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.
| Nebius Group | Applied Digital | Marvell Technology | |
|---|---|---|---|
| Moat rating | narrow The FY2025 20-F positions Nebius as 'one of the few global, at scale, multi-tenant clouds purpose built for AI,' but the same filing names its central dependency plainly: 'We currently rely on Nvidia for the GPU chips we use' — a purpose-built neocloud renting a supply the hyperscalers it competes with also control, which is a real but narrow position. | none The FY2026 10-K (filed 2026-07-29) shows contracted revenue, not a demonstrated competitive edge. About 1,410 MW is leased under 15-year take-or-pay, non-cancellable base terms worth about $36.2 billion, but only about 100 MW of the roughly 1.5 GW that is contracted and either operating or under construction was operating and earning revenue at May 31, 2026, and Item 1A says "lessees may have the right to terminate applicable leases if there are significant delays in construction." Item 1A also concedes "We do not have the resources to compete with larger providers of similar products or services at this time," and the Competition section names 13 power-advantaged developers the company competes with. Signed leases give revenue visibility, but the filing does not show a durable advantage. | 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 |
| Moat type | cost scale The moat the filing describes is engineering: designs that 'optimize power and cooling efficiency, lower latency' plus 'a consistent track record of being one of the first-to-deploy the latest generation of NVIDIA GPU chips' — density and time-to-deploy at scale, not a switching lock. | none The 10-K claims three advantages: power-advantaged sites (it believes securing power and interconnection ahead of demand is 'the principal constraint on new HPC capacity and a core differentiator for us from many of our competitors'), a standardized 'franchise-style' design, and hyperscaler master service and master telecom service agreements 'that are difficult to obtain.' The filing does not show any of them to be durable. Its Competition section says competition 'centers on securing and developing sites with access to large-scale, reliable, and cost-competitive power and interconnection' and names 13 power-advantaged developers going after the same leases, and Item 1A concedes it lacks the resources to compete with larger providers. Signed leases are take-or-pay and non-cancellable, so a tenant leaving for convenience owes 'the full remaining contractual value,' but that is contractual lock-in on each lease rather than a moat source, so no moat type is assigned. | 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 |
| Leadership | fast follower The 20-F's own claim is 'one of the FEW global, at scale' AI clouds — a differentiated challenger to the hyperscalers it names as the competitive field, not a claimed leader of it; the first-to-deploy record is a follower's speed advantage, not category leadership. | behind The 10-K makes no leadership claim and gives no ranking or share figure. Item 1A concedes "We do not have the resources to compete with larger providers of similar products or services at this time" and that some rivals have "substantially greater liquidity and financial resources than we do." Its Competition section places APLD against established operators (Digital Realty, Equinix), hyperscalers that build their own capacity, independent developers and 13 named power-advantaged developers (IREN, Cipher Digital, TeraWulf, Hut 8, Riot, CleanSpark, HIVE, Core Scientific, Bitdeer, Galaxy Digital, Fermi, Keel Infrastructure, MARA). | 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. |
| Pricing power | weak The filing frames a 'highly competitive' market with 'frequent introduction of new or improved solutions' and a GPU cost base set by a single supplier (NVIDIA) — a renter of compute competing on efficiency has little list-price control. | weak Item 1A says "Due to the limited number of hyperscalers, we expect that a limited number of customers will continue to account for a high percentage of our revenue for the foreseeable future," and that if customers' equipment usage declines or they discontinue use of its facilities, APLD "may be compelled to lower our lease prices in some instances or risk losing a significant customer." One customer was 59% of FY2026 revenue from continuing operations. Take-or-pay, non-cancellable terms protect contracted revenue over the base term, and Note 19 reports a $39.1M HPC Hosting segment profit on $385.3M of segment revenue in FY2026, but those terms are agreed with a small group of concentrated buyers. | 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. |
| Summary | Nebius sells a 'unified full-stack AI cloud that spans the complete AI journey – from compute capacity to software and services,' with hardware and software built in-house — the neocloud pitch of hyperscaler reliability at purpose-built efficiency, and a first-to-deploy record on new NVIDIA silicon. Two things bound the moat, both from the filing: it 'currently rel[ies] on Nvidia for the GPU chips,' the same constraint every neocloud shares, and it is a Nasdaq 'Controlled Company' whose founding shareholder holds concentrated voting power. The three non-core segments (Toloka, Avride, TripleTen) are separate businesses, not the cloud moat. | Applied Digital designs, builds and operates purpose-built, liquid-cooled HPC data centers, which it calls 'AI factories', and leases the capacity to CoreWeave and investment-grade hyperscalers. At May 31, 2026 its 10-K lists five campuses (Polaris Forge 1-3 and Delta Forge 1-2) with about 1,410 MW contracted under roughly 15-year take-or-pay, non-cancellable leases worth about $36.2 billion over the base terms. The filing claims three sources of advantage: it controls power-advantaged sites, it uses a standardized 'franchise-style' design built to deliver about 150 MW in about 14 to 18 months, and it holds hyperscaler master agreements that are 'difficult to obtain.' The same document shows how early the company is. About 100 MW was operating and earning revenue. One customer was 59% of FY2026 revenue from continuing operations. It competes with Digital Realty, Equinix, hyperscalers that build their own capacity and 13 named power-advantaged developers, and it concedes that it lacks the resources to compete with larger providers. Signed leases give long-dated revenue visibility, but the filing does not show a durable competitive advantage. | 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. |
| Chain position | Layer-8 neocloud — a purpose-built AI compute provider reselling NVIDIA silicon at scale. | Developer and landlord of power-advantaged, liquid-cooled AI data-center capacity, leased long-term to CoreWeave and investment-grade hyperscalers. | 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). |
| Products (share / barrier) |
|
|
|
| Long-horizon vote | -0.01 at weight 0.20 · swarm neutral Editorial prior, not backtested. | -0.20 at weight 0.20 · swarm neutral Editorial prior, not backtested. | +0.13 at weight 0.20 · swarm bullish Editorial prior, not backtested. |