Broadcom vs. Marvell: The AI Supercycle Is Big Enough for Both. Here's the Better Buy.

Broadcom projects AI semiconductor revenue of about $58 billion in fiscal 2026, rising to roughly $115 billion in fiscal 2027 and $230 billion in fiscal 2028; its Q3 AI chip revenue increased 221% year over year to $16.7 billion. Marvell raised its fiscal 2027 revenue outlook to $12 billion (+45%) and fiscal 2028 outlook to $18 billion (+50%), supported by an expanded Google custom-silicon partnership and data-center demand. The article favors Broadcom on risk-reward, citing a lower forward valuation of 21.6x fiscal-2027 earnings and 14.2x EBITDA versus Marvell's 57.3x earnings and 30.2x EBITDA, despite Broadcom's faster expected revenue growth.

Analysis

Google's multi-sourcing is more important for AI supply-chain economics than for near-term share transfer. It lowers its dependence on any one ASIC vendor and creates negotiating leverage on unit pricing, while AVGO's broader rack content and software cash flows can absorb custom-silicon pricing pressure better than MRVL. For MRVL, the warrant structure effectively makes incremental Google revenue lower-quality than headline revenue: dilution and likely customer-specific engineering costs mean investors should focus on gross-margin progression and design-win conversion, not booked revenue alone.

The valuation spread creates an asymmetric pair setup over the next 1-3 months. AVGO has multiple avenues to monetize AI buildout—custom compute, switching, interconnect and software—whereas MRVL must convert a narrow set of hyperscaler programs into sustained volume without margin erosion; a single TPU generation delay or lower-than-expected optical attach rate would disproportionately impair MRVL's earnings multiple. Conversely, MRVL can outperform sharply if Google volume ramps ahead of vesting thresholds and connectivity growth demonstrates that it is gaining content per AI cluster rather than merely participating in a cyclical optics upturn.

The consensus risk is that the market extrapolates hyperscaler capex into linear semiconductor revenue through 2028. AI infrastructure spending can remain high while shifting toward internally designed accelerators, lower-cost networking architectures, or longer asset lives, compressing merchant supplier growth rates. Near-term evidence comes from Alphabet capex commentary and AVGO/MRVL backlog, lead-time, and gross-margin disclosures; a deceleration in custom-silicon ramps would hit MRVL first, while an AVGO de-rating would likely require both AI guidance pressure and weaker infrastructure-software renewals.

Key Decisions for Investors

• Initiate a 6-12 month dollar-neutral pair: long AVGO / short MRVL. The trade captures a more diversified AI monetization base against higher execution and dilution sensitivity at MRVL; target 15-20% relative outperformance, with a 10% relative stop if MRVL raises FY2028 revenue or margin outlook materially above current expectations.
• Use AVGO pullbacks following earnings or sector-wide AI-capex volatility to build the long rather than chase strength. Thesis is falsified by a material reduction in custom-silicon revenue outlook, deterioration in semiconductor gross margin, or evidence that a major hyperscaler is reducing Broadcom content per rack.
• Avoid adding directional MRVL exposure until management discloses Google-related revenue timing, warrant vesting economics, and gross-margin impact. Upgrade to a tactical long only if data-center growth is accompanied by sequential gross-margin expansion and backlog visibility extending beyond one TPU cycle.
• Monitor GOOG capex guidance and TPU deployment cadence over the next two earnings cycles as the key cross-asset catalyst. A capex increase with accelerating TPU adoption supports both names, but a spend increase focused on internal infrastructure or power/real-estate constraints may not translate proportionally into MRVL silicon revenue.