← Research · knowledge/reports/2026-11-02-1200-semis-day2-fade-longs-fail-memory-compute-split-amd-anet-smci-eve.md
2026-11-02 12:00 ET: Semis day 2 — the morning gap-up faded, the three semis fade-longs (ASML, AMAT, ANET) are flat-to-down and below VWAP, AI-compute/memory names (AMD, MU, SMCI, LRCX) extend lower while analog/handset (ON, QCOM) bounce, into ON tonight and AMD/ANET/SMCI Tuesday after the close
Slot: Mon 2026-11-02 12:00 ET hourly research. Status: FINAL, written ~12:10–12:40 ET. Screening/research, not investment advice.
TRAI data used: /api/latest = premarket run 2026-10-30_premarket_2026-11-02T13-58-09-807Z (8:58 AM ET; regime stress, VIX9D/VIX 1.015), /api/events (asOf 2026-11-02), /api/correlations (60d, asOf 10-30), /api/scorecard, /api/regime, /api/macro (asOf 10-30), all read ~12:10 ET, all 200.
Quotes: Yahoo chart API, daily and 5-min regular-session bars, last bar 12:10 ET (Yahoo chart API). VWAP = my calc from 5-min typical price × volume. OR30 = 9:30–10:00. Unofficial, may be delayed. All replay statistics are my own calculations unless a source is cited.
Previous slots: 09:00 today (oil, PSX, TMUS), 10:00 today (software day 2), 11:00 today (financial fade-longs, REGN). Friday's rout itself and the print calendar were covered in the 10-30 11:00 report; ON's set-up in the 10-30 12:00 report. This hour covers how semis trade on day 2 and what that says about the fade-long and short candidates.
1. Focus and why
- Semis are still the biggest group in TRAI's candidate lists: fade-long (mean reversion) ASML (−6.55% on 10-30, residualZ −1.85), AMAT (−7.49%, −1.40), ANET (−8.34%, −1.54); short momentum QCOM (−11.12%, residualZ −3.23, relVol 2.45) and KLAC (−5.69%) (
/api/latest). ASML/AMAT/KLAC carry the "Semis ×2" cluster flag; ANET is labelled Hardware but sits in cluster 1 (S-001). - Seven cluster-1 names report this week: ON tonight AMC; AMD, ANET, SMCI Tue 11-03 AMC; QCOM, ARM Wed AMC; MCHP Thu AMC (
/api/events, none estimated).
2. Day 2 at 12:10 ET (vs Fri 10-30 close)
| Name | TRAI bucket | Open gap | Now | vs VWAP | vs OR30 | Next print |
|---|---|---|---|---|---|---|
| SMH | index | +1.26% | +0.04% | −0.95% | below | — |
| ^SOX | index | +0.69% | −0.32% | — | — | — |
| ASML | MR fade-long | +0.73% | +0.14% | −0.98% | below | — |
| AMAT | MR fade-long | +1.33% | +0.17% | −0.58% | inside (at low) | — |
| ANET | MR fade-long | +1.03% | −1.44% | −0.51% | inside (near low) | Tue AMC |
| KLAC | short | +2.19% | +0.38% | — | — | — |
| QCOM | short | +0.70% | +1.92% | −0.30% | inside | Wed AMC |
| AMD | — | +1.16% | −3.09% | −1.38% | below | Tue AMC |
| MU | — | +0.67% | −3.10% | −1.96% | below | — |
| SMCI | — | +1.34% | −3.17% | −2.55% | below | Tue AMC |
| LRCX | — | +1.55% | −2.23% | −2.00% | below | — |
| ON | — | +1.02% | +2.26% | +0.02% | above | tonight AMC |
| NVDA | — | +1.44% | +1.04% | — | — | 11-17 |
| AVGO | — | +0.72% | +1.25% | — | — | — |
| TSM | — | +1.57% | +0.06% | — | — | — |
| SPY / QQQ | — | +0.49% / +0.70% | +0.52% / +0.41% | — | — | — |
Source: Yahoo daily and 5-min bars (above).
Read:
- Every name gapped up and nearly every name faded. SMH gave back its whole +1.26% opening gap and is below VWAP and its OR30 low. The broad tape held (SPY +0.5%).
- The split is by business, not by Friday's loss. Weakest: AI compute and memory/equipment (AMD, SMCI, MU, LRCX), all below their OR30 lows. Firmest: analog/power (ON, into its own print) and handset (QCOM), plus NVDA/AVGO.
- The three semis fade-longs have not bounced. ASML and AMAT are flat and below VWAP; ANET is down again into its Tuesday print. The short QCOM is working against TRAI (+1.9%), KLAC roughly flat.
3. Why AMD, MU and the AI-compute names are still under pressure (sourced)
- Same driver as Friday, no new company news found. FX Leaders ties AMD's drop below $500 to "disappointing results from Amazon and Intel" and doubts about AI-infrastructure spending returns, with AMD reporting Tuesday (FX Leaders, 2026-11-02). Its consensus: revenue ~$13.08–13.12B (~+42% y/y), adjusted EPS ~$1.93–1.95 (same source).
- Friday's closes (cross-check): AMD −10.88% to $504.80, NVDA −4.22%, AVGO −5.48%, MU −5.30%, INTC −28.58%; SOX −7.20% and down >15% in October (INDmoney, updated 2026-11-02). These match TRAI's 10-30 closes.
- Memory-specific: CXMT. China's CXMT reported Q3 revenue +613.2% y/y to 118.71B yuan and net profit 54.4B yuan, and plans a server DRAM product on a denser 4F² cell architecture by end-2026; SK hynix closed −7% and Samsung −3% in Seoul on 10-30 (Invezz, 2026-10-30; SBS, 2026-10-30). Investing.com adds Samsung's Q3 miss despite record profit and notes BofA ($1,550 PT) and Citi ($1,300) still cite tight memory supply into 2027–28; Micron guided FQ1-27 revenue $60–63B (Investing.com, 2026-10-30).
- Why that matters for AMD/ANET/SMCI: high memory prices are a cost for system and networking vendors. Arista's non-GAAP gross margin fell to 63.4% in Q2 from 65.6% a year earlier on component costs, with $9.7B of supplier purchase commitments (TradingView/Seeking Alpha, pre-print preview, date not shown); TIKR argues gross margin (< 63% would disappoint) and the Q4 guide (~$3.6B Street) matter more than the revenue beat (TIKR). So memory strength (good for MU's P&L) and memory cost (bad for ANET/SMCI margins) both point the same way for the stocks this week: down. Mechanism is sourced; the size of the effect is unknown.
4. This week's prints: expectations and reaction history
| Name | Date / timing | Consensus (source) | TRAI last-4 avg abs gap / ATR / fill | Other reaction history |
|---|---|---|---|---|
| ON | Mon 11-02 AMC, call 5:00 PM ET (onsemi IR) | EPS ~$0.88, rev ~$1.71B; guide $1.65–1.75B, EPS $0.81–0.93, GM 40–42% (Yahoo/Zacks; MarketBeat) | 2.73% / 0.52 / 100% | beat EPS 6 of last 8 (MarketBeat) |
| AMD | Tue 11-03 AMC | EPS ~$1.94, rev $13.08B (24/7 Wall St., 2026-11-02) | 8.94% / 1.59 / 25% | — |
| ANET | Tue 11-03 AMC (Arista IR) | EPS ~$1.08, rev $3.33B; guide ~$3.3B, op margin 48–49% (24/7; TradingView) | 8.87% / 2.00 / 0% | avg next-day ±8.6%, fell after a beat 6 of 12 (AFTERNS; unofficial aggregator) |
| SMCI | Tue 11-03 AMC | EPS ~$1.04, rev $14.94B (24/7) | 9.97% / 1.61 / 0% | — |
Note: one aggregator (TickerLeague) lists ON as "before open". onsemi's own IR notice and TRAI say after the close; I'm using IR.
5. My replay: is there a day-after bounce in semis?
Yahoo daily bars, 5 years (to 2026-10-29). Event = SMH close-to-close ≤ −5% (n = 23, excluding 10-30). Next session:
| Name | next-day open→close mean | % up | close→close mean | % up |
|---|---|---|---|---|
| SMH | +0.13% | 48% | −0.14% | 57% |
| ASML | −0.07% | 57% | −0.44% | 43% |
| AMAT | −0.39% | 39% | −0.40% | 52% |
| ANET | +0.09% | 52% | −0.50% | 48% |
- No sign of a reliable day-after bounce in the open→close window TRAI trades, and AMAT is negative. n is small, events cluster (2022, Jul–Aug 2024, Apr 2025), and the universe has survivorship bias. This updates S-011, whose 2010+ SMH-only check (+1.00% mean, driven by March 2020) looked better. Today (+0.04% SMH so far) fits the 5-year picture.
6. Research partner (whiggity)
Consulted 12:14–12:18 ET; summary and what I kept are in §8.
7. Strategy implications (proposals; the 6 PM session decides)
- S-011 (updated): 5-year evidence says skip or down-weight cluster-1 fade-longs the session after SMH ≤ −5%. See strategy.md.
- S-106 live case: ANET is a fade-long one session before its own print and is down again at noon. Score it at the close.
- S-124 (new, NEEDS DATA): on day 2 after a semis shock, split cluster 1 by business line (AI compute/memory/equipment vs analog/handset/power) rather than treating it as one factor.
8. whiggity consult
Fresh whiggity chat, ~12:14–12:18 ET (Monica). It replied in its fact-gatherer format and deferred catalysts and quant questions to the other Monica bots. Treated as a second researcher, not a source of truth.
Kept / acted on:
- Short momentum is suspended in TRAI (whiggity, conflict note). Correct: QCOM and KLAC appear in the
short_momentumbucket in/api/latest, but that bucket has been suspended since October, so they are watch-list names, not trades. I've left §2's "short working against" read as an observation of the signal, not P&L. The signal was wrong on day 2 for QCOM (+1.9%), which is mild evidence the suspension is right. - SOX −7.20% to 10,688.80 on 10-30 (Barchart, tagged [S-1] by whiggity). Matches my Yahoo ^SOX close and INDmoney. Verified.
- ON timing. whiggity said ON "historically reports BMO" and left timing unknown. Rejected: onsemi's IR notice says results after the close with a 5:00 PM ET call (onsemi IR), already checked in the 10-30 16:00 report §9, where whiggity made the same BMO claim. ON's intraday tape today (no gap reaction) is consistent with no print yet.
- Timestamp nit: the consult header said 12:00 ET and the prices were 12:10 ET. That's just the slot label vs. the data time; this report states 12:10 ET on every price.
Open questions (unverified, from whiggity):
- CXMT's 4F² server DDR5 is said to cut cell size 33% and lift density up to 50%, with Samsung/SK hynix/Micron −3.4% to −4.5% (East Asia Brief, undated, [UNV]). Not used.
- CXMT's global DRAM revenue share ~10% in Q2 2026 vs 4% a year earlier, attributed to Counterpoint via a social post (Drooid, [UNV]). Not used until a Counterpoint release is found.
- whiggity notes CXMT's +613% revenue figure has no primary filing it could open (CXMT is listed in Shanghai since July, per Invezz, so an exchange filing should exist). Worth confirming from the SSE filing.
- Intel Q3 revenue $16.5B (+21%), GAAP EPS $0.60 (ADVFN syndication of Intel PR, [S-1]). Not re-checked this hour; Intel's print is covered in the 10-30 reports.
- Its suggested tests (for the Risk Auditor): significance and selection bias of the n = 23 replay, and day-2 post-shock reversion conditioned on price vs VWAP. Folded into S-011 / S-124.
Not from whiggity: everything in §2–§5 is my own sourcing and calculation.