Better Buy: Marvell or Broadcom?

Sales are growing at two semiconductor companies as demand for artificial-intelligence infrastructure continues to expand. The article provides no financial figures, guidance changes, or other company-specific operating metrics to quantify the growth.

Analysis

The relevant read-through is not headline AI demand but the split between merchant silicon economics and custom silicon execution. AVGO's hyperscaler programs create higher visibility and operating leverage once volumes ramp, while MRVL remains more exposed to the timing of cloud capex conversion, customer concentration, and the profitability of ramping bespoke designs. In a normalized AI-capex environment, AVGO should retain the superior multiple because its semiconductor, networking, and infrastructure-software cash flows reduce dependence on any single accelerator cycle.

Near term, the likely catalyst is quarterly evidence that custom ASIC and optical-interconnect revenue is scaling faster than the market's implicit expectations, rather than another broad AI-spending narrative. MRVL has greater percentage upside if design-win revenue converts on schedule, but also greater downside from a one-quarter pushout: fixed R&D and lower scale make gross-margin disappointment more consequential. NVDA remains the gating variable—if its platform continues absorbing most incremental AI budgets, ASIC adoption may be deferred even as total data-center spending rises.

Consensus may be underestimating the networking bottleneck. Higher cluster density increases the value of switching, DSPs, and optical connectivity; this favors AVGO directly and can support MRVL, but spending can migrate toward Ethernet fabrics and optics without proportionately benefiting compute vendors. The key falsifier is hyperscaler guidance: evidence of AI capex flattening or a shift toward internally designed networking silicon would compress expectations for both companies within 1-3 months, while a sustained ASIC ramp would matter over 6-18 months.

Key Decisions for Investors

• Maintain an overweight AVGO versus NVDA pair for the next 1-3 months: long AVGO / short NVDA in dollar-neutral sizing. The trade captures rotation from GPU-concentrated AI exposure into custom silicon and networking; exit if AVGO's next data-center outlook fails to accelerate or NVDA raises forward supply/revenue expectations materially.
• Treat MRVL as an earnings-event watch rather than a core long until management quantifies custom-silicon revenue, program timing, and gross-margin trajectory. Initiate only after independently verifiable guidance supports a ramp; target a 15-20% upside versus 20%+ downside on a material program delay.
• Use AVGO 6-9 month call spreads rather than outright calls if implied volatility is elevated: buy an at/near-the-money call and sell a 15-20% out-of-the-money call. This expresses upside from ASIC/networking revisions while limiting exposure to broad semiconductor multiple compression.
• Monitor hyperscaler capex guidance and Ethernet/optics supplier commentary during the next reporting cycle. A sequential deceleration in AI infrastructure orders is the trigger to reduce both AVGO and MRVL exposure; a confirmed acceleration supports adding AVGO first, then MRVL as the higher-beta follow-on.