Haesung DS (195870.KS) — Korean Lead-Frame #2 Pivoting to AI Heat-Spreader Second-Source: 1Q26 Margin Shock, DDR5 Utilization, and Why the Re-rating Sits in Multiple Compression, Not Revenue

Thu, 07 May 2026 20:00:00 +0900

🔗 Related: AI PCB & Substrate Hub · Korea Semi · HBM Hub · Korea AI PCB Ecosystem — 10 Companies

The AI PCB thesis post framed AI substrate demand as a system-wide bottleneck — not a single-chip story but a rack-scale BOM. This piece adds the eleventh name to that cluster: Haesung DS (195870.KS), where the foreign-allocator angle is unusual. Unlike the other ten — all PCB, substrate, or CCL plays — Haesung DS sits at a different node: automotive lead-frame compounder potentially pivoting to AI heat-spreader second-source, in a market that the named incumbents (Shinko, Honeywell/Solstice, Jentech, I-Chiun) have ~85% locked but no longer have the capacity to serve.

TL;DR

Haesung DS is not a clean AI play. It is a three-axis re-rating candidate. Axis 1: automotive lead-frame recovery (≈75% of revenue mix). Axis 2: DDR5 memory package substrate utilization normalization (1H25 ~30% → 2026 ~70% per KB Securities). Axis 3: AI data-center heat-spreader second-source optionality. The three axes don’t need to fire in the same quarter — they accumulate sequentially, which is what makes the asymmetry interesting.
1Q26 was a strong-revenue / weak-margin timing shock, not a structural break. Revenue ₩188.7bn (+37.2% YoY, near-record), but OP ₩11.0bn at 5.8% OPM — versus 4Q25 OPM 12.1%. The driver was Cu/Au/Ag/Pd input cost inflation hitting before the customer ASP pass-through caught up (~1-quarter lag). The 2Q26 OPM print is the first verification gate. Sell-side consensus assumes ₩208.5bn revenue and 10% OPM in 2Q.
The AI heat-spreader entry is “technically plausible” — but reading it as direct top-bin GPU-lid supply is overreach. The IC package heat-spreader market is ~85% concentrated among Shinko (Japan), Honeywell/Solstice (US), Jentech (Taiwan), and I-Chiun (Taiwan). Shinko capacity is reportedly stretched as AI accelerator power crosses 700W+, and Korean lead-frame players are receiving big-tech development requests. Realistic base case: flat heat-slug / heat-spreader second-source, not advanced vapor chamber lid or micro-channel lid (those sit with Jentech / I-Chiun).
Direct revenue contribution is too small to move the P&L. The re-rating mechanism is multiple compression, not earnings power. TAM is ~US$567m in 2024 with 9.7% CAGR. A 3% share at 2026 TAM (~US$682m) at USD/KRW 1,450.98 implies ~₩29.7bn — about 4.5% of FY25 revenue ₩653.4bn. The investment payoff is the company being re-classified from “automotive lead frame + DDR substrate cyclical” to “AI thermal supply-chain second-source” — and the discount compressing toward Korean AI-substrate peers.
86,000 won (₩1.46tn cap) is a “good name / wrong price” zone. Closing +10.3%, 20D return +65.4%, 26% above 20DMA, RSI14 71.9. Chasing makes sense only as a 30–50bp pilot. The asymmetric entry sits at the 82,000–84,000 retest of the prior 83,800 high. 76,000 break = re-evaluate the technical thesis.

1. What Haesung DS actually is — for the foreign allocator who hasn’t seen this name before

Haesung DS is not a chipmaker. It is not a foundry, not a memory company, not an OSAT. It manufactures the metal and substrate parts that sit inside the semiconductor package — the lead frame that connects the die to the PCB, and the memory package substrate that connects DRAM dice to the motherboard.

Item	Detail
Company	Haesung DS / HAESUNG DS
Ticker	KOSPI 195870
HQ / Sites	Seoul Gangnam HQ + Changwon plant + China / Japan / Philippines sales-production presence
Business segments	Lead frame + Memory package substrate + Tape substrate + Graphene
Reference date	2026-05-07
Closing price / market cap	KRW 86,000 / ~KRW 1.46 trillion
Industry positioning	Global lead-frame #2 (Mitsui High-Tec is #1)

The two revenue pillars are Lead Frame and Package Substrate. FY25 revenue was ₩653.4bn — Lead Frame ~₩493.8bn (75.6%) and Package Substrate ~₩159.6bn (24.4%) per Kyobo Securities’ segment estimate (the company itself reports consolidated revenue and OP, not full segment splits).

For the allocator who has not previously held this name, the two-line takeaway:

Lead Frame is the “metal skeleton” inside a chip package: the structure that supports the die, routes signals out to the PCB, carries current in, and dissipates heat. Automotive lead frames in particular are difficult to substitute on price because automotive ICs require long-cycle qualification, traceability, AEC-Q100 reliability data, and very low PPM defect rates. This is the first moat layer.
Package Substrate is a small circuit board between memory die and motherboard. Haesung DS’s edge is the Reel-to-Reel (R2R) continuous lamination process — closer to roll-to-roll volume manufacturing than panel-by-panel. R2R does not compete with high-layer-count AI FC-BGA (that field belongs to Samsung Electro-Mechanics, Daeduck, Ibiden, AT&S etc.), but it has cost-and-yield advantages on volume DDR4/DDR5 memory substrate.

The new optional layer being priced in at the 86,000 won close is the AI data-center heat spreader / heat slug opportunity sitting on top of these two pillars.

2. The 1Q26 print — strong revenue, broken margin

2-1. Quarterly P&L flow

(KRW bn)

Period	Revenue	GP margin	Operating profit	OPM	Δ
FY24	603.0	18.5%	56.9	9.4%	base
FY25	653.4	15.0%	46.5	7.1%	rev +8.4% / OP -18.3%
4Q25	179.9	19.7%	21.8	12.1%	OP QoQ +35%
1Q26	188.7	12.3%	11.0	5.8%	rev +37.2% YoY / OP QoQ -49.6%
1Q26 TTM	704.6	15.3%	57.1	8.1%	rev +7.8% / OP +22.8% vs. FY25

The shock is sharp. Top-line print was near-record — automotive segment alone delivered ~₩91.0bn (≈48% of total). But OP halved sequentially, and OPM dropped from 12.1% in 4Q25 to 5.8% in 1Q26.

2-2. The cause is mechanical, not structural — but verification matters

Cu, Au, Ag, and Pd input prices rose ahead of customer ASP pass-through, with a roughly one-quarter lag. 1Q26 closed with expensive raw materials still in COGS and ASPs not yet repriced. This is the textbook “cost-pull margin” pattern in a mid-cap component company.

Three readings, in order of probability:

Timing issue (high probability) — 2Q26 ASPs catch up, OPM normalizes to ~10%. 1Q is a one-off. This is what consensus is pricing.
Structural margin pressure (medium probability) — customers resist price hikes, or low-cost Chinese LF entrants close the spread. OPM stalls at 7–8%.
Demand softening (low probability) — 2Q top line itself rolls over. The 1Q +37.2% YoY pace makes this the lowest-probability scenario right now, but EV / auto-IC inventory swings are not impossible.

Sell-side consensus (Meritz) assumes 2Q26 revenue ₩208.5bn, OP ₩20.8bn, OPM ~10%. The 86,000 won close is priced for scenario 1. If 2Q OP comes in below ₩15bn, the recovery thesis breaks and price retracement is rational.

3. Splitting the business into “core” vs. “optional” axes

The cleanest mental model for Haesung DS — and the one that makes the asymmetry visible — is to separate what is already happening from what is being priced as optionality.

Axis 1 — Core: recovery already underway

Item	1Q26 status	2026 path
Automotive lead frame	~₩91bn quarterly (48% of total, near-record)	EV / auto-IC demand recovery → quarterly ~₩90bn run-rate stabilization
Package substrate utilization	1H25 ~30% → 1Q26 recovery underway	KB Securities models 70%+ in 2026 → fixed-cost leverage
DDR5 / D1b / D1y substrate	Anonymous Korean / Chinese / US customers in qualification or early ramp	4Q25 was the first profitable substrate quarter in seven; 2026 mix shift up

This axis alone supports 2026 revenue of ₩733–784bn and OP of ₩79–109bn (sell-side range: iM ₩733.5bn / ₩79.0bn / 10.8% OPM, conservative; KB ₩784.3bn / ₩108.9bn / 13.9% OPM, aggressive). KB’s bull case turns on substrate utilization reaching 70%.

Axis 2 — Optional: AI data-center heat-spreader entry

This is the axis the +10.3% reaction on May 7 was about.

The IC package heat-spreader space is ~85% concentrated in Shinko (Japan), Honeywell/Solstice (US, post the Honeywell Advanced Materials spin-off in October 2025), Jentech (Taiwan), I-Chiun (Taiwan). Shinko is the long-standing server CPU heat-spreader incumbent — stamping plus surface finish is its core capability stack.
AI accelerator power has crossed 700W+ (Blackwell, MI300 class) and is rising. Heat spreader thermal demand is climbing with package size, warpage, and TIM contact requirements. Shinko’s capacity is reportedly stretched, and big-tech buyers are explicitly soliciting development from lead-frame suppliers as second-source candidates.
Haesung DS’s lead-frame manufacturing stack — etching, stamping, Ag plating, PPF plating, down-set, ±2μm/1m tooling, deep down-set, AEC-Q100 Grade 0 surface treatment — is the necessary starting point. The question is whether it is sufficient.
Reporting (EBN, Samsung Securities’ Lim Eun-young commentary) indicates Haesung DS is conducting heat-slug quality testing with a customer in 2Q26 with 2H26 revenue contribution if the order is confirmed.

This axis is not yet base case for the P&L. But it is already partially priced in the equity.

4. Mapping where Haesung DS can — and cannot — go in AI thermal

4-1. Four-tier AI-server thermal value chain

Tier	Component	Key competitors	Haesung DS penetration probability
Package metal lid / heat spreader	Heat slug, heat spreader, lid, stiffener	Shinko, Honeywell/Solstice, Jentech, I-Chiun	Medium
Advanced lid	Vapor chamber lid, micro-channel lid	Jentech, I-Chiun	Low
TIM	TIM1/1.5/2, graphite film, liquid metal, indium	Solstice, Henkel, Indium Corp	Low
System cooling	Cold plate, liquid cooling module, CDU, manifold	Delta, AVC, Auras, Jentech, Vertiv	Very low

Only tier 1 is in scope. The investment thesis is that Haesung DS becomes a package-level metal heat-spreader second-source, not that it leapfrogs into liquid cooling or TIM materials.

4-2. Penetration probability by product difficulty

Product	Difficulty	Penetration probability	Rationale
Generic CPU/GPU heat spreader	Medium	Medium–High	Strong overlap with existing lead-frame process
AI ASIC / edge-AI heat slug	Medium-High	Medium	Customer co-design needed; easier than top-bin GPU
Blackwell / MI300-class large AI accelerator lid	High	Low–Medium	Incumbent qualification + warpage / TIM optimization barrier
Vapor chamber lid	Very high	Low	Jentech / I-Chiun ahead
Micro-channel lid	Very high	Very low	Requires fluidics / liquid-cooling integration
TIM materials	High	Very low	Solstice / Honeywell domain
Cold plate / liquid cooling system	High	Very low	Outside current business

The realistic 2026 read: “AI thermal entry” should be interpreted as a flat heat-slug second-source, not a top-bin GPU lid contract.

4-3. Why the AI lid is not just a “bigger lead frame”

Lead-frame skill is necessary but not sufficient. AI package lids add requirements that aren’t on the automotive LF spec sheet:

Requirement	Why it matters
Flatness / warpage control	Die-to-TIM contact failure spikes thermal resistance
Surface roughness / plating uniformity	Determines TIM1/TIM2 contact quality
CTE-mismatch management	Silicon, substrate, copper-lid thermal expansion differences create package stress
Outgassing / contamination control	Affects high-end package assembly yield
Large-area machining yield	AI packages are big — small deformation is critical
Customer co-design	Simultaneous optimization with Nvidia / AMD / ASIC vendor / foundry / OSAT / TIM vendor

Flatness, plating uniformity, and surface treatment reliability are areas Haesung DS already demonstrates in automotive lead frames. CTE-mismatch management, large-area yield, and co-design with hyperscaler ASIC teams are AI-specific territory where incumbent qualification is a binary outcome.

5. Revenue scenarios — base / bull / bear

5-1. TAM math

IC package heat spreader: ~US$567m in 2024, CAGR ~9.7% (third-party market research; treat as directional).

2026E TAM ≈ US$567m × 1.097² ≈ US$682m
2028E TAM ≈ US$567m × 1.097⁴ ≈ US$821m

At USD/KRW 1,450.98 (May 7 print):

Share assumption	Revenue translation
1% of 2026E TAM	~₩9.9bn
3% of 2026E TAM	~₩29.7bn
5% of 2026E TAM	~₩49.5bn
5% of 2028E TAM	~₩59.6bn

Direct revenue is not P&L-changing in 2026. ₩29.7bn at 3% share is ~4.5% of FY25 revenue ₩653.4bn. The investment value is in multiple compression, not contribution margin.

5-2. Three scenarios

Scenario	Product / customer	Timing	2026 contribution	2027 contribution	Investment implication
Base	Flat heat slug / heat spreader, anonymous big-tech or package company second-source	2Q26 qualification → 2H26 small revenue	₩5–20bn	₩20–50bn	“Confirmed AI-supply-chain entry” → discount compresses
Bull	AI ASIC / GPU heat-spreader second-source, ≥2 customer qualifications	2026 end → 2027 volume	₩10–30bn	₩50–100bn	Re-classified from lead-frame to AI-thermal supplier — meaningful multiple re-rating
Bear	Stuck at sample / development; customer testing delays; yield miss; incumbent capacity normalizes	2026 revenue stillborn	₩0–5bn	₩0–10bn	Recent rally retraces; thesis discarded

Heat slugs look like simple metal parts. They are actually package reliability components. Failing customer qualification = no revenue. The bear case is non-trivial.

6. Where Haesung DS sits in the Korean AI substrate cluster

Adding Haesung DS as the eleventh name to the Korea AI PCB ecosystem 10-companies framework:

Company	Primary exposure	AI direct exposure	Unique angle
Samsung Electro-Mechanics	FC-BGA + MLCC	Very strong	AI substrate + MLCC dual exposure
Daeduck Electronics	FC-BGA / MLB	Strong	Korean #2 FC-BGA volume
Isu Petasys	MLB	Strong	AI GPU board MLB
Simmtech	Memory package substrate	Medium	DDR5 memory substrate
Korea Circuit	Package substrate	Medium	Memory + mobile
Doosan Electro-BG	CCL	Strong	Low-Dk CCL
Kolon Industries	Materials	Medium	Polyimide / film
Pamicell	CCL materials	Medium	Doosan Electro-BG proxy
TLB	Memory module PCB	Medium	DDR5 module PCB
Taesung	Equipment	Medium	PCB equipment
Haesung DS (new)	Lead frame + DDR substrate	Medium (with strong optionality)	Auto LF + AI heat-spreader second-source

Haesung DS’s unique angle in the cluster: it is the only name that does not sit on a PCB / substrate / CCL line. The other ten are all on the substrate or substrate-feeder axis. Haesung DS is on a parallel evolutionary track — lead frame → AI thermal — that no one else in the cluster owns.

This is also the structural reason for the +10.3% reaction on May 7. Axis 1 (auto LF) and axis 2 (DDR5 substrate) are largely understood by the market, in line with the May 6 Korean semis rally synthesis. Axis 3 (heat spreader) is not yet fully in consensus numbers — and that is what the multiple is starting to reflect.

7. Entry strategy at 86,000 won

7-1. Where the price actually is

Metric	Value	Read
Daily change	+10.3%	52-week breakout
20D return	+65.4%	Short-term spike
Distance from 20DMA	+26.2%	High chase risk
Distance from 50DMA	+41.9%	Overheated
RSI14	71.9	Near overbought
Volume	1.7× 20D average	Breakout has volume confirmation
Close in daily range	81% from low	Strong end-of-day buying

7-2. Multiple check

At 2026E OP ₩79–109bn against ₩1.46tn market cap, cap/OP is 13.4–18.5×. Korean substrate peers (Simmtech, Daeduck) trade at 17–22× OP. The upper bound is stretched, but the lower bound is reasonable — the multiple is not yet at peer-average.

7-3. Entry zones

Zone	Action
85,000–87,000 (current)	Chase only as 30–50bp pilot
82,000–84,000	Preferred first entry — retest of prior 83,800 high
78,000–80,000	Best risk-reward — May 6–7 breakout box bottom
88,000+ on close	Add only if turnover re-expands and foreign / institutional flows confirm
76,000 break	Re-evaluate technical structure

ATR14 ~5,843 won (~6.8% daily volatility). A normal 1-ATR pullback lands around 80,000. The 82,000–84,000 retest zone is the asymmetric one.

7-4. Why not full-size now

1Q26 was a miss, not a beat. Top line was good but ₩11.0bn OP fell short. Today’s price is already pricing 2Q recovery and partial heat-spreader optionality.
Recovery hasn’t been confirmed yet. 2Q OP ≥ ₩20bn justifies the level. ≤ ₩15bn means breakout fails.
The chart is strong but extended. “Strong names go further” works as a tactical bias, but not at full position size from this entry zone.

The actionable framing for an institutional book: good name, wrong price. The retest is the higher-quality entry.

8. Checkpoints & invalidation

Window	Checkpoint	Read
2Q26 print	OPM recovery + heat-slug qualification mention	Recovery + axis 3 progressing
2Q–3Q26	New CAPEX or line conversion disclosure	Volume ramp signal
3Q26	New revenue line item	Theme → numbers
2H26	≥2 customers in qualification	Single-customer concentration risk reduced
2027	Heat-spreader revenue ≥ ₩30bn	Minimum re-rating evidence
2027	Mention of advanced lid / VC lid / micro-channel	Bull case entry signal
Anytime	2Q26 OPM persistently ≤ 7% / heat-slug effort stalls	Rally retraces, thesis is invalidated

9. Risks

Technology

R2R substrate is structurally not for high-layer FC-BGA / 2.5D packaging. Reading Haesung DS as a direct beneficiary of the AI-packaging trend is overreach.
AI thermal qualification is binary. Heat slug / spreader is attractive optionality, but customer / size / margin are not yet publicly verified.

Business / customer

Customer concentration is opaque. Per-customer revenue isn’t disclosed; quantifying memory-customer or auto-IDM dependence is hard.
Auto exposure ~75–77%. EV / autonomous-driving slowdown hits directly. Memory mix is the lowest in the four-name peer cluster — the direct memory super-cycle benefit is the smallest.
Raw-material pass-through lag. Cu / Au / Ag / Pd spikes can replay 1Q26-type margin damage.

Macro / regulatory

EV demand and policy volatility drive the auto LF axis.
US-China supply chain rebalancing affects the global customer mix, export controls, and FX.

Valuation

The discount is double-edged. Cheap-for-a-reason is partially structural (auto weight + substrate latecomer) and may not compress in a single quarter.

10. FAQ

Q: Is Haesung DS a clean HBM proxy? A: No. Haesung DS does not produce HBM stacks. AI infrastructure expansion lifts DDR4 / DDR5 substrate demand, which is a second-order beneficiary position. For direct HBM exposure, see the SK Hynix HBM market-share analysis.

Q: Can Haesung DS displace Shinko? A: Not in the near term. Shinko holds incumbent qualification on server CPU heat spreaders. The realistic outcome is coexistence as second-source — capacity gap absorption, not replacement. Big-tech buyers want supply diversification; Haesung DS fits that allocator demand, not the disruption narrative.

Q: Is 1Q26 OPM 5.8% structural? A: Probably not. The +37.2% YoY top line and the documented one-quarter raw-material pass-through lag both point to a timing issue. The 2Q26 OPM print is the verification gate — recovery to 10%+ resolves the question; persistence at ≤7% would force a structural re-read.

Q: Isn’t 75% automotive exposure too concentrated? A: This is the principal two-sided risk. Strong EV / auto-IC demand carries axis 1; auto-cycle softness hits directly. The thesis depends on axis 2 (DDR5 substrate utilization) and axis 3 (heat spreader) accumulating to dilute the auto dependency. As a single-axis name, this is risky; as a three-axis re-rating candidate, it is viable.

Q: Is the 86,000 close chaseable? A: Not for a full-size book. The asymmetric entry is at the 82,000–84,000 retest. Chase only as a 30–50bp pilot. 76,000 break = re-evaluate the technical structure. Allocator translation: “good name, wrong price.”

Q: How does this fit alongside other Korean AI substrate names? A: It occupies a slot none of the other ten cluster names hold. Samsung Electro-Mechanics, Daeduck, Isu Petasys, Simmtech, Korea Circuit, Doosan Electro-BG, Kolon, Pamicell, TLB, and Taesung are all on the substrate / CCL / equipment axis. Haesung DS is the only “lead-frame → AI thermal” path. From a cluster-construction viewpoint, this adds genuine angle diversification rather than another correlated AI-substrate proxy.

Q: What if axis 3 (heat spreader) fails entirely? A: Axis 1 (auto LF recovery) + axis 2 (DDR5 substrate utilization) alone supports 2026 OP ₩70–90bn and cap/OP ~16–21× — a defensible stand-alone case. The 86,000 close has partially priced in axis 3, so failure could send the stock back toward ~70,000 — but the company itself does not require axis 3 to be solvent.

Q: When is the cleanest moment to add? A: Three combinations make the thesis “confirmed” rather than “priced”:

2Q26 OPM ≥10% + heat-slug qualification mention,
Disclosed CAPEX or line conversion in 2Q–3Q26,
A new revenue line item in 3Q26.

Two of those three converging is sufficient for a position upgrade from “watchlist / pilot” to “core.”

This article is for research and informational purposes only and does not constitute investment advice. All figures are sourced from the company’s IR materials, sell-side reports (Kyobo, SK, iM, KB, Meritz, Samsung Securities), and third-party market research. Anonymous customer designations follow sell-side report conventions; actual customer names, order sizes, and pricing terms have not been publicly verified. Before any investment decision, consult the company’s IR materials, the latest quarterly report, and applicable filings directly.

Disclaimer: For research and information purposes only. Not investment advice. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.

Lead Frame on Korea Invest Insights