The next AI data-center bottleneck is moving down to optical interconnect. Tens of thousands of GPUs talking to each other require 800G and 1.6T optical modules, and Co-Packaged Optics (CPO) — placing the optical engine right next to the switch ASIC — is emerging as the architectural answer. Mapping the seven Korean listed names — OE Solutions, Optocore, Daehan Optical Communications, BWE, WooriRo, Lycom, Coset — only OE Solutions sits genuinely close to CPO via its ELSFP external laser source and in-house 100G EML laser chip. The other six are downstream beneficiaries or thematic plays. And six of seven are up +300% to +905% YTD with operating losses still in place. The price moved before the earnings did. The actionable read is: OE Solutions on the watchlist (with a wait for either pullback or 3Q ELSFP customer-sample confirmation), and the rest until overheating clears.
Samsung Electronics and SK hynix could turn the AI memory boom into KRW 90–120T of incremental tax capacity, KRW 30–40T of household cash income and a KRW 120–150T pension buffer versus 2024.
KOSPI 7,498 in confirmed Bull regime. RFHIC tops a rare 5-screener sweep; Wonik IPS and HD Korean Shipbuilding follow with earnings + flow confirmation.
Korea exported USD 11.43 billion of cosmetics in 2025, a record high. It is now one of the world's top three cosmetics exporters, and in some classifications it competes with the U.S. for the number-two position behind France. Yet Korea does not have a L'Oreal or Estee Lauder. Its edge is a fast experimentation ecosystem built on ODM manufacturers, demanding consumers, Olive Young, Korean content, digital distribution and skincare-first culture.
Korea has more than ten listed semiconductor substrate and PCB-adjacent companies, while the U.S. and Europe have very few large-scale commercial substrate manufacturers. The reason is not that the West cannot make substrates. It is that chip design, software, tools and materials were prioritized for 30 years while high-volume plating, lamination, etching and yield learning moved to Asia.
Haesung DS is not a clean HBM proxy. It is a three-axis re-rating candidate: automotive lead-frame recovery (≈75% of revenue), DDR5 memory package substrate utilization normalization (1H25 ~30% → 2026 ~70%), and the optional AI data-center heat-spreader second-source entry. 1Q26 printed revenue ₩188.7bn (+37.2% YoY) but OPM collapsed to 5.8% on raw-material pass-through lag. The IC package heat-spreader market — Shinko, Honeywell/Solstice, Jentech, I-Chiun ~85% concentrated, 2024 TAM ~US$567m, CAGR 9.7% — is showing capacity shortage as AI accelerator power crosses 700W+. Haesung DS is qualifying heat-slug samples in 2Q26 with 2H26 revenue potential. Base case: flat heat slug second-source, ₩5–20bn 2026 contribution rising to ₩20–50bn in 2027 — too small to move the P&L, but large enough to trigger a multiple re-classification from automotive lead-frame to AI thermal supply chain. Closing 86,000 won (₩1.46tn cap) prices in some of axis 3; chasing here vs. waiting for the 82,000–84,000 retest is the actionable trade-off.
Mirae Asset lifted its Samsung Electro-Mechanics target price from KRW 530,000 to KRW 1.3 million by applying a 37x multiple to 2028F EPS. The important question is not the target itself, but whether the market accepts a 2017-style MLCC shortage multiple for today's AI server MLCC and FC-BGA cycle.