Marvell Technology vs. Broadcom: Comparing Revenue Trends Between These Artificial Intelligence Companies

Broadcom generated $29.6 billion in fiscal Q3 2026 revenue, up 86% year over year, versus Marvell's $2.7 billion in fiscal Q2 revenue, up 37%. Broadcom's revenue scale and growth trajectory substantially exceed Marvell's, supported by its broader infrastructure portfolio, a new OpenAI hardware collaboration, and a 54% operating margin versus Marvell's 17%. Marvell's optical-interconnect specialization could position it to capture hyperscaler AI data-center demand and potentially narrow the gap over time.

Analysis

The relevant distinction is not revenue scale but operating leverage and product mix. AVGO’s combination of custom silicon, networking and software creates materially higher incremental margins and gives it capacity to fund leading-edge packaging and supply commitments; this can reinforce share gains as hyperscalers prioritize execution certainty. MRVL is more exposed to the interconnect bottleneck created by AI cluster scale-out, but its lower-margin profile means equivalent revenue growth will translate into less EPS and FCF upside unless gross margin expands alongside optical volume.

Over the next 1-3 months, the key catalyst is whether MRVL converts design-win rhetoric into disclosed production ramps, backlog, and optical revenue guidance rather than merely incremental customer qualifications. A Google-related custom-hardware award is strategically valuable but should be discounted until management identifies volume timing, NRE economics, and whether it displaces incumbent merchant or captive silicon. AVGO’s near-term risk is increasingly high expectations around AI semiconductor growth: even strong results could de-rate if the next-quarter AI revenue guide implies normalization or if large customers shift more workloads to internally designed ASICs.

The contrarian view is that optical connectivity may be a more durable beneficiary of rising model size than accelerator vendors because it scales with cluster topology, not only chip unit growth. However, MRVL’s upside is likely nonlinear and lumpy: qualification wins can precede revenue by several quarters, while price competition from COHR, LITE and Cisco optical platforms can prevent the expected margin conversion. For 6-18 months, monitor co-packaged optics adoption; broad deployment would favor differentiated DSP and optical-component suppliers, but could also compress legacy pluggable-transceiver demand during the transition.

Key Decisions for Investors

• Maintain/establish AVGO overweight versus MRVL for the next two earnings cycles; use a long AVGO / short MRVL dollar-neutral pair only if AVGO’s valuation premium is below its trailing five-year relative range. Thesis is superior FCF conversion and lower execution risk; exit the pair if AVGO’s AI revenue guide decelerates below 25% year-over-year or MRVL guides optical growth above 50% with gross-margin expansion.
• Treat MRVL as a catalyst watch rather than a full-size directional long until the company discloses production-volume timing and margin economics for its hyperscaler custom programs. Add on confirmation of two consecutive quarters of accelerating data-center revenue and gross margin; the risk is that design-win headlines remain NRE-heavy and fail to produce meaningful EPS leverage.
• For a 6-12 month thematic basket, consider a modest long MRVL paired with partial exposure to COHR or LITE rather than viewing AVGO as the direct optical substitute. This captures a potential optical-interconnect capex cycle while diversifying single-customer qualification risk; reassess if hyperscaler capex guidance weakens or component lead times normalize sharply.
• Do not chase AVGO immediately after a strong earnings-driven move; wait for a pullback or confirmation that incremental AI revenue retains company-level operating-margin support. A material decline in semiconductor gross margin or evidence that custom ASIC customers are dual-sourcing advanced packaging would invalidate the quality-premium thesis.