护城河对比
最多三家已覆盖的公司，逐项对照。每一项判断都是人工编辑的判断，而非公司披露的数字——并且每一项都附有其引用依据。
| Marvell Technology | NVIDIA | Space Exploration Technologies Corp. (SpaceX) | |
|---|---|---|---|
| 护城河评级 | 窄 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 | 宽 宽阔：CUDA 软件平台及其开发者既有用户基础在整个 AI 训练技术栈上形成高昂的转换成本，由累计 $76.7B 研发投入护持，并被 FY2026 约 71% 的毛利率以及同比增长 68% 的数据中心营收（约占总额 90%）所强化——按 FY2026 10-K，这些是持久且难以复制的优势。 | 窄 The moat is narrow rather than wide. The filing names real entry barriers in launch — capital, technological expertise, licences and established government and commercial relationships — and adds spectrum and orbital resources in connectivity, while warning that the company's "leadership position in various markets, especially in orbital launch services" could attract competition-law scrutiny. The AI segment has barriers too: "Our AI businesses likewise compete in markets characterized by significant barriers to entry." Durability is capped by three-to-five-year satellite lives that force perpetual replacement launches, but the moat is not eroding: subscribers are 12.0M against 6.0M. |
| 护城河类型 | 无形资产与知识产权 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 | 无形资产与知识产权 持久优势建立在专有 IP 与软件之上：可在所有 NVIDIA GPU 上运行的全栈 CUDA 开发平台，加上数百个专有领域库/SDK/API，以及累计 $76.7B 研发投入所带来的 "inventions that are essential to modern computing"（NVIDIA 于 1999 年发明了 GPU）。按 FY2026 10-K 的业务章节，这一优势还被开发者生态系统的网络效应与 CUDA 转换成本进一步强化。 | 成本规模 The advantage is cost and scale. The stated model is extreme vertical integration plus reusability driving unit cost down: per NASA, the first Falcon 9 cut launch cost to about $2,700/kg, roughly 85% below the $18,500/kg historical average, and Falcon 9 has "demonstrated the ability to refly a first-stage 34 times" (3/31/26). The filing credits rising cadence, cargo capacity and "declining unit costs—driven by rapid reusability—have generated a compounding competitive advantage". SpaceX also launches its own constellation, and Competition notes "some of these service providers are also launch customers of SpaceX". Spectrum and licence intangibles reinforce the cost advantage. |
| 领先地位 | 并列领先者 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. | 明确领先者 FY2026 Data Center revenue was $193.7B (~90% of the $215.9B total), up 68% YoY on the Blackwell ramp, and NVIDIA describes itself as "a data center scale AI infrastructure company reshaping all industries." The 10-K frames named rivals as parties who "provide or intend to provide" GPUs/accelerators — incumbent-leader positioning. | 明确领先者 SpaceX leads clearly in its two established segments but not in its newest. The prospectus calls it "the primary launch provider for the U.S. government" — 11 of 12 NSSL medium and heavy lift missions and all five U.S. crew and cargo ISS missions for NASA in 2025 — and says Starlink is "the sole low-latency network available globally", low-latency self-defined as under 70ms; it flags competition-law scrutiny risk from that leadership. AI is the exception, with OpenAI, Anthropic, Google, Meta and Microsoft named as competitors, some with greater resources, in a segment the filing calls early stage. Cadence slipped: H1 2026 Falcon launches 77 versus 81, mass to orbit 1,041t versus 1,102t. |
| 定价权 | 中等 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. | 强 FY2026 gross margin was 71.1% (75.0% in FY2025); per the 10-K MD&A the ~3.9pt decline reflects the Hopper HGX→Blackwell full-system mix shift and a one-time $4.5B H20 excess-inventory/purchase-obligation charge, not competitive price erosion. A low-70s% hardware gross margin evidences strong pricing power. | 中等 Pricing power is moderate — a deliberate give-back from a cost position rather than weakness. Management states its strategy is "rather than prioritizing increases in ARPU" and expects Starlink ARPU to keep declining as it adds lower-priced plans and adjusts monthly service fees; the Q2 10-Q shows ARPU of $66 versus $85 while Connectivity operating income rose 79.4%. Launch sells on fixed-price contracts, one to five years for Launch Services and up to fourteen for Launch and Development. Concentration is material: about one-fifth of 2025 revenue came from U.S. federal agencies, with all launch contracts firm fixed-price and terminable at the government's convenience without advance notice. |
| 综述 | Marvell is a fabless data-infrastructure silicon supplier whose center of gravity has shifted decisively to the AI data center: the data center end market was $6,100.3M, 74% of FY2026 revenue, up from 40% two fiscal years earlier (FY2026 10-K). Its strongest position is electro-optics — in high-end PAM4 optical DSPs for 400G+ transceivers it holds roughly 60% share on Inphi-inherited SerDes/FEC IP, with Broadcom above 30%, the two together over 90% (36kr, 2026-06-27). In custom AI silicon it is the structural #2 design partner at an estimated 20-25% of the custom AI ASIC design-services market versus Broadcom's ~70%, anchored by AWS Trainium and Microsoft Maia wins (hashrateindex, 2026-05-13). The Q1 FY2027 release (2026-05-27) shows the flywheel turning — record $2.418B revenue (+28% YoY), Q2 guided to $2.7B mid-point (+35% YoY), management citing 'exceptional AI-related bookings' across 800G/1.6T optics, 51.2T switches, CPO/NPO and custom XPU — and the Celestial AI (Photonic Fabric) and XConn (PCIe/CXL switching) acquisitions closed in February 2026 extend the interconnect moat toward scale-up fabrics. The offsets that keep the moat narrow are in Marvell's own filing: intense competition (AMD, Alchip, Astera, Ayar, Broadcom, Credo, GUC, Lightmatter and others), hyperscaler in-housing risk, and heavy customer concentration. | NVIDIA 将市场领先的加速计算硬件（Blackwell 数据中心平台）与专有的全栈软件护城河——CUDA 加上数百个领域库——结合起来，由累计 $76.7B 的研发投入支撑；其 "large and growing number of developers and installed base... strengthens our ecosystem and increases the value of our platform for our customers"（FY2026 10-K）。来自 AMD、Intel 与 Huawei 的竞争正在加剧，超大规模云厂商（Alphabet、Amazon、Microsoft）也在自研 AI 芯片，但竞争者必须先跨越 NVIDIA 根深蒂固的 CUDA 软件生态系统才能取而代之。 | Two real moats and one expensive bet. Launch and Starlink rest on facts the filing states outright: barriers of capital, technology, licences, spectrum and orbital resources; about 9,600 satellites, roughly 75% of all active maneuverable satellites in orbit; the only globally available low-latency network; a NASA-cited cost per kilogram about 85% below the historical average; a first stage reflown 34 times. Connectivity rivals buy launches from it. What caps it: satellites live three to five years and must be perpetually replaced; H1 2026 Falcon launches and mass to orbit both fell and Falcon 9 flights are expected to decline; about 20% of 2025 revenue is U.S. federal, competitively bid and terminable at will; 86% of Q2 capex ($15.8bn of $18.4bn) went to an AI segment whose cloud revenue sits with few customers on 90-day-terminable contracts. Q2 operating loss $143M; H1 loss $2,086M. |
| 产业链位置 | 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). | Upstream compute-platform supplier: NVIDIA sells full-stack data-center systems (GPU + Arm CPU + DPU + NVLink/InfiniBand networking + CUDA software) to cloud providers and enterprises, and relies on third-party foundry/assembly-test-packaging partners (e.g., SPIL, Amkor, Wistron, Fabrinet). Per the FY2026 10-K, several of its largest customers (hyperscalers such as Amazon, Alphabet, Microsoft) are simultaneously customers and emerging competitors developing internal accelerated-computing silicon. | — |
| 产品（份额 / 壁垒） |
|
|
|
| 长周期投票 | +0.13 权重 0.20 · 群体判断 bullish 编辑先验，未经回测。 | +0.42 权重 0.20 · 群体判断 bullish 编辑先验，未经回测。 | +0.20 权重 0.20 · 群体判断 bearish 编辑先验，未经回测。 |