We independently reproduce the 2030 HBM claim of 26.7EB demand versus 10.6EB supply (a 2.5x shortage) from the published formula, then cross-check it against Goldman's 24x token forecast, DeepSeek MLA and TurboQuant efficiency counter-evidence, and the Big 3 capacity schedule. The conclusion separates direction from magnitude: tight through 2027 and supply relief from 2028 are supported by data, but a precise 2.5x shortage in 2030 is a bull scenario requiring many demand assumptions to hold at once. This is supply-demand structure analysis, not investment advice.
A July 10, 2026 study of 320 non-semiconductor companies from the top 200 KOSPI and KOSDAQ market-cap universes. KOSPI leadership spread into grids, construction, retail, financials and refining, while KOSDAQ remained an extreme winner-takes-most market with a -12.4% median YTD return.
Reverse-engineering the 2027 EPS consensus dispersion for Samsung, SK Hynix, Micron and SanDisk shows the two Korean names trade at 11.7x the most bearish street estimate, not a discount to the average. The market has already adopted the worst case as its base case. The real question is whether that worst case gets rejected on paper within three weeks.
SK Hynix surged 5.3% on institutional buying and ADR launch news as TrendForce projects Q3 2026 server DRAM prices rising 13–18%. Korean market breadth tells a subtler story.
A structured look at why US and Chinese AI inference infrastructure is diverging, and how power, HBM, SRAM/LPU, packaging, and Korean listed companies fit into the map.