Key Takeaways
• Marvell is the only player simultaneously covering custom ASIC design, 1.6T optical DSP, silicon photonics (Celestial AI), and CXL switching — a full-stack moat no single competitor replicates today.
• NVIDIA's $2B equity investment (March 31, 2026) into Marvell via NVLink Fusion is not a component deal — it is an ecosystem architecture deal. NVIDIA is financially aligned with Marvell's success.
• Google TPU partnership talks (disclosed April 20, 2026) represent a potential second ecosystem anchor alongside NVIDIA, diversifying Marvell away from its Amazon/Trainium concentration risk.
• The optical interconnect business is the more stable growth engine: management guided interconnect at 50%+ YoY growth in FY2027, well above the ASIC segment's 20%+ guide. Optical is the thesis, not ASIC alone.
• At ~$134, the stock is priced for FY2028 execution. The investment thesis requires believing Marvell can deliver ~$15B in FY2028 revenue (+40% YoY) — not a stretch given multi-year purchase orders in hand, but there is no margin for error.
What Is Driving This Perspective?
Structural Shift: From GPU-Centric to Fabric-Centric AI Infrastructure
The core thesis for Marvell in 2026–2028 rests on a single structural insight: as AI clusters scale from a single rack to hundreds of XPUs operating as one coherent system, the value of the silicon that connects and controls those chips is converging with — and in some cases exceeding — the value of the chips themselves. Marvell has positioned itself at precisely this inflection point, offering custom silicon design, 1.6T optical DSPs, silicon photonics fabric, and CXL memory switching in one integrated portfolio. No other merchant semiconductor company covers all four layers simultaneously.
The NVIDIA NVLink Fusion Partnership: An Ecosystem Deal, Not a Component Deal
On March 31, 2026, NVIDIA invested $2B in Marvell via Series A convertible preferred stock and formally integrated Marvell into the NVLink Fusion platform — NVIDIA's rack-scale interconnect ecosystem. This allows hyperscalers to build semi-custom AI infrastructure where Marvell's XPUs remain fully compatible with NVIDIA's software stack. Critically, NVIDIA made equivalent $2B investments in Lumentum and Coherent for laser/optical components in the same period — confirming that Marvell sits at the platform architecture layer, not the component layer. The sequencing is also notable: Marvell closed the $3.25B Celestial AI acquisition (Photonic Fabric technology) just six weeks before the NVIDIA announcement. The photonic fabric was the technical prize NVIDIA was investing behind.
The Google Angle: A Second Ecosystem Anchor
On April 20, 2026, The Information reported that Google is in active discussions with Marvell to co-develop two AI chips: a Memory Processing Unit (MPU) designed to work alongside Google's TPUs, and a new inference-optimized TPU. The MPU targets the data movement bottleneck — exactly the problem Marvell's optical and CXL portfolio is built to solve. If confirmed, this positions Marvell inside both the NVIDIA ecosystem and Google's custom silicon supply chain simultaneously, meaningfully diversifying away from its Amazon concentration and directly countering Broadcom's Google TPU lock-in narrative.
Optical > ASIC as the Primary Growth Driver
The market's narrative around Marvell has historically focused on custom ASIC (XPU) design wins. But the Q4 FY2026 earnings call revealed a more important insight: management guided the interconnect/optical business at 50%+ YoY growth in FY2027, versus only 20%+ for custom ASIC. The optical business — driven by 800G and 1.6T DSP penetration, CXL switch attach rates, and eventual Photonic Fabric ramps — is the more durable and predictable engine. Amazon Trainium uncertainty matters less if optical carries the growth.
How Have These Catalysts Been Quantified?
FY2026 actuals (confirmed): Revenue $8.195B (+42% YoY). Non-GAAP EPS $0.80 for Q4, full year non-GAAP EPS $2.84 (+81% YoY). Data center revenue $6.1B (+46.5% YoY), now 74.4% of total. Q4 non-GAAP gross margin 59.0%.
FY2027 guidance (management-issued): Revenue approaching $11B (+30%+ YoY). Data center growth 40%+ YoY. Interconnect 50%+ YoY. Custom ASIC 20%+ YoY (bias to the upside). Q1 FY2027 revenue guidance: ~$2.4B (+/-5%).
FY2028 guidance (management-issued): Revenue ~$15B (+40% YoY). Data center growth ~50% YoY. Non-GAAP EPS well over $5.00. Custom ASIC expected to roughly double vs FY2027.
Celestial AI (Photonic Fabric): Revenue contribution begins H2 FY2028. $500M annualized run-rate by Q4 FY2028. $1B annualized by Q4 FY2029. Additional opex: ~$50M annually post-acquisition.
Street consensus (April 20, 2026): 32 analysts, Strong Buy consensus. Average 12-month price target range: $121 (MarketBeat) to $125 (TipRanks). Stifel raised to $140 (April 16). Oppenheimer raised to $150 (April 15). Craig-Hallum target at $164.
My FY2028 DCF inputs: Revenue $15B. Non-GAAP net margin ~33% (EPS $5+). Shares ~930M. At 28x forward earnings = $140 FV (base). Discount rate 10.8%. Sensitivity: at 32x = $160; at 24x = $120.
Fair Value Analysis — Three Scenarios
My base case fair value is US$140, representing approximately 4% upside from the current price of ~$134. This is not a screaming buy — it is a quality compounder at fair price, where the return profile comes from FY2028 execution rather than near-term multiple expansion. The street consensus at $121–$125 is conservative because it does not yet price in the Google TPU deal (unconfirmed) or the full potential of Celestial AI optical revenues.
Note: Current price as of April 21, 2026 approximately $134. All scenarios use FY2028 earnings power as the anchor, discounted at 10.8% per annum. The bear case does not assume a catastrophic loss of business — only that ASIC growth disappoints at 10–15% YoY instead of 20%+, and that the optical ramp takes one additional quarter.
Competitive Positioning — Why Marvell, Not the Others?
Marvell occupies a unique intersection in the AI infrastructure stack that no direct competitor fully replicates:
Broadcom (~70% ASIC market share) is the primary competitor, but operates in a different ecosystem — Google's TPU stack, not NVIDIA's. Broadcom leads the UALink standard as a counterweight to NVLink. Google simultaneously uses Broadcom (existing TPU) and is in talks with Marvell (MPU + new TPU) — this is coexistence, not winner-take-all.
Credo Technology dominates the Active Electrical Cable (AEC) market — a different layer entirely. Credo's risk is that copper-to-optical transition (CPO adoption) erodes its core AEC business. Marvell is already positioned in optical. Credo is not a substitute for Marvell; they are complementary positions.
Astera Labs competes in PCIe retimers and CXL switching — one of four layers Marvell covers. Marvell's XConn acquisition directly entered Astera's turf. Astera has no custom ASIC or photonics capability. Cannot replace Marvell's full stack.
Risks — What Could Invalidate This Narrative?
Key Events to Monitor — Ordered by Priority
The following triggers are the active checkpoints that will validate or invalidate the narrative. These are not general monitoring items — each one has a specific pass/fail threshold.
Position Management Framework
Summary — The Investment Case in Plain Language
Have other thoughts on Marvell Technology?
Create your own narrative on this stock, and estimate its Fair Value using our Valuator tool.
Create NarrativeHow well do narratives help inform your perspective?
Disclaimer
The user sorkdhkddlek holds no position in NasdaqGS:MRVL. Simply Wall St has no position in any of the companies mentioned. Simply Wall St may provide the securities issuer or related entities with website advertising services for a fee, on an arm's length basis. These relationships have no impact on the way we conduct our business, the content we host, or how our content is served to users. The author of this narrative is not affiliated with, nor authorised by Simply Wall St as a sub-authorised representative. This narrative is general in nature and explores scenarios and estimates created by the author. The narrative does not reflect the opinions of Simply Wall St, and the views expressed are the opinion of the author alone, acting on their own behalf. These scenarios are not indicative of the company's future performance and are exploratory in the ideas they cover. The fair value estimates are estimations only, and does not constitute a recommendation to buy or sell any stock, and they do not take account of your objectives, or your financial situation. Note that the author's analysis may not factor in the latest price-sensitive company announcements or qualitative material.
