护城河对比

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.

来源：sec.gov

Real but unbuilt. The Q2 2026 10-Q shows the moat resting on regulatory standing rather than delivered product: the NRC completed the Environmental Assessment for the Dow Long Mott construction permit application on 18 May 2026 with a Finding of No Significant Impact, yet no Xe-100 has been constructed or operated and two counterparties supplied nearly all income, with the U.S. Government at 90.3% and Dow at 4.4% of total revenue and grant income for the three months ended 30 June 2026.

来源：sec.gov

The FY2025 10-K grounds real stickiness — a trailing-12-month dollar-based net retention rate of "about 120%" as of December 31, 2025 and "approximately 84% of our customers were using two or more products" out of approximately 32,700 customers — but the same filing caps it. It names IBM, Microsoft and SolarWinds (on-premise infrastructure monitoring), Cisco, New Relic and Dynatrace (APM), Cisco and Elastic (log management) and "native solutions from cloud providers such as Amazon Web Services, or AWS, Microsoft Azure, and Google Cloud Platform" as competitors, plus "home-grown and open-source technologies", and concedes "many of our competitors have greater financial, technical and other resources, greater brand recognition, larger sales forces and marketing budgets". It further discloses an AI-native cohort "which cohort includes our largest customer and represented approximately seven percentage points of our year-over-year revenue growth for the quarter ended December 31, 2025" whose members "have rapidly increased their usage of our product and then optimized or may in the future optimize their usage". Sticky but bounded: narrow, not wide.

来源：sec.gov

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.

来源：sec.gov

The IPO prospectus describes an 'intellectual property-driven business model': revenue is to come from technology fees for licensing the proprietary Xe-100 design plus proprietary TRISO-X fuel, and the company states it will not construct, own or operate the power plants. The durable assets are therefore patents and NRC licences, not plant, installed base or scale.

来源：sec.gov

The 10-K locates the durable hold in platform integration rather than protected IP. A single agent collects "metrics, traces, logs, and other data"; under "One Data Model" every ingested datum is "consistently tagged with metadata regardless of its type", so different data types can be "queried together, correlated, alerted on, and visualized in a common user interface"; more than 1,000 out-of-the-box integrations bind it to the customer's stack; and the attach ladder deepens (approximately 84% of customers on two or more products, 55% on four or more, 33% on six or more and 18% on eight or more as of December 31, 2025). Displacing Datadog means re-instrumenting an estate the filing describes as "frequently deployed across a customer's entire infrastructure, making it ubiquitous".

来源：sec.gov

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.

来源：sec.gov

The prospectus records X-energy as one of two awardees under the DOE Advanced Reactor Demonstration Program, and it holds the Oak Ridge HALEU fuel-fabrication licence; but it is not the NRC design-approval leader — the NRC issued a standard design approval to NuScale for the US460 small modular reactor, published at 90 FR / Federal Register document 2025-10123 on 4 June 2025, while the Xe-100 is being licensed indirectly through a customer's Part 50 construction permit and holds no design certification or approval of its own.

来源：sec.gov

The 10-K claims only that "We believe that we compete favorably with respect to the factors listed above" — never category leadership — and names a distinct credible rival set in each category it serves, while conceding many of those rivals have greater resources and brand recognition. It does claim one first: being "the first to combine the 'three pillars of observability' - metrics, traces, and logs - into a single end-to-end platform" with log management in 2018. That reads as the leading independent among several credible rivals, not a clear leader.

来源：sec.gov

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.

来源：sec.gov

Nothing has been priced in a market yet. The prospectus states the Xe-100 technology fee is a benchmarking assumption to be 'negotiated on a case by case basis' with no assurance customers will accept the proposed structure, and that Amazon holds first-priority manufacturing queue slots across 2031-2039, a right of first refusal and most-favoured pricing; the 10-Q records a 2025 warrant issued to a related-party customer with a $162.7 million grant-date fair value accounted for as consideration payable to a customer, i.e. a reduction of the transaction price.

来源：sec.gov

Expansion is real but volume-driven rather than price-driven. The 10-K attributes the increase in trailing-12-month dollar-based net retention to about 120% (from "high-110%'s" a year earlier) to "increased usage growth from existing customers", and describes self-service expansion by "adding hosts or volumes of data monitored". The same filing warns that if customers "reduce their usage, fail to renew their subscriptions or renew on different terms", then "our revenue and dollar-based net retention may decline" — a usage-metered model hands the customer a dial that seat-based pricing does not.

来源：sec.gov

Layer-8 neocloud — a purpose-built AI compute provider reselling NVIDIA silicon at scale.

Upstream supplier of reactor technology and TRISO fuel to industrial-heat and data-centre power buyers; it licenses the design and fabricates the fuel rather than owning or operating the plants.

A software layer above the cloud rather than a supplier into it: the 10-K describes the platform as "cloud agnostic", deployable across "public cloud, private cloud, on-premise, multi-cloud, and hybrid environments", and monetizes the AI build-out through LLM Observability, which traces LLM chains and correlates them with APM.

• Nebius AI cloud (GPU compute + managed AI/ML platform) 挑战者 · 中等 来源：sec.gov
• Toloka / Avride / TripleTen (non-cloud segments) 未知 · 低 来源：sec.gov

• Reactor development and lifecycle services 细分市场 · 中等 来源：sec.gov
• TRISO-X HALEU fuel fabrication (TX-1, Oak Ridge) 挑战者 · 深 来源：federalregister.gov
• XENITH microreactor 未知 · 中等 来源：sec.gov
• Xe-100 small modular reactor 挑战者 · 深 来源：federalregister.gov

• Application Performance Monitoring (APM) 未知 · 深 来源：sec.gov
• Bits AI SRE 未知 · 中等 来源：sec.gov
• Cloud Security 未知 · 中等 来源：sec.gov
• Infrastructure Monitoring 未知 · 深 来源：sec.gov
• LLM Observability 未知 · 中等 来源：sec.gov
• Log Management 未知 · 中等 来源：sec.gov

编辑先验，未经回测。

查看它究竟如何投票 →

编辑先验，未经回测。

查看它究竟如何投票 →

编辑先验，未经回测。

查看它究竟如何投票 →