Better AI Chip Stock to Buy Today: Broadcom vs. Marvell

Broadcom is presented as the more attractive AI-infrastructure investment than Marvell, supported by projected ASIC revenue doubling in each of the next two fiscal years and reaching $230 billion in fiscal 2028. Broadcom is projected to retain a 60% share of the AI ASIC design market in 2026 and trades at 18x fiscal-2027 and 11.5x fiscal-2028 forward earnings, versus Marvell at 37.5x fiscal-2028 and 25x fiscal-2029 estimates. Marvell has risen more than 200% year-to-date versus Broadcom's 3% gain, leaving Broadcom with a substantially lower valuation despite its leadership in custom chips and AI data-center networking.

Analysis

The actionable divergence is not simply ASIC exposure but earnings-quality and customer-concentration asymmetry. AVGO monetizes proprietary switching silicon, custom silicon design, and software against the same hyperscale capex pool; this gives it more opportunities to offset a delayed customer program than MRVL, whose custom-silicon ramp is more dependent on a smaller set of program milestones. MRVL's optical exposure remains a legitimate second-order beneficiary of scale-out architectures, but optical design wins typically convert more slowly and carry greater supply-chain sensitivity to foundry, packaging, and transceiver availability.

The valuation spread only works if AVGO's custom-chip bookings translate into durable gross-margin dollars rather than low-margin pass-through content. Management's stated ASIC revenue trajectory—particularly the cited 2028 figure—requires verification against filings and earnings-call transcripts before underwriting it; the magnitude appears inconsistent with a standalone segment forecast. A 1-3 month catalyst is evidence of volume production or incremental hyperscaler commitments; the 6-18 month risk is that internal accelerators displace GPU spend without expanding total AI budgets, converting the narrative from incremental TAM to vendor substitution.

Consensus may be over-extrapolating MRVL's recent relative performance from a single program-win cycle while underpricing AVGO's ability to bundle networking and custom silicon. Conversely, AVGO is not a clean defensive value trade: a major customer deferral can compress both custom-silicon estimates and its current multiple because the market increasingly values it as AI infrastructure rather than diversified semis/software. The clean falsifier for the relative thesis is MRVL demonstrating sustained data-center growth above 50% with rising gross margin while AVGO's semiconductor solutions growth decelerates despite customer capex remaining intact.

Key Decisions for Investors

• Initiate a 6-12 month long AVGO / short MRVL dollar-neutral pair after confirming next-quarter ASIC backlog and customer concentration disclosures. Target 20-30% relative outperformance; exit if MRVL raises fiscal-2028 data-center guidance while AVGO cuts semiconductor revenue or gross-margin guidance.
• For directional exposure, accumulate AVGO on post-earnings weakness only if custom-silicon revenue growth and networking margins hold; use a 10-12% underlying stop. The risk/reward is favorable versus MRVL because AVGO has multiple monetization paths into the same AI capex cycle.
• Do not chase MRVL until the next earnings report provides independently measurable evidence that its Alphabet-related components and optical products are entering volume production. Treat customer-program commentary without shipment, margin, and backlog detail as an alert rather than a catalyst.
• Monitor AMZN, GOOG, MSFT and META capex guidance and leading-edge foundry/advanced-packaging constraints over the next two quarters. Broad hyperscaler capex cuts invalidate both longs; a shift from training toward inference favors AVGO networking and MRVL optical content, but does not by itself validate custom-ASIC revenue forecasts.