护城河对比

FY2026 10-K (filed 2026-03-11): differentiated platform IP — over 10,000 issued patents and pending applications as of 2026-01-31, plus a proven custom ASIC platform leveraging ultra-high-speed SerDes, silicon photonics, co-packaged optics and custom HBM — but Marvell itself calls its markets 'intensely competitive' with 'pricing pressures', notes customers 'have chosen to develop certain semiconductor products internally', and discloses two >=10% customers with the ten largest at 82% of FY2026 net revenue. Real, defensible IP in a concentrated, contestable customer base = narrow, not wide. https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm

来源：sec.gov

The FY2025 10-K (fiscal year ended August 29, 2025) documents design-in niches but no durable excess return. In favour: the Competition section says some large rivals 'tend to have limited customization and service capabilities and are generally focused on higher-volume memory, storage, or compute products that are manufactured to industry-standard specifications'. It says the company's 'close collaboration with customers, customer-specific designs, long-lifecycle solutions, superior products and proprietary supply chain services create significant customer loyalty that may provide an advantage when competing against large international companies'. Against: the same filing says many competitors have 'substantially greater financial, technical, marketing, distribution and other resources' and 'lower cost structures'. Sales are made 'primarily pursuant to customer purchase orders and are not based on long-term supply agreements', and 'customers are not obligated to purchase our products even if we achieve a design win'. Large customers 'have exerted' pressure for price concessions, and the ten largest were 66% of fiscal 2025 net sales. GAAP operating income was 0.6%, 1.6% and 4.2% of net sales in fiscal 2023, 2024 and 2025, and fiscal 2024 had a net loss from continuing operations of $41.8 million. The 10-Q for the quarter ended May 29, 2026 (filed 2026-07-07, https://www.sec.gov/Archives/edgar/data/1616533/000161653326000043/peng-20260529.htm) adds that Advanced Computing nine-month sales fell 20.6%, reflecting 'both the ongoing Penguin Edge wind down and hyperscale hardware sales in 2025 that did not recur in 2026'. The filing's own advantage claim is hedged ('may provide an advantage'): design-in niches exist, but the 10-K shows no durable excess return.

来源：sec.gov

The FY2025 10-K describes a purpose-built AI cloud with a real performance edge, but its own risk section discloses the defining constraint: 'approximately 67% of our revenue from our top customer, Microsoft, for the year ended December 31, 2025' — a specialized platform whose economics rest on one buyer that is also a hyperscaler competitor is narrow, not wide.

来源：sec.gov

The moat rests on hard-to-replicate mixed-signal IP: the 10-K describes the custom ASIC platform built on ultra-high-speed SerDes, ARM compute, security, storage, silicon photonics and advanced packaging (die-to-die interconnects, chiplets, CPO, custom HBM), with multiple 5nm designs executed, 3nm in progress and a 2nm platform in development; a secondary switching-cost element comes from multi-year custom design wins co-developed to individual customer specifications. https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm

来源：sec.gov

The strongest candidate moat source in the 10-K is design-in and qualification, a switching-cost argument, and the filing limits it too far to count. The risk factors say 'Our products are often incorporated into customers' systems at the design stage', and many specialty products 'are specifically designed for our OEM customers' systems or products'. But qualification is 'both product-specific and platform-specific', so a design win covers one platform. The filing notes that customers 'sometimes require us to re-qualify our products' for new platforms, which 'can be time-consuming and cause reductions in our net sales', so each new platform reopens the socket. It also says 'customers are not obligated to purchase our products even if we achieve a design win', and warns that OEMs designing in 'standardized or commodity components' could reduce demand for its 'higher priced specialized or customized solutions'. Scale is ruled out because the filing says many competitors have lower cost structures. Patents are ruled out as a durable source. The company holds about 1,650 of them and calls intellectual property 'an important aspect of our business', but it says its patents 'do not cover all of our technologies' and that competitors 'may design around our patented technologies'. It also says some products are 'built around mature industry standards and have less patent protection', so 'we cannot prevent our competitors from reverse-engineering and duplicating those products'.

来源：sec.gov

The 10-K grounds the advantage in purpose-built infrastructure — first-to-deploy NVIDIA GB200/GB300 NVL72 systems and a data-center fabric 'designed to harness the full potential of each GPU' — a performance-per-dollar edge at scale, not a customer lock.

来源：sec.gov

Leader in its optics niche, #2 in custom silicon: ~60% of high-end PAM4 DSP share (36kr, 2026-06-27, https://eu.36kr.com/en/p/3870758441178373) but an estimated 20-25% of custom AI ASIC design services versus Broadcom's ~70% (hashrateindex, 2026-05-13, https://hashrateindex.com/blog/design-partners-ai-asic-market-part-2/) — net, a co-leader in AI data-center connectivity/custom silicon behind Broadcom overall.

来源：sec.gov

The 10-K claims no share figure or rank in any segment. It says 'Our businesses compete with numerous global and local companies'. Its HPC and AI business 'competes primarily with global manufacturers of HPC and AI products and services'. For fault-tolerant solutions it says 'we primarily compete with manufacturers of enterprise servers and industrial computers'. Its memory business competes 'against memory module providers and, to a lesser extent, large semiconductor manufacturers'. The risk factors warn that larger competitors 'may be able to respond better to new or emerging technologies, such as generative AI'. Against that, the company's stated edge is customization and service, where it says some large rivals 'tend to have limited customization and service capabilities'. Its leadership language is self-description without share figures: Cree LED 'has been a leader in LED lighting technology', Stratus Technologies is 'a global leader in simplified, protected and autonomous computing platforms and services', and the memory business is 'a primary supplier of longer-lifecycle solutions to OEM customers'. A specialist that claims no rank and faces many larger, lower-cost rivals, while neither leading nor clearly trailing in its niches, is parity.

来源：sec.gov

The 10-K positions CoreWeave as a first-to-deploy specialist against hyperscalers who offer AI compute 'as part of a broader product portfolio' — a leader of the purpose-built neocloud niche, not of the AI-cloud market its own filing says it competes in.

来源：sec.gov

The 10-K characterizes Marvell's markets as having 'pricing pressures' and intensifying competition (https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm), yet the Q1 FY2027 release reports 52.1% GAAP / 58.9% non-GAAP gross margin on record revenue (https://www.sec.gov/Archives/edgar/data/1835632/000183563226000014/q127_8kx522026ex-991.htm) — differentiated-IP margins, tempered by hyperscaler buyer power.

来源：sec.gov

The FY2025 10-K describes price as set by customers and the market. It says its large customers 'are able to exert, have exerted and we expect will continue to exert, pressure on us to make concessions on price and on terms and conditions'. It says 'The markets for our Integrated Memory products have historically been characterized by declines in average selling prices', and that 'Competitive pressure has led in the past and may continue to lead to intensified price competition resulting in lower net sales and lower profit margins'. Gross margin held in a narrow band: 28.8% in fiscal 2023, 29.1% in fiscal 2024 and 28.8% in fiscal 2025. The filing attributes each move to mix, for example higher-margin Advanced Computing service revenue in 2024, not to pricing. The 10-Q for the quarter ended May 29, 2026 (filed 2026-07-07, https://www.sec.gov/Archives/edgar/data/1616533/000161653326000043/peng-20260529.htm) reports gross margin of 27.8% versus 29.3% a year earlier (27.7% versus 28.9% for nine months), which it attributes to the Penguin Edge wind-down and sales mix. Third-quarter Integrated Memory sales rose 111.4% (75.6% for nine months) as 'accelerating AI-driven demand drove favorable pricing and increased volume', and the 10-Q says that business has 'gross margins which are lower than the Company average'.

来源：sec.gov

A first-to-deploy performance edge on scarce new NVIDIA systems supports pricing while the hardware is scarce, but 67% single-customer concentration is buyer leverage the filing states outright.

来源：sec.gov

Fabless supplier spanning 'data center core to network edge': it sits between hyperscaler AI compute (custom XPU/XPU-attach ASICs) and the optical layer (PAM4/coherent DSPs, CPO/LPO, DCI, AEC, PCIe retimers), outsourcing fabrication to independent CMOS foundries; the Feb-2026 Celestial AI and XConn acquisitions push it further into scale-up photonic fabric and PCIe/CXL/UALink switching (FY2026 10-K, https://www.sec.gov/Archives/edgar/data/1835632/000183563226000011/mrvl-20260131.htm).

Downstream integrator. Per the 10-K it buys memory from Samsung, Micron, SK hynix and Kioxia and processors from Intel and AMD, and turns them into specialty memory modules and managed AI/HPC clusters for OEM, enterprise, government and cloud customers.

Layer-8 purpose-built AI neocloud — first-to-deploy NVIDIA systems for training and inference at scale.

• Custom AI XPUs & ASICs 前三 · 深 来源：hashrateindex.com
• Data center interconnect & CPO/LPO 未知 · 深 来源：sec.gov
• Ethernet switching (scale-out & scale-up) 未知 · 中等 来源：sec.gov
• PAM4 & coherent optical DSPs 领先 · 深 来源：eu.36kr.com
• Storage controllers & fibre channel 未知 · 中等 来源：sec.gov

• AI and HPC infrastructure (OriginAI, ICE ClusterWare) 未知 · 低 来源：sec.gov
• Cree LED chips and components (XLamp, J Series, high-brightness SMD) 未知 · 低 来源：sec.gov
• SMART Modular specialty memory and storage (DRAM modules, Zefr, CXL, flash/SSD) 细分市场 · 中等 来源：sec.gov
• Stratus fault-tolerant computing (ztC Endurance, ftServer, ztC Edge, everRun) 细分市场 · 低 来源：sec.gov

• CoreWeave Cloud (GPU compute platform) 领先 · 深 来源：sec.gov
• Weights & Biases (AI developer tooling) 挑战者 · 中等 来源：sec.gov

编辑先验，未经回测。

查看它究竟如何投票 →

编辑先验，未经回测。

查看它究竟如何投票 →

编辑先验，未经回测。

查看它究竟如何投票 →