Marvell Technology Company Overview
Marvell Technology, Inc. (NASDAQ: MRVL) is a fabless semiconductor company specializing in data-center infrastructure. It designs custom AI accelerators, high-speed optical and electrical interconnect chips, Ethernet switches, data-processing units, storage controllers and networking processors.
Marvell’s central strategy is to supply both the processing silicon inside AI systems and the connectivity required to link processors, memory, racks and data centers. The company does not manufacture its own chips; production, packaging and testing are outsourced to semiconductor foundries and manufacturing partners.
Marvell Technology Financials
Marvell’s fiscal year ends near the end of January. Its latest reported quarter was the first quarter of fiscal 2027, ending May 2, 2026.
| Financial metric | Result | Year-over-year change |
|---|---|---|
| Q1 FY2027 revenue | $2.418 billion | 28% |
| Data-center revenue | $1.833 billion | 27% |
| Data-center share of revenue | 76% | Approximately flat |
| GAAP gross margin | 52.1% | Improved |
| Non-GAAP gross margin | 58.9% | Relatively stable |
| GAAP net income | $34.5 million | Positive |
| Non-GAAP net income | $718 million | Strong growth |
| Non-GAAP diluted EPS | $0.80 | 29% |
| Operating cash flow | $638.8 million | Record quarterly level |
| Non-GAAP operating margin | Approximately 35% | Expanded |
For fiscal 2026, Marvell generated revenue of $8.195 billion, up 42% year over year. Data-center revenue reached $6.1 billion, representing 74% of total revenue. Communications and other markets generated approximately $2.09 billion, or 26%.
Fiscal 2026 non-GAAP net income was $2.466 billion, with non-GAAP diluted EPS of $2.84. Operating cash flow was approximately $1.75 billion. Reported GAAP net income of $2.67 billion included a substantial gain from the sale of the automotive Ethernet business and therefore overstates the profitability of continuing operations.
Marvell’s non-GAAP operating margin reached approximately 35.3% in fiscal 2026, up materially from the previous year. Management expects additional operating leverage as revenue grows faster than research and development and other operating expenses.
Marvell Technology Bull Case
The bull case for Marvell centers on its transformation from a diversified communications-chip supplier into a focused AI data-center infrastructure company.
Custom AI silicon could become a major revenue engine. Hyperscalers increasingly want custom accelerators that reduce cost per token, improve performance per watt and lessen their dependence on Nvidia. Marvell’s custom-silicon revenue reached approximately $1.5 billion in fiscal 2026 and management is targeting more than $10 billion in fiscal 2029.
Inference increases demand for specialized silicon. Large-scale inference workloads are repetitive, power-sensitive and highly dependent on memory bandwidth. These characteristics favor custom accelerators, custom HBM architectures, CXL memory expansion and specialized networking. Marvell supplies both the accelerator and many of the chips attached to it.
Marvell can earn more content per AI accelerator. After winning a custom XPU program, Marvell can also supply network-interface controllers, CXL memory-attach chips, retimers, Ethernet switches, optical DSPs and data-center interconnect products. This “XPU plus XPU-attach” strategy expands the revenue opportunity beyond the primary processor.
Optical interconnect is becoming essential. Copper connections become less efficient as AI clusters scale across racks and campuses. Marvell has strong positions in PAM4 DSPs, coherent DSPs, silicon photonics, active electrical cable technology and co-packaged optics. Its acquisition of Inphi established the company as a major optical-connectivity supplier, while Celestial AI adds photonic-fabric technology for scale-up networks.
Marvell is positioned across multiple AI networking architectures. The company supports Ethernet, UALink, PCIe, CXL and Nvidia NVLink Fusion. This allows Marvell to participate whether customers build systems around merchant GPUs, custom XPUs or heterogeneous combinations of both.
Long design cycles create durable revenue. Advanced custom chips can take two or more years to develop. Once qualified, a successful design may ship across multiple product generations. The engineering investment and risk involved in changing suppliers can make customer relationships relatively sticky.
Operating leverage could increase earnings faster than revenue. Management expects fiscal 2028 operating expenses to rise in the mid-to-high teens while total revenue grows approximately 45%. If achieved, non-GAAP operating margin could approach the upper end of the company’s 38% to 40% target range.
Marvell Technology Bear Case
The fiscal 2029 custom-silicon target is exceptionally ambitious. Custom revenue is expected to rise from approximately $1.5 billion in fiscal 2026 to more than $10 billion in fiscal 2029. Achieving this requires several hyperscaler programs to enter high-volume production on schedule. A delay, redesign or lost socket could materially reduce the forecast.
Revenue is concentrated among a small number of customers. Marvell’s ten largest customers represented approximately 82% of fiscal 2026 revenue. One direct customer represented 14%, while one distributor handled 37%. Distributors may serve multiple end customers, but the concentration still creates forecasting and collection risk.
Hyperscalers have significant bargaining power. Marvell’s largest customers purchase enormous volumes and can negotiate pricing, use multiple design partners or move additional chip-development work internally. Custom ASIC revenue can carry lower gross margins than proprietary merchant products, particularly when expensive manufacturing and packaging costs pass through Marvell’s income statement.
Broadcom is a formidable competitor. Broadcom has a larger and more established custom AI accelerator business, long-standing hyperscaler relationships, leading Ethernet switches and substantial SerDes expertise. Marvell must continue winning advanced designs while competing against a company with greater scale and financial resources.
Nvidia remains the dominant AI platform. Custom ASICs can offer better economics for stable workloads, but Nvidia provides a flexible architecture, a powerful software ecosystem and a rapid product roadmap. If models continue changing faster than expected, customers may favor programmable GPUs rather than committing to workload-specific ASICs.
AI capital expenditure may be cyclical. Marvell’s outlook assumes continued aggressive spending by cloud providers. Slower AI monetization, lower infrastructure utilization, power constraints or a shift in hyperscaler capital allocation could reduce demand.
Manufacturing capacity is a constraint. Marvell depends on external foundries and packaging suppliers for leading-edge wafers, HBM integration and advanced packaging. The company expects approximately $1 billion of supply-chain prepayments in fiscal 2027. These commitments secure capacity but create risk if customer demand falls below forecast.
Acquisition and integration risk has increased. Marvell has made substantial investments in Celestial AI, XConn and other technologies. These transactions increase operating costs, integration requirements, intangible assets and share dilution before all acquired products generate meaningful revenue.
Marvell Technology Management Outlook Following Its Latest Earnings
Management expects Marvell’s growth to accelerate throughout fiscal 2027, supported by exceptional AI-related bookings and programs already in development.
| Management outlook | Forecast |
| Q2 FY2027 revenue | $2.7 billion, plus or minus 5% |
| Q2 FY2027 non-GAAP EPS | $0.93, plus or minus $0.05 |
| FY2027 total revenue | Nearly $11.5 billion |
| FY2027 total revenue growth | Approximately 40% |
| FY2027 data-center growth | Approximately 50% |
| FY2027 interconnect growth | More than 70% |
| FY2027 custom-silicon growth | More than 20% |
| FY2027 scale-out switch revenue | More than $600 million |
| FY2028 total revenue | Approximately $16.5 billion |
| FY2028 total revenue growth | Approximately 45% |
| FY2028 data-center growth | Approximately 55% |
| FY2028 custom-silicon growth | More than 100% |
| FY2028 DCI module run rate | Approximately $1 billion annually |
| FY2028 scale-out switch run rate | More than $1 billion annually |
| FY2028 target operating margin | Progressing toward 38%–40% |
| FY2029 custom-silicon target | More than $10 billion |
Management expects fiscal 2028 custom-silicon growth to come from three areas: continued expansion of its existing flagship XPU, more than ten XPU-attach programs reaching higher production volumes, and a new Tier-1 XPU entering volume production.
The company also expects CXL and custom NIC demand to benefit from inference-related memory requirements and KV-cache expansion. Newly won custom designs should begin contributing after development cycles of approximately two years.
The fiscal 2027 and fiscal 2028 figures represent management forecasts based on current demand and secured programs. The fiscal 2029 custom-silicon figure is a longer-term target model and carries substantially more execution risk.
Marvell Technology Total Addressable Market and CAGR
Marvell estimates that its addressable data-center semiconductor market will reach approximately $94 billion in calendar 2028, up from approximately $21 billion in 2023. This implies an estimated compound annual growth rate of approximately 35%.
| Addressable market | Estimated 2028 TAM | Expected CAGR |
| Custom XPUs | $40.8 billion | 47% |
| XPU-attach silicon | $14.6 billion | 90% |
| Total custom accelerated compute | $55.4 billion | 53% |
| Data-center interconnect | $19.0 billion | 35% |
| Ethernet switching | $13.2 billion | 17% |
| Data-center storage silicon | $6.5 billion | 9% |
| Total Marvell data-center TAM | $94.0 billion | 35% |
Custom accelerated compute is the largest component of Marvell’s projected opportunity. It includes custom AI processors and the memory, connectivity and interface silicon attached to those processors.
Marvell is targeting approximately 20% of its total addressable data-center market by 2028, compared with an estimated share of approximately 13% in 2024. These figures are company estimates and depend on continued AI infrastructure investment, custom-silicon adoption and Marvell winning a meaningful portion of new hyperscaler designs.
Marvell Technology Products and Revenue Breakdown
Marvell reports revenue by end market rather than individual product family. Consequently, exact revenue percentages for optical DSPs, switches, storage controllers and processors are not publicly disclosed.
The following table uses fiscal 2026 reported end-market revenue and management’s custom-silicon disclosure. Custom silicon is included within the data-center segment.
| Product or service category | Products and applications | FY2026 revenue share | Principal competing products |
| Custom AI silicon and XPU-attach | Custom training and inference accelerators, Arm-based CPUs, custom NICs, CXL memory-attach devices, HBM interfaces, SRAM, chiplets and advanced packaging | Approximately 18% | Broadcom custom XPUs, Nvidia GPUs and custom-silicon services, AMD Instinct accelerators, hyperscaler internal designs |
| Optical and electrical interconnect | 800G and 1.6T PAM4 DSPs, coherent and coherent-lite DSPs, laser drivers, transimpedance amplifiers, silicon photonics, active electrical cable DSPs, PCIe retimers, near-package optics and co-packaged optics | Included within the approximately 56% merchant data-center residual | Broadcom, Credo, Coherent, Lumentum, MaxLinear and MACOM |
| Data-center interconnect modules | Coherent optical modules connecting data centers across campuses, metropolitan regions and longer distances; includes secure 800G and emerging 1.6T ZR/ZR+ products | Included within the approximately 56% merchant data-center residual | Ciena, Cisco/Acacia, Coherent, Nokia and Infinera |
| Ethernet switches and controllers | Teralynx and Prestera switches, 12.8T, 51.2T and 102.4T switching platforms, Ethernet PHYs, controllers and network adapters for scale-out and scale-up networks | Included within the approximately 56% merchant data-center residual | Broadcom Tomahawk and Jericho, Nvidia Spectrum-X, Cisco Silicon One and Intel |
| Data-processing and security products | OCTEON DPUs, infrastructure processors, SmartNIC technology, NITROX security processors and LiquidSecurity hardware security modules | Included within the approximately 56% merchant data-center residual | Nvidia BlueField, AMD Pensando, Intel Infrastructure Processing Units and Broadcom |
| Storage controllers | Bravera HDD and SSD controllers, storage preamplifiers, Fibre Channel adapters, NVMe, SAS and SATA controller technology | Included within the approximately 56% merchant data-center residual | Broadcom, Microchip, Phison, Silicon Motion and internal customer designs |
| Communications and other | Enterprise Ethernet switching, carrier routers, broadband access, 5G base-station processors, wireless infrastructure, industrial networking, security appliances and selected consumer connectivity products | 26% | Broadcom, Qualcomm, MediaTek, Intel, Cisco, Nokia and NXP |
| Total data-center products | Custom silicon, interconnect, switching, storage, processors and security products | 74% | Broadcom, Nvidia, AMD, Credo, Cisco and Intel |
| Total company | Data center plus communications and other | 100% | Broad semiconductor and infrastructure-chip market |
Approximately $1.5 billion of fiscal 2026 revenue came from custom silicon. Subtracting that amount from reported data-center revenue leaves approximately $4.6 billion, or 56% of total company revenue, from merchant data-center products. Marvell does not disclose how that residual divides among interconnect, switching, storage, processors and security.
Marvell Technology Business Model
Marvell operates a fabless semiconductor business model. It designs chips and owns or licenses the underlying intellectual property, while external foundries manufacture the wafers and specialized contractors handle packaging, assembly and testing.
The company has two principal commercial models:
Merchant semiconductor products: Marvell develops standardized products such as optical DSPs, Ethernet switches, storage controllers and DPUs that can be sold to multiple customers. These products generally offer attractive gross margins because development costs are spread across a broader customer base.
Custom silicon: Marvell collaborates with a hyperscaler or equipment manufacturer to design a chip optimized for that customer’s architecture. The customer may contribute proprietary compute cores, workload requirements or software specifications, while Marvell supplies system architecture, SerDes, memory, Arm, networking and packaging technology.
Custom engagements can range from physical-design services to full turnkey programs covering architecture, chip design, verification, packaging, testing, manufacturing and supply-chain management. Revenue can include engineering-related payments and subsequent production-chip sales, with the majority of the long-term opportunity typically coming from volume production.
Marvell benefits from lengthy design cycles and high qualification requirements. Once its technology is designed into a cloud platform, switch, storage system or optical module, the product may generate revenue over several years. However, the same structure creates risk because Marvell must invest substantially before knowing whether a design will reach high-volume production.
The company outsources production to leading foundries, including advanced-node manufacturing partners, and relies on external packaging and memory ecosystems. This keeps capital expenditures below those of an integrated chip manufacturer but makes Marvell dependent on foundry capacity, advanced packaging availability and
Invest Alpha 