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 | MediaTek | |
|---|---|---|---|
| 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 Counterpoint's public quarterly AP-SoC table, published 8 September 2026, shows MediaTek first in all ten quarters it covers (Q1 2024 through Q2 2026), never below 31%, with the nearest rival never above 30% -- a sustained position, not one good quarter. What the same table does not show is a widening one: the share oscillates inside a 31% to 41% band, the Q1 readings step down 41% (2024), 38% (2025), 33% (2026), and the 31% of Q2 2026 matches the lows already set in Q4 2024 and Q4 2025 rather than breaking them; Q2-on-Q2 the series runs 34%, 36%, 31%. Counterpoint also records MediaTek's shipments down more than 30% year-over-year in the quarter, the mainstream and entry tiers hit by the memory shortage and the premium tier by a 'slowdown in sales of Dimensity 9000 series powered devices.' MediaTek's own 2Q26 remarks show the gross line holding and the operating line not: gross margin 46.2%, down only 0.1 percentage points sequentially and guided at 46% plus or minus 1.5 for Q3, while operating income fell 22.2% year-over-year on revenue up 1.2% as operating expenses rose to NT$47.4 billion from NT$44.5 billion. A first place held for ten quarters without extending it, and earning less on it, is defensible rather than compounding. |
| 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。 | 无形资产与知识产权 MediaTek manufactures nothing itself -- its 2Q26 remarks describe 'deep design-technology co-optimization (DTCO) with TSMC' and close collaboration with advanced-packaging partners -- so the durable asset cannot be plant scale, and the remarks do not claim customer lock-in either. They locate it in a reusable design library: 'Building on our industry-leading IP portfolio, we offer best-in-class, pre-validated subsystems solutions for memory, I/O, and connectivity' -- which, the company says, 'largely reduces design complexity and shortens time to market for our data center customers'. The same remarks itemise 448G SerDes with a Co-Package Copper system solution, CPO development on TSMC's COUPE platform, and an end-to-end 3.5D platform with 'best-in-class 3.5D IP, packaging, and design flows,' and say design wins come 'backed by our industry-leading IPs, deep ecosystem partnerships, and proven execution capabilities'. The spending behind that library shows in the cost line: operating expenses of NT$47.4 billion in the quarter against NT$44.5 billion a year earlier, with the Q3 operating expense ratio guided at 31% plus or minus 2 points. It is the same library MediaTek re-amortises across mobile, edge and now data-center ASIC. |
| 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. | 并列领先者 Counterpoint has MediaTek ahead of every rival in all ten quarters of its public table -- 31% against Qualcomm's 23% and Apple's 19% in Q2 2026 -- but that count covers smartphone application processors only, and MediaTek's 2Q26 remarks put mobile phone at 41% of group revenue, no longer its largest business. Smart Edge Platforms is larger at 53%, and while the remarks claim 'continued market share gains across connectivity, computing and automotive products', no independent count of those positions appears in the sources used here. In data center MediaTek holds nothing yet: production of its first AI accelerator ASIC is scheduled for the fourth quarter of 2026, and 15% to 20% is a 2027 target raised from 10% to 15% a quarter earlier. Inside its core market the flagship tier is a stated plan rather than a claimed position -- a 2nm SoC launching in the third quarter and, 'together with ongoing engagements with global players, we will increase our presence continuously in the flagship smartphone market in 2027' -- while Counterpoint records MediaTek's premium-tier shipments declining on a 'slowdown in sales of Dimensity 9000 series powered devices.' |
| 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 The 2Q26 remarks describe price as cost pass-through rather than capture: 'As rising costs across the supply chain have become an industry-wide reality, we are taking pricing actions to ensure these increases are appropriately reflected in our product pricing', with the full-year aim, 'through our disciplined pricing strategy', to 'deliver our full year gross margin within the current quarter guidance range' -- holding a band, not widening it. The realised figures show that defence working: gross margin 46.2%, down 0.1 percentage points sequentially, with Q3 guided at 46% plus or minus 1.5 points. The 2.9-point year-over-year decline is not an erosion rate; the company states it 'was mainly due to a one-time benefit in the year-ago quarter.' Holding a mid-40s gross margin through a market the company expects to decline 'by about 15% in units this year' is genuine discipline, but passing rising costs through to defend a guided band is not the ability to price above it. |
| 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。 | MediaTek is the largest supplier of smartphone application processors by units and has been for as long as Counterpoint's public quarterly table runs. Across all ten quarters it covers, Q1 2024 through Q2 2026, MediaTek is first every time and never below 31%, while the nearest rival never clears 30%; in Q2 2026 it holds 31% against Qualcomm's 23%, Apple's 19%, UNISOC's 13%, Samsung's 9% and HiSilicon's 5%. That is a real, independently counted position, and it is the only part of MediaTek's business for which such a count exists in the sources used here. The same table is the clearest evidence that the position is being defended rather than extended. MediaTek's share moves inside a 31% to 41% band across the ten quarters, and the top of that band has come down -- 41% in Q1 2024, 38% in Q1 2025, 33% in Q1 2026 -- while the 31% of Q2 2026 matches the lows already set in Q4 2024 and Q4 2025 rather than breaking them; measured like for like against the same quarter a year earlier the series runs 34%, 36%, 31%. Counterpoint records MediaTek's shipments falling more than 30% year-over-year in the quarter -- hit in the mainstream and entry tiers by the memory shortage, and in the premium tier by a slowdown in Dimensity 9000 series devices -- while Samsung's Exynos climbed to 9% on Galaxy S26 base variants and mid-tier A-series wins. MediaTek's own 2Q26 remarks show the gross line holding and the operating line not: gross margin of 46.2%, down 0.1 percentage points sequentially and guided at 46% plus or minus 1.5 for the third quarter, against operating income down 22.2% year-over-year on revenue up 1.2%, with operating expenses of NT$47.4 billion against NT$44.5 billion a year earlier. Unit leadership held while the return on it did not. What MediaTek re-uses is design IP. It manufactures nothing itself -- the remarks describe deep design-technology co-optimization with TSMC and close work with advanced-packaging partners -- and they locate the durable asset in a pre-validated library of memory, I/O and connectivity subsystems, 448G SerDes with a Co-Package Copper system solution, CPO development on TSMC's COUPE platform, and an end-to-end 3.5D platform. The spending behind that library is visible in the cost line: operating expenses of NT$47.4 billion in the quarter against NT$44.5 billion a year earlier, with the Q3 operating expense ratio guided at 31% plus or minus 2 points against a gross margin guided at 46% plus or minus 1.5 -- an operating cost base running near a third of revenue. That library is now pointed at two adjacencies. Smart Edge Platforms -- connectivity, computing, automotive and TV SoCs -- is now the larger business, at 53% of revenue in Q2 against mobile's 41%, and grew 26% year-over-year while mobile fell 20%. And the company says it has built its first AI accelerator ASIC for a major US cloud customer, with production scheduled for the fourth quarter of 2026. One adjacency is a target and one is a contribution, and this profile treats them as exactly that. The data center share figure of 15% to 20% is a 2027 target against a served market MediaTek itself sizes at US$80 billion, raised from 10% to 15% only one quarter earlier, and production of the first ASIC was still scheduled for the fourth quarter at the date of the remarks. The RTX Spark AI PC work is corroborated by NVIDIA's own announcement, which credits MediaTek with collaborating on the custom CPU design and contributing to its power efficiency, performance and connectivity -- but the superchip, its Grace CPU, its NVLink-C2C interconnect and its software stack are NVIDIA's, and the systems ship from PC makers, so what MediaTek holds is a design contribution rather than a franchise with a share to count. The claimed share gains in connectivity, computing and automotive carry no number and no third-party count. The honest reading is a company with one durable, independently verified franchise it is holding rather than extending, funding a credible but still unproven move into higher-value silicon on the strength of its IP portfolio. |
| 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. | A fabless designer sitting between its customers and the TSMC-centred manufacturing chain. The 2Q26 remarks describe 'deep design-technology co-optimization (DTCO) with TSMC' and close work with advanced-packaging partners to build 'high-performance ASICs across a broad range of very-large chip sizes using CoWoS and EMIB-T technologies' at nodes down to 2nm, and say MediaTek creates 'significant value beyond semiconductor by orchestrating other key components throughout the supply chain such as memory and substrate' -- a coordinator role as much as a design one. |
| 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。 | +0.13 权重 0.20 · 群体判断 neutral 编辑先验，未经回测。 |