Verdict: Thesis confirmed on both the demand side (structural memory tightness, not a normal cycle) and the equipment side (HBM physically requires different bonding/testing). The 2026-04-11 blind trace surfaced the right names — BE Semiconductor/BESI (routed through scanner ticker BESIY), ATEYY/KLIC/COHR strong, ONTO lone. Coverage is good on the cohort but has two specific equipment-side gaps (ATEYY and KLIC have no deep-dives). Actionability is mixed: the cohort is mid-cycle on the thesis but late on the tape — RSIs 60–70, +50–100% 3M across the board, BESIY/STX/KLIC within 3% of 52-week highs. Entry timing matters more than name selection now.

What we're asking

Blind test #6 from research/projects/thesis-ideas.md:65 — @aleabitoreddit flagged the $DRAM ETF and prior coverage put MU and STX in our watchlists at +468% and +66%-above-SMA200. The thesis-ideas questions were physics-first:

• HBM vs regular DRAM — what's physically different? (TSV stacking, interposer bonding)
• Who has HBM capacity? (SK hynix dominant, Samsung, Micron catching up)
• What EQUIPMENT is unique to HBM? (die thinning, TSV etching, micro-bump bonding)
• Who makes the HBM testing equipment?
• What's the yield situation? (HBM yield reportedly 30–50% vs 90%+ for regular DRAM)
• Who supplies the advanced packaging for HBM stacks? (AMKR, ASX — already in picks-and-shovels)

Equivalently: WHY did the memory complex explode (MU +468% 1Y at the time of the prompt), and whose equipment gets pulled along regardless of which memory maker wins.

What we already had before today

This is post-trace consolidation, same shape as the tungsten verdict (research/investigations/2026-05-25-tungsten-supply-chain-blind-test-verdict.md).

So the blind test isn't actually blind anymore. The right shape today is a verdict that consolidates trace + perspective + deep-dive + operator-side input and closes the backlog item with a state-of-coverage map.

The physics chain — what HBM actually needs differently

The thesis-ideas physics questions are answerable now. Compressing the chain:

The blind trace converged on layers 2 (BE Semiconductor/BESIY), 3 (ATEYY), 4 (KLIC), 5 (ONTO), 6 (COHR) — five of the eight layers. Layers 1 (Lam/DISCO) and 7 (AMKR/ASX) and 8 (MU/SK hynix) the trace didn't surface because of how the prompt was framed (it asked for equipment, not for the OSATs or memory makers themselves) — those names were either already in watchlists or now marked research-only in stonks/watchlists/_securities.json.

State-of-coverage by ticker

Pricing as of 2026-05-23 precompute (no fresh tape pulled for this investigation — cd stonks && deno task summary <watchlist> --json).

Equipment cohort (the blind trace finds)

Memory cohort (the demand-side)

Read on the cohort: The 2026-05-19 R14 staging snapshot had the cohort uniformly 7-day red just before the Samsung NSEU strike day 0 (2026-05-21) — interpreted then as profit-taking, not a demand-side break. Six days later (current data), the cohort is uniformly strong-up again on the 30D + 3M view. That's the structural tightness thesis playing out: the strike was a transient catalyst layered on top, but the contract-price acceleration is the actual driver, and the cohort recovered the strike-window weakness with the supercycle intact. The R14 entry note that called the de-rate "profit-taking + recycled bear-narrative cover" looks correct on the 7–10 day follow-through.

Verdict and reasoning

Demand-side thesis (WHY memory exploded): CONFIRMED. The 2026-05-19 memory-supercycle perspective documents the chain — TrendForce Q2 2026 contract-price acceleration (DRAM +58–63% QoQ, NAND +70–75% QoQ), IDC ~15-year-worst supply deficit, no new fab capacity in volume before late 2027/2028. The cohort is +50–100% on 3M with the supercycle thesis intact through the Samsung strike window. This isn't a normal cycle — it's a multi-year structural tightness that survives transient labor catalysts. Nothing in this investigation challenges that read.

Equipment-side thesis (WHAT's physically different): CONFIRMED. HBM manufacturing genuinely requires a different equipment stack — micro-bump / hybrid bonding (BE Semiconductor/BESIY), KGSD-grade memory test (ATEYY), advanced packaging beyond wire-bond (KLIC), stack metrology (ONTO), laser singulation (COHR). The 2026-04-11 blind trace got to the right physics layer and surfaced merchant-leader names without prompting on memory at all. The convergence with NuttyCLD's "Last Micrometer" (BSPDN cross-thesis) and @ParadisLabs' independent operator-side core-themes list is the second and third source.

Actionability: MIXED, with a clear shape.

• One name fully covered. BE Semiconductor has a deep-dive at high conviction, routed through scanner ticker BESIY (OTC proxy for AMS:BESI). Entry zone $260–280; current $315 is +14% above the top of that zone. Watchlist add / hold, do not chase.
• One name is the laggard. ATEYY is the only cohort member negative 30-day and -14.6% from 52w high, with RSI 46.6. If the trace's "HBM test equipment" reasoning holds and HBM volumes accelerate, ATEYY is the structural beneficiary that hasn't yet rallied with the cohort. No deep-dive yet — this is the highest-leverage coverage gap to close. Filed.
• Two names are running hard. KLIC (+51.6% 3M, -2.4% from ATH) and COHR (+51.7% 3M) — both confirmed as advanced-packaging beneficiaries by the trace; KLIC has no deep-dive (also the highest-leverage filing item alongside ATEYY); COHR is already in optical-supercycle so the HBM use case is a thesis-stack refinement, not a new name.
• One name is cross-thesis. ONTO is currently in optical-supply-chain for packaging metrology there; the HBM line is a parallel use case. Worth a one-line note in the optical-supercycle log that ONTO has cross-thesis exposure (memory + optical), but no new deep-dive needed yet.
• The memory makers themselves are extended. Per the memory-supercycle perspective: HOLD the structural read; do NOT chase the cohort. Late on the trade, mid-cycle on the thesis. The cohort is +50–100% 3M and most names are within 3–8% of 52w highs. Wait for a pullback that doesn't break trend.

Net. The system did good work on the blind side. The trace converged on the right names at the right layers; the memory-supercycle perspective independently nailed the demand-side mechanics; the two views meet at "BESIY/BE Semiconductor is the cleanest pure-play, ATEYY is the laggard, KLIC is the dark horse, memory makers are late." The two specific coverage gaps worth closing — both filed below — are ATEYY and KLIC deep-dives.

Coverage gaps / follow-ups

Two new tasks filed in TASKS-RESEARCH.md (deep-dives section); one optional cross-thesis log note.

1. ATEYY (Advantest) deep-dive — Layer-3 HBM test equipment. Highest-leverage filing because it's the cohort laggard with a clean structural thesis (KGSD test requirement on HBM3/HBM4 scales with HBM volumes, not memory bit volumes). Questions for the deep-dive: HBM-specific revenue split vs SoC/logic test; Samsung vs SK hynix vs Micron exposure mix; competitive position vs Teradyne (TER); whether the laggard tape is structural (ASP / mix concern) or just cohort rotation.
2. KLIC (Kulicke & Soffa) deep-dive — Layer-4 advanced packaging. Wire bonding leader pivoting into hybrid bonding adjacencies. Questions: revenue mix between legacy wire bonding (cyclical) and advanced packaging (HBM-driven); competitive position vs BE Semiconductor/BESIY on the hybrid-bonding side; OSAT customer concentration; whether the +51% 3M run already prices in the HBM ramp.
3. Optional: log entry in optical-supercycle/log.md — single-paragraph note that ONTO has cross-thesis exposure (memory stack metrology in HBM lines, not just optical packaging). Does not require its own event; surfaces it in the optical perspective's log so future scans don't miss the dual driver. Not filed as a [ ] task — flagged for whichever next scan touches optical-supercycle.

Also worth knowing but deliberately not filed as separate tasks:

• DISCO Corp (6146.T) — Japanese wafer-thinning leader, would be Layer-1 in the chain. deno task fetch 6146.T --period 30d works as of 2026-05-25, but this remains research-only until JP-listing coverage is authorized.
• Samsung Electronics (005930.KS) — Memory-supercycle perspective already lists it in the cohort context. deno task fetch 005930.KS --period 30d works as of 2026-05-25, but tracking it separately would duplicate the demand-side coverage we already have via HY9H.F/MU/STX/WDC/DRAM and the perspective's competitor-pricing watch.
• TER (Teradyne) — Competitor to ATEYY on the test-equipment side, would naturally land in the same coverage decision as the ATEYY deep-dive. Roll into that task rather than file separately.

What didn't get done (transparency)

• No fresh OHLC fetch. Prices are 2026-05-23 precompute (per the do not run broad scans constraint in the task prompt). The 30D/3M numbers are recent enough to make the actionability call; entry-timing precision needs fresh tape on whichever specific name a user wants to act on.
• No Q2 / FY filings read. ATEYY, KLIC, and BE Semiconductor all report on different fiscal calendars; their HBM-specific revenue disclosures aren't extracted here. That's the deep-dive work, deferred.
• No picks-and-shovels.json cross-check beyond what's in scope. The picks-and-shovels thesis flags AMKR/ASX for advanced packaging; not re-running that analysis in this verdict — it's already documented elsewhere.
• No Samsung NSEU strike postmortem. The strike was a transient catalyst layered on top of the supercycle (per the memory-supercycle perspective); the verdict here is on the structural thesis. A separate write-up on how the strike actually resolved vs the prior staging is a memory-supercycle log entry, not this investigation's job.

Receipts

Source backlog item:

• research/projects/thesis-ideas.md:65 — Priority 1 Blind Test #6 (HBM/Memory + Why Memory Stocks Exploded). Updated to RESOLVED with verdict pointer.

Cited and consolidated:

• research/projects/autonomous-research/traces/2026-04-11-hbm-memory-equipment-bottleneck.json — original blind trace (BE Semiconductor/BESI unanimous, ATEYY/KLIC/COHR strong, ONTO lone).
• research/perspectives/2026-05-19-memory-supercycle/README.md — demand-side thesis (TrendForce contract prices, IDC deficit, Samsung strike posture).
• research/perspectives/2026-05-19-memory-supercycle/log.md — R14 strike-window staging entry that called the de-rate correctly.
• research/deep-dives/2026-05-01-besi-deep-dive.md — pure-play deep-dive at high conviction.
• research/api-data/events/2026-05-22-ParadisLabs-core-themes-cpu-hbm-nand-burn-in-test-glass-core-substrates-.json — independent operator-side core-themes list (HBM/NAND, burn-in/test, glass-core substrates) that mirrors the equipment-side layers traced.
• stonks/watchlists/memory.json, stonks/watchlists/supply-chain-traces.json — current cohort coverage.
• published/research-notes/2026-05-18-memory-supply-shock-verify.md — spark research note feeding the memory-supercycle perspective.

Filed follow-ups (this investigation, in TASKS-RESEARCH.md):

• [ ] ATEYY (Advantest) deep-dive — Layer-3 HBM test equipment, cohort laggard.
• [ ] KLIC (Kulicke & Soffa) deep-dive — Layer-4 advanced packaging.

Commands run for this artifact:

# scaffold the investigation event
deno task beta:emit investigation \
  --topic "HBM/Memory Supercycle blind-test verdict — supply-chain trace consolidation" \
  --slug "hbm-memory-supercycle-blind-test-verdict" \
  --tickers "BESIY,ATEYY,KLIC,COHR,ONTO,MU,STX,WDC,HY9H.F" \
  --perspectives "memory-supercycle" \
  --watchlists "memory,supply-chain-traces" \
  --cites "2026-05-01-besi-deep-dive,2026-05-19-memory-supercycle-spark"

# project the event into ticker / perspective / source feeds
deno task research:feeds

# cohort price summaries used in the body
cd stonks && deno task summary supply-chain-traces --json
cd stonks && deno task summary memory --json
cd stonks && deno task summary optical-supply-chain --json