MRVL Rides on the AI Capex Wave: What's Ahead for Investors?

Marvell expects custom-silicon revenue to grow more than 20% in fiscal 2027 and more than double in fiscal 2028, while interconnect revenue is projected to rise over 70% in fiscal 2027 on AI networking demand. TIAs and drivers are expected to exceed a $1 billion annualized run rate within several quarters, with DCI modules targeting roughly a $1 billion rate in fiscal 2028; fiscal 2027 EPS consensus implies 47% year-over-year growth and has risen over the past 30 days. Upside from AI infrastructure, 800G/1.6T optics and an expanded NVIDIA partnership is tempered by a 12.84x sales multiple versus 5.07x for the semiconductor industry, intensifying Broadcom/AMD competition, and China/Taiwan supply-chain exposure to export restrictions and tariffs.

Analysis

MRVL’s upside is increasingly tied to the mix and timing of optical/interconnect ramps rather than aggregate AI capex. If 1.6T adoption arrives ahead of the customary network refresh cycle, MRVL can gain content per cluster; however, early ramps often carry lower yields, expedited freight and customer-qualification costs, limiting near-term gross-margin conversion. The key earnings question over the next 1-3 quarters is whether connectivity growth offsets any concentration-driven volatility in custom silicon.

AVGO is the stronger competitive beneficiary of hyperscaler custom-silicon budgets because its packaging scale, incumbent ASIC programs and software/customer integration create switching costs. MRVL’s differentiation is potentially greatest in the attach layer—optics, PAM and data-center interconnect—where cluster scaling increases port count independently of accelerator share. NVDA’s broader fabric strategy is a two-edged catalyst: ecosystem access can accelerate MRVL design wins, but deeper vertical integration could cap MRVL’s addressable content and bargaining power over 6-18 months.

The market is likely underwriting management’s multiyear revenue targets before evidence of sustained gross-margin expansion. At a substantial sales multiple premium, even a modest delay in customer deployment schedules, China/Taiwan-related compliance friction, or a slower transition from 800G to 1.6T would produce disproportionate multiple compression. Contrarian view: the most investable expression may be relative—MRVL can outperform in a continuing AI-networking buildout, but absolute upside requires estimates to rise faster than an already optimistic revenue narrative.

Key Decisions for Investors

• Maintain MRVL as a tactical overweight only against a semiconductor benchmark over the next 1-3 months; add on post-earnings weakness only if interconnect backlog, 1.6T qualification progress and gross-margin guidance validate conversion rather than just revenue growth.
• Use a pair trade: long MRVL / short AVGO in a small, beta-neutral size into an MRVL-specific connectivity catalyst, with a 3-month horizon. Exit if MRVL fails to raise its next-fiscal interconnect outlook or if AVGO discloses incremental hyperscaler ASIC wins that signal share loss; AVGO’s execution and valuation support make this unsuitable as a large structural short.
• For 6-12 month AI infrastructure exposure, prefer a core long AVGO over chasing MRVL outright: AVGO has more resilient custom-silicon economics if optical ramps slip, while MRVL offers higher operating leverage but greater customer, geography and execution sensitivity.
• Set a risk alert around export-control or tariff actions affecting China/Taiwan shipments. Any guidance quantifying compliance-driven shipment delays or elevated inventory should trigger a reduction in MRVL, since a premium multiple leaves limited tolerance for a one-quarter digestion event.
• Do not initiate directional MRVL calls solely on stated multiyear targets; wait for missing evidence—segment revenue mix, gross-margin trajectory and customer concentration—to establish whether earnings revisions can support the valuation.