Jeju Semiconductor — Revenue ₩180.5bn, OP ₩67.1bn, OPM 37.2%, +28% in One Day. Not an 'AI Chip Stock' — a Beneficiary of the 'Legacy Memory Squeezed by AI'

Fri, 15 May 2026 08:30:00 +0900

🔗 Related: Samsung Electronics Citi TP ₩460,000 — Memory-Cycle Frame Reset · KOSPI May 13 V-Reversal — All-Time High Through ₩3.76tn Foreign Selling · Why Korea Part 4 — $6.7B Inflows, Korea-Discount Dissolving or Value Trap?

Jeju Semiconductor closed +28% on May 14. The trigger is a 1Q26 print that came in extraordinarily strong — revenue ₩180.5bn, OP ₩67.1bn, OPM 37.2%; YoY revenue +273% and OP +1,713%. Most market participants are unfamiliar with what the company actually does. It isn’t Samsung or SK hynix. It doesn’t make HBM. It’s a fabless designer of low-power memory (MCP, LPDDR) for IoT devices, mid-tier smartphones, and automotive electronics. So why is it earning so much money?

TL;DR

1Q26 print. Revenue ₩180.5bn (+273% YoY), OP ₩67.1bn (+1,713%), NI ₩78.1bn. OPM 37.2%.
Stock reaction. +28.4% on May 14, close ₩75,600 (near limit-up ₩76,500). Turnover ~₩760.9bn.
What this company actually does. Not an AI-server HBM business. A fabless designer of low-power memory (MCP, LPDDR) for IoT, mid-tier smartphones, and automotive electronics. No fabs — design only.
Why earnings exploded. Samsung, SK hynix, and Micron concentrated fab capacity on HBM, shrinking supply of ordinary memory (LPDDR4X, legacy DRAM). Demand from IoT, smartphones, and autos kept going — supply down + demand steady = prices up. Jeju captured a large slice of that.
The single test. Is this earnings level repeatable? If 1Q was a one-off peak, the current price is rich. If 2Q sustains it, the price still has room.
Caveats. Customer A = 72.1% of revenue; China = 72.9% of revenue. Customer and geography concentration are extreme. Convertible-bond (CB) dilution overhang exists. The stock is under an “investment caution” designation by the exchange.

1. First — what does Jeju Semiconductor actually do

1.1 A “memory company” but completely different from Samsung / SK hynix

The Samsung Electronics analysis framed memory as the “work desk (DRAM)” and “filing cabinet (NAND).” Samsung and SK hynix operate the giant fabs that produce both, plus the AI-server-grade HBM at the high end.

Jeju Semiconductor plays in a completely different market.

Samsung Electronics / SK hynix:
→ Own their fabs
→ AI servers, PCs, smartphones — large-capacity memory
→ HBM, DDR5, high-performance NAND are the core
→ Global #1-3 megacaps

Jeju Semiconductor:
→ No fab (fabless = designs only, manufacturing outsourced)
→ IoT, mid-tier smartphones, automotive electronics — small-capacity memory
→ MCP, LPDDR, eMCP are the core
→ Market cap \~₩2.6tn (small-to-mid cap)

1.2 MCP, LPDDR — what these are

Two concepts unlock Jeju’s core products.

LPDDR (Low Power DDR): “low-power DRAM.” Uses less power than standard DRAM. Essential for battery-powered devices — smartphones, tablets, IoT.

MCP (Multi-Chip Package): multiple memory chip types packaged into one.

Analogy: a bento box
Standard memory = buying rice, side dishes, soup separately
MCP = rice + sides + soup all in one bento

NAND MCP = NAND (storage) + LPDDR (memory) bundled
eMCP = eMMC (embedded storage) + LPDDR bundled

Why it matters:
→ Fewer components → smaller, cheaper devices
→ Ideal for IoT sensors, smartwatches, mid-tier smartphones
→ Especially useful where space and power are constrained

1.3 Why an “ordinary memory” company suddenly earned this much

In the Samsung analysis, the explanation was that “HBM absorbs fab capacity.” That phenomenon directly impacted Jeju.

The AI-era memory supply structure:

Samsung / SK hynix / Micron fabs:
→ Concentrate output on HBM, DDR5, high-performance server memory
→ These carry much higher prices and margins
→ LPDDR4X and legacy DRAM output declines

Meanwhile, on the demand side:
→ IoT devices keep growing (5G IoT, smart-home, industrial sensors)
→ Mid-tier smartphone demand continues (especially China / SE Asia / LATAM)
→ Auto electronics — memory content per vehicle keeps rising

Supply down + demand steady → prices up

Jeju Semiconductor = specialist designer in that "ordinary memory" segment
→ Price uplift = revenue and earnings explosion simultaneously

One-line summary: Jeju is not an “AI chip stock.” It’s a beneficiary of the “legacy memory squeezed by AI.”

2. 1Q26 print — how strong are the numbers

2.1 Headline numbers

Item	1Q26	YoY	QoQ
Revenue	₩180.5bn	+273%	+97%
OP	₩67.1bn	+1,713%	+383%
NI	₩78.1bn	+1,716%	+417%
OPM	37.2%	+29.5pp	+22.0pp
NPM	43.3%	+34.4pp	+26.8pp

Cross-check: OPM = 67.1 / 180.5 = 37.2% ✓

2.2 Why the numbers are remarkable

OP rose 18× YoY. From ₩3.7bn a year ago to ₩67.1bn. Revenue rose 3.7×, but OP rose 18× — extreme operating leverage at work.

What operating leverage means:

Fabless (no-fab) cost structure:
→ Fixed: salaries, office, R&D — doesn't move much with revenue
→ Variable: contract manufacturing, raw materials — scales with revenue

If revenue doubles:
→ Variable cost doubles
→ Fixed cost stays the same
→ Profit grows much more than double

For Jeju:
Revenue +273% → OP +1,713%
→ Revenue 3.7×, OP 18×
→ Operating leverage was extreme

Adding to that, memory prices rose. Same unit volume × higher unit price = revenue rises. Costs don’t rise as fast — margin spikes.

2.3 Why NI exceeds OP

NI (₩78.1bn) is higher than OP (₩67.1bn). Normally NI is lower due to taxes and interest. NI > OP means non-operating income (FX gains, financial income, etc.) was material. The breakdown will appear in quarterly-report footnotes. As a company with >90% export ratio, KRW weakness produces FX gains.

Caveat: if the non-operating gain is one-off, annualizing NI directly is risky.

3. Revenue quality — where and to whom

3.1 Geography — China is 73%

Region	1Q revenue	Share
China	~₩131.6bn	72.9%
Other	~₩48.9bn	27.1%

3.2 Customer — Customer A is 72%

Customer	1Q revenue	Share
Customer A	~₩130.1bn	72.1%
Other	~₩50.4bn	27.9%

3.3 Application (2025 annual basis)

Application	Share	YoY growth
IoT	51.9%	+46%
Consumer	26.3%	+400%
Mobile	8.2%	+16%
Auto	8.0%	+126%
Network	5.5%	+6%

3.4 What the data says

The positive: IoT, consumer, and auto are growing simultaneously. Consumer at +400% YoY is striking. Auto at +126%.

The concerning: Customer A alone = 72%. China alone = 73%. Extreme concentration in one customer + one geography. If that customer reduces orders or China-bound exports face friction, earnings collapse fast.

The “200 global customers” narrative exists, but the data shows the core is one customer. Whether that customer’s order pattern repeats is the central 2Q variable.

4. Technical moat — real but not absolute

4.1 What Jeju is good at

Strength	Description
MCP productization	Designs combining NAND + LPDDR in a single package. Optimal for space/power/cost-constrained IoT and mobile
Platform certification	Qualcomm / MediaTek 5G IoT chipset certification. Without this, you can’t supply memory compatible with those chipsets
Multi-SKU coverage	200+ customers, covering IoT, mobile, auto, networking
Outsourced production network	Powerchip, Winbond partners. Mass production without owning fabs

4.2 What Jeju is not good at

Weakness	Description
No own fab	Dependent on partner capacity. When supply is tight, you can’t make more even if you want to
Not cutting-edge	LPDDR4X / legacy on mature processes, not HBM, DDR5, or latest LPDDR5X
Pricing power is cycle-dependent	Earns big when supply is tight; loses pricing when supply normalizes

4.3 Core read

Jeju’s moat exists — platform certification, MCP productization, customer-handling experience are not easily replicated. But it’s not the absolute moat that Samsung / SK hynix HBM or TSMC leading-edge processes offer. If large memory players re-engage with legacy or Chinese players catch up, competition intensifies.

So Jeju is more accurately framed as “a niche memory specialist whose earnings explode when cycle and product mix align,” not as a “tech-monopoly” name.

5. Fair value — looks cheap but contains a trap

5.1 Simple math

1Q NI ₩78.1bn × 4 = ₩312.4bn (naive annualization)
Market cap \~₩2.6tn ÷ ₩312.4bn = PER \~8.3×
→ PER 8× looks very cheap

But this math is dangerous. If 1Q is the cycle peak, annualization is meaningless. 2Q earnings could collapse.

5.2 Scenario fair value

Scenario	2026 revenue	OPM	OP	NI estimate	EPS	Applied PER	Implied price
Bear (1Q peak)	₩540.0bn	24%	₩129.6bn	₩95.0bn	₩2,758	15×	~₩41,000
Base	₩680.0bn	31%	₩210.8bn	₩165.0bn	₩4,790	16.5×	~₩80,000
Bull (cycle sustained)	₩820.0bn	35%	₩287.0bn	₩225.0bn	₩6,532	18×	~₩118,000

Cross-checks: Base EPS = ₩165bn / 34.44M shares ≈ ₩4,790 ✓; implied price = 4,790 × 16.5 = 79,035 ≈ ₩80,000 ✓

5.3 Where the current price sits

₩75,600 is very close to the base-case implied price (₩80,000). Upside ~+5.8%.

Upside / downside:
Bear case: -46%
Base case: +6%
Bull case: +56%

→ Even if the base case is right, additional upside is small
→ If bear case is right, the drawdown is large
→ Chasing the day after +28% is unfavorable risk-reward

6. Caveats — three risks

6.1 Customer / geography concentration

Customer A = 72.1%, China = 72.9%. One customer’s order pattern drives the whole P&L. Whether Customer A is a long-term contract or a one-off large order isn’t disclosed.

6.2 Convertible-bond (CB) dilution

Jeju issued CB and BW (warrants) totaling ₩117.0bn. Conversion / exercise price ₩44,300, ~2.64M latent shares (~7.7% of existing share count). With the stock well above the strike, conversion incentive is strong — diluting EPS.

CB dilution effect:

Current share count: \~34.44M
Latent conversion: \~2.64M (7.7%)

If converted:
→ Shares 34.44M → 37.08M
→ EPS \~7% lower

6.3 Investment-caution designation

KRX has designated Jeju as an investment-caution stock. Further sharp rises can trigger a 1-day trading halt (investment-warning). This affects short-term flow independent of earnings.

7. Flow — foreigners selling, retail buying

May 14 flow:

Participant	Net
Foreign	-₩13.4bn (net sell)
Institutional	+₩9.5bn
Retail	+₩5.0bn
Program	-₩14.7bn (net sell)

Foreigners and program sold; retail bought. This is closer to “retail-driven momentum buying” than “clean institutional accumulation.” Sustained move requires foreign / program flow to turn net-positive.

The May 13 KOSPI V-reversal piece noted the same pattern at the index level — ₩3.76tn foreign selling absorbed by domestic flows. Jeju mirrors that structure: domestic capital catching what foreigners drop. How long it lasts is the key short-term variable.

8. What to watch — 2Q decides

8.1 1Q vs 2Q

Item	1Q (confirmed)	2Q (to verify)
Revenue	₩180.5bn	≥₩170bn = positive; <₩130bn = 1Q-as-peak confirmed
OPM	37.2%	≥33% = structural; <25% = cycle peak
Customer A orders	72.1%	Repeatability is core
Memory pricing	Up	Sustained / rising = good; reversal = bad

8.2 Key triggers

Trigger	Window	Why
LPDDR4X price trajectory	Monthly	Prices holding → 2Q stays strong; rolling over → margin collapses
2Q print	August	Verdict on 1Q (one-off vs structural)
Proprietary LPDDR4X mass production	2H onward	Transition from sourcing margin to design-margin
Investment-caution removal	Days	Direct effect on short-term flow
CB / BW conversion activity	Ongoing	EPS-dilution monitoring

8.3 Linking back to the Samsung analysis

The Samsung piece argued “HBM absorbs fab capacity → legacy memory prices rise.” Jeju is the direct beneficiary of that legacy price uplift.

Samsung: HBM / DDR5 / SOCAMM2 → makes and sells directly
Jeju: LPDDR4X / MCP → benefits where Samsung isn't producing

Samsung AI memory capex → legacy memory undersupply → Jeju pricing leverage

This linkage breaks if:
→ Samsung / SK hynix resume legacy production at scale, OR
→ LPDDR4X demand contracts, OR
→ Chinese memory players ramp supply

9. Bottom line

Jeju Semiconductor’s 1Q26 print is strong. Revenue ₩180.5bn, OP ₩67.1bn, OPM 37.2%. OP up 18× YoY. The numbers themselves are undeniable.

Whether this is a “repeatable structure” or a “one-off cycle peak” is settled in 2Q. Customer A = 72%, China = 73% are extreme concentrations. CB dilution and the investment-caution designation are real risks. After a +28% day, the chase upside is only ~+6% even in the base case.

Jeju is not an “AI chip stock.” It’s a beneficiary of the “legacy memory squeezed by AI.” How long that benefit lasts depends on LPDDR4X pricing, Customer A’s order repeatability, and the pace of transition to proprietary-design products. The most disciplined approach is to wait for the 2Q print (August) and confirm OPM staying ≥33%.

Whether 1Q is the “start” or the “peak” — that answer hasn’t arrived yet.

FAQ

Q: Is Jeju Semiconductor an AI chip company? A: Not directly. It doesn’t make AI-server HBM or cutting-edge GPU memory. It’s a fabless designer of low-power memory (MCP, LPDDR) for IoT, mid-tier smartphones, and automotive electronics. But Samsung/SK hynix concentrating capacity on HBM tightened supply of ordinary memory, which lifted prices and benefited Jeju.

Q: Is OP rising 18× from ₩3.7bn to ₩67.1bn normal? A: Two effects coincided. (1) Revenue rose +273%. (2) Memory prices rose. Fabless has small fixed costs, so revenue growth produces disproportionate earnings growth (operating leverage). But the structure runs in reverse too — revenue declines compress earnings disproportionately.

Q: After a +28% day, is it safe to buy now? A: Base-case implied price is ~₩80,000. Vs current ₩75,600, upside is only ~+6%. Bear-case downside (1Q-as-peak) is -46%. The asymmetry is unfavorable for chasers. Waiting for the 2Q print is more disciplined.

Q: Who is Customer A? A: Not disclosed. Only the 72.1% revenue share appears in the quarterly report. Market speculation points to global 5G IoT chipset-related OEMs (Qualcomm / MediaTek ecosystem) or Chinese IoT/mobile OEMs. Whether Customer A’s order pattern repeats or proves one-off is the central 2Q variable.

Q: What does the investment-caution designation mean? A: KRX designates stocks with sharp short-term gains. Further sharp rises can escalate to “investment-warning” with a 1-day trading halt. Affects short-term flow regardless of earnings.

Q: How significant is the CB dilution? A: Latent conversion is ~2.64M shares (7.7% of existing count). With the strike at ₩44,300 and the stock well above, conversion incentive is strong. Conversion lowers EPS ~7%. Fair-value calculations should reflect dilution for conservatism.

Q: What’s the fastest signal in 2Q? A: Two. (1) Monthly LPDDR4X market pricing — sustained or rising. (2) 2Q print (August) — revenue ≥₩170bn and OPM ≥33%. Both confirm “structural”; either failing supports “1Q-as-peak.”

This article is for research and informational purposes only and does not constitute investment advice. 1Q figures based on the DART quarterly report (submitted 2026-05-14). Customer and geographic concentration percentages are from the quarterly report; Customer A’s identity is undisclosed. Application-mix figures are 2025 annual basis (USD), not 1Q. Fair-value scenarios are analyst estimates; the company has not published guidance. Public consensus is thin, limiting quantitative versus-consensus comparison. CB/BW dilution scale (~2.64M shares, 7.7%) is from press reports; exact conversion conditions require separate verification. The stock is under an investment-caution designation, with potential trading halt on further sharp rises. May 14 flow showed foreigners and program as net sellers. Analysis can be wrong. Data cut: May 15, 2026 KST.

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.

Legacy Memory on Korea Invest Insights