Discovery › screen
Screen tech crash hidden value
Date: 2026-07-29
Criteria: Korean tech crash cascading into US tech — is there real hidden value, or just a deflating parabola?
Framework: analysis_notes.md §0 (great+cheap = value / poor+cheap = trap), §1 health, §4 small-mid overlay
Executive Answer
Yes — but not where the crash is loudest, and not for the reason I initially hypothesised.
A crash manufactures value and value traps simultaneously. Triaging the wreckage produces three distinct buckets, and only one of them is actually cheap:
| Bucket | What it is | Verdict |
|---|---|---|
| A. Peak-cycle memory | MU, SNDK, WDC, AMKR — 4-5x forward multiples | 🔴 Trap. The low multiple is the warning. |
| B. Quality dislocation | KLAC, ONTO — best assets, multiples compressed below their own history | 🟢 Real value. Already actioned on KLAC. |
| C. Wrong-cycle collateral damage | Analog / auto / industrial semis — sold with SOX despite being at the opposite end of their cycle | 🟢🟢 The actual hidden value. This is the answer. |
The core finding: the semiconductor industry is not one cycle, it's several. Memory/AI is at a peak. Analog/auto/industrial is at a trough. They are being sold together because they share an index. That is the dislocation.
⚠️ A Hypothesis I Tested and Had to Discard
I went in expecting the cleanest contrarian trade to be memory-price deflation benefiting memory consumers — server OEMs, storage vendors, device makers whose input costs would fall as the memory bubble burst. The watchlist already documented memory shortage as a tax on CVLT ("hardware supply constraints"), AMKR (Android −20% on memory supply), and IBM (clients diverting capex to supply-constrained memory/servers). Reversing that looked obvious.
The data killed it. Memory prices are still rising.
- DRAM contract prices +13-18% QoQ in Q3 2026; NAND +10-15% (decelerating, but still up)
- DRAM/LPDDR5x/NAND saw 90-95% QoQ price surges in Q1 2026
- SK Hynix reported HBM, DRAM and NAND capacity sold out through 2026; Micron's CEO guides tightness into 2027
- The price deceleration is being caused by OEMs being unable to absorb higher costs — i.e. it's demand destruction at the buyer, not supply relief
Two consequences that matter more than the discarded idea:
- The equity crash is a multiple de-rating running ahead of a fundamental turn, not a response to one. Physical memory is still tight and still pricing up. The equities fell on anticipated 2027 supply (CXMT's $8.6B raise, Kioxia/SanDisk capex +41%, YMTC), SK Hynix slowing its HBM4 transition, Meta selling AI compute capacity, and Burry's disclosed AMAT short. Near-term memory earnings will probably still be fine. The trap is the 2027 estimate, and the market is front-running it.
- Memory consumers are hurt, not helped. DELL, HPE, SMCI, NTAP are eating rising input costs they can't pass through. That's a margin headwind. Any thesis on them has to be built on something else.
🟢 THE SHORTLIST — Bucket C: Wrong-Cycle Collateral Damage
The setup: analog/auto/industrial semiconductors spent 2023-2025 in a brutal destocking cycle after pandemic over-ordering. That cycle is bottoming now, mid-2026:
- TXN explicitly signalled an inflection point on its last print — revenue and earnings beat, industrial recovering, automotive resilient, and critically lead times stabilising and channel inventory reducing (the textbook end-of-destocking signal)
- ADI ended FQ2 with channel inventory stable at 6-7 weeks, which management calls healthy — versus the bloat of two years ago
Meanwhile these names are −29% to −42% off their highs because they sit in the same SOX and semi ETFs as the memory complex. They have essentially zero AI-datacenter revenue exposure — the thing being repriced.
Ranked
| # | Company | Ticker | Sector | Mkt Cap | Price | %off High | fwd P/E | Div | Target | Upside |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Analog Devices | ADI | Semiconductors | ~$175B | $353.37 | −20.8% | 23.5x | 1.20% | $457.73 | +30% |
| 2 | NXP Semiconductors | NXPI | Semiconductors | ~$61B | $240.98 | −29.1% | 13.3x | 1.52% | $314.10 | +30% |
| 3 | Microchip | MCHP | Semiconductors | ~$39B | $71.37 | −32.6% | 17.0x | 2.34% | $113.08 | +58% |
| 4 | Texas Instruments | TXN | Semiconductors | ~$248B | $271.30 | −18.8% | 26.1x | 2.05% | $321.16 | +18% |
| — | ON Semiconductor | ON | Semiconductors | ~$33B | $78.86 | −41.6% | 18.2x | — | $113.12 | +43% |
Health Check (§1) — 4 fiscal years, newest first
| Metric | ADI | NXPI | MCHP | TXN | ON |
|---|---|---|---|---|---|
| Revenue trend | 11.02 ← 9.43 ← 12.31 ← 12.01B ✅ inflected | 12.27 ← 12.61 ← 13.28 ← 13.21B ⚠️ still falling | 4.71 ← 4.40 ← 7.63 ← 8.44B ✅ inflected off −44% | 17.68 ← 15.64 ← 17.52 ← 20.03B ✅ inflected | 6.00 ← 7.08 ← 8.25 ← 8.33B 🔴 still falling −15% |
| FCF trend | 4.28 ← 3.12 ← 3.56 ← 3.78B — 4yr HIGH | 2.28 ← 1.91 ← 2.51 ← 2.67B — recovering | 0.87 ← 0.77 ← 2.61 ← 3.13B — −72% from peak | 2.60 ← 1.50 ← 1.35 ← 5.92B — recovering | 1.42 ← 1.21 ← 0.44 ← 1.60B — recovering |
| Gross margin | 61.4% | 54.8% | 57.7% (was 67.5%) | 57.0% | 🔴 33.0% (was 49.0% — −16pts) |
| Debt / Assets | ✅ 18.0% | ⚠️ 46.0% | 38.6% | 40.6% | ✅ 24.0% |
| Equity / Assets | ✅ 70.5% | ⚠️ 37.9% | 44.7% | 47.0% | 61.3% |
| ROIC proxy (OpInc/InvCap) | 7.1% (goodwill-heavy, understated) | 13.6% | 4.4% (trough) | 20.3% | 7.1% |
| Shares outstanding | ✅ −5.1% (497M ← 523M) | ✅ −3.7% (254M ← 264M) | ~flat (545M ← 557M) | ~flat (913M ← 926M) | ✅ −8.1% (412M ← 448M) |
| Dividend covered by FCF? | ✅ Yes — $1.92B div + $2.16B buyback vs $4.28B FCF | ✅ Yes — $1.92B total vs $2.28B FCF | 🔴 No — $0.98B div vs $0.87B FCF | 🔴 No — $5.00B div vs $2.60B FCF | n/a |
Rationale per candidate
1. ADI — the quality pick. Best business in the group and it's already turning: FCF at a four-year high ($4.28B) while revenue inflects off the trough, 61.4% gross margin, and the cleanest balance sheet here (18% debt/assets, 70% equity). Shares down 5.1% over three years with dividend and buyback fully covered by FCF. The knock is honest: at 23.5x forward it's the least cheap of the four, so you're paying for quality rather than getting a dislocation. Target +30%. Earnings Aug 19.
2. NXPI — the cheapest quality. 13.3x forward for a 54.8% gross-margin auto/industrial franchise with 13.6% ROIC, a covered dividend, and a shrinking share count. Two real caveats: revenue has not inflected yet (12.61 → 12.27B, still declining), and debt/assets at 46% is the highest in the group with equity only 38% of assets. Trailing EPS $10.47 against a 13.3x forward multiple implies a large expected earnings ramp — that ramp is the thesis, and it isn't visible in the numbers yet. Target +30%. Earnings Oct 26 — a long wait for confirmation.
3. MCHP — the highest torque, the highest risk. This is the deepest trough in the group by a wide margin: revenue fell 44% ($8.44B → $4.40B), net income went to a loss, gross margin compressed 10 points, and ROIC sits at 4.4%. Trailing P/E of 324x is not a valuation — it's an earnings trough, and that's the point. Operating leverage that savaged them on the way down reverses on the way up, which is why the consensus target implies +58%, the highest here. But the dividend is not covered ($984M paid against $871M FCF) at exactly the wrong moment, and debt/assets is 38.6%. This is a cyclical recovery bet, not a compounder — size it accordingly. Earnings Aug 6.
4. TXN — best evidence, worst price. The strongest confirmation of the thesis (management called the inflection; lead times stabilising, channel inventory drawing down) plus the best returns profile in the group at 20.3% ROIC. But it's the least distressed (−18.8% off high), already up ~64% YTD, at 26.1x forward — and the $5.00B dividend is not covered by $2.60B FCF while the fab buildout runs (capex still −$4.55B, debt up to $14.05B). You're buying a recovery that's already partly priced. Watch, don't chase.
— ON: not yet. Screens cheap and shows a real −8.1% buyback, but revenue is still falling 15% YoY and gross margin has lost 16 points (49% → 33%). Its EV/SiC exposure is a structural demand question, not an inventory question. Fails §0's first test — I can't call this a good business right now. Revisit when revenue turns and GM stabilises above 35%.
🟢 Bucket B — Quality Dislocation (already actioned)
| Ticker | Price | %off High | fwd P/E | Target | Note |
|---|---|---|---|---|---|
| KLAC | $170.19 | −44.6% | 25.9x | $234 (+37%) | ✅ Already the top pick (watchlist-scan-2026-07-29). Beat, guided above consensus, fell 10.8%; forward multiple 37.5x → 25.9x, now below the 29.6x sector median. |
| ONTO | $218.31 | −43.5% | 22.0x | $369.60 (+69%) | 🆕 Worth noting. Process control / metrology — the same toll-booth economics as KLA, strong_buy consensus, largest implied upside in the complex. ⚠️ But more memory-levered than KLA, so it carries the Bucket-A risk KLA partly avoids. Needs an /analyze before any capital. |
🔴 Bucket A — The Traps (name them so they're not mistaken for value)
The low multiple is the warning, not the opportunity. These forward P/Es are computed on peak-cycle earnings that are real today (capacity sold out through 2026) but face a 2027 supply wall.
| Ticker | Price | fwd P/E | Why it's a trap |
|---|---|---|---|
| MU | $739.00 | 4.8x | Held in Beta at −$2,146. 52wk path $103 → ~$1,255 → $739. The 4.8x is 2026 earnings; CXMT/Kioxia/YMTC supply lands in 2027. Top sell candidate — and being a loss, it's tax-cheap to exit. |
| SNDK | $1,096 | 5.1x | [3.5] AVOID. GM 22.5% → 78.3% in four quarters on pure NAND price. Honest FV range $250–$2,200. |
| WDC | $462.04 | 24.6x | strong_buy consensus, target +42% — but it's the same NAND cycle in a different wrapper. |
| AMKR | $45.69 | 16.0x | [4.0] Value trap. 16.8% GM, FCF guided −$1.0 to −$1.6B, Graham IV $27.10. |
| FORM · ACLS · KLIC · TER | — | 20-30x | ⚠️ Look like Bucket C but aren't. Probe cards (FORM), ion implant (ACLS), packaging (KLIC) and HBM test (TER) are all memory-levered equipment — −38% to −48% off highs for the same reason as Bucket A. Don't buy these as "cheap semis." |
Also checked and rejected: SMCI ($25.70, 7.8x forward, −58.8% off high) is the cheapest AI-adjacent name on the board, but rising memory input costs are compressing it directly and the accounting history remains unresolved. HPE (11.1x, target +45%) and DELL (16.9x, target +36%) are cheap but face the same input-cost squeeze. MPWR at 40.8x and LSCC at 48.6x are not in this conversation.
Portfolio Fit — Why This Screen Matters More Than Usual
Current semiconductor exposure across both accounts is entirely one cycle.
| Account | Semi holdings | Cycle |
|---|---|---|
| Self | AVGO, INTC, QCOM, SNPS (AMD sold 7/28) | AI capex + handsets |
| Beta | INTC (15.7% of account), AVGO (8.6%), MU, AMD | AI capex + memory |
There is zero analog / auto / industrial semiconductor exposure in either account. Every semi position the portfolio owns is levered to the cycle that is currently deflating.
This matters against two flags already on the books: - The Concentration Map names "AI capex" as a 13-name cluster to be sized once, not thirteen times - Beta is 78% technology, with QQQ + INTC alone at 38% of the account
A portfolio-specific passage was removed from the public build.
Recommendations
Add to Watchlist — 🔧 Re-Rating Plays sleeve (these are cycle-recovery bets with a datable thesis, not evergreen compounders):
| Ticker | Suggested entry | Conviction | Flag |
|---|---|---|---|
| ADI | Attractive $320-345; strong <$300 | [—] pending analyze | 📋 Quality anchor of the group |
| NXPI | Attractive at $241; add $215-230 | [—] pending analyze | ⭐ Cheapest quality — but wait for revenue to inflect |
| MCHP | Stage $65-72; strong <$60 | [—] pending analyze | ⚠️ Uncovered dividend — size small |
| TXN | $230-250 only | [—] | 📋 Honorable mention — best business, worst entry |
| ONTO | Analyze before any level | [—] | ⚠️ Memory-levered despite the KLA resemblance |
Do not add: ON (revenue still falling, GM −16pts), and none of Bucket A.
Next steps — pick any
/analyze NXPI— the cheapest quality name, and the one where the open question (has revenue inflected?) is answerable with one more data point. My recommendation for the first full analysis./analyze ADI— the quality pick; earnings Aug 19 makes it timely./analyze MCHP— highest upside, but the dividend-coverage question needs resolving before capital./compare ADI NXPI MCHP— probably the most efficient single move: one head-to-head resolves which of the three to own rather than three separate analyses./analyze ONTO— only if you want a second process-control name alongside KLAC.
---
PART II — Spiral Rings 4-6
Rings 0-3 above were run organically. Rings 4-6 apply the explicit spiral method: each ring is one causal step further from the epicentre, and rings ≥4 require a number, not a mechanism — if the only support is "it stands to reason," it doesn't make the cut.
Ring 4 — Second-Order Infrastructure
Test: was it bid up because of the epicentre's boom, one step removed? (power, grid, cooling, construction)
| Ticker | Price | %off High | fwd P/E | Target | Read |
|---|---|---|---|---|---|
| TLN | $316.19 | −29.9% | 10.9x | +48% | 🟢 IPP. Existing generation fleet earns regardless of AI PPAs. Cheapest here. |
| NRG | $124.23 | −34.6% | 11.0x | +59% | 🟢 Same structure. 1.39% yield. |
| VST | $142.81 | −35.0% | 13.6x | +56% | 🟢 strong_buy. Same structure. |
| ATKR | $71.43 | −20.8% | 11.8x | +15% | 🟡 Electrical conduit — mostly construction, not datacenter. The ring-2 structure repeated. But ⚠️ no trailing earnings and target only +15%. |
| EME | $672.48 | −29.4% | 20.4x | +47% | 🟡 Electrical/mechanical construction. Reasonable multiple, real backlog. |
| MOD | $178.04 | −44.9% | 15.7x | +91% | 🔴 Thermal mgmt. Biggest drawdown + biggest implied upside — but now AI-datacenter-levered. High beta to the thing breaking. |
| VRT | $223.04 | −41.3% | 25.3x | +69% | 🔴 Pure AI derivative (datacenter power/cooling). Higher beta than the semis. |
| POWL | $186.39 | −43.2% | 27.1x | +70% | 🔴 Switchgear, real datacenter revenue. Same trap. |
| GEV · PWR · FIX · AGX | — | −25 to −39% | 25-36x | — | 🔴 Too expensive after the fall. |
| DLR · EQIX | — | −9.6% / −10.7% | 64.2x / 52.2x | — | 🔴 Datacenter REITs barely fell and are expensive. Not value. |
Verdict: mostly traps. Most of ring 4 is a higher-beta expression of the same AI-capex bet already over-owned — buying VRT or MOD here is doubling down on the epicentre, not diversifying from it.
The one real find: independent power producers at 10.9-13.6x. TLN, NRG and VST were bid up on AI datacenter PPAs, but their existing generation fleets produce cash whether or not the next datacenter gets built. That's a genuine "narrative exposure > actual exposure" setup — the ring-2 structure repeating one ring out. ⚠️ Caveat with teeth: if AI power demand disappoints, the marginal PPA pricing that justified the re-rating goes with it. These are cheaper than they look only if you believe the base fleet economics stand alone.
Ring 5 — Capital-Flow Destination
Test: where did the money go (SCHD +4.7%/month) — and what there has not yet re-rated?
| Ticker | Price | %off High | fwd P/E | Yield | Target vs spot | Read |
|---|---|---|---|---|---|---|
| CI | $296.47 | −6.0% | 8.9x | 2.07% | +15% | 🟢 Cheapest large-cap healthcare on the board. |
| ELV | $375.98 | −13.8% | 12.7x | 1.78% | +19% | 🟢 Managed care, already through its own de-rating. |
| CVS | $105.92 | −4.3% | 12.6x | 2.43% | +6% | 🟡 strong_buy but only −4.3% off high. Late. |
| MCK | $888.56 | −11.1% | 17.6x | 0.42% | +6% | 🟡 Quality distributor, fairly priced. |
| COR | $318.61 | −15.6% | 16.1x | 0.75% | +12% | 🟡 Same. |
| HUM | $365.41 | −14.8% | 22.8x | 0.93% | −5% | 🔴 Target below spot. |
| CAG | $15.50 | −23.7% | 10.0x | 7.94% | −7% | 🔴 Trap signature. |
| GIS | $37.92 | −26.1% | 11.8x | 6.66% | −1% | 🔴 Trap signature. |
| KHC | $27.62 | −5.4% | 13.2x | 6.10% | −12% | 🔴 Trap signature. |
Verdict: mostly late, with one trap cluster and one genuine sub-ring.
The rotation's beneficiaries have already re-rated — most sit within 15% of their highs, which is the whole point of being the destination. And the names that screen cheapest are a textbook trap cluster: CAG, GIS and KHC all pair a 6-8% yield with no trailing earnings, a "hold" consensus, and an analyst target at or below spot. That is packaged food in structural volume decline paying you to wait for something that isn't coming — high yield as compensation, not dislocation. §0 calls this correctly: poor + cheap = trap.
The exception worth naming: managed care. CI at 8.9x forward and ELV at 12.7x are genuinely cheap for businesses that size. Healthcare went through its own de-rating in 2025-26 (utilization/MLR pressure, DOJ probes) — so this is the ring-2 structure a third time: a sector that already troughed, now receiving inflows. Also relevant to portfolio fit: healthcare exposure across both accounts is one 3-share ISRG position.
Ring 6 — Narrative-Only Correlation
Test: sold purely by association — geography, index membership, thematic ETF — with no business link.
This was my highest-expectation ring. It was falsified, and the reason is the most interesting finding in the whole screen.
What actually happened in Korea
| KOSPI circuit breakers fired two consecutive days (Jul 28 and 29) | a market-historic first |
| Two-session index decline | −18%+ |
| July on track for | worst calendar month in KOSPI history, >30% loss |
| Samsung + SK Hynix | each −13%+; ₩396T (~$270B) of combined value destroyed in two sessions |
| Trigger | SK Hynix posted the most profitable quarter in its history — and still missed consensus |
| Samsung + SK Hynix share of the KOSPI | ~60%, up from ~40% two years ago |
| KRX Bank Index, past month | +2.33% — the only positive KRX sector index, vs KOSPI −20.19% |
| Bank driver | first base rate hike in 3.5 years → expanded net interest income |
Why the ring came up empty
When 60% of an index is two companies, an "index crash" is a two-stock crash — there is no indiscriminate spillover to create a dislocation. The Korean market discriminated with precision: memory got destroyed, banks rallied on their own catalyst, and the price action was correctly targeted throughout.
The quotes confirm it — KB is only −10.5% off its high and SHG −7.2%. They never got swept up.
| Ticker | Price | %off High | fwd P/E | Yield | Read |
|---|---|---|---|---|---|
| KB | $112.40 | −10.5% | 8.1x | 2.71% | ❌ Never crashed. Rate-hike winner. Not a dislocation. |
| SHG | $68.47 | −7.2% | 7.3x | 2.68% | ❌ Same. |
| WF | $65.37 | −22.8% | 6.1x | 4.18% | 🟡 The one anomaly — banks rallied as a sector, so this is likely a 52wk-high timing artifact rather than a crash. Verify before believing. |
| KT | $18.35 | −25.3% | 4.6x | 4.38% | 🟡 Fell, but for a fundamental reason — the rate hike hurts bond-proxy telecoms. Not narrative. Korean telecoms are also perennially cheap (state influence, low growth) = trap risk. |
| PKX | $48.68 | −47.3% | 9.1x | 3.17% | 🔴 Steel — crushed by China oversupply, its own problem. Not narrative. |
| LPL | $2.79 | −52.1% | 26.2x | — | 🔴 LG Display — panels are semi-adjacent. Legitimately correlated. |
| CPNG | $15.16 | −55.5% | 52.4x | — | 🔴 Fell hard but 52x forward. Not value. |
| EWY | $144.21 | −34.7% | 14.3x ttm | 1.01% | 🔴 Buying the index means buying ~60% Samsung/SK Hynix — i.e. buying ring 0. |
Nothing here clears the bar. Every Korean name that fell, fell for a reason I can identify. The ones that didn't fall are the ones a panic would have mispriced — and there was no panic, there was a repricing.
The transferable lesson
Ring 6 pays only when the index is broad. Narrative correlation creates dislocations when a crash drags unrelated companies down through shared index or ETF membership. When the index is concentrated, the crash is correctly aimed and there is no collateral damage to buy. Check index concentration before spending effort on ring 6.
Spiral Summary — Where the Value Actually Was
| Ring | Distance | Yield |
|---|---|---|
| 0 · Epicentre (memory) | — | 🔴 Nothing. All trap. |
| 1 · Adjacent function (semi-cap) | 1 step | 🟢 KLAC — actioned |
| 2 · Same sector, different cycle (analog) | 2 steps | 🟢🟢 ADI · NXPI · MCHP — the answer |
| 3 · Counterparties (memory consumers) | 3 steps | ❌ Hypothesis falsified — costs rising, not falling |
| 4 · Second-order infra (power/grid) | 4 steps | 🟡 TLN · NRG · VST at 10.9-13.6x — mostly traps otherwise |
| 5 · Capital-flow destination (defensives) | 5 steps | 🟡 CI · ELV managed care — rest already re-rated or trap |
| 6 · Narrative-only (Korea) | 6 steps | ❌ Falsified — index too concentrated for collateral damage |
Payoff peaked at ring 2 and decayed monotonically outward. Two of the three outer rings produced falsifiable hypotheses that failed — which is the method working, not failing. Both failures are now documented so they don't get re-derived.
Bottom Line
- The hidden value is real, and it sits at ring 2. Not memory (peak-cycle), not the AI-capex names that fell on their own earnings — the analog/auto/industrial semis sold by association while sitting at the bottom of a different cycle that is visibly bottoming.
/compare ADI NXPI MCHPis the highest-value next step. ADI has the quality and the inflection, NXPI has the price, MCHP has the torque.- Two secondary finds worth a look: CI at 8.9x forward (ring 5 — and healthcare exposure across both accounts is currently one 3-share ISRG position), and TLN/NRG/VST at 10.9-13.6x (ring 4 — but only if base-fleet economics stand without AI PPAs).
- Three hypotheses tested and killed — memory-consumer deflation (costs are rising; SK Hynix sold out through 2026), most of ring 4 (higher-beta versions of the bet already over-owned), and Korean narrative correlation (60% index concentration meant the crash was correctly aimed). Documented so they aren't re-derived.
- The best single fact in the screen: SK Hynix posted the most profitable quarter in its history and still missed consensus, triggering back-to-back KOSPI circuit breakers. That is what a cycle top looks like — and it is the clearest argument yet for exiting MU rather than holding for a recovery.
- This remains the one screen where adding semiconductor exposure reduces concentration risk rather than compounding it.
Screen run 2026-07-29. Prices live. Health check via .mcp/fin.py (4 fiscal years, newest-first). Related: watchlist-scan-2026-07-29.md · compare-korea-memory-selloff-CVLT-SNDK-KLAC-AMKR-GLW-2026-07-28.md
Sources: TXN analog cycle inflection · TXN inventory strategy · Memory prices still rising Q3 2026 (TrendForce) · Memory price surge cooling on affordability limits · Memory/flash prices not coming down through 2027