# Huawei as Risk Factor for Optical in Data Center – Implications for $LITE $MRVL
**Final Research Report – Deep Research Map-Reduce (Wave1 Layer1 + Reducers)**
**Date:** 2026-07-10 UTC
**Prepared by:** Final Deliverable Coordinator (Layer 3, PID 1 child: task_agent_2ac3bf3)

---

## Executive Summary

**Question:** Huawei has ability to make optical telecommunications system. Is Huawei a risk factor for optical in data center? What does this mean for LITE, MRVL?

**Direct Answer – Two-market reality:**

Huawei **can** make the full stack – from QSFP-DD/OSFP 400G/800G intra-DC grey optics (VR4, SR8, DR4, FR4, OSFP-800G-SR8-D 100m OM4, OSFP-800G-2FR4-D 2km) to DCI transport OptiXtrans DC908/DC908 Pro (2019-09-29 GA Q4 2019, Pro Sept 2023 per PRNewswire: 19.2T/2U, 1.2T/wave to 240km, 96T/fiber C120+L120) and carrier OSN 9800 K36/K12/M24/U32E (64T cross-connect to 100T, Super C+L). 800G tunable coherent module launched 2020-02-26 London per Huawei press release is industry-first (CMS algorithm, 7th-gen oDSP). Evidence: Huawei e-file portfolio PDF copyright 2025 lists QSFP-DD-400G-VR4/-SR8/-DR4/-FR4/-SR4, QSFP112-400G variants, OSFP-400G-SR4/DR4, OSFP-800G-SR8/2xFR4 – proven high confidence via huawei.com/e.huawei.com/marketingcloud primary sources (P1).

However **telecom optical strength ≠ datacom hyperscale intra-DC dominance**. Telecom uses coherent QPSK/16QAM, 80-5000km, DWDM EDFA, OTN, chassis hundreds of watts, $1000+ per wave, cost per bit*km optimized. Datacom intra-DC uses IM-DD PAM4 100G/200G per lane, 50m-2km SR/DR/FR, QSFP-DD/OSFP <12W pluggable, $100-300, extreme cost/volume pressure, >60% of transceiver market per LightCounting. DCI (10-120km) is hybrid – Huawei leads enterprise DCI per GlobalData 2020-2023 and has 32.6% all-optical DCI ex-NA per Omdia 2019 calculation, but cases are Tier2 colocation (aixit Germany, Serverius, Converge ICT Philippines) not US hyperscalers. Intra-DC AI fabric new attempt is OptiXtrans DC808 MEMS OXC all-optical switch (Interop Tokyo 2025 Special Prize) – photon-electron hybrid, 98% port power saving claim, 20% fault reduction, but unproven at scale vs Google Apollo.

**Regulatory = hard wall for Western AI DC:** BIS Entity List effective May 16 2019 16:15 EDT (Federal Register May 21 2019 per Lexology), license presumption denial, Temporary General License to Aug 19 2019; Foreign Direct Product Rule amendment May 15 2020 + final expansion Aug 17 2020 (adds 38 affiliates total 152, removes TGL, expands to any subject product with knowledge Huawei purchaser or incorporated). Impact: TSMC 7nm/5nm cutoff for HiSilicon optical DSP, forces SMIC N+1/N+2 7nm DUV multi-patterning (Kirin 9000S teardown 2023), 5nm DUV quad patterning +40-50% cost, 1/3 yield. FCC: Secure Equipment Act signed Nov 11 2021 instructing FCC, FCC 22-84 Report & Order adopted Nov 25 2022 unanimous per Rosenworcel statement docs.fcc.gov/public/attachments/FCC-22-84A2.pdf – bans new equipment authorization via Certification or SDoC for Covered List (Huawei/ZTE all telecom eq). NDAA Sec 889(a)(1)(A) effective Aug 13 2019 prohibits federal agencies procuring covered telecom as substantial component, 889(a)(1)(B) Aug 13 2020 prohibits contracting with entity that USES covered eq anywhere. Hyperscalers as federal contractors (JWCC, GovCloud) must certify no Huawei anywhere per FAR 52.204-24/25/26. UK: purchase ban Dec 31 2020, removal by end 2027 per NCSC. EU 5G Toolbox Jan 29 2020 voluntary, only 11/27 implemented Aug 2024 per Euronews/IEEE Spectrum, no direct DC optics ban but hyperscalers globally exclude Huawei for single BOM compliance. Result: **Zero US hyperscaler market for Huawei optics – high confidence 90%+**.

**Market share context:** Dell'Oro via ElectronicsWeekly Feb 2025 summary FY2024 optical transport: Huawei 33% #1, Ciena 19% #2, ZTE +1pp, Cisco +1pp. Omdia WDM bitrate: Huawei #1 100G 21%, 200G 38%, 400G 31.2-35% four quarters 2022. Overall telecom equipment Dell'Oro 30% 2023 vs 28% 2022. But ex-APAC/NA top is Ciena #1, Huawei still in top5 combined >80% but share single-digit in NA. LightCounting overall transceiver total (telecom+datacom) 2023: #1 Innolight, #2 Coherent, #3 Huawei, #4 Cisco, #5 Accelink, #6 Hisense, #7 Eoptolink, #8 HGGenuine, #9 Source Photonics, #10 Marvell. Reading method change 2022 includes internal equipment manufacturing explains Huawei jump; 2022: Innolight/Coherent/Cisco/Huawei >50% market ~$1.4B each top2. Datacom pure 800G AI: Cignal AI 4Q24 – module leaders Innolight, Coherent, Eoptolink; component sources Coherent, Broadcom, Lumentum. Coherent #1 800G datacom 3Q24, Innolight leads 400G, Nvidia via Fabrinet #3. Global 800G shipments 20M+ 2024, 800GbE +60% 2025, 24M 800G+ units 2025 → 63M 2026 (TrendForce 2.6x), 60% of shipments 800G+ by 2026 per Google Ironwood. Huawei not in top3 800G AI Western chain – Innolight + Eoptolink capture ~60% incremental Nvidia GB200 orders (18,432 DR8 per 576-GPU cluster) per P5.

**China ecosystem nuance:** Innolight & Eoptolink are **independent private-origin** (Innolight founded Apr 2008 Suzhou by Osa Mok/Hsing Kung, Google Capital backed, merged into SZSE:300308 Zhongji Electrical; Eoptolink Apr 15 2008 Chengdu SZSE:300502) – evidence high confidence. Huawei is only 5% customer vs Alphabet 22%, Amazon 11%, Meta 6.4% per Bloomberg (digitaltoday.co.kr), Alibaba 1.9%. Meanwhile Accelink is quasi-SOE (WRI/FiberHome 36.4% via Interfax 2011, origin 1976 P&T Ministry Solid-state Device Institute, listed 2009 first Chinese optoelectronic supplier, formal 2023 joint development with Huawei Asia-Pac). Hisense Broadband (2003 Qingdao, JV Ligent Photonics 2002) municipal SOE background, Delphi Inc Delaware, OEM for Huawei/Ericsson fronthaul. Source Photonics strategic partnership with Huawei 2018 PAM4 showcase + Jintan fab co-built with Economic Development Zone. Huagong Tech/HG Genuine Wuhan Optics Valley anchor 12.8T XPO first Mar 2026 per Xinhua Silk Road June 2026. HiSilicon Optoelectronics captive (OM3538SX101 100GE QSFP28, OM9380ZS100 80km ZR, OM3660FX102 QSFP-DD 400G FR4) – #3 module vendor 2020-2021 captive, owns Caliopa 2013 EUR7m imec Ghent spin-off + CIP UK 2012 for silicon photonics, now TFLN-on-Si 400G/lane 110GHz BW Ge PDs demo ECOC 2025 per Gazettabyte – ready for pluggables and scale-out CPO per IPAC ELS work with PIC Magazine interview. Localization >90% in YRD/PRD/Wuhan clusters per Xinhua, 7/10 top10 vendors Chinese >60% global share. US DoD 1260H list additions Oct 2025-June 2026 includes Innolight, Eoptolink, BYD/Baidu/Alibaba per FDD – risk for US cloud BOM.

**Tech roadmap gap:** Western CPO shipping: Broadcom Bailly 51.2T CPO Tomahawk5 + 8x 6.4T SCIP shipping to Meta 2024 >30% power saving >4M flap-free hrs, Davisson 102.4T Tomahawk6 3nm 512x200G. Nvidia Spectrum-X Photonics 409.6Tb/s 512x800G + Quantum-X 800Tb/s 1.6T/port 5x efficiency, TSMC/Corning/Foxconn partners, available H2 2026 production Q1 2025 per some. Huawei tactical workaround: LPO (linear-drive pluggable) 800G OSFP up to 18 per XPU per LightCounting 2025, saves 2-4W vs 7-9W @400G but needs pristine link, Google/Meta OFC analysis said problematic at hyperscale, hype waning ECOC 2025 per Chris Cole. Strategic: 100T CPO target same as Western, needs 200G per fiber multi-λ, external laser source standard, but DSP node constrained to 7nm due sanctions unless LPO bypass.

**Implications for LITE and MRVL:**

*Is Huawei risk factor for optical in DC?* **Direct risk in Western AI DC: NO** (0% share, regulatory excluded, procurement policy exclusion, no hyperscaler adoption). **Indirect risk YES** via three channels:
  1. **China domestic competition** – Huawei + HiSilicon + Accelink/Hisense/HG Genuine dominate China telco/metro and China cloud (Alibaba, Tencent, ByteDance, Huawei Cloud). Reduces TAM for LITE/MRVL inside China (true China OEM <10% for MRVL, ~20-25% historically for LITE down to Hong Kong 24.2% shipment proxy). Pricing spillover: Chinese overbuild + 60% global share (Innolight/Eoptolink cost advantage) → 10% YoY ASP compression in 400G+ (Omdia via ad-hoc-news), market share slide 30% →25% peak for LITE.
  2. **Supply chain headline risk** – DoD 1260H listing of Innolight/Eoptolink (60% of Nvidia supply) could force hyperscalers to dual-source Western, benefiting LITE (EML bottleneck 50-60% share, sold out via LTAs through 2026-2027) and MRVL (PAM4 DSP 60-70% revenue share historically up to 80% shipment, Spica first production-ready 800G Mar 3 2022, Nova first 200G/lane 5nm 2023, Ara 3nm 2024, Orion first 800G coherent 5nm 130+GBd, Electra first 2nm 1.6T ZR, Libra 2nm 800G ZR – neutral supplier to Innolight/Eoptolink/Coherent/Lumentum). But also logistics risk via China contract manufacturing exposure (MRVL 39-45% shipment to China but majority is non-China customers factories; LITE Thailand hub 17.7% vs Shenzhen divestiture Mar 5 2025 gain $34.9M).
  3. **Technology fork** – Huawei LPO push + TFLN 400G/lane could enable low-cost China 1.6T domestic, pressure Western pricing if exported via gray channels, and forces Western CPO acceleration – threat to LITE high-power EML if SiPh TFLN replaces EML, but near-term mitigated by CW laser deliveries for 800G SiPh (LITE Q1 FY2026) + ultra-high-power for CPO scaling Q4 2025.

**LITE:** Cloud & Networking 86% FY2025 $1410.8M (+21% YoY), 4 consecutive quarters AI growth 56-65% YoY, OCS backlog >$400M Q2 FY2026, EML record shipments 100G lane, 200G ramping, gross margin 28% FY2025 vs 18.5% FY2024 (Huawei obsolescence reduction $20.8M). Huawei direct 0% now vs 11% FY2018, 15% YTD FY2019, 18% Q3 FY2019, guidance cut $375-390M vs $405-425M (-7.8%) May 20 2019. Hong Kong 24.2% shipment, Thailand 17.7% manufacturing derisking. Bullish scenario sanctions durable → scarcity rents GM path >40%, quarterly $500M → $750M target, 40% Western 800G share with Coherent. Bearish: Chinese price spillover 10% ASP compression continues, share 30%→20%, 200G EML second-sourcing 2027-28, relocation tariffs 100 bps headwind.

**MRVL:** Data center 70-75% revenue $6.1B FY26 74.4%, Q4 FY25 $1.37B 75%. Electro-optics $1B+ annual estimate substantial base per CEO Q1 FY25. Innolight/Eoptolink customers not Huawei; HiSilicon internal PAM4/coherent competition China-only, 2Tbps muxponder 0.1W/Gbps vs 0.125W industry claim via vertical integration but larger node. True China OEM <10% vs shipment headline 39-45% (10-Q May 26 2023 disclaimer majority shipments to China relate to non-China customers factories) explains -10.1% stock drop July 17 2024 on Bloomberg tightening report volatility despite fundamentals intact. Bullish: DSP moat intact, AI interconnect $1.5B beat → $2.5B FY26, TAM share 10%→20% of $75B by 2028. Bearish: China mandates domestic DSP, US tightens destination rules forces relocation, Broadcom BCM83640 3nm 400G/lane Taurus 2026 catch-up.

**Net assessment:** Huawei is **not a direct competitive threat to 800G/1.6T AI DC fabric in US/EU**; it is an **indirect catalyst increasing concentration risk (two hyperscaler customers >15% each for LITE, China shipment 40% for MRVL headline) and pricing pressure via its ecosystem proxies (Innolight/Eoptolink independent but China-centric cost base)** and a **structural benefit via regulatory shelter** that preserves LITE/MRVL pricing power short-term but creates overvaluation risk (>90x forward earnings LITE priced for perfection) and CPO disruption risk.

---

## Research Contract, Methodology, Map-Reduce Coverage

**Contract:** /workspace/orchestration/research_contract.md – operator question, scope 2020-2026 outlook 2027-28, hypotheses H1-H6, deliverables report with LITE/MRVL separate, bullish/bearish, acceptance requires regulatory, telecom vs datacom distinction, source-grounded provenance.

**Methodology:**
- **Layer 1 wave1:** 10 mapper partitions P1-P10 (P1 portfolio, P2 telecom vs datacom, P3 regulatory, P4 market share, P5 competitive DC 800G/1.6T, P6 LITE business, P7 MRVL optical, P8 China ecosystem, P9 tech roadmap, P10 stock implications) – each required browser.search/open, primary source preservation URL/publisher/date/retrieval, Markdown handoff with Scope/Layer/Parent/Methods/Sources/Findings/Structured Rows/Contradictions/Errors/Confidence/Follow-Up/Memo.
- **Reducers:** R1 (P1+P2+P9 capability), R2 (P3+P4 regulatory & market), R3 (P5+P8 competitive & ecosystem), R4 (P6+P7+P10 LITE/MRVL impact) – fan-in 2-3, inspectable Markdown, support/contradiction/unresolved/source quality/confidence/rejected/audit/next-wave/layer-4 blocked.
- **Coverage:** 80-100 distinct URLs, 11 huawei.com product pages + PDFs, 2 Lexology, 1 FCC PDF FCC-22-84A2, Dell'Oro via ElectronicsWeekly, LightwaveOnline LightCounting, Cignal AI press releases Q3/4 2024/2025, TrendForce, momoview, FS.com, SEC EDGAR FY2025 10-K Aug 19 2025 (lite-20250628), Marvell 10-Q May 26 2023 + earnings transcripts Motley Fool, Marvell.com product pages, Nvidia official Spectrum-X, Gazettabyte ECOC reflections, PIC Magazine HiSilicon interview, Wikipedia/Web.
- **Point-in-time:** Regulatory dates BIS May 16 2019 16:15 EDT effective (Federal Register May 21 2019), FDPR May 15 & Aug 17 2020, TGL expiry; Secure Equipment Act Nov 11 2021, FCC 22-84 Nov 25 2022; UK ban Dec 31 2020 removal 2027; EU toolbox Jan 29 2020 voluntary 11/27 Aug 2024; product launches DC908 Sep 29 2019 GA Q4 2019, 800G tunable Feb 26 2020 London, DC908 Pro Sept 2023 PR, OptiXtrans E6600 fgOTN MWC 2024, Bailly 51.2T 2024 shipping Meta, Spectrum-X H2 2026, TFLN demo ECOC 2025, DoD 1260H Oct 2025-June 2026.
- **Quant validation:** When measurable – market share numbers (Dell'Oro 33% FY2024), transceiver ranking (LightCounting #1 Innolight 2023), LITE segment $1410.8M 85.8% FY2025, MRVL data center 75% Q4, shipment vs OEM distinction; technical – PAM4 vs coherent power 15-25W vs 5.5W 100G.
- **Review:** Coordinator wave1 audit sampling 5 primary dates/numbers – all pass high-medium.

**Three-layer decomposition considered:** Layer1 executed wave1 fully, follow-up next-wave partitions promoted (P1a support.huawei.com datasheets, P2b DCI vs intra-DC TAM, P3b SMIC yield impact, P4b Dell'Oro quarterly exact curve, P5a-c EML LTAs MRVL vs Broadcom design wins 1.6T, P6b/P10b dual-sourcing risk 2027-28, P7b/c silicon photonics 3D engine export control, P8b DoD 1260H impact US BOM, P8c HiSilicon external sale verification, P9a patent search TFLN) – blocked layer4 requires hardware/BOM teardown, FCC counsel opinion, paywalled LightCounting/Cignal AI Excel, reliability laser data – recorded as Blocked Layer-4 Recommendations.

---

## Evidence Synthesis

### Huawei Capability – 400G/800G DC Optics? DCI vs Intra-DC

**Structured inventory (P1 high confidence 90-95%):**

| Product | Spec Evidence | DC Relevance | Source Publisher Date Retrieved |
|---|---|---|---|
| OptiXtrans DC908 | 2RU, 12.8T max 4.8T/RU, 8 svc slots, line 100-800G prog, 48T/fiber C-ext, CFP2-DCO, client GE-400GE OTU4 FC SFP+ QSFP28/DD | High – DCI enterprise | https://e.huawei.com/au/products/optical-transmission/dc908 + GlobalData blog 2019-09-29 launch |
| OptiXtrans DC908 Pro | 19.2T/2U, 1.2T single-wave to 240km, 96T/fiber C120+L120, single-slot XC 2.4T, ROADM integrated | Very High – AI era DCI | PRNewswire/Huawei Enterprise search, e.huawei spec table, Sept 2023 inferred PR |
| OSN 9800 M12/M24/U32E/U64E/P32/P32C | 64T XC exp 100T, 800G/wave, 96T/fiber Super C+L, SDH/Eth/SAN/OTN/Video services | Medium-High – Carrier backbone hub DCI | https://e.huawei.com/... OSN 9800 OTN |
| OSN 9800 K36/K12 | K36 large cap, front-to-rear airflow DC-oriented, 400G OTN, energy efficient | Very High – DC-oriented OTN | e.huawei.com K series spec page active 2023-2024, id 2719093 search |
| E6600/E9600/E9624X | MS-OTN PCM/PDH/SDH/PKT/OTN/fgOTN, first fgOTN MWC 2024 | Medium – Enterprise industry DCI campus | https://e.huawei.com solutions DCI 5375074706005167959 |
| 800G Tunable Ultra-High-Speed Module | 200-800G tunable, 48T/fiber, CMS algo, 7th gen oDSP, +20% distance | Core line tech backbone/metro/DCI | https://www.huawei.com/en/news/2020/2/800g-tunable-ultra-high-speed-optical-module 2020-02-26 |
| QSFP-DD-400G-SR8/VR4/DR4/FR4/SR4 | 850nm 100m OM4 MPO16, 50m VR4 OM4 OM5 MPO12, 1311nm 500m DR4 MPO12, 2km FR4 duplex | High – intra-DC spine-leaf | Huawei Data Center Short-Distance Optical Module Portfolio PDF 2025 copyright + individual datasheets 2025 |
| QSFP112-400G-VR4/SR4/DR4/FR4 | Same reach next-gen lower power | High – 800G switch breakout | Same portfolio PDF 2025 |
| OSFP-400G-SR4/DR4 | OSFP 400G | High – CloudEngine switches | Same PDF |
| OSFP-800G-SR8-D | 850nm 0.1km MPO16 8x100G PAM4 | Very High – 800G intra-DC AI cluster | Portfolio PDF line: OSFP,800G, Multimode module (850nm,0.1km, MPO 1x16 APC) 2025 |
| OSFP-800G-2FR4-D | Dual LC 2x400G FR4 2km SMF | Very High – intra-DC/DCI edge breakout | Same PDF + product page |

**Finding:** YES Huawei makes 400G/800G grey optics for intra-DC (50m-2km) – confirmed via official enterprise PDFs – but **not merchant competitive in Western hyperscaler supply** – use confined to Huawei CloudEngine ecosystem + client side of DC908. No standard QSFP-DD-400G-ZR/ZR+ coherent pluggable retail catalog; strategy platform CFP2-DCO/MSA proprietary, not ZR.

**DCI vs intra-DC distinction (P2, R1):**
- **Long-haul backbone >1000km**: Coherent QPSK/16QAM/64QAM, EDFA/Raman, OTN OSN 9800 claims 200G 5000km no regen – Strong leader, low datacom relevance.
- **Metro 80-600km**: Coherent 100/200/400G DWDM MS-OTN E9600, fgOTN – Strong APAC/EMEA carriers, medium DCI relevance overkill intra-DC.
- **DCI 10-120km enterprise**: DC908 chassis leader GlobalData – strong but buyers Tier2 colocation/finance not hyperscale.
- **DCI hyperscale IPoDWDM 80km+**: 400G ZR/ZR+ coherent QSFP-DD – Huawei July 2025 backbone-focused press (LinkedIn scrape) not proven hyperscaler router insertion vs Acacia/Marvell.
- **Intra-DC 0-2km AI leaf-spine**: PAM4 DR/FR 400G/800G OSFP/QSFP-DD IM-DD – NO competitive portfolio proof; focus OCS.
- **AI back-end OCS fabric DC808**: MEMS OXC, transparent rate/protocol, no O/E, 98% port power reduction, 20% network energy reduction, PoD granularity vs Google Apollo – award-winning concept but deployment unproven scale.

**Interpretation:** Telecom strength gives strong DCI story for carriers/enterprises outside US, does NOT translate automatically to intra-DC due to design centers: ultra-low cost PAM4 (<12W, $100-300 vs 15-25W @400G coherent per EDGE Optical), massive scale, Broadcom Tomahawk ASIC integration, hyperscaler operational model Huawei excluded by geopolitics + PAM4 portfolio gap.

### Regulatory & Market Access – BIS Entity List May 16 2019 FDPR May15/Aug17 2020 FCC 22-84 Nov25 2022 NDAA 889, UK/EU

**Timeline validated primary (P3/R2 high confidence 95%):**

| Regulation | Date | Scope | Impact DC Optical | Jurisdiction | Source URL |
|---|---|---|---|---|---|
| BIS Entity List addition | 2019-05-16 16:15 EDT Federal Register May 21 2019 | Huawei +68 affiliates all EAR items license presumption denial TGL May20-Aug19 2019 | US firms cannot supply chips/tools/EDA to Huawei for optical modules | US BIS | Lexology 116d5a8c-c781-4f06-948f-74965ef7e768 |
| FDPR amendment | 2020-05-15 | Non-US direct product of ECCNs 3D001/3E001 produced/developed by Huawei need license | Blocks TSMC HiSilicon DSP | US BIS | Lexology 1628ea1b-3799-4156-989c-f36e39f93aaa |
| FDPR final + 38 affiliates TGL expiry | 2020-08-17 savings ship Sep14 | Any subject product knowledge Huawei party or incorporated needs license, captures COTS EDA test | Captures COTS optical components, halts SMIC without US tech (impossible) | US BIS | same |
| NDAA Sec 889(a)(1)(A) | 2019-08-13 Part A | Federal agencies cannot procure covered telecom as substantial component | Federal DC cannot buy Huawei optical | US Congress FAR | congress.gov |
| NDAA 889(a)(1)(B) | 2020-08-13 Part B | Agencies cannot contract with entity USES covered eq anywhere | Hyperscalers as federal contractors must certify no Huawei anywhere global ban | US Congress FAR | same |
| Secure Equipment Act | 2021-11-11 signed Pub L117-18 | Instruct FCC adopt rules no later than Nov11 2022 | Enables FCC ban | US law | Wiley law summary |
| FCC Covered List first | 2021-03-12 | Lists Huawei ZTE Hytera Hikvision Dahua | Basis Equipment Act | FCC | FCC 22-84 doc |
| FCC 22-84 Final Rules | 2022-11-25 unanimous Rosenworcel statement | Prohibits Certification+SDoC + import/marketing Covered List telecom eq (Huawei ZTE all telecom) existing not auto revoked | No new Huawei optical modules telecom class including 400ZR SDoC can be authorized/imported/marketed US | FCC | https://docs.fcc.gov/public/attachments/FCC-22-84A2.pdf |
| UK NCSC + Supply Chain Review | 2020-07-14 decision purchase ban Dec31 2020 (ext Dec31 2021) removal 2027 end | Huawei high-risk vendor direction ban 5G + full fibre core removal | UK operators must remove OptiX from 5G/core transport enterprise DCI prohibited for operators | UK DCMS NCSC | gov.uk + networkworld |
| EU 5G Toolbox | 2020-01-29 voluntary | Exclude high-risk vendors radio/core diversification | No direct ban DC optics private DC allowed but hyperscalers exclude globally | EU Commission | livemint/IEEE Spectrum 11/27 Aug2024 |
| EU proposed mandatory | 2024-2025 examinations Bloomberg Nov2025 | Commission examining mandatory phase-out | Would extend to optical transport telco DCs not intra-DC private | EU | livemint search |
| HiSilicon process limitation via ASML EUV ban | 2020-present | No EUV to China SMIC DUV ArF immersion multi-patterning | HiSilicon DSP stuck ~7nm effective no 5nm/3nm until domestic EUV breakthrough | US/NL controls | Tom's Hardware Wccftech aInvest |

**Hyperscaler procurement:** EAR direction restricts exports TO Huawei not FROM, but FCC marketing/import ban + NDAA 889 B federal contractor compliance + ICTS supply chain rule 2021 (EO14028) + internal supply chain security – effectively zero. No evidence US hyperscalers purchase Huawei optics; industry analyses show Innolight, Coherent, Lumentum, Acacia/Cisco, Marvell/Inphi, Broadcom.

**HiSilicon node:** After FDPR TSMC stopped, SMIC no EUV, DUV N+1/N+2 7nm equiv via multi-patterning Kirin 9000S Mate60 Pro 2023 teardown, low yields; 5nm DUV quad patterning +40-50% cost 1/3 yield worse power per Winbuzzer/aInvest/WCCFTech – forces Huawei power-tolerant metro, LPO workaround for DC.

### Market Share – DellOro 33% 2024 Optical Transport Leader, LightCounting Transceiver #3 2023, DCI vs LH

**Dell'Oro Optical Transport (P4/R2):**
- FY2024: Huawei 33% revenue (ElectronicsWeekly Feb 2025 summarizing Dell'Oro 7719664090174722548), Ciena 19%, Nokia+Infinera ~19% if merged, Huawei +3pp 2024, +4pp first 9M, +6pp YoY 2Q24 despite market -19% YoY – China-driven counter-cyclical.
- 2021-2023 baseline: 1Q21 Huawei Ciena ZTE Nokia trailing, 2022-23 trailing Huawei Ciena Nokia ZTE Infinera FiberHome ~$15B 2021 -2% (5007062693931893151).
- Overall telecom equipment: 31% 2020 (TechBlog), 28% 2019, 30% 2023 vs 28% 2022 (SDxCentral via Dell'Oro).
- Bitrate: Omdia via LightReading Huawei #1 100G 21%, 200G 38%, 400G 31.2-35% four quarters leader – high-speed coherent strength (3180452833449176973).
- Ex-APAC/ROW: Top ex-APAC Ciena #1, Huawei top5 combined top5 ~80% (Ciena Huawei Nokia Infinera Cisco/Fujitsu) – Huawei ROW single-digit NA.
- China vs ROW inference: Global 33% implies >50% China share (China ~40-45% global optical spend) + single-digit ROW. China +8% YoY 1Q24 while other regions declined.
- Methodology change 2022: Include internal manufacturing spurred Huawei top10 entry.
- Total transceiver: 2022 – Innolight, Coherent, Cisco, Huawei >50% ~$1.4B each top2; 2023 – #1 Innolight (~$1.5B), #2 Coherent, #3 Huawei.
- DCI: Global top3 Dell'Oro DCI Ciena Nokia Cisco NA concentrated. Huawei ex-NA all-optical DCI 32.6% (Omdia 2019 calc) largest non-NA but NA declined -2.1% – strong non-NA, weak NA where spend dominates. So LH/metro >30%, DCI global <15% <5% NA.

**Interpretation:** Huawei global optical transport leader inflated by China dominance; in Western AI DC relevant segments (hyperscale DCI + intra-DC) share minimal.

### Competitive Landscape DC Optical 800G/1.6T – Innolight #1 Coherent #1 800G Q3 24 Eoptolink top3 Nvidia Chain PAM4 DSP Marvell 60-70% vs Broadcom timeline to 1.6T

**Leadership (P5/R3):**
- LightCounting 2023 total: 1 Innolight, 2 Coherent, 3 Huawei, 4 Cisco, 5 Accelink, 6 Hisense, 7 Eoptolink, 8 HGGenuine, 9 Source Photonics, 10 Marvell.
- Cignal AI 2024: 400G leader Innolight, 800G leader Coherent (took top 3Q24 after Innolight), Nvidia Fabrinet third. 4Q24 report: largest datacom module suppliers Innolight/Coherent/Eoptolink, largest component sources Coherent/Broadcom/Lumentum. LITE telecom leader 2024, #2 component behind Coherent, ~40% Western 800G with Coherent (AnandCapital), 18.7% non-GAAP op margin late 2025.
- Nvidia AI: Innolight+Eoptolink ~60% incremental 800G orders GB200/H100, ~40% Coherent/Lumentum/Broadcom. 576-GPU GB200 = 18,432x 800G DR8 single-end (3832794605794927700).
- Market size: Datacom components $9B 2024 → ~$12B 2026 Cignal or $16B+ 2025 including 400G/800G, 20M+ 400G+800G units 2024 Cignal +4x YoY, SR8 >3M 2024 LightCounting via Lessengers.
- Innolight $3.3B 2024 revenue ~35% global 800G (momoview secondary), >500k 400G H1 2024, 800G ZR with Marvell Orion.
- Eoptolink #7 2023 → top3 2024 $1.2B, fastest capacity expansion targeting AWS, 500k/month plan 2026 combined 800G+1.6T.
- Coherent ~22% total market momoview, #1 800G 3Q24.
- Lumentum not top3 module volume but EML depth moat.
- Cisco/Acacia #4 overall 2023 coherent port 10k 400ZR/quarter run rate new 800G pluggable 2024.

**Adoption speed:**
- 2020-2022 400G dominant Innolight first 800G ECOC 2020.
- 2023 800G ramp H100/HGX Google TPUs Nvidia new segment.
- 2024 explosive 20M+ 400G+800G SR8 >3M.
- 2025 800G fastest growing 60% (Cignal) 24M 800G+ units TrendForce 1.6T <1M early volume Broadcom shipped 1.6T Sept 2024 200G/lane.
- 2026-2027 800G+ >60% share 63M units 2.6x, transition 200G/lane PAM4 IEEE 802.3df ratified 2024 802.3dj mid-2026 OSFP same cage 1.6T/3.2T.
- 2028+ 3.2T over 8 lanes 400G/lane discussion CPO no material impact next 3 years Cignal May 2025.

**PAM4 DSP:**
- Marvell historically 60-70% revenue up to 80% shipment 50G-100G lane FS.com, $1.2B FY25 optical interconnect, Alaska C Spearfish + Orion coherent, Ara 3nm 200G/lane first 1.6T. 80%+ share hyperscaler AI per Investing.com note Google Nvidia. Broadcom #1 revenue 2025 per Dataintelo largest, Humboldt/Bayside families 100/400/800G 43% YoY Q1 2024 800G/1.6T launch, direct catch-up Taurus BCM83640 3nm 400G/lane Mar 2026 targeting 1.6T. Top5 68.4% of $4.2B PAM4 market 2025 Broadcom/Marvell/Credo/ADI/MACOM. Marvell both competitor (#10 module) and enabler (Innolight 800G ZR uses Orion).

**Implication:** Huawei not in Nvidia supply chain – Innolight/Eoptolink/Coherent/Lumentum – so AI cluster optical volumes Western dominated by non-Huawei.

### China Ecosystem – Innolight/Eoptolink Independent vs Accelink/Hisense/HG Genuine State-Linked HiSilicon Captive Huawei Ties

| Supplier | Ownership/Link to Huawei | Role Global | Evidence Source |
|---|---|---|---|
| Zhongji Innolight (Innolight Technology) | No equity; Huawei ~5% revenue customer; supplies Huawei Cloud; independent US-funded origin Google Capital | Global #1 transceiver >40% 800G 50-70% 1.6T (Xinhua Jun2026), $10.7B CNY 2023 revenue, 6,166 employees 2023 SZSE:300308, DoD 1260H list Oct2025/Jun2026 | Wikipedia/Crunchbase 2008 founded Osa Mok, DigitalToday 2025 customer split Alphabet 22% Amazon11% Meta6.4% Huawei5% Alibaba1.9%, Xinhua Silk Road, LightCounting Jan2023 Chinese dominance |
| Eoptolink | No equity; indirect Huawei customer rival same cluster private 2008 Chengdu | Global #2 800G Gen2 1.6T OSFP multicore fiber OFC2025 Nvidia ecosystem | Wikipedia Apr15 2008, SimplyWallSt SZSE:300502, LightCounting |
| Accelink | Quasi-SOE lineage WRI FiberHome 36.4% (Interfax2011), origin Solid-state Device Inst P&T Ministry1976 founded2001 listed2009 first Chinese optoelectronic public, merged WTD Dec2012 acquired IPX Denmark2013 JV Armayi, formal 2023 joint product dev Huawei Asia-Pac targeting 20% share + IEEE supporters list + 50GE China Mobile/Huawei/Accelink/Hisense | Top5 global 2016 telecom fronthaul DC | Interfax, GlobeNewswire, ResearchInChina, telecoms.com 50GE |
| Hisense Broadband / Ligent Inc | Municipal SOE origin Hisense Group Qingdao founded2003 JV Ligent Photonics US2002, consolidates Delaware USA PRWeb, 1k-5k employees | PON/GPON 1M for Verizon historically fronthaul | PRWeb formation, EDGE Optical Q&A |
| Huagong Tech/HG Genuine | No direct Huawei equity Wuhan Optics Valley HUST spin-off anchor listed, subsidiary HG Genuine second largest wholly China 2006 Luminent acquisition | 12.8T XPO first Mar2026 ByteDance sales surge 2021, localization | Luminent 2006 announcement, Xinhua, LightCounting ByteDance surge |
| Source Photonics | US HQ private strategic partnership Sep25 2018 PAM4 50G/400G showcase co-built Jintan fab with Economic Development Zone + Taiwan Hsinchu InP fab 2+ decades | InP laser fab PAM4 specialist | BusinessWire China Sep25 2018, compoundsemiconductor, optics.org Jintan |
| Huawei/HiSilicon Optoelectronics | Captive fully owned Caliopa 2013 EUR7m imec Ghent spin-off + CIP UK 2012 + imec Leuven + Canada Advanced Photonics + Wuhan Optics Valley process corner, products OM3538SX101 QSFP28 OM9380ZS100 80km OM3660FX102 QSFP-DD 400G FR4 etc patent US10560209B2 PAM4 | #3 global module 2021 captive self-supply OptiXtrans, not merchant EML supplier at scale high-end EML still import dominated Lumentum/Coherent/Sumitomo/Mitsubishi per finance.biggo domestic substitution gap | hisilicon.com product pages, LightReading Caliopa 2013 EUR7m 7769245939241242162, PIC Magazine 115352 Eric Bernier 100T CPO |

**BAT + ByteDance procurement:** Alibaba historically 1.9% Innolight small vs US. Tencent/Baidu less disclosed but domestic supply chain Accelink/Hisense/Eoptolink/Huagong. ByteDance driver HG Genuine 2021 surge. Huawei Cloud second/third China cloud uses Ascend AI servers prefers Accelink/Hisense/Huagong/HiSilicon own for full stack independence Entity List (P8 medium confidence 60-70% due limited public contracts). Trend: Domestic substitution >90% upstream localization YRD/PRD/Wuhan per Xinhua June2026, finance.biggo high-end EML low domestic historically Everbright Photonics first laser chip surge 300%, customs Mar2026 export price +23% YoY modules fiber/cable +263.8% value, government work report 2025-26 compute-power coordination first time prompting domestic fab expansion.

**Western impact:** 7/10 top10 vendors Chinese >60% global share 2026, cost manufacturing scale, forced exit Western transceiver vendors AOI divested 2022, remaining Western leaders DWDM coherent Cisco Acacia/Marvell Inphi.

### Tech Roadmap – Silicon Photonics Caliopa/CIP, TFLN 400G/lane 110GHz ECOC 2025, CPO vs LPO Huawei LPO Tactical, Western CPO Broadcom Bailly 51.2T Shipping Meta, Nvidia Spectrum-X 409.6T, Process Node Limits SMIC 7nm

**Huawei timeline:**
- Acquisitions: CIP Centre for Integrated Photonics UK from EEDA 2012, Caliopa NV Belgium 2013 EUR7m imec/Ghent spin-off → HiSilicon Advanced Photonics (Ghent + Canadian lab CEO Eric Bernier interview PIC Magazine 115352).
- imec Leuven joint R&D continuous, Politecnico di Milano OFC 2024 Th1F.1 papers.
- 2020: first 800G tunable coherent 200-800G channel-matched shaping CMS +20% distance 48T/fiber 7th-gen oDSP.
- F5.5G roadmap: 400G/800G OTN backbone 50G PON 10Gbps everywhere ETSI Sep2022 whitepaper (545914593081443543).
- 2024-2025 LPO: using 800G OSFP LPO up to 18 per XPU scale-up per LightCounting snippet 2382824564275216101 – brute-force vs CPO – and StarryLink 800G SR8 datasheet.
- ECOC 2025 last-day TFLN-on-Si demo per Gazettabyte 5596893651799700275: 110GHz EO BW MZM, integrated high-speed Ge PDs, low-loss edge fiber attach, reliability data first public near ready for pluggables and scale-out CPO. Wuhan Optics Valley process corner for TFLN.
- IPAC external laser source ELS form factor/system architecture projects targeting 100Tbps CPO.

**LPO vs CPO:**
- LPO removes DSP saves 50% power 7-9W→2-4W @400G (SemiEngineering 2024 Dell Tech), retains pluggable, but requires pristine channel short reach host SerDes strong FEC/equalization. Google/Meta rigorous OFC analysis problematic hyperscale variance. Hype waning ECOC 2025 per Chris Cole. Huawei controlled environment own switches/XPU tolerable.
- CPO integrates optical engine mm from ASIC eliminates PCB traces ~3pJ/bit ~5W per engine vs 10-16W pluggable 70% lower power Broadcom Bailly, 8x area efficiency. Hurdles: external laser reliability remote module, fiber attach density 1024 fibers 51.2T SN-MT connectors, serviceability whole switch replacement hours, thermal crosstalk.

**Western:**
- Broadcom: 2019 program Alexis Black Near Margalit, Bailly 51.2T CPO Tomahawk5 + 8x 6.4T SCIP SiPh chiplets 2024 shipping volume with Micas to Meta >30% switch power saving >4M link flap-free hrs (4571941553342458688), Davisson 102.4Tbps Tomahawk6 3nm 512x200G, 4th gen doubling per channel 400G/lane goal SiGe EIC then CMOS.
- Nvidia: Spectrum-X Photonics Ethernet 200G SerDes CPO 512x800G = 409.6Tb/s, Quantum-X Photonics InfiniBand 800Tb/s 1.6T/port 5x power 10x resiliency NCCL improvements 11 partners TSMC/Corning/Foxconn available H2 2026 production Q1 2025 per some reports (2992356792197262673), scale-up NVL576 CPO + co-packaged copper.
- Marvell: Custom CPO for AI accelerators 6.4T engines $36B invested Inphi etc collaboration Nvidia 200G/400G PAM4.
- Intel: 100G SiPh 5y 400G commercial roadmap 800G licensing to Source Photonics/Jabil+MaxLinear 5nm 800G PAM4 DSP on Intel platform (1560959103736080375) focus OCI optical compute interconnect.

**Lead/Lag:** Huawei lead early 800G tunable coherent 2020 vs Western 2021-22, owns oDSP CMS vertically, TFLN-on-Si 400G/lane ECOC2025 at par/ahead in high-baud modulators needed 1.6T/3.2T. Lag: DC Ethernet CPO switch commercial deployment shipping already Western, Huawei research target 100T still LPO-based, advanced CMOS blocked sanctions reliance Wuhan TFLN not TSMC.

**Technical barriers:** DSP node 1.6T needs 3nm to meet power 5nm 800G; sanctions prevent HiSilicon TSMC 3/5nm – LPO attractive workaround. EML tight supply global China dependence Lumentum/Coherent, SiPh high-speed MZM Si or TFLN alternative reduces EML reliance explains Huawei TFLN push.

---

## LITE Specific Analysis

**Segments & Financials – high confidence 90-95% primary SEC 10-K FY2025 Aug19 2025 filing lite-20250628.htm:**
- Reorg Q1 FY2024 from OpComms/Commercial Lasers to Cloud & Networking + Industrial Tech.
- FY2025 Total $1,645.0M +21% YoY (FY2024 $1,359.2M FY2023 $1,767.0M):
  - Cloud & Networking $1,410.8M 85.8% (FY2024 $1,084.9M FY2023 $1,322.5M)
  - Industrial Tech $234.2M 14.2% (FY2024 $274.3M FY2023 $444.5M)
- Gross margin 28.0% FY2025 vs 18.5% FY2024 improvement due $20.8M reduction excess obsolete charges related US trade restrictions (Huawei proxy) + lower Cloud Light acquisition costs lower excess capacity.
- Segment profit Cloud & Networking $264.5M +112% YoY Industrial Tech $12.1M -51.8%.
- Products: EMLs 50-60% high-end market share per Capital Blueprint/FinancialContent medium confidence – bottleneck InP fab rare in-house, 100G-per-lane EML record shipments Q1 FY2026 remarks Michael Hurlston, ramping 200G-per-lane for 1.6T. DMLs, CW lasers pump narrow-linewidth detectors. CW lasers for silicon photonics ELS initiated deliveries for 800G transceiver manufacturers milestone Q1 FY2026, plus ultra-high-power for CPO scaling Q4 2025 volumes. Transceivers via Cloud Light $750M acquisition 2023 $200M+ revenue base >90% 400G+ >50% 800G last quarter, active optical cables. OCS for AI clusters backlog >$400M Q2 FY2026, growth 65% Q2 58% Q1 56% Q4 FY2025.
- Roadmap: AI infrastructure OCS + transceivers ~25% today → projected 60% two years Morningstar Investor Day target 40% op margin vs ~18.7% non-GAAP op margin late 2025 thesis.

**Customers – concentration:**
- 10-K anon: Customer A 16.0% Customer B 15.4% FY2025, A 11.4% B 18.9% FY2024, A 15.3% C12.1% D10.5% FY2023. Historical identities (2017-2020 filings + press): Accelink, Alphabet/Google, Apple ~21% FY2018 VCSEL, Arista, Ciena 13-20%, Cisco 11-13%, Huawei 11% FY2018 15% YTD FY2019 18% Q3 FY2019, Microsoft, Nokia. Risk flag limited number no long-term volume VMI MFN pricing.

**China/Huawei exposure:**
| Dimension | Value | Source |
|---|---|---|
| Direct Huawei FY2025 | 0% halted after Entity List | 10-K risk factor cessation & TheStreet |
| Historical peak | 11% FY2018 15% YTD May2019 18% Q3 Mar30 2019 quarter | SEC + Lumentum 8-K May2019 |
| Impact 2019 ban | Guidance cut $405-425M → $375-390M -7.8% midpoint EPS 85c-$1 → 65-77c | Zacks press lite |
| Geographic International | 81% FY2025 73.8% FY2024 86.3% FY2023 | 10-K |
| Hong Kong shipment | $398.6M 24.2% FY2025 ($261.9M 19.3% FY2024 $246.7M 14% FY2023) | 10-K |
| Thailand | $291.8M 17.7% FY2025 $183.8M FY2024 | 10-K |
| US | $312.3M 19% FY2025 | 10-K |
| China-adjacent HK+Thailand+Other APAC | ~54% FY2025 | Derived |
| Historical China 20-25% down | Estimate FinanceCharts profile | Secondary |
| Manufacturing footprint | Closed 2 NeoPhotonics factories China Dec2023 per Q2 FY2024 CFO, sold Shenzhen assets Dec2024 closed Mar5 2025 proceeds $47.8M net cash $17.6M costs $1.1M carrying $12.9M gain $34.9M withholding tax $4.4M | 10-K Note |

**Pivot strategy:** Q2 FY2024 earnings Feb8 2024 CEO Alan Lowe response JP Morgan Samik Chatterjee: “deployments about to happen in China that we're part of through non-Huawei customers, and we're in a meaningful way, a very large part of the share of non-Huawei in China NEMs” – ZTE/FiberHome/Alcatel Shanghai Bell.

**Datacom DC opportunity vs risk:**
- Driver: Cloud & Networking growth $325.9M YoY $193.2M cloud AI/ML including Cloud Light full-year + $132.7M NEM recovery inventory normalization. AI OCS rapid scaling.
- Risk: hyperscaler capex concentration few customers, inventory cycle 2023-24 excess build-down $1,767M→$1,359M→rebound, InP capacity sold out through2026 booked 2027 via LTAs moat + constraint, tariff ~100bps GM Q3 FY2025 submounts/capacitors mitigation Thailand ongoing.

**EML/CW to transceiver makers:** YES LITE dominant upstream chip supplier to downstream module makers Innolight/Eoptolink plus internal Cloud Light modules simultaneous competing/partnering. EML 50-60% share claim medium confidence secondary but corroborated record shipments and 10-K commentary bottleneck. CW lasers milestone initiation Q1 FY2026 for 800G manufacturers + ultra-high-power CPO.

**Implication for Huawei risk:** Direct Huawei customer loss absorbed already 2019; remaining risk indirect via Chinese transceiver pricing wars compressing ASPs 10% YoY per Omdia ad-hoc-news 400G+ share sits 25% down 30% peak. Coherent SiPh push threatens InP. Yet tight 200G EML supply + second-sourcing risk 2027-28 double-edge. US manufacturing base mitigation vs Asia-heavy peers tailwind due Huawei excluded Western supply chains. >90x forward earnings priced perfection concentration Nvidia $2B backing link per SimplyWall – sensitive mix shift where hyperscalers concentrate 800G purchasing lower-cost Chinese suppliers while reserving 1.6T/OCS premium vendors.

### MRVL Specific Analysis

**PAM4 DSP leadership – high confidence product official marvell.com/solutions/data-center/optical-dsp.html:**
- History via Inphi acquisition Apr2021 $10B bring world-class optical DSP IP 7 generations.
- Market share 60-70% revenue historically up to 80% shipment 50G-100G per lane FS.com snippet medium need Cignal/LightCounting primary but consistent.
- Portfolio: Perseus/Porrima/Alcor/Polaris/Atlas 50G-1.6T 3nm-16nm, Spica 800G PAM4 7nm first production-ready Mar3 2022 8x100G/2x400G integrated drivers/TIA, Spica Gen2-T 5nm Tx-only low-power award Electronic Products 2024, Nova 5nm first 200G/lane 1.6T DSP 2023, Ara 3nm 1.6T platform 2024 Leading EDGE winner 2025.
- Customer validation Innolight – shipping 800G optics based on Spica CMO Osa Mok quotes 2022, ecosystem.

**Coherent portfolio:**
- Canopus/Deneb market-leading 400G coherent widely adopted carriers/clouds Innolight Molex – Deneb 100G QSFP-DD 5G backhaul.
- Orion 5th gen industry first 800G coherent DSP pluggable 5nm 2x bandwidth vs Deneb/Canopus 130+ GBd 112G SerDes PCS advanced FEC QSFP-DD/OSFP/CFP2.
- Libra first 2nm 800G ZR/ZR+ powering 2nd-gen COLORZ 800 MACsec 800G@1000km 600G@2000km 400G@3000km.
- Electra first 2nm 1.6T ZR/ZR+ powering COLORZ 1600 campus 20km metro 120km regional 1000km 1.6T.
- Aquila coherent-lite O-band 400G 16QAM 2-20km intra-campus bridging PAM4 coherent.

**COLORZ modules:**
- COLORZ 800 first family 800G ZR/ZR+ pluggable QSFP-DD/OSFP silicon photonics single die initially 500km @800G PCS extended 1000km full BW OpenZR+ OpenROADM C+L bands award Most Innovative Pluggable ECOC2024, 2nd-gen with Libra MACsec.
- COLORZ 1600 first 1.6T ZR/ZR+ pluggable MACsec Electra.

**Alaska family:** Alaska P PCIe Gen6/CXL PAM4 retimers extending signal reach + Alaska Ethernet PHY platform branding – part of electro-optics ecosystem 6.4T optical engines + LPO TIA laser driver chipset.

**Revenue & Data Center exposure – high confidence 85-90% earnings transcripts:**
- Data center end market 70-75% total: Q4 FY25 $1.37B 75% (Motley Fool Feb1 2025), FY26 full year $6.1B 74.4% (Zacks Mar26 2026), Q3 FY26 73% $1.52B, early FY24 only 33% – massive shift to AI.
- Electro-optics inside data center substantial base per CEO Matt Murphy Q1 FY25 complementary custom AI ramp, estimated $1B+ annual not broken out, part of ~$6B data center total.
- Trend AI deployment 800G then 1.6T.

**Customers – not Huawei:**
- Confirmed partners Innolight (Deneb/Canopus/Spica/Orion), Eoptolink preference neutral supplier not competing lasers vs Broadcom substack inference medium, Coherent company (module maker) Lumentum Coherent demo 500km ElectronicsWeekly Marvell Lumentum Coherent interoperability, Molex 400G coherent.
- Huawei NOT customer vertical integrated HiSilicon Entity List bans Marvell sales to Huawei.

**Huawei competition:**
- YES HiSilicon owns internal coherent + PAM4 DSP only Chinese capability: Wuhan factory CNY1.8B 10k R&D only company capable coherent optical DSP chipset per huaweicentral/droidverse, history 100G coherent internal Glen Fang quote “We only do the pieces such as the DSP”. PAM4 DSP internal consumption optical networking router line cards DC switches per dataintelo PAM4 DSP market report – not merchant.
- Next-gen: 2Tbps-capable coherent DSP Q1 2026 per Dell'Oro Mar2026 embedded muxponder 2x2T larger node vs industry 3nm but 0.1W/Gbps vs 0.125W industry avg via vertical integration tunable laser TIA driver InP modulator 30% lower parasitic capacitance own fab.
- Dynamic: China domestic Huawei leading coherent supplier regional competitor carrier inside China carriers use Huawei in-house DSP MRVL TAM inside China limited; PAM4 internal reduces MRVL addressable China DCs Tencent Baidu Alibaba maybe Huawei but foreign module makers still dominate; Huawei does NOT rely MRVL. So NOT direct competitor merchant PAM4 market cloud Google Meta MSFT buying Innolight etc because Huawei modules not sold US DC. Regional carrier competition China.

**Primary merchant competitors:** Broadcom Taurus BCM83640 monolithic 3nm 400G/lane targeting 1.6T announced Mar2026 direct catch-up, Acacia Cisco, Lumentum internal, Coherent internal but Marvell neutral merchant leader.

**China exposure – high confidence 90% primary filings:**
| Metric | Value | Source |
|---|---|---|
| Shipment dest China | $515.7M 39% Apr29 2023 $644M 45% Apr30 2022 | Marvell 10-Q May26 2023 |
| Disclaimer | Substantial majority shipments to China relate to sales to non-China based customers factories/contract manufacturing within China | Same 10-Q repeats Aug30 2024 May31 2024 Dec04 2024 |
| OEM customers China true | <10% total company revenue declined Q4 FY2023 Mar2 2023 Matt Murphy transcript | Motley Fool transcript |
| Q1 FY2025 >45% revenues came from China by shipment | Nasdaq/Zacks Jul2024 article tightening could severely impact | Secondary |
| <5% revenue from China regulatory uncertainty spooked – true PRC OEM? | AInvest Oct17 WSJ piece <5% vs shipment confusion | Secondary |
| Stock headline risk observed | -10.1% Jul17 2024 Bloomberg tightening report | Market data |

**Implication for Huawei risk:** Huawei internal DSP reduces China TAM but not US. Main risk headline volatility shipment vs OEM confusion + Broadcom catch-up. Benefit high-end optical DSP premium offsets custom ASIC margin dilution per SeekingAlpha Khaveen.

---

## Scenarios – Bullish vs Bearish

### Bullish – Sanctions Durable, Western Share Gain, Scarcity Rents GM>40%

**Drivers:** US BIS Entity List + FDPR Aug2020 remains, FCC 22-84 no revocation, NDAA 889 enforcement, UK removal 2027 completed, EU moves to mandatory phase-out 2026/2029 per Bloomberg Nov2025 examinations. Huawei stays on 7nm DUV limited power, no EUV breakthrough, TFLN 400G/lane ready but CPO commercial lag vs Broadcom Bailly/Davisson commercial + Nvidia Spectrum-X 409.6T production H2 2026. Chinese vendors Innolight/Eoptolink remain on DoD 1260H list June2026 but US hyperscalers continue buying via fab diversification (Thailand/Japan) and avoid grey channels – ASP compression limited to <5% YoY.

**Density:** 800G +60% 2025 24M units TrendForce, 63M 2026 2.6x, 800G+ >60% share by 2026 Google Ironwood 4M TPUs →6M+ modules, 1.6T <1M 2025 → volume 2026-27 IEEE 802.3dj mid2026.

**LITE outcome:** EML scarcity 50-60% share sold out through 2026 booked 2027 via LTAs sustains pricing power Hurlston quote record shipments path >40% GM. Cloud & Networking $500M quarterly exiting CY25 → $750M target (Morningstar Investor Day). OCS backlog >$400M converts, OCS/CPO/transceivers mix shift 25% today →60% two years. US manufacturing base mitigates tariff headwind 100bps. Revenue $1.41B FY25 86% Cloud → $2B+ FY27. Gross margin >40% long-term, operating margin 18.7% →25%+. Customer concentration risk remains but hyperscaler preference for US-made high-performance EML/CW reduces Chinese substitution. Huawei 0% direct, indirect non-Huawei NEM share in China stable. Stock acts as scarcity rent + CPO option.

**MRVL outcome:** Data center 70-75% $6.1B FY26 full year → $8-9B FY28 driven custom AI + electro-optics. PAM4 DSP 60-70% share retained despite Broadcom Taurus catch-up due neutral supplier moat (Innolight/Eoptolink prefer Marvell no laser competing). Spica/Nova/Ara design wins at Google/Meta/Amazon + COLORZ 800/1600 incremental metro DCI revenue $1.5B AI interconnect target beaten → $2.5B FY26. Gross margin corporate average lifted by premium optical DSP offsetting 5G/custom ASIC dilution per SeekingAlpha. True China OEM <10% limits fundamental regulatory hit; shipment domicile relocated partially to Thailand expansion. TAM share doubling 10% →20% of $75B per bullish thesis.

**Probability directional:** Regulatory shelter maintains high short-term, but requires monitoring DoD 1260H enforcement and Smic EUV progress.

### Bearish – Chinese Price Spillover 10% ASP Compression, Dual-Sourcing 2027-28 EML, DoD 1260H Listing Risk, Relocation Costs

**Drivers:** Chinese transceiver vendors 7/10 top10 >60% global share 2026 Xinhua + >40% 800G Innolight 50-70% 1.6T, localization >90% upstream, cost advantage scale domestic relationships, push to maintain utilization after China AI overbuild (ByteDance 100k+ GPUs 2025-26). If China domestic demand softens (5G CAPEX reduced, carrier inventory), vendors dump 400G/800G into international markets via non-US customers or gray channels, compressing global ASP 10-15% YoY already observed Omdia ad-hoc-news packaging LITE market share 400G+ 30% peak →25% →20% risk. Hyperscalers (Alphabet 22% of Innolight, Amazon 11%, Meta 6.4% per Bloomberg) already have dual-source economics incentive to allocate 800G volume to lower-cost Chinese suppliers while reserving 1.6T/OCS premium to Western suppliers – LITE components still inside competitor modules “we often win regardless who supplies module” Hurlston Q3 2025 cushions but not immunizes.

**Second-sourcing EML 2027-28:** Industry notes SeekingAlpha/SimplyWall tight 200G EML currently drives record shipments double end CY25 but second-sourcing risk 2027-28 likely pricing pressure volume share loss – Everbright Photonics first laser chip stock surge 300% domestic substitution narrative; Sumitomo/Mitsubishi Japanese also expanding; silicon photonics TFLN 400G/lane (Huawei demo) could replace high-power EML in LPO/CPO reducing LITE content.

**DoD 1260H:** Inclusion Innolight/Eoptolink Oct2025/June2026 (Bloomberg/Reuters Wikipedia citations + FDD note Est control over 60% market) could force Alphabet/Amazon/Meta/Google to diversify BOM to US vendors – positive for LITE short-term but creates supply disruption, inventory write-down if abrupt. Conversely if US expands Covered List to optical transceiver vendors, LITE/MRVL contract manufacturing China exposure headline hits.

**Relocation costs:** LITE tariff headwind ~100bps GM Q3 2025 submounts/capacitors moving China → Thailand, NeoPhotonics factories closed Dec2023, Shenzhen asset sale Mar2025 gain $34.9M but operational transition cost, validation. MRVL 39-45% shipment to China headline despite <10% true OEM creates equity volatility -10.1% July17 2024 + 8-15% tariff headlines observed Jul2024, may need relocate contract manufacturing increasing cost disrupting supply chain grounding optimistic 31.5% revenue growth consensus 2025.

**LITE outcome:** ASP -10% YoY continues, share loss 30%→20%, gross margin stall sub-35% despite path 40%, Industrial Tech $234.2M declining -51.8% segment profit tech imaging/sensing competition offset insufficient. Inventory cycle repeat $1,767M FY2023 → $1,359M FY2024 drop. Valuation >90x forward earnings priced perfection collapses on miss. Industrial China soft 5G delays.

**MRVL outcome:** China mandates domestic DSP/optical for BAT/ByteDance/Huawei Cloud, China design wins lost reduces TAM; US tightens destination shipment export control requires relocation out of China increasing cost logistics. Broadcom 3nm Taurus BCM83640 catches up 1.6T design wins eroding PAM4 share 60-70% →40-50%. Custom AI margin dilution persists GM challenged. Stock reacts to regulatory headlines more than fundamentals but eventually supply chain realignment cost hits cash flow.

**Probability directional:** Medium-term 2027-28 second-sourcing and price spillover credible; near-term scarcity protects.

---

## Investment Implications for $LITE, $MRVL – Risk Factor Assessment

### Question: Is Huawei a risk factor for optical in data center?

**Structured answer with confidence:**

| Layer | Assessment | Evidence | Confidence |
|---|---|---|---|
| Direct competition – Western AI DC intra-DC 0-2km PAM4 + DCI ZR | NO – Huawei tech excluded by regulation, zero hyperscaler adoption, portfolio confined to Huawei CloudEngine + enterprise DCI | FCC 22-84 Nov25 2022 + NDAA 889 B + no evidence US hyperscaler buying + Cignal AI top3 800G Innolight/Coherent/Eoptolink absent Huawei | 90% |
| Direct competition – China DC | YES – Huawei + HiSilicon leads China telecom transport 33% global inflated by >50% China share + DCI ex-NA 32.6% + HiSilicon only Chinese coherent DSP capability + 2Tbps DSP 2026 | Dell'Oro + Omdia + Lightwave + P7 Dell'Oro Mar2026 2Tbps | 85% |
| Indirect risk – price spillover global ASP | YES 10% YoY compression observed Omdia, Chinese vendors >60% global share cost advantage overbuild dump via non-US customers gray channels | Ad-hoc-news Omdia + R3 Chinese 7/10 >60% + P10 analyst | 75% |
| Indirect risk – supply chain relocation / headline volatility | YES LITE Hong Kong 24.2% + Thailand 17.7% + Other APAC 12.2% =54% China-adjacent shipment; MRVL 39-45% shipment to China vs <10% OEM true; DoD 1260H Innolight/Eoptolink + tariff 100bps GM | SEC filings + Marvell 10-Q + July17 2024 -10.1% | 85% |
| Indirect risk – technology fork CPO/LPO vs EML moat | Medium – Huawei TFLN 110GHz 400G/lane ready ECOC2025 reduces EML reliance + LPO workaround bypasses DSP node block could enable low-cost domestic 1.6T; LITE CW laser for 800G SiPh + ultra-high-power CPO mitigant; MRVL neutral DSP vs Broadcom catch-up | Gazettabyte + LightCounting LPO + LITE Q1 FY26 CW deliveries + Broadcom Taurus | 70% |
| Regulatory shelter benefit | YES short-term +5-10% pricing power scarcity rents GM>40% path, Western 40% 800G share Coherent+LITE | Hurlston record EML + Cignal 40% Western + MRVL 60-70% DSP | 80% |

**Conclusion:** Huawei is **not a direct competitive threat to LITE/MRVL inside Western AI data centers** today; it is an **indirect structural risk factor** via its ecosystem proxies (state-linked Accelink/Hisense/HG Genuine/Source) and independent but China-centric cost leaders Innolight/Eoptolink that dominate global supply, create pricing pressure, and increase headline regulatory risk. It simultaneously **creates regulatory shelter** that benefits LITE/MRVL pricing power short-term. This dual nature explains contradictory investor narratives (benefit vs risk) in P10 analyst commentary.

### $LITE Specific

**Bull triggers:**
- EML scarcity 50-60% high-end market sold out through 2026 via LTAs booked 2027 (P4/R3).
- CW laser deliveries for 800G SiPh initiated Q1 FY2026 milestone (P6/R4).
- OCS backlog >$400M Q2 FY2026 + CPO ultra-high-power lasers scaling Q4 2025 (P6).
- Cloud & Networking 86% $1.41B FY2025 + AI OCS growth 65% YoY Q2 FY2026.
- US manufacturing base mitigates China risk unlike Asia-heavy peers tailwind Huawei excluded.
- Nvidia $2B backing linked capacity access per SimplyWall note (secondary) – upside transceiver revenue.

**Bear triggers:**
- Hong Kong 24.2% shipment proxy China exposure + Thai hub cost + Shenzhen divestiture gain one-off.
- Customer concentration 2 >15% anon hyperscaler/NEM; no long-term volume VMI MFN.
- Pricing wars -10% YoY ASP Omdia; share 30%→25% peak; second-sourcing 200G EML 2027-28 Everbright Photonics domestic substitution + Japanese expansion + TFLN SiPh threatening InP.
- Industrial Tech 14% declining $234.2M -51.8% profit offset.
- Valuation >90x forward earnings priced perfection; mix shift 800G volume allocated lower-cost Chinese suppliers while 1.6T/OCS premium reserved Western – reduces LITE leverage if not qualified for 1.6T.
- DoD 1260H Innolight/Eoptolink if enforced abrupt BOM shift causes inventory fab mismatch.

**Scenario math illustrative:**
- FY2025 Cloud $1,410.8M. If EML price +15% scarcity → GM 28% →35% → $197M incremental gross profit. If share loss -5pp + ASP -10% → -$141M revenue downside (10% of Cloud). Net bull still positive near-term.
- Long-term 2027-28 second-sourcing 200G EML could erode 10-15% pricing power ~$150-200M revenue at risk if Chinese EML qualifies at Google/Meta.

**Investment phrasing:** LITE structurally benefits from Huawei regulatory exclusion – “we often win regardless who supplies module” but not immune to China cost chain pricing spillover. Huawei 0% direct customer vs 11-18% pre-2019 cut now removed; remaining China exposure via non-Huawei NEMs (ZTE FiberHome) meaningful + large part non-Huawei share – resilient. Near-term scarcity rent sustains bullish thesis, medium-term dual-sourcing risk underwrites bearish hedge.

### $MRVL Specific

**Bull triggers:**
- Data center 70-75% $6.1B FY26 (Zacks) – custom AI chip + electro-optics base substantial → differentiated from telecom optical.
- PAM4 DSP 60-70% revenue historically up to 80% shipment moat Spica first production-ready 800G Mar3 2022, Nova first 200G/lane 5nm 2023, Ara 3nm 2024 Leading EDGE winner, Orion first 800G coherent 5nm 130+GBd, Electra/Libra 2nm 1.6T/800G ZR first – enabler not competitor.
- Customers Innolight/Eoptolink/Coherent/Lumentum – all non-Huawei – Nvidia chain 60% incremental orders shared Innolight+Eoptolink – MRVL inside their modules.
- True China OEM <10% (Matt Murphy Q4 FY2023 transcript Mar2 2023) vs shipment headline 39-45% – fundamental risk lower than stock reaction suggests.
- Premium high-end optical DSP offset custom ASIC margin dilution per SeekingAlpha Khaveen.
- COLORZ 800/1600 incremental metro DCI revenue 500-1000km saves carrier capex.

**Bear triggers:**
- Broadcom BCM83640 monolithic 3nm 400G/lane Taurus Mar2026 targeting 1.6T – direct catch-up could erode 60-70% share to 40-50% – 43% YoY datacenter growth Broadcom Q1 2024 evidence.
- Accelink/Hisense/HG Genuine internal HiSilicon PAM4 reduces China TAM for MRVL – China AI cluster domestication BAT/ByteDance + Huawei Cloud full stack independence mandate reduces foreign DSP.
- Shipment domicile China 39-45% creates regulatory headline volatility -10.1% Jul17 2024 Bloomberg tightening report + 31.5% revenue growth consensus grounding risk if forced relocation.
- Smic 7nm limitation explains Huawei LPO workaround but also means Huawei 2Tbps DSP at larger node 0.1W/Gbps claim via vertical integration could eventually compete in non-Western markets, reducing MRVL TAM inside China.
- Custom silicon concentration similar to LITE – data center revenue highly linked to cloud capex cycles.

**Scenario math:** Electro-optics $1B+ annual – if DSP share drops -20pp to 40% share at same ASP → -$300-400M revenue at risk; if AI interconnect demand +60% 2025 24M →63M 2026 units TrendForce + Google Ironwood → +2x volume offsets pricing pressure.

**Investment phrasing:** MRVL not exposed to Huawei as customer/competitor in Western AI DC; Huawei vertical integration own DSP insulated but constrained; true China OEM <10% limits fundamental impact; however headline risk + Broadcom catch-up + Chinese internal DSP for domestic market creates indirect risk. Regulatory shelter benefits high-end DSP premium.

---

## Contradictions, Source Quality, Confidence Ranges

**Contradictions resolved:**
- P1 existence 400G/800G grey optics vs P2 claim no evidence competitive PAM4 portfolio – Resolved: capability exists as Huawei-branded internal ecosystem CloudEngine + DC908 client, but not merchant competitive Western open market – both true (R1).
- OSN 9800 K as “first DC-oriented OTN” suggesting DC focus vs architecture still OTN/telecom not Ethernet fabric – marketing vs technical (R1).
- FCC ban all Huawei telecom vs intra-DC <2km grey modules classification – FCC order says all telecom equipment for Huawei/ZTE, includes base station, closes SDoC loophole; intra-DC modules still Part15 unintentional radiators arguably telecom scope – likely covered but legal opinion needed (P3). Treat as covered.
- Huawei 35% global optical share vs Dell'Oro 33% – within year difference/error not contradiction (R2).
- Omdia bitrate shares vs Dell'Oro revenue – port vs revenue metrics different (R2).
- Huawei self-calc DCI 32.6% ex-NA vs Dell'Oro global DCI top3 excluding Huawei – strong non-NA weak NA global rank lower (R2).
- Dataintelo 14-16% Huawei optical vs Dell'Oro 33% – Dataintelo lower-quality secondary reject preferring Dell'Oro/Cignal primary (R2).
- Innolight independence vs DoD 1260H Chinese military linkage – not contradictory dual nature independent private but flagged military-linked per US gov assessment due control over 60% market (R3).
- Accelink private vs SOE – resolution: public listed but controlling shareholder state-linked FiberHome Technologies Group 36.4% WRI SASAC lineage quasi-SOE (R3).
- Source Photonics partnership vs ownership – press showcase not acquisition no equity transfer (R3).
- HiSilicon supply to Innolight/Eoptolink vs internal only – no evidence merchant chip supply high-end EML still import dominated (finance.biggo) – internal only conclusion high confidence (P8/P7).
- Market size $9B Cignal datacom components 2024 →$12B 2026 vs $16B+ 2025 TrendForce all 400G/800G growth – different inclusion definitions (R3).
- Innolight $3.3B 2024 revenue momoview vs LightCounting #1 2023 # – consistent magnitude secondary aggregation (R3).
- LITE China exposure 20-25% profile vs Hong Kong largest contributor could include China final demand routed via HK distributors – ambiguous but both proxies indicate significant China-adjacent shipment (R4).
- LITE US manufacturing mitigates China risk vs other sources tariff headwinds China components still – both true different aspects (R4).
- Alaska P retimer vs Alaska PHY naming conflation – both connectivity family (R4).

**Source Quality Grading:**

| Tier | Sources | Confidence Basis |
|---|---|---|
| High – Primary regulatory/legal filings | SEC 10-K FY2025 lite-20250628.htm, Marvell 10-Q May26 2023 shipment disclaimer, Marvell.com product pages Spica Mar3 2022 Orion press kit, Huawei e.huawei.com product pages + marketingcloud PDFs copyright2025 + 2020-02-26 press release, Gazettabyte ECOC2025 TFLN, PIC Magazine HiSilicon interview, FCC 22-84 PDF docs.fcc.gov, Lexology BIS Entity List analyses with Federal Register dates, Motley Fool Q4 FY2023 transcript <10% China OEM, TheStreet Huawei halt | Direct evidence provenance date publisher retrieval timestamp |
| Medium-High – Trade press primary republish | ElectronicsWeekly Dell'Oro Feb2025 summary FY2024 33%, LightwaveOnline LightCounting top10 2023, Cignal AI press releases Nov2024 800G 60% growth, Nvidia official Spectrum-X 409.6T 800Tb/s, Broadcom Bailly Davisson data sheets, BusinessWire Source Photonics Jintan partnership | Secondary trade but consistent cross-checked |
| Medium – Secondary synthesis + earnings call notes | TradingView 10-K summary, FinanceCharts China 20-25% profile, Ad-hoc-news Omdia ASP -10% share 25% down 30% peak, Zacks/Nasdaq China shipment articles, SimplyWallSt ownership FiberHome 36.4%, Xinhua Silk Road June2026 Innolight >40% 800G localization >90% bias note state media, Interfax 2011 WRI merger | Needs primary verification but directionally consistent |
| Low – Syndicated market reports marketing | Dataintelo Ciena 18-20% Huawei 14-16% conflicting, Momoview Innolight $3.3B revenue, Dataintelo PAM4 $4.2B market 68.4%, TrendForce 24M→63M unit forecast methodology unclear | Use caution, prefer primary, flag low confidence |

**Confidence Ranges (aggregated from reducers):**
- Huawei makes 400G/800G grey optics intra-DC: 90-95% high.
- DC908 existence 2019 GA + Pro Sept2023 + OSN K/M/U specs: 85-90% (date medium for Pro).
- No ZR pluggable retail strategy: 70%.
- Telecom focus not translating to hyperscale intra-DC: 85%.
- SiPh Caliopa/CIP history + TFLN ECOC2025 demo ready 400G/lane 110GHz: 80-85%.
- Western CPO shipping Broadcom 51.2T Meta 2024, Nvidia 409.6T H2 2026: 75% (vendor optimism).
- LPO usage 18 per XPU: 65% single LightCounting snippet.
- BIS timeline May 2019 May/Aug2020 FDPR: 95%.
- FCC ban Nov25 2022 no new authorizations: 95%.
- US hyperscalers zero Huawei market NDAA+FCC procurement: 90%.
- Huawei #1 optical transport 33% 2024: 85-90%.
- Huawei #3 transceiver 2023 due rule change: 80-85%.
- Weaker in DCI NA vs long-haul: 80%.
- HiSilicon constrained DUV 7nm no EUV: 80%.
- Innolight #1 2023/2024, Coherent #1 800G datacom 3Q24, Eoptolink top3 2024: 85-90%.
- Chinese vendors 7/10 top10 >60% global share: 85%.
- Innolight independence vs Huawei 5% exposure: 80%.
- Accelink state-owned via FiberHome/WRI: 80%.
- Only Chinese coherent DSP capability HiSilicon: 85% (P7).
- LITE Cloud & Networking 86% $1410.8M Huawei historical halt geography HK/T: 90-95%.
- EML dominance 50-60%: 70% secondary but corroborated.
- MRVL PAM4 60-70% historical leadership + data center 70-75%: 85-90%.
- MRVL true China OEM <10% vs 39-45% shipment: 90%.
- Huawei HiSilicon vertical not merchant only Chinese coherent DSP: 85%.
- Stock implications benefit Western share gain: 80%, risk price spillover: 65%, bull/bear scenarios: 60-70%.

---

## Unresolved Questions & Next-Wave / Blocked Layer-4 Recommendations

**Per coordinator_wave1.md Follow-Up Partitions promoted to Layer2 (requires wave2 if available):**

| Partition | Question | Why Promoted |
|---|---|---|
| P1a | Huawei CloudEngine 800G switch CE9860/CE9855 compatibility matrix for OSFP-800G modules power validation | Need exact power per 800G module FEC/breakout modes |
| P1b | Support.huawei.com datasheets OSFP-800G-SR8-D/2FR4-D power | Validate intra-DC power budget vs Broadcom Tomahawk |
| P2b | DCI vs intra-DC TAM model % Huawei addressable Western vs China | Quantify TAM at risk $B split |
| P3b | SMIC 7nm yield teardown impact Huawei 2Tbps DSP power claim 0.1W/Gbps vs 0.125W industry – Dell'Oro Mar2026 claim verification | Assess process node disadvantage offset vertical integration |
| P4b | Dell'Oro quarterly optical transport Excel 2020-2024 exact curve + China market size model to derive implied China share | Derive China >50% vs ROW single-digit with model |
| P4c | LightCounting segment split Huawei transceiver fronthaul/WDM/Ethernet 2020-2024 | Quantify datacom pure share vs captive |
| P5a-c | LITE EML capacity LTAs pricing power sold out 2026/2027, MRVL vs Broadcom DSP power design wins Google/Meta, 1.6T early wins | Competitive moat durability |
| P6b/P10b | Chinese vendor 800G pricing vs LITE dual-sourcing risk 2027-28 200G EML second-sourcing | Bear scenario timing |
| P7b | Marvell silicon photonics 3D engine vs Huawei InP modulator advantage | Tech moat vs process |
| P7c | Export control impact can Marvell sell PAM4 DSP to Chinese module makers Innolight China vs US entity does Huawei absence inflate US market share | Clarify MRVL China exposure granular |
| P8b | Innolight DoD 1260H listing impact US cloud BOM exclusion risk timeline Oct2025/June2026 – could benefit LITE/MRVL? | Regulatory upside/downside for LITE |
| P8c | HiSilicon external sale vs internal verification OM series captive vs merchant | Confirm HiSilicon not supplying peers |
| P9a | Patent search HiSilicon photonics 2020-2025 TFLN Espacenet | Quantify R&D lead/lag |

**Blocked Layer-4 Recommendations (needs operator approval beyond 3-layer cap):**
- L4-1: BOM teardown Huawei OSFP-800G-SR8 vs Innolight 800G DR8 cost – requires hardware purchase teardown.
- L4-2: CPO supply chain reliability laser failure rate model – requires vendor reliability data Meta field 4M flap-free hrs vs private.
- L4-3: FCC OET classification of QSFP-DD SR4 intentional vs unintentional radiator – need FCC counsel legal opinion for optical module FCC ID database pre-2022.
- L4-4: Detailed China vs Western transceiver pricing ASP teardown -$100-300 vs $1000 coherent – needs paywalled LightCounting component pricing.
- L4-5: TAM model 2024-2028 split customer Google Meta Amazon Microsoft ByteDance Huawei-exposed vs Western – needs paywalled Cignal AI Excel + LightCounting quarterly.

**Outstanding Gaps for Final Investment Decision:**
- Exact LITE 10-K Note 19 geographic table precise China vs Hong Kong vs Thailand vs US vs Mexico – TradingView snippet incomplete.
- LITE EML LTAs pricing terms booked through 2027 – supplier contract search paywalled.
- MRVL electro-optics revenue granularity exact $B inside data center segment not disclosed.
- Huawei OSFP-800G power consumption and FEC type for thermal comparison to Innolight DR8.

---

## Appendix: Handoff Index, Manifest, Discarded Branches, Memo Status

### Handoff Index – All Inputs Consumed

| ID | Role | File Path Under /workspace/handoffs | Transport Status | Merge/Discard Decision | Size Lines | Confidence Inputs |
|---|---|---|---|---|---|---|
| P1 | mapper Huawei portfolio | /workspace/handoffs/P1_huawei_optical_portfolio.md | Accepted via SendMessage recovered from /agent/inbox/nwt84sv | Merge | 130 lines | High – primary huawei.com |
| P2 | mapper telecom vs datacom | /workspace/handoffs/P2_telecom_vs_datacom.md | Accepted | Merge | 123 lines | High |
| P3 | mapper regulatory | /workspace/handoffs/P3_regulatory.md | Accepted + /agent/inbox/bltadfe duplicate | Merge | 125 lines | High – Lexology/FCC PDF |
| P4 | mapper market share | /workspace/handoffs/P4_market_share.md | Accepted + hq4ww8k | Merge | 108 lines | Medium-High – Dell'Oro via trade |
| P5 | mapper DC competitive | /workspace/handoffs/P5_dc_competitive.md | Accepted + frgif6j | Merge | 132 lines | High – Cignal/LightCounting |
| P6 | mapper LITE business | /workspace/handoffs/P6_LITE_business.md | Accepted recovered tvk2msx inbox | Merge | 165 lines | High – SEC 10-K primary |
| P7 | mapper MRVL optical | /workspace/handoffs/P7_MRVL_optical.md | Accepted + 8b9enna | Merge | 113 lines | High-Medium |
| P8 | mapper China ecosystem | /workspace/handoffs/P8_China_ecosystem.md | Accepted + pr2kxdb | Merge | 105 lines | Medium-High |
| P9 | mapper tech roadmap | /workspace/handoffs/P9_tech_roadmap.md | Accepted + cxdvjqs | Merge | 110 lines | Medium-High |
| P10 | mapper stock implications | /workspace/handoffs/P10_stock_implications.md (alias) + original /agent/inbox/hj1d3nn 16055B | Recovered inbox | Merge – create alias | 101 lines transcribed | Medium |
| R1 | reducer capability | /workspace/handoffs/reducer_R1.md + auto version | Manual creation + auto | Merge manual | 78 lines | High – aggregates P1+P2+P9 |
| R2 | reducer regulatory & market | /workspace/handoffs/reducer_R2.md + auto | Manual + auto | Merge manual | 88 lines | High |
| R3 | reducer competitive & ecosystem | /workspace/handoffs/reducer_R3.md + auto | Manual + auto | Merge manual | 90 lines | High |
| R4 | reducer LITE/MRVL impact | /workspace/handoffs/reducer_R4.md + auto | Manual + auto | Merge manual | 102 lines | High |
| coordinator_wave1 | coordinator | /workspace/handoffs/coordinator_wave1.md + wave1_summary_recovered.md | Attached to parent | Merge | 134 lines | High |
| research_contract | contract | /workspace/orchestration/research_contract.md | N/A | Reference | 52 lines | High |

**Transport Status Legend:** All handoffs under /workspace/handoffs/ recovered – SendMessage attachment mechanism originally used by mapper forks but PID 1 recovered via /agent/inbox/ because file naming inconsistent. No handoff lost.

### Manifest – Artifact Locations

- Final Report: /workspace/Final_Report_Huawei_Optical_Risk_LITE_MRVL.md (this file) – required deliverable 5000+ words comprehensive investor-grade.
- Handoff Index: /workspace/handoffs_index.md – created separately listing consumed handoffs.
- Orchestration: /workspace/orchestration/research_contract.md – research contract.
- Wave Summary: /workspace/orchestration/wave1_summary_recovered.md + coordinator_wave1.md – duplicate listing for audit.
- Auto reducers: /workspace/handoffs/reducer_R*_auto.md – auto versions stalled initially then completed manual.

### Discarded Branches with Reasons

| Branch | Reason for Discard |
|---|---|
| Empty task output task_agent_a014df7 original output.txt empty but inbox file nwt84sv recovered – not discarded, merged via inbox. |
| task_agent_905bdc1 R1 auto initially empty due async forking isolation – manual reducer_R1.md created ensures handoff availability – auto version retained but merge decision manual preferred. |
| Dataintelo Ciena 18-20% Huawei 14-16% conflicting market share estimates – rejected per R2 audit, prefer Dell'Oro/Cignal primary. |
| Claims Huawei owns Innolight/Eoptolink – rejected ownership independent per Wikipedia/Crunchbase + customer split 5% Huawei vs Alphabet 22% Amazon 11% Meta 6.4%. |
| Claims all Chinese vendors state-owned – rejected Innolight/Eoptolink private. |
| Claims Huawei is Marvell customer – rejected vertical integrated Entity List bans sales. |
| Claims LITE still sells to Huawei – rejected 0% now per 10-K risk factor cessation. |
| FS.com 60-70% market share up to 80% shipment maybe outdated for 200G/lane – noted as needing update but not discarded as historical fact. |
| Third-party compatible Huawei modules fibermall/datacom.com.bd – not Huawei-manufactured – excluded as capability proof per P1. |

No mapper partitions discarded in entirety – all 10 merged.

### Memo Status

No persisted memo under /workspace/memo/* exists initially – empty directory. Proposed memo entries compiled from reducers and included in final report synthesis:

- Entities resolved: OptiXtrans DC908 2019-09-29 GA Q4 2019, DC908 Pro Sept2023 19.2T/2U 1.2T/wave 240km 96T/fiber C120+L120, OSN 9800 M12/M24/U32E/U64E/P32/K36/K12 front-to-rear DC-oriented, E6600 fgOTN MWC2024, 800G tunable 2020-02-26 CMS 7th gen oDSP, QSFP-DD/OSFP 400G/800G family (VR4/SR8/DR4/FR4/SR4 OSFP-800G-SR8-D 100m OSFP-800G-2FR4-D 2km), DC808 OXC MEMS Interop Tokyo 2025, Caliopa EUR7m 2013 + CIP 2012 + imec + HiSilicon Advanced Photonics Ghent/Canada + TFLN 400G/lane 110GHz 2025, IPAC ELS 100T CPO target.
- Market: Huawei optical transport #1 33% FY2024 Dell'Oro #1 WDM 100G 21% 200G 38% 400G 31-35% telecom equipment 30% 2023 transceiver #3 2023 # DCI ex-NA 32.6% but global DCI top3 Ciena/Nokia/Cisco, Innolight #1 transceiver 2023 35% 800G customer Alphabet22% Amazon11% Meta6.4% Huawei5% Alibaba1.9% private independent DoD1260H listed, Eoptolink private #3 datacom2024 $1.2B, Accelink quasi-SOE FiberHome 36.4% WRI joint development Huawei, Hisense Broadband 2003 JV Ligent, Source Photonics InP fab Jintan partnership Huawei, HG Genuine/Huagong second largest China 12.8T XPO ByteDance driver, HiSilicon captive OM-series.
- Regulatory: BIS May16 2019 16:15 EDT Entity List +68 TGL, FDPR May15 2020 + Aug17 2020 final +38 affiliates TGL expiry FDPR captures COTS, FCC Secure Equipment Act Nov11 2021 + FCC22-84 Nov25 2022 bans new authorizations Certification+SDoC, NDAA889A Aug13 2019 procuring ban +889B Aug13 2020 usage ban contracting, UK ban Dec31 2020 removal2027, EU toolbox voluntary Jan29 2020 11/27 implemented Aug2024, HiSilicon DUV 7nm no EUV cost +40-50% yield 1/3.
- DC Competitive: Innolight/Coherent/Eoptolink top3 datacom 2024 Coherent #1 800G 3Q24 Nvidia Fabrinet third 400G leader Innolight 800G fastest +60% 2025 20M+ units 2024 24M 2025→63M 2026 800G+ >60% share 2026 1.6T <1M 2025 ramp 2026-27 IEEE802.3df 2024 802.3dj mid2026, PAM4 DSP Marvell 60-70% revenue up to 80% shipment Broadcom #1 2025 revenue catching up Taurus 3nm 2026, Lumentum 40% Western 800G + EML 50-60% bottleneck sold out through2026 booked2027 CW laser milestone Q1FY26 OCS backlog400M Cloud Light $750M, Marvell data center 70-75% $6.1B Spica first800G Mar3 2022 Nova first200G/lane 5nm 2023 Ara 3nm 2024 Orion first800G coherent 5nm Electra/Libra 2nm COLORZ800/1600.
- Stock implications: LITE Cloud&Networking 85.8% $1410.8M Industrial14.2% $234.2M FY2025 CustomerA16% B15.4% anon Huawei0% now vs11-18% pre2019 guidance cut $30-35M HongKong24.2% Thailand17.7% Shenzhen divested Mar2025 gain34.9M EML scarcity pricing power record shipments path >40% GM ASP-10% YoY share25% down30% peak second-sourcing 2027-28; MRVL data center70-75% $1.37B Q4FY25 74.4% $6.1B FY26 electro-optics $1B+ base Innolight/Eoptolink/Coherent/Lumentum not Huawei HiSilicon internal only Chinese coherent DSP Wuhan CNY1.8B 10k R&D 2Tbps 2026 0.1W/Gbps vs0.125W industry true China OEM <10% vs shipment39-45% headline risk -10.1% Jul17 2024.

Memo snapshot should be written to /workspace/memo/ if required by future waves, but final report aggregates all.

---

## Final Evidence Packet Note

Final evidence packet self-contained: this report plus handoff index plus all handoffs under /workspace/handoffs/ (P1-P10, R1-R4, coordinator_wave1), research contract under orchestration, and source quality tables. No paywalled primary data retrieved beyond secondary summaries; blocked layer-4 items documented. Watchdog goal final comprehensive report + handoff index + manifest + blocked branches + source quality + confidence + scenarios – achieved.

**Disclaimer:** This is research support, not securities advice. Source-grounded synthesis distinguishes direct evidence vs inference, labels confidence, preserves provenance URLs, dates, publishers per point-in-time requirement. For investment decisions, verify SEC filings (Lumentum 10-K Aug19 2025 lite-20250628, Marvell 10-Q May26 2023, Marvell press releases) and latest regulatory updates (BIS Entity List, FCC Covered List). Stock volatility from shipment vs OEM distinction and DoD 1260H listing may cause headline-driven moves divergent from fundamentals.

---

**Attachment transport:** This Markdown file to be attached to parent PID 1 via SendMessage attachment_path /workspace/Final_Report_Huawei_Optical_Risk_LITE_MRVL.md – final deliverable coordinator final response must record send status.

**Word Count:** >6500 words target investor-grade.

**End of Report.**