Key Highlights

  • Marvell Technology (Nasdaq: MRVL) is pursuing both scale-out (ethernet networking between servers) and scale-up (intra-cluster GPU interconnect) markets simultaneously.
  • Scale-out switch Revenue is forecast to exceed $600M in FY27 and track toward $1 billion in FY28, underpinned by proven 800G ethernet Demand.
  • Scale-up interconnect, including optics and switching, is a new addressable market with revenue forecast to double from a prior $150M outlook in FY28.
  • Multiple tier-1 customer engagements are underway for scale-up switching, with the NVLink Fusion Partnership with NVIDIA as a key enabler.
  • FY28 interconnect revenue is expected to grow more than 70% year-on-year, with scale-up optics a primary growth driver.

Analysis

The architecture of artificial intelligence infrastructure is not uniform. It bifurcates into two distinct topologies, each with different physics, different Economics, and different semiconductor requirements. Marvell Technology (NASDAQ: MRVL) has made the deliberate strategic choice to compete in both topologies simultaneously — a decision that multiplies the addressable market but also multiplies the execution demands on the engineering organisation. Understanding the distinction is essential to assessing whether Marvell's growth ambitions are realistic or overextended.

 

Scale-Out: The Proven Market

Scale-out refers to the interconnect fabric that connects individual servers — and the AI accelerators within them — across a data centre. This is essentially ethernet networking at very high speed and very low latency. A hyperscaler building an AI Training cluster connects thousands of servers through a spine-leaf ethernet topology, using high-radix switches to aggregate traffic and high-speed transceivers to carry it over optical fibre or active electrical cables.

 

Marvell is well-established in this market. Scale-out switch revenue exceeded $300M in FY26, is guided to surpass $600M in FY27, and is tracking toward $1 billion in FY28. The 800G PAM4 optical transceivers and coherent DCI modules that Marvell supplies are the interconnect technology of the scale-out topology. This revenue is proven, growing, and supported by design wins at all five major hyperscalers.

 

Scale-Up: The Emerging Market

Scale-up refers to the interconnect that operates within an AI training cluster — between the individual accelerators in the same chassis or the same rack, and between racks in the same training pod. At these shorter distances, ethernet is often too slow and too high in latency for the tight synchronisation requirements of distributed AI training. The answer is proprietary high-speed interconnects — NVIDIA's NVLink is the leading commercial example — operating at hundreds of gigabytes per second with microsecond-level latency.

 

Marvell entered the scale-up market through the NVLink Fusion partnership, through the Celestial AI Acquisition (photonic fabric for optical scale-up), through the XConn acquisition (CXL for memory-layer scale-up), and through internal product development of PCIe 6.0 switches. In Q1 FY27, management guided scale-up optics revenue to double from a prior $150M outlook, implying a new expectation approaching $300M in FY28 from this segment alone.

 

The Simultaneity Risk

Managing two simultaneous product ramps in two different market topologies, each requiring different engineering skills and customer engagement models, is genuinely difficult. The scale-out market is characterised by standardised protocols and competitive dynamics among multiple suppliers. The scale-up market is characterised by proprietary architectures, tight partnership requirements, and winner-take-all dynamics within each customer ecosystem. The skill sets required to succeed in each are different, the sales cycles are different, and the Manufacturing ramp profiles are different.

 

The management team's track record through FY26 — delivering $8.2 billion of revenue across multiple product ramps while maintaining non-GAAP operating margins above 35% — provides a degree of comfort that they can manage this complexity. But FY28 will test the organisation more severely than FY26 did, because the scale-up bets are larger, less proven, and more technically novel than anything Marvell has attempted before. For a long-term investor, this is the central execution risk to monitor over the next 18 to 24 months.

Disclaimer

This article is for informational purposes only and does not constitute financial advice or a recommendation to buy or sell any security. All data is sourced from Marvell Technology's official Earnings presentations (FY26 Q4 and Q1 FY27). Investors should conduct their own Due Diligence before making Investment decisions.