Not Nvidia. Not Broadcom. But This Semiconductor Stock Is Set to Grow Faster Than Any Other Major AI Chipmaker Through the End of the Decade.

Marvell Technology is projected to grow revenue 45% in the current fiscal year, 50% next year, and faster again in fiscal 2029 as custom AI-chip demand ramps. Management expects its custom AI-chip business to more than double year over year in fiscal 2028, supported by Microsoft Maia chip design work and an expanded Alphabet agreement tied to revenue targets of up to $120B through fiscal 2033. Analysts expect Marvell EPS to rise from $4.20 this year to $6.72 next year and $10.28 in 2029, though its 53x forward P/E already prices in substantial acceleration.

Analysis

MRVL is a higher-beta expression of hyperscaler ASIC adoption, not a clean substitute for NVDA. The key debate is whether design-win revenue converts into durable, high-margin platform economics: custom silicon typically carries lower gross margins, longer qualification cycles, and materially greater customer concentration than merchant accelerators. The more investable second-order exposure may be MRVL's networking, optical, and scale-up attach content, where each incremental ASIC deployment can pull through connectivity silicon without requiring MRVL to win the primary compute socket.

Alphabet-linked warrants create an underappreciated asymmetry: they align customer incentives and validate a large opportunity, but can transfer a meaningful portion of upside to the customer through dilution if revenue thresholds are achieved. Investors should model fully diluted share count, warrant vesting triggers, XPU gross margin, and customer-specific receivables before underwriting consensus EPS; headline revenue growth alone is insufficient. A miss in any one hyperscaler program could produce outsized multiple compression because the current valuation leaves little room for a normal ASIC schedule slip.

Over the next 1-3 months, the relevant catalyst is evidence that design activity has converted to production volumes—especially backlog, supply commitments, and networking attach-rate commentary—not further TAM claims. Over 6-18 months, the structural risk to NVDA is not MRVL share capture alone but a broader shift in inference workloads toward vertically integrated ASICs, which would reduce the market's assumed durability of NVDA's accelerator mix and pricing. Consensus may be too optimistic on MRVL's revenue slope but too pessimistic on AVGO's ability to retain the highest-value custom-silicon programs; AVGO's scale, IP portfolio, and manufacturing leverage make it the lower-risk beneficiary of the same secular trend.

Key Decisions for Investors

• Maintain MRVL as a watch-list long rather than chase spot: initiate only after the next earnings release confirms XPU revenue conversion, stable-to-improving gross margin, and quantified production timing. Target 15-20% upside on successful de-risking versus 20-30% downside if guidance is pushed by one quarter or more; use a stop/review trigger on a material reduction in fiscal-year revenue outlook.
• Express the custom-ASIC theme through a 6-12 month pair trade: long AVGO / short MRVL in dollar-neutral size. AVGO should better absorb program timing volatility and capture higher-value ASIC content, while MRVL's premium multiple is more exposed to concentration and dilution; cover the short if MRVL discloses a second large production ASIC customer or materially raises long-term margin targets.
• For a higher-conviction structural hedge on NVDA, own a modest long AVGO position against an NVDA trim rather than an outright NVDA short. The trade benefits if hyperscaler ASIC adoption gains share in inference, but should be reassessed if NVDA demonstrates sustained accelerator pricing and networking attach resilience despite customer ASIC ramps.
• Set an event-driven alert for MRVL's fully diluted share count, Alphabet warrant milestones, XPU gross margin, and the percentage of revenue attributable to its largest two customers. Do not underwrite the bullish case until those disclosures allow a revenue-to-free-cash-flow bridge; absent that data, the claimed long-term EPS trajectory is not independently verifiable.