HSBC Upgrades Marvell to Buy: AI Networking Super Cycle Drives Valuation Re-rating

Overview

HSBC has released a report upgrading Marvell Technology (MRVL.US) from “Hold” to “Buy” and raising its target price dramatically from $85 to $300. The core thesis is that despite a 124% stock price rebound since late March, the market still severely undervalues Marvell’s growth opportunities in AI networking infrastructure—particularly the revenue potential from optical interconnects and Compute Express Link (CXL). As AI clusters scale from single racks to multi-rack “AI factories,” traditional copper cabling is hitting physical limits, creating a structural shift toward optical interconnects that will deliver a multi-year super-cycle tailwind for Marvell.

Core views

The optical interconnect business is the primary driver behind this upgrade. The report notes accelerating AI infrastructure capex, with the top five U.S. cloud providers projected to spend approximately $750 billion combined in 2026 and a consensus estimate of $1 trillion in 2027—providing a massive funding base for optical module procurement. As cluster sizes expand beyond 500,000 GPUs, the number of optical transceivers required per GPU increases, and exponential growth in AI inference token consumption ensures sustained infrastructure build-out. Marvell holds a 70% share in the 800G PAM4 DSP market and is expected to maintain a 50% share in the 1.6T DSP market. Accordingly, HSBC has raised its FY27 and FY28 optical interconnect revenue forecasts to $5.2 billion and $8.8 billion respectively—both representing 70% YoY growth, far exceeding consensus estimates of 55% and 36%. CXL (Compute Express Link) is identified as another major growth vector. Agentic AI is driving explosive demand for memory capacity and bandwidth, causing DRAM shortages and forming a “memory wall.” Marvell strengthened its CXL product portfolio through the acquisition of XConn Technologies. Its CXL switches and memory expanders decouple memory from servers, enabling memory pooling and sharing to overcome traditional DIMM slot limitations. The report argues CXL will be critical in resolving memory bottlenecks, fueling growth in the “XPU-attached” business. Thus, HSBC’s FY28 and FY29 ASIC (including XPU and attachments) revenue forecasts are 16% and 24% higher than consensus, respectively. Longer term, the custom silicon (ASIC) business outlook is improving. Beyond its existing NVIDIA NVLink Fusion partnership, media reports suggest Marvell may win development contracts for Google’s next-generation TPU and MPU. Sensitivity analysis shows that if Marvell ships 1–3 million TPU/MPU combo chips in FY29, EPS could be 28%–85% higher than the base case—a potential large-scale order offering significant upside to long-term revenue. Near-term results are also positive: the report expects Marvell to report Q1 FY27 revenue of $2.47 billion—slightly above management’s guidance ceiling and consensus—and non-GAAP EPS of $0.82, beating the $0.80 consensus. For Q2 FY27, revenue is forecast to grow 11% QoQ to $2.74 billion, driven by continued strength in optical and ASIC businesses.

Analysis framework

The report combines top-down industry trend analysis with bottom-up company fundamentals. First, it confirms the AI infrastructure super-cycle at the macro and industry level—especially the structural shift from copper to optical interconnects and the memory bottleneck caused by agentic AI—establishing a high-beta industry growth thesis. Second, it deeply analyzes Marvell’s competitive positioning in key segments, such as its market share in 800G/1.6T DSPs, CXL technical advantages, and acquisition integration effects, to assess its alpha-generating capability. Third, detailed financial modeling breaks down revenue forecasts by optical interconnects, ASIC, and other segments, comparing them against consensus to identify expectation gaps. The report emphasizes market underestimation of both optical interconnect growth rates and CXL potential. Finally, on valuation, the report raises the forward P/E multiple from 26x FY27 (historical average) to 42x FY28 (near historical peak), reflecting a re-rating for high growth and scarcity, and derives the new target price using the upgraded EPS.

Methodology notes

• Industry/ Sector Analysis FrameworkSupply-demand framework

  Physical interconnect and memory bottlenecks driven by AI compute scaling

  The report analyzes how AI cluster scaling exposes physical limits of copper cabling in transmission distance and bandwidth (supply constraint), while agentic AI creates explosive memory demand exceeding DRAM supply (demand surge), leading to the inevitable adoption of optical interconnects and CXL as industry solutions.

• Valuation MethodPE/PEG valuation

  Forward P/E Re-rating

  Instead of static PE, the report rolls the valuation base to FY28 and applies a high 42x P/E multiple (between +2 standard deviations above historical mean and the historical peak) to reflect market premium expectations for high-growth tech stocks in a super-cycle.

• Event Arbitrage & Behavioral FinanceExpectation Gap / Expectation Management

  Consensus Underestimation

  The core argument rests on 'severe market consensus underestimation.' By showing large gaps between HSBC’s forecasts and consensus—e.g., FY28 optical revenue 37% higher—the report contends current share price hasn’t priced in future earnings surprises.

Asset mapping & comparison

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

• MARVELL TECHNOLOGY INC (MRVL.US)

  Core beneficiary: Directly benefits from AI cluster scaling and memory bottleneck resolution via leadership in 800G/1.6T DSP markets and strategic CXL product positioning.

  Strengths

  70% market share in 800G DSP; projected 50% share in 1.6T DSP; comprehensive CXL product line addressing AI memory wall; partnership with NVIDIA on NVLink Fusion.

  Comparison

  Compared to competitors like Broadcom, Marvell holds a first-mover advantage and high market share stickiness in high-speed DSPs.

  Risks

  Optical transceiver market growth falls short of expectations; faster-than-expected loss of DSP market share; delays in ASIC project delivery.

Key data

• New Target Price300.00 USDSharply raised from previous $85 target, implying ~53% upside
• FY28 EPS Forecast7.12 USDUp 61% from prior $4.41 forecast and 31% above consensus $5.45
• FY27 Optical Interconnect Revenue Forecast5.176 billion USD70% YoY growth, 10% above consensus
• 800G DSP Market Share70%Marvell maintains leadership in 800G PAM4 DSP market
• 1.6T DSP Market Share Forecast50%Expected to capture half the market as 1.6T adoption accelerates

Impact & implications

The report argues Marvell is positioned at the epicenter of the AI networking hardware super-cycle. Strong demand for optical interconnect DSPs and widespread CXL adoption will drive rapid revenue growth over the next two years and deliver significant margin expansion. Current market pricing reflects only partial optimism and fails to account for the pace of optical module ramp-up and the incremental revenue from CXL solving memory bottlenecks. Confirmation of large customer orders like Google TPUs would further lift the long-term valuation ceiling. For investors, this implies Marvell offers above-industry-average earnings growth potential and a compelling re-rating opportunity.

Risks

• Optical transceiver market growth slower than expected
• Faster-than-expected erosion of DSP market share
• Delays in ASIC projects

What to watch

• Q1 FY27 optical interconnect revenue guidance and gross margin performance
• Adoption progress of CXL products among major cloud providers
• Official confirmation or updates on rumored Google TPU/MPU collaboration
• Customer qualification and mass production timeline for 1.6T DSP products