Marvell Technology: Qualcomm Just Entered Its AWS Territory, But The Buy Case Still Holds

Marvell's data-center revenue rose 46% year over year in Q2 FY2027, and management raised its revenue outlook to approximately $12B for FY2027 and $18B for FY2028. Qualcomm's entry into Amazon's AI custom-silicon and optical-connectivity ecosystem introduces a competitive risk, but Marvell retains broad AWS exposure through its interconnect, switching, Photonic Fabric and outstanding AWS warrants. The article maintains a buy view based on Marvell's AI data-center growth trajectory and differentiated connectivity portfolio.

Analysis

The key question is not whether Qualcomm can win an AWS design slot, but whether it can displace Marvell content across the full rack-level connectivity stack. Custom compute ASIC programs create high switching costs once firmware, SerDes qualification, optical modules, and network topology are validated; a new entrant is more likely to pressure incremental design-win pricing than rapidly remove incumbent revenue. Marvell's risk is therefore gross-margin dilution and lower future content-per-cluster, not an immediate revenue cliff.

AWS warrant economics are the underappreciated issue. If warrant vesting is tied to purchase commitments, reported revenue growth may overstate the quality of underlying gross profit and free-cash-flow conversion; investors should focus on disclosed warrant charges, customer concentration, and data-center gross-margin progression rather than headline revenue. A 100-200bp miss in segment gross margin or evidence that AWS is dual-sourcing optical/connectivity components could drive a sharper multiple reset than a modest top-line revision.

Over the next 1-3 months, Qualcomm's opportunity is principally narrative-driven until a specific production deployment, volume commitment, or optical design win is independently confirmed. Over 6-18 months, rising AI cluster bandwidth should expand the addressable market for optical interconnects, benefiting component suppliers such as COHR and LITE as well as networking incumbents; the more material competitive risk to Marvell remains hyperscaler insourcing and Broadcom's entrenched custom-silicon ecosystem rather than Qualcomm alone.

Contrarian view: the market may initially treat any Qualcomm-AWS announcement as evidence of broad Marvell displacement. That is likely overstated unless it includes switching, DSP, or optical attach-rate content rather than a narrowly scoped accelerator engagement. Conversely, Marvell's growth expectations leave little room for a slower AI capex cadence: any AWS capex moderation or order pushout could expose how much of the valuation rests on a small number of hyperscaler programs.

Key Decisions for Investors

• Maintain or initiate a tactical long MRVL only on pullbacks following Qualcomm-related headlines; use a 3-6 month horizon and size modestly until AWS customer concentration, warrant accounting, and segment gross-margin data are verified. Thesis is invalidated by a material FY2028 revenue-guide reduction, a sustained data-center gross-margin decline exceeding 200bp, or confirmed AWS dual-sourcing of Marvell's core interconnect content.
• Avoid a standalone long QCOM solely on the AWS custom-silicon narrative. Treat a disclosed production-volume award, associated revenue range, and timing of commercial deployment as required confirmation; without those data, the likely valuation impact is limited relative to QCOM's broader handset and licensing earnings base.
• For a relative-value expression, consider long MRVL / short QCOM only after MRVL underperforms by 10-15% on an unverified competitive headline and only with a 1-3 month horizon. The trade captures the distinction between near-term narrative pressure and slower qualification cycles; stop out if Qualcomm announces a broad AWS connectivity win with quantified production volumes.
• Monitor COHR and LITE as second-order beneficiaries of higher optical intensity in AI clusters. Do not initiate on this signal alone; upgrade the thesis if hyperscaler capex commentary confirms accelerating 800G/1.6T deployments, while reducing exposure if AI infrastructure spending shifts toward lower-bandwidth scale-up architectures.