Marvell Technology: The Hidden Growth Engine of AI - Beyond GPUs

Context

Marvell Technology (MRVL, Nasdaq) is an American semiconductor company specializing in connectivity, storage, and network processing chips. While Nvidia is the visible engine of AI, Marvell is its silent infrastructure: its components equip the data centers of hyperscalers (Amazon AWS, Google, Microsoft Azure) to manage data flows between GPUs, between racks, and toward storage servers. A discreet but structurally indispensable role.

Data

• Marvell’s AI revenue: the AI segment now reaches ~$2.5 billion on an annualized basis (Marvell guidance, FY2026 Q4), representing ~45% of the company’s total revenue-up from less than 10% in 2022.
• YoY growth of the data center segment: +78% (FY2026 vs FY2025).
• Forward P/E: ~40x on FY2027 estimates-a premium compared to its historical range (15-20x pre-AI) but below Nvidia’s (>50x).
• Custom ASICs: Marvell develops custom inference chips for hyperscalers (notably Google TPU v5, Amazon Trainium 2). Multi-year contracts, gross margins >55%.
• Hyperscaler capex 2026: Amazon, Google, Microsoft, and Meta have collectively announced over $220 billion in data center capex for 2026-with a significant portion allocated to interconnects and network chips (Marvell’s market).

Analysis

Marvell’s positioning is based on an investment thesis distinct from Nvidia’s: while Nvidia is exposed to the GPU purchasing cycle (volatile, dependent on AI model releases), Marvell is exposed to the hyperscalers’ infrastructure capex-more stable, more predictable, and less cyclical.

Custom ASICs (Application-Specific Integrated Circuits) represent the core of the long-term thesis: hyperscalers want to reduce their dependence on Nvidia by developing their own inference chips, and Marvell is often their foundry/design partner. This is a long-term B2B relationship with multi-year contracts-a rare revenue visibility in semiconductors.

The PAM4 DSP market (optical chips for ultra-high-speed 400G/800G interconnects in data centers) is dominated by Marvell, with an estimated ~70% market share (Dell’Oro Group). This is a defensive oligopoly in a rapidly growing market (inter-GPU bandwidth in AI clusters doubles every 18 months).

Marvell is not without risks: it remains exposed to the sluggishness of the storage segment (HDDs, enterprise SSDs), which still accounts for ~20% of revenue. However, the AI pivot now dominates its mix.

Probability-weighted scenarios

• Base case (55%): hyperscaler capex maintains its pace in 2026-2027, ASIC contracts drive AI revenue growth of +40-50% YoY. MRVL reaches $85-100 (upside of ~20-35% vs. June 2026).
• Bull case (20%): a major hyperscaler announces an ASIC order exceeding expectations, or Marvell wins a contract with a new customer (Meta, Mistral AI/European infrastructure). Re-rating toward $120.
• Bear case (25%): hyperscalers reduce capex in 2027 (macro slowdown, disappointing AI ROI). The ASIC segment stagnates. MRVL retraces to $55-60.

Portfolio implications

Marvell offers a more defensive AI exposure than Nvidia: less volatility in GPU cycles, more contractual visibility. For a portfolio seeking to diversify its AI exposure without concentrating solely on Nvidia/AMD, MRVL provides an attractive entry point at ~40x forward P/E-with high operational leverage if AI revenue growth continues. To be combined with positions in data center cooling (Vertiv VRT) and power suppliers.

Risks & blind spots

• Technological risk: if hyperscalers further internalize network chip design (Amazon Annapurna Labs), Marvell could lose ASIC market share.
• Geopolitical risk: US export restrictions to China (BIS controls) on advanced chips-Marvell has ~15% China exposure in its data center revenue.
• Blind spot: the rise of co-packaged optical cables (CPO) in AI data centers could reduce demand for Marvell’s traditional DSPs-a risk of obsolescence to monitor over 3-5 years.

To watch

• Marvell’s FY2027 Q1 earnings (August 2026): AI segment progress and annual guidance.
• TSMC announcements on advanced ASIC foundry capacity (N3/N2)-Marvell heavily depends on TSMC.
• Updated hyperscaler capex guidance (Amazon AWS re:Invent, Google Cloud Next-H2 2026).