Compare moats
Up to three covered companies, band by band. Every call is a curated editorial judgment, never a disclosed figure — and every band carries its cited basis.
| Marvell Technology | SK Hynix | Fabrinet | |
|---|---|---|---|
| Moat rating | narrow 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 | narrow SK hynix held ~62% of the AI-critical HBM market in Q2 2025 (Counterpoint, via TweakTown) and was first to mass-produce HBM4, but memory remains a cyclical three-player business where Samsung and Micron are closing the HBM gap — a real but contestable edge. | narrow The FY2026 10-K grounds a real but bounded defence. On the defence side: Item 1 states that in the manufacturing services market "there are significant barriers to entry in our existing and target markets, including lengthy sales cycles, the need to demonstrate complex precision optical and electro-mechanical engineering and manufacturing capabilities to a prospective customer and the ability to protect a customer's intellectual property," and that qualification of a program "may take three to six months or longer to complete." On the limiting side, the same filing shows the protection does not reach price: gross profit was 12.0% of revenues in FY2026 against 12.1% in FY2025 and 12.4% in FY2024 even as revenue rose 35.7% to $4.64 billion, four customers each exceeded 10% of revenue (Cisco 19.9%, NVIDIA 16.3%, Nokia 10.7%, Amazon 10.5%), and "reliance on a small number of customers gives those customers substantial purchasing power and leverage in negotiating contracts with us." Sales are made on "individual purchase orders that have short lead times and are subject to revision or cancellation," so the qualified program is sticky while the contract is not. |
| Moat type | intangibles ip 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 | intangibles ip The advantage is process-and-packaging technology leadership: SK hynix achieved industry-first HBM4 mass production in September 2025 and has positioned itself as NVIDIA's key HBM partner, per its record FY2025 results. | switching costs The 10-K locates the stickiness in qualification rather than in patents or scale. Customers "do not purchase our services until they qualify the services and satisfactorily complete factory audits and vendor evaluations"; qualification "may take three to six months or longer"; production is transferred "copy-exact: the setup of a production process identical to the one used by our customer"; and the filing's own experience of changing a component source is that it "resulted in our customers or their end customers requiring requalification and validation of components, a process that can often be lengthy and has negatively impacted the timing of our revenue" — the same friction a customer would face moving the program elsewhere. The filing rules out an IP-based moat itself: "Historically, patents have not played a significant role in the protection of our proprietary rights," and any process improvement developed for a customer's product is "immediately assigned to that customer." The Thailand cost base is a genuine advantage but the filing warns it is not durable on its own — "Wage increases may impact our ability to sustain our competitive advantage and may reduce our profit margin." |
| Leadership | co leader 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. | clear leader Clear leader in HBM (#1 at ~62% share, Q2 2025) and became the world's top overall memory supplier in Q2 2025; caveat: it sits #2 to Samsung in commodity DRAM by volume and Micron/Samsung are narrowing the HBM gap heading into HBM4. | co leader The leadership claim in the filing is the company's own qualitative assessment: "We believe we are a leader in manufacturing products for the optical communications market," and, on breadth of process technology, "Based on our experience with customers and our qualitative assessment of our capabilities, we believe we provide a broader array of process technologies to the optics industry than any other manufacturing services provider." The same Item 1 names a crowded field against it — "Benchmark Electronics, Inc., Celestica Inc., InnoLight Technology (Suzhou) Ltd., Jabil Inc., Sanmina Corporation, Venture Corporation Limited and Eoptolink Technology Inc., Ltd., as well as the internal manufacturing capabilities of our customers" — and no independent share ranking is given, so the record supports front-of-field standing but not sole primacy. |
| Pricing power | moderate 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. | strong FY2025 operating margin hit a record 49% (operating profit +101% YoY) as HBM revenue more than doubled — unusually strong pricing power for a memory maker, per SK hynix's FY25 release. | weak Gross profit was 12.0% of revenues in FY2026, 12.1% in FY2025 and 12.4% in FY2024 — flat to down across three years in which revenue grew from $2.88 billion to $4.64 billion — and the FY2026 increase in gross profit is attributed to "sales volume and product mix," not price. Item 1 says so directly: "we expect the prices we charge for our manufactured products to decrease over time (partly as a result of competitive market forces)," with the offset coming from cycle-time, mix, yield and material-cost work rather than from rate. The risk factors add that customer consolidation gives buyers "increased leverage that may result in, among other things, decreases in our average selling prices," and that new competition "could result in price reductions for our services, reduced gross profit margins or loss of market share." |
| Summary | 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. | SK hynix's durable advantage rests on an HBM process-and-packaging lead: it held roughly 62% of the HBM market in Q2 2025 (Counterpoint) as NVIDIA's key HBM partner, and delivered industry-first HBM4 mass production. That lead drove record FY2025 results with a 49% operating margin, though rivals are qualifying competing HBM parts. | Fabrinet is the neutral outsourced factory for complex photonics. The FY2026 10-K describes a business that wins on qualification and IP hygiene rather than on product ownership: a "factory-within-a-factory" that physically segregates each customer's engineers and floor space, copy-exact line transfers, Telcordia-grade environmental qualification, and a claim that "there is no other manufacturing services provider with a similar breadth and depth of optical and electro-mechanical engineering and process technology capabilities that does not directly compete with its customers in their end-markets." That neutrality is claimed as a belief, not demonstrated — the filing's own framing is "we believe" — and Item 1 names the field it is measured against without characterising any of them: "Benchmark Electronics, Inc., Celestica Inc., InnoLight Technology (Suzhou) Ltd., Jabil Inc., Sanmina Corporation, Venture Corporation Limited and Eoptolink Technology Inc., Ltd., as well as the internal manufacturing capabilities of our customers." Once a program is qualified it tends to stay, and "in many cases, we are the sole outsourced manufacturing partner used by our customers for the products that we manufacture for them." What the moat does not buy is margin. Revenue grew 35.7% in FY2026 to $4.64 billion on the AI-datacenter build-out — data center products are now 47.9% of revenue — yet gross margin slipped to 12.0%, and Item 1 concedes the company expects "the prices we charge for our manufactured products to decrease over time." The two live erosion paths are named in the risk factors: customer consolidation that "may result in, among other things, decreases in our average selling prices," and customers who "acquired the capacity to manufacture products in-house." Concentration cuts both ways this year — FY2025 had two customers above 10% of revenue, FY2026 had four, so the book broadened even as its scale grew. |
| Chain position | 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). | Primary supplier of HBM (high-bandwidth memory) stacked onto NVIDIA's AI accelerators; HBM represented roughly 77% of Q2 2025 revenue. | A contract manufacturer sitting between optical components and the network and AI-datacenter OEMs whose brands ship the finished box. FY2026 revenue is 47.9% data center, 33.3% communications infrastructure and 18.8% automotive, industrial and other markets (FY2025: 46.2% / 30.7% / 23.1%), with Cisco, NVIDIA, Nokia and Amazon each above 10% of revenue. Fabrinet also integrates one layer down, designing and fabricating its own customized optics and glass — crystals, ferrules, precision glass tubing — at Fuzhou, China and Mountain Lakes, New Jersey, both for its own assemblies and for the merchant market, while volume manufacturing runs from the Pinehurst and Chonburi campuses in Thailand. |
| Products (share / barrier) |
|
|
|
| Long-horizon vote | +0.13 at weight 0.20 · swarm bullish Editorial prior, not backtested. | +0.24 at weight 0.20 · swarm bullish Editorial prior, not backtested. | +0.05 at weight 0.20 · swarm neutral Editorial prior, not backtested. |