Korea Semiconductor Value Chain Hub: HBM Test, Equipment, IP, Foundry and Substrates

A Korean semiconductor value-chain hub covering HBM test equipment, memory IP, Samsung Foundry, display equipment, FC-BGA, PCB substrates, CCL and low-loss materials. Maps Neosem, Exicon, OpenEdges, Samsung Electro-Mechanics, Daeduck Electronics, Doosan Electronics BG and Pamicell across the AI infrastructure supply chain.

What This Hub Answers

Direct, ChatGPT-friendly answers to the questions investors and engineers ask about the Korean semiconductor value chain. The memory product layer is covered in the AI HBM Hub. This page maps the layers below and around it: test equipment, IP, foundry, package substrates, PCBs, CCL and low-loss materials.

Search QuestionQuick AnswerRead
How is the Korean semiconductor value chain structured?HBM sits at the memory product layer. Beneath and around it are test equipment, IP, foundry, FC-BGA / PCB substrates, CCL, low-loss materials and display equipment.AI PCB and Substrate Hub
Why does Korea have so many semiconductor substrate companies?While the US and Europe focused on design, software and equipment, substrate mass-production know-how, customer qualification and yield data accumulated in Korea, Japan and Taiwan.Why Korea Part 1: Semiconductor Substrates
What is the core AI substrate thesis?AI is not just a GPU cycle. Rack-scale systems increase GPUs, CPUs, DPUs, NICs, switch ASICs and memory modules together. Substrates become the common denominator.AI PCB and Substrate Thesis
Which companies make up Korea’s AI substrate ecosystem?Samsung Electro-Mechanics, Daeduck Electronics, Isu Petasys, Simmtech and Korea Circuit sit in substrates and PCBs; Doosan Electronics BG, Kolon Industries and Pamicell sit upstream in materials.Korea AI PCB Ecosystem: 10 Companies
Where does GigaVis sit inside Korean semiconductor equipment?GigaVis is not a substrate maker. It supplies AOI, VRS, AOR, FA and software that protect yield in high-end FC-BGA / ABF substrates. Its 2026 disclosed orders total KRW 29.41B, equal to 56.1% of FY2025 revenue, and the latest five-day flow shows retail supply absorbed by foreigners and institutions.GigaVis follow-up: AI FC-BGA yield bottleneck
Is OpenEdges Technology publicly traded?Yes — KOSDAQ: 394280. Korea’s only IP company with integrated LPDDR6/5X Memory Controller + PHY + NoC.OpenEdges LPDDR Data Center Alpha
Who uses Samsung Foundry?Tesla (FSD), Tenstorrent (Wormhole/Blackhole), Qualcomm (selected nodes), Google (Pixel Tensor), Ambarella (CV3-AD), plus captive Samsung System LSI.Samsung Foundry Customer List
Who makes HBM test equipment?Korean memory test equipment names — Neosem, Exicon — sit at the front of HBM test capacity ramp.SemiScope Neosem/Exicon/OpenEdges Re-Rank
Where does TechWing sit in HBM inspection equipment?TechWing combines memory test handlers and recurring C.O.K revenue with HBM Cube Prober expansion across Samsung supply, the first SK hynix order and Micron evaluation. The key checks are repeat SK hynix orders, Micron production conversion and Q2-Q3 earnings leverage.TechWing HBM Cube Prober analysis
Which equipment and substrate names could beat Samsung and SK Hynix over two months?The Top 50 semiconductor plus 1,137 Naver ETF constituent scan compresses the first watchlist into Korea Circuit, HPSP, TES, VM and FADU. Equipment names need to be separated from substrates and AI storage.Korean semiconductor Top 50 two-month alpha candidates
Is Hanmi Semiconductor’s June 1 IR notice an order?No. It is an IR meeting notice about 2H26 outlook and overseas new-market entry. The key message is TC-bonder TAM expansion from HBM into 2.5D packaging, AI system semiconductors, OSAT and HBF, with CEO Kwak Dong Shin’s latest KRW 315,407 purchase price as a secondary checkpoint.Hanmi IR notice and TC-bonder TAM expansion
What does Jensen Huang’s HBM4 three-vendor comment mean for Hanmi Semiconductor?It increases the option that the HBM4 equipment cycle broadens from SK hynix into Samsung Electronics and Micron. Direct upside still requires actual TC-bonder order confirmation from Samsung, Micron, OSAT or foundry customers.Jensen Huang’s HBM4 three-vendor comment
What does Samsung’s HBM4E 12-high sample mean for Hanmi Semiconductor?It is not a direct order; it raises the option value of a future Samsung TC-bonder order. Higher-stack HBM increases bonding precision and yield bottlenecks, which may push Samsung to diversify equipment vendors.Samsung HBM4E 12-high samples
Why does CXL / SSD / ATE matter for Neosem?Neosem’s CXL and ATE-class capability has been the central thesis for the memory test re-rating cycle.SemiScope Neosem CXL/SSD/ATE Turnaround
What does the OpenEdges + Samsung 4/5/8nm story look like?OpenEdges silicon-proven on Samsung SF5A (LPDDR5X 8,533 Mbps); SAFE Sub-License partner; LPDDR6/5X first license disclosed April 2026.OpenEdges Samsung LPDDR6 IP Option
What’s the Korean display equipment picks-and-shovels thesis?BOE Apple-line utilization weakness + Korean display panel makers + AP System / Soulbrain / Dongjin Semichem on the OLED 8.6G capex cycle.Korea Display Equipment Picks & Shovels
What is OpenEdges Technology?A Korean fabless / IP company (KOSDAQ: 394280) integrating Memory Controller, DDR PHY, and NoC interconnect — the three blocks an AI inference SoC needs to attach LPDDR memory.OpenEdges Technology Platform Overview
What exactly does Samsung Electro-Mechanics make?Not chips. MLCC (power-stability components) and FC-BGA package substrates (the substrate connecting AI chips to motherboards), plus camera modules. The “invisible infrastructure” of AI silicon.SEMCO Three-Division Deep Dive
Is SEMCO’s rerating from “smartphone parts” to “AI infrastructure” already priced in?At KRW 1.02M (~38x 2027E PER), much of the structural thesis is reflected. Base-case fair value KRW 1.15M (+12%); bear KRW 730K (-29%). Earnings upgrades must continue.SEMCO Three-Division Deep Dive
What does SEMCO’s KRW 1.5T silicon-capacitor contract mean?Reclassification from “MLCC / substrate maker” → “AI-package power-integrity component company." KRW 1.557T over 2 years (annualized KRW 779B, ~6.9% of FY25 revenue), recognized 2027–2028. Globally the only player able to bundle FC-BGA + MLCC + Si-Cap. But May 20 +7.5% and May 21 pre-market KRW 1,109,000 already price it in meaningfully.Si-Cap KRW 1.5T — A Reclassification Event (May 21)
Why do Intel EMIB-T and SEMCO Si-Cap belong to the same AI package power-delivery debate?EMIB-T adds TSVs to reinforce power delivery and signal routing inside 2.5D packaging, while Si-Cap sits near GPU/HBM packages to dampen transient voltage noise. Both point to the same internal AI-package PDN bottleneck.SEMCO Si-Cap and Intel EMIB-T
Does SEMCO at KRW 100T mean it can overtake Hyundai Motor or Murata?A temporary overtake is possible. Hyundai common stock requires about KRW 1.80M per SEMCO share, while a KRW 150T benchmark requires about KRW 2.01M. Sustaining it needs a credible 2028 OP path around KRW 3.8-4.5T and group OPM near 20%.SEMCO at KRW 100T: Murata and Hyundai comparison
Can SEMCO command a higher premium than Murata and Ibiden after reaching Hyundai Motor’s weight class?The premium thesis is real. The Si-Cap long-term contract and MLCC/FC-BGA/Si-Cap bundle reclassify SEMCO as an internal AI-package power-integrity supplier. But at KRW 138T, it is already 94% of Murata’s size; a KRW 180–205T bull case needs repeat orders and a 2027 OP path near KRW 4T.SEMCO at KRW 138T and peer premium
What is the AI server passive-component bottleneck?It is not a GPU shortage. It is the need for higher-spec MLCCs, silicon capacitors and inductors that buffer, filter and stabilize the transient power consumed by GPU/HBM packages.AI server passive-component bottleneck: SEMCO explainer
How should investors map the full AI data-center power bottleneck?Split it into generation and backup power, fuel cells and distributed power, ESS/UPS, power conversion and transformers, cables and bus ducts, and cooling/EPC. The direct-evidence names are GNC Energy, VinaTech and Seojin System, while Sanil Electric, Iljin Electric, Gaon Cable and SK Gas need follow-up diligence.AI data-center power bottleneck map
Can SK Gas’s LNG cold energy be treated as a confirmed AI data-center cooling beneficiary?Not yet. The confirmed fact is the SK Gas-linked DevOcean technical logic: LNG produces roughly 0.2 MWh of cold energy per ton during vaporization, and that waste cold energy could support data-center cooling. But commercial contract, target PUE, tariff, capex payback and SK Gas margin attribution remain undisclosed. Treat it as a moat candidate, not a confirmed moat.SK Gas LNG cold energy
Where does VinaTech sit in the AI data-center power chain?If Bloom SOFC units provide on-site baseload power near the data center, VinaTech’s supercapacitor system sits in the power-quality layer that buffers transient AI-server load, voltage movement and peak current. The KRW 41.2B Bloom contract is 50.12% of FY2025 revenue, but the key is repeat POs and system margin.VinaTech and Bloom Energy
Where does GNC Energy sit in the AI data-center power chain?GNC Energy is not an AI power-generation stock in the broad sense. It sits in the backup-power EPC layer. LG Uplus AIDC cooperation, the KRW 29.748B Samsung SDS Gumi AI data-center generator contract, FY2025 revenue of KRW 262.6B and OP of KRW 49.4B show real business substance. But after a 45.5% five-day rally, foreign selling and 1Q26 margin weakness argue for pullback discipline.GNC Energy rally analysis
How do SK Gas and SK Oceanplant differ in the AI data-center power theme?SK Gas is the more direct power-infrastructure candidate through Ulsan GPS, KET LNG and SK Group’s Ulsan AI data-center power structure. SK Oceanplant is an offshore-wind foundation maker, so it is a higher-beta second-line candidate only if the power-shortage theme expands into renewable generation infrastructure.SK Gas vs SK Oceanplant
What does ARM’s rally mean for Korea’s semiconductor value chain?ARM’s rally is a signal that AI rack bottlenecks are moving from GPU compute into CPU orchestration, memory movement, optical interconnect, package power integrity, high-speed substrates and test sockets.ARM rally and Korea’s next AI bottlenecks
What should Korea investors watch before Marvell and Broadcom earnings?If Broadcom’s $10.7B AI semiconductor guide and Marvell’s 74% data-center mix hold, Korea’s read-through expands from HBM into AI ASICs, Ethernet, optical links, FC-BGA, silicon capacitors, high-speed PCBs and test sockets.Marvell / Broadcom earnings preview and Korea’s AI bottlenecks
What did Marvell Q1 FY2027 confirm after the preview?The post-earnings read-through is that custom XPU / ASIC, optical interconnect, scale-up networking and XPU attach are moving Korea’s AI bottleneck from HBM alone into FCBGA, MLCC, SiCap, test sockets and high-speed substrates.Marvell Q1 FY2027 and Korean semiconductors
What is the Broadcom read-through from Jensen Huang’s Marvell call-out?Directly positive for Marvell, but a 70/100 positive category signal for Broadcom: custom XPU, AI networking and optical / SerDes bottlenecks are being re-rated. In Korea, the first-order language is FC-BGA, AI network PCB, Si-Cap and MLCC, with Samsung Electro-Mechanics, Korea Circuit and Isu Petasys as the key names to check.Marvell’s trillion-dollar story and the Broadcom read-through
Why do laser light sources matter after the Marvell / Broadcom connectivity thesis?Moving one layer below AI networking brings the bottleneck into EML and CW-DFB laser sources. As AI optical transceivers and CPO scale, external laser sources, ELSFP, InP lasers and optical packaging become critical. In Korea, OE Solutions is the most direct technical proxy, but customer qualification and repeat orders still need proof.Marvell follow-up: EML and CW-DFB laser sources
Who is the Korean proxy for the AI storage/NAND bottleneck?FADU is the cleaner structural candidate in eSSD controllers/firmware, while Jeju Semiconductor is the tactical LPDDR/MCP scarcity candidate. Foreign flow is beginning to treat FADU as a Korean high-beta proxy for Sandisk.FADU and Sandisk AI storage beta
What is the latest FADU 2Q26 earnings bar after the June contract amendments?The updated base case is revenue of KRW 69-72bn and operating profit of KRW 11-12.5bn, above the KRW 60.5bn / KRW 9.3bn consensus but not yet a mega-surprise. The key is how much of the disclosed 2Q contract gross amount is actually recognized before the cutoff.FADU 2Q26 earnings preview
Can FADU become an AI infrastructure bottleneck that beats memory?It can, but only if P, Q, C and a new segment open together. The June 2 framework set the long-duration question; the July 4 earnings preview lowers the near-term proof bar to recognized Q2 revenue, controller mix, gross margin and backlog delivery visibility.FADU P, Q and new-segment check
From a chip-design perspective, where is Korea’s AI semiconductor bottleneck moving?As data-movement cost becomes more important than headline FLOPS, HBM, FC-BGA, high-speed PCB, Ethernet / optical links, power integrity and testing move as one system bottleneck.After NVIDIA: the AI semiconductor bottleneck
AI back-end — should substrates and test sockets be lumped as one “AI theme”?No. On 1Q26 OPM, Daeduck Electronics 14.8% vs ISC 35% vs LEENO Industrial 47.4% — ~3x spread. Substrates are an “AI server volume beta” (strong short-term momentum, CAPEX cycle risk); test sockets are a “chip-complexity consumables beta” (high margin, low volatility). Different answers by holding horizon.Substrate vs Test Socket: Two Betas of AI Back-End
Within the 11 AI back-end names, who has the best value-for-growth?Simmtech (2027E PER 20.0x × OP +59%), Haesung DS (15.4x × +37%), Daeduck Electronics (27.1x × +34%). Memory megas screen cheapest (SK hynix 2027E PER 5.2x), but cycle-peak risk is a separate class. New-capital priority: substrates > test sockets > memory megas.AI Back-End — 11 Stocks Side by Side
What should investors watch if CloudMatrix-style systems expand?The alpha is not direct Huawei supply. It is the inspection, metrology and high-difficulty consumables demand created when China has to scale its own fabs. TCK is the confirmed consumables recovery, Nextin is the 2H26 China PO option, and Komico is global quality with a China option.CloudMatrix Expansion Scenario

Start Here

StepQuestionRead
1Where is the physical bottleneck in Korea’s AI semiconductor value chain?AI PCB and Substrate Hub
2Why does Korea have so many substrate and PCB companies?Why Korea Part 1: Semiconductor Substrates
3What’s the SemiScope framing — Neosem vs Exicon vs OpenEdges?SemiScope Re-Rank: Neosem / Exicon / OpenEdges
4Why is OpenEdges the most direct LPDDR-to-AI-inference alpha?OpenEdges LPDDR Data Center IP Alpha Thesis
5How does OpenEdges connect to Samsung Foundry’s roadmap?OpenEdges + Samsung 4/5/8nm LPDDR6 IP Option
6Who actually uses Samsung Foundry?Samsung Foundry Customer List 2026
7What does the Korean display equipment picks-and-shovels look like?Korea Display Equipment Picks & Shovels
8What exactly does Samsung Electro-Mechanics make, and is it still investable at KRW 1.02M?SEMCO MLCC / FC-BGA / Optics Deep Dive
8+What does the KRW 1.5T silicon-capacitor contract change — and is KRW 967K already pricing it in?Si-Cap KRW 1.5T Supply Contract (May 21)
8++What does ARM’s rally say about the next AI infrastructure bottlenecks — CPU orchestration, optical links, power integrity and high-speed substrates?ARM rally and Korea’s next AI bottlenecks
8+++Will Marvell and Broadcom confirm that the next bottleneck is ASIC networking, packaging and power integrity?Marvell / Broadcom earnings preview and Korea’s AI bottlenecks
8++++What did Marvell actually confirm in Q1 FY2027 for Korean AI bottlenecks?Marvell Q1 FY2027 and Korean semiconductors
8+++++How does Jensen Huang’s Marvell call-out connect Broadcom’s Q2 call to Korean substrate and power-integrity bottlenecks?Marvell’s trillion-dollar story and the Broadcom read-through
8+++++AIf the Marvell / Broadcom bottleneck moves into laser sources, what should Korea investors watch?Marvell follow-up: EML and CW-DFB laser sources
8+++++BCan FADU open price, volume and a new AI storage segment together?FADU P, Q and new-segment check
8++++++Why do data movement, FC-BGA and power integrity become the next AI bottlenecks when viewed from the bottom of chip design?After NVIDIA: the AI semiconductor bottleneck
8++++++AWhat do Intel EMIB-T and SEMCO Si-Cap say about internal AI-package power delivery?SEMCO Si-Cap and Intel EMIB-T
8+++++++Is SEMCO’s KRW 100T market cap the start of a Murata-style AI passive-component rerating, or already priced in?SEMCO at KRW 100T
8+++++++ACan SEMCO command a premium over Murata and Ibiden after reaching Hyundai Motor’s weight class?SEMCO at KRW 138T and peer premium
8++++++++How should non-engineers understand AI server passive-component bottlenecks?AI server passive-component bottleneck
9Why do substrates and test sockets behave so differently, even when both benefit from AI?Two Betas of AI Back-End: Substrate vs Test Socket
10Of the 11 AI back-end names, who actually has the best value-for-growth? — 2027E PER × OP-growth matrixAI Back-End — 11 Stocks Side by Side
11How does CloudMatrix-style inefficiency flow into Korean inspection and consumables names?CloudMatrix and Korean Inspection / Consumables Alpha

Korean Semiconductor Value Chain Layer Map

LayerKorean NamesFunction in the AI Cycle
Memory products (covered in AI HBM hub)SK hynix, Samsung Electronics MemoryHBM3E, HBM4, server DRAM, LPDDR-based SOCAMM2
AI storage / eSSD controllerFADU, Jeju Semiconductor, adjacent Neosem SSD-test exposureEnterprise SSD controllers, firmware qualification, LPDDR/MCP scarcity and KV-cache/data-center storage bottlenecks
Memory test equipmentNeosem (KOSDAQ: 253590), Exicon (KOSDAQ: 092870)HBM and high-bandwidth memory test capacity for AI server memory
HBM die inspection / handlingTechWing (089030)Memory Test Handler, C.O.K and HBM Cube Prober exposure. This is the layer where individual-die inspection and customer qualification become more important as HBM stacks move higher
Test sockets (AI back-end consumables)LEENO Industrial (058470), ISC (095340), TSEChip-complexity beta. 1Q26 OPM 35–47%. Fresh revenue each AI chip generation
Inspection / metrology / high-difficulty consumablesTCK (064760), Nextin (348210), Komico (183300), Park SystemsThe layer where China’s brute-force AI systems and domestic fab buildout translate into yield-control and process-stability demand
Memory subsystem IPOpenEdges Technology (KOSDAQ: 394280)LPDDR6/5X Controller + PHY + NoC for AI inference SoCs
Foundry / contract manufacturingSamsung Electronics — Foundry segment (KOSPI: 005930)Production for Tesla, Tenstorrent, Qualcomm, Google, Ambarella, Samsung System LSI
Package substrates / FC-BGASamsung Electro-Mechanics, Daeduck Electronics, Korea CircuitHigh-performance package substrates for GPUs, CPUs and ASICs; the lower-stack bottleneck that physically shortens the path between memory and compute
Substrate inspection / repair equipmentGigaVis (420770)AOI, VRS, AOR, FA and software for AI GPU / ASIC FC-BGA and ABF substrates; the yield-equipment layer beneath high-end package-substrate capex
Passive components / power integritySamsung Electro-Mechanics, MLCC peers, MurataMLCCs stabilize board and package-adjacent power; silicon capacitors move closer into GPU/HBM packages as die-near PDN components
PCB / MLB / module substratesIsu Petasys, Daeduck Electronics, TLB, Simmtech, Korea CircuitServer boards, network switch boards, SoCAMM and DDR5 modules
CCL / low-loss materialsDoosan Electronics BG, Kolon Industries, PamicellCopper-clad laminates, mPPO, resins and hardeners for high-speed signal integrity
Display equipmentAP System, Soulbrain, Dongjin SemichemOLED 8.6G capex cycle — picks-and-shovels for Korean display panel rebuild
Display panel makersLG Display (KOSPI: 034220), Samsung Display (Samsung Electronics subsidiary)iPhone OLED supply, AI display content

For memory products, see the AI HBM Hub. For deeper substrate, PCB and CCL coverage, see the AI PCB and Substrate Hub. For Korean AI customers of these foundries and IP, see Korean AI Companies Hub.


SemiScope Series — Neosem, Exicon, OpenEdges

The SemiScope series tracks the three Korean memory test equipment + IP names that form the cleanest picks-and-shovels exposure to the AI memory cycle.

DateTopicRead
2026-04-25Three-way re-rank: Neosem vs Exicon vs OpenEdgesSemiScope Re-Rank
2026-04-25Neosem CXL / SSD / ATE turnaround thesisSemiScope Neosem CXL/SSD/ATE
2026-04-25OpenEdges Technology IP platform overviewOpenEdges IP Platform
2026-04-30OpenEdges + Samsung Foundry LPDDR6 optionOpenEdges Samsung LPDDR6 IP Option
2026-04-30OpenEdges as the most direct LPDDR-to-AI-inference alphaOpenEdges LPDDR Data Center Alpha

AI Back-End Comparison — Substrates vs Test Sockets

When AI silicon sells, two back-end regions benefit together — but the structures are completely different. Substrates = “AI-server volume beta” (short-term momentum, CAPEX-cycle risk). Test sockets = “chip-complexity consumables beta” (high margin, low volatility). 1Q26 operating-margin spread of ~3x quantifies that difference. The 5/16 follow-up extends this into an 11-stock data comparison on a 2027E PER × OP-growth matrix.

DateTopicRead
2026-07-04FADU 2Q26 earnings preview — uses 1Q financials, Q2 contract disclosures and June amendment cutoffs to estimate Q2 revenue of KRW 69-72bn and operating profit of KRW 11-12.5bn, implying a clear beat versus consensus but not a mega-surpriseFADU 2Q26 earnings preview
2026-07-04SK Gas LNG cold energy and AI data-center cooling option — separates the confirmed DevOcean technical logic from unproven commercial economics. The Ulsan KET / SK-AWS data-center location matters, but target PUE, tariff, capex, payback and SK Gas margin attribution still need proofSK Gas LNG cold energy
2026-07-04AI data-center power bottleneck map — screens listed Korean companies by generation and backup power, fuel cells and distributed power, ESS/UPS, power conversion and transformers, cables and bus ducts, and cooling/EPC. GNC Energy, VinaTech and Seojin System have the cleanest direct evidence, while Sanil Electric, Iljin Electric, Gaon Cable and SK Gas remain diligence candidatesAI data-center power bottleneck map
2026-07-04GNC Energy AI data-center backup power — moves the AI data-center power bottleneck into the backup-generator EPC layer. LG Uplus AIDC cooperation, the KRW 29.748B Samsung SDS generator contract and five-day institutional buying of KRW 7.66B are positive, but foreign selling and 1Q26 margin weakness make pullback discipline essentialGNC Energy rally analysis
2026-07-04SK Gas vs SK Oceanplant as AI data-center power second-line candidates — compares power directness after SK Group’s AI data-center announcement. SK Gas is tied to Ulsan GPS, KET LNG and the Ulsan AI data-center power structure, while SK Oceanplant is an indirect offshore-wind infrastructure option for a later phase of the themeSK Gas vs SK Oceanplant
2026-07-04VinaTech and Bloom Energy power-buffer chain — reframes the AI data-center power bottleneck as transient load and power quality, then maps VinaTech’s KRW 41.2B Bloom supercapacitor-system contract as a buffer layer between SOFC units and AI serversVinaTech and Bloom Energy
2026-06-21TechWing HBM Cube Prober analysis — frames TechWing as a potential HBM inspection platform candidate after Samsung supply, the first SK hynix order and Micron evaluation were added to its memory handler and C.O.K base. The practical entry checks are support at KRW 58,000-60,000 or a reclaim of KRW 65,300TechWing: can HBM Cube Prober expand to the Big 3?
2026-06-05Jensen Huang’s HBM4 three-vendor qualified comment — translates the Vera Rubin HBM4 three-vendor structure into Hanmi Semiconductor TC-bonder, inspection/metrology and packaging-equipment demandJensen Huang’s HBM4 three-vendor comment
2026-06-05GigaVis follow-up: AI FC-BGA yield bottleneck — classifies GigaVis as an AOI / AOR inspection and repair equipment name, not a substrate maker, and checks KRW 29.41B of 2026 disclosed orders plus the latest retail-to-foreign/institutional ownership handoffGigaVis follow-up: AI FC-BGA yield bottleneck
2026-06-03Marvell follow-up: EML and CW-DFB laser sources — moves the Marvell / Broadcom connectivity bottleneck down into EML, CW-DFB, external laser sources, ELSFP and CPO; in Korea, OE Solutions is the structural watchlist name, while the cleaner current flow sits in Solid, RFHIC and Samji ElectronicsMarvell follow-up: EML and CW-DFB laser sources
2026-06-02Marvell’s trillion-dollar story and the Broadcom read-through — interprets Jensen Huang’s Marvell call-out as a custom XPU, AI networking and optical-interconnect bottleneck signal, then maps Broadcom’s $10.7B AI semiconductor Q2 guide, Q3 guide, Tomahawk 6 and CPO checklist into Samsung Electro-Mechanics, Korea Circuit and Isu PetasysMarvell’s trillion-dollar story and the Broadcom read-through
2026-06-02FADU P, Q and new-segment check — sets the original AI-storage bottleneck framework around price, volume, cost and new segments. The July 4 preview updates the near-term earnings bar after June amendments to KRW 69-72bn of Q2 revenue and KRW 11-12.5bn of operating profitFADU P, Q and new-segment check
2026-06-02Hanmi Semiconductor IR notice and TC-bonder TAM expansion — the June 1 filing is not an order but an IR meeting notice. The message is a strategic expansion from HBM TC bonders into 2.5D packaging, AI system semiconductors, OSAT and HBF. CEO Kwak Dong Shin’s accumulated open-market buying is a strong secondary signal, but the stock still needs to regain the latest KRW 315,407 purchase price and confirm a flow turnHanmi IR notice and TC-bonder TAM expansion
2026-06-01Samsung HBM4E 12-high samples and Hanmi’s TC-bonder option — decomposes Samsung’s HBM4E sample into customer qualification, mass production and equipment-order paths; Hanmi’s foreign buying is the first positive piece, but institutional selling and short interest still make it a watchlist turn, not confirmed accumulationSamsung HBM4E 12-high samples
2026-05-31Is FADU Korea’s Sandisk beta? — maps the AI storage/NAND bottleneck across FADU’s eSSD controller/firmware, Jeju’s LPDDR/MCP scarcity, and global Sandisk/WDC, then tests it with foreign flowsFADU and Sandisk AI storage beta
2026-05-31AI Infrastructure Multiple Map — maps why GPU, HBM, CPU, MLCC and FC-BGA deserve different multiples inside the same AI cycle, using pricing power, LTAs, customer lock-in, capex burden and peak-earnings doubt. Samsung Electronics has the cleaner relative-value case; SEMCO’s direction is right but the chase risk is highAI Infrastructure Multiple Map
2026-05-30AI token futures and cost per token — from a performance race to a cost race — once AI usage gets a market price, the axis shifts from performance to cost per token. This maps how the packaging / substrate layer of that bottleneck (SEMCO FC-BGA / MLCC / silicon capacitor, Daeduck as a second-order FC-BGA beta) connects to the token-cost thesis — with the discipline that the direction is right but the price is richAI token futures and cost per token thesis
2026-05-29Dell Q1 FY2027 and Korea AI-server margin read-through — the print was a big beat, but AI-server gross margin compressed from 21.6% to 18.1%. The server OEM not keeping the margin signals that pricing power sits upstream with memory (Samsung / SK Hynix) and high-value FC-BGA / MLCC / silicon capacitors (SEMCO); low-margin assembly is the part to avoidDell Q1 FY2027 and Korea AI-server margin
2026-05-28SEMCO at KRW 138T and peer premium — checks whether Samsung Electro-Mechanics can command a premium over Murata and Ibiden after reaching Hyundai Motor’s common-share weight class, using Si-Cap references and a KRW 150–205T market-cap rangeSEMCO at KRW 138T and peer premium
2026-05-28SEMCO Si-Cap and Intel EMIB-T — separates the official KRW 1.5T Si-Cap contract from the still-inferred Google TPU / EMIB-T link, and reframes the story as an internal AI-package PDN bottleneckSEMCO Si-Cap and Intel EMIB-T
2026-05-28Marvell Q1 FY2027 and Korea semiconductor read-through — Q1 revenue $2.418B and Q2 guide $2.70B ±5%; the key Korean read-through is not SOCAMM alone but custom XPU / optical / scale-up networking into FCBGA, MLCC, SiCap and test socketsMarvell Q1 FY2027 and Korean semiconductors
2026-05-26AI server passive-component bottleneck explainer — a non-engineer map of GPU power spikes, MLCCs, silicon capacitors and inductors as the technical foundation behind SEMCO’s power-integrity reratingAI server passive-component bottleneck
2026-05-26SEMCO at KRW 100T — what it would take to overtake Hyundai Motor or Murata, and why a durable KRW 150T valuation requires a 2028 OP path near KRW 4T-plusSEMCO at KRW 100T: Murata and Hyundai comparison
2026-05-24After NVIDIA: the AI semiconductor bottleneck — translating Reiner Pope’s bottom-up chip-design discussion into an investment map linking data movement, HBM, FC-BGA, power integrity and testingAI chip design and Korea’s lower-stack bottlenecks
2026-05-23Marvell / Broadcom earnings preview and Korea’s AI bottlenecks — using Broadcom’s $10.7B AI semiconductor guide and Marvell’s 74% data-center mix to test whether the Korea read-through broadens from HBM into AI ASICs, Ethernet, optical links, FC-BGA, silicon capacitors and test socketsMarvell / Broadcom preview
2026-05-22ARM rally and Korea’s next AI bottlenecks — why ARM itself may be expensive, while Marvell connectivity, SEMCO silicon capacitors, memory, high-speed substrates and test sockets may carry the better risk-rewardARM rally and Korea’s next AI bottlenecks
2026-05-17CloudMatrix expansion scenario — not direct Huawei suppliers, but where China’s inefficient AI buildout creates demand for inspection, metrology and high-difficulty consumables across TCK, Nextin and KomicoCloudMatrix and Korean inspection / consumables alpha
2026-05-17US-China technology decoupling and the semiconductor value chain — what the H200 refusal and Huawei CloudMatrix imply for memory, optical interconnect, substrates and back-end supply-chain exposureUS-China summit follow-up
2026-05-17Google I/O / NVIDIA earnings preview — how NVIDIA’s Q2 guide and Data Center revenue flow through HBM, FC-BGA, MLCC and test socketsKorean semiconductor event preview
2026-05-16Data follow-up — 11 stocks in one table (2026 YTD, 2026E/2027E PER, OP YoY, OPM), PEG-style ranking, priority order for new capitalAI Back-End — 11 Stocks Side by Side
2026-05-15Head-to-head comparison — 14.8% vs 35% vs 47.4% OPM, priority names by holding horizon, portfolio constructionSubstrate vs Test Socket: Two Betas of AI Back-End
2026-05-15Substrate-side stock — Samsung Electro-Mechanics divisional dissectionSEMCO MLCC / FC-BGA / Optics Deep Dive
2026-05-07Substrate industry structure — Why Korea?Why Korea Part 1: Semiconductor Substrates
2026-05-05The Korean substrate-cluster 10 companiesKorea AI PCB Ecosystem: 10 Companies

For deeper substrate-side analysis, continue at the AI PCB and Substrate Hub.


Samsung Electro-Mechanics Deep Dives

SEMCO is the Korean name that sits at the intersection of MLCC supply tightness and FC-BGA capacity shortage — the two passive-component / substrate bottlenecks the AI server cycle has exposed.

DateTopicRead
2026-05-31AI Infrastructure Multiple Map: Samsung Electronics owns HBM catch-up plus AI memory-hierarchy optionality at a lower multiple, while SEMCO’s MLCC / FC-BGA / Si-Cap thesis is already priced aggressivelyAI Infrastructure Multiple Map
2026-05-28Market cap at KRW 138T: after reaching Hyundai Motor’s weight class, what conditions would let SEMCO command a premium over Murata and Ibiden? Reasonable ceiling KRW 150–165T, bull case KRW 180–205TSEMCO at KRW 138T and peer premium
2026-05-28Intel EMIB-T and Si-Cap: TSV-enabled power delivery, internal AI-package PDN, SEMCO’s official KRW 1.5T contract, and the boundary between confirmed facts and Google TPU / EMIB-T inferenceSEMCO Si-Cap and Intel EMIB-T
2026-06-02Marvell / Broadcom connectivity lens: why custom XPU, AI networking, Tomahawk 6 and optics strengthen the SEMCO FCBGA / MLCC / Si-Cap thesisMarvell’s trillion-dollar story and the Broadcom read-through
2026-05-28Marvell Q1 FY2027 lens: why custom XPU, optical interconnect and scale-up networking strengthen the SEMCO FCBGA / MLCC / SiCap thesisMarvell Q1 FY2027 and Korean semiconductors
2026-05-26AI server passive-component bottleneck: a rack-to-package power-path explainer for why SEMCO is being reclassified as an AI package power-integrity companyAI server passive-component bottleneck
2026-05-26Market cap at KRW 100T: the share price and earnings bridge needed to overtake Hyundai Motor and Murata. Temporary overtake is possible; durability requires a 2028 OP path around KRW 3.8-4.5T and group OPM near 20%SEMCO at KRW 100T
2026-05-24Chip-design lens: why SEMCO’s FC-BGA and silicon capacitor exposure sit in the same bottleneck when AI performance becomes a data-movement-cost problemAfter NVIDIA: the AI semiconductor bottleneck
2026-05-23Marvell / Broadcom earnings-preview lens: whether AI custom chips and AI networking turn SEMCO’s silicon capacitor and FC-BGA exposure into a broader AI package bottleneckMarvell / Broadcom preview and Korea’s AI bottlenecks
2026-05-22ARM-rally lens: if AI racks become CPU + XPU + HBM + power-integrity systems, SEMCO’s silicon capacitor becomes one of the clearest second-order bottlenecksARM rally and Korea’s next AI bottlenecks
2026-05-15Three-division dissection: MLCC / FC-BGA / Optics, why it’s an “AI infrastructure parts platform,” and what KRW 1.02M already prices inSEMCO MLCC / FC-BGA / Optics Deep Dive
2026-05-07Mirae Asset lifts target to KRW 1.3M — the valuation frame shift from “Korean parts maker” to “global AI peer”SEMCO Mirae Target Frame Shift
2026-04-301Q26 head-to-head: Samsung Electronics vs Samsung Electro-Mechanics earnings qualityBig-Tech 1Q26: Samsung vs SEMCO
2026-04-21First rerating call: SEMCO as an AI infrastructure rerating candidateSEMCO AI Infrastructure Rerating

FAQ — Korean Semiconductor Value Chain

Q: Is OpenEdges Technology publicly traded? A: Yes. OpenEdges Technology is listed on KOSDAQ under ticker 394280. It is a Korean memory subsystem IP company licensing LPDDR6/5X Memory Controller, DDR PHY, and NoC interconnect to AI inference SoC fabless designers.

Q: How is this different from the AI PCB and Substrate Hub? A: This page is the top-level map for the Korean semiconductor value chain below HBM. The AI PCB and Substrate Hub is the deeper page for FC-BGA, MLB, CCL and low-loss materials.

Q: Who uses Samsung Foundry? A: External customers include Tesla (FSD SoCs across HW3/HW4 generations), Tenstorrent (Wormhole, Blackhole), Qualcomm (modem and selected Snapdragon variants), Google (Pixel Tensor SoCs), Ambarella (CV3-AD ADAS), plus Korean and Asian fabless customers (Rebellions, FuriosaAI, DeepX, etc.). Captive workload comes from Samsung’s own System LSI division.

Q: Is Samsung Foundry the same as Samsung Electronics? A: Samsung Foundry is the contract-manufacturing division inside Samsung Electronics’ DS (Device Solutions) division. It is not a separately listed entity. Public-market exposure runs through Samsung Electronics (KOSPI: 005930).

Q: Is Neosem publicly traded? A: Yes — Neosem is listed on KOSDAQ under ticker 253590. It is a Korean memory test equipment maker positioned for the HBM and CXL test capacity cycle.

Q: Is Exicon publicly traded? A: Yes — Exicon is listed on KOSDAQ under ticker 092870. It is a Korean memory test equipment maker, comparable benchmark to Neosem.

Q: Is AP System publicly traded? A: Yes — AP System is listed on KOSDAQ. It is a Korean display equipment maker exposed to OLED capex cycles, including the 8.6G OLED build.

Q: What is SF2 / SF3 / SF4? A: Samsung Foundry’s process node naming. SF4 = 4nm-class, SF3 = 3nm-class, SF2 = 2nm-class. SF2 is the next major node target whose external customer roster will materially shape Samsung Foundry’s positioning vs TSMC through 2027–2028.

Q: Does Samsung Foundry compete with TSMC? A: Yes. Samsung is widely regarded as TSMC’s #2 challenger at advanced nodes. TSMC retains the bleeding-edge AI hyperscaler customer base; Samsung competes effectively in automotive, AI accelerator startups, mobile mid-tier, and Korean/Asian fabless segments.

Q: What is SOCAMM2? A: A new server memory module standard based on LPDDR5X, providing higher bandwidth (up to 153.6 GB/s per module) and ~70% better power efficiency vs DDR5 RDIMM. SK hynix has begun mass production of 192GB SOCAMM2 optimized for NVIDIA Vera Rubin. Samsung also makes SOCAMM2.

Q: What is the LPDDR6 SOCAMM2 standard? A: A JEDEC standard in development extending LPDDR6 into AI server memory modules (and a complementary LPDDR6 PIM standard for in-memory inference). This is the standardization vector that converts LPDDR from “mobile memory” into “AI inference server memory.”


This hub is updated as new SemiScope, equipment / IP, foundry, substrate and PCB coverage publishes. For memory products, see the AI HBM Hub. For deeper substrate coverage, see the AI PCB and Substrate Hub. For AI company exposure, see Korean AI Companies Hub.

Last updated on 2026-07-04 12:43 KST
Built with Hugo
Theme Stack designed by Jimmy