A relative-value comparison of SK Hynix and Micron across HBM leadership, U.S. AI-memory scarcity premium, Vera Rubin HBM4, SOCAMM2 and Gen6 SSD narratives.
A relative-value note on why SK Hynix's forward P/E has moved above Samsung Electronics, and why Samsung's HBM4E catch-up may now be the cleaner alpha.
Goldman Sachs sees AI agents lifting token usage 24x by 2030 while cost per token falls 60-70% per year. J.P. Morgan sees the year-over-year gains in DRAM and NAND ASP rolling over from 2027. Split the two seemingly opposed calls into P, Q and C, and one coherent path emerges: alpha rotating out of 2026 memory beta and into a post-2027 token-cost-reduction stack.
PharmaResearch's multiple compression reflects the 4Q25 miss, a pause in medical-device exports, domestic skin-booster competition and skepticism toward cosmetics durability. This note tests whether Rejuran Cosmetics' U.S. Sephora and Olive Young sell-through can reopen the re-rating path.
GPU, HBM, CPU, MLCC and FC-BGA all sit inside the same AI infrastructure cycle, but they do not deserve the same multiple. This note decomposes the spread through pricing power, LTAs, customer lock-in, capex burden and peak-earnings doubt, then links Samsung Electronics and Samsung Electro-Mechanics in one relative-value frame.
The posts on Korea Invest Insights aren't written one by one from scratch — they're produced through a structure called the Thesis Investment OS. Alpha gathers the evidence, Lattice turns that evidence into judgment, and Arki keeps the whole system healthy. Together these three roles form an open-source research operating system. This post explains that structure in plain terms and invites you to visit the GitHub repository.
On May 29, 2026 the KOSPI printed a fresh all-time high at 8,476pt, yet the same day's ADR sank to roughly 52% — the lowest since March 2020, a six-year trough. The index made a record high while nine out of ten stocks fell. Here is a single, clear read on whether the Samsung–SK Hynix concentration tightens further or whether the neglected names are setting up to rebound.
China's Shanghai Futures Exchange is early-designing AI-token-price-linked futures, while CME and ICE are preparing GPU compute futures. Once AI usage gets a market price, the industry shifts from a performance race to a cost-per-token race. The conclusion: don't buy generic AI stocks — buy the bottleneck parts, chips and platforms that lower the customer's cost per token. The Korea read-through runs Samsung Electronics (HBM, eSSD, KV-cache) and Samsung Electro-Mechanics (FC-BGA, MLCC, silicon capacitor).
Dell's FY2027 Q1 print was a massive earnings surprise — revenue +88%, EPS +214% — yet its AI-server gross margin fell from 21.6% to 18.1%. The takeaway: pricing power and the durable margin do not sit with the server assembler (Dell); they sit upstream with memory and high-value component makers. The Korea read-through runs Samsung Electronics (HBM, server DRAM, eSSD) and Samsung Electro-Mechanics (FC-BGA, high-end MLCC, silicon capacitor).