NVIDIA and Marvell deepen NVLink and silicon photonics collaboration, highlighting AI networking bottlenecks and the value of MRVL's interconnect assets
AI summary card
NVIDIA and Marvell deepen NVLink and silicon photonics collaboration, highlighting AI networking bottlenecks and the value of MRVL's interconnect assets
Morgan Stanley believes NVIDIA's $2bn investment in Marvell and the two companies' collaboration in NVLink Fusion and silicon photonics strengthen Marvell's strategic position in the AI interconnect and optical-module supply chain, but valuation keeps the rating at Equal-weight.
- NVIDIA and Marvell are integrating the NVLink ecosystem with Marvell's XPU and scale-up networking portfolio, and Marvell will also receive NVIDIA's $2bn investment.
- The report argues that demand for networking and optical components continues to outstrip supply and is becoming a key bottleneck as AI clusters scale from single racks to larger systems.
- Marvell can embed NVLink Fusion as an IP module so that custom XPUs can communicate through NVLink, creating a new greenfield opportunity in scale-up networking.
- The silicon photonics collaboration aligns with NVIDIA's long-term CPO roadmap, and Morgan Stanley sees optical solutions entering scale-up networks as a matter of when, not if.
- At the stock level, the report is positive on the growth driver from the optical business over the next 12 months, but remains somewhat cautious on the custom silicon business, and valuation limits an upgrade.
Report interpretation
Overview
This report provides a quick take on the collaboration between NVIDIA and Marvell. The partnership includes integrating NVIDIA's NVLink ecosystem with Marvell's XPU and scale-up networking portfolio, advancing NVLink Fusion, and pursuing silicon photonics collaboration, alongside NVIDIA's $2bn investment in Marvell. Morgan Stanley believes the event underscores how severe the networking and optical interconnect bottlenecks are in AI infrastructure and reinforces the strategic value of Marvell's connectivity platform in the AI scaling cycle.
Core views
The core view is fourfold: first, as AI clusters scale from single racks to larger systems, they require higher bandwidth, longer transmission distances, and tighter system coupling, making networking and optical interconnect a key constraint. Second, Marvell's participation in NVLink Fusion means that the previously relatively closed NVLink ecosystem is beginning to open to custom XPUs and third-party networking solutions, giving Marvell incremental scale-up networking opportunities. Third, the silicon photonics and CPO direction aligns with NVIDIA's long-term roadmap, and Marvell's prior investments in the area give it the potential to become an important partner. Fourth, although the optical business and AI interconnect opportunities improve growth visibility, the report maintains an Equal-weight rating because of valuation and high stock-based compensation expense.
Analysis framework
The report combines an event-driven framework with a risk-reward lens: it first assesses the strategic implications of the NVDA/MRVL collaboration for AI supply-chain bottlenecks, the NVLink Fusion ecosystem, and the silicon photonics roadmap, then maps those implications to Marvell's optics, networking, and custom-silicon opportunities, and finally uses Morgan Stanley ModelWare earnings forecasts, target price, scenario valuation, and peer comparisons to assess the stock impact.
Methodology notes
P/E valuation based on CY27e ModelWare EPS
The base case uses 24x CY27e ModelWare EPS of $4.31 to derive a $103 target price; the bull case cites 34x CY27e ModelWare EPS of $4.98; the bear case cites 14x CY27e ModelWare EPS of $3.73.
The probabilities and drivers of upside and downside scenarios relative to the target price
The report uses the target price, current share price, bull and bear cases, and option-implied probabilities to show MRVL's 12-month risk-reward distribution.
The bottleneck role of networking, optical components, silicon photonics, and CPO in AI cluster expansion
The report treats NVIDIA's investment in key constrained supply areas as a signal that AI supply-chain bottlenecks are intensifying, with a focus on optical interconnect and scale-up networking.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- MRVLThe main subject of the report, NVIDIA partner, and beneficiary of AI interconnect and silicon photonics
- Strengths
- It has a connectivity platform, an optical business, silicon photonics investments, and custom silicon capabilities; it can participate in NVLink Fusion scale-up networking and benefit from AI cluster networking bottlenecks.
- Weaknesses
- The custom silicon business is more project-based and carries higher uncertainty; valuation and high stock-based compensation expense limit room for a rating upgrade.
- Comparison
- Compared with some peers, Marvell's capabilities span optics, networking, and custom silicon, giving it broader strategic reach; however, the report notes that ALAB also has a strong position in the NVLink Fusion ecosystem, so this collaboration should not be read as NVIDIA favoring Marvell.
- Risks
- AI opportunities falling short of expectations, weaker-than-expected recovery in storage and networking, continued drag from enterprise data centers, and persistent customer or channel inventory pressure.
- NVDAThe initiator of the collaboration and the core ecosystem player
- Strengths
- Through the NVLink ecosystem, NVLink Fusion, and investment in key supply-chain nodes, it strengthens system-level AI cluster scaling capability.
- Weaknesses
- Large-scale AI infrastructure expansion is constrained by networking and optical supply bottlenecks.
- Comparison
- Historically, NVLink has been a relatively closed ecosystem, while the Fusion model opens room for third-party XPU and networking solutions.
- Risks
- If networking, optics, or CPO supply-chain expansion falls short, it may limit the efficiency of deploying larger GPU clusters.
- ALABAnother beneficiary of the NVLink Fusion theme
- Strengths
- The report says ALAB has a strong position in the NVLink Fusion ecosystem, and scale-up opportunities have the biggest impact on it by revenue mix.
- Weaknesses
- It faces the risk of market share loss due to intensifying competition.
- Comparison
- The report is positive on the scale-up networking roles of ALAB, MRVL, AVGO, and SMTC, with ALAB having the highest revenue sensitivity.
- Risks
- A pause in AI spending, slower data center investment, delays in CXL server adoption, and delays in 1.6T port upgrades.
- AVGOA peer in AI ASIC and networking
- Strengths
- It has AI growth and long-term ASIC opportunities, and its valuation method reflects AI growth potential.
- Weaknesses
- It may face competition from NVIDIA Mellanox in networking.
- Comparison
- It is in the same AI semiconductor and custom silicon beneficiary chain as MRVL, but the business mix and scale differ.
- Risks
- Loss of networking share, insufficient ASIC competitiveness, and customer attrition.
- SMTCA peer related to optical connectivity and data center opportunities
- Strengths
- It benefits from the deployment of LPO/CPO in optical connectivity and new data center projects.
- Weaknesses
- Broad market exposure reduces pure AI data center leverage.
- Comparison
- It benefits from optical interconnect upgrades alongside MRVL, but the scale, business mix, and AI exposure differ.
- Risks
- New data center opportunities not materializing, slower capex from hyperscale cloud providers, macro weakness, and tariff or geopolitical pressure.
Key data
- NVIDIA investment in Marvell$2bnDisclosed together with the NVLink Fusion and silicon photonics collaboration to strengthen the strategic relationship between the two companies.
- MRVL ratingEqual-weightThe report is constructive on AI optical and interconnect opportunities, but valuation limits upside.
- MRVL target price$103.00Base-case target price.
- MRVL current share price$99.05Closing price as of March 31, 2026.
- Implied upside to target price+3.99%Calculated from the $103.00 target price relative to the $99.05 current price shown in the chart.
- Bull-case price$169.00The chart shows about +70.62% upside versus the current share price, with an option-implied probability of about 3.0%.
- Bear-case price$52.00The chart shows about -47.50% downside versus the current share price, with an option-implied probability of about 19.7%.
- Base-case valuation assumption24x CY27e MW EPS $4.31Used to support the $103 target price.
- Bull-case valuation assumption34x CY27e MW EPS $4.98Corresponds to upside from stronger-than-expected AI opportunities and higher revenue and valuation multiples.
- Bear-case valuation assumption14x CY27e MW EPS $3.73Corresponds to weak storage and enterprise data center demand and continued network inventory pressure.
- Bull-case revenue assumptionCY27 Revenue $15bn+The report says that if Marvell's AI opportunity exceeds expectations, it could drive higher revenue and valuation multiples.
- Global revenue exposureMainland China 40-50%, Asia-Pacific ex-Japan/Mainland China/India 20-30%, Europe ex-UK 20-30%, North America 0-10%Revenue exposure disclosure by region from the risk-reward page.
Impact & implications
The main significance of this collaboration for Marvell is that it raises the strategic visibility of its AI interconnect assets. NVLink Fusion gives Marvell the chance to participate in the custom XPU interconnect ecosystem, while the silicon photonics collaboration strengthens its potential role in next-generation GPU cluster optical interconnect and the CPO roadmap. For the industry, NVIDIA's continued investment in optical and critical interconnect suppliers indicates that AI infrastructure bottlenecks are expanding beyond compute alone to include networking, optics, and system-level interconnect. For the stock, the optical business may become a clearer driver over the next 12 months, but the custom XPU business remains more project-based and binary, and valuation already reflects a meaningful amount of growth expectation.
Risks
- Marvell's AI opportunity falls short of expectations, limiting upside to revenue and valuation.
- Weaker-than-expected recovery in storage, enterprise data center, and networking businesses continues to weigh on results.
- Customer and channel inventory pressure offsets the tailwind from custom ASIC business.
- Custom XPU projects are highly binary, and uncertainty beyond CY27 remains elevated.
- High stock-based compensation expense weighs on investor acceptance of earnings quality and valuation.
- If AI capex or data center investment slows, demand for optics and interconnect could be below expectations.
- Intensifying competition could affect Marvell's share in the NVLink Fusion, optical, or custom silicon ecosystems.
What to watch
- After NVIDIA and Marvell's collaboration, does MRVL win more NVLink Fusion scale-up networking projects?
- In custom XPU ecosystems such as AWS Trainium4, does Marvell secure actual design wins on the networking side?
- Does optical interconnect demand remain undersupplied, and how do orders and supply evolve across PAM4, 400ZR, LPO, and CPO product lines?
- What is the pace of real adoption of silicon photonics and CPO in large-scale GPU clusters?
- Can CY27 revenue move closer to the $15bn+ bull-case scenario?
- Do storage, enterprise data center, and legacy networking businesses show signs of recovery?
- What are the order, customer-win, and competitive developments at peers such as ALAB, AVGO, and SMTC in scale-up and optical interconnect?