Discovery › screen
Screen us tech optical crypto and round synthesis
Date: 2026-07-29 · Command: /screen (round 3 — US-only tech) · Framework: analysis_notes.md §0–§4, §6
Origin: User rejected round-1 (defensive/income) names and asked to lean into software/hardware/tech/emerging tech on the hypothesis that "the recent massive downslide might help with surfacing value." Then asked whether America had been saturated, and why we were looking abroad.
Companion files from rounds 2–3: - screen-semi-test-ate-and-tech-sweep-synthesis-2026-07-29.md — ATE + round-2 synthesis - screen-robotics-automation-machine-vision-physical-ai-2026-07-29.md - screen-gaming-adtech-consumer-internet-2026-07-29.md - screen-japan-tech-semis-2026-07-29.md - screen-health-tech-software-2026-07-29.md - screen-semi-materials-consumables-subsystems-2026-07-29.md
PART A — Answering the user's question: was America saturated?
No — and the detour abroad was the weaker call.
The 7/29 screen-global-vs-us-value had already concluded: "The cheapest equity risk on the board is in America, and it always was" and "nothing abroad is cheap enough to pay for its own structural discount — except Japan." Round 2 chased that single exception. The Japan sweep then killed it: EWJ's −8.4% "drawdown" is 100% currency (TOPIX set a record close of 4,101.96 on July 6; USD/JPY ~163.5, yen −8.41% over 12 months — the same number). So the global screen's primary finding stands untouched and its lone exception is now closed. That cost one forager.
Verified US-tech coverage gaps (company-name matches across 47 prior screens, substring noise removed):
| Field | Prior mentions | Status |
|---|---|---|
| Optical / photonics | Lumentum 0, Viavi 0, Ciena 1 | Only coverage was two March-21 screens, pre-crash |
| Semi materials / consumables | Entegris 0, MKS 1 | Never screened as a field |
| Health tech / techbio | Doximity 0, Certara 0, Schrödinger 0 | Never screened |
| Crypto / retail fintech | Coinbase 0, Robinhood 0 | Never screened; user holds BTC directly |
| Digital infra REITs | Equinix 0, Digital Realty 0 | Still unswept (deferred) |
The library's ~47 screens cluster hard on enterprise software and the semi design/equipment layers, leaving the materials, optical, health-software and crypto layers untouched. Round 3 swept four of the five.
PART B — ⭐ The cross-round meta-finding
Every AI-levered field trades ABOVE its own historical multiple band. The only value found was in corners the AI trade never touched.
Nine tech fields swept across rounds 2–3. The pattern did not break once on the AI side:
| Field | AI-levered names vs own band | Evidence |
|---|---|---|
| Semi test / ATE | 11 of 11 above | TER 76% above its FY2025 high; P/B 17.97 vs band 3.67–11.45 |
| Optical / photonics | 10 of 13 above | LITE at 2.6x its own recent-high P/B, 3.2x its recent-high EV/Sales |
| Semi materials | Above EV/EBITDA medians throughout | MKSI above its entire 10-year range |
| Robotics | Quality names 32–35x | ROK 4.8% off high, mean target at spot |
| Japan tech | AI names carry no discount | Advantest 60x P/E, 28x P/B |
The mechanism: the drawdown is smaller than the run-up. Names are −15% to −60% off their highs and simultaneously +110% to +582% off their lows. "Down 40%" repeatedly meant "down 40% from three times its own ceiling."
Where value actually was found — all non-AI-levered:
| Name | Field | Why the AI trade never touched it |
|---|---|---|
| DOCS | Healthcare software | Crashed on a different narrative (agentic AI in healthcare); revenue is pharma marketing budgets, not CMS |
| ZBRA | Machine vision / warehouse | Mislabeled as a barcode-scanner company; 3PL and industrial end-markets |
| NTES | Chinese gaming | Cheap for geopolitical reasons, entirely decoupled from AI |
| HOOD | Retail fintech | Own cycle; grew +32% through a −50% BTC drawdown |
| HLIT / ADTN | Carrier/cable capex | ~0% AI exposure; depressed since 2022 |
Corollary worth keeping: the user's portfolio is already saturated with AI-capex correlation (AVGO, ANET, INTC, QCOM, ORCL, GOOGL, MSFT, META, QQQ×2, plus SMR/LEU). Buying into any AI-levered field now would be concentration at peak multiples, not diversification. The diversification constraint repeatedly disqualified the highest-quality names — Teradyne owns the best collaborative-robotics assets (UR/MiR) but its 104% growth is semi test riding the same wave; same for MKS.
PART C — The board after three rounds
| # | Ticker | Field | Price | Valuation | vs own band | Conv | Status |
|---|---|---|---|---|---|---|---|
| 1 | DOCS | Healthcare software | $21.24 | 11x EV/FCF, 13x fwd | at the FLOOR of its 5yr band | 7.0 | ✅ in zone (<$23) |
| 2 | ZBRA | Machine vision | $288.94 | ~15.6x adj fwd | vs own 3yr avg ~40x | 7.0 | ✅ −18% off high |
| 3 | NTES | Gaming (China) | — | 12.5x fwd | vs ~26x peers | 7.0 | ✅ great + cheap |
| 4 | HOOD | Retail fintech | $86.60 | 28.2x fwd | mid-band FY25, rich vs FY24 | 7.0 | ⚠️ entry <$75 |
| 5 | HLIT | Cable/broadband | $11.22 | 14.4x fwd, 7.9% FCF yield | mid-band | 5.5 | fair, not cheap |
| 6 | COIN | Crypto exchange | $163.58 | 38.8x fwd | bottom decile of own history | 5.0 | overlaps BTC |
| 7 | ADTN | Carrier access | $8.42 | 11.1x fwd | below own band | 5.0 | no moat, neg. ROIC |
| 8 | ENTG / Q | Semi materials | $117 / $131 | 24.9x / ~49x adj | mixed | 5.0 | zero growth |
DOCS is the only name sitting at the floor of its own historical band — the exact test every AI-levered field failed.
PART D — The Optical / Photonics Screen
D1. The field is not crashed. It is a bubble that gave back a third.
| Ticker | Price | Cap | Off HIGH | Off LOW | P/E ttm | P/E fwd | EV/EBITDA | P/B |
|---|---|---|---|---|---|---|---|---|
| LITE | 693.24 | 53.9B | −36% | +582% | 122 | 37.8 | 85.2 | 16.7 |
| CIEN | 372.07 | 52.7B | −42% | +341% | 124 | 38.6 | 60.0 | 18.2 |
| AAOI | 90.11 | 7.2B | −61% | +387% | neg | 18.9 | neg | 6.4 |
| VIAV | 35.05 | 8.7B | −42% | +264% | neg | 27.3 | 36.6 | 9.7 |
| ALAB | 299.69 | 51.4B | −40% | +206% | 204 | 65.7 | 178.5 | 34.4 |
| COHR | 249.06 | 48.7B | −43% | +195% | 118 | 30.0 | 34.1 | 4.6 |
| SMTC | 114.61 | 10.7B | −35% | +149% | neg | 29.2 | 65.9 | 18.6 |
| EXTR | 29.84 | 3.9B | −12% | +121% | 249 | 22.8 | 40.7 | 50.1 |
| MTSI | 248.71 | 19.0B | −41% | +110% | 105 | 35.9 | 73.2 | 13.4 |
| POET | 7.28 | 1.3B | −65% | +88% | neg | neg | neg | 2.5 |
| HLIT | 11.22 | 1.2B | −37% | +36% | 140 | 14.4 | 28.4 | 3.4 |
| CLFD | 30.89 | 420M | −41% | +30% | 134 | 21.3 | 66.3 | 1.7 |
| ADTN | 8.42 | 682M | −58% | +18% | neg | 11.1 | 13.1 | 4.9 |
The bottom three rows are a different asset class — the names that never participated.
vs own bands (roic.ai free tier caps at 2 years, so AI-name results are understated — the 2-year window is itself inside the bubble): - LITE — P/B 16.7 vs FY2025 band 2.36–6.40 (avg 4.30); EV/Sales 17.4 vs band 2.68–5.47. 2.6x its own recent-high P/B, 3.2x its recent-high EV/Sales. - CIEN — P/B 18.2 vs FY25 band 2.54–10.01 and an FY24 high of 3.54. 5.1x its FY24 ceiling on book. - HLIT — P/B 3.43 vs band 2.26–3.97; P/FCF 12.6 vs band 9.2–16.1. Mid-band, fairly priced. - ADTN — EV/EBITDA 13.1 vs its FY25 band of 13.5–19.3 — below its own two-year low. The only name in the field below its own band.
D2. The two demand streams, separated
Stream 1 — AI datacenter interconnect: booming and fully priced past the point of return. Coherent's datacenter/comms segment is now 75% of revenue (+41% YoY). Lumentum collapsed to a single reporting segment in FY2026 — the telecom/datacom split is no longer disclosed, which is itself a tell. ALAB, MTSI, SMTC, AAOI are ~pure AI plays.
Two reasons to distrust durability: LightCounting flags a 2H-2026 oversupply warning (800G+ units 24M → 63M), and the 400G→800G jump created a "toxic inventory environment" in mid-speed components. 800G pricing has already settled to $360–450/module; 1.6T is guided from $1,300–1,500 down to ~$1,100 within two years. This is a business with contractual annual price deflation — current 30–41% gross margins are peak-cycle prints on a deflating product, not moat margins.
Stream 2 — telecom/carrier and cable: genuinely depressed, and the only non-AI-correlated candidates. But note the sequencing: BEAD money only reaches operators materially from late 2026; cable DAA spending jumped ~40% in Q1'26 on first-year DOCSIS 4.0 — and Charter's network-evolution project is substantially complete by end-2027. A defined two-year window, not a secular stream.
D3. Top picks
HLIT — Harmonic · $11.22 · $1.22B · conviction 5.5/10
−37% off high, +36% off low · 14.4x fwd · P/FCF 12.6 (mid-band) · FCF yield 7.9% · net cash ($124M vs $133M debt) · GM 48%
FY25 FCF $96.9M (from −$1.4M in 2023) and $79M of buybacks — 6.5% of market cap retired in one year, the best capital allocation in the field. Headline revenue CAGR of −16.8% is the video business being divested (closing Q2'26); broadband alone is guided to $475–495M for FY2026 vs $360M total in 2025, raised on record backlog. AI vs carrier mix: ~0% AI, ~100% cable/broadband operator. Moat — switching costs, the strongest genuine moat in the field. cOS is deployed at 150 operators serving 45.7M cable modems across all tier-1 accounts; virtualizing a cable access network is a multi-year architectural commitment you do not swap out. Bear case: severe concentration in a handful of tier-1 operators (Comcast, Charter, Vodafone) — one deferral resets the year. The moat protects a shrinking pie: cable is losing broadband share to fiber and fixed wireless, and Charter's upgrade is done by end-2027. Post-divestiture margin structure unproven. Fair value $15–19 · entry below $10.
ADTN — ADTRAN · $8.42 · $682M · conviction 5/10
−58% off high, +18% off low · 11.1x fwd · EV/EBITDA below its own band · FCF yield 8.9% on cap · GM 39% and rising (35.1%→38.4%)
FCF trend is the real story: −$61M (2022) → −$89M (2023) → +$38M (2024) → +$60.5M (2025). A genuine cash inflection. Revenue +15.5% YoY in Q1'26. ~0% AI, ~100% carrier/broadband access — the purest trough expression in the field. ⚠️ EV data flag: cap $682M + debt $245M − cash $96M = $831M, but reported EV is $1.20B. The ~$370M gap is the DPLTA put obligation to Adtran Networks SE minority holders — a real liability. Equity is only 57% of EV. Bear case: no moat, negative ROIC, book equity of $146M against a $370M minority put, 18% short float, competing against Nokia and Calix. Fails §2's quantitative test outright. Fair value $9.50–12.50 · entry below $8.00.
VIAV — conviction 3/10, WATCH NOT BUY
Best business in the field (61% GM, genuine intangibles moat in network test & certification), but it rallied +148.65% YTD by April precisely because it re-rated from a carrier-test story into an AI-networking-test story. FY25 revenue of $1.08B is still below FY22's $1.29B; FCF $62M against an $8.65B cap = P/FCF ~140; 3yr FCF CAGR −16.3%. Growth is acquired (Spirent, $425M), not organic. Fails the diversification test. Revisit under $22.
Also rejected: CLFD (only sub-2x book in the field but revenue −15%, negative op margin, EV/EBITDA 66); LITE/CIEN/COHR/MTSI/SMTC/ALAB/AAOI (all above own bands); EXTR (P/B 50, current ratio 0.91); POET (pre-revenue, P/S 890). 🔎 Housekeeping: JNPR and INFN return 404 — HPE's Juniper and Nokia's Infinera acquisitions have both closed. DZSI delisted.
D4. Verdict by stream
AI interconnect — NO GEMS, emphatically. Structural price deflation, a 2H'26 oversupply warning, 2–3 hyperscaler concentration, and no nameable moat. Low multiples on peak AI-optics earnings would be a trap; we don't even have low multiples.
Carrier/cable — modestly interesting, not deep value. "Not re-rated" turned out to mean mid-band, not cheap. Both survivors are thin-moat businesses with negative-to-marginal ROIC, dependent on spending programs with defined end-dates. Cyclical rentals, not compounders.
Most worth /analyze: HLIT — not ADTN. ADTN wins the valuation test but fails §2 completely. HLIT is the better business at a fair price; §0 says great-and-cheap is the goal, nothing here is both, and HLIT is the only name where the "great" half is defensible.
PART E — The Crypto Infrastructure / Retail Fintech Screen
E1. Screened mid-bear-market — the ideal moment
BTC peaked ~$126K (Oct 2025) and trades in the low-$60Ks — a ~50% drawdown. The "do these earnings survive a bear market" question did not need modelling; it is in the print.
The single most useful number in the sweep
Revenue YoY, same quarter Verdict COIN −31% substantially coin beta HOOD +32% (record $1.31B) genuinely diversified 63 points of divergence in the same market. Reality ran the experiment.
E2. Regulatory state of play (mid-2026)
- GENIUS Act is law (signed 2025-07-18); six agencies hit the 2026-07-18 rulemaking deadline; issuance prohibition on non-permitted issuers bites ~November 2026. This hands incumbents a licensing moat — regulatory cost is now a barrier to entry.
- CLARITY Act (market structure) is stalled — passed House, Senate Banking approved 15–9 on 2026-05-14, never reached the floor; the 2026 window shuts ~Aug 10. Do not underwrite CLARITY passage in 2026.
- SEC is routing around Congress — three proposals targeted for July 2026 (token offerings, broker-dealer custody, trading-venue structure). Constructive but slower than the 2025 narrative priced in.
- Rate cross-current: 10yr 4.71% + ~72% odds of a hike is a tailwind to float income (COIN stablecoin line, HOOD net interest, CRCL reserves) and a headwind to risk appetite. Net negative for COIN (transaction loss > float gain), net positive for HOOD.
E3. HOOD — Robinhood · $86.60 · $77.98B · conviction 7/10
−43.7% off high, +36.3% off low · P/E fwd 28.2 · P/S 16.9 · P/B 8.54 · FCF $1.58B (from −$207M in FY24) · Revenue CAGR 3yr 48.8% · ROE 21% · diluted share CAGR +1.5%
Recurring vs transactional — the decisive test. Q2'26: $1.31B revenue +32% YoY, with 13 separate business lines each above $100M annualized. Gold at 4.8M subscribers (pure recurring); net interest on customer cash (rate-levered recurring); banking and credit card; and prediction markets at $156M, up 10x YoY — a revenue line with zero BTC correlation.
Dilution is the standout. 1.5%/yr for a company growing this fast is remarkable. $1.5B buyback authorized; $664M / 7.5M shares repurchased YTD at ~$94 average — buying back into the drawdown. Cleanest capital-allocation signal in the field. ⚠️ Data gap: press citing "SBC = 49% of revenue" is wrong — the $641M figure is combined adjusted opex AND SBC. True standalone SBC % is not cleanly disclosed. Needs the 10-Q.
Moat — network effects (moderate) + switching costs (weak but strengthening). Zero-commission trading is commoditized; the real asset is the conversion of a trading app into a primary financial relationship (Gold, banking, retirement matching, credit card) — assets arriving via retirement transfer and direct deposit don't leave. The most defensible piece is the prediction-markets first-mover position: a CFTC-regulated venue where liquidity begets liquidity, with genuine licensing scarcity.
Coin-beta verdict: REAL BUSINESS — proven, not asserted, by +32% growth during a 50% BTC drawdown. The only name in the field that is true diversification for a BTC holder.
Bear case: Schwab and Fidelity can copy every product with vastly more assets. Prediction markets face Kalshi and Polymarket and invite regulatory backlash the moment a scandal hits — politically fragile revenue. The 32% growth is flattered by a retail-speculation regime a hawkish Fed should eventually end. And valuation is the real problem: P/S 16.9 / P/B 8.5 vs FY24 averages of 6.5 / 2.4 — a ~3.5x re-rating on the multiple alone. The 44% drawdown removed froth, not the premium. Fair value $85–110 (base ~$95) · entry below $75, strong below $68. Today is fair, not cheap. §0 quadrant: great company, bad entry.
E4. COIN — Coinbase · $163.58 · $43.10B · conviction 5/10
−59.3% off high · P/E fwd 38.8 · P/B 3.20 · P/TBV ~4.6x · FCF $2.43B (−21.6%) · diluted share CAGR +8.9% 🚩
Subscription & Services = $555M, 48% of net revenue (up from 29% in Q4'24) — genuine progress. But stablecoin revenue is $292M, over half that line, and COIN captures ~50% of USDC economics on a record $20B average balance; staking and custody scale with token prices. So "recurring" here means "not volume-linked" — not "not crypto-linked." S&S also missed ($555M vs $599M consensus); third consecutive quarterly revenue and EPS miss.
Moat — the strongest in the field: regulatory/licensing intangible (strengthening under GENIUS), institutional network effects, brand/trust. Crypto trading market share hit a record 10.3%, a third straight quarter of gains — winning share while the pie shrinks.
vs own band — the genuinely interesting part. P/B 3.20 vs FY25 low 2.58 / avg 5.05 / FY24 avg 5.34. P/TBV ~4.6x vs FY25 avg 8.10x. P/S 6.86 vs FY25 avg 10.11. Bottom decile of its (brief) valuation history — the first time COIN has been objectively cheap on assets.
Coin-beta verdict: ~60/40 coin beta / real business. Best ratio in the field and improving — but still majority coin beta, and the user already owns BTC. With beta 3.35, this is leverage on an existing position more than diversification. Bear case: retail take-rate is structurally doomed (highest-margin revenue, every competitor undercutting). CLARITY passage would legitimize competitors as much as COIN. 8.9% annual dilution means the business must compound ~9% just to hold per-share flat. Fair value $140–210 (base ~$175) · entry below $150, strong below $135. Method: recurring S&S annualizes ~$2.2B; at 8–10x EV/S = $17–22B, plus $11.3B cash less $7.83B debt ≈ $21–26B, or ~$79–97/share for the recurring business alone. The market pays ~$17–22B for the trading franchise — defensible, not a bargain.
E5. Sub-field verdicts
🟢 Exchanges — the only sub-field with real businesses. GENIUS implementation actively widens the incumbent moat.
🔴 Miners — avoid the sub-field. The AI pivot is real; the pricing is not.
| 3yr diluted share CAGR | Op margin | |
|---|---|---|
| CLSK | +95.4%/yr 🚩🚩 | −246% |
| MARA | +46.3%/yr 🚩 | −558% |
| RIOT | +34.7%/yr 🚩 | −280% |
Shareholders in CLSK, MARA and RIOT were diluted into oblivion to fund BTC accumulation the user can simply own directly, at zero fee and zero dilution. The cleanest argument in this report.
Contracted-vs-energized reality check on the AI-pivot names:
| Contracted book | Actually energized | Balance sheet | |
|---|---|---|---|
| IREN | $9.7B (Microsoft, IG counterparty) | 200MW IT load contracted | $3.7B converts |
| CORZ | $10.2B / 12yr (CoreWeave) | ~200MW billing | BVPS −$7.57 🚩 |
| WULF | $12.8B "signed" | only 39MW online 🚩🚩 | $5.7B debt, BVPS −$0.18 |
| CIFR | ~$11.4B | site in development 🚩 | $1.7B secured notes |
WULF and CIFR at EV/Rev 60.5x and 53.8x are the 2024 MOU pattern in better clothes. IREN is the least-bad (conviction 3/10, §6 SPECULATIVE) — but note it has converted from a BTC proxy into a leveraged single-tenant AI-datacenter landlord, which diversifies away from BTC straight into the AI-capex concentration the user already carries. Trading one bubble for another.
🟡 CRCL — the value trap of the field, flagged explicitly. Do not be seduced by −66%. Reserve income rose 17% to $653M on 39% USDC growth, but net income fell 15%, gross margin is 8% and net margin −3% because distribution costs eat it ($907.9M paid to Coinbase in 2024). Two structural problems: the Coinbase distribution deal renews August 2026 — next month — with Coinbase holding all the leverage, and Coinbase and BlackRock are backing OUSD, a yield-sharing rival. Circle's largest partner is also its largest competitor and controls its margin. That is an anti-moat. Textbook §0 trap.
Also screened: XYZ/Block — genuinely cheap (16.3x fwd, PEG 0.87) but at its 52-week high and barely crypto-infra; worth a separate look on valuation grounds. NU (12.5x fwd, ROE 30%, +44% growth — excellent, but LatAm fintech, wrong field). ETOR (10.8x fwd but revenue −36% — coin beta). BULL, AFRM, GLXY (295x fwd) — no.
E6. Verdict
Most worth /analyze: HOOD — the only candidate passing the decisive test on evidence rather than argument. But the open question is valuation, not quality; §0 puts it in "great company, bad entry." Watchlist as 👀 On Deck, not ⭐ Shortlist. Entry <$75.
Secondary: COIN at conviction 5 — cheapest vs its own book ever, best moat in the field, but ~60/40 coin beta that overlaps the user's existing BTC.
Honest partial no-gems answer: for miners and stablecoins there is no gem. The user's direct BTC holding is a strictly better way to own crypto price exposure than any miner on this list.
PART F — Semi materials (summary; full file separate)
No gems — but it fails differently from ATE. The thesis is true: in the 2023 downturn semiconductor materials revenue fell −8.2% vs WFE −18.8% (SEMI) — genuinely ~2.3x less cyclical. Four reasons it doesn't make the trade:
- Stability without growth. Every name ~zero 3yr revenue CAGR: ENTG −0.9%, Q −0.0%, ESI 0.0%, MKSI +3.5%. An annuity that doesn't grow is a bond, and none trades at a bond price.
- The moat was capitalised away. Qualification lock-in is real (GM 42–47%), but ROIC is 5.7–11.0% — at or below cost of capital. Every name is a roll-up (ENTG/CMC, MKS/Atotech written off ~$1.8B in 2023, Q a DuPont carve-out with 53% goodwill). The economics went to the sellers.
- The razor-blade annuity barely exists as a listed asset. Only ENTG is genuinely per-wafer (75–80% unit-driven). MKSI is 40% recurring and only 43% semiconductor; Q's faster-growing half is AI packaging. Photoresist and wafers have no US-listed pure play.
- The field is consolidating away.
⚠️ ESI is disqualified — it is being acquired. Solstice agreed to buy Element Solutions for $14.5B on 2026-07-06. Its "below-decade-low" 17.2x forward P/E is meaningless, and its headline "+56% revenue growth" was acquisition-driven, not organic — a naive screen would have been misled twice over.
⚠️ PART F ADDENDUM — late primary-source research partially OVERTURNS the above
A follow-up research pass using primary filings and SEMI market data materially refined this field. Three corrections:
1. The materials-is-defensive thesis FAILED in this actual selloff. Measured from the SOX peak close (14,634.72 on 2026-06-22) to 2026-07-29:
| Drawdown | Drawdown | ||
|---|---|---|---|
| ENTG (purest materials, 75% unit-driven) | −41.8% 🚩 | AMAT | −31.8% |
| MKSI | −38.7% | TER | −30.1% |
| LRCX | −38.4% | Q (Qnity) | −29.7% |
| ONTO | −37.2% | ASML | −19.5% |
| KLAC | −36.8% | ^SOX | −28.6% |
Entegris — the name whose model best matches the thesis — was the worst performer in the entire group, worse than every equipment stock. And it was not earnings-driven (ENTG doesn't report until Aug 4); it was pure sector beta plus China-localization fear. ASML, an equipment name, was the best true-semi performer. In a multiple-compression event, the equipment/materials axis explained almost none of the cross-section — leverage and AI-narrative beta explained far more (the two most levered names, ENTG 3.6x and MKSI 3.5x, were the two worst materials performers).
2. ⚠️ SOX is −28.6%, not −24%. The −24% figure in circulation measured to the 7/17 intraday low and is stale; the rout extended sharply in the final week (−4.9% 7/27, −4.5% 7/28, −5.3% 7/29). Watchlist.md corrected.
3. The mechanism was wrong, and the correction generalizes. Materials consumption tracks fab utilization / wafer starts; equipment tracks fab capex. These are different cycles:
| Materials | Equipment | Cycle type | |
|---|---|---|---|
| 2019 | −1.1% | −7.3% | capex downturn → materials defensive ✅ |
| 2023 | −8.2% | −1.3% | utilization downturn → thesis INVERTS ❌ |
(SEMI, 2020-03-31 / 2020-04-14 / 2024-05-06)
Materials is defensive against capex cuts but fully exposed to utilization declines — and a demand shock hits utilization FIRST, with capex lagging on backlog. SEMI explicitly named CMP among the biggest 2023 contractions — the exact line ENTG and Qnity dominate. (Caveat: the 2023 equipment number is flattered by a China/CHIPS stockpiling bid, China +29% to $36.6B while Taiwan −27%, Korea −7%.)
4. Ranking reversed — Q (Qnity) is the better name, not ENTG. Qnity's 10-K states ">90% of 2025 revenue from consumable or unit-driven products"; leverage is ~2.0x (vs ENTG 3.6x), coverage ~6.5x (vs ~4.8x), FY26 guidance was raised, and it fell least in the rout (−29.7%). ENTG has declined in revenue two consecutive years (FY24 −8.0%, FY25 −1.4%) while WFE grew, and its apparent 2023 resilience was a CMC Materials acquisition artifact — organic was down.
5. 🔴 CBT must be excluded entirely — it is not a semiconductor company. The CMP slurry business left Cabot in 2000 (spun as Cabot Microelectronics → CMC Materials → acquired by Entegris in 2022). Cabot today is 63% carbon black for tires. It was the only name that held up (+0.9%) precisely because it has no AI-narrative multiple to compress. Its inclusion in any "semi materials" basket flatters the basket thesis — strip it out and the materials group's average drawdown moves from ~−27% to ~−33%, converging on equipment.
6. ROG is also not semi materials — power-electronics packaging with zero wafer-start exposure; belongs in an EV/industrial bucket. Best balance sheet in the cohort though (net cash, beta 0.49).
7. ESI update: closed $34.79 vs implied deal value ~$50.10 — a 31% discount, and that gap is SOLS' own share price collapsing, not deal risk. ESI is now a leveraged proxy on Solstice, not a semi-materials expression.
8. MKSI deleveraging is better than reported — ~$2.3B of gross debt retired since the 2022 Atotech close ($6.4B → $4.05B), with a Feb-2026 €1.0B notes issue prepaying $1.274B of USD TLB and cutting the spread SOFR+200 → +175. Interest expense −15% YoY; coverage ~5.8x. Genuine, well-executed. But at 3.5x leverage and beta 1.93 it remains a WFE beta trade, not a defensive holding.
🚩 Critical unclosable gap: SEMI materials market figures for 2001, 2002, 2008, 2009, 2010 are unobtainable (semi.org blocks automated fetch; pre-2015 releases unindexed; requires SEMI's paid subscription). This means the two deepest downturns in semiconductor history cannot be used to test this thesis at all. Anyone asserting "materials held up in 2009" is asserting it without a citable number.
🚩 Wire-service error caught: a circulating claim that MKS reported Q2 revenue of $920M on June 24 is FALSE — it contradicts the company's own scheduling release (2026-07-13). Q2 actuals are not yet out for ENTG (Aug 4), Q (Aug 4) or MKSI (Aug 5). All Q2 figures in this file for those three are guidance.
Revised top of field: Q (Qnity) — best consumable mix, lowest leverage, raised guidance, smallest drawdown. ENTG demoted to #2 on leverage, coverage and two years of revenue decline.
Top: Q (Qnity) $130.52 conviction 5 (best business — 23% op margin, 2.1x leverage, +17.6% YoY — but no valuation band exists post-Nov-2025 spin, and standalone-adjusted P/E is ~49x not 42x; entry $90–100). ENTG $117.11 conviction 5 (best moat and ROIC 11%, the only true volume annuity — but revenue shrank three years and interest coverage is 2.9x; entry $75–90). MKSI rated AVOID, conviction 2 — above its entire 10-year EV/EBITDA range, tangible book −$1.82B, $1.4B due within 12 months against $569M cash.
⚠️ Data corruption flagged: ENTG's FY2025 operating and EBITDA margins are wrong in BOTH Yahoo and roic.ai — a mis-mapped Q4 shows negative operating expense. The widely-quoted "41% EBITDA margin" is fiction; true operating margin is 13–17%.
PART G — Recommended next actions
In zone now:
1. /analyze DOCS — the only name at the floor of its own historical band. Crux: is 18.9% SBC a one-year AI-investment spike or the new run-rate? It swings fair value from ~$20 to ~$40. Note the dual-class structure: screeners using the 128.86M Class-A count overstate cheapness by ~10%.
2. /analyze ZBRA — widest gap between what the market thinks it owns and what it does. Crux: how much of the 2026 surge is one-time device refresh, and how large is the machine-vision/robotics business (Zebra doesn't disclose it — and it's the entire optionality).
3. /analyze NTES — only "great + cheap" name in consumer software. ⚠️ Would be a third Chinese position alongside BABA and TCEHY; size the Structural-Risk sleeve, not the name.
Watchlist with hard triggers: | Ticker | Trigger | Note | |---|---|---| | HOOD | <$75 | Great business, bad entry. On Deck, not Shortlist | | FORM | <$70, ideally <$60 | Purest consumable annuity in semi test | | CGNX | <$48 | Best switching-cost moat in automation | | HLIT | <$10 | Real moat, shrinking pie, 6.5% of shares retired/yr | | COIN | <$150 | Cheapest vs own book ever; overlaps BTC | | SONY | $19.50–21.50 | Only frictionless Japan access; the only Japan name that diversifies |
Still unswept (deferred): digital-infrastructure REITs (EQIX, DLR, IRM), EDA/semi-IP beyond SNPS (CDNS, ARM), enterprise storage/PC-adjacent (DELL, HPE, PSTG), government/defense IT beyond LDOS/BAH (CACI, SAIC, PSN).
Data gaps and hazards across round 3
- roic.ai is on the FREE plan (2 years of history). No name in any field was tested against a true 5–10yr band. This makes AI-name results conservative — the 2-year window is itself inside the bubble.
fin.pysupplies ~4 annual periods, so all CAGRs across rounds 2–3 are 3-year, not the §1-preferred 5–8yr. Primary reason no conviction exceeds 7.0.- ENTG FY2025 margins corrupted in Yahoo AND roic.ai — the quoted "41% EBITDA margin" is fiction (true: 13–17% operating).
- ADTN's EV requires a +$370M DPLTA minority-put adjustment; market cap alone understates it materially.
- COIN share count inconsistent (snapshot 222.43M vs balance-sheet 267.84M — likely Class A vs total).
- HOOD standalone SBC % not cleanly disclosed; widely-cited "49% of revenue" conflates opex + SBC and is wrong.
- DOCS dual-class — Class-A-only screener caps overstate cheapness ~10%.
- Miner energized-MW figures come from secondary sources, not filings — verify against 10-Qs.
- VIAV FY25 diluted EPS/share count returned null from Yahoo.
- JNPR, INFN, DZSI no longer investable (acquisitions closed / delisted).
- claude.ai connectors Alpha Vantage, FactSet, S&P Global unauthorized this session.
Scouts: Optical/Photonics · Semi Materials & Consumables · Health Tech/Techbio · Crypto & Retail Fintech. ~120 names screened in round 3, ~370 across rounds 2–3. Data from .mcp/fin.py (Yahoo), roic.ai (FY2024–25) and Public.com live quotes, retrieved 2026-07-29.