Trainium 3 and the Optical Business Support Near-Term Growth, With Upside to Marvell's CY27–CY28 Data Center Expectations

Overview

This report previews Marvell's F2Q27 results and discusses F3Q27 guidance and the outlook for its CY27–CY28 data center business. JPMorgan believes that optical DSPs, switching chips, the Trainium and Maia programs, and expanding XPU attach-chip opportunities will collectively support medium- to long-term revenue and earnings above market consensus, and therefore reiterates its Overweight rating.

Core views

JPMorgan expects Marvell's F2Q27, the quarter ended in July, to meet or slightly exceed market consensus. The primary support comes from the data center business: 1.6T and 800G programs are sustaining strong demand for optical DSPs, Teralynx 10/51.2T switching chips continue to gain market traction, and the custom chip business is beginning to benefit from the initial ramp of AWS's next-generation 3-nanometer Trainium 3 XPU ASIC program. Trainium 3 shipments are expected to increase more meaningfully in the second half of the year, although the near-term incremental contribution could be partially offset by the gradual production decline of Trainium 2. For F3Q27, the quarter ending in October, the report expects the company's revenue guidance to exceed the market's current expectation of US$3.028 billion and potentially approach US$3.1 billion, corresponding to total sequential revenue growth of 13%–14%. The data center business is expected to grow faster than the company overall, potentially increasing approximately 16%–18% quarter over quarter. Key drivers include Trainium 3 entering a more meaningful volume-ramp phase, continued strength in optical DSPs and switching chips, expanding XPU attach-chip opportunities, and further progress on Microsoft's 3-nanometer Maia program. The 3-nanometer Maia program remains scheduled to ramp around year-end and has a strong CY27 order backlog; the report also believes design work has begun on the next-generation 2-nanometer Maia program. More important than the quarterly results is how management characterizes the outlook for the CY27 and CY28 data center business. The market currently expects CY27 data center revenue to grow 55% year over year, but JPMorgan believes there is meaningful upside to this estimate. Its confidence has increased over the past 90 days based on optical DSP demand, progress with Teralynx 10, ramps in storage and CXL controllers, Trainium 3 shipments, visibility into Trainium 4 and Maia programs, and an expanding pipeline of XPU attach-chip programs. These factors are viewed as capable of supporting CY26–CY28 revenue and earnings per share above market consensus while enhancing the breadth and durability of data center growth. Marvell's expanded partnership with Google further validates the company's position as a commercial silicon and ASIC partner for cloud providers. The report emphasizes that this is not a core TPU program win but rather an opportunity involving chips attached to TPUs or XPUs, including SRAM-based AI inference offload engines, PCIe Gen5/Gen6 storage controllers, NICs, SmartNIC or DPU-class controllers, CXL-based memory interface controllers, and other high-value chips surrounding accelerators. JPMorgan believes that most potential Google-related revenue is not yet reflected in market expectations and could therefore provide incremental revenue and earnings upside, making the path to CY28 earnings per share approaching US$11.00 clearer, substantially above the market's current expectation of approximately US$9.64. Consequently, Marvell's custom chip addressable market is expanding from accelerators themselves to a broader range of AI infrastructure attach chips. On this basis, the report maintains Marvell's status as a top pick within its semiconductor coverage and reiterates Overweight.

Analysis framework

The report first assesses F2Q27 results by examining product lines such as optical DSPs, switching chips, and custom chips, and then derives F3Q27 revenue guidance based on the ramp schedules of Trainium 3, Maia, and XPU attach-chip programs. It subsequently maps individual customer programs, the product pipeline, and the Google partnership to CY27–CY28 data center revenue and earnings per share, comparing its conclusions with market consensus to identify potential expectation gaps.

Methodology notes

• Industry/Sector Analysis FrameworkPrice-Volume Decomposition

  Decomposing revenue growth by product demand and program ramp schedules

  The report separately examines demand strength and shipment ramps for optical DSPs, switching chips, and custom chips, and uses these factors to project quarterly revenue growth; the incremental contribution from Trainium 3 is partially offset by declining Trainium 2 production.

• Event-Driven Strategy and Behavioral FinanceExpectation Gap/Expectation Management

  Comparing company guidance with market consensus

  The report compares potential F3Q27 revenue guidance with the market expectation of US$3.028 billion and compares its CY27 data center growth view with the current consensus of 55% year-over-year growth to assess potential upside revisions.

• Event-Driven Strategy and Behavioral FinanceEvent-driven analysis

  Analyzing the incremental impact of the expanded Google partnership

  The report views the Google partnership as an event validating Marvell's XPU attach-chip capabilities. It distinguishes the core TPU program from incremental opportunities such as peripheral controllers and offload engines, and then analyzes their potential impact on CY28 revenue and earnings.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Marvell Technology (MRVL.US)

  Opportunities in optical DSPs, switching chips, Trainium and Maia custom chips, and XPU attach chips collectively support near- and medium- to long-term growth.

  Strengths

  Growth drivers span multiple areas, including 1.6T/800G optical DSPs, Teralynx switching chips, custom ASICs for cloud providers, and controllers surrounding accelerators, while its customer programs and serviceable addressable market continue to expand.

  Weaknesses

  Declining Trainium 2 production could partially offset the near-term incremental contribution from the Trainium 3 ramp.

  Comparison

  The report expects F3Q27 revenue guidance to exceed the market consensus of US$3.028 billion and believes CY28 earnings per share could approach US$11.00, above the market expectation of approximately US$9.64.

  Risks

  Competition in AI ASICs, optical business market share, and the sustainability of cloud capital expenditure remain key areas of market debate.

Key data

• F3Q27 Market Revenue ExpectationUS$3.028 billionThe report expects company guidance to exceed this market consensus
• Potential F3Q27 Revenue GuidanceApproximately US$3.1 billionCorresponding to approximately 13%–14% sequential growth in total company revenue
• F3Q27 Data Center Revenue GrowthApproximately 16%–18% quarter-over-quarter growthExpected to outpace overall company revenue growth
• Current CY27 Data Center Revenue Expectation55% year-over-year growthJPMorgan sees meaningful upside
• CY28 Earnings-per-Share PotentialApproximately US$11.00The potential earnings level estimated in the report
• CY28 Market Earnings-per-Share ExpectationApproximately US$9.64Below the potential level proposed in the report
• Current Share PriceUS$237.04As of August 21, 2026
• Most Recently Disclosed Price TargetUS$240Price target in the May 28, 2026 rating record

Impact & implications

The report believes Marvell's AI data center growth drivers are expanding from optical DSPs and core custom XPU chips to switching, storage, CXL, and other chips surrounding accelerators. This broadening of product and customer opportunities could improve the durability of growth and potentially drive upward revisions to CY27 data center expectations and CY26–CY28 revenue and earnings-per-share forecasts.

Risks

• Competition in the AI ASIC market could affect the extent to which Marvell realizes its custom chip opportunities.
• The market remains focused on whether Marvell can maintain its optical business market share.
• The sustainability of capital expenditure by cloud service providers could affect AI data center demand.
• The gradual production decline of Trainium 2 could partially offset growth from the near-term Trainium 3 volume ramp.

What to watch

• Whether F2Q27 results meet or slightly exceed market consensus.
• Whether F3Q27 revenue guidance exceeds US$3.028 billion and approaches US$3.1 billion.
• Management's latest commentary on the outlook for the CY27 and CY28 data center business.
• The actual degree of offset between the Trainium 3 volume ramp and declining Trainium 2 production.
• Whether the 3-nanometer Maia program begins ramping around year-end as planned, as well as progress on the 2-nanometer Maia design.
• Whether Google-related XPU attach-chip opportunities translate into incremental revenue and CY28 earnings upside.