~16 min read · 3,615 words ·updated 2026-04-29 · ⚠ speculative ·
confidence 96%
Marvell Risk Register — Seven Categories
This register codifies the structured-risk view that complements bear case (narrative form) and bull case (positive thesis). Each risk is identified by category prefix (T = Technical, C = Customer, F = Foundry/Supply, K = Capital, IP = Intellectual Property, X = Competitive, M = Market). Likelihood and Impact are scored 1-3 (1 = low, 2 = medium, 3 = high). The detection trigger field describes the public signal an analyst would observe if the risk were materializing. The mitigation field summarizes management’s commentary or structural defenses where applicable.
The register is calibrated against today’s date (2026-04-29), Marvell’s current operating posture (FY27 in progress, three M&A deals integrating, NVIDIA $2B investment closed, $1B 5.300% senior notes settled), and a spot price of $154.83 per data/STOCK_PRICE_DATA.json close 2026-04-28.
T — Technical risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| T1 | Polariton POH commercial yield gap vs. lab demos. Horst et al. Optica 2025 1.1 THz EO bandwidth (DOI 10.1364/OPTICA.544016) was achieved on hand-fabricated MPW-shuttle devices. Production-grade volume yield, organic-chromophore poling stability, and TSMC silicon photonics process integration are unproven at COLORZ-3200 module scale. | 2 | 3 | First TSMC qualification lots in H2 2026 fail to clear automotive/industrial-grade temperature cycling; OFC 2027 (March 2027) Polariton POH demo reverts to thin-film LiNbO3 backup. | Marvell precedent: Inphi 2021 integration successfully transitioned PAM4 IP to TSMC volume in 18 months (inphi integration retrospective). Polariton ETH-trained engineers retained for design-center continuity. |
| T2 | Celestial Photonic Fabric scale-up architecture immature. OMIB and Photonic Fabric Switch chiplets validated at >14 Tbps per edge in pre-acquisition demos, but never at hyperscaler-rack volume; scale-up topology requires TSMC SoIC + CoWoS-L co-packaging. | 2 | 3 | Q2 FY27 / Q3 FY27 calls fail to disclose first hyperscaler design win; OFC 2026/2027 demos delayed; Lazovsky departs (Lazovsky is now EVP & GM Data Center Networking per session handoff). | Murphy Q3 FY26: Celestial brings “deep IP in optical interconnect, scale-up fabric architecture, and a world-class engineering team” (✓). $50M/quarter OpEx absorption already in Q1 FY27 (earnings calls Q4 FY26). |
| T3 | 1.6T DSP transition slippage. Murphy stated 1.6T DSP “in production at lead customers” Q3 FY26 (✓), but Marvell’s revenue framing assumes 1.6T becomes “meaningful revenue contributor in fiscal 2027” (Murphy Q2 FY26 ✓, earnings calls). Module OEM ramp at Innolight, Eoptolink, Coherent, Lumentum is the gating step. | 2 | 2 | Q1-Q3 FY27 optical revenue grows below 30% YoY (vs. baseline expectation of 40%+ implicit in FY27 +40% data-center guide). | Murphy Q2 FY26: “We continue to ship every 800G port we can build. We are not seeing meaningful share loss to alternative architectures” (✓, Q1 FY26 verbatim, earnings calls). |
| T4 | Custom xPU first-pass silicon respin risk. Marvell touted “100 billion-plus transistor chips…high volume production on first-pass silicon without any respins” (Murphy Q3 FY25 ✓). Trainium 3, Maia 200 are next-generation programs at advanced-node TSMC; respin probability at advanced nodes is non-trivial. | 1 | 3 | Slip in custom xPU revenue cadence quarter-over-quarter; mgmt commentary citing “yield optimization” or “design margin”; channel checks reporting hyperscaler delivery delays. | Murphy Q3 FY25: “Our success in ramping these highly complex…100 billion-plus transistor chips from initial samples to high volume production on first-pass silicon without any respins is a testament to Marvell’s robust design methodology” (✓). |
| T5 | Plasmonic-organic hybrid material thermal stability. Polariton’s POH active material is an organic chromophore (specifically LWLG’s Perkinamine series 3 per Horst et al. 2025 acknowledgements ✓ DOI 10.1364/OPTICA.544016). Long-term Telcordia-grade thermal/photo stability for hyperscale 5+ year deployment lifetimes has historically been the limiting factor for EO-polymer commercialization. | 2 | 2 | Polariton/Marvell H2 2026 qualification lots fail GR-1221 (Telcordia generic thermal cycling) or 85/85 humidity testing; first commercial COLORZ 3200 lot returns from hyperscaler customers. | LWLG-Polariton joint PR Mar 2022: “world-class enhanced stability” claim (PRNewswire ◐). Polariton’s 2024 Sep 24 PR cites “decade of experience in plasmonic circuits” + LWLG material. |
C — Customer risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| C1 | AWS Trainium 3 / Trainium 4 share loss to Alchip (Stifel customer-cliff thesis). Stifel (Ruben Roy) March 10 2026 note analyst coverage Shift 8 contended that Marvell loses Trainium 3/4 to Alchip; JPMorgan (Sur) channel checks countered. AWS estimated 40-50% of Marvell custom xPU revenue (dcf sum of parts §2.2). | 2 | 3 | Hot Chips 2026 (Aug 25) or ECOC 2026 (Sep 21) channel checks confirm Alchip Trainium 3/4 win; AWS re:Invent 2026 (Dec) names second-source ASIC vendor; FY28 custom xPU revenue prints below $3.4B ($3.6B base). | Murphy Q4 FY26: “Do you see me blinking? You do not.” (✓, Hari question). Multi-generational program continuity language repeated 6 quarters running (earnings calls cross-quarter synthesis). |
| C2 | Microsoft Maia 200/300 rotation to Broadcom. Microsoft is the most likely candidate to rotate to AVGO if Marvell’s Q1-Q2 FY27 execution slips. (broadcom — AVGO holds Google TPU + Meta MTIA Iris but not Maia today). | 1 | 3 | Microsoft Build 2026 (May) or Ignite 2026 (Nov) names AVGO on Maia 300 / future generation; Microsoft Azure DC capex commentary cites “multi-source ASIC strategy.” | No public Marvell disclosure on Microsoft contract terms. Multi-generational program design (Q3 FY25 ✓) implies Maia 100, 200 are contractually programmed; future-generation contention is the realistic risk vector. |
| C3 | Meta MTIA Arke ramp slips OR Broadcom expands into Arke. Broadcom’s April 14 2026 multi-gigawatt extension with Meta covers Iris (training); the bear extension is AVGO bidding into Arke on a future generation (broadcom ✓ globenewswire 2026-04-14). | 2 | 2 | Meta Connect 2026 (late Sep / early Oct) names AVGO on inference; Marvell Arke revenue guide softens; channel reporting on Meta inference fleet refreshes. | Marvell positions Arke as inference-side franchise complementing AVGO’s Iris training role; market segmentation reduces head-to-head pressure (analytical inference; no primary disclosure). |
| C4 | Single-customer concentration disclosure (Distributor A 37% / Customer A 14% per FY26 10-K). A single distributor or end-customer cancellation has outsized P&L impact. (segment revenue, 10-K Item 1A ✓). | 2 | 3 | Q1 FY27 10-Q discloses elevated concentration (>40% distributor or >18% customer); concentrated customer files for capex deferral; Murphy commentary acknowledges “specific customer dynamics.” | S&P Oct 2024 ratings commentary cited concentration as the principal monitored risk (credit market positioning §4 ✓); Moody’s Dec 2025 upgrade nonetheless held outlook stable. |
| C5 | Hyperscaler capex deceleration (industry-wide). AWS, Microsoft, Google, Meta combined FY26-FY28 AI infrastructure capex is the demand engine. Any pause — recession, regulatory pressure, capacity-utilization concerns — compresses the entire custom-silicon ramp simultaneously. | 1 | 3 | Hyperscaler capex guidance trimmed in their Q4 CY26 (Jan-Feb 2027) calls; AWS, MSFT, GOOG, META consensus capex midpoints decline; AI-revenue-per-dollar-of-capex commentary turns cautious. | Murphy Q3 FY26: “We see a massive opportunity ahead” (✓). The capex cycle is the macro variable Marvell cannot manage; but the Inphi-driven optical interconnect TAM grows even at moderated GPU growth. |
F — Foundry / Supply chain risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| F1 | TSMC CoWoS-S allocation cut. Marvell, NVIDIA, AMD, AVGO, Apple all compete for CoWoS-S advanced-packaging allocation. Marvell’s custom xPU + Celestial Photonic Fabric (SoIC + CoWoS-L) ramp depends on uninterrupted allocation. | 2 | 3 | TSMC Q2-Q3 CY26 capex calls cite CoWoS bottleneck; Marvell guides Q3 / Q4 FY27 below baseline; channel reporting on packaging-shortage prioritization. | Murphy Q4 FY26: “We have ample capacity commitments for the FY28 ramp” (✓ Q&A response to Aaron Rakers). Marvell is a long-term TSMC partner; Inphi was acquired 2021 with TSMC relationship intact. |
| F2 | HBM supply allocation (Micron / SK Hynix / Samsung). Custom xPUs co-packaged with HBM3E / HBM4 require base-die partnership. Marvell’s “value bundle” (Pillar 3 in bull case) includes HBM-base-die partnership, but supply is gated by the three HBM vendors. | 2 | 2 | Hyperscaler customer requests Marvell-alternative HBM source; HBM4 ramp delays in CY26-CY27; SK Hynix or Micron capacity guide cuts. | Cross-license / co-development with HBM partners is identified as a Celestial moat-extension catalyst (celestial ai ✓). |
| F3 | Innolight + Eoptolink China-bound module shipment compression. Marvell’s PAM4 / Ara DSPs are sold to Chinese module OEMs (Innolight, Eoptolink) that ship globally; a portion of volume is China-destined. BIS expansion to optical modulators / DSPs is a tail risk. | 1 | 2 | BIS publishes new export-control rules adding optical interconnect to controlled categories; Innolight or Eoptolink files China-export disclosure changes; channel volume drops at Chinese hyperscalers. | No primary mitigation disclosure; risk is pure macro-regulatory. Stress test in dcf sum of parts §6 sizes -$575M FY28 impact at the modeled 10-15% China-bound share. |
| F4 | TSMC silicon photonics process maturity for POH. Polariton’s POH stack requires TSMC silicon photonics (likely 65nm/45nm SiPh or N6/N5 monolithic if co-integration is pursued, polariton). TSMC SiPh is less mature than logic processes; PDK and verification flows are still evolving. | 2 | 2 | TSMC Q2 CY26 silicon photonics roadmap update; Polariton-Marvell first qualification lots delayed beyond H2 2026; OFC 2027 (March) demo slipped. | Marvell controls the Inphi PDK relationship and can leverage existing TSMC SiPh infrastructure; Polariton’s MPW-shuttle background gives ETH-side process familiarity. |
K — Capital risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| K1 | $1B senior notes refinance risk if rates rise. The April 2026 5.300% / 2036 issuance cleared at +97 bps over UST (424B2 2026-04-08 ✓). Future maturities (2028 $750M, 2029 $500M, 2030 $500M, 2031 $750M) reprice through 2032; if 10Y UST climbs >5% with widening spreads, refinancing costs increase materially. | 2 | 1 | UST 10Y crosses 5%; IG semi spreads widen >150 bps; Marvell credit rating outlook revised negative on customer-concentration concern. | Total liquidity $4.2B covers all maturities through FY2029 (credit market positioning §1). FCF $1.8-2.5B annually covers debt service 15-20×. |
| K2 | Celestial earnout dilution if FY29 cumulative revenue trips trigger. Up to 27.2M earnout shares (1/3 trigger at $500M cumulative Celestial revenue by FY29; full at $2.0B cumulative — full earnout unlikely on management’s stated ramp). dcf sum of parts §3.2; credit market positioning §6. | 2 | 1 | Celestial revenue trajectory tracks toward $500M cumulative by end-FY29; mgmt discloses earnout-fair-value adjustment in 10-Q quarterly earnings. | Earnout structure is performance-aligned: dilution only happens if Celestial delivers; the SOTP model already prices 50% PV expected ($1.5B in dcf sum of parts §5.2). |
| K3 | Buyback pace forced down to absorb three M&A integrations. Q1 FY27 is the first quarter Celestial $50M OpEx + XConn + Polariton OpEx hit simultaneously with no offsetting revenue. Free cash redeployment toward M&A integration reduces share count reduction and per-share growth. | 2 | 1 | Q1 FY27 buyback prints below $300M (vs. Q1 FY26 $340M ✓); Q2 FY27 commentary cites “balanced capital allocation”; Meintjes hedge language on capital returns. | Use-of-proceeds language in the April 2026 senior-notes raise explicitly listed dividends, repurchases, and acquisitions (credit market positioning §1a) — signaling buyback continuity. |
| K4 | Goodwill impairment trigger on Celestial / Polariton if revenue ramp disappoints. $3.25B Celestial purchase against pre-acquisition revenue <$50M run-rate implies most of consideration is goodwill + IPR&D. If FY28 / FY29 ARR targets slip materially, impairment risk materializes. | 2 | 2 | Q3 / Q4 FY28 disclosure of Celestial revenue meaningfully below the $500M ARR target; Marvell 10-Q goodwill-impairment-test commentary ratchets cautious; auditor commentary on intangibles. | Goodwill cushioned by integration into existing Marvell data-center segment (revenue cross-allocation possible); Inphi 2021 has carried $11.1B goodwill without impairment despite varied revenue cadence (segment revenue ✓). |
IP — Intellectual Property risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| IP1 | NVIDIA cross-license disputes around DPU / NIC / NVLink IP. Marvell’s OCTEON DPU and custom NIC programs interoperate with NVIDIA fabrics. NVIDIA’s NVLink Fusion ecosystem terms gate which third-party silicon participates. (nvidia ✓). | 1 | 2 | NVIDIA 10-K / 10-Q discloses litigation against Marvell or third-party DPU vendor; NVLink Fusion gating commentary; Marvell-NVIDIA joint-product disclosures stop. | $2B NVIDIA investment in Marvell (Tom’s Hardware ✓) signals strategic alignment, not adversarial posture. |
| IP2 | LWLG / NLM material-supplier dependency on Polariton POH stack. Confirmed primary-source: Polariton’s commercial POH modulators use LWLG’s Perkinamine™ chromophore (Horst et al. Optica 2025, DOI 10.1364/OPTICA.544016 acknowledgements ✓; Heni quotes in joint PRs March 2022 / September 2024). polariton lwlg verification, confidence ◐ partial. Marvell’s commercial 3.2T POH module roadmap is gated on a third-party supplier (LWLG). | 2 | 2 | LWLG (NASDAQ:LWLG) supply-chain disruption; LWLG-Marvell licensing dispute; Marvell publicly qualifies NLM Photonics’ Selerion-HTX (US 12,187,827) as alternative; Marvell builds in-house chromophore. | Material-supply not patent-license: Polariton’s device IP is ETH-Zurich-derived (US 10,571,724; US 11,764,873), now Marvell-owned; <5% claim-overlap risk vs. LWLG patents per LWLG patent KB §7.2. Substitution paths exist but are 2-3 year requalification. |
| IP3 | ETH Zurich license back to Polariton / Marvell post-acquisition. Polariton holds licensed rights to ETH Leuthold-lab foundational POH patents. Survival of the license through change of control is not publicly disclosed. (polariton deal terms “Outstanding monitoring items” ✓). | 1 | 2 | ETH Zurich Transfer files dispute or notice of license modification; Marvell 10-Q footnote on Polariton IP ownership; Swiss federal IP-court filing. | Inphi 2021 precedent: Marvell honored upstream university licensing (SiPh-related agreements). Polariton press release silent — typical for license-survival clauses. |
| IP4 | Pending Polariton App 18/357,334 prosecution / opposition. Polariton has at least one pending US patent application (App 18/357,334 plus EU/CH counterparts per polariton). Patent prosecution outcomes are not under Marvell’s control. | 2 | 1 | USPTO publishes office action narrowing claims; competitor (HyperLight, Lumentum, Coherent) files opposition or interference; granted scope materially narrower than expected. | Patent-prosecution risk is industry-standard; Marvell holds the prosecution rights post-acquisition and can amend claims through standard practice. |
| IP5 | 2008 SEC settlement legacy / governance overhang. Historical: Weili Dai 2008 SEC settlement for stock-option backdating (SEC Press Release 2008-82; Litigation Release LR-20559; SEC v. Marvell Technology Group, Ltd. and Weili Dai, Case CV 08-2367-HRL N.D. Cal. ✓). 5-year officer/director bar expired ~May 2013. Resolved, not active. | 1 | 1 | Not expected to recur; included for completeness. | Murphy-era governance (since July 2016) post-dates founder departures (April 2016, leadership ✓). 2016 PwC → Deloitte audit transition. No active SEC enforcement. |
X — Competitive risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| X1 | Astera Labs CXL fabric pricing pressure. ALAB pure-play CXL/PCIe re-timer + fabric specialist trades at 18-22× EV/Sales (Apr 2026, dcf sum of parts §5.1). Marvell’s XConn-derived CXL switch revenue ramp competes for the same hyperscaler RFQs. | 2 | 2 | XConn / Apollo 2 design wins announced by hyperscalers cite ALAB as alternate vendor; ALAB earnings call references Marvell-XConn as competitor for the first time. | Marvell’s PCIe re-timer + XConn switch portfolio is end-to-end (re-timer + switch); ALAB is re-timer-strong but switch-weaker. Apollo 2 (CXL 3.1 + PCIe Gen 6) sampling differentiates on protocol completeness (xconn ✓). |
| X2 | Broadcom Tomahawk-Bailly CPO disrupts merchant DSP. AVGO’s co-packaged optics roadmap is positioned as scale-up alternative to merchant pluggables (broadcom ✓). Marvell’s Inphi-derived 1.6T DSP volumes face structural pressure. | 2 | 3 | AVGO Q1-Q2 CY26 calls cite Bailly CPO design wins at Microsoft, Google, or Meta; pluggable optical port volume guidance softens at module OEMs. | Marvell counters with Polariton POH for next-gen pluggable + Celestial Photonic Fabric for scale-up — an end-to-end optical stack response (bull case Pillars 2-3). |
| X3 | AMD Pensando-derived DPU competitive expansion. AMD’s Pensando acquisition ($1.9B, 2022) targets the same DPU market as Marvell OCTEON. AMD’s bundled GPU+DPU+CPU stack creates pull-through pressure. | 1 | 1 | AMD MI400 / MI500 launch bundles Pensando DPU as default; OCTEON revenue trajectory softens. | OCTEON has installed-base inertia; merchant DPU market is large enough for multi-vendor coexistence; Marvell DPU revenue is a small fraction of total (dcf sum of parts §2.1). |
| X4 | Alchip second-source ASIC threat (custom xPU). Taiwan-based Alchip is the secondary merchant ASIC vendor cited in the Stifel customer-cliff thesis (analyst coverage Shift 8). | 2 | 3 | AWS, Microsoft, or Meta name Alchip on a future ASIC generation; Hot Chips 2026 / IEDM 2026 disclosures reveal Alchip-led custom-silicon programs. | Alchip lacks Marvell’s optical DSP + custom NIC + DPU stack; the multi-product bundle is part of Marvell’s hyperscaler value proposition. Murphy Q4 FY26 “do you see me blinking?” (✓) |
| X5 | NVIDIA in-house CPO / silicon photonics vertical integration. NVIDIA’s CPO program (announced 2024-2026) targets vertically-integrated optical interconnect for Blackwell Ultra and Rubin platforms (nvidia Risks ✓). | 1 | 3 | NVIDIA Q2-Q4 CY26 calls cite CPO production ramp; SC26 / GTC 2027 keynotes detail in-house optical roadmap; merchant DSP volume guides soften. | $2B NVIDIA investment in Marvell (Tom’s Hardware ✓) implies optical is part of the strategic alignment, not a target for disintermediation. NVLink Fusion ecosystem invites Marvell-enabled custom XPUs to interoperate. |
M — Market / Macro risks
| ID | Risk | L | I | Detection trigger | Mitigation / management commentary |
|---|---|---|---|---|---|
| M1 | AI capex deceleration (peak hyperscaler thesis). Hyperscaler capex moderation is the single largest macro variable. The credit market’s “tight spreads + 50× P/E” divergence (credit market positioning §7 ✓) suggests both equity and credit have priced in continued 40%+ DC growth. A 5-10% revenue miss could trigger synchronized repricing (equity -15-20%, credit spreads +40-60 bps). | 2 | 3 | Hyperscaler Q3 / Q4 CY26 capex guidance trimmed; CY27 capex midpoints decline; AI-revenue-per-dollar-of-capex commentary turns cautious. | Marvell’s “we don’t do an annual guide” discipline (Murphy Q2 FY26 ✓) is itself a hedge against macro miscalibration. |
| M2 | Multiple compression from “high-growth premium” rotation. MRVL trades 35-45× forward P/E (Apr 2026 levels, comps valuation ◐) — near 10-year highs vs. 20-25× 10-year median. Multiple-compression to 30-35× on any miss is plausible. | 2 | 3 | Q1 FY27 EPS prints below mgt midpoint $0.90; FY27 revenue guide softened; multiple-comp peers (NVDA, AVGO) re-rate. | Probability-weighted DCF/SOTP fair value $113 (dcf sum of parts §7) implies 27% downside vs. spot $154.83 — multiple compression risk is the dominant valuation risk. |
| M3 | AI capex cycle inflection (2027-2028). Long-cycle hyperscaler refresh dynamics; if FY28-FY29 marks the peak of generation-1 custom-silicon deployment, FY30+ growth depends on next-gen acceleration which is not yet committed. | 2 | 2 | FY29 hyperscaler capex guidance flat or down; second-generation custom-silicon programs not yet announced. | Murphy Q4 FY26: “several years of exceptional performance” (✓) explicitly framed multi-year. Multi-generational programs are a 3-4 year cadence. |
| M4 | Geopolitical escalation (Taiwan-China). TSMC Taiwan exposure is the single supply-chain dependency that would shock the entire merchant-silicon industry. | 1 | 3 | Geopolitical tension escalation; TSMC contingency-planning disclosures; insurance/risk premia repricing. | Industry-wide risk; Marvell’s fab-light model concentrates exposure but is no different from peers. |
| M5 | Recession / IT-spend pullback. Slower enterprise + carrier infrastructure spend (the 26% of FY26 revenue not in data-center). Communications & Other segment trajectory in dcf sum of parts §2.1 assumes modest growth FY27-FY30. | 2 | 1 | Carrier capex guides cut at telco operators; Enterprise networking softness at Cisco / Arista / others; Marvell guides communications revenue flat or down. | Communications & Other is now only 26% of FY26 revenue (vs. 60% in FY24, segment revenue ✓); risk is structurally smaller than data-center concentration. |
Aggregated risk-scenario sensitivity
| Scenario | Risks materializing simultaneously | FY28 revenue impact | DCF $/share impact (vs. base $98) |
|---|---|---|---|
| Single-tail bear | C1 (AWS Trainium 3 share loss, -25%) | -$950M / -6% | -$11 → $87 |
| Optical pricing bear | X2 (AVGO CPO disruption, optical -15%) | -$700M / -5% | -$8 → $90 |
| Geopolitical bear | F3 (China BIS expansion, optical -10-15%) | -$575M / -4% | -$7 → $91 |
| Soft-bear combined | C1 + M1 + GM compression | -$1.4B / -9% | -$24 → $74 |
| Hard-bear combined | C1 + M1 + X2 + F3 + IP2 (LWLG supply shock) | -$2.5B / -17% | -$40 → $58 |
The hard-bear scenario lands at $58/share DCF / $74/share SOTP — roughly a 50-60% drawdown from spot $154.83. That asymmetric tail is the structural reason for active position-sizing rather than passive accumulation; even at the base case fair value $98-$114 (dcf sum of parts §7), the implied downside is 25-37% from spot.
Cross-references
- bull case — positive thesis (where the ramp pillars are described)
- bear case — narrative bear thesis (the prose form of this register)
- catalysts — forward catalyst calendar (when each risk will be tested)
- open questions — research gaps that map to specific risk IDs above
- q1 fy27 preview — Q1 FY27 watch list (first risk-resolution window)
- dcf sum of parts — stress-test scenarios in §6 align to the risk IDs here
- earnings calls — verbatim management commentary on each risk where addressed
- analyst coverage — sell-side framing of customer-concentration / ASIC-share risks
- credit market positioning — credit-market view of customer concentration + leverage risks
- polariton deal terms — Polariton-specific risk register
- celestial ai — Celestial-specific risks
- polariton — Polariton-specific risks
- broadcom — primary competitive risk vector
- nvidia — coopetition risks
- segment revenue — concentration math
- polariton lwlg verification — IP2 dependency primary-source audit