• Type: spark (subtype: new-perspective)

• event_id: 2026-05-19-memory-supercycle-spark

• Sources:

  • Research note — Memory Supply Shock: Samsung Strike Verify & Supercycle Read (2026-05-18)
  • Capture — Weekly Capture/Source Sweep (2026-05-18)
  • R10 scan bucket — memory-cohort scan: published/scan-result-buckets/2026-05-19-r10-stale-research-refresh.json

• Shift: Perspective stood up. The R9 weekly capture sweep flagged a memory cluster (MU/DRAM/SNDK/STX/WDC) plus a Samsung-strike headline; the R10 cycle verified the lead and routed it to SCANS to stand up the perspective the capture's frontmatter had already named but which did not yet exist.

  Thesis (paradigm cycle): memory is in a structural tightness cycle, not a normal trough-to-trough one. Independent of any strike, the evidence is: TrendForce Q2 2026 contract prices DRAM +58-63% / NAND +70-75% QoQ (after ~95% Q1 jumps); an IDC-sized supply deficit DRAM 4.9% / NAND 4.2% / HBM 5.1%, the worst since ~2011; HBM consuming ~23% of DRAM wafers; new fab capacity not in volume before late 2027/2028.

  The Samsung NSEU strike is a transient catalyst, logged separately. ~45,000 workers, 18-day general strike from 2026-05-21, government-mediated talks collapsed 2026-05-12; Samsung's April test-walkout cut memory fab output ~18% / foundry ~58%, and Samsung is pre-emptively scaling down wafer input. It is bullish for competitor pricing near term — but it is a bonus-formula labor dispute that could settle fast (union open to talks post-2026-06-07). A settlement kills the strike angle and does not touch the contract-price thesis. The strike does not carry the thesis; the cycle data does.

• Effect on thesis: Establishes the perspective at status: active, priority: high. Cohort = MU/SNDK/STX/WDC/DRAM/HY9H.F. Posture: HOLD the structural read, do NOT chase — the cohort is strong-up but uniformly 7-day red, and the Samsung-strike news produced no competitor-rotation pop into MU (it sold off with the KOSPI). Late on the trade, mid-cycle on the thesis.

• Per-ticker (initial reads):

  • MU → (anchor; most direct Samsung-strike competitor-pricing beneficiary; +63% 3M but -9.1% 7D — extended, no rotation pop)
  • SNDK → (NAND leader; monster-grade, +115% 3M, -7.2% 7D)
  • STX → (HDD/storage; +81% 3M, -2.9% 7D)
  • WDC → (NAND/HDD; +61% 3M, -5.4% 7D)
  • DRAM → (Roundhill fund; sentiment proxy, flagged "parabolic" in the memory watchlist note)
  • HY9H.F → (SK Hynix GDR; most direct strike beneficiary, already extended +90% 3M)

• Open questions:

  • Does the Samsung strike actually start 2026-05-21, and does post-2026-06-07 mediation settle it fast?
  • How much of the 18-day strike hits memory vs foundry lines? (April test: memory -18% / foundry -58%.)
  • Does MU outperform Samsung (005930.KS) during the strike window — the real test of the competitor-pricing mechanism?
  • Does the cohort's 7-day pullback resolve up or break trend?
  • Capacity rule: ACTIVE is approaching the editorial-playbook 8-active soft cap — review whether a decayed perspective should be resolved when the next one is added.

• Methodology note: This perspective is deliberately born with the transient catalyst (the strike) and the durable thesis (the contract-price cycle) logged as separate things. The capture sweep surfaced the strike headline; the research verification confirmed the strike but established that the thesis stands on cycle data independently. Keeping them separate means a fast strike settlement does not falsely "break" the perspective.

🔄 2026-06-25 full-scan tape check: CONFIRMING (textbook) — MU RSI 63 (+30% 30D, −7% off high), SNDK 65 (+43%, −3.6% off high), WDC +29%, STX +21%, DRAM ETF +28%, HY9H.F +30% — whole cohort strong-up near highs; the dispersion bear-case (commodity DRAM rolling while HBM holds) did NOT fire — DRAM is leading. Confirmed but extended/consensus (do-not-chase). (Data tables spliced fresh to the 2026-06-25 close; full cross-scan read in 2026-06-25-full-scan-market-brief §6.)

🔄 2026-06-29 first-principles validation + complex deep-dives: Ran a full first-principles-validation investigation (2026-06-29-memory-supercycle-first-principles-validation-hbm-dram-nand-tightness-as-durable) — VERDICT: validated with cyclical caveats, now triple-confirmed (our own tape: the complex +110–240%/3m strong-up while NVDA, the demand driver, is weak-down −7.8%/30d; our own financials: MU Q3 FY26 rev $41.46B > all of FY25, SNDK swung from a FY25 loss to +$3.6B net/qtr; Apollo macro: hyperscaler FCF collapsed $300B→$60B with capex ~75% of operating cash flow — the cash the Mag 7 burns is landing on the suppliers). Filed deep-dives on EWY (the cleanest, least-extended HBM-duopoly proxy — added to key_tickers), STX and WDC (the storage vector, less extended than the MU/SNDK parabolas), and MSFT (the demand anchor — broken tape but a hyperscaler memory buyer). Added SIMO (controllers) to key_tickers. Operative frame (user): bubble volatility (equity) and physical undersupply (buildout) are orthogonal — both true → high-volatility secular long. Formalized the #1 risk as a standing signal: 2026-06-29-memory-capex-glut-watch (dormant; fires when ≥2 of contract-price-deceleration / new-fab-volume / cohort-trend-break / hyperscaler-capex-cut co-confirm). Corroborating discovery surface: @jiahanjimliu "Memorable 3 > Mag 7" + CXL-not-an-HBM-threat, and the @quxiaoyin/@kimmonismus "US building DCs too slowly" threads (bullish tightness; China/Huawei mostly an NVDA-TAM risk near-term). Posture unchanged: HOLD / add on pullbacks, do-not-chase the parabola; EWY the least-chased entry.

🔄 2026-07-05 operating book: Ratified as PRIMARY 1 of the operating book (operating-book) — mature/manage mode. Breaks section added to README citing the capex-glut-watch signal as the standing exit tape; nearline-storage folds in as a layer (its capacity-discipline break counts here). Book cut applied in ACTIVE.json (15→6 active).