護城河比較
最多三家已涵蓋的公司，逐項對照。每一項判斷都是人工編輯的判斷，而非公司揭露的數字——而且每一項都附有其引用依據。
| Snowflake | Vistra | Applied Digital | |
|---|---|---|---|
| 護城河評級 | 窄 Narrow, not wide, because the FY2026 10-K documents a strong installed base and a competitive position the company itself says is under erosion. On the asset side: revenue of $4.7 billion (29% growth in each of the last three fiscal years), 13,328 total customers up from 10,996, 790 of the Forbes Global 2000 contributing about 43% of revenue, 733 customers above $1 million in trailing-12-month product revenue up from 576, a 125% net revenue retention rate, and more than 1,050 issued U.S. patents. Against that, Item 1A states plainly that adopting open data formats like Apache Iceberg means 'there is less customer “lock in” when our products are used in external environments' and that 'our support of open data formats may also reduce switching costs between us and our competitors'; that AWS, Azure and GCP 'generally compete in all of our markets' while also supplying the infrastructure a 'substantial majority of our business is run on'; and that the company remains loss-making at $1.3 billion of net loss for the year. | 窄 The 10-K states that "the majority of our facilities operate as “merchant” facilities without long-term power sales agreements" and that Vistra is "not guaranteed any rate of return on our capital investments". Against that, the scarcity is real: six NRC-licensed nuclear units totalling 6,448 MW, licences running 2036-2053, inside a 43,641 MW fleet, plus 20-year PPAs with AWS (1,200 MW) and Meta (2,609 MW). The 2025 gas additions - Lotus (2,600 MW, closed October 2025) and pending Cogentrix (5,500 MW) - extend the merchant gas side, not the nuclear scarcity. Only the 433 MW of uprates extends the moat asset. | 無 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. |
| 護城河類型 | 轉換成本 The filing makes its own affirmative claim of network effects — 'Our business benefits from powerful network effects. ... The more customers adopt our platform, the more data can be exchanged with other Snowflake customers, partners, data providers, and data consumers' — but the load-bearing, quantified evidence in the document points to switching costs. The platform is sold as the way to 'consolidate data into a single source of truth,' and the disclosed economics of that consolidation are a 125% net revenue retention rate and 733 customers above $1 million in trailing product revenue. Item 1A confirms the mechanism by naming what is at risk: open formats produce 'less customer “lock in”' and 'may also reduce switching costs.' The filing frames lock-in, not network density, as the thing erosion would take away. | 無形資產與智慧財產 The intangibles are non-replicable regulatory assets rather than patents: six nuclear licences (Comanche Peak 2050/2053, Perry 2046, Davis-Besse 2037, Beaver Valley 2036/2047), fuel "contracted to support all our refueling needs through 2030", section 45U credits "recognizing the value of existing carbon-free nuclear power", and TXU Energy, sold "for over 20 years" and "registered and protected by trademark law". Read 45U as a floor, not moat strength: the 2025 credit was $220m against $545m in 2024, and it "provides increasing levels of support as unit revenues decline". Efficient scale does not apply. | 無 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. |
| 領先地位 | 並列領先者 The 10-K contains no ranking, market-share figure, or claim of leadership, and it names no non-hyperscaler competitor by name. The band rests on disclosed scale — $4.7 billion of revenue, 13,328 customers, 9,060 employees across 36 countries — set against the filing's own statement that 'many of our competitors have substantially greater brand recognition, customer relationships, and financial, technical, and other resources than we do.' Co-leader among independent cloud data platforms; not a leader over AWS, Azure and GCP, which the filing says compete in all of its markets. | 並列領先者 Vistra describes itself as "one of the largest producers of power in deregulated markets in the U.S." with over 230 TWh generated, "one of the largest competitive power generators in the U.S. as measured by MWh of generation capacity", "one of the largest electricity generators in the U.S.", and "one of the largest competitive residential retail electricity providers". That hedged phrasing appears four times and is the strongest claim the filing makes. The 10-K names no competitor and assigns no rank, so co-leader is the ceiling the disclosure supports. | 落後 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). |
| 定價權 | 中等 The consumption model plus 125% net revenue retention shows real expansion pricing, and the filing argues it competes on 'pricing transparency and optimized price-performance.' But Item 1A limits how far that goes: competition 'may negatively impact our ability to acquire new customers ... put downward pressure on our prices and gross margins'; the company 'may not be able to ... offer as many discounts or free services as our competitors'; results depend on 'changes in our pricing model, including in response to significant price discounts by our competitors' and on 'customer optimization efforts that result in reduced consumption.' On the cost side, 'our costs and gross margins are significantly influenced by the prices we are able to negotiate with these public cloud providers, which in certain cases are also our competitors.' | 弱 Vistra is a merchant price taker. Price formation rests on "the highest variable cost unit that clears the market", prices are "unpredictable and may fluctuate substantially", hedging markets have "limited liquidity after two to three years", and competing retailers "may offer different products, lower electricity prices and other incentives". ERCOT's $2,000/MWh figure is the low system-wide offer cap, applied conditionally when the peaker net margin exceeds three times CONE or under the PUCT Emergency Pricing Program, not a standing cap. PJM has "announced that it would propose" extending its capacity cap to 2028-29 and 2029-30, subject to FERC approval. | 弱 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. |
| 綜述 | Snowflake's advantage in its FY2026 10-K rests on being the consolidation point for enterprise data: a multi-cluster shared-data architecture with proprietary columnar storage and automatic micro-partitioning, delivered across three major public clouds and 53 interconnected regional deployments, that customers adopt as a single governed source of truth and then expand on — 125% net revenue retention, 790 of the Forbes Global 2000 as customers. The filing layers a collaboration claim on top, with sharing 'generally without copying or moving the underlying data' and a Marketplace of 'hundreds of live, ready-to-query third-party data sets and data products.' The same document is unusually candid about the counter-pressure: Iceberg and open formats reduce lock-in by the company's own account, the three hyperscalers compete across every market while setting the cloud costs that 'significantly influence' gross margins, and frontier AI model providers 'may seek to vertically integrate ... by expanding into the data storage and management layers.' | Vistra's moat is one scarce asset wrapped in a commodity business. Six NRC-licensed nuclear units - 6,448 MW, licences running 2036-2053 - sit inside a 43,641 MW fleet that the 10-K says operates in the majority as "merchant" facilities with no long-term power sales agreements and no guaranteed rate of return. That block cannot be rebuilt by a rival, and is now partly de-risked by 20-year PPAs with AWS (1,200 MW from Comanche Peak) and Meta (2,609 MW from the PJM plants) plus section 45U credits. Everything else - 26,989 MW of gas, 8,743 MW of coal, the 5m-customer retail book - competes on price in markets Vistra does not set, against entrants the filing says keep building "despite relatively low power prices". The 2025 growth was gas M&A (Lotus, pending Cogentrix), which widens the commodity-exposed side. Narrow, for a specific reason: the moat is 15% of the fleet. | 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. |
| 產業鏈位置 | Sits at the enterprise data and governance layer of the AI stack, and the AI exposure is explicit rather than incidental: the filing brands the product the 'AI Data Cloud,' lists AI as a product category, and put Snowflake Intelligence, Cortex Agents and a Managed MCP Server into general availability during the fiscal year. It is a buyer of hyperscaler compute and of third-party frontier models — 'strategic partnerships with foundational model providers deliver state-of-the-art models natively within Snowflake Cortex AI,' with stated 'model neutrality' — and a supplier of governed enterprise data and GPU-backed managed compute to AI applications built on top. | Merchant IPP: sells energy, capacity and ancillary services into ISO/RTO spot and short-term wholesale markets (ERCOT, PJM, ISO-NE, NYISO, CAISO, MISO) and resells to ~5m retail customers. Emerging role as long-term nuclear offtake supplier to hyperscalers (AWS, Meta). Not a price setter at any link. | Developer and landlord of power-advantaged, liquid-cooled AI data-center capacity, leased long-term to CoreWeave and investment-grade hyperscalers. |
| 產品（市佔 / 門檻） |
|
|
|
| 長週期投票 | +0.13 權重 0.20 · 群體判斷 neutral 編輯先驗，未經回測。 | +0.05 權重 0.20 · 群體判斷 neutral 編輯先驗，未經回測。 | -0.20 權重 0.20 · 群體判斷 neutral 編輯先驗，未經回測。 |