# Huawei Data-Center Optics Risk: Implications for Lumentum (LITE) and Marvell (MRVL)

**Observed cutoff:** Official named-feed audit through **2026-07-10T03:21:37Z**; selected exact-product checks through **2026-07-10T03:48:40Z**. The evidence is **not** complete through the end of July 10, 2026.

## Executive Summary / Direct Answer

**Huawei’s optical-telecom capability does create a real data-center optical risk, but the risk is narrower and more segmented than a simple “Huawei enters AI optics” thesis implies.** Huawei is not telecom-only: public records establish exact Huawei 800G optical modules supported on named Huawei 800GE switches with software-version gates. That is credible integrated product capability inside Huawei’s own ecosystem, not merely a research paper or catalog aspiration.[^1]

The evidence does **not** establish three stronger propositions: externally qualified hyperscaler-scale Huawei 800G supply, a current public Huawei/HiSilicon 1.6T intra-data-center client pluggable, or a fully resolved captive short-reach bill of materials and foundry/assembly route. Public Huawei client-optics evidence therefore supports **integrated 800G presence**, not demonstrated parity with the leading global 1.6T merchant ecosystem.[^2]

The competitive channel is geographically and commercially segmented. Direct Huawei risk is materially lower at U.S. hyperscalers because of system-level legal barriers, private qualification and security requirements, firmware/provenance scrutiny, and the absence of public Huawei 800G qualification evidence. This is a strong **de facto barrier** conclusion, not a claim that every Huawei optical component is universally illegal or that private access is literally zero. China and some accessible non-U.S. markets present the more credible direct Huawei channel.[^3]

Huawei-specific risk must also be separated from the broader China-headquartered module ecosystem. Eoptolink and InnoLight have far stronger public evidence of global volume scale and 1.6T commercialization than Huawei’s externally qualified short-reach franchise; Accelink is more China-weighted. Those companies can compete with Western module vendors while still buying merchant optical components or Marvell silicon.[^4]

For **Lumentum**, current direct Huawei revenue is effectively zero: the company says it stopped all Huawei shipments at the **beginning of calendar year 2024**, that current Huawei business is completely restricted, and that it cannot earn Huawei revenue. The relevant risks are now indirect—finished-module displacement, retained or lost Lumentum component content, price/mix pressure, localization, geography, customer architecture, concentration, and an unresolved BIS/DOJ matter. LITE has the **more direct competitive channel**, but the evidence does not show that it necessarily has the larger expected dollar, EPS, valuation, or stock-price loss.[^5]

For **Marvell**, the latest Q1 FY2027 data-center share was about **76%** and China ship-to was **44%**, but neither is an optical or Huawei exposure percentage. Marvell says a substantial majority of China-destination shipments support factories or contract manufacturers for non-China customers. The separate “roughly half” optical-interconnect framing applies to FY2026 data-center revenue and remains an unaudited management proxy, not a Q1 product table. MRVL’s China-vendor effect is two-sided: module-vendor growth can preserve Marvell DSP/analog content, or remove it through LPO, captive, domestic, or alternative-supplier architectures.[^6]

The independently reviewed sensitivity grid does not identify a likely outcome. At a coordinate of **25% AI growth and 25% displacement**, evaluated extrema range from modestly negative to materially positive for both issuers; with **zero AI growth and 25% displacement**, the evaluated ranges are negative or non-positive. These are blocker-spanning assumptions, not forecasts, probabilities, or empirical confidence intervals, and the mechanisms cannot be added without evidence that they do not overlap.[^7]

## Scope and Assumptions

This report assesses Huawei-related competitive risk across four distinct optical domains:

1. **Carrier/metro transport:** OTN, coherent line systems, ROADMs, long-haul and metro optical DSPs.
2. **Scale-across:** campus and data-center interconnect, including coherent ZR/ZR+/coherent-lite links between buildings, campuses, or regions.
3. **Scale-out:** rack-to-rack and leaf-spine fabrics using 400G/800G and emerging 1.6T Ethernet or InfiniBand pluggables.
4. **Scale-up:** accelerator-to-accelerator connectivity within tightly coupled AI systems, where copper, active electrical cables, linear optics, photonic fabrics, NPO and CPO compete under stringent power and latency constraints.

The report distinguishes product existence from external qualification, named deployment, shipment scale, yield, cost and market share. It also distinguishes ship-to geography from final demand, Huawei from other China-headquartered vendors, and a competitive mechanism from a quantified financial exposure.

No real-time share price, valuation, price target, probability-weighted forecast or personalized action recommendation is included. A dedicated finance lookup for LITE and MRVL failed, so business-risk analysis is deliberately separated from market-price claims.[^8]

## Findings: Value-Chain and Technology Analysis

Carrier optics and AI-data-center optics share lasers, modulators, photonic integration, packaging, FEC, signal processing and reliability disciplines, but commercialization does not transfer automatically. The binding differences are often host electrical lanes, form factor, thermals, firmware, qualification, yield, latency, cost and customer trust.[^9]

| Value-chain layer | What matters in AI data centers | Huawei public evidence | Lumentum relevance | Marvell relevance |
|---|---|---|---|---|
| **Systems and switches** | Port density, switch ASIC/SerDes, firmware, telemetry, security and platform support | CloudEngine systems include exact 800GE platforms and supported Huawei 800G optics | Indirect: system architecture determines module/component demand; Lumentum also sells OCS | Direct in merchant switching/custom silicon, but switch revenue is not optical revenue |
| **Pluggable modules** | 400G/800G/1.6T OSFP/QSFP-DD, CMIS, power, thermal and host qualification | Exact Huawei 800G products are vendor-host supported; external 800G scale is unproven | Direct: finished modules and Cloud Light exposure | Some direct complete-module exposure through COLORZ/DCI; most short-reach role is enabling silicon |
| **Lasers, PICs and optical components** | EML, VCSEL, CW/UHP laser, SiPh, detectors, packaging and yield | Meaningful internal optoelectronic capability; exact current 800G content is unresolved | Core exposure: laser chips, assemblies, EMLs, coherent components, SiPh-related content | Indirect/component-adjacent through drivers/TIAs, DSPs and SiPh platforms |
| **DSP, analog and electrical interface** | PAM4 DSP, coherent DSP, FEC, gearbox, drivers/TIAs, 112G/224G lanes | Coherent-DSP and family-level module oDSP capability are credible; exact short-reach chip provenance is not public | Lumentum exited in-house coherent DSP/RFIC development but remains active in coherent optics and modules | Core strength: PAM4/coherent DSPs, drivers/TIAs and related merchant silicon |
| **LPO/CPO/photonic fabric** | Host co-design, low-power linear channels, switch-package integration, KGD, external lasers and serviceability | Deep research/patent activity and exhibit-level 400G LPO; no public shipping Huawei 800G LPO or CPO switch platform | Potential beneficiary through lasers/components; possible module-DSP architecture disruption | Two-sided: LPO can remove module DSP; CPO/scale-up optics can expand SiPh, custom and interconnect opportunity |
| **Qualification and market access** | Exact supported-part lists, thermal/reliability testing, firmware security, provenance and customer AVL | Strong captive-host evidence; limited public external support at 400G; no equivalent exact 800G external record found | Existing Western ecosystem and NVIDIA relationship are offsets, not immunity | Merchant design relationships and NVIDIA collaboration are offsets, not immunity |

The transfer from Huawei’s carrier strength is strongest in coherent DCI and campus links, where Huawei already has meaningful product and deployment evidence. It is weaker in mainstream short-reach scale-out and especially scale-up, where retimed PAM4, LPO, CPO, switch SerDes, exact firmware and hyperscaler qualification dominate. Huawei’s 800G products show that the company has crossed the categorical entry barrier; they do not prove that it has crossed the scale, economics and external-qualification barriers.

## Huawei Capability and Limitations

### What is established

- **Integrated data-center systems and modules through 800G.** Huawei’s public support database maps exact 800G part numbers—including 2FR4, 2VR4, DR8 and 2DR4 variants—to named CloudEngine/XH switches and current software versions. Huawei also publishes a 128×800GE switch and direct-sale/standalone module positioning.[^1]
- **Real datacom product engineering.** HiSilicon publishes exact 400G/800G Ethernet module briefs with ordering information and CMIS 5.2 for selected parts. One exact 400G HiSilicon module appears in an NVIDIA ConnectX-7 supported-parts document, although the document’s third-party disclaimer prevents treating that as broad platform approval.[^1]
- **Serious internal technology capability.** Huawei has public evidence in coherent optical DSPs, network silicon, silicon photonics, optoelectronic chips, packaging, CPO research and module-level oDSP functionality. This is more than simple resale or branding.[^10]
- **A non-China route to market in optical transport and DCI.** Named or partially corroborated Huawei DCI relationships exist in Europe and the Middle East. That demonstrates international systems reach, although it is not proof of external 800G client-optics volume.[^2]

### What is not established

- **No public hyperscaler-scale Huawei 800G client-optics franchise.** No exact non-Huawei 800G host qualification, named production customer, shipment volume, capacity, yield, cost or market-share record was established for the target parts.
- **No current public Huawei/HiSilicon 1.6T intra-data-center client pluggable.** Public client catalogs and exact searches stopped at 800G. Huawei’s 1.6T records concern coherent transport or trials and cannot be converted into a short-reach client-pluggable claim.[^2]
- **No fully resolved captive short-reach stack.** Public evidence does not identify the exact current 800G DSP, PIC, laser, TIA/driver, assembler, foundry, OSAT, substrate or yield. The best-supported manufacturing characterization is mixed internal capability plus outsourcing, not “fully captive” or “fully outsourced.”[^10]
- **No public shipping 800G LPO or CPO platform.** Huawei has credible research, patents and exhibit-level LPO evidence, but not a disclosed production 800G LPO module or CPO switch/optical-engine deployment.

The investment-relevant conclusion is that Huawei has enough capability to matter in integrated systems, China localization and accessible non-U.S. markets. The public record does not justify treating Huawei as a proven global hyperscaler-scale 800G/1.6T merchant-optics leader.

## Market Access and China/Global Segmentation

U.S. restrictions and procurement practices materially reduce Huawei’s direct opportunity, but the mechanisms must remain separate. Entity List and foreign-direct-product rules constrain many technology inputs and transactions involving Huawei; FCC rules and the Covered List are clearest for complete telecommunications systems; Section 889 applies to federal procurement and contractor use; private hyperscalers separately apply security, firmware, reliability and vendor-qualification standards.[^3]

A complete Huawei switch or optical-transport chassis is a much clearer restricted fact pattern than a bare laser or fixed passive component. A standalone smart optical module is likely to face serious FCC, procurement and customer-security barriers, but the packet found no product-specific ruling that makes every Huawei optical component universally illegal in every private or non-U.S. transaction. The practical conclusion is **low direct Huawei access at U.S. hyperscalers**, not a universal legal zero.

| Market segment | Direct Huawei systems/module channel | Broader China-vendor module channel | Principal reason |
|---|---|---|---|
| **U.S. hyperscalers and federal-aligned environments** | Low | Material but separately qualified | Legal/procurement barriers, private AVLs, firmware/security, exact host qualification; other China-headquartered vendors have stronger global merchant evidence |
| **China domestic market** | High relative to other regions | High | Huawei ecosystem integration, localization policy and domestic procurement; multiple domestic vendors participate |
| **Huawei-aligned or accessible non-U.S. enterprise/carrier markets** | Moderate to high, product- and country-dependent | Moderate to high | Huawei systems/DCI route to market exists; U.S. private qualification barriers do not apply uniformly |
| **Global merchant hyperscale market outside China** | Unproven/limited on current public 800G evidence | Material | Eoptolink and InnoLight have stronger overseas scale and 1.6T evidence; Huawei external 800G qualification remains unproven |

PRC localization is material, but the evidence does not establish Huawei-specific national dominance in high-speed client optics. The relevant competitive set includes Huawei, Eoptolink, InnoLight, Accelink, Opticore and other domestic networking/module suppliers. Treating Huawei as a proxy for all China-headquartered optical competition would understate the scale of non-Huawei competitors and misstate their merchant-component relationships.[^4]

## Differentiated Lumentum Analysis

### Current position

Lumentum’s latest public operating baseline in the packet is Q3 FY2026, ended March 28, 2026: revenue was **$808.4 million**, with **$533.3 million (66.0%)** in Components and **$275.1 million (34.0%)** in Systems. Two anonymous customers accounted for 26% and 12% of quarterly revenue. Initial ship-to was **23.0% United States** and **36.0% total Americas**; the latter must not be mislabeled as U.S. exposure. Ship-to is not final demand.[^5]

Lumentum says it stopped Huawei shipments at the **beginning of calendar year 2024** and that current Huawei business is completely restricted. Accordingly, there is no credible current direct-Huawei revenue pool to haircut again. Historical lost opportunity and an unresolved BIS/DOJ inquiry remain separate issues.[^5]

The portfolio spans high-speed EMLs and other lasers, photonic components, coherent components and modules, cloud transceivers, OCS, and emerging 1.6T/CPO-related products. The company discontinued in-house coherent DSP/RFIC development, not coherent optics. Its strategic pivot toward upstream components can reduce direct box-for-box competition with Huawei systems, but it increases dependence on component retention inside third-party modules and architectures.

### Risk mechanisms

1. **Finished-module displacement:** Huawei or another China-headquartered module vendor can displace Lumentum complete modules in contestable markets.
2. **Component retention or loss:** A competing module may still use Lumentum EML, CW/UHP laser or other optical content. Public evidence makes retention plausible but does not establish Lumentum content in exact Eoptolink, InnoLight, Accelink or Huawei 800G/1.6T products.[^4]
3. **Price and mix:** Competition can pressure module ASPs or shift demand toward lower-value architectures even without unit-share loss.
4. **Architecture:** LPO, SiPh, CPO and customer-specific module designs can change which Lumentum components remain in the bill of materials.
5. **Geography and localization:** Lumentum is effectively shut out of direct Huawei business and faces lower opportunity in Chinese hyperscalers, while Chinese optical components can still qualify into Western supply chains.
6. **Customer and capacity concentration:** AI demand and strategic capacity commitments can offset Huawei risk, but they increase execution, bargaining-power and utilization sensitivity.
7. **Compliance:** The unresolved BIS/DOJ matter is a contingent legal, penalty, export-privilege and reputation risk, not current Huawei revenue.

NVIDIA’s $2 billion preferred investment, multiyear agreements, purchase commitment and capacity-access rights provide strong strategic validation of Lumentum’s advanced-laser position. They do not guarantee realized revenue, margins, capacity execution or customer diversification; they also carry concentration and dilution considerations.[^11]

**LITE conclusion:** The Huawei/China optical threat reaches Lumentum through the most direct module-and-component channel of the two issuers. However, exact end-market mix, product margins, final geography and retained content are not disclosed, so a larger expected dollar or earnings loss versus MRVL cannot be established.

## Differentiated Marvell Analysis

### Current position

Marvell’s latest operating baseline is Q1 FY2027, ended May 2, 2026: revenue was **$2.4178 billion**, with **$1.8327 billion (75.8%, rounded to 76%)** from data center. China was **44% of ship-to**, but Marvell states that a substantial majority of China-destination shipments support factories or contract manufacturers for non-China customers. One direct customer was 16% of revenue and one distributor was 45%; the distributor represents a channel serving multiple customers and geographies, not one end user.[^6]

The “optical interconnect represents roughly half of data-center revenue” statement belongs to FY2026, not Q1 FY2027. It is useful evidence that optical is material but narrower than the entire data-center segment. A 40%/50%/60% translation is an analyst sensitivity around the word “roughly,” and rolling it into Q1 is an assumption, not a Marvell disclosure.[^12]

Marvell’s core optical franchise includes PAM4 optical DSPs, coherent and coherent-lite DSPs, drivers/TIAs, silicon photonics/CPO/LPO where revenue exists, and COLORZ DCI modules. AEC is electrical and should not be folded into optical; generic retimers are dual-use; switching and storage are separate; complete module revenue should be counted once. Marvell says its existing Ara 1.6T DSP was shipping in mass volume, while newer variants were sampling—different maturity stages that should not be conflated.[^13]

### Risk mechanisms

1. **Merchant-content retention:** InnoLight, Eoptolink and Accelink have selected public designs or collaborations using Marvell DSPs. Growth by those vendors can therefore expand Marvell content even when it pressures Western finished-module suppliers.[^4]
2. **DSP removal or substitution:** LPO explicitly removes the module DSP; captive Huawei/Chinese silicon, another merchant supplier, or domestic analog/DSP content can also reduce Marvell attach.
3. **Complete-module displacement:** Marvell-branded DCI modules face a more direct module-level competitive channel, but current product revenue is not separately disclosed.
4. **Custom and scale-up programs:** Customer-specific XPUs, networking and photonic-fabric programs have long design cycles and may be less directly substitutable by Huawei in restricted Western platforms, but they add customer and ecosystem concentration.
5. **China localization and export controls:** These can reduce China customer demand or accelerate indigenous substitutes, while the 44% ship-to figure materially overstates final China demand if read literally.
6. **Portfolio dilution:** Storage, switching, custom compute and electrical interconnect reduce the company-wide proportion directly exposed to Huawei optical competition.
7. **Acquisition and manufacturing execution:** Celestial AI, XConn and Polariton expand the interconnect roadmap but add integration, acquired-intangible and process-execution risk; Marvell remains dependent on external foundry, packaging and substrate capacity.

NVIDIA’s $2 billion investment and the NVLink Fusion/custom-XPU, scale-up networking and silicon-photonics collaboration are meaningful strategic validation. They do not eliminate execution, customer concentration, dilution or ecosystem-dependency risk and do not identify Marvell’s anonymous customer or distributor relationships.[^14]

**MRVL conclusion:** Huawei-specific direct optical risk is narrower and more architecture-dependent than the headline data-center and China percentages imply. Broader China-vendor growth can be negative, neutral or positive depending on merchant content, LPO/captive substitution and complete-module exposure.

## Conditional Scenario and Stress Table

The following scenarios are **conditions**, not probability-weighted forecasts.

| Conditional state | Huawei / China-vendor development | Content and architecture | Likely directional read-through |
|---|---|---|---|
| **Constructive** | Huawei remains primarily captive to its own systems and China-accessible markets; no public external 1.6T breakthrough; broader AI demand remains strong | Lumentum component retention is high; Chinese modules retain Marvell DSP/analog content; CPO/scale-up programs execute | Competitive loss can be more than offset by AI optical growth; LITE benefits from lasers/components, MRVL from merchant/custom content |
| **Mixed / central uncertainty** | Huawei expands integrated 800G deployment in China and selected non-U.S. markets; Eoptolink/InnoLight continue global gains | Partial Lumentum retention; mixed retimed and LPO architectures; Marvell wins some designs and loses others | LITE has the clearer direct pressure channel; MRVL’s sign varies by BOM; issuer-level outcomes remain dominated by mix and execution |
| **Adverse** | PRC localization accelerates, Huawei/other domestic vendors gain share, and external qualification broadens | Low Lumentum retention, more captive/domestic components, LPO removes Marvell DSPs, price/mix weakens, AI demand stalls | Broad architecture downside becomes material for both; exact ranking remains unresolved because revenue pools, margins and overlaps are undisclosed |

### Evaluated-grid coordinate comparison

| Assumed coordinate | LITE module-only broader China-vendor stress | LITE broad module/system architecture stress | MRVL broader China-vendor stress |
|---|---:|---:|---:|
| **25% AI growth; 25% displacement** | **−0.3% to +12.9%** | **−4.6% to +12.0%** | **−8.9% to +11.4%** |
| **0% AI growth; 25% displacement** | **−7.7% to −4.2%** | **−18.2% to −5.0%** | **−17.4% to 0.0%** |

These are one-quarter-equivalent net-revenue extrema among evaluated grid points. The coordinates and extrema span assumptions and unresolved blockers; they are not forecasts, probabilities, market-share estimates or empirical ranges. Huawei competition, PRC localization, broader China-vendor competition, route/geography, price/mix, customer delay, compliance, acquisitions and profit effects are **non-additive** without an evidenced non-overlap partition.[^7]

## Stock Implications and Larger Risk Factors

Because current prices and valuation inputs were unavailable through the dedicated finance tool, the defensible stock-level conclusion is about **risk channels**, not fair value.

### LITE

- A renewed direct Huawei-revenue loss is not the central issue; that direct relationship was already stopped.
- The important variables are AI/cloud optical demand, 200G-lane and 1.6T execution, component retention inside third-party modules, module ASP/mix, CPO/OCS ramp, customer concentration, fab utilization/capacity, and the BIS/DOJ outcome.
- Huawei’s integrated 800G capability increases competition in China and selected non-U.S. markets, but the more globally evidenced module threat comes from the wider China-headquartered vendor set.
- The NVIDIA relationship validates technology and capacity demand, while increasing dependence on successful program execution and a concentrated strategic ecosystem.

### MRVL

- The 76% data-center mix and 44% China ship-to figures should not be used as direct optical or Huawei haircuts.
- The relevant pool is narrower: PAM4/coherent DSPs, TIAs/drivers, DCI modules, SiPh/CPO/LPO and selected scale-up optical content, with explicit uncertainty around product revenue and overlap.
- The largest Huawei/China-specific downside arises if captive/domestic silicon and LPO reduce merchant content; the offset is that global China-headquartered module vendors can remain Marvell customers.
- Larger company-wide risks include custom-program execution, customer/distributor concentration, foundry and advanced-packaging capacity, acquisition integration, product-mix margins and the pace of AI infrastructure spending.

**Relative judgment:** LITE has the more direct competitive channel. The evidence does **not** support a claim that LITE necessarily has the larger expected dollar, EPS, valuation or stock-price loss. MRVL’s broader company portfolio dilutes direct optical exposure, but its absolute outcome can still be materially affected by architecture, China localization and customer-program concentration.

## Monitoring and Falsification Triggers

The conclusion should be revisited if any of the following occurs:

1. A named non-Huawei or U.S. hyperscaler qualifies and deploys an exact Huawei 800G client module at material scale.
2. Huawei or HiSilicon publishes an exact current 1.6T intra-data-center client-pluggable datasheet, host-support record, orderability record and deployment evidence.
3. An authenticated teardown or supplier/foundry record resolves the Huawei 800G DSP, PIC, laser, driver/TIA, assembler, fab or OSAT path.
4. A product-specific U.S. agency determination materially expands or narrows Huawei optical-module access, or a named U.S. hyperscaler discloses a contrary procurement decision.
5. Representative PRC award/order data shows Huawei—not the broader domestic vendor set—holding a dominant high-speed client-optics share.
6. Lumentum discloses exact AI/cloud, module, component-retention, margin, customer and final-geography cross-tabs.
7. Marvell discloses product-level optical revenue, merchant/custom overlap rules, complete-module treatment and vendor-wide content attach.
8. Representative BOMs show near-universal Marvell content in China-vendor modules, or near-complete removal through LPO/captive/domestic alternatives.
9. Multi-period results show AI demand, pricing, capacity and displacement behaving materially differently from the one-quarter sensitivity bridge.
10. Later official publications change the product, regulation, issuer or model facts beyond the observed timestamps.

## Uncertainty and Source Quality

Evidence quality is strongest for exact product existence and vendor-host support, SEC-reported issuer results, formal regulations, standards, packet integrity and model arithmetic. It is weaker for external Huawei qualification, shipment scale, market share, physical BOM, private AVLs, final-demand geography, product-level issuer revenue, margins, content attach and calibrated probabilities.

Vendor product pages and management statements establish what the company publicly claims; they do not independently prove cost, yield, reliability, qualification or share. Negative search results are bounded to the inspected official corpus and observation time. Ship-to and distributor data are routing information, not final-demand identity.

The quantitative model was independently reproduced and passed deterministic, arithmetic, monotonicity, boundary and manifest checks. Its economic calibration remains deliberately broad because important inputs are private or undisclosed. Qualitative evidence grades are not probabilities.

No material named-feed delta was observed through **2026-07-10T03:21:37Z**; selected exact-product checks continued through **03:48:40Z**. Publications later on July 10 were not observed.[^15]

## Final Answer

**Yes—Huawei’s optical-telecom capability creates a credible data-center optical competitive risk, because Huawei already has integrated, vendor-host-supported 800G data-center systems and modules. But the current public evidence supports a segmented 800G ecosystem threat, not a proven globally qualified Huawei hyperscaler franchise or a public 1.6T client-optics leadership position.** The more robust global China-vendor threat includes Eoptolink, InnoLight and other firms independent of Huawei.

For **Lumentum**, current direct Huawei revenue is effectively zero; the risk is indirect and comparatively direct at the module/component layer. For **Marvell**, the relevant optical pool is material but narrower than total data-center revenue, and China-vendor growth is two-sided because it may retain or remove Marvell content. AI optical growth can offset competitive loss under some evaluated assumptions and fail under others. The evidence does not identify a likely coordinate, rank expected dollar losses between LITE and MRVL, or support a valuation conclusion.

**Disclaimer:** This report is evidence-based research support, not investment, legal, tax or personalized financial advice. It provides no price target, valuation or action recommendation.

## Endnotes

[^1]: Huawei exact-product and host-support evidence: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W2-L2-A_handoff.md` and `W2-L2-B_handoff.md`. Primary sources include Huawei InfoFinder Hardware Center (dynamic; observed 2026-07-10), https://info.support.huawei.com/info-finder/search-center/en/enterprise/switch/cloudengine-58-68-78-88-98-pid-252837181/hardwarecenter ; Huawei OSFP-800G-2FR4 datasheet (copyright 2025; PDF metadata 2025-12-15), https://e.huawei.com/marketingcloud/pep/asset/20000001/Material/8e3c3421b4db4f25b887acd0d2c1068d/M3T1A590N1207289789466906885/Huawei%20OSFP-800G-2FR4%20Optical%20Module%20Datasheet.pdf ; Huawei XH9330-128EO datasheet (PDF created 2026-02-02), https://e.huawei.com/marketingcloud/pep/asset/20000001/Material/2afbd2f426954fb4a3ec290adf3c82c8/M3T1A590N1225541068559077532/Huawei%20CloudEngine%20XH9330-128EO%20Data%20Center%20Switch%20Datasheet.pdf . Vendor-host support does not establish external qualification or volume.

[^2]: Generation and deployment boundary: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W2-L2-A_handoff.md`. HiSilicon current datacom catalog, observed 2026-07-10: https://www.hisilicon.com/en/products/Optical-Communication/Datacom . Customer-side Huawei 1.6 Tb/s transport trial, dated 2023-12-18: https://www.eand.com/en/news/18-dec-etisalat-by-eand-completes-the-trial-of-world-s-first-ultra-high-speed-1-6Tbps-optical-solution.html . The latter is coherent transport/OTN evidence, not a client pluggable.

[^3]: Legal and procurement analysis: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W2-L2-C_handoff.md`. Primary materials include eCFR/BIS Parts 734 and 744, versioned 2026-07-08, https://www.ecfr.gov/api/versioner/v1/full/2026-07-08/title-15.xml?part=734 and https://www.ecfr.gov/api/versioner/v1/full/2026-07-08/title-15.xml?part=744 ; FCC Covered List DA 26-585, released 2026-06-12, https://docs.fcc.gov/public/attachments/DA-26-585A1.pdf ; and FAR 4.2101/4.2102, FAC 2026-01, https://www.acquisition.gov/far/4.2101 and https://www.acquisition.gov/far/4.2102 . Product and transaction facts remain necessary; this report is not legal advice.

[^4]: Broader China-vendor evidence: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W2-L2-E_handoff.md`. Eoptolink 2025 Annual Report filed 2026-04-24: https://static.cninfo.com.cn/finalpage/2026-04-24/1225172598.PDF . Zhongji InnoLight 2025 Annual Report filed 2026-03-31: https://static.cninfo.com.cn/finalpage/2026-03-31/1225056459.PDF . Accelink 2025 Annual Report filed 2026-04-23: https://static.cninfo.com.cn/finalpage/2026-04-23/1225148212.PDF . Chinese-language passages were analyzed with working translations; customer and supplier identities remain largely anonymous.

[^5]: Lumentum Q3 FY2026 Form 10-Q, filed 2026-05-06 for quarter ended 2026-03-28: https://www.sec.gov/Archives/edgar/data/1633978/000162828026030777/lite-20260328.htm . Exact issuer analysis is preserved at `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W1-L1-P05_handoff.md` and `W2-L2-F_handoff.md`. Shipment geography is initial ship-to, not final demand.

[^6]: Marvell Q1 FY2027 Form 10-Q, filed 2026-05-28 for quarter ended 2026-05-02: https://www.sec.gov/Archives/edgar/data/1835632/000183563226000019/mrvl-20260502.htm . Issuer analysis: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W1-L1-P06_handoff.md` and `W2-L2-F_handoff.md`.

[^7]: Canonical evaluated grid and validation: `quantitative/scenario_report.md`, `quantitative/quantitative_summary_and_guardrails.md`, and nested canonical archive reference `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/quantitative/g_recovered/W2-L2-G_recovered_package.tar.gz`. The recovered-G package was independently reviewed and reproduced; coordinates are non-probabilistic and mechanisms are non-additive without a non-overlap partition.

[^8]: Dedicated market-data limitation: `evidence/market_data/market_data_status.md`, dated 2026-07-10 UTC. No authoritative real-time price or valuation result was available.

[^9]: Technology-transfer analysis: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W1-L1-P03_handoff.md`. Primary standards include IEEE 802.3df/802.3dj task-force materials, https://www.ieee802.org/3/df/index.html and https://www.ieee802.org/3/dj/index.html ; OIF 800LR announcement dated 2025-04-22, https://www.oiforum.com/oif-releases-800lr-coherent-implementation-agreement-enabling-interoperable-low-power-high-capacity-10-km-optical-solutions/ ; LPO MSA v1.0 dated 2025-03-19, https://www.lpo-msa.org/files/live/sites/lpomsa/files/specs/LPO_MSA_Specification_v1p0_final.pdf ; and OIF CPO framework dated 2022-02-10, https://www.oiforum.com/oif-releases-co-packaging-framework-implementation-agreement/ . Standards readiness is not customer qualification.

[^10]: Huawei BOM/manufacturing/CPO assessment: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W2-L2-B_handoff.md`. Huawei 2025 Annual Report published 2026-03-31: https://www-file.huawei.com/dam/asset/view/dec782afd1544f558c32fc76e3141334.pdf . Huawei Research Issue 04, March 2023: https://www-file.huawei.com/-/media/corp2020/pdf/publications/huawei-research/2023/huawei-research-issue4-en2.pdf . Corporate and research evidence does not identify an exact current 800G BOM, foundry, OSAT or yield.

[^11]: NVIDIA–Lumentum transaction and partnership, filed/announced 2026-03-02: https://www.sec.gov/Archives/edgar/data/1633978/000119312526085412/d41019d8k.htm and https://www.sec.gov/Archives/edgar/data/1633978/000119312526085412/d41019dex991.htm . The agreements are non-exclusive/forward-looking and do not identify current anonymous-customer revenue.

[^12]: Marvell 2026 definitive proxy, filed 2026-05-13: https://www.sec.gov/Archives/edgar/data/1835632/000110465926060253/tm261486-1_def14a.htm . Final proxy-definition audit: `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W3-L3-P03_handoff.md`. “Roughly half” is FY2026 management framing, not an audited product line or Q1 FY2027 disclosure.

[^13]: Marvell 1.6T optical DSP release dated 2026-03-12: https://www.marvell.com/company/newsroom/marvell-1-6t-optical-dsp-ai-data-center-connectivity.html . Existing Ara mass-volume language and newer-product sampling language are distinct vendor claims; no product revenue or independent share is disclosed.

[^14]: NVIDIA–Marvell partnership announced 2026-03-31: https://investor.marvell.com/news-events/press-releases/detail/1019/nvidia-ai-ecosystem-expands-as-marvell-joins-forces-through-nvlink-fusion . The relationship validates strategic positioning but does not remove concentration, execution, dilution or ecosystem risk.

[^15]: Temporal and final adversarial controls: `evidence/adversarial/final_adversarial_review.md` and nested `evidence/scheduler/scheduler_final_evidence_packet.tar.gz::huawei_optics_coordinator_packet/evidence_handoffs/W3-L3-P04_handoff.md`. No end-of-day completeness is claimed.