Japan Earthquake

The map highlights each case's selected facility IDs and draws its illustrative radius; it calculates neither seismic intensity nor production losses. Supplier shares and each Herfindahl remain independent of the selected case.

Each case has a chosen magnitude, epicentre, map radius, severity and list of locations. HERP probabilities provide separately dated context for the broader earthquake classes; they do not assign probabilities to these exact combinations of settings.

The authored site list selects the highlighted locations independently of the circle. The chart draws a geodesic circle from the stored epicentre and radius; the optional navigable map applies pixel-size limits at different zoom levels. Ground motion, tsunami inundation and damage are outside it. The inventory includes equipment-company headquarters as well as production sites; a company’s headquarters cannot locate all of its manufacturing capacity.

A Herfindahl adds a market’s squared supplier shares. One is printed here only where the shares come from a tier 1 or 2 source with a link a reader can open and a single date; 5 of the 6 markets have theirs withheld, each with its reason in the market’s record. The supplier shares stay as recorded. All 6 market-share records contain estimated values.

The Korean assessment uses DRAM shares from the concentration table. Its discussion of Japanese input shortages is a separate assessment of supply dependencies, made outside the cascade calculation; no inventory-depletion timescale is available. The cascade simulator’s Japanese-input preset follows recorded customer relationships from Japanese suppliers, including Shin-Etsu, SUMCO, Tokyo Electron and Advantest; those paths reach Samsung but not SK Hynix, so the preset covers only part of the Korean exposure assessed here and none of the earthquake probabilities or HBM-qualification detail.

Estimating lost output requires the share of production at each selected plant and the extent of the interruption there. Company market shares supply neither input. Equipment companies also require a distinction between the headquarters shown on the map and the factories and service operations that supply their customers.

Recovery dates refer to specific milestones. Sony’s end-July 2016 figure describes wafer inputs at full capacity; NXP’s March 2021 announcement records an initial restart. Neither date measures when customers received their delayed orders.

The GPU SKU shares in a location’s detail are inferred. HBM market share is a quarterly aggregate by supplier (Q2 2026, Counterpoint Research, Global DRAM and HBM market share); it is not a direct per-fab measurement of HBM bit output. Per-fab → GPU SKU shares apply the canonical share to a qualified-suppliers subset that is per-SKU, not per-generation.

HBM3 NVIDIA H100 draws from {SK Hynix, Micron} (Samsung HBM3 NVIDIA qualification fell through); HBM3 AMD MI300X draws from {SK Hynix, Samsung} (TrendForce 2024-03-13: Samsung HBM3 received AMD MI300 series certification by 1Q24; Micron is not in the MI300X HBM3 chain, and Micron’s MI3-series collaboration is on MI350X HBM3E); HBM3E SKUs (H200, B200, GB200) draw from {SK Hynix, Samsung, Micron} after Samsung NVIDIA-qualification in late Q3 2025. The assumption: each qualified supplier contributes HBM to a SKU in proportion to its merchant HBM share, not a measured per-SKU bit allocation, which would require per-SKU primary backing (Counterpoint, TrendForce or SemiAnalysis SKU tables) not yet ingested.

Authority tiers: HBM merchant shares are tier 2 (Counterpoint Research, Global DRAM and HBM market share, Q2 2026; SK investor disclosures and TrendForce as corroborating sources). The per-SKU projection itself is tier 3 (analyst inference from the canonical supplier shares). The projection applies Q2 2026 vendor revenue shares to supplier qualification sets recorded through Q3 2025. It assumes those sets remain applicable; subsequent qualification and product-mix changes are not modelled.

Sources: HERP long-term seismic hazard evaluation (jishin.go.jp), TrendForce quarterly DRAM/NAND/HBM market share reports, Omdia semiconductor equipment and silicon wafer trackers, USGS earthquake catalog (earthquake.usgs.gov), Micron 10-K FY2024, Kioxia IPO prospectus, Shin-Etsu Chemical Annual Report 2024, SUMCO Annual Report 2024, Tokyo Electron Annual Report 2024, Advantest Annual Report 2024. All data is open-source or from publicly filed corporate disclosures. Market share estimates are authority tier 2-3. Seismic probabilities are authority tier 1 (HERP official assessments).

No completed review is recorded for the full analysis. HBM revenue shares were checked against Counterpoint’s Q2 2026 table in September 2026; the DRAM figures were aligned with the concentration table without a new source check. The newest input comes from the Q2 2026 HBM revenue shares (Counterpoint); the DRAM share figures, which include CXMT, are two quarters older, at Q4 2025. Review state for every scenario is on the Scenarios page.