Marvell Technology highlights its custom-ASIC pipeline for AI inference and networking, positioning the company to capture design wins in next-generation hyperscaler silicon.

# Marvell Stock: Custom ASIC Pipeline Suggests Higher Valuations Ahead

As of June 1, Marvell Technology (NASDAQ: MRVL) is trading at $204.76 within an upward-sloping channel, hovering just above the MA50 line at $199.86, with the MA100 at $188.31. The Relative Strength Index (RSI) stands at 57.38 in neutral-bullish territory with no divergence and steady momentum. Following Marvell's Q1 FY2027 beat ($1.41 billion in revenue, with Data Centre segment up 42% year-over-year to $678 million), this analysis examines the structural reasons why this $678 million quarterly run rate understates the company's near-term revenue potential: recent custom ASIC design wins signed this and last quarter have not yet translated into revenue due to the long production cycle ahead.

The 12 to 24-month lag between a design win and revenue production is the primary reason the Data Centre segment will look significantly different in Q4 FY2027 compared to Q1. Understanding this timeline is crucial to assessing Marvell's growth trajectory.

**The Custom ASIC Revenue Lag: Why $678 Million Is a Floor, Not a Ceiling**

Marvell's custom ASICs operate on an entirely different revenue timeline than its merchant silicon and standard networking products. A merchant silicon product—such as an Ethernet switch or PAM4 DSP—converts to revenue within weeks of sale. A custom ASIC, by contrast, is a purpose-built chip for a specific hyperscaler use case: an inference accelerator, custom networking chip, or storage controller. When Marvell wins a custom ASIC design, it signs a multi-year supply agreement followed by 12 to 24 months of design, tape-out, product validation, and qualification before the hyperscaler deploys the chip into production and revenue begins.

Design wins announced during Q1 FY2027 are unlikely to generate production revenue until Q1 to Q3 FY2028 at the earliest. This revenue lag creates a structural understatement in today's reported numbers. When CEO Matt Murphy announces new custom ASIC design wins on earnings calls, he is effectively pre-announcing revenue 12 to 24 months out—revenue not currently factored into Marvell's stock price. The $678 million Data Centre revenue booked in Q1 FY2027 originated from design wins signed approximately Q1 FY2025 to Q1 FY2026; design wins signed this quarter will not appear until fiscal 2028. This visibility into future revenue establishes a structural argument that Marvell's Data Centre segment has not peaked but rather sits at the beginning of a multi-year ramp.

**Optical DSPs and High-Speed Ethernet: The Near-Term Revenue Drivers**

While custom ASICs follow a long revenue cycle, optical DSPs and high-speed Ethernet switches are poised to drive near-term growth. Optical DSPs—chips that convert electrical signals to optical signals within data centre transceivers—are necessary components in every AI data centre rack and between data centre buildings. As AI clusters scale to gigawatt-scale facilities, transceiver count grows much faster than GPU count, since distributed AI training requires dense optical connections between all nodes. Marvell's PAM4 DSPs are best-in-class for powering these optical connections, and critically, they follow standard product order cycles rather than the 12 to 24-month custom ASIC timeline.

Marvell's high-speed Ethernet switches represent another near-term revenue opportunity. The industry is currently debating whether InfiniBand (Nvidia's Quantum) or high-speed Ethernet should serve as the AI cluster interconnect protocol. Ethernet holds advantages: it is open-standard, more familiar to enterprise networking teams, and benefits from a diverse vendor ecosystem (Marvell, Broadcom, Cisco). As hyperscalers shift toward Ethernet, Marvell has significant room to expand market share. Both AWS and Meta have publicly endorsed Ethernet as their AI cluster interconnect protocol—a clear tailwind for Marvell's networking business.

**MRVL Technical Setup: Ascending Channel, Targets $222 and $234**

On the 4-hour chart, MRVL trades at $204.76 within an ascending channel from the $154.42 base. Price is riding the top side of the MA50 ($199.86) and MA100 ($188.31) as dynamic supports. The RSI reads 57.38 in neutral-bullish conditions with no divergence visible, and volume is expanding on green price legs.

Recent candles are testing horizontal resistance around $203.86 to $208.69 after clearing previous resistance levels. A decisive break above $208.69 would target the channel extension toward $222.62 and $234.30.

**Trade Setup:**

- **Entry:** Long above $208.69, upon clearing horizontal resistance

- **Target 1:** $222.62 (channel extension)

- **Target 2:** $234.30 (upper channel extension)

- **Stop Loss:** Close below $199.86 (MA50 floor; invalidates rising channel)

**The Mechanics of Custom ASIC Revenue Lag**

A custom ASIC is an application-specific integrated circuit tailored to exact workload specifications for a given hyperscaler. After the design win is awarded, the chip undergoes 12 to 24 months of design, tape-out, validation, and qualification before deployment in the hyperscaler's production environment. Unlike merchant silicon, custom ASICs cannot be shipped on a buy-to-order basis. Because custom ASICs represent long-term supply contracts, Marvell recognizes revenue only once the ASIC is deployed by the hyperscaler. This is why Q1 FY2027 Data Centre revenue ($678 million) reflects design wins awarded approximately two years prior, not the design wins being announced this quarter.

**The Ethernet vs. InfiniBand Debate**

AI GPU clusters can be constructed using either InfiniBand (Nvidia's Quantum is the primary supplier) or high-speed Ethernet. InfiniBand is typically preferred for tightly-coupled distributed model training requiring ultra-low latency, whereas Ethernet is a standard protocol, more familiar to enterprise networking engineers, and supported by a large vendor ecosystem. AWS and Meta have both announced Ethernet-based AI cluster networking strategies. The greater the migration by hyperscalers toward Ethernet for AI clusters, the larger Marvell's high-speed Ethernet switch portfolio can capture in competition with Broadcom and Cisco.

**The Bull Case for MRVL**

Marvell's Data Centre segment delivered 42% year-over-year growth in Q1 FY2027, and the company has secured new custom ASIC design wins that will ramp into revenue throughout fiscal 2028. Near-term growth will also come from optical DSP and high-speed Ethernet switch expansion. Technically, the stock's setup is solid, and more importantly, the business is positioned for accelerated growth over the next two years driven by custom ASIC design win ramp. Key monitoring points include the Q2 FY2027 earnings report and any hyperscaler AI CapEx guidance released in the coming weeks.

**The Bottom Line**

Marvell's Q1 FY2027 Data Centre revenue of $678 million reflects design wins secured two years ago. Today's newly announced design wins will begin ramping into the top line in late fiscal 2028. The 12 to 24-month lag is significant because it means current top-line growth in the Data Centre segment represents a floor for future growth rates, not a ceiling. Near-term catalysts include growth of the company's optical DSP business and expansion of its high-speed Ethernet switch portfolio into the hyperscaler AI cluster space.