic.
The honest read: silicon capacitor is the most defensible new leg of the SEMCO thesis. It probably justifies bumping the fair value range from KRW 400-600K up to KRW 800K-1.25M. Still implies 32-57% downside from current. The gap has narrowed since the May 20 swarm, but the bull case is still asking you to pay full price for three businesses simultaneously executing at peak.
Briefly: it does not work. Samsung Electronics (005930.KS) holds 23.7% of SEMCO directly. At current SEMCO market cap of W140T, that stake is worth W33.2T — roughly 1.7% of Samsung Electronics’ market cap. Buying $1 of Samsung Electronics gives you about 1.7 cents of SEMCO mark-to-market, with the rest being memory, foundry, display, mobile, consumer electronics, and other cross-holdings. The chaebol holding-company discount of 25-30% applies to the entire cross-holding basket, not specifically to SEMCO. There is no clean way to isolate SEMCO exposure through the parent. If you want SEMCO, own SEMCO directly.
Supercapacitors are a completely separate device class from MLCC and silicon capacitors. Where MLCC stores charge in microfarads (millionths of a farad), supercapacitors store charge in farads (whole units, a million times larger). The physics is electrostatic charge storage at an electrode-electrolyte interface, similar to an electrolytic capacitor but with much higher surface area achieved through activated carbon, graphene, or similar nanostructured electrode materials. Lithium-Ion Capacitors (LIC) are a hybrid class that uses a Li-doped negative electrode to push energy density well beyond conventional EDLC supercaps, sitting in the middle of the energy/power Ragone plot between traditional supercaps and Li-ion batteries.
Supercaps are not for decoupling, filtering, or high-frequency power delivery. They are bulk energy storage devices for rapid-response applications where batteries are too slow and conventional capacitors are too small. Use cases: backup ride-through during grid sag or transient power loss, energy buffering for regenerative braking in transit and heavy vehicles, 48V mild-hybrid powertrain energy recovery, grid stabilization for renewable energy ramp-rate management, and increasingly, AI data center backup power and load smoothing as hyperscaler facilities deal with multi-megawatt transient loads from GPU training runs.
The relevant industry trend is power-quality and backup at the rack and facility level. A modern AI training rack draws 40-80 kilowatts. When all GPUs in a rack synchronize on a large all-reduce operation, the instantaneous load swings can exceed 30% of nominal in milliseconds. This causes voltage sag at the rack input, and if the facility-level UPS is too slow to respond, downstream blade-level power conditioning hardware sees brownouts that corrupt training jobs. The fix is to put rapid-response energy buffers at the rack shelf level — exactly the supercap use case.
SeekingAlpha covered this in May 2025: Delta demonstrated at OCP Summit a rack-level Li-ion capacitor bank that reduced AC fluctuation from 73% to 6% during simulated AI training workload swings. The article did not name suppliers but the demonstration was a proof point that LIC has a real role in next-generation AI data center power architecture.
The 800V EV story we covered in the MLCC primer is the other half. SiC and GaN inverters in 800V EVs benefit from supercap buffering at the inverter DC link because the switching dynamics are too fast for chemical batteries to follow without parasitic heat losses. Mild-hybrid 48V systems are already the production-scale feedback loop driving supercap manufacturing cost down — and that scale is what gives AI data center applications a viable supply base.
This is where supercap diverges from silicon cap. There is no listed pure-play with the disclosure quality of SEMCO on silicon caps. The relevant peers split into three buckets.
Listed pure-play and meaningful exposure: - Vinatech (126340.KQ — KOSDAQ, not 006440.KS as I’d originally noted; 006440 is Hanil Engineering & Construction) — the cleaner LIC pure-play. Supercaps and LIC are >90% of revenue. Profiled in detail below. - Musashi Seimitsu (7220.T) — Honda-group auto parts company that acquired Musashi Energy Solutions (formerly JM Energy from JSR Corporation) around 2020-2021. LIC is buried under the auto-parts business at under 2% of consolidated revenue. Detailed analysis below.
Listed with thin/diluted exposure: - Nichicon (6996.T) — Japanese capacitor manufacturer with EDLC supercap product lines. Diversified business; supercap is a small slice. - CAP-XX (CPX.AX) — Australian small-cap with thin-form prismatic supercap technology for portable electronics. Different end market.
Private: - Skeleton Technologies (Estonia) — the closest comp to Musashi on volumetric metrics. SCX5000 product line published at 16 Wh/L (Skeleton’s website). Active in industrial and transit applications. Took meaningful venture funding through 2024. - AOWEI (China) — Chinese LIC manufacturer, significant production capacity for the domestic transit and renewable energy storage markets. Not listed. - Maxwell Technologies — historically the dominant supercap brand, acquired by Tesla in 2019 and now an in-house Tesla component, not a merchant supplier.
Tier-1 majors that produce supercaps as part of a broader passive components catalog: - Eaton — XL60 series hybrid Al polymer capacitors and conventional supercaps. Small relative to Eaton’s $25B+ revenue base. - Panasonic — long-running supercap product line within the broader passive components business. - JSR Corporation — the original parent of JM Energy (which became Musashi Energy Solutions). JSR exited the LIC business via the Musashi sale.
Irrational Analysis published a “1-hour” supercap note on May 28 framing Musashi as “uber monopoly of a rapidly growing, mission-critical niche” with industry-leading volumetric energy density and 3.8V rated voltage per cell. The note argued Musashi “vaporizes everyone else” on density and that ESR (high) does not matter for the application.
Profiling the consolidated stock against this thesis revealed three problems with using 7220.T as the LIC expression. First, Musashi Energy Solutions revenue is not separately disclosed, and best-estimate puts it at under 1-2% of ¥347B consolidated revenue today. Even at full ramp by FY2027 — 6.5M cells per year multiplied by generous ASP and healthy margins — the segment remains structurally too small to dominate consolidated economics in any 3-5 year window. Roughly 99% of the stock’s economic gravity is the legacy Honda-driveline business, which has 50.1% revenue concentration to Honda and faces structural BEV-transition risk because BEVs use fewer transmission and differential components than ICE vehicles.
Second, the stock is already priced for the LIC story to be substantial. ¥7,960 at last close is 260% above the 52-week low of ¥2,210 and 67% above the 5-analyst mean price target of ¥4,760. Forward P/E is 44.8x on a flat-revenue, 5%-EBIT-margin auto parts business. The Irrational Analysis note appears to be the proximate retail catalyst for the recent rally. Sell-side does not model the LIC subsidiary separately, which is both the analytical edge (a careful Musashi Energy Solutions DCF built from the bottom up would yield insights the consensus does not have) and the reason the stock has detached from consensus PTs.
Third, the Irrational Analysis density claim has a verification gap. The Musashi public technology page does not publish exact volumetric energy density numbers. Skeleton Technologies publishes its SCX5000 at 16 Wh/L. The 3.8V per cell that Irrational Analysis highlighted as a Musashi advantage is also present in Vinatech LIC products — it is a consequence of the Li-doped electrode chemistry, not a Musashi-specific moat. To validate the “uber monopoly on density” framing, you would need to request the Musashi cell catalog PDF directly and compare specifications side-by-side against Skeleton and AOWEI. Irrational Analysis did not publish the comparison table.
Verdict: PASS at ¥7,960. Watch zone is ¥4,500-5,500 (the mean PT region, where the legacy auto-parts business is fairly valued and LIC optionality is closer to free). Buy only if Musashi separately discloses Energy Solutions revenue and shows greater than 5% revenue contribution from the segment.
Vinatech is a Korean KOSDAQ-listed supercap and hydrogen fuel cell MEA manufacturer with HQ in Jeonju, Korea and production concentrated in Vietnam (three factories near Hanoi running approximately 30M units per month combined). Supercaps and LIC are more than 90% of revenue. The Hung Yen 172,000 square meter consolidated campus targeted for 2028 represents a roughly 5x capacity uplift; an India plant is planned within three years.
The hardest commercial datapoint in the entire supercap space is Vinatech’s Bloom Energy exclusive supply agreement signed May 22, 2025, running through 2027 for SOFC-based AI data center systems. Bloom Energy serves Oracle, Equinix (19 sites, 100MW+ deployment), and AEP (up to 1 GW capacity). This is the only named exclusive AI data center customer relationship among listed LIC vendors. Musashi’s Flex CESS partnership is the structural equivalent but Flex has not disclosed downstream hyperscaler customers; Bloom’s hyperscaler customer base is public.
Specifications correct the Irrational Analysis density framing. Vinatech publishes datasheet specs on its VPC product line: 50F variant at approximately 45-80 Wh/L (usable/full range), 250F variant at 61-108 Wh/L. Musashi does not publish exact Wh/L numbers on its public technology page; the “10x EDLC density” claim implies approximately 50-80 Wh/L. The two are in the same density band. Skeleton Technologies’ SCX5000 at 16 Wh/L is an apples-to-oranges comparison because Skeleton SCX is EDLC chemistry, not LIC. Both Vinatech and Musashi run 3.8V per cell, which is the standard chemistry voltage from the Li-doped negative electrode — not a Musashi moat as the original Irrational Analysis note implied.
Valuation reflects the thematic re-rating. Market cap KRW 1.057T (about $770M), enterprise value approximately $840M, EV/Revenue 14.0x (vs Musashi consolidated 1.6x), forward P/E approximately 122x on a single analyst, FCF margin -31% in FY25, debt/equity 131%. A KRW 41B zero-coupon convertible bond floated in March 2026 maturing 2031 creates near-term dilution overhang. Per dollar of LIC revenue per market cap, Vinatech is roughly 4-5x cheaper LIC exposure than Musashi — but you give up Musashi’s Honda cash-cow underpinning and the $110M FCF base that protects the auto-parts equity from negative LIC operating leverage. CEO Do-Kyung Sung owns roughly 30% (founder since 1999), top-5 holders aggregate roughly 50%. Single analyst coverage (PT KRW 160K, +5% upside). KOSDAQ small-cap liquidity and English-language disclosure gap apply.
Verdict: WATCH at KRW 152,100. Watch entry zone KRW 90-110K (down 30-40% from current). BUY zone below KRW 60K (mid-2025 levels, genuine value). Stock has run +491% off the 52-week low and -26% off the April KRW 204,500 high; thematic re-rating risk is acute and the March 2026 convertible is the near-term clearing event.
If you want listed LIC exposure to AI data center backup and 800V EV inverter buffering, the choice is between two compromises. Musashi (7220.T) buries a small, real LIC business inside a 50%-Honda-concentrated auto parts company; at ¥7,960 the consolidated equity prices in LIC optimism that the segment cannot deliver economically because it is too small. Vinatech (126340.KQ) is the cleaner pure-play with the only named exclusive hyperscaler-adjacent supply contract — but trades at 14x EV/Revenue on a loss-making, dilution-prone, single-analyst KOSDAQ small-cap that has tripled in a year. Neither is buyable today. Both are worth watching for pullbacks: Musashi at ¥4,500-5,500 (where the auto-parts business is fairly valued and LIC is closer to free) and Vinatech at KRW 90-110K (where the thematic premium compresses and the convertible overhang is partly cleared).
The cleaner single-stock supercap exposure for the AI data center thesis would be Skeleton Technologies, which remains private. For listed exposure with embedded supercap optionality and no thematic premium, Nichicon and Panasonic both have supercap product lines but the optionality is buried inside larger businesses.
If the AI data center backup-power thesis matters more than the specific capacitor chemistry, the more reliable expressions are upstream: UPS specialists like Vertiv (VRT) or Eaton (ETN) capture the rack-level power architecture spend regardless of whether the storage element is supercap, LIC, or lithium-ion battery. The specific capacitor chemistry is not the bottleneck; the system integration is.
The MLCC peer swarm established that Murata, SEMCO, Yageo, Taiyo Yuden, TDK, and the upstream BaTiO3 names (Sakai, Nippon Chemical) are the structural beneficiaries of the AI MLCC content step-change from 4,000 capacitors per GPU today to 12,000 by 2030, plus the 800V EV automotive ramp. Murata is the only name in that universe worth scaling into at current price, and even Murata is at the top of its historical valuation range.
Silicon capacitor extends that thesis to a more specialized bottleneck. The same Tier-1 names (Murata, SEMCO) plus TSMC and AP Memory cover the four-supplier silicon capacitor oligopoly. The structural setup is more concentrated than MLCC (semiconductor-grade fab capacity is harder to add than ceramic kilns), the qualification cycles are similarly long, and the AI infrastructure demand step-change is similarly real (Google v8e, Intel EMIB scale-up, Amazon Trainium). SEMCO is the cleanest disclosed expression with W1.6T order visibility. But the current price already embeds aggressive assumptions about both the MLCC business and the silicon capacitor ramp; the rework above puts fair value at KRW 800K-1.25M against current KRW 1.85M.
Supercap and LIC are not the same thesis. They occupy a different layer of the AI infrastructure stack (rack and facility level, not chip and board level), serve a different physical function (bulk energy buffering, not high-frequency decoupling), and have a different competitive structure (no listed Tier-1 oligopoly, the technology leadership is private at Skeleton). The Musashi-as-LIC-proxy thesis does not hold at ¥7,960 because the LIC business is buried inside a much larger and structurally challenged auto parts business. For AI data center power architecture exposure, look upstream at UPS and power conditioning specialists rather than capacitor chemistry.
AP Memory (6531.TW) silicon capacitor exposure is named by Zephyr but not sized. TSMC’s internal silicon capacitor business via Integrated Passive Devices is opaque by design. Vinatech (006440.KS) as the closest LIC listed peer to Musashi has not been profiled. Skeleton Technologies private financials and customer disclosure would materially change the LIC ranking if available. The “Z = Zephyr” identity match with Citrini Research is unconfirmed — Pink’s voice note suggested Z is affiliated with Citrini but public information does not confirm this.
Research compiled May 20-28, 2026. Sources: Citi sell-side coverage on SEMCO silicon capacitor business, Jukan (@jukan05) Twitter posts on Intel EMIB and silicon capacitor adoption (May 27-28), Zephyr (@zephyr_z9) Twitter PSA on silicon capacitor supply tightness (May 24), Collyer Bridge “Small Parts, Big Cycle” (May 28), GS sell-side meeting with Murata (May 27), The Elec on SEMCO-Marvell supply (May 20), Irrational Analysis Musashi Seimitsu supercap note (May 28), full profile of Musashi 7220.T, full profile of SEMCO 009150.KS, full profile of Kingboard 0148.HK and Kingboard Laminates 1888.HK, plus prior MLCC peer swarm research. Prices are snapshots from May 20-28; verify before acting.