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。
| Nebius Group | Hut 8 | Salesforce | |
|---|---|---|---|
| 护城河评级 | 窄 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. | 窄 The FY2025 10-K says 'we believe we have established a defensible competitive advantage through our power-first, innovation-driven strategy', resting on a power-native team, an application-agnostic framework for digital infrastructure design, end-to-end greenfield development capabilities, and the use of ASIC compute builds to rapidly and cost-effectively secure and monetize power. What separates this from no moat is scarce power: the filing says demand for energy capacity continues to outpace supply and that 'grid interconnection bottlenecks have further constrained access to power and digital infrastructure development', and Hut 8 reports 330 MW of utility capacity under construction at River Bend, commercialized in December 2025 through a 15-year triple-net lease supporting 245 MW of IT capacity, plus 1,230 MW of utility capacity under development. The same filing says 'The industries in which we operate are highly competitive and continuously evolving.', warns that these factors 'might not provide the competitive advantage we anticipate, or if they do, such competitive advantage might not endure', cites the River Bend lease (valued at $7.0 billion over the base term) as an example of expecting a significant portion of revenue from a limited number of customers, and says the company has not maintained consistent profitability. That supports narrow at most, not wide. | 窄 Salesforce's own language caps this at narrow: Item 1 calls its market "highly competitive, rapidly evolving and fragmented, and subject to changing technology with low barriers to entry." What holds it above none is scale and contract duration: $72.4B total RPO at 2026-01-31 versus $63.4B (+14%), current RPO $35.1B versus $30.2B (+16%), on revenue of $41.5B (+10%). But cRPO growth "was positively impacted by three percent" by currency, and total RPO "includes approximately $2.2 billion" from Informatica, acquired November 2025. Adjusted, the contracted base still outgrows revenue, but by less than the headline implies. |
| 护城河类型 | 成本规模 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. | 成本规模 The 10-K says 'we believe success depends on the ability to secure scarce assets like power, data center equipment, and customers', and places part of its claimed edge in 'our ability to use ASIC compute infrastructure development to rapidly and cost-effectively secure and monetize power'. Its examples: Vega was energized less than a year after acquisition for an all-in cost of about $455,000 per megawatt, which the filing calls 'a fraction of traditional data center costs', and Salt Creek was completed for about $250,000 per MW; Vega also has immediate access to some of the lowest locational wholesale power prices in North America. These are low-redundancy ASIC compute builds, which the 10-K says carry relatively low capital intensity by design. The 'fraction of traditional data center costs' comparison is against a different facility tier, and the filing gives no peer cost comparison and no cost figure for its AI campuses. | 转换成本 Switching costs, with a caveat: the filing's clearest lock-in sentence is about rivals' customers, not its own. "Many prospective customers have invested substantial personnel and financial resources to implement and integrate their current enterprise software... may be reluctant or unwilling to migrate away" is a headwind to Salesforce's own growth. Evidence for its own base is architectural: Data 360 unifies data into "a single, governed source of truth," Agentforce runs on existing processes "without requiring customers to remove and rebuild," behind $72.4B of contracted revenue. Contracts run "typically 12 to 36 months" with no renewal obligation. |
| 领先地位 | 快速追随者 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. | 快速追随者 The 10-K says 'For example, we signed our first large scale single tenant AI data center lease and commenced construction of the site at our River Bend campus.', offered under the statement that to remain competitive with peers it may need to modify aspects of its business model. River Bend was commercialized in December 2025 through a 15-year triple-net lease supporting 245 MW of IT capacity; Fluidstack is expected to serve as tenant and Google is expected to provide a financial backstop, and initial delivery is targeted for Q2 2027. On speed, the filing says 'Through early engagement with Entergy Louisiana, we assumed key interconnection and development functions typically managed by the utility, such as the development of transmission and distribution infrastructure, materially accelerating originally quoted power delivery timelines.', and that Vega was energized 'less than a year after acquiring it'. It also warns that Hut 8 competes against companies 'that may be more established or have greater financial and other resources and/or expertise'. That describes a newer entrant moving quickly, not an established leader. | 明确领先者 Clear leader in the category as it exists today, runner-up in the one it is being rebuilt into. Per IDC's Worldwide Semiannual Software Tracker on 2025 data, Salesforce was first in worldwide CRM applications for a thirteenth consecutive year at 20.0% share, and first in Sales (fourteenth year), Customer Service (thirteenth), Marketing (seventh), model-driven application platform (eleventh) and Integration (fourth). These come from Salesforce announcing its own IDC results via a trade outlet, in translation, not IDC primary material. In the same tracker's new Agent Build and Deploy segment it ranked second at 17.5%, albeit as fastest-growing vendor. |
| 定价权 | 弱 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. | 弱 The 10-K says ASIC compute, run through American Bitcoin, can only be profitable if hardware and electricity costs are below the price of the Bitcoin mined, which makes it a price-taking business. It also says Traditional Cloud and AI Cloud compete with cloud services providers for customers, and that the company has not maintained consistent profitability. River Bend's contracted triple-net lease economics are not in service yet; initial delivery is targeted for Q2 2027. | 中等 Gross margin rises across the Item 7 table — gross profit of $32,255M on $41,525M of revenue, 78% of total revenues in FY2026 against 77% in FY2025 and 75% in FY2024 — yet MD&A denies price as the cause: growth "was primarily caused by volume-driven increases from new business," and "Pricing was not a significant driver of the increase in revenues for the period." Item 1A names consumption-based pricing among attrition factors, says Agentforce and Data 360 monetization "remain relatively new and uncertain," and concedes "limited experience with determining optimal pricing for our consumption-based contracts." |
| 综述 | 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. | Per its FY2025 10-K, Hut 8 is a power-first infrastructure developer. It originates powered land and interconnections, uses ASIC compute builds, which the filing says combine relatively low capital intensity with rapid deployment, as a transitional load to monetize sites sooner, and aims to transition suitable power assets to higher-return uses; its River Bend campus was commercialized through a 15-year AI data center lease. The filing claims a defensible advantage but warns it might not provide the advantage anticipated, or might not endure. It says Hut 8 competes for powered land with digital infrastructure developers and large-scale Bitcoin miners, and its first AI campus is still under construction. Its majority-owned Bitcoin miner, American Bitcoin, can only be profitable when mining costs are below the price of the Bitcoin mined, and the filing says it believes American Bitcoin must keep acquiring miners to keep up with a growing global network hashrate. | Salesforce is the incumbent of record in CRM applications and is paid like one: $41.5B of FY2026 revenue (+10%), 78% gross margin, $72.4B of contracted revenue not yet recognized, and first place in worldwide CRM for a thirteenth consecutive year at 20.0% share per IDC's tracker as announced by Salesforce. Enterprise data and workflow embedded through Data 360 and Agentforce is costly to unwind, and the agentic layer is deliberately built atop existing processes rather than requiring re-implementation. The limits are the company's own: it twice describes its market as having low barriers to entry, names AI-native startups whose solutions "may bypass traditional business process workflows or displace established user interfaces," ranks second at 17.5% in IDC's new Agent Build and Deploy segment, and concedes pricing did not drive FY2026 growth. |
| 产业链位置 | Layer-8 neocloud — a purpose-built AI compute provider reselling NVIDIA silicon at scale. | Power-first developer building AI data center capacity, beginning with River Bend under a 15-year lease (Fluidstack is the expected tenant and Google is expected to provide a financial backstop; initial delivery is targeted for Q2 2027), and runs Bitcoin mining through majority-owned American Bitcoin. | Applications-layer incumbent: sells enterprise CRM and agentic software to end customers across industries, consumes cloud infrastructure (third-party IaaS partners plus its own Hyperforce estate) and third-party LLMs, and sits upstream of no hardware supply chain. Exposure is to enterprise software budgets and to whether the agentic layer is captured by incumbent suites or AI-native entrants. |
| 产品（份额 / 壁垒） |
|
|
|
| 长周期投票 | -0.01 权重 0.20 · 群体判断 neutral 编辑先验，未经回测。 | -0.01 权重 0.20 · 群体判断 neutral 编辑先验，未经回测。 | +0.20 权重 0.20 · 群体判断 neutral 编辑先验，未经回测。 |