Who actually makes money when AI silicon sells? It is not just the GPU and HBM makers. The 'back end' also has major winners. Two regions matter — substrates and test sockets. Both are grouped under 'AI back-end,' but the investment structures are completely different. Substrates are a direct CAPEX beta on AI-server build-out. Volumes rise, packages get bigger, ASPs lift. Short-term momentum is strong: Daeduck Electronics 1Q26 OPM 14.8%, Samsung Electro-Mechanics package solutions revenue +45%. Test sockets are a high-margin consumables beta on chip complexity. As chips get more complex, testing gets harder, and the sockets have to be remade. ISC 1Q26 OPM 35%, AI revenue 81% of total. LEENO Industrial OPM 47%. Same AI tailwind, but the margin structures differ by ~3x. Don't lump the two as one 'AI theme.' For 3–6 month momentum, substrates. For 1–2 year holding, test sockets.
Pearl Abyss said on its 1Q26 call that it is 'exploring various ways including DLC' for Crimson Desert. Foreign outlets reported it as 'DLC confirmed,' the market twitched, and then sold off. Even after a record quarter (revenue KRW 328.5B, operating profit KRW 212.1B), the stock stayed weak. But the meaning of the DLC comment is being misread. The point is not 'KRW 60B in DLC revenue.' The point is that Crimson Desert is being proven as a recurring-revenue franchise IP rather than a one-shot package game. The real reason the market is applying a single-digit PER to Pearl Abyss is the assumption that '2026 base-game revenue is real, but 2027 is a cliff.' If DLC is formalized, that assumption breaks. The reclassification — package-game maker → franchise-IP holder — moves the multiple far more than KRW 60B of incremental revenue moves EPS.
On May 15, the KOSPI crashed -6.12% — the day right after it breached the 8,000 all-time high. KOSDAQ fell -5.14%. The sell-side circuit breaker fired. On days like this, the temptation is to hunt the names that did NOT fall in search of the next leadership stock. Hana Micron +18.6%, Jeju Semiconductor +8.9%, Samsung Electro-Mechanics -1.4% (RS vs index +4.7pp). Relative-strength (RS) names clearly exist. But you cannot simply treat them as 'Monday buy candidates.' The real read is the macro gate — US 10-year 4.46%, Brent 108, USD/KRW near 1,500, VIX 18.6 — and unless at least two or three of those settle, chasing even good RS names is inefficient. The next leadership stocks are not found AFTER macro recovers; they are the names that did NOT fall WHILE macro was bad, recorded, and bought ONCE macro recovers.
Markets convulsed between May 14 and 15. US April CPI +3.8% YoY, PPI +6.0% (the largest since 2022). Japan April PPI +4.9%, JGB 10-year at 2.55% (highest since 1997). US 10-year 4.46%, 30-year 5.02%. Brent 108 USD. The Strait of Hormuz is functionally closed. The Xi–Trump summit ended in a 'small deal.' Markets read these as separate stories: an Iran story, a US-CPI story, a Japan-BOJ story, a US-China story. They are not five independent events. They are one cycle. Iran/Hormuz is the origin. Oil is the first transmission. US inflation is the second. US long rates are the third. Japan's BOJ is the fourth. A higher global discount rate is the terminus. The US-China summit is a side variable that accelerates or delays the cycle, not its cause. Once you see the picture, you can trace where the next headline will land in other asset classes.
Jeju Semiconductor reported 1Q26: revenue ₩180.5bn (+273% YoY), OP ₩67.1bn (+1,713%), OPM 37.2%. The stock closed limit-up +28.4% on May 14. This is not an AI-server HBM business. Jeju is a fabless memory designer for IoT, mid-tier smartphones, and automotive electronics — MCP and LPDDR. So why the earnings explosion? With Samsung, SK hynix, and Micron concentrating fab capacity on HBM, supply of 'ordinary memory' (LPDDR4X, legacy DRAM) tightened while IoT / smartphone / auto demand kept running — and prices rose. The single test now: is this earnings level repeatable, or was 1Q the cycle peak? If 1Q is the peak, the current price is rich. If 2Q sustains it, the price is still attractive.
On May 13, 2026, the KOSPI opened down -1.69% and slid toward 7,400, before closing at 7,844.01 (+2.63%) — a fresh all-time high. The same session saw foreign investors net-sell ₩3.76tn, absorbed by retail (+₩1.89tn) and domestic institutions (+₩1.69tn). The KOSDAQ printed -0.20%, and breadth was negative — 955 advancers vs. 1,455 decliners in the local DB. This is not a 'broad rally.' It's a compressed large-cap rally. SK hynix +7.68%, Samsung Electro-Mechanics +7.41%, Daeduck Electronics +11.64%, Hyundai Mobis +18.43% — what was working worked harder, and what was lagging stayed flat. The most analytically interesting fact is that the move ignored a full set of risk-off macro inputs — hot US CPI, WTI back above $100, Middle East tension, soft SOX the day before. Macro had its head turned by flow and large-cap news flow.
KOSPI near 7,844 as KR holds Neutral vs US Bull. Six 3-screener overlap candidates led by Gigabis, TES, and Jeju Semiconductor screen for quality re-rating.
The two names at the top of Korea's robotics value chain: Robotis (actuator specialist — hand and joint motion) and Rainbow Robotics (humanoid platform — Samsung Electronics 35% stake). Robotis 2025 revenue ₩38.9bn, OP turn-positive ₩3.4bn, market cap ~₩5.1tn. Rainbow 2025 revenue ₩34.1bn, OP -₩2.5bn (loss), market cap ~₩16.2tn. Rainbow's market cap is 3× Robotis's while revenue is smaller and unprofitable — the gap is the price of the 'Samsung option.' Two distinct questions: for Robotis, 'does the actuator-supplier earnings trajectory deliver?'; for Rainbow, 'does Samsung's robotics strategy convert into revenue?' Both are expensive, but only one is verifiable on a quarterly cadence.