Back to Writing

Report Board

MLCC, EMIB and silicon capacitors: core AI upstream tickers

A layered AI infrastructure watchlist separating high-end MLCC pricing, EMIB and advanced packaging, silicon capacitors and on-package decoupling, with pure-play versus U.S.-tradable ticker mapping.

MLCCEMIBAI infrastructureadvanced packagingsilicon capacitorVSHINTCTSM
Layer Map

Do not merge the three stories: they solve different physical problems

MLCCs, silicon capacitors and EMIB/CoWoS all benefit from AI accelerator complexity. MLCCs sit around boards and power modules, silicon capacitors move into packages, and EMIB/CoWoS are high-density interconnect platforms.

01
Demand trigger

AI server / accelerator

NVDAAMDAVGO ASICMRVL customCSP AI capex

Rising power, HBM bandwidth, rack-scale interconnect and custom ASIC complexity push bottlenecks into power integrity, package integration and passive components.

02
MLCC line

Board-level decoupling

MurataSamsung Electro-MechanicsTaiyo YudenTDKYageoVSH

MLCCs are core decoupling parts around CPUs, GPUs, memory, PMICs and VRMs. AI servers raise content and specification mix, but U.S. pure-play access is weak.

03
Silicon capacitor

Package-level decap

Samsung Electro-MechanicsMurataTSMC ecosystemIntel packaging

Silicon capacitors move power stabilization closer to GPUs and HBM with low ESL and ultra-thin form factors. They complement, not replace, EMIB.

04
EMIB / CoWoS

Die-to-die interconnect

INTC EMIBTSM CoWoSASE FOCoSAMKR advanced packaging

EMIB and CoWoS solve how chiplets, HBM and I/O dies communicate at high bandwidth and lower power. This is a different layer from MLCCs.

05
Packaging capacity

OSAT / tools

AMKRASXAMATKLACONTOCAMTBESI

If advanced-packaging capex keeps rising, OSATs and packaging/inspection tools can benefit from the broader capacity cycle rather than a single EMIB customer.

06
Execution layer

U.S.-tradable basket

INTCTSMAMKRASXVSH

For U.S.-listed execution, the practical answer is a layered basket. Do not force VSH into being a pure MLCC leader.

Logic Chain

A Reviewable Logic Chain

Each card stays open and maps one transmission node without collapsible controls or pseudo-precise scores.

01
FIRST PRINCIPLE

AI hardware complexity rises

Higher GPU/HBM/ASIC power and package density create three bottlenecks: power noise, package-level decoupling and die-to-die interconnect.

02
MLCC

board-level decap

MLCCs stabilize power around CPUs, GPUs, DRAM, PMICs and VRMs. The pricing story depends on high-end specification tightness and mix shift.

03
SILICON CAP

closer to the package

Silicon capacitors are useful for low-ESL, high-frequency, ultra-thin package-level decoupling. They can displace some high-frequency decap locations, but mostly reinforce the power system.

04
EMIB / COWOS

multi-die / HBM interconnect

EMIB and CoWoS compete more with advanced packaging routes such as interposers and bridges than with MLCCs.

05
STOCK FILTER

purity conflicts with tradability

The pure MLCC names are mostly in Japan, Korea and Taiwan. The U.S.-tradable names cover only parts of the chain.

06
FAILURE TEST

orders must become revenue

Without evidence of high-end MLCC ASP, AI-server customer demand, silicon capacitor mass supply and advanced-packaging orders, this remains a narrative spillover.

Research note

Bottom line

EMIB is not an MLCC substitute. MLCCs stabilize power at the board and module level; EMIB is an advanced packaging interconnect; silicon capacitors push decoupling closer to the package. They benefit from the same AI hardware complexity, but they solve different physical problems.

The pure MLCC names are mostly outside U.S. common stocks: Murata and Samsung Electro-Mechanics first, then Taiyo Yuden, TDK, and Yageo/Walsin. In U.S. equities, VSH is only a passive-components proxy, not a pure high-end AI MLCC story.

MLCC: structural high-end pricing, not a broad commodity upcycle

The MLCC case comes from both unit-count expansion and specification mix. Murata has said AI servers use roughly five to ten times more MLCCs than conventional servers. More important, high-capacitance, low-ESL, high-reliability parts consume more difficult capacity.

TrendForce points to a polarized market: high-end MLCCs tied to AI servers and ASIC packaging orders are tight, while consumer-spec demand remains weaker. This should be treated as a high-end mix and ASP story, not a blanket MLCC price cycle.

EMIB: INTC is pure, but TSM has the stronger packaging certainty

EMIB is Intel's embedded multi-die interconnect bridge. It connects multiple dies inside a package, so its real comparables are silicon interposers, CoWoS, fan-out bridges and other advanced packaging routes, not MLCCs.

If the trade is specifically EMIB, INTC is the purest ticker. If the trade is the broader AI advanced-packaging bottleneck, TSM has higher ecosystem certainty through 3DFabric and CoWoS, while AMKR and ASX provide OSAT leverage and AMAT, KLAC, ONTO, CAMT and BESI provide equipment and inspection exposure.

Silicon capacitors: the package-level decap extension

Silicon capacitors move decoupling closer to GPUs and HBM. Their low ESL and ultra-thin form factor make them useful for high-frequency power integrity inside advanced packages. Samsung Electro-Mechanics' KRW 1.5T two-year silicon capacitor contract is the most important recent proof point.

Samsung Electro-Mechanics is high-purity because it spans MLCCs, silicon capacitors and package substrates. Murata also has both MLCC and silicon-capacitor capability. TSM and Intel matter more as advanced-packaging platforms than as direct capacitor suppliers.

U.S. execution: do not force a pure-play that does not exist

For U.S.-listed execution, the practical basket is INTC, TSM, AMKR, ASX and VSH. INTC is EMIB beta; TSM is advanced-packaging certainty; AMKR and ASX are OSAT capacity leverage; VSH is a passive-components proxy, not a high-end AI MLCC leader.

The next verification test is not price action alone. Look for high-end MLCC ASP evidence, AI-server customer long-term agreements, silicon capacitor mass-supply contracts, advanced-packaging order visibility and capex increases in the next earnings cycle.

Ticker Map

Core tickers by purity, tradability and failure condition

This is not a target-price table. It separates industry purity from U.S.-listed execution. The cleanest MLCC exposure is not in U.S. common stocks.

Theme Purity / role Ticker Access Priority Read-through
MLCC high-end highest purity Murata 6981.T / MRAAY Japan / OTC ADR Core Global MLCC leader with the cleanest high-end AI-server mix. OTC liquidity needs separate execution checks.
MLCC + silicon cap highest purity Samsung Electro-Mechanics 009150.KS Korea Core Spans MLCCs, silicon capacitors and package substrates. The KRW 1.5T silicon capacitor contract reinforces the AI package decap thesis.
MLCC high-end medium-high purity Taiyo Yuden 6976.T / TYOYY Japan / OTC Watch High-end MLCC beneficiary, but with weaker status than Murata/SEMCO. Track pricing and utilization.
Passive components medium purity TDK 6762.T / TTDKY Japan / OTC ADR Watch MLCC, magnetic components and batteries create diversified exposure that dilutes the AI-server MLCC signal.
MLCC catch-up medium-low purity Yageo 2327.TW / Walsin 2492.TW Taiwan Track Potential beneficiaries of price sentiment and channel restocking, but consumer-spec versus AI high-end mix must be separated.
US passive proxy low-medium purity VSH NYSE Proxy Closest U.S. passive-components proxy, but not a pure MLCC leader and not equivalent to Murata/SEMCO.
EMIB highest purity INTC Nasdaq Beta Purest EMIB ticker if the market trades external advanced-packaging orders, HBM integration or CSP ASIC packaging wins.
CoWoS / 3DFabric high certainty TSM NYSE ADR Core Not EMIB, but the clearest AI advanced-packaging ecosystem core. Higher certainty, usually lower distress-beta.
Advanced OSAT medium-high beta AMKR Nasdaq Leverage Advanced packaging outsourcing and U.S. packaging supply-chain expansion. Verify customer mix, ramp timing and margins.
Advanced OSAT medium-high beta ASX NYSE ADR Leverage ASE is a major OSAT with advanced packaging platforms such as FOCoS that can capture AI/HPC spillover.
Packaging tools equipment spillover AMAT / KLAC / ONTO / CAMT / BESI US / Europe Basket Not the content itself, but advanced-packaging capex lifts deposition, inspection, metrology and bonding demand.
Source Trail

AI infrastructure component watchlist · Static research snapshot · 2026-05-29

Earnings releases, announcements, filings, estimate tables, and reviewable sources.

Core signal
TrendForce points to tight high-end MLCC supply driven by AI servers and ASIC packaging orders; Murata has said AI servers use roughly 5-10x more MLCCs than conventional servers; Samsung Electro-Mechanics disclosed a KRW 1.5T silicon capacitor contract; Intel EMIB and TSMC CoWoS/3DFabric map to advanced-package interconnect.
Current read
The investable answer is layered: Murata and Samsung Electro-Mechanics are the pure MLCC/silicon-capacitor cores, INTC is the pure EMIB beta, TSM is the higher-certainty AI advanced-packaging core, and the practical U.S. basket is INTC + TSM + AMKR + ASX + VSH.
Next question
If AI upstream attention moves to the next bottleneck layer, which tickers map to MLCCs, EMIB/advanced packaging and silicon capacitors?
Core conclusions
  • EMIB does not replace MLCCs; it competes with advanced packaging interconnect routes, while MLCCs belong to the power-integrity stack.

  • The MLCC price thesis is structural high-end tightness tied to AI servers, not a broad consumer-electronics MLCC bull market.

  • For U.S. execution, VSH is only a proxy. The clearer U.S.-listed basket is INTC, TSM, AMKR, ASX and VSH.

  • Silicon capacitors are the highest-signal adjacent line because they move decoupling into GPU/HBM packages.

Next review
01

Whether Murata, Samsung Electro-Mechanics and Taiyo Yuden keep disclosing AI-server MLCC orders, high-end utilization and ASP support in the next earnings cycle.

02

Whether Samsung Electro-Mechanics expands silicon capacitor contracts beyond the first large-scale customer and clarifies 2027-2028 ramp timing.

03

Whether Intel lands quantified external advanced-packaging orders tied to HBM, ASICs or CSP accelerators.

04

Whether TSM, AMKR and ASX keep raising advanced-packaging capex, capacity and customer visibility.

05

Whether VSH reports revenue specifically attributable to AI-server or high-reliability capacitors, not only a generic passive-components recovery.