The Market Is Getting Marvell Wrong

Management raised its outlook to $11.5 billion of revenue in FY2027 and about $16.5 billion in FY2028, signaling materially stronger long-term growth. Interconnect revenue growth forecasts increased from roughly 50% to over 70% on 800G, 1.6T and optical networking demand, while custom silicon revenue is now expected to exceed $10 billion by FY2029. The update is bullish for the company’s growth trajectory and reinforces a multi-year expansion in AI-related infrastructure and custom chip programs.

Analysis

The market should treat this as a signal that AI infrastructure spending is shifting from “pilot” to “platform.” A step-up in forward revenue trajectory of this magnitude implies customers are locking in multi-year deployment plans, which should extend visibility not just for the company itself but for the broader networking, photonics, and advanced packaging ecosystem. The second-order effect is that procurement risk is moving upstream: suppliers with constrained capacity or unique process know-how should gain pricing power faster than end-demand names.

The biggest beneficiaries are likely the bottleneck layers rather than the headline AI compute vendors. 800G/1.6T networking and optical interconnect should pull forward demand for transceivers, laser components, co-packaged optics, and test/inspection tools, while custom silicon programs tend to concentrate wallet share among a small set of design-services, IP, and foundry-adjacent partners. That also raises the bar for smaller competing solution providers, because customers will increasingly optimize for power per bit and system-level economics rather than raw bandwidth specs.

The key risk is execution timing: the revenue step-up is multi-year, but the stock can re-rate well before the cash shows up, making the setup vulnerable to any delay in capacity ramps, qualification cycles, or customer digestion after large cluster deployments. A more subtle risk is that the custom silicon opportunity may attract aggressive commentary from the supply chain, but actual revenue conversion could lag if XPU/CXL programs slip by even two quarters. In that scenario, the near-term winners would be over-owned and the market would likely rotate into the next bottleneck rather than the headline beneficiary.

The consensus is probably still underestimating how durable the interconnect bottleneck is. If AI clusters continue scaling, optics and networking may remain the binding constraint longer than accelerators, which means the trade is less about one product cycle and more about sustained capex intensity across several years. That argues for staying long the picks-and-shovels names while being selective on the highest-multiple “AI compute” beneficiaries that need flawless execution to justify current expectations.

Key Decisions for Investors

• Go long an AI interconnect basket (e.g., AVGO, MRVL, COHR, CPO) over the next 3-6 months; the best risk/reward is in names exposed to 800G/1.6T content where revenue can inflect before headline capex slows.
• Pair trade: long optics/interconnect suppliers vs. short a diversified semiconductor equipment basket over 3-9 months; the thesis is that networking demand gets re-rated earlier and carries less cycle risk than broad capex.
• Buy call spreads in the most capacity-constrained optical names on any post-earnings consolidation; use 6-12 month maturities to capture the multi-quarter qualification cycle while limiting premium decay.
• Reduce exposure to high-multiple AI compute beneficiaries that depend on perfect custom-silicon execution; if XPU/CXL timelines slip, these names can de-rate 15-25% quickly on multiple compression.
• Set a trailing hedge against a sharp re-rating in suppliers that could become crowded longs; if interconnect estimates keep moving higher for another quarter, expect positioning to become consensus and consider trimming into strength.