Scanner symbol: ATEYY (OTC proxy, USD, otc_broker_dependent). Primary listing: 6857.T (Tokyo JPX, JPY, foreign_broker_required). Family registry: the Advantest security-family group.

Thesis (one line): Advantest is the dominant supplier of HBM stack test (KGSD — known-good-stacked-die — testing) at HBM3/HBM3E/HBM4 bandwidths; it sits at Layer-3 of the 8-layer HBM equipment chain documented in 2026-05-25-hbm-memory-supercycle-blind-test-verdict.md. Cohort peers (BESIY +37.6% 3M, KLIC +51.6% 3M, COHR +51.7% 3M) all ran with the HBM ramp; ATEYY didn't (+2.7% 3M). If the KGSD-test thesis holds — and the physics says it does — ATEYY is the structural laggard with the cleanest "the rest of the cohort already moved, this one hasn't" setup. Conviction medium because (a) the tape laggard isn't yet showing constructive structure (RSI 46.6, still rolling), and (b) HBM-test ASP/mix questions aren't yet primary-source verified.

Setup

Prices as of the 2026-05-23 scan (no fresh tape pulled per the producer-only constraint; entry/stop levels are structural, not technical-print specific).

• Last: $169.43
• 52w high: ~$198 (implied from -14.6%)
• 30D: -2.9% — only cohort member negative on 30D
• 3M: +2.7% — vs cohort +37–52%
• RSI: 46.6 — neither oversold nor breaking; flat
• Watchlists: supply-chain-traces (HBM trace 2026-04-11 — 0.67 convergence)

Entry zone: $155–170 — current $169 is at the top of the zone. The reason for a structural zone vs a technical-print target is that ATEYY is mid-cohort-rotation: a clean entry print needs either (a) RSI recovery above 55 with rising 20D (the cohort-rotation tell), or (b) a fresh HBM-test ASP datapoint in an Advantest quarterly or a Samsung/SK hynix capex disclosure. Stop: $140 — closes below would signal the cohort-laggard thesis was wrong because either HBM-test isn't ASP-positive (Teradyne stealing share) or HBM volumes are slipping vs forecast. Target: $240–260 (12-month) — if ATEYY does the cohort-laggard catch-up trade analogous to HY9H.F (which ran +93% 3M as the rotation-laggard memory name once it caught a bid), $240–260 puts the multiple back in line with where BESIY/KLIC sit relative to their 52w highs.

Watch for the ADR-vs-primary spread question separately — ATEYY OTC vs 6857.T JPY-translated. The deep-dive task's question (e) is whether ADR float depth is adequate. As of 2026-05-23, ATEYY US OTC pricing tracks 6857.T within a few percent on a 30D window after currency translation; not a blocker for sizing through ATEYY.

Bull case

1. HBM test equipment is the cohort laggard on a structurally accelerating end-market. HBM unit shipments are growing far faster than DRAM bit volume because each HBM stack is 8–12 die stacked vs 1 die for commodity DRAM. ATEYY's HBM-test ASP scales with stack volume (every die tested, every stack tested, KGSD requirement before assembly). The five-name cohort all converged on the same thesis in the 2026-04-11 trace; four of five have already moved.

2. KGSD test is structural, not cyclical. The "known-good-stacked-die" requirement is physics, not market preference. You cannot rework a stacked HBM die after bonding — if one die in the stack has a latent defect, the entire stack is scrapped. As stack heights climb (HBM3 8-Hi → HBM3E 12-Hi → HBM4 16-Hi), the cost of a failed stack scales superlinearly with capacity; pre-stack and post-stack test become indispensable. ATEYY dominates this segment of memory test (~50%+ share by various trade-press estimates), with Teradyne (TER) primarily in SoC/logic test.

3. Customer concentration is the right shape. ATEYY's three largest memory customers are SK hynix, Samsung, and Micron — exactly the three names with documented HBM capacity. The memory-supercycle perspective documents demand-side acceleration on all three (TrendForce Q2 contract prices +58–63% QoQ; 15-year-worst supply deficit). The same demand wave that lifted MU/HY9H.F/STX +50–100% 3M should pull through to memory test, with a quarter-lag on equipment-cycle timing.

4. No deep-dive done elsewhere in our system. ATEYY surfaced in the 2026-04-11 blind trace at 0.67 convergence ("HBM test equipment — every HBM stack needs testing at multiple stages") and was filed straight into supply-chain-traces. No conviction-doc, no in-house analysis, no operator-side confirmation specific to Advantest. This is genuine coverage gap, not a stale-replay of a known name.

5. Cohort rotation reads as buyable, not breaking. The other four trace names (BESIY/KLIC/COHR/ONTO) all show the same shape — RSI 60–70, +37–52% 3M, near ATHs. ATEYY is flat through the entire cohort move, not down, which is the cleaner laggard pattern. (If it had broken down -20% while the cohort ran, the right read would be "thesis is wrong about this name." Flat tape during a cohort rally is the "next to rotate" tell.)

Bear case

1. The tape is the tape. Four cohort peers ran +37–52% over three months and ATEYY did +2.7%. The market sees something the structural thesis doesn't. Possible reads: (a) HBM-test ASP is less ASP-positive than expected because Samsung/SK hynix are pushing for lower per-stack test cost as stack heights rise (volume offset by ASP compression); (b) Teradyne is taking memory-test share at the margin via integrated SoC+memory platforms that win on a multi-die package basis; (c) ATEYY's exposure is more in SoC test (gold-finger of the Apple A-series, Snapdragon, and NVDA test programs) than in memory test, and SoC test is decelerating as discrete logic shipments soften. Without HBM-segment revenue disclosure to bound this, the bear case is plausible.

2. HBM yield improvement reduces test intensity. If SK hynix and Samsung crack the HBM yield problem (current ~50–70% per various trade estimates) and push closer to logic-DRAM-like yields, the number of test-and-scrap cycles per shipping stack drops. That's net-net negative for memory-test equipment volumes, even if HBM unit shipments grow. The structural ASP thesis assumes yield stays in the 50–70% band; a yield breakout invalidates it.

3. ADR mechanics + JP-mover-share-class drag. ATEYY trades OTC USD vs the 6857.T JPX primary; currency translation and ADR creation/redemption can drag the OTC line behind the JP primary on tape-day moves. Not a thesis problem — but a sizing problem if you're trying to time a print.

4. Mid-cycle on positioning. Even if the thesis is right, the entry geometry isn't here yet. The cohort-laggard catch-up trade typically needs (a) cohort to stop making new highs on the leg-leaders (a pause, not a break), and (b) the laggard to show the first sign of constructive RSI/20D structure. Neither is in evidence as of 2026-05-23. The right buy is a watchlist add today, sized scale-in on the structural signal — not a fat entry at the top of the structural zone.

5. Comparative — Teradyne is the obvious "cleaner US-listed alternative." TER has a direct US listing without OTC mechanics and sits in the ai-infrastructure watchlist — not ai-power-delivery (corrected 2026-06-05 per 2026-05-29-ter-ateyy-comparative.md; ai-power-delivery carries no key_tickers list at all). If a user wants memory + SoC test exposure with cleaner trade access, TER is the default. The case for ATEYY over TER specifically is: ATEYY is more HBM-pure on the test side; TER is more SoC-pure with memory mix. If you want HBM-test laggard, you want ATEYY. If you want diversified test exposure, you want TER.

Catalysts

Competitive positioning

The cleanest narrative for ATEYY is "the test gate on the BESIY bonding line." Every HBM stack BESI hybrid-bonds gets ATEYY-tested (or Teradyne-tested) before it ships to AMKR/ASX for accelerator-package assembly. ATEYY's TAM scales with BESI's bonder-utilization curve, with a quarter-lag on equipment-cycle timing.

Correction (2026-06-05, from 2026-05-29-ter-ateyy-comparative.md): HBM-stack test is a contested market, not an Advantest near-monopoly. TER's memory tester (Magnum 7) is explicitly HBM+DRAM and trade press credits TER with >50% of HBM wafer-stack test in a current "win-back" phase; TER guides to low-single-digit memory-test share gains for 2026. This confirms bear-case #1c (Teradyne taking share — now from TER's own guidance, not speculation) and bounds bull-case #2 ("ATEYY dominates ~50%+"): Advantest's ~60–70% memory-test share is DRAM-weighted; on the narrow HBM-stack-test axis the thesis rests on, TER is arguably the share gainer. Also: both names are ~80% SoC / ~17–19% memory by revenue — neither is an "HBM-test pure-play"; ATEYY's memory segment is itself ~90% DRAM, with HBM the fast-growing slice inside it.

Cross-thesis exposure

ATEYY is not cross-thesis with optical-supercycle (unlike COHR/ONTO). It's a pure HBM-test name with secondary SoC-test exposure on Apple/Snapdragon/NVDA. The right cross-references are:

• memory-supercycle (downstream demand-side driver).
• supply-chain-traces (the watchlist housing).
• Eventual ai-power-bottleneck indirect — HBM4 scales with watts-per-rack rack-power; same physical chain.

Financials addendum — Advantest FY ended 31 Mar 2026 segment actuals (2026-06-05, closes TASKS-RESEARCH.md:81)

The original deep dive carried a financials_waiver (Massive has no ATEYY ADR financials) and flagged "no Advantest quarterly read." Advantest's full-year results are now public (fiscal year ended 31 March 2026; note Advantest labels this "FY2025" — the year starting April 2025 — so it equals what our notes call FY26). Segment actuals from Advantest IR (corroborated across the results summary + the FY25 slide recap):

The read — bearish for the memory-test-laggard thesis:

1. Memory test was the slowest-growing segment: +8.7%, vs SoC test +74.3% and total +44.7%. The "memory supercycle" demand the bull case leans on is not showing up in Advantest's memory-test revenue — the record year was overwhelmingly an SoC/AI-logic-test story.
2. Advantest's memory-tester share slipped to 61% (from 63%) as the memory-test market expanded to ~$2.1B. The incumbent's own disclosure now confirms the contested-market / Teradyne "win-back" finding — this hardens bear-case #1c a second time (first from TER's guidance, now from Advantest's filing).
3. ~90% of memory-test revenue is DRAM → HBM is the minority growth-slice inside memory, and HBM-specific revenue is not separately disclosed. The task's ideal "HBM revenue mix" datapoint does not exist in Advantest's reporting; Memory test (¥171.5B, ~15% of revenue, ~90% DRAM) is the best available proxy and the ceiling on HBM exposure.
4. FY-ending-Mar-2027 guide ("FY2026"): total ¥1,420B (+25.8%); memory test guided to ¥201.0B (+17%) — reaccelerates but still trails the company average. The HBM4 ramp is the swing factor and the first place to watch.

Net: the actuals push ATEYY's risk/reward toward the bear case — the memory-test segment the thesis rests on is its slowest grower and it is ceding share, even as the company prints record SoC-driven results. The cohort-laggard "catch-up" now explicitly depends on HBM-test ASP/volume inflecting in the FY-ending-Mar-2027 numbers (the ¥201B memory guide is the marker); until then, the laggard tape is consistent with laggard fundamentals. (Data: Advantest IR, FY ended 31 Mar 2026 results + FY25 slide recap — see Sources. Segment %s are simple composition ratios of the reported ¥ figures.)

Sources

• 2026-05-25-hbm-memory-supercycle-blind-test-verdict — Layer-3 placement, cohort-state map, cohort-laggard read.
• Advantest IR — FY ended 31 Mar 2026 results (2026-06-05 pull): total net sales ¥1,128.6B (+44.7%), SoC test ¥767.4B (+74.3%), memory test ¥171.5B (+8.7%, ~90% DRAM, memory-tester share 63%→61%), Services ¥109.2B, OI ¥499.1B (44.2%), NI ¥375.4B; FY-ending-Mar-2027 guide ¥1,420B / memory ¥201B. (Advantest IR results summary, Investing.com FY25 slide recap, MarketScreener FY-ended-Mar-2026).
• 2026-05-25-hbm-adr-foreign-primary-backfill-decision — ATEYY scanner routing, ADR-vs-primary policy.
• 2026-04-11-hbm-memory-equipment-bottleneck — original blind trace (0.67 convergence on ATEYY).
• README — demand-side TrendForce + IDC mechanics.
• _securities — security-family registry; ATEYY routing.
• supply-chain-traces (data_as_of 2026-05-23) — price/RSI/30D/3M used in setup and cohort-relative read.

What didn't get done (transparency)

• No fresh OHLC fetch. Prices are from the 2026-05-23 scan; the structural entry zone holds across the few-percent move that's likely between then and any user action.
• No Advantest 10-K / quarterly read. Done 2026-06-05 — FY-ended-Mar-2026 segment actuals folded (see Financials addendum). HBM-specific is not separately disclosed; Memory test ¥171.5B / ~90% DRAM is the proxy ceiling, and it grew only +8.7% (slowest segment).
• No ATEYY ADR float-depth fetch. The ADR-vs-primary spread question (deep-dive question (e)) is informally checked but not formally measured against 6857.T JPY-translated; deferred until the next focused fetch on Advantest.
• No Teradyne comparative. Treated as the "cleaner US alternative" in the bear case but no side-by-side comp. If user wants a head-to-head, that's a separate ~30-min comp note.

Filed follow-ups

Two narrow follow-ups, filed via this artifact (not yet in TASKS-RESEARCH.md — bundled into the closeout batch update at the end of this session):

1. ATEYY HBM-segment revenue extraction Done 2026-06-05 (TASKS-RESEARCH.md:81) — FY-ended-Mar-2026 actuals: Memory test ¥171.5B (+8.7%, the slowest segment vs SoC +74.3%), ~90% DRAM, Advantest memory-tester share slipped 63%→61%. HBM not separately disclosed. Folded into the Financials addendum; the read tilts bearish. Next watch: the FY-ending-Mar-2027 memory guide ¥201B + the HBM4 ramp.
2. Teradyne (TER) memory-test comparative note — ~30-min side-by-side on memory vs SoC mix, customer overlap, and HBM-test platform positioning. Decide whether TER deserves a sister deep-dive or stays as cohort context. Done 2026-05-29 (verdict: stays cohort context; home is memory-supercycle + supply-chain-traces, not ai-power-delivery; HBM-stack test is contested — see Correction callout above).

Commands run for this artifact

the desk's own tooling
  --ticker ATEYY \
  --topic "ATEYY (Advantest) deep-dive — Layer-3 HBM test equipment, cohort laggard" \
  --slug "ateyy-deep-dive-hbm-test-equipment" \
  --tickers "ATEYY,TER,KLIC,BESIY,MU,HY9H.F" \
  --perspectives "memory-supercycle" \
  --watchlists "supply-chain-traces,memory" \
  --cites "2026-05-25-hbm-memory-supercycle-blind-test-verdict,2026-05-25-hbm-adr-foreign-primary-backfill-decision" \
  --thesis "..." \
  --conviction medium