初创公司尽调
尽调报告 Edge Infrastructure / Sovereign AI Series B 2026-05-24

Armada

主权边缘基础设施尽调 — 面向偏远和受监管环境的模块化 AI 数据中心

Armada 抓住了可信的主权边缘基础设施切口,早期部署证据异常扎实;但经济性不透明、工厂级放量风险高,2026 年价格也偏紧,因此仍应继续研究。

封面要素

最新轮次 01
$230M Series B [CO021]
最新估值 02
2000 USD M [CV001]
累计融资 03
465 USD M [CO023]
成立时间 04
Late 2022 [CO001]
总部 05
San Francisco [CO003]
FY25-FY26 订单额增长 07
540 % [CO025]
估值判断 08
Stretched [CV043]

公司概况

Armada 是一家总部位于旧金山的私营边缘基础设施公司,由 Dan Wright 和 Jon Runyan 于 2022 年底创立。公司销售全栈主权边缘平台,覆盖 Galleon 模块化数据中心、Atlas 车队与连接管理、Bridge GPU 编排,以及面向第一方和第三方应用的 Marketplace。公开牵引力最强的部分,是与 U.S. Navy、Alaska DOT&PF、Washington DNR 和 Aker BP 的具名部署,以及 Microsoft、Carahsoft 和 Johnson Controls 提供的商业化与制造支持。Armada 2026 年 5 月 Series B 将投前估值定在 $2.0 billion,披露融资总额约 $465 million,但核心经济性、客户广度和工厂执行细节大多仍未公开。

官网
www.armada.ai
创始人
Dan Wright, Jon Runyan
创立地点
San Francisco, USA
总部
San Francisco, USA
产品
Armada 的产品是一套全栈边缘基础设施平台:Galleon 加固型模块化数据中心、Atlas 车队 / 连接管理、Bridge 本地 GPU 编排与 GPUaaS 软件,以及一个面向应用和合作伙伴硬件的 Marketplace,服务偏远或受监管环境中的主权 AI。
客户
国防、公共部门、海上能源、电信及其他工业运营方;这些客户需要在传统集中式云之外获得低延迟、主权化或离线运行的 AI 和数据基础设施。
商业模式
硬件加软件收入模式,把 Galleon 系统销售与部署,同 Bridge 和 Atlas 软件、Marketplace 与编排层,以及 Azure Marketplace、Carahsoft 等伙伴主导的采购渠道结合起来。
阶段
Series B private company
融资情况
2026 年 5 月以 $2.0 billion 投前估值完成 $230 million Series B;该轮之后,披露融资总额达到约 $465 million。
[CO001, CO003, CO004, CO006, CO010, CO021, CO023, CU001]

执行摘要

主要优势

  • Armada 的全栈产品有差异化,覆盖坚固型模块化数据中心、舰队 / 连接管理、GPU 编排和应用市场。
  • U.S. Navy、Alaska DOT&PF、Washington DNR 和 Aker BP 等公开部署案例说明,产品确实在解决断网和恶劣环境问题。
  • Microsoft、Carahsoft 和 Johnson Controls 明显补强采购通道、主权云可信度和制造能力。
  • 主权与边缘 AI 基础设施需求加速时,2026 年 5 月 Series B 给了 Armada 可观资本和外部背书。

主要风险

  • 公开披露仍缺收入、ARR、毛利率、积压订单转化和现金跑道,预订增长无法对应到可持续经济性。
  • Forge One、Leviathan 以及更大范围的制造爬坡带来高执行、供应商、散热和营运资本风险。
  • 客户证据仍集中在少数国防、公共部门和工业案例,对 Microsoft、Carahsoft 和 Johnson Controls 的依赖也不低。
  • 主权 AI 扩张会抬高出口管制、网络安全和采购合规负担,但公开材料只露出一部分。
  • 仅看公开证据,2026 年 5 月估值偏高;若软件附加率或收入转化不及预期,估值可能重估。

未决问题

  • 未公开已确认收入、ARR、毛利率、EBITDA、烧钱速度、现金跑道,或从预订到收入确认的桥接。
  • 未公开客户数、续约或留存指标,也未按金额披露账户级集中度。
  • Forge One 和 Leviathan 生产的吞吐量、良率、capex 责任和营运资本细节均未公开。
  • 未公开董事会构成、优先股堆叠或详细治理权利。
  • 非 Atlas 安全认证、完整合规包,以及律师级诉讼或事件尽调的公开细节有限。

目录

Chapter 01

01公司概览

1.1 身份、平台与商业模式

Armada 由 Dan Wright 和 Jon Runyan 于 2022 年底创立,2023 年 12 月走出隐身状态,总部位于旧金山。公开材料一直把公司定位为私营、全栈边缘基础设施企业,主张把计算、存储、连接和 AI 推近数据产生地,以此「弥合数字鸿沟」,而不是让所有负载都回流到集中式云。2026 年 5 月 Series B 之后,这一定位已经和公司阶段绑定:Armada 不再只是在销售加固型边缘计算产品,而是在推一个更宽的主权 AI 平台,面向工业、国防、公共部门和受监管负载。 到 2026 年,运营栈比发布初期清晰得多。Armada 当前组合包括:Atlas 用于监测和管理联网资产,Galleon 用于加固型模块化数据中心部署,Marketplace 用于在边缘部署第三方硬件和软件,Bridge 用于把 GPU 集群编排并货币化为主权 AI 工厂。合起来看,商业模式是硬件加软件,而不是纯 SaaS:Armada 销售可部署基础设施,再叠加控制软件、应用分发和伙伴集成,让客户在连接不稳定或主权要求严格的地方运行推理、分析或私有云负载。[CO001, CO002, CO003, CO004, CO005, CO006]

关键 KPI 快照表
指标数值 / 状态日期置信度缺口 / 注意事项
总部San Francisco, CA2026-05-19CNBC 人物报道和当前公开材料共同佐证
阶段未上市 Series B2026-05-19最新披露融资为 2026 年 5 月 Series B
最新估值$2.0B2026-05-19Series B 估值
已披露累计融资$465M / 近 $500M2026-05-19CNBC 给出精确数字;官方来源做了取整
2024 年战略轮$40M2024-07-11由 M12 领投,并与 Azure Marketplace 上线绑定
2025 年战略轮$131M2025-07-24与 Leviathan 发布同一时间
计划工厂面积最高 400,000 sq ft2026-05-19Arizona 的 Galleon Forge One
计划工厂岗位5002026-05-19公司和 Johnson Controls 估计
FY25-FY26 订单额增长540%2026-05-19公司披露的订单额指标,不是经审计收入
FY27 Q1 同比订单额增长~2,000%2026-05-19公司披露的订单额指标
全球部署足迹43 个国家2024-07-11代表已披露地理覆盖,不代表当前客户数
具名部署证明U.S. Navy、Alaska DOT&PF、Washington DNR 与 Aker BP2025-12 至 2026-05示例性参考,不是完整客户名单
当前收入 / ARR2026-05-24已抓取 2026 年公开披露中没有
当前员工数2026-05-24已抓取 2026 年公开披露中没有

快照混合了官方公告和第三方佐证报道。空值表示已抓取 2026 年来源集中没有公开披露该指标;应在管理层材料中索取,而不是自行推算。

[CO003, CO004, CO017, CO018, CO019, CO021]
FO002: 公司快照逻辑

Armada 把模块化硬件、连接管理、合作伙伴软件和战略资本接成一套主权 AI 技术栈,服务远程和受监管环境。

[CO005, CO006, CO007, CO008, CO009, CO010]

1.2 创始人、领导层与治理

Armada 的公开叙事由创始人驱动。Dan Wright 是联合创始人兼 CEO,也是融资、使命叙事和外部合作中最显眼的公众面孔;公开资料反复把 Armada 的起源故事锚定在 Wright 此前在 AppDynamics 和 DataRobot 的运营岗位上。联合创始人兼 COO Jon Runyan 则带来 Okta 的企业法务和公司建设经验,在 Armada 早期产品和投资人沟通中,扮演 Wright 商业化画像的运营补位。技术侧,Pradeep Nair 一直被标注为创始 CTO,Armada 官方资源页和 Forbes 报道则将 Prag Mishra 标注为 Chief AI Officer。 这组领导班子对企业销售和应用 AI 商业化有利,但也带来集中风险。Wright 贯穿 CNBC、投资人、伙伴和公司材料,是关键公众面孔;一旦领导层出现扰动,融资、招聘和战略叙事很可能受到不成比例的影响。治理披露仍是薄弱环节。已抓取的 2026 年公开材料能支撑高管画像,但没有给出清晰的现任董事会名单、委员会结构或投资人控制权视图。因此,后续阶段尽调可以把公开领导层可见度视为足以做团队映射,但不足以完成完整治理或控制权分析。[CO011, CO012, CO013, CO014, CO015, CO016]

领导层和创始人表
人物角色背景职能覆盖 / 创始人-市场匹配关键人物依赖
Dan Wright联合创始人兼 CEO曾任 DataRobot CEO、AppDynamics COO企业级运营负责人、融资者、外部发言人和任务架构师关键
Jon Runyan联合创始人兼 COO曾在 Okta IPO 前后担任总法律顾问覆盖运营、法律和企业结构搭建,支撑受监管基础设施公司规模化
Pradeep Nair创始 CTO曾任 VMware 和 Microsoft Azure 工程负责人核心平台架构、模块化计算设计和产品执行
Prag Mishra首席 AI 官曾任 Amazon Health AI/ML 负责人、Microsoft 研究负责人应用 AI、模型策略和边缘用例的工作负载转化中高

枚举覆盖了 Armada 公开材料和为本章抓取的一线报道中反复出现的创始人及具名技术 / AI 负责人。公开来源对董事会构成和委员会结构仍然披露稀少。

[CO011, CO012, CO013, CO014, CO015, CO016]

1.3 资本基础、投资方与工业化扩张

2024 至 2026 年间,Armada 的资本形成明显提速。2024 年 7 月,由 M12 领投的战略轮融资 $40 million,把披露融资总额推高到 $100 million 以上,也让公司和 Azure Marketplace 分发绑定得更紧。2025 年 7 月,Armada 在发布 Leviathan 这一兆瓦级模块化数据中心产品的同时,宣布 $131 million 战略融资。到 2026 年 5 月,公司以 $2 billion 估值完成超额认购的 $230 million Series B;CNBC Disruptor 50 资料列示总融资为 $465 million,Armada 和 Wilson Sonsini 则把累计金额描述为接近半十亿美元。 投资人组合也不只是变大,而是更具战略性。Founders Fund、Lux Capital、Shield Capital、8090 Industries 等早期和复投方之外,又加入或强化了 M12、Veriten、Glade Brook、Overmatch、BlackRock 和 Johnson Controls。Johnson Controls 这条线在运营上最重要,因为它超出了融资:双方宣布 Global Framework Agreement,并计划在 Arizona 建设最多 400,000 平方英尺、约 500 个岗位的工厂 Galleon Forge One。这给 Armada 的主权 AI 叙事配上了工业制造故事,但也把执行、营运资本需求和供应链纪律的门槛抬高了。[CO017, CO018, CO019, CO020, CO021, CO022]

利益相关方 / 投资者图谱
利益相关方角色记录在案的进入点控制 / 经济重要性尽调问题
M12战略投资者和渠道伙伴领投 2024 年 7 月 $40M 轮把 Armada 连接到 Azure Marketplace 采购和 Microsoft 商业化验证由 Azure credits 与 MACC 式支出带动的销售管线
Founders Fund早期且重复下注的风险投资方在早期融资和 2025 年后续跟投中被具名长周期支持者,具备国防和基础设施可信度厘清董事会权利、储备资金和未来资本需求胃口
Lux Capital早期风险投资方在早期融资和 2025 年后续跟投中被具名释放对硬科技和主权基础设施逻辑的支持信号测试其支持制造重型规模化的意愿
8090 Industries重复投资方和 Series B 共同领投方早期参与,并共同领投 2026 年 5 月 Series B跨多轮融资的重要重复信号了解持股集中度和治理条款
OvermatchSeries B 共同领投方共同领投 2026 年 5 月 Series B对最新估值重置和成长融资有影响力询问其在董事会或观察员结构中的角色
BlackRock新战略金融投资者共同领投 2026 年 5 月 Series B在风险投资之外增加机构资本市场信号评估参与是战略性、财务性,还是两者兼有
Johnson Controls战略投资者和制造伙伴加入 2026 年 5 月 Series B 和工厂协议对工厂、热系统和部署规模化至关重要审阅排他性、定价和供应链依赖条款
Veriten能源相关战略投资者在 2025 年 7 月 $131M 轮和 2026 年现有投资者名单中被具名有助于闲置能源和工业部署逻辑检查商业引荐和任何能源市场集中度
Glade Brook成长投资者在 2025 年 7 月轮和 2026 年现有投资者名单中被具名战略轮与规模化轮之间的延续性确认按比例跟投支持和持股水平
Mitsui / Singtel Innov8新战略投资者2026 年 5 月 Series B 中被具名潜在亚洲和工业分销杠杆判断这些关系是否转化为已签部署

图谱覆盖已审阅 2024-2026 融资和制造公告中明确披露的具名投资者与战略利益相关方。它不是完整股权结构表,也不披露持股比例、清算优先权或观察员权利。

[CO017, CO018, CO019, CO021, CO022, CO023]
FO003: 快照 KPI

截至 2026 年 5 月,Armada 的可投资性靠几项东西支撑:大额已披露资本、快速订单额增长、具名部署证明,以及核心运营指标上仍然重要的披露缺口。

KPI 图是面向尽调的摘要,不是管理层仪表盘。它把硬数字和披露缺口数量放在一起,突出订单额动能已经超过当前公开指标透明度。

[CO021, CO023, CO024, CO025, CO028, CO030]

1.4 部署、伙伴与商业足迹

Armada 的商业证明,在具名部署和行业宽度上强于传统 SaaS 指标。公司 2024 年披露,客户已经把其技术带到 43 个国家;到 2025-2026 年,公开来源点名了与 U.S. Navy、Alaska DOT&PF、Washington DNR、Aker BP 及其他工业或公共部门运营方的标杆部署。用例内部一致:在延迟或连接性会打断云工作流的地方处理数据,再把 AI 或自动化用于海上钻井平台、野火作业、国防演练和偏远现场基础设施等环境。 客户案例呈现出同一种模式。Alaska DOT&PF 使用 Atlas 和 Galleon,把无人机影像周转从多日或 28 小时以上缩短到当天或实时输出。Washington DNR 用 Atlas 集中管理约 45 台 Starlink,支撑野火响应和远程作业。Armada 参与 UNITAS 2025,把 Galleon 和 Atlas 带入 U.S. Navy 的岸上和舰上演练;Aker BP 则同意在挪威大陆架部署一套 Galleon,用于海上钻井工作流。围绕这些部署,Armada 搭起了包括 Microsoft、NVIDIA、Palantir、Dell、Skydio 以及 Carahsoft 关联公共部门渠道的伙伴层,强化了公司的核心主张:它销售的是系统化的主权 AI 能力,而不是单个盒子或一张软件许可。[CO026, CO027, CO028, CO029, CO030, CO031]

1.5 里程碑与尽调警示

Armada 的里程碑推进异常紧凑。公司从 2022 年底创立,到 2023 年 12 月退出隐身,再在 2024 年叠加 Microsoft/M12 战略融资,2025 年伴随 $131 million 融资发布 Leviathan,同年展示海军和公共部门部署,到 2026 年 5 月又把 $230 million Series B 与 Arizona 制造扩建并行。这是从品类创建到工业化扩张的快速切换,也给后续市场、客户和估值分析提供了有用时间线。 同一条时间线也暴露出主要风险。第一,Armada 资本强度高:公司制造模块化数据中心,不只是授权软件;最新增长故事取决于工厂产能爬坡和持续获得战略资本。第二,当前收入、ARR、员工数、客户数、董事会构成和投资人控制条款的公开披露仍有限,意味着订单额增长还不能等同于经审计的运营成熟度。第三,创始人过往包袱并非完全从公开记录中消失:Forbes 和 CNBC 仍持续提到 Dan Wright 离开 DataRobot 一事,尽管近期动能很强,声誉和治理尽调仍然相关。最后,已抓取的公开材料没有发现针对 Armada 的诉讼或执法行动,但这不等于没有案卷级审查也能出具干净的法律健康证明。[CO001, CO002, CO017, CO019, CO021, CO024]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2022-Q4Armada 于 2022 年底成立创立公司成立Dan Wright;Jon Runyan当前边缘基础设施逻辑的起点
2023-12Armada 走出隐身模式创立披露 >$55M 早期融资创始人;早期投资者公司从概念转向公开运营姿态
2024-01Commander Connect 和创始人愿景内容发布产品初始软件控制层定位Dan Wright;Jon Runyan;Armada 团队显示商业化路径是软件加硬件,而非纯硬件
2024-07-11宣布 M12 领投战略轮,并上线 Azure Marketplace融资$40M;当时累计融资超过 $100MM12;Microsoft;Armada加深战略云分销和采购杠杆
2025-07-24Armada 宣布 $131M 战略轮并发布 Leviathan融资$131MPinegrove;Veriten;Glade Brook;现有投资者推动 Armada 从加固边缘盒子走向兆瓦级 AI 基础设施
2025-12-04Armada 携 Galleon 和 Atlas 参与 UNITAS 2025规模化海上和岸上运营演示U.S. Navy Fourth Fleet;Microsoft;行业伙伴验证国防部署叙事
2025-12-16DCD 公开报道 Alaska DOT&PF 部署规模化决策窗口从 28+ 小时缩短到近实时Alaska DOT&PF;Armada展示公共部门工作流压缩,不只是硬件部署
2026-03-23Aker BP 签署海上 Galleon 部署协议合作初始参考部署Aker BP;联盟伙伴;Armada如果首个安装成功,将形成可复制海上蓝图
2026-03-31Armada 与 Microsoft 推出 Azure Local 主权 AI 合作合作现已可用Armada;Microsoft强化受监管行业和国防定位
2026-05-19宣布 Series B,并披露 Johnson Controls 制造协议融资$230M,估值 $2BOvermatch;BlackRock;8090;Johnson Controls 等为工业化规模化提供资金,并把 Armada 重新框定为制造支撑的 AI 基础设施公司
2026-05-19披露 Galleon Forge One 工厂计划规模化最高 400,000 sq ft;500 个岗位Armada;Johnson Controls把未来增长绑定到工厂执行和供应链表现
2026-05-24公开治理和法律披露仍不完整反向董事会名单、完整运营指标和案卷级法律审查不可得仅限公开来源集在承销治理或法律风险前,需要私有尽调材料

这是从 2022 年底到本章运行日期,已抓取来源集中浮现的单一公开里程碑时间线。最后一行记录披露缺口,因为没有发现更干净的公开治理或法律里程碑。

[CO001, CO002, CO017, CO018, CO019, CO021]
FO001: 公司里程碑时间线

Armada 从 2022 年底创立走到 2026 年 5 月工厂规模融资,大约用了三年半;在完整公开运营指标披露前,国防、公共部门和海上部署已经出现。

融资和工厂节点精确到公开公告日期;公开材料只支持 2022 年底这个时间,因此创立节点按季度粒度展示。

[CO001, CO002, CO017, CO019, CO021, CO024]

1.6 图表

Chapter 02

02市场分析

2.1 市场边界、纳入支出,以及为什么这不是通用边缘计算

Armada 的市场边界从问题定义出发,而不是从一个泛化基础设施品类出发。公司明确销售四部分边缘栈——Atlas、Galleon、Bridge 和 Marketplace——让客户把 AI 就绪计算、编排、连接控制和主权运行环境直接带到传统数据中心建设周期或公有云假设失效的地方。因此,纳入支出不是所有 AI 基础设施,也不是所有边缘支出,而是更窄的一组预算:为恶劣、离线或受监管场景采购加固型模块化数据中心硬件、本地供电与冷却、边缘编排、安全连接和主权私有云能力。 这一框定很关键,因为 Armada 对自己想切入的楔子讲得异常直接。公司首页和 Galleon 页面强调数周内可运行、模块化扩容,以及对离场数据的完全控制。Microsoft 也用 Azure Local 语言强化同一框定:双方方案面向间歇连接、受争夺或完全离线的环境,在这些环境中,国防、公共安全、能源和关键基础设施运营方需要云一致的 AI 能力,同时不能放弃本地性和控制权。因此,排除支出应包括核心超大规模建设、通用托管容量、普通公有云消耗,以及不能解决加固部署或主权问题的纯软件工具。这些类别相邻,有时也能替代,但把它们视为直接 SAM 会高估市场。 实际含义是,Armada 应被当作一家主权边缘基础设施公司来评估,而不是小号超大规模云厂商。它的市场存在于部署速度、离线运行、加固和本地治理是采购标准而非锦上添花的地方。因此,正确买方集中在国防、公共部门、海上能源、制造、采矿和电信,而不是那些可以依赖标准云或都会区托管的普通企业 IT 资产。[CM001, CM002, CM003, CM004, CM005, CM006]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方Armada 相关性
坚固型模块化 AI 基础设施面向远程站点的集装箱化或预制计算、存储、网络、冷却、本地电力接入和快速部署服务传统绿地数据中心建设和普通都市圈托管租赁基础设施、运营、任务或数字化转型预算最接近硬件的市场边界
主权私有云和本地 AI 运行时Azure Local、本地推理、隔离或客户可控云环境,以及边缘侧治理和安全控制没有本地控制要求的通用公有云消耗安全、合规、主权云或任务 IT 预算受监管和断联买家的关键需求
边缘编排和连接控制跨远程资产的设备群监控、工作负载编排、连接管理和应用部署不控制坚固型边缘基础设施的通用 ITSM 或监控工具网络运营、平台工程或现场运营预算让设备群部署和变现成为可能
垂直解决方案部署需要本地处理的国防任务、海上钻井平台、应急响应、采矿、制造和电信 AI 网格铺设可稳定接入核心云的标准企业数据中心更新周期业务线加 IT/OT 预算界定 Armada 所选垂直行业的 SAM
排除的相邻市场N/A核心超大规模建设、通用云、通用托管,以及没有可部署基础设施的软件纯工具广义企业 IT 预算可作背景,但不是直接 SAM

纳入支出只包括购买决策里包含坚固型部署、主权或断联运行要求的部分。排除行是可能影响买方但不应直接计入 Armada 可寻址市场的相邻预算。

[CM001, CM002, CM003, CM004, CM005, CM006]

2.2 多个规模测算视角,而不是一个泛化 TAM

Armada 周边最好的公开市场证据,是一组基于视角且彼此重叠的证据。最宽的公开视角是 AI 基础设施本身:IDC 称 2025 全年 AI 基础设施支出达到 $318 billion,预计 2026 年约 $487 billion,并预期市场到 2029 年超过 $1 trillion。这证明周边资本开支周期真实存在,但显然太宽,不能直接叫 Armada 的 TAM。更接近的视角是模块化数据中心;两个可获得的 2026 年估算已经明显分化:Future Market Insights 预计 2026 年为 $29.3 billion,Research and Markets 预计 2026 年为 $47.75 billion。两者都指向快速增长,但超过 $18 billion 的差距本身就是一条尽调事实,说明「模块化数据中心」还不是一个干净、稳定的品类。 JLL 在这些市场估算和 Armada 实际用例之间提供了最有用的桥。其 2026 年展望称,2026 至 2030 年将新增约 100 GW 数据中心容量,CAGR 为 14%;推理可能在 2027 年超过训练;推理需求将要求地理分布和嵌入式边缘系统。这正是 Armada 需要的宏观条件:AI 负载离开纯集中式集群模型,转向分布式、延迟敏感和监管敏感部署。Deloitte 又补上主权视角,认为仅欧洲就可能在五年内围绕主权云、AI 数据中心、半导体及相邻基础设施投入超过 €100 billion 的公共和私人资本。 因此,正确的承销动作不是把每个大类简单相加。Armada 可触达市场,是这些更大资金池里加固、主权敏感、离线的子集。如果用已发布模块化数据中心区间的大约 13-17%,再用 IDC 2026 年 AI 基础设施预测的大约 1-1.6% 交叉校准,2026 年实用 SAM 约 $4-8 billion 可以成立。相对周边资本开支热潮,这一区间刻意保守;它比一个巨大但无差异的 TAM 更有决策价值,因为它尊重 Armada 真实部署边界。方向性三年 SOM 约 $0.2-0.6 billion 也只有在 Armada 把今天的标杆和渠道关系转成可重复多站点项目时才算合理。[CM017, CM018, CM019, CM020, CM021, CM022]

TAM / SAM / SOM 或规模测算视角表
口径2025 / 2026 锚点2029 / 2036 终点方法 / 解读置信度局限
IDC AI 基础设施支出2025 年实际 $318B;2026 年预测约 $487B到 2029 年 >$1T最宽的 AI 基础设施资本开支口径,覆盖服务器、存储和配套基础设施远宽于 Armada 的坚固型主权边缘切口
JLL 全球数据中心建设周期2026-2030 年新增约 100 GW 容量;CAGR 14%2030 年全球总容量约 200 GW用容量口径展示新基础设施建设规模,以及推理为何转向区域部署不是直接收入市场,仍宽于 Armada 的细分赛道
Future Market Insights 模块化数据中心市场2026 年 $29.3B2036 年 $106.7B模块化数据中心公开、偏硬件相邻口径的下限类别范围似乎窄于一些竞争报告
Research and Markets 模块化数据中心市场2026 年 $47.75B2030 年 $104.98B可获取摘要中 2026 年模块化数据中心公开口径的上限定义显著宽于 FMI,可能包含更多通用模块化容量
国防 AI 与自主系统预算锚点五角大楼 FY2026 申请 $13.4BN/A垂直需求锚点,说明 Armada 一个目标行业已经有大额 AI / 自主系统预算预算授权不等于可供 Armada 获取的支出
Armada 分析性 SAM / 3 年 SOM2026 年 SAM $4-8B;3 年方向性 SOM $0.2-0.6BN/A三角测算:取模块化市场区间 13-17%,并取 IDC 2026 年 AI 基础设施支出的约 1-1.6%分析区间,不是公开市场研究或已披露公司指标

这些口径刻意不相加。IDC 和 JLL 勾勒外围基础设施景气;模块化数据中心报告提供最接近的公开硬件代理;国防预算数据锚定一个目标垂直行业;Armada SAM/SOM 行则是其间保守的分析桥接。

[CM020, CM021, CM022, CM023, CM024, CM025]
FM001: 市场规模测算视角

Armada 身处庞大的 AI 基础设施资本开支周期,但它真正切入的是小得多的模块化和分布式部署子集:坚固化、主权化、断连场景。

金字塔刻意采用视角法,不做加总。每往下一层,都是上方更大资本池的更窄子集。

[CM021, CM023, CM031, CM032, CM043, CM044]
FM002: 市场估算区间

公开的 2026 年模块化数据中心估算已经明显分化,因此 Armada 的市场应按区间承销,而不是押一个单点 TAM。

前三行是财务市场视角,最后一行是最重要的部署瓶颈。它们放在一起,是因为该品类可触达价值离不开通电时间约束。

[CM026, CM043, CM044, CM045, CM046, CM047]

2.3 买方分层、预算归属与采用路径

Armada 的买方地图之所以有吸引力,正因为它高度碎片化。国防和公共安全交易通常靠韧性指挥、感知和本地分析来论证;能源和海上交易靠降低停机时间、在本地处理运营数据来论证;制造和采矿交易靠自动化、工人安全和预测性维护来论证;电信交易则靠把分布式 AI 容量货币化,同时把延迟敏感负载留在用户和网络资产附近来论证。在这四类场景里,买方很少只是一个通用 CIO。预算权往往分散在运营、OT、安全、IT 基础设施、数字化转型,有时还包括任务或产品负责人之间。 这种碎片化提高了销售复杂度,但也扩展了有效落地切入点。Microsoft 的 Azure Local 信息面向政府和受监管行业,Carahsoft 为联邦、州、地方、教育和医疗买方把公共部门路径具体化,Ericsson 加 NTT DATA 则显示,私有 5G 加边缘 AI 正在制造、采矿、能源、交通和智慧城市环境中成为可重复的工业采购模式。Mitsui 的投资逻辑从工业角度强化了同样的需求形态:当连续性、自主性和预测性维护必须在数据生成点工作时,本地 AI 才有价值。 采用路径通常不是从第一天就企业级全面铺开。买方先从一个痛点明确、紧迫度高的工作流开始——例如 Alaska 的无人机影像周转、海军或应急响应任务、海上钻井工作流,或电信区域 AI 部署——等本地计算证明价值后再标准化。因此,伙伴渠道和实施能力本身就是市场的一部分。在 Armada 的品类里,分发不只是看谁签 PO,也要看谁能足够快地降低部署风险,让客户从单个加固站点走向车队、区域或主权云标准。[CM010, CM011, CM012, CM013, CM014, CM015]

细分 / 买方地图
细分核心买方核心用户付款方 / 预算负责人工作流触发点采用触发点
国防 / 军事 / 公共安全任务 IT 负责人、C6ISR、行动指挥官或数字化转型负责人操作人员、情报团队、野外单位、安全团队项目、任务或现代化预算断联或受干扰时仍需运行 AI 和云工作负载通信受限下需要弹性的本地算力和主权控制
能源 / 海上钻井、数字化运营或资产绩效负责人钻井平台人员、钻井工程师、OT 团队资产、运营或数字油田预算本地数据流量大,回传岸上不稳定在恶劣环境中降低停机并本地处理数据
制造 / 采矿工厂运营、OT、自动化或工业 CIO 负责人OT 工程师、可靠性团队、安全团队运营改进或智慧工厂 / 矿山预算远程站点的预测性维护、自动化、质量和工人安全不用等中心化基础设施,就把 AI 带到数据源旁
州和地方 / 公共部门机构 CIO、应急管理、交通或公共安全负责人外勤人员、分析师、响应人员机构现代化或韧性预算灾害响应、遥感和对延迟敏感的一线决策把依赖云的工作流压缩成本地实时运营
电信 / 服务提供商网络平台、边缘云或产品负责人网络运营、平台工程、AI 服务团队网络投资和平台变现预算需要在既有资产群上变现低延迟分布式 AI 服务用一个控制平面协调 AI 工厂、区域枢纽和边缘站点

预算归属因垂直行业而异,但共同点是分散在运营、OT、安全、IT 和数字化转型负责人之间,而不是落在单一通用预算线。

[CM008, CM010, CM011, CM012, CM013, CM014]
FM003: 买方 / 阻碍因素图

Armada 的市场横跨多个职能:买方随垂直行业变化,但都需要本地控制、坚固化部署,并证明边缘 AI 能改善运营工作流。

[CM008, CM012, CM013, CM016, CM017, CM036]

2.4 国防、能源、公共部门、工业与电信的垂直需求证明

Armada 的公开证明集很重要,因为它显示这个品类不只是营销抽象。在海上能源领域,Aker BP 的理由很直接:关键钻井决策依赖大量井下和运营数据,但到岸上和云基础设施的连接并不总能保证,因此需要本地计算来支撑韧性、网络安全和更快的模型驱动决策。在公共部门,Alaska 工作流从 28 小时滞后改善到 4 小时甚至实时输出,说明离线环境不是贬义上的边缘个案;它们正是集中式云工作流在经济和运营上失效的场景。 公共部门分发同样重要,因为 Carahsoft 的体验中心把产品从创业公司叙事推到采购界面。联邦、州、地方、教育和医疗买方现在可以看到一个为云触达不了的地方打造的自给式 AI 计算环境,这为受监管和任务关键预算提供了具体采用桥梁。在电信侧,Armada 的 NVIDIA AI Grid 定位把机会从向远程站点发货扩展出去。软件主张是,电信或服务提供商可以把现有数据中心、AI 工厂、区域枢纽和边缘位置缝合成一张可货币化的 AI 网络。 这就是主权为何有经济意义,而不只是政治意义。在这些买方眼里,主权意味着能决定智能在哪里运行、数据如何流动、回传链路失败时怎么办。当替代方案不是更便宜的公有云 SKU,而是不可接受的运营妥协时,价值主张最强。这创造了真实市场,但也意味着 Armada 的增长绑定在艰难垂直行业里;在这些行业中,证明、合规和现场执行比抽象云弹性更重要。[CM009, CM010, CM011, CM014, CM016, CM017]

FM004: 采用漏斗或价值链地图

市场采用通常从一个紧迫的远程工作流开始;只有买方信任本地计算 ROI 和运营模式后,才会扩展。

这张采用逻辑图综合了 Alaska、Aker BP、Carahsoft,以及 Armada 以渠道带动市场推进的公开信号, 不是已发布的转化漏斗。

[CM010, CM011, CM014, CM016, CM017, CM050]

2.5 增长驱动、采用约束,以及什么会打破多头逻辑

最强的增长驱动清晰可见,而且相互强化。IDC 的支出数据表明,AI 基础设施已经进入多年资本周期。JLL 称推理正变得比训练更重要,并将需要更多区域和边缘部署。Deloitte 说明主权不是小众话术,而是活跃投资议程,在欧洲尤其如此。国防和工业来源又补上第二个驱动:确有一些环境需要本地计算,因为间歇连接、延迟或现场自主性会让集中式处理太慢或太脆弱。Armada 还有一个产品层面的部署速度楔子:在建设、许可和场地准备可能耗时数年的市场里,数周可运行的基础设施本身就有价值。 约束同样真实。JLL 称主要市场并网等待超过四年,AI 装配成本可达每 MW $25 million;Vertiv、Schneider 和 Data Center Knowledge 都把电力和高密度基础设施描述为设计限制因素。Uptime 又给出两个警示信号:市场仍面临人员短缺和成本压力,而且大量 AI 基础设施需求集中在超大规模云厂商和其他资本充足的玩家手里。换句话说,这个品类很大、很急迫,但并不顺滑。具体到 Armada,这意味着制造执行、渠道杠杆和部署复杂度,与顶线需求一样重要。 因此,模块化数据中心估算互相矛盾,并不是记账麻烦,而是在提醒:品类成熟度仍不均衡。如果电力、许可和集中度继续拖慢市场转化,或者电信和主权 AI 项目持续停留在大量试点阶段,可触达市场的扩张速度就可能慢于标题级资本开支数字。正确的尽调重点不是抽象地判断市场是否很大——它确实很大——而是判断 Armada 能否在这个品类的基础设施瓶颈和买方复杂度开始偏向更大在位者之前,拿到可重复预算线。[CM020, CM021, CM022, CM023, CM024, CM025]

增长驱动因素与约束表
驱动 / 约束方向时间对 Armada 的影响尽调问题
推理从核心训练集群转向区域和边缘部署正向2026-2030扩大对地理分布式、本地 AI 基础设施的需求验证 Armada 管线中有多少由推理驱动,而不是通用现代化
主权、本地控制和受监管行业要求正向当前支撑面向政府和受监管买家的 Azure Local 与主权私有云定位询问哪些在谈交易把主权或数据驻留要求列为购买核心原因
本地处理带来的远程站点 ROI正向当前海上、公共安全和工业用例可以靠停机或延迟降低纳入预算索取 Alaska 之外更多客户站点的量化前后对比指标
数周投运,相对多年建设正向当前传统数据中心建设周期太慢时,形成价值兑现速度切口按产品层级测量实际部署时间、场地准备负担和服务附着
渠道和合作伙伴杠杆正向当前Microsoft、Carahsoft、Johnson Controls 和垂直合作伙伴可加速市场进入和规模化厘清哪些伙伴是需求生成引擎,哪些只是履约或背书层
电力可用性和通电周期瓶颈负向当前至结构性如果本地或电网电力不可用或太贵,即便模块化部署也会变慢评估 Armada 多常能靠本地发电或分阶段部署绕过电网延迟
资本密集度和制造执行负向当前至结构性Armada 不是纯软件,工厂产能和营运资本纪律很关键审查工厂爬坡假设、供应商集中度,以及按硬件层级拆分的毛利率
集成和现场部署复杂度负向当前坚固型站点要在恶劣条件下接好连接、算力、安全和应用量化实施周期、伙伴依赖和部署后支持负担
能力集中在超大规模云厂商和既有大厂手中负向当前可能压缩可触达市场,或让买方更偏好资产负债表更大的供应商对标与既有厂商、捆绑替代方案竞争的胜率,以及客户对小供应商风险的顾虑
市场定义彼此矛盾,部分细分需求仍以试点为主负向当前可能制造嘈杂 TAM 叙事并拖慢变现,尤其是电信和主权 AI 试点按垂直行业索要需求证据:已签多站点铺开,还是展示部署或试点

上行驱动是真实的,但最影响估值的风险是电力、资本开支,以及把好看的参考案例转成工业规模标准化设备群铺开的难度。

[CM003, CM010, CM011, CM020, CM021, CM022]

2.6 图表

Chapter 03

03竞争对手

3.1 直接同业比通用 AI 基础设施筛选看起来更少

公开证据不支持把每个 GPU 云或模块化数据中心供应商都当成 Armada 的同类竞争对手。Armada 自己的产品页描述的是一种组合:加固型、集装箱化的 Galleon 硬件,AEP/Bridge 编排,多租户 GPU 货币化,以及一个伙伴 Marketplace,目标是把远程基础设施变成托管式主权 AI 运营环境。按这个定义,公开证据支撑的最接近创业公司替代品是 Crusoe Spark 和 Nscale,因为两者都使用模块化 AI 工厂语言,而不只是泛泛谈云容量。Crusoe 明确推低延迟、主权和本地用例的交钥匙预制模块化 AI 工厂,Nscale 则把主权模块化数据中心同完整软件和车队运营栈结合起来。 Lambda 是重要替代品,但证据把它放在安全私有集群和 GPU 云替代方案的位置,而不是加固型可部署现场数据中心同业。它最强的公开差异点是透明 GPU 定价、单租户和托管私有云 Kubernetes。这很重要,因为它揭示了一种真实买方替代路径:许多客户可以通过租用安全 GPU 容量或搭建单租户集群来完成任务,而不必购买加固、集装箱化的部署系统。因此,本章把直接竞争视为一个较薄的集合,把替代 / 私有集群竞争单列为一类,也把物理模块化供应商单列为另一类;后者从下方攻击 Armada 的硬件护城河。[CP001, CP002, CP003, CP004, CP005, CP008]

竞争对手画像表
替代方案类别产品 / 范围最适合客户 / 站点公开规模或包装信号相对 Armada 的局限
Armada直接基准坚固型模块化 Galleon 硬件,加 AEP/Bridge 编排和合作伙伴市场国防、公共部门、海上能源、电信和远程工业站点从 Beacon 到 Leviathan;20+ 伙伴;Carahsoft 和 Microsoft 渠道需要软件采用和伙伴杠杆,才能守住硬件之外的防线
Crusoe Spark直接模块化同业交钥匙预制模块化 AI 工厂,加 Crusoe Cloud 和托管推理低延迟、主权、本地和成组训练部署声称最快 3 个月部署,Spark 模块覆盖数百 kW 到数百 MW在坚固、受干扰或公共部门现场部署上的公开证据少于 Armada
Nscale直接主权同业全栈 AI 云平台,加模块化主权数据中心和设备群运营需要主权枢纽容量的企业和政府买家公开园区从 30MW 到 240MW+,并声称具备全生命周期管理和模块化预制公开足迹更像枢纽级园区基础设施,而不是手提箱或 20 英尺现场部署
Lambda私有集群替代单租户 GPU 云、私有云 Kubernetes 和 1-Click 集群希望快速获得安全 GPU 容量的企业和研究团队公开小时计价,集群规模 16 到 2,000+ GPU不是坚固型模块化现场数据中心供应商
AWS Outposts既有混合云由 AWS 安装的机柜和服务器,在本地运行 AWS 服务需要本地延迟或数据本地化的现有 AWS 账户42U 机柜形态、AWS 安装、3 年定价模型较少为恶劣断联环境和报价驱动采购专门打造
Azure Local既有混合 / 本地部署由 Azure Arc 启用的分布式基础设施,跑在合作伙伴或已验证硬件上以 Microsoft 为中心的企业和主权买家按核心计价、合作伙伴硬件目录、断联本地控制平面选项伙伴硬件模型不如 Galleon 那样专为坚固型现场部署打造
Google Distributed Cloud既有混合 / 本地部署面向边缘和数据中心站点的全托管 Google 硬件和软件希望使用 Google 技术栈的受监管或隔离运营方Gemini 本地部署、隔离选项、从一个到数千个地点的叙事对便携式坚固容器化硬件的强调较少
HPE Private Cloud AI交钥匙企业私有 AI预配置 HPE/NVIDIA 私有云,通过 GreenLake 交付在自建与交钥匙之间比较的企业 AI 团队从开发者到大型配置,GSI 生态强更瞄准数据中心私有云,而不是前沿部署边缘站点
Dell AI Factory交钥匙企业 AI 基础设施端到端数据平台、模块化架构、服务和 NVIDIA 技术栈大型企业账户把试点推向生产4,000+ 客户部署,OEM 规模强在企业账户中竞争最强,而不是远程断联任务

这些行混合了直接同业、既有混合云技术栈和替代方案,因为 Armada 买家可以用几种方式解决同一任务。证据不足的单元格写成局限或未知,而不是猜测。

[CP001, CP004, CP006, CP010, CP013, CP016]
功能 / 能力矩阵
购买标准ArmadaCrusoe SparkNscaleAWS OutpostsAzure LocalGoogle Distributed CloudLambda
面向恶劣远程站点的坚固型模块化部署混合混合混合
全断联或隔离运行混合选择性选择性选择性
跨既有和新站点的控制平面混合混合混合混合混合
多租户 GPU 变现 / GPUaaS混合混合
公开标价透明度UnknownUnknownUnknown混合混合混合
既有账户企业渠道深度混合混合混合混合
公共部门 / 主权渠道证据选择性选择性选择性
运行在客户自有或既有基础设施上混合

矩阵分值是有证据支撑的定性评分,不是基准测试。“强”表示留存来源明确描述该能力;“混合”表示支持不完整或依赖伙伴;“选择性”表示只有窄用例证据;“未知”用于公开证据不足的地方。

[CP004, CP008, CP009, CP010, CP013, CP015]
FP001: 竞争定位图

Armada 在加固本地部署上位置很高,在集成控制平面和渠道力量上也相对靠前; 但几家既有厂商的分销更强,物理设备厂商则挤压纯硬件层。

分值是基于保留证据的序数综合,并非基准测试结果。x 轴反映各方案离加固或现场可部署本地算力有多近, y 轴反映厂商是否还掌握广泛的软件平面和市场路径。

[CP002, CP006, CP010, CP013, CP016, CP019]

3.2 在位混合云和私有云栈掌握最广渠道与最安全采购路径

AWS Outposts、Azure Local、Google Distributed Cloud、HPE Private Cloud AI 和 Dell AI Factory 是主流企业和主权买方最重要的实际替代方案,因为它们把熟悉的控制平面和既有客户关系打包在一起。Outposts 在三年商业结构下,把部分 AWS 服务延伸到客户设施和托管站点。Azure Local 把 Azure Arc 延伸到合作伙伴或验证硬件上,并包含一条离线控制平面路径;Microsoft 与 Armada 的合作本身也说明,该公司可以一边与 Armada 合作,一边争夺主权私有云控制权。Google Distributed Cloud 给出更清晰的主权和隔离反击:它是一套由 Google 全托管的硬件加软件栈,用于数据中心和边缘位置,Gemini 可本地运行,并提供隔离选项。 HPE 和 Dell 重要,是因为它们缩短了超大规模云软件与企业 OEM 采购之间的距离。HPE 把购买决策明确框定为自建、参考架构服务和交钥匙之间的选择;Dell 则推动一套模块化架构,拥有超过 4,000 个客户部署和广泛专业服务支持。这一组合是 Armada 最大的分发挑战。买方如果想要本地 AI,但不需要加固型前置部署盒子,就可以留在既有 Microsoft、AWS、Google、HPE 或 Dell 关系内,同时拿到 Armada 宣传的大部分主权、延迟和治理收益。[CP006, CP010, CP011, CP012, CP013, CP014]

定价 / 包装对比
替代方案公开定价情况合同 / 打包方式包含能力影响
Armada未找到公开标价证据硬件叠加 AEP/Bridge 软件与合作伙伴生态;定制化企业销售坚固化模块化站点、编排、连接管理、Marketplace、主权部署叙事定价不透明,ACV、服务附加率、硬件与软件收入占比都需要尽调
AWS Outposts有官方定价结构,但配置选择需定制3 年期;可全额预付、部分预付或不预付;必须购买 Enterprise Support交付、安装、服务、EC2/EBS/S3 基础容量、本地 AWS 服务商务结构对 AWS 买家熟悉,但可能把客户锁进 AWS 支持与续约机制
Azure Local按物理核心收取月度服务费已验证合作伙伴硬件,或在符合条件的硬件上自安装;60 天试用Azure Arc 管理,AKS on Azure Local 不额外收费,可选 Windows Server 订阅Microsoft 存量客户切入摩擦较低;仍需要 Azure 订阅和合作伙伴硬件路径
Google Distributed Cloud 联网版公开起步价基于 Google 认证硬件的托管基础设施;每站点至少 96-vCPU面向容器和 VM 的托管 Kubernetes 基础设施与存储适合想要 Google 技术栈、但不需要定制坚固化硬件的受监管站点
Google Distributed Cloud 气隙隔离版报价制气隙隔离托管部署面向主权和断网运行的气隙隔离软硬件栈主权叙事强,但公开定价仍不透明
HPE Private Cloud AI按报价 / 范围核定GreenLake 主导的交钥匙私有云,配置按需匹配预配置且验证过的技术栈、统一数据层、NVIDIA 软件、可观测性、合作伙伴生态靠降低集成负担竞争,目标是原本会自建的企业买家
Dell AI Factory有用量制和即用即付选项,但未找到标价模块化架构叠加服务,覆盖桌面到数据中心服务器、网络、软件、服务、液冷、自动化平台即便没有公开标价,强企业销售动作也会挤压小厂商
Lambda透明的公开小时定价按需实例,加上 1-Click 集群和预留容量销售单租户安全 GPU 云、私有云集群、托管 Kubernetes透明 GPU 价格给不透明的定制基础设施方案设下可见锚点
AWS 公有云 GPU 区域公开实例族页面;按用量计费的云定价集中式云区域消费无需本地站点部署的加速计算当本地主权、延迟或断网运行并非决定因素时,它就是默认替代方案

公开定价可得性差异很大。Armada、HPE、Dell 以及大多数模块化物理设施厂商仍需尽调实际商务条款;Azure Local、Google Distributed Cloud 联网版和 Lambda 至少给出部分公开价格锚点。

[CP002, CP012, CP014, CP018, CP019, CP022]
GTM / 分销能力表
参与方控制什么谁受益最大竞争影响锁定或杠杆效应
MicrosoftAzure Local 软件平面、主权私有云品牌、既有企业和政府账户Azure Local、Azure 技术栈合作伙伴,以及任何借 Microsoft 主权叙事切入的厂商可以与 Armada 合作,同时仍掌握控制平面和客户关系提高买家信任度,但也可能限制 Armada 独立拥有账户
NVIDIA参考架构、GPU 路线图、AI Enterprise 软件和生态验证Dell、HPE、Armada、Carahsoft 及其他系统构建商GPU 生态替 Armada 背书,但多家厂商都继承同一套参考栈,差异化也被压低权力转向最会把 NVIDIA 技术栈打包进客户账户的一方
Carahsoft公共部门合同工具、经销商生态、演示 / 采购界面Armada、NVIDIA 及其他瞄准政府和受监管买家的厂商公共部门进入市场可以由合作伙伴撮合,而不必直销不只靠技术,也靠合同和经销商关系抬高切换成本
DellOEM 制造规模、服务、装机基础和账户覆盖Dell AI Factory 以及对齐 NVIDIA 的企业 AI 交易Dell 可以把 AI 基础设施变成既有企业账户里的标准 OEM 增购大型 OEM 触达面会压缩小厂商赢下通用企业交易的窗口
HPE 和全球系统集成商GreenLake 消费模式和全球系统集成商网络HPE Private Cloud AI 及共同开发的 NVIDIA 方案企业可以在熟悉的集成商支持下购买交钥匙私有 AI,而不用重新拼一套新供应商栈集成商主导交付,把流程和运营锁定嵌进去
Armada Marketplace / 合作伙伴生态预集成软件、连接合作伙伴和垂直应用界面在集成时间关键的远程或主权边缘部署中,Armada 受益最大Armada 最好的渠道防线,是在坚固化环境中成为从箱体到可用技术栈的最快路径如果 Marketplace 使用量一直偏薄,渠道杠杆会回到既有厂商和分销商手里

主权和公共部门 AI 基础设施里,分销能力往往比功能是否齐平更重要。本表关注谁能最快触达、验证并签下买家。

[CP005, CP006, CP007, CP021, CP024, CP025]
FP002: 护城河 / 就绪度图

需要加固部署和软件控制同时成立时,Armada 最强;既有厂商渠道更深, 物理设备厂商大多仍是箱体和制冷生意。

[CP004, CP009, CP016, CP019, CP026, CP029]

3.3 硬件箱体、供电与冷却层越来越容易商品化

Armada 的物理产品目前靠便携性、加固和集成形成差异化,但更广泛的基础设施市场正快速走向预集成、AI 就绪的 pod、skid 和模块化系统。Vertiv 围绕 SmartMod、MegaMod 和 OneCore 宣传工厂集成、多 MW 规模、快速运输,以及 time-to-token、密度和 TCO 上的大幅承诺。Eaton 和 Flexnode 推 3.5MW 到 35MW 数据大厅的交钥匙预制 AI 工厂。Schneider 的 AI pod 架构已经围绕 1MW 以上高密度集群、液冷和预组装交付设计;Rittal 则把 OCP 启发的机架和水冷对齐 NVIDIA 新兴 DC 电源要求。换句话说,物理层并非静止,也并非空白。 这不会消除 Armada 的边缘优势,因为这些供应商多数在优化企业、colo、超大规模和 AI 工厂部署,而不是手提箱、20 英尺箱体或受争夺任务现场。但这意味着 Armada 不能假设箱体、供电链或冷却栈会长期独特。如果客户主要把 Galleon 看成快速模块化 AI 容量盒子,资本更充足的供应商就能用相邻产品和既有现场服务能力压迫这个品类。这会把护城河耐久性推向更上层:编排、伙伴分发、主权运营模式,以及 Galleon 部署上线后仍能保有价值的证据。[CP001, CP002, CP003, CP035, CP036, CP037]

物理基础设施既有厂商表
厂商物理产品部署速度信号密度 / 功率信号对 Armada 施压的位置缺失的软件 / 控制平面层
Armada Galleon从 Beacon 到 Leviathan 的集装箱式坚固化模块Galleon 系列宣称数天到数周可部署从 3 个机架、5 个机架到兆瓦级液冷为可部署的坚固化 AI 就绪硬件设定基线AEP/Bridge 已包含,并非缺失
VertivSmartMod、MegaMod 和 OneCore 集成模块化方案预制方案节省 40%+ 时间;OneCore time-to-token 最多快 50%单机架最高 600 kW,并支持多 MW 机柜行直接冲击企业和主权运营方看重的快速部署与密度叙事未找到类似 Armada 的分布式边缘控制平面或 Marketplace 证据
Eaton / Flexnode交钥匙预制 AI 工厂数据大厅通过模块化 NX 计算模块快速部署3.5MW 到 35MW 数据大厅,配 800 VDC 电力基础设施买家不需要野外坚固化箱体时,它靠集成电力的预制交付竞争未找到多站点 AI 工作负载编排证据
Rittal对齐 OCP/NVIDIA 的机架,以及紧凑型冷却 / 电力基础设施以标准化换速度的叙事>1MW 水冷能力,紧凑占地;兼容 800 VDC靠高密度 AI 硬件模块竞争未找到主权工作负载控制或应用层证据
Schneider ElectricEcoStruxure Modular Data Center 和 AI pod 架构快速部署,预组装 pod 支持快速铺开1MW+ pod,支持液冷和高密度机架靠 AI pod 架构、合作伙伴生态和现场服务覆盖竞争未找到类似 Armada 的 GPU 变现或统一边缘站点控制平面证据

本表只拆出物理层竞争。它有意把箱体、电力、冷却竞争同软件 / 控制平面竞争分开,因为如果把这些层合成一个无差异产品分析,Armada 的护城河会变弱。

[CP001, CP002, CP003, CP035, CP036, CP037]

3.4 分发能力和切换成本塑造真实输赢边界

本章最强的竞争证据,不是跑分功能,而是谁掌握通往买方的路径。Microsoft、HPE、Dell、NVIDIA 和 Carahsoft 都显示,渠道杠杆在主权和企业 AI 基础设施里有多重要。Armada 自己的 Carahsoft 和 Microsoft 公告也强化同一点:公共部门和受监管买方通常希望在自己与原始基础设施供应商之间,有分销商、经销商网络、验证栈或在位云关系。Carahsoft 的 NVIDIA 页面把这一教训扩展到 Armada 之外;公共部门 AI 市场靠集成商生态和合同载体流动,而不只是创始人直接销售。 现状替代同样重要。买方往往可以留在集中式公有云 GPU 基础设施上,借助 Outposts 或 Azure Local 延伸既有云栈,或者把计算留在安全私有集群里,而不是采用一种新的可部署边缘形态。锁定来源说明了这件事为何重要:计算本身可以迁移,但数据模型、身份、应用集成、IaC 和组织工作流会让退出云变得昂贵。这意味着 Armada 不只是比性能;它还要对抗买方不愿增加新运营模式的惯性。AEP 和 Bridge 越能跨既有设施和客户自有基础设施运行,Armada 越能降低切换问题;如果越依赖专有硬件采购,买方池就越窄。[CP004, CP005, CP006, CP007, CP008, CP009]

护城河耐久度 / 竞争风险登记表
护城河主张支撑证据威胁 / 反向证据严重性尽调问题
坚固化模块化封装Armada 可在数天或数周内部署从便携式到兆瓦级的硬件,并在恶劣环境中气隙隔离运行Vertiv、Eaton、Schneider 和 Rittal 正在快速工业化预制 AI 就绪基础设施要求提供详细赢单 / 输单数据,证明 Armada 胜出主要靠坚固化,而不是软件
AEP / Bridge 控制平面Armada 宣称可统一控制、编排 AI Grid、变现 GPU,并横跨既有与新建站点运行Microsoft、Google、AWS、Dell、HPE 和 Nscale 都控制相邻的软件平面或生命周期管理器查看 AEP/Bridge 脱离硬件出货后的实时产品采用指标
主权与断网运行Armada 和 Microsoft 明确把 Galleon 上的 Azure Local 定位给断网和受监管环境Azure Local、GDC 气隙隔离版和 HPE 私有 AI 都在不依赖 Armada 硬件的情况下主打主权列明 Armada 部署相对合作伙伴技术栈独有的合规或认证材料
公共部门渠道准入Carahsoft 体验中心和合同工具创造采购入口渠道权力可能意味着 Carahsoft、Microsoft 或 NVIDIA 掌握账户经济性量化销售管线、订单和续约中,合作伙伴来源与直销各占多少
不透明企业定价保护利润率需求特殊时,定制打包可支撑更高价值销售Lambda 公布明确 GPU 价格,Azure/GDC 也暴露部分公开锚点,溢价更难证明将 Armada 的实际性价比同 Lambda 私有集群和既有私有 AI 技术栈对标
部署速度优势Armada 核心卖点是在传统数据中心跑不通的环境里,用数周而不是数年交付无论是否模块化,电力约束和供应瓶颈都可能拖慢整个品类梳理哪些交易卡在本地电力、互联或场地准备,而不是软硬件就绪度

本登记表有意把支撑证据和反向证据混在一起。本章判断应取决于哪些护城河主张能扛住既有厂商响应和硬件商品化的冲击。

[CP002, CP006, CP009, CP016, CP019, CP024]

3.5 Armada 的护城河真实存在,但主要在软件、渠道和部署速度耐久性

证据支持对 Armada 差异化作更细的判断。公司确实有真实楔子:加固型模块化部署、隔离和离线运行、Azure Local 对齐、借 Bridge 和 AEP 形成的 GPU 货币化与编排层,以及借 Carahsoft 获得的具体公共部门渠道。这比纯 GPU 云更有差异化,也比标准企业私有 AI 栈更能现场部署。它也解释了为什么 Armada 可以看起来像 Microsoft 和 NVIDIA 的伙伴,同时又与其他购买本地 AI 容量的方式竞争。 同一组证据也给护城河乐观设下硬边界。Lambda 透明定价、HPE 和 Dell 交钥匙栈、Google 和 AWS 混合产品,以及快速进步的模块化基础设施供应商,都说明盒子层面的独特性会被压缩。耐久问题在于,Armada 的控制平面、伙伴触达和远程站点执行质量,能否成为主权和离线 AI 部署的默认运营层。如果能,Armada 可以在集中式云和商品化模块化硬件之间占据可防守品类;如果不能,在位者可以吸收软件层,物理供应商可以压缩硬件层,Armada 则有风险变成一个形象鲜明但更窄的系统集成商。[CP004, CP006, CP007, CP009, CP023, CP030]

FP003: 护城河 / 就绪度 KPI

这些紧凑信号勾勒出竞争压力最具体的区域,也说明 Armada 的投资逻辑如今多大程度上靠渠道和编排, 而不只是加固硬件。

数值混合了不同单位,只作为压力指标呈现,不是估值模型。之所以选取这些指标, 是因为它们公开、可比较且与决策相关。

[CP005, CP012, CP018, CP023, CP030, CP034]

3.6 图表

Chapter 04

04财务

4.1 收入模式明显是硬件加软件和渠道赋能,但实际组合仍未公开

Armada 的公开材料支持多线收入模式,而不是纯产品或纯 SaaS 框架。Galleon 是物理系统销售:从较小现场单元到兆瓦级 Leviathan 的加固型模块化数据中心。Bridge 是最清晰的软件货币化界面。Armada 现在称 Bridge 是用于管理、扩展和货币化 GPU 集群的软件;更新的 Bridge 材料进一步说明,定价基于活跃 GPU 使用量,并按 GPU/year 或 GPU/hour 结构计费。同一组来源也明确,Bridge 可以运行在客户自有基础设施上,不只跑在 Armada 硬件上;这很重要,因为它意味着即使客户不购买新的 Galleon,控制平面也可能成为经常性软件收入线。Atlas 和 Marketplace 扩展了这幅图景:Atlas 是带有 pooled data-plan 和 Azure 集成语言的管理平台,Bridge 和 Marketplace 材料则描述了伙伴软件和模型服务,能把基础设施转成可创收的 AI 能力。 公开记录没有披露的是这些收入线之间的实际组合。没有发布 Galleon 硬件、Bridge、Atlas、Marketplace、部署工作或支持之间的拆分。这一缺口对承销至关重要。如果硬件占主导,Armada 的估值应更接近资本密集型基础设施供应商;如果软件和服务挂载率随时间占优,毛利率和倍数画像可能显著上移。因此,本章把多重收入界面的存在视为已确认,但每个界面的实际经济权重仍视为私有信息。[CI001, CI002, CI003, CI004, CI005, CI006]

收入来源表
收入来源机制单位当前数值 / 状态收入质量关键尽调问题
Galleon 硬件和交钥匙部署模块化数据中心销售或合同部署,外加调试交付按单元 / 按站点产品系列已公开;实际 ASP 和付款时点未公开可能波动大,且由硬件驱动索取 SKU ASP、验收条款、客户预付款,以及租赁 / 销售占比
Bridge GPU 编排面向 GPU-as-a-Service 的用量制软件和控制平面GPU/年或 GPU/小时明确计费单位已公开;实际费率表未公开若单独销售,可能形成经常性软件收入线索取费率表、折扣阶梯、附加率和独立收入
Atlas 运营平台监控、管理、共享数据套餐和资产软件账户 / 资产 / 数据套餐口径未公开量化产品已存在;定价和收入贡献未披露可能是经常性软件收入,但经济性尚无公开证据索取 Atlas 合同结构、ARPA、留存和毛利率
Marketplace 合作伙伴软件和硬件发现、购买、部署,以及可能的交易或推荐收益抽佣率 / GMV 单位未披露购买流程已公开;变现机制未公开若按抽佣收费,可能毛利较高,但目前未被证明索取 GMV、抽佣率、附加率和合作伙伴分成条款
部署、集成和支持调试交付、集成、现场支持,以及可能的托管运营按部署 / 按支持期交钥匙叙事暗含服务层,但缺少公开定价若有续约,可平滑硬件周期性索取服务收入占比、续约率和现场服务毛利率

公开资料确认了机制和渠道,但没有披露各收入来源的实际组合或确认收入。

[CI001, CI002, CI003, CI004, CI005, CI006]
定价 / 变现表
产品 / 路径公开单位或价格合同 / 计费结构标价 vs. 实际未知项 / 注意事项来源
Bridge 用量定价仅按活跃 GPU 用量,以 GPU/年或 GPU/小时计费用量制软件计费计费单位公开;实际费率表未披露未公开最低消费、折扣阶梯或预留容量条款Armada Bridge 博客
Bridge GPUaaS / ModelaaS 变现未公开美元价格运营方可以推出 GPU-as-a-Service 或 Model-as-a-Service机制公开;实际价格未知未发布租户定价或抽佣率Bridge 产品页 + 发布稿
Galleon 硬件无公开标价可能是报价制硬件或部署合同未披露标价与实际价差ASP、付款节点和安装经济性未披露Galleon 页面 + Carahsoft 体验中心博客
Atlas未公开美元价格企业软件,带共享数据套餐表述和 Azure 集成未披露标价与实际价差订阅口径和毛利率未披露Atlas 页面
Azure Marketplace / MACC通过既有 Azure 承诺消费渠道采购机制采购路径公开;商业分成不公开未公开 Microsoft 收费结构或利润率影响Business Wire 2024 轮融资
Carahsoft 公共部门渠道基于合同工具的采购,而非公开标价SEWP / ITES / NASPO / TIPS / OMNIA / Quilt 下单路径路径公开;任务订单价格不公开经销商利润、阶梯折扣和机构专属定价未披露Armada Carahsoft 公告

只有 Bridge 有明确的公开计费单位;其他 Armada 标价仍是报价制或未披露。

[CI003, CI005, CI008, CI009, CI024, CI025]
FI001: 收入模型桥接——从客户需求到确认收入

公开需求信号和产品层可能如何转化为确认收入和毛利。

该结构有来源支撑,但硬件与经常性软件之间的组合未公开披露。

[CI001, CI002, CI003, CI005, CI010, CI011]

4.2 唯一明确的定价单位在 Bridge 上,披露的增长指标是订单额而非收入

对一家现在同时销售物理基础设施和软件的公司来说,Armada 的定价披露异常狭窄。最好的公开定价信号来自 Bridge 新的用量语言:Armada 称定价基于活跃 GPU 使用量,并按 GPU/year 或 GPU/hour 结构计费。这有价值,因为它证明公司存在明确的计量软件结构。但价格点、折扣阶梯、最低承诺,以及预留用量与突发用量之间的组合仍未披露。公开来源也没有给出 Galleon、Leviathan、部署服务或 Atlas 的标价。Azure Marketplace 和 Carahsoft 降低采购摩擦,但不揭示实际成交价或毛利率。 收入质量是更严肃的问题。Armada 最强的公开牵引指标是订单额增长:FY25 至 FY26 增长 540%,FY27 Q1 同比增长 2,000%。这些数字由公司披露,并被 CNBC 独立复述,但它们不是收入。硬件加软件公司常常在订单、开票现金和已确认收入之间出现时间差。Pure Storage 和 Equinix 提供了公开类比:硬件和安装现金可能先于会计确认或以不同方式确认,而经常性软件和服务义务可能按期确认。没有 Armada 的合同组合、确认政策、积压订单排期或递延收入数据,订单额数字只能证明需求,不能证明盈利质量。[CI003, CI007, CI010, CI011, CI012, CI021]

FI002: 单位经济性桥接——Armada 要产出利润率,哪些环节必须跑通

公开可见的定价结构和隐藏成本驱动因素,共同决定 Armada 的实际经济性。

由于 Armada 未披露实际 ASP、单位 COGS 或现金转化指标,这张图是定性分析。

[CI003, CI021, CI026, CI027, CI030, CI031]

4.3 公开可比公司给出很宽的利润率包络,这正说明 Armada 未披露组合很关键

Armada 不发布毛利率、贡献利润率、硬件物料清单或软件挂载率。因此,公开承销只能先从外部基准开始,而不是公司披露的单位经济学。在这组来源中,Vertiv 是最干净的硬件偏重基础设施类比:其 FY2024 结果显示约 $8011.8 billion 销售额,服务约占收入 20.2%,隐含综合毛利率约 36.6%。当物理层占主导时,设备加服务基础设施模式可能就是这种样子。Pure Storage 展示了截然不同的组合:FY2025 结果隐含产品毛利率约 66.1%、订阅服务毛利率 74.1%、综合毛利率 69.8%。如今更偏软件的 Nutanix,在 FY2024 Q4 报告 GAAP 毛利率 85.2%,全年 84.9%,并同时给出 ARR 和 ACV 定义,且明确与 GAAP 时点脱钩。 这些类比重要,因为 Armada 处在它们之间。Galleon 销售叠加部署、质保和现场支持,不应被硬塞进纯 SaaS 单位经济学模板。但 Bridge、Atlas 和 Marketplace 也不能被忽略,因为它们是通往更高经常性毛利率的最明显路径。当前公开答案因此是一个包络,而不是一个点估计:如果 Armada 仍由硬件主导,利润率更可能像基础设施供应商而不是软件平台;如果 Bridge 和其他经常性层增长快于硬件,利润率扩张可能显著。公开材料尚未显示哪条路径正在发生。[CI026, CI027, CI028, CI029, CI030, CI031]

单位经济模型表
指标公开数值 / 代理指标置信度为什么重要关键尽调问题
Bridge 定价单位活跃 GPU 用量按 GPU/年或 GPU/小时计费这是 Armada 软件层最清楚的公开变现单位获取当前 Bridge 费率表和企业折扣阶梯
Galleon / Leviathan 实际 ASP硬件 ASP 决定订单质量、营运资本和硬件毛利索取已签报价、硬件 BOM 假设和验收标准
Atlas 变现没有定价,Atlas 就无法建模为经常性软件贡献索取 Atlas 合同、定价口径和客户数
硬件主导基础设施利润率基准~36.6% Vertiv FY2024 隐含综合毛利率可作为产品和服务占比较高的基础设施模型底线与 Armada 硬件 BOM、部署人工和质保成本核对
硬件 + 订阅混合基准Pure FY2025 隐含 66.1% 产品毛利率 / 74.1% 订阅服务毛利率 / 69.8% 综合毛利率说明经常性软件 / 服务如何抬升硬件平台经济性索取 Galleon 与软件 / 服务分别的毛利率
软件控制平面基准Nutanix FY2024 ARR $1.91B;Q4 FY2024 GAAP 毛利率 85.2%展示软件式估值层所需的经济性索取 Bridge 或 Atlas 的独立经常性收入和利润率
订单到收入转换没有合同组合,公开订单增长不能直接转成 GAAP 收入索取收入确认政策、在手订单转化表和递延收入明细
营运资本负担工厂、库存、部署和应收账款需求有所暗示,但未披露它比醒目的融资金额更能驱动烧钱和外部融资需求索取库存、WIP、应收、应付和客户预付款明细
客户集中度少数主权或工业合同可能主导订单和现金转换索取头部客户在订单、收入和在手订单中的占比

Armada 的实际利润率未公开;外部上市公司数据只是基准,不是 Armada 结果。

[CI003, CI021, CI022, CI026, CI027, CI029]
FI003: 财务估算区间——与 Armada 相关的公开毛利率基准区间

公开基准区间显示,Armada 的经济性高度取决于硬件和经常性软件最终各占多少。

这些是上市公司的基准区间,不是 Armada 的实际利润率;Armada 尚未披露自己的毛利率。

[CI027, CI030, CI031, CI038]

4.4 工厂规模、模块化部署和兆瓦级产品让资本强度在客户自有经济性披露之前就已经成立

资本充足性问题与公司概览里的历史融资时间线有实质差异。财务承销需要看的公开事实是,Armada 披露总融资 $465 million,其中包括 2026 年 5 月 $230 million Series B、2025 年 7 月 $131 million 战略轮,以及 2024 年 7 月 $40 million 轮。最新一轮与 Johnson Controls 制造协议和 Arizona Galleon Forge One 计划同步发生。Johnson Controls 另行确认,面积最高 400,000 平方英尺、约 500 个岗位,并从 Leviathan 开始连续生产。这组事实意味着资本强度并非假设:Armada 正围绕兆瓦级液冷产品爬坡一个真实制造足迹。 仍不清楚的是,已披露资金是否足以覆盖最关键用途。公开来源没有说明 Series B 中有多少预留给工厂工装、预付款、GPU 库存、部署营运资本或运营消耗。公开可比公司显示,数字基础设施可以很快变成资产负债表重型业务。Equinix 作为自有基础设施模式,指引 2025 年总 capex 为 $3.222-$3.472 billion,并解释即使经常性安装经济性也可能与收入确认不同步。Moody's 和 Data Center Knowledge 进一步警告,如果需求或技术对已安装基础不利,交钥匙 AI 数据中心资产会面临过度建设、过时和资本开支更新风险。Armada 近期资本可能仍然充足,但公开记录不够细,无法有信心地承销这一结论。[CI013, CI014, CI015, CI016, CI017, CI018]

资本充足性表
类别金额 / 状态日期 / 来源为什么重要关键未知项
已披露累计融资$465M 已披露May 2026;CNBC Disruptor 50这是公开可见资本基数的上限,可用于尽调测算剩余无限制现金未公开
B 轮$230M,估值 $2BMay 19 2026;Armada / CNBC / Wilson Sonsini 材料最近一次资产负债表补强,也明确提供增长资本募资用途如何分配到工厂、GPU、运营支出和营运资金仍未知
战略轮$131MJuly 24 2025;Armada / DCD为 Leviathan 发布和能源侧战略定位提供资金资金还剩多少、已花多少仍未知
M12 领投轮$40MJuly 11 2024;Business Wire早期资本,同时带来 Azure Marketplace 分销支持剩余募资款和稀释条款未知
工厂承诺Galleon Forge One 最高 400,000 sq ft;500 个岗位;JCI 投资和框架协议May 2026;Johnson Controls / Armada确认公司在真实扩大制造规模,也可能带来营运资金需求Armada 与 JCI 的出资责任未披露
账上现金判断近期偿付能力、以及能否撑过制造爬坡时必须看未公开资产负债表数字
月度烧钱和现金跑道没有烧钱速度和流动性,就无法判断资本能撑多久未公开烧钱速度、现金跑道或月度现金桥
债务 / 设备融资 / 项目融资未披露公开融资额度截至 2026-05-24资本密集型硬件公司通常会用结构化融资补充股权融资需要确认债务、供应商融资或客户预付款支持

本表只使用本地融资事实来判断资本充足性;完整历史时间线仍放在公司概况。

[CI013, CI014, CI015, CI016, CI017, CI018]
FI004: 资本强度 / 现金流图——公开融资可能被哪些环节吃掉

已披露融资可能与模块化 AI 基础设施业务中的真实制造和营运资本压力同时存在。

由于 Armada 未发布工厂预算、库存水平或现金跑道,该流程是定性分析。

[CI017, CI018, CI019, CI037, CI038, CI039]

4.5 关键承销问题不是需求是否存在,而是隐藏的组合和现金画像能否支撑估值

财务章节可以支持需求叙事,但不能闭合承销回路。公开证据支持 Armada 是一家真实业务,具备硬件、软件、渠道触达、主权 AI 定位和快速增长的订单额。公开证据也支持一个判断:制造和部署成本真实存在,并非理论问题。问题在于,最影响决策的指标仍然缺失:收入、ARR、实际价格、硬件与软件组合、分层毛利率、账上现金、烧钱速度、跑道、积压订单、递延收入和客户集中度。公开材料也没有解释,尤其是在 Forge One 和 Leviathan 纳入范围后,制造到收入确认之间的营运资本高峰由哪一方出资。 这导向一个有纪律的结论。订单额增长应被视为需求证明,而不是收入质量证明。没有证据时,不应套用 NRR、CAC 回收期或 SaaS 毛利率等软件式指标。资本充足性也不能只凭标题融资额来假设,因为当前模式可能需要在经常性软件经济性变得重要之前,先投入工厂开支、GPU 采购、部署人力和应收账款融资。正确下一步不是建立一个英雄式模型,而是集中索取组合、利润率、积压订单和流动性资料。[CI021, CI022, CI023, CI024, CI025, CI037]

公开财务缺口表
缺失指标对尽调测算的影响为什么公开记录仍未披露精确尽调路径
收入和 ARR无法把预订额动能映射到估值或经常性质量私营公司;未公开财务报表或 KPI 材料索取月度收入桥、ARR 滚动表和已签署的收入确认备忘录
预订额定义和绝对基数540% 和 2,000% 增速无法换算成美元金额公开来源只披露百分比索取 FY25、FY26 和 Q1 FY27 的绝对预订额,以及按合同类型划分的定义
硬件与软件 / 服务收入结构决定毛利率路径,也决定软件式倍数是否站得住未公开分部披露索取 Galleon、Bridge、Atlas、Marketplace 和服务的产品线收入拆分
分层毛利率没有各层毛利率,无法判断利润率路径未公开损益表或分部成本数据索取硬件、软件、服务和支持的毛利率
现金、烧钱速度和现金跑道即使融资新闻很大,资本充足性仍无法验证未公开资产负债表或现金流披露向 CFO 索取现金余额、月度烧钱速度和现金跑道桥
工厂资本开支和 JCI 经济条款制造爬坡可能吃掉 B 轮的大部分资金框架协议经济条款仍属私下信息索取 Forge One 预算、资本开支责任矩阵和 JCI 定价条款
GPU 采购承诺和库存Leviathan 经济性取决于芯片预付款和库存周转未公开 GPU 采购或供应商融资披露索取 GPU 供货协议、定金和库存政策
客户集中度、积压订单和递延收入少数大合同可能主导风险和现金节奏已公开具名客户,但未公开集中度指标索取大客户占比、积压订单计划、递延收入和取消权
实际成交价与公开计价结构用量单位和采购路径无法揭示实际定价权公司披露机制,但不披露折扣索取发票样本、价格瀑布和各产品平均实际 ASP

空值有意保留:这些是截至 runDate 公开记录中缺失的最重要项目。

[CI021, CI022, CI023, CI024, CI025, CI039]

4.6 图表

Chapter 05

05产品与技术

5.1 Armada 的产品定义是一套四产品边缘平台,支点是可被证据支撑的 Galleon 家族

Armada 的官网和产品页面呈现的是一套连贯的全栈方案,而不是一家单产品公司。Armada Edge Platform 明确由四款产品构成:Atlas、Galleon、Bridge 和 Marketplace。关键在于,产品叙事覆盖了客户工作流:从连接和资产群可见性,到本地计算,再到 GPU 云管理和打包应用交付。最能被公开证据支撑的硬件细节在 Galleon 家族里。Armada 公开称,这条产品线从手提箱大小的 Beacon,到带 3 个计算机架的 20-foot Cruiser,再到带 5 个计算机架的 40-foot Triton,最后延伸到兆瓦级液冷 Leviathan。同一组来源也支撑一个务实部署判断:Galleon 预装计算、网络、存储、供热和制冷;Armada 称它可在数周内投运,而不是数年;本地处理工作流也明确设计成只把任务关键的信息回传云端,从而减少回传链路负担。仍缺的是各 SKU 在功耗、存储和网络上的精确边界,公开材料只给了这些宽泛描述。[CE001, CE002, CE003, CE004, CE005, CE006]

产品模块 / 资产矩阵
模块 / SKU主要用户 / 买方公开成熟度可由公开资料支撑的技术细节差异化尽调缺口
Atlas车队 / 网络运营团队已有产品页和试用 CTA在一个控制台管理 Starlink、SD-WAN、无人机、摄像头和传感器把连接能力和资产遥测收进同一个控制界面未公开 API 范围、可用性 SLA 或价目表
Beacon空间受限的远程现场团队已公开命名的 SKU手提箱大小的 Galleon,用于轻量级远程边缘处理最小形态把产品线下探到集装箱尺度以下未公开机架数量、功耗或载荷物料清单
Cruiser现场站点和紧凑型工业 / 防务部署已公开命名的 SKU20 英尺 Galleon,配 3 个计算机架,可选气隙配置把集装箱化计算带进空间受限场景未公开电力输入、GPU 组合或调试清单
Triton更重的工业、离岸和防务工作负载已公开命名的 SKU,且产品已部署40 英尺 Galleon,配 5 个计算机架,可选气隙配置Leviathan 之外最大的形态,重点面向加固现场未公开单机架密度、存储架构或可用性目标
LeviathanAI 工厂、主权云和高密度训练 / 推理买方2025 年公开发布,2026 年扩展兆瓦级液冷 Galleon,声称具备 10x Triton 计算能力,选址可贴合能源条件把 Armada 从边缘节点延伸到模块化 AI 工厂具体 MW 额定值、CDU 拓扑和支持的芯片 / 机架组合未公开
BridgeGPU 云运营商、数据中心、电信、大学、企业公开产品加文档界面IaaS / PaaS GPU 云平台,支持 Kubernetes、SLURM、JupyterHub、NIM、MIG、计费和 RBAC让 Armada 覆盖客户既有基础设施,而不只限于 Armada 硬件未公开性能基准或外部安全认证包
Marketplace平台管理员和解决方案团队公开产品页将自有 OpsAI 应用、合作伙伴应用和客户容器部署到 Galleon / AEP加快应用上架,而不是每次都强迫定制集成未公开收入分成、审核流程或支持的运行时矩阵
Azure Local on Galleon 部署需要边缘主权云的政府和受监管行业公开合作已可用Azure Local 加 AEP,运行在加固型模块化数据中心上,支持多机架和断连运行为 Armada 硬件加入与云一致的运营模型已验证的参考架构和认证未完整公开

本公开产品矩阵综合 Armada 产品页、文档、Microsoft 和 DCD;未知的 SKU 级规格有意留空,不做推断。

[CE001, CE003, CE004, CE005, CE006, CE010]
FE002: 客户工作流 / 运营流

Armada 的公开运营流把数据从受限边缘资产带入本地处理、策略控制和选择性云同步。

该流程抽象了工业、电信和公共部门部署,不是体现价值实现时间的 SLA 图。

[CE009, CE015, CE024, CE026, CE028, CE029]

5.2 控制平面有足够公开证据证明是软件定义:Atlas 管资产群运营,Bridge 管 GPU 云编排

Armada 的软件叙事足够具体,可以承销一个真实控制平面。Atlas 被定位为 Starlink 终端、SD-WAN、无人机、摄像头、传感器及其他连接资产的运营界面,配有共享数据套餐经济性、预测性监控,以及包含 SSO、RBAC、审计日志和 SOC 2 / ISO 27001 表述的安全面。Bridge 在技术上更关键,因为文档跳出了营销口径,展示出一套真正的 GPU 云平台。Armada 将 Bridge 写成结合 IaaS 和 PaaS 的层,支持多租户、Kubernetes 和 SLURM、JupyterHub 和 Ray 模板、NVIDIA NIM 推理模板、按 GPU 利用率自动扩缩,以及面向受支持 GPU 的 UI 驱动 MIG 分区。Marketplace 则补齐架构,让客户部署 Armada 自有工业应用、合作伙伴软件,或自己的容器化工作负载。一个开发者信号来源也强化了这个判断:Armada 正在招聘在裸金属 Kubernetes 上做本地 CaaS 和 GPUaaS 的岗位,技术栈包括 KVM、容器运行时、KubeVirt 和 vGPU;这和公开文档一致,不像无关招聘噪音。[CE015, CE016, CE017, CE018, CE019, CE020]

工作流 / 用例表
用户任务当前工作流痛点Armada 解决路径声称的可衡量收益限制
远程工业监测传感器、摄像头和无人机数据受弱回传或延迟审核拖住Atlas + Galleon 本地处理 + OpsAI 应用延迟更低、带宽占用更少,因为只有任务关键的信息上行公开来源未量化延迟降幅或误报率
受对抗或断连环境下的主权 AI公有云可能不可用,或不满足合规要求Azure Local on Galleon,配 AEP 编排和多路径连接在本地运行与云一致的 AI 和分析,同时保留主权与可审计性公开来源未披露认证边界文件或任务 SLA
既有 GPU 集群变现闲置或碎片化的 GPU 资产难以分配和计费Bridge,提供 GPUaaS 计费、RBAC 和可观测性利用率更高,多余 GPU 容量可变现公开来源未披露实际成交价或客户利用率提升
电信分布式 AI 网格对延迟敏感的推理需要 GPU 靠近用户和网络边缘AEP + NVIDIA AI Grid 参考对齐,可选 Galleon 模块在集中式、区域和边缘站点之间统一放置工作负载公开证据偏预生产架构,KPI 很少
公共部门快速响应部署应急和现场任务等不起新建工程加固型 Galleon,配本地计算、连接能力,以及 Azure Local / Palantir 生态选项现场部署更快,本地处理更有韧性站点前置条件、培训负担和持续现场服务义务未公开
数据科学和模型服务团队GPU 资源分散在不同集群和工具链中Bridge 模板支持 Ray、带 KAI Scheduler 的 JupyterHub 和 NVIDIA NIM更快拉起集群,且预装面向具体工作负载的软件公开文档未发布基准吞吐量或租户上限

工作流行把公开用例表述映射到 Armada 相应产品界面;收益为公开资料声称,除非另有说明,未独立量化。

[CE009, CE015, CE018, CE021, CE022, CE024]
技术 / 运营架构表
层 / 组件作用公开证据依赖主要风险
连接层用多种链路让分布式站点保持在线Atlas 产品页以及 Microsoft / Armada Azure Local 材料提到 Starlink、SD-WAN、卫星、LTE/5G 和 RF运营商 / 卫星接入和策略设计公开来源未量化故障切换行为或链路预算
Galleon 物理模块在现场提供加固型计算、存储、网络、供热和散热Galleon 产品页和产品族概览站点电力、运输、调试和现场维护单 SKU 的电气和热工边界未公开
AEP 编排层跨边缘站点做中央监控、生命周期控制和工作负载放置Armada / Microsoft 博客以及 NVIDIA AI Grid 新闻稿Armada 软件成熟度和合作伙伴集成质量公开材料偏描述,未披露控制平面架构图
Bridge GPU 云层多租户基础设施、集群创建、计费、可观测性和 GPU 控制Bridge 产品页和文档Kubernetes / SLURM 运维和 GPU 虚拟化栈未公开基准、渗透测试摘要或客户证明包
Marketplace 应用层将自有、合作伙伴和客户工作负载交付到平台上Marketplace 页面和合作伙伴生态提及应用验证和运行时兼容性未公开认证规则或应用商店治理文件
主权云层加入 Azure Local 控制平面、托管集群、存储选择和本地 AI 运行Microsoft Azure 博客和 Armada 合作页面Microsoft 参考架构和客户认证要求具体已验证拓扑和存储性能数据未公开
制造与服务层标准化生产,并在全球部署模块Forge One 和 Johnson Controls 框架协议供应链、热管理设备和现场服务执行工厂吞吐和供应商瓶颈仍属私下信息

架构层仅覆盖公开资料可支撑的组件;隐藏的内部服务或芯片级拓扑有意排除。

[CE008, CE015, CE018, CE021, CE024, CE026]
FE001: 产品架构图

Armada 的公开架构把加固物理模块放在底层,上面叠加编排、主权云和应用部署。

该堆栈根据公开产品、文档和合作伙伴材料综合而成,不是官方系统图。

[CE001, CE008, CE015, CE018, CE024, CE026]

5.3 Azure Local 集成和具名连接栈让主权边缘架构更可信,但公开保证仍不到完整认证细节

Microsoft 合作是本章最强的公开技术旁证之一,因为 Armada 和 Microsoft 描述了同一套基本架构。Azure Local 跑在 Galleon 模块化数据中心上,并与 AEP 互操作,形成主权私有云参考设计。Microsoft 称该设计支持多机架托管集群、超融合或 SAN 支撑的存储,以及跨卫星、LTE/5G、RF 和 SD-WAN 的韧性多网络连接。Armada 也提出类似说法:该方案能在通信受限环境中保持关键系统在线,并可在数周而不是数月内部署加固模块。主权叙事也与 Armada 的本地处理表述一致,Carahsoft 和 Dell/Palantir 的合作页面同样强调在客户控制的基础设施上运行敏感工作负载。但信任图景仍不均衡。Atlas 的公开控制集最清楚,Azure Local 被描述为加固且可审计,Bridge 文档写到硬隔离和 RBAC;但已审阅公开材料没有给出 Bridge 或 Galleon 专属认证清单、正式物理隔离验证包,或公开可用性承诺。[CE010, CE017, CE018, CE019, CE020, CE021]

信任 / 质量 / 合规表
控制 / 质量项状态范围产品区域剩余缺口
SSO公开提及Microsoft Entra、Google、Okta 等身份提供商Atlas未公开映射到非 Atlas 产品
RBAC公开提及多账户支持和细粒度权限Atlas 和 Bridge未公开政策样例或职责分离材料
审计日志公开提及详细记录用户活动和变更Atlas;合作伙伴材料还加入更广的可审计性表述未公开留存或防篡改策略
SOC 2公开提及平台认证表述Atlas未公开报告或控制矩阵
ISO 27001公开提及平台认证表述Atlas未公开说明其覆盖 Bridge 或 Galleon 运营
气隙运行已公开说明适用于特定硬件Cruiser 和 Triton 可配置为完全断连运行Galleon 系列未公开证明材料或认证备忘录
加固型主权云架构公开提及Azure Local 加固安全、合规和断连运行Azure Local on Galleon / AEP 组合具体认证和边界文件未公开

本表只记录获取到的公开材料中明确提到的控制;这里缺失某项控制,并不证明公司私下没有。

[CE010, CE017, CE018, CE019, CE026, CE028]
FE003: 关键依赖图

Armada 架构对合作伙伴云、GPU 生态、连接网络、热管理 / 制造执行的依赖, 不亚于对箱体本身的依赖。

这个 DAG 只标出外部可见依赖;内部零部件供应商和未公开软件服务未纳入。

[CE018, CE024, CE026, CE028, CE030, CE032]

5.4 部署速度、能源灵活性和液冷是实在的产品主张,Forge One 则是把架构变成制造能力的押注

Armada 的公开部署主张刻意激进:数周投运,而不是数年。支撑这一主张的最强产品证据来自预配置 Galleon、Microsoft 与 Armada 的 Azure Local 集成叙事,以及一条公开信息:某盟友在 6 天内部署了一台 Triton。Leviathan 把这个故事从边缘计算延伸到模块化 AI 工厂基础设施。Armada 和 DCD 的独立报道都把 Leviathan 描述为液冷、兆瓦级产品,Armada 还称它可以与闲置天然气、太阳能、核能及其他替代能源共址。Forge One 是把这份雄心落地的运营层。Armada 和 Johnson Controls 称,Arizona 工厂最高可达 400,000 sq ft,创造约 500 个岗位,并以 Leviathan 启动连续生产;Johnson Controls 则带来先进热管理能力和庞大全球现场服务队伍。这应能提升标准化和部署可重复性,但也把公司的执行负担从单纯软件交付,推向制造吞吐、热设计和能源设备。公开来源仍未披露 Forge One 能生产多少单元、供应商瓶颈在哪里,或哪些场址准备假设必须成立,数周上线的主张才站得住。[CE002, CE011, CE012, CE013, CE014, CE026]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
2025-07Leviathan 发布已公开发布把 Galleon 从边缘模块延伸到兆瓦级液冷 AI 基础设施Armada 融资 / Leviathan 博客;DCD Leviathan 报道
2025-07 起计划在能源 / 工业站点部署第三方报道公开提及说明 Leviathan 和 Galleon 绑定真实部署项目,而不只是概念图DCD Leviathan 报道
2026-03Azure Local on Galleon / AEP 合作已公开宣布,且现已可用为技术栈加入主权私有云和断连 AI 路径Armada 和 Microsoft 博客
2026-03NVIDIA AI Grid 对齐已公开宣布把 AEP 定位为分布式 AI 网格的控制平面,可覆盖既有和新建站点PRNewswire Armada / NVIDIA AI Grid 新闻稿
2026-05Galleon Forge One 与 Johnson Controls已宣布框架协议推动 Armada 走向连续制造,并复用热管理 / 现场服务支持Armada / Johnson Controls Forge One 公告
当前文档界面Bridge 工作负载模板和 MIG 运维公开文档可用说明 Bridge 正在成熟为面向运营者的平台,而不只是落地页Armada 文档概览、Kubernetes、SLURM 和 MIG 页面

路线图行记录截至报告日期可在公开来源看到的产品和制造里程碑,而不是私下的内部发布计划。

[CE011, CE012, CE022, CE026, CE030, CE032]
FE004: 产品成熟度 / 能力图

Armada 的公开成熟度在产品定义和编排广度上最强,在已发布运营规格和工厂吞吐披露上最弱。

单元格反映从抓取的公开证据得出的尽调成熟度,不是内部 KPI。

[CE004, CE011, CE015, CE018, CE024, CE026]

5.5 公开记录显示架构可信,但最影响决策的运营规格仍是缺口

本章可以以相当高置信度支撑 Armada 的产品逻辑:硬件家族是真的,软件控制平面是真的,主权云集成路径是真的,Leviathan 背后也有真实制造计划。约束同样真实。两篇独立 TechCrunch 文章有用,因为它们解释了为什么模块化 AI 基础设施并不能绕开行业普遍的电力和散热瓶颈:电力转换损耗很实在,涡轮机和其他能源设备交期很长,AI 基础设施的经济性越来越取决于能源可得性,不只是算力可得性。Armada 自己的公开材料没有解答这些问题。除宽泛形态外,公司没有发布各 SKU 的功耗或制冷边界,没有发布场址前置条件矩阵或可用性 SLA,也没有披露 Bridge 或 Galleon 的详细安全认证集。因此,产品架构作为概念可以投资,成熟度也明显高于 PPT 堆栈;但完整承销仍取决于一套私有技术包,量化功耗、部署可重复性、安全认证和工厂吞吐。[CE007, CE023, CE035, CE038, CE039, CE040]

5.6 图表与证据

Chapter 06

06客户情况

6.1 客户模式:主权、远程、运营受限的买家

Armada 的公开客户证据不是广泛的横向企业采用,而是集中切入那些连接、主权或延迟让中心云工作流失效的环境。最清楚的具名终端用户是 Alaska DOT&PF、Washington DNR、U.S. Navy 在 UNITAS 期间的使用,以及 Aker BP。行业不同,但这些买家有相似模式:政府或工业运营方在远程或受监管环境中承担任务;一线用户需要本地计算或集中治理的连接;买家还需要快速部署,不能等传统数据中心建设。Alaska 的无人机和地理空间工作流、Washington DNR 的野火连接运营、Navy 的海上演习环境,以及 Aker BP 的海上钻井数据流,都符合这个模板。Carahsoft、Microsoft、Second Front、Skydio、DOE Genesis Mission、WinDC 和 Aramco 又把图景从单个证明点扩展到相邻需求面。结论是一个连贯的客户逻辑:边缘基础设施不是优化项而是运营前提时,Armada 最先赢单。[CU001, CU002, CU004, CU007, CU010, CU011]

客户分群表
分群买方 / 用户 / 付款方代表性证明用例当前判断
州交通与应急运营州机构采购;现场运营人员和 GIS / 无人机团队使用Alaska DOT&PF无人机影像、滑坡 / 雪崩 / 洪水响应、地理空间处理活跃使用证据强
野火和远程公共安全运营州机构采购;事故响应团队使用Washington DNR托管 Starlink 连接、野火协同、远程团队工作流证据强,但公开案例只有一个
防务 / 海事运营方联邦买家;作战人员和任务系统操作员是用户美国海军 UNITAS 2025断连环境下的计算、网络态势感知、多源情报任务负载证明能在演训中运行,但不是正式项目经济性
海上能源运营方工业买家;钻井和远程运维团队是用户Aker BP钻井平台上的钻井数据处理和模型执行已签约参考部署,具备复制逻辑
公共部门渠道买家机构通过经销商采购;IT 和运营人员是用户Carahsoft 触点合同载体采购、演示、Commander Connect / Starlink 接入渠道接入已验证;终端客户规模未披露
可再生能源驱动的 AI 基础设施合作伙伴基础设施合作伙伴采购;未来企业租户是用户WinDC可再生能源站点的便携式 AI 工厂证明行业可扩张,但终端客户名单尚未铺开

各行区分已确认的具名部署触点,以及让 Armada 看起来具备需求相关性的买家—用户—付款方模式。

[CU001, CU002, CU007, CU011, CU015, CU018]
具名客户验证表
客户 / 渠道细分领域部署 / 用例生产还是试点结果或证据局限
Alaska DOT&PF州级公共部门无人机影像、地理空间分析、远程灾害响应运行中部署 / 案例研究工作流从 >28 小时缩短到接近实时;据报道有两台 Galleon未披露合同金额、续约或全州推广经济性
美国海军 UNITAS 2025防务岸上和舰载模块化边缘计算,加上 Atlas 监控作战演习部署Flankspeed Edge 和 Minotaur 工作负载在断连海上环境下运行未披露正式项目授标、合同上限或舰队推广
Aker BP海上能源钻井平台上的钻井和运营数据处理已签约参考部署单台参考 Galleon 用作新增资产蓝本仍处早期,还不是多钻井平台生产推广
Carahsoft公共部门渠道体验中心加合同载体采购触点渠道接入 / 演示基础设施联邦、州、地方、教育和医疗买家可通过已知载体评估和采购渠道证据不等同于已披露终端客户 ARR
Second Front + Microsoft 在 Armada 上防务软件生态Armada Galleon 内 Azure Local 上的 Frontier成功的合作伙伴部署证明关键任务应用可迁移到 Armada 基础设施合作伙伴主导的证据点,不是具名机构合同
WinDC可再生能源 AI 基础设施澳大利亚可再生能源站点的便携式 AI 工厂已签署部署计划已宣布 11 MW,首台单元已在澳大利亚终端客户名称、定价和利用率仍未公开

清单有意只覆盖一部分,涵盖截至 2026-05-24 对客户尽调最相关的 Armada 公开具名证据触点。

[CU004, CU011, CU015, CU018, CU019, CU025]
FU003: 客户验证矩阵

Armada 的证明质量在运营细节上最强,在已披露合同经济性或留存可见度上最弱。

矩阵评估公开证据质量,而不是客户质量;收入可见度低,反映的是披露缺口,不代表产品反馈为负面。

[CU004, CU009, CU011, CU015, CU018, CU030]

6.2 部署证明真实存在,但账户成熟度不同

Armada 客户章节最强的地方在于证明很具体。Alaska DOT&PF 描述了一个工作流:从依赖存储卡、延迟超过 28 小时,转为近实时决策支持;Data Center Dynamics 报道称,该部门现在运营两台 Galleon。Washington DNR 的故事也偏运营而非愿景:Atlas 把碎片化、靠 P-card 采购 Starlink 的方式,替换为面向野火和远程政府场景的集中管理连接。UNITAS 2025 在任务可信度上更进一步,因为 Armada 称一台 Galleon 和 Atlas 在岸上和舰上使用,多方来源称该平台支撑了 Flankspeed Edge 和 Minotaur 工作负载。Aker BP 又不同:它不是大规模铺开,而是一项签署的单钻井平台参考部署,明确目标是成为可复制蓝图。这些差异很重要。Alaska 和 Washington 像活跃使用场景,Navy 像一次运营演习部署,Aker BP 像签约后的先落地再扩张起点。这比客户名单证明更强,但还不是一个成熟、已披露安装基数且能展示复购经济性的客户基础。[CU003, CU004, CU005, CU008, CU009, CU011]

客户增长 / 采用轨迹表
信号公开细节来源依据含义缺失分母
阿拉斯加工作流压缩从超过 28 小时压到接近实时Armada 案例研究 + DCD实际使用收益清晰,不只是品牌露出未披露合同金额或用户数
阿拉斯加装机足迹两台 Galleon 投入运行DCD证明部署单元不止一台未披露利用率或支出数据
Washington DNR 治理迁移35 个未托管 Starlink,转为约 45 个通过 Atlas 管理的 StarlinkArmada 案例研究买家痛点是治理和现场连接,不只是计算没有合同期限或续约数据
UNITAS 作战使用Galleon 和 Atlas 在岸上和舰上使用Armada + PR Newswire + DCD防务验证已进入作战演习条件未披露后续授标或正式项目
Aker BP 推广阶段单台参考 Galleon 计划用作新增资产蓝本Armada + World Oil先落地再扩张的逻辑很明确除首座钻井平台外,没有时间表或资产数量承诺
WinDC 扩张信号已宣布 11 MW,首台单元已落地澳大利亚Armada + DCD + PV + W.Media 来源组合显示行业和地域扩张未披露具名终端需求客户或收入条款
重复客户经济性未公开披露无公开来源最大采用缺口仍是持久性和变现缺少 ARR、NRR、客户数和合同金额

轨迹表跟踪公开部署和渠道里程碑,不是内部 CRM 漏斗或收入台账。

[CU003, CU004, CU005, CU008, CU009, CU011]
FU002: 采用 / 部署漏斗

公开证据从大量合作伙伴和行业信号,很快收窄到极少数已披露的经常性经济性指标。

数值统计的是本章审阅过的公开证据项,不代表销售管线、客户数或收入指标。

[CU004, CU011, CU016, CU022, CU030, CU041]

6.3 采购渠道和行业扩张让触达速度快于终端客户披露

Armada 下一层客户证据并不总是具名付费终端用户;很多时候,它是让未来获客更可信的渠道或合作伙伴入口。Carahsoft 是最干净的例子。Reston 体验中心和转售商页面显示,Armada 已经拥有跨多个合同工具的公共部门采购路径;更早的 Commander Connect 文章也显示,Starlink 加 Armada 软件可以通过 NASPO 触达州和地方买家。Microsoft 通过把 Armada 包进 Azure Local 和面向国防、公共安全、能源及受监管行业的主权私有云叙事,进一步加深了这种入口。Second Front 在这套基础设施之上补充了任务关键国防软件证明,Skydio 则延伸了国家安全和公共安全无人机故事。美国公共部门之外,WinDC 和 Aramco 展示了向可再生能源驱动的 AI 基础设施和工业分布式云部署扩展;DOE Genesis Mission 则显示公司进入了大型联邦科学计划的合作者名单。这些都不能替代硬性的续约或 ARR 证据,但合在一起说明 Armada 正围绕同一核心用例搭建可复制分销:在传统云或地面基础设施太慢、太脆弱或太中心化的地方,交付主权计算和托管连接。[CU018, CU019, CU020, CU021, CU022, CU023]

采购渠道与行业扩张表
触点公开内容客户相关性证据质量局限
Carahsoft 体验中心Reston 的实体 Galleon 演示中心加快机构、教育和医疗买家的评估演示基础设施,不是已预订收入
Carahsoft 合同载体SEWP、ITES-SW2、NASPO、TIPS、OMNIA、Quilt 等合同通道公共部门买家的可复制采购路径量级和续约未披露
Microsoft Azure Local已可用的主权私有云方案,并明确提到已有部署切入受监管行业和防务买家的讨论未披露客户数和 ACV
Armada 上的 Second FrontFrontier 在 Galleon 内 Azure Local 上成功部署证明任务软件可在战术边缘的 Armada 上运行合作伙伴证据,不是具名终端客户合同
DOE Genesis Mission大型联邦科学生态将 Armada 列为合作方释放联邦相关性和关系入口信号不是付费客户部署证据
WinDC / Aramco澳大利亚可再生能源 AI 基础设施,以及沙特工业分布式云证明地域和行业扩张超出美国公共部门商业条款和终端需求客户仍未公开

本表跟踪可复制接入触点和相邻市场信号,而不是假设每条合作伙伴公告都已转化为 ARR。

[CU018, CU020, CU022, CU025, CU028, CU030]
FU001: 客户旅程图

Armada 的公开客户旅程先从远程或主权运营痛点起步,再经由渠道或证明性部署推进, 之后才可能变成重复基础设施支出。

这张旅程图反映案例研究和合作伙伴公告暗示的公开采用路径,不是已披露的内部销售漏斗。

[CU001, CU010, CU018, CU020, CU024, CU038]

6.4 主要客户风险不是相关性,而是转化、留存和集中度

反向读法很直接。Armada 的公开客户故事更擅长说明产品能做什么,而不是这些账户的收入随时间如何表现。已审阅来源没有披露客户数、NRR、GRR、客户流失、任务订单历史或账户级扩张经济性。因此,公开证明以案例和合作伙伴为主,而不是以队列数据为主。这带来三类承销风险。第一,集中度风险可能很有意义,因为具名证明集仍然小,并集中在少数展示账户。第二,即便底层技术可行,试点转生产风险仍真实存在;独立企业 AI 研究仍警告,成功试点常常因治理、运营模式和生产执行落后于演示而无法规模化。第三,即便有有用的转售商和采购改革,公共部门转化仍可能受采购流程中介。Carahsoft 降低摩擦,FY2026 NDAA 改革也旨在加快商业技术采购,但两者都不等于已披露复购订单。简言之,Armada 的客户证据足以证明相关性和早期采用,但公共数据还不足以证明耐久性或广泛收入多元化。[CU019, CU024, CU032, CU035, CU036, CU037]

留存 / 重复使用 / 满意度表
指标或代理指标公开数值最佳可用代理置信度尽调要求
净留存率未公开披露索取按队列和垂直领域拆分的 NRR
总留存率未公开披露索取 GRR、客户流失和收缩数据
续约时间阿拉斯加、Washington DNR、美国海军或 Aker BP 均未披露索取开始日期、续约日期和当前合同状态
重复推广证据部分Aker BP 新增资产蓝本;WinDC 多站点计划中低区分蓝本表述和已签署后续订单
客户满意度只有案例研究和合作伙伴引述索取 NPS,或超出精选故事的客户访谈证据
渠道可复制性UnknownCarahsoft 采购载体和体验中心已公开,任务订单历史未公开索取按载体拆分的经销商预订、续约和机构数量

空值表示所审阅来源中没有公开指标;代理行展示在不把精选案例研究当成留存证据的前提下可推断的内容。

[CU016, CU019, CU030, CU035, CU036]
扩张与集中度风险表
风险或上行方向重要性公开信号尽调路径
具名证据集偏小风险少数展示型客户可能主导路线图和客户背书公开证据集中在阿拉斯加、Washington DNR、UNITAS、Aker BP、Carahsoft 和 WinDC索取头部客户收入集中度,以及按垂直领域拆分的管线
演习到正式项目的转化风险防务演示能验证相关性,却未必带来持久收入UNITAS 显示作战使用,但没有披露后续授标索取 CRADA 里程碑、任务订单和任何正式项目路径
单钻井平台参考部署风险只有蓝本能复制,Aker BP 才有说服力明确从一座钻井平台起步索取安装后成功标准和后续资产承诺
Carahsoft 采购杠杆上行已知载体降低机构采购摩擦SEWP、ITES-SW2、NASPO、TIPS、OMNIA 和 Quilt 已公开索取每个载体下的实际订单数和机构集中度
主权云合作伙伴栈上行Microsoft 和 Second Front 拓宽面向受监管买家的方案广度Azure Local 和 Frontier 在 Armada 触点上运行索取哪些垂直领域或机构从演示转入付费生产
地域和行业扩张上行WinDC 和 Aramco 显示需求不只来自美国公共部门这一切入点澳大利亚和沙特证据已公开浮现索取按地区和工业细分拆分的收入结构

本表把真实的扩张期权,与同样真实的风险拆开:当前公开证据集仍然窄,且高度依赖参考客户。

[CU017, CU020, CU024, CU032, CU037, CU040]
Chapter 07

07风险

7.1 执行风险占主导,因为 Armada 正从证明点扩到工业化产出

Armada 不像一家还在找用例的公司。公开证据显示,国防、州政府、海上能源和主权私有云都有真实需求面。风险在于,公司正大幅跨越运营复杂度,同时外部预期也在抬升。Series B、Johnson Controls 框架协议和 Forge One 发布,把 Armada 从交付加固单元推到证明 Leviathan 及更广泛 Galleon 需求可以连续生产。承销问题因此从产品是否可信,变成制造、部署和现场运营能否在客户耐心或资本市场容忍度耗尽之前规模化。 公开记录中最强的反向事实不是需求弱,而是收入不透明。Armada 披露的是订单额增长,不是收入、ARR、毛利率、待交付订单转收入,或收入确认政策。投资者因此能看到需求加速,却看不出它多快转为已确认收入或耐久的装机经济性。公开客户证明集也仍相对集中在少数具名参考,其中几项是演习或单一参考部署,而不是已披露的大规模部署。因此,最主要的剩余风险是工厂执行、订单额转收入、集中度和合规复杂度,而不是市场缺席。[CR001, CR002, CR003, CR004, CR005, CR006]

按严重度排序的风险登记表
风险证据依据可能性影响缓释成熟度剩余敞口投资含义
工厂执行 / Forge One 爬坡连续生产、Leviathan 发布和 400,000 sq. ft. 工厂计划,把 Armada 推入真正的制造执行阶段。严重中等必须先尽调吞吐量、供应商和良率,才可纳入投资假设。
预订转收入不透明Armada 披露预订增长,但不披露收入、ARR、毛利率、积压订单转化或收入确认政策。严重不能把公告层面的需求当成持久经济性的证据。
公共部门 / 防务集中Carahsoft、UNITAS、阿拉斯加和 Washington DNR 让切入点可信,但公开证据集很小,且集中在政府相关用例。中等预算节奏和采购摩擦仍可能显著影响收入。
主权 AI 合规复杂度跨境 AI、出口管制、AI 治理规则和买家特定网络安全义务都在收紧。中等中高商业推进速度取决于可复用合规包,不只看产品能力。
电力 / 冷却 / 场地就绪约束行业研究显示,电力接入、并网等待、冷却设计和施工延迟是关键瓶颈。低中即便客户需求不变,Leviathan 规模的增长也可能延后。
战略合作伙伴依赖Johnson Controls、Microsoft 和 Carahsoft 都能降低摩擦,但也占据关键制造、平台和渠道路径。中等中高任何合作伙伴放慢速度,都会收窄 Armada 最快规模化路径。
严苛环境下的网络 / OT 韧性防务、海上和远程部署会放大安全与可靠性故障的后果。低中中高控制成熟度必须跟上部署野心。
竞争挤压大型云和 AI 基础设施厂商可以捆绑更广的方案,并争夺同一批有电力保障的站点和买家。中高商业化路径和运营证据必须保持差异化,不能只靠硬件形态。

风险排序基于已引用的公开证据,不基于私有尽调或管理层指引;剩余敞口假设没有新的私有披露。

[CR003, CR004, CR005, CR006, CR022, CR028]
FR001: 风险热力图

剩余风险集中在工厂执行、收入转化不透明和集中度,而不是某个已知法律事件。

[CR003, CR005, CR022, CR032, CR033, CR038]

7.2 主权 AI 扩张放大市场,也增加出口、采购和网络义务

Armada 的主权 AI 定位之所以有商业吸引力,正是因为它落在比普通云转售更难的监管和运营环境里。Microsoft 合作明确瞄准国防、政府和受监管行业,Carahsoft 与 UNITAS 也显示公共部门和军事暴露不是假设。但同样的市场路径也带来更多义务。BIS 对先进计算出口的更新提高了芯片来源、最终用途筛查和转移控制的重要性。EU AI Act 实施栈为跨境部署再加一层文档、透明度,以及高风险 / 通用模型义务。与此同时,FY2026 NDAA 及相关国防指引显示,DoD 买家正在提高对 AI 治理、测试、采购要求和能源感知数据中心规划的期待。 网络安全和安全负担会随着部署环境而上升。CISA 的 AI-in-OT 指南和 DoD 的 AI 网络安全定制指南,都把运营 AI 描述为生命周期治理问题,而不是简单的软件补丁问题。对 Armada 而言,这意味着在对抗性、断连或工业环境中的主权部署,不仅要满足可用性目标,还要满足模型、数据、基础设施和供应链控制。已审阅来源包中没有发现 Armada 专属公开执法行动或已披露重大网络事件,但这种缺席应被解读为证据边界,而不是合规负担低的证明。[CR008, CR009, CR010, CR011, CR012, CR013]

监管 / 法律风险登记表
规则 / 敞口司法辖区当前公开状态可能性严重度缓释措施剩余敞口尽调路径
先进计算出口管制与转移筛查美国 / 全球BIS 在 2025 年更新了 AI 芯片管制立场,并强调尽调和转移风险红旗。及早梳理芯片来源、终端用户和受限国家工作流。中高审查 ECCN、供应商证明和各国推广控制。
高风险和通用 AI 使用的《欧盟 AI 法案》义务欧盟实施工具已上线;透明度、文档以及风险 / 网络安全义务正在分阶段生效。按产品与用例打包合规材料,不要只套一个全球模板。按部署原型和买方类型取得 EU 法律顾问备忘录。
DoD AI 治理、采购和测试要求美国国防NDAA 和 DoD 指引抬高了对治理、采购控制、测试、数据中心能源规划的要求。在产品和销售流程中内嵌可复用的 ATO、网络安全和测试材料。索取按买方划分的 ATO/cATO 状态、测试计划和所需网络安全材料。
临时拨款决议与采购时点风险美国联邦GAO 称,CR 会延后合同、推高成本,并限制新项目启动或增产。中-高尽可能用工业和非联邦需求平衡受预算影响的客户。中-高将管线映射到拨款来源、合同工具和预期授标时间。
Armada 公开执法或诉讼可见度美国 / 全球本章审阅的公开来源没有发现直接针对 Armada 的执法行动或披露的重大安全事件。把当前判断视为受公开记录限制,不要当成健康证明。Unknown开展律师级别的案卷、制裁、禁入和事件检索。

各行按当前剩余严重性排序。本表是基于公开记录的风险清单,不能替代出口、采购或诉讼法律顾问审查。

[CR022, CR023, CR024, CR025, CR026, CR027]

7.3 工厂爬坡、电力接入和恶劣环境交付,是最尖锐的运营风险

Forge One 是风险章节最重要的运营催化剂,因为它把 Armada 从一家能组装和部署模块化系统的公司,变成一家必须在更大规模上证明吞吐、质量、供应商协调和现场服务的公司。Johnson Controls 的帮助很实在:该合作方带来热管理专长、全球现场队伍和正式制造协议。但同一安排也集中风险。如果 Forge One 在时间、良率或供应商准备度上失手,Armada 回旋空间更小,因为 Leviathan、主权 AI 需求和客户预期现在都已经进入市场叙事。 更广的行业背景让这件事更难,而不是更容易。Belfer、Deloitte、JLL 和 CBRE 都指出,2025-2026 年 AI 基础设施项目的核心约束是电力并网、设备瓶颈和施工延迟。JLL 称,电力而非成本或地理位置,是首要选址变量;CBRE 显示空置率创历史低位、预租活跃;Belfer 强调需求超过可用容量时,电网压力和项目延迟会带来风险。Armada 还卖进比标准企业数据机房更难运营的环境。Aker BP 的海上参考系统和 Navy 的海上部署证明了差异化,但也意味着每个已部署资产都要承担更高的物流、维护、安全和可靠性负担。[CR003, CR004, CR019, CR020, CR021, CR028]

运营 / 质量 / 安全风险清单
失效模式当前证据可能性严重性缓解成熟度剩余暴露未解决缺口
Forge One 爬坡在吞吐、良率或供应商就绪度上失手公司承诺连续生产,但没有公开吞吐量、BOM 和交期细节。严重中等没有公开的单台吞吐量、良率或供应商承诺数据。
电力和场地就绪延误拖慢 Leviathan 部署行业研究显示,电力获取、并网和电气设备交期会造成延误。低-中Armada 没有公开场地就绪矩阵或电力来源披露。
热管理和冷却设计在高密度下表现不及预期Leviathan 面向高密度 AI 工作负载销售;整个行业的冷却复杂度都在上升。中等中-高没有公开的热测试、正常运行时间或现场故障数据。
离岸 / 海上等严苛环境推高物流和维护负担UNITAS 和 Aker BP 证明该方案适用于偏远、争议或离岸场景。低-中中-高这些部署没有公开 MTBF、服务 SLA 或备件披露。
网络 / OT 攻陷影响已部署的 AI 基础设施CISA 和 DoD 都把运营中的 AI 视为治理、供应链和监控问题。低-中中-高公开认证和事件历史细节仍然单薄。

模块化制造、高密度 AI 基础设施和现场部署复杂度交叉处,运营风险最高。

[CR003, CR004, CR019, CR020, CR021, CR028]
FR002: 风险传导图

运营延迟传导到收入转化、融资需求和估值压力时,影响最大。

[CR003, CR006, CR027, CR032, CR035, CR043]

7.4 Armada 最好的商业助推器也是集中点

Armada 的合作伙伴栈是优势,也是公司主要依赖关系的地图。Microsoft 扩展了主权云控制平面,并给 Armada 一个面向受监管买家的验证叙事。Carahsoft 降低联邦、州、地方、教育和医疗账户的采购摩擦。Johnson Controls 提供制造和热管理深度。这些关系降低了获客和交付风险,但也意味着规模越来越依赖 Armada 不能完全控制的交易对手。合作伙伴路线图停滞、联合销售放慢,或渠道优先级下降,都可能收窄 Armada 进入那些当前让故事可投资的行业的路径。 即使没有私有收入数据,客户集中度风险也看得见。公开证明集仍集中在少数具名公共部门、国防和工业参考账户周围,几项旗舰胜单也被明确描述为演习或单一参考系统,而不是大规模铺开。竞争压力会放大集中度风险。CoreWeave 的 10-K 是有用的行业类比,因为它显示 AI 基础设施公司可以长期高度集中,并面对更大云厂商的竞争;这些厂商能打包更广产品,并利用既有客户关系。Armada 仍未披露通常能缓释这些风险的指标:收入、ARR、利润率、待交付订单转化和营运资本结构。因此,即便完成大额 Series B,融资依赖仍然真实存在。[CR007, CR012, CR013, CR016, CR017, CR018]

合作伙伴 / 依赖风险清单
依赖对手方角色集中度 / 暴露失效场景严重性缓解措施剩余暴露
制造与热部署合作伙伴Johnson Controls工厂爬坡、热系统、现场支持Forge One 在进度、质量或规模上交付不足。框架协议叠加合作伙伴投资,能对齐激励。中-高
主权私有云平台合作伙伴MicrosoftAzure Local / Foundry Local 控制平面与企业验证联合路线图或销售动作放缓,削弱受监管行业可信度。Armada 保留自有硬件和 AEP 栈。中-高
公共部门分销渠道Carahsoft合同工具、转售商生态、演示中心中-高渠道准入没能转化为可持续复购。直接工业渠道和其他合作伙伴存在,但公开验证较少。
公共部门和国防标杆客户U.S. Navy、Alaska DOT&PF 与 Washington DNR标杆部署与可信度中-高预算、采购或任务优先级调整拖慢后续需求。工业和主权 AI 邻近需求存在,但公开规模仍小。
资本提供方和战略投资者Overmatch、BlackRock、Johnson Controls 等增长资本与可信度工厂和营运资本需求超过已披露融资证明。中-高近期大额融资有助于短期。

帮助 Armada 扩张的依赖,也界定了公司当前的集中点。

[CR002, CR007, CR008, CR012, CR013, CR035]
人员 / 执行风险清单
角色 / 职能依赖或缺口可能性严重性缓解措施尽调路径
创始人 / CEO 对外领导力Dan Wright 在工厂、产品和 GTM 同时扩张期间,仍是最主要的公开发言人。扩大可见运营班底,并下放客户 / 合作伙伴责任。索取当前组织架构、继任覆盖和运营节奏。
制造与热运营领导力Forge One 和 Leviathan 需要超出创业公司式集成的工业执行力。借助 Johnson Controls 经验,并招聘有经验的工厂运营人员。索取具名工厂负责人、KPI 和制造就绪评审。
政府合规与安全领导力国防和主权 AI 买方需要 ATO、出口、网络安全和治理打包能力。尽早建立专门合规工作流,不要逐单处理。索取合规组织责任归属、控制矩阵和客户特定例外清单。
现场服务 / 严苛环境运营离岸、海上和远程部署推高维护与支持需求。借助合作伙伴现场覆盖,并标准化服务手册。按部署类型索取 SLA、备件模型和升级路径。
财务 / 营运资本规划硬件加软件扩张可能在经常性经济性显现前形成现金低谷。让制造支出匹配已承诺需求和付款条款。索取现金使用桥、客户付款条款和库存融资计划。

执行风险不只是技术问题,也是组织问题,尤其是在 Armada 同时扩张多个职能时。

[CR041, CR042, CR043, CR044, CR045, CR047]
FR003: 依赖关系图

Armada 目前的规模化路径,穿过制造、平台、渠道、客户、电力和资本这几类小而关键的依赖。

[CR002, CR008, CR012, CR040, CR047]

7.5 有缓释项,但投资逻辑仍取决于尚未公开的证据

缓释故事足够可信,能让 Armada 仍具可投资性。Johnson Controls 降低热管理和制造执行风险。Microsoft 与 Carahsoft 降低渠道和主权叙事风险。恶劣环境案例证明,标准云模式失效时,这个产品有价值。但缓释逻辑仍主要来自架构和合作伙伴;还没有配套的公开运营披露,让投资者可以实质性下调风险。我们仍没有 Forge One 的公开吞吐数据、从订单额转收入的披露、按金额计算的客户集中度、详细合规包,或达到法律顾问级别的诉讼和事件检索结果。 因此,风险章节应被读作一个可监测的承销框架,而不是静态风险信号清单。如果 Forge One 爬坡明显失手,如果公共部门采购周期拉长且客户集中度仍高,如果合规打包落后于主权 AI 雄心,或公司不得不在尚未证明硬件加软件经济性之前为硬件式营运资本融资,投资逻辑就会断裂。当前公开证据支持真实机会和真实缓释,但还不足以让人忽视狭窄执行窗口。[CR005, CR022, CR027, CR032, CR040, CR042]

缓解措施与否决标准表
风险可监控触发项阈值 / 事件行动含义
Forge One 执行工厂投产和单元产出爬坡明显延误,或到下一个运营周期吞吐量仍未披露。升级尽调;在看到制造证明前降低确信度。
订单额质量收入 / 订单积压 / 确认披露管理层仍只披露订单额,回避收入转化证据。可把需求视为真实,但经济性仍未验证;避免按软件逻辑承销。
公共部门集中新披露非政府生产客户具名部署仍集中在国防 / 公共部门,缺乏多元化。假设波动更高、扩张更慢。
出口 / 主权合规面向买方的合规包和出口控制跨境销售在可复用合规文件可见前先行扩张。加大法律 / 合规尽调,并下调部署推进假设。
电力 / 场地就绪并网和通电场地交期项目依赖通电场地,但没有锁定时间表或备用发电方案。提高部署缓冲和资本假设。
合作伙伴依赖联合销售和合作伙伴交付证据Microsoft、Carahsoft 或 Johnson Controls 活动放缓,或更像宣传而非运营。重新评估销售路径韧性。
网络 / OT 韧性独立安全证据部署扩大后,仍没有有意义的认证、ATO 或事件管理证明出现。把安全视为尽调闸门,而不是勾选项。
领导层带宽运营班底可见度工厂、现场和合规范围扩大时,创始人集中度仍高。重新评估执行能力和继任覆盖。

否决标准是承销触发项,不是预测。每一项都被选中,是因为可以从公开或尽调获得的运营证据中监控。

[CR003, CR005, CR022, CR027, CR032, CR040]
反向信号日志
信号证据为何反向抵消证据后续监控项
披露订单额,但缺少收入质量指标Armada 宣传订单额增长,但不披露收入、ARR、利润率或订单积压转化。需求可能真实,但经济性质量仍未知。大额融资和多个具名部署意味着市场拉力存在。收入转化、订单积压时间表和利润率披露。
公开证明样本小具名部署集中在 Alaska、Washington DNR、UNITAS、Aker BP 和合作伙伴渠道。标杆集中可能夸大多元化程度。用例真实,且跨多个行业。新的具名生产客户和复购订单。
标杆系统多于成批铺开UNITAS 是一次演习,Aker 也从单个参考安装开始。从试点走向成批部署的执行风险仍然不低。这些标杆确实证明方案能在真实运营中跑通。成批扩张、续约和多站点铺开证据。
全行业电力和延误瓶颈JLL、CBRE、Belfer 和 Deloitte 都描述了电力稀缺和交付延误。尽管形态差异化,Armada 仍必须锁定电力、冷却和场地。模块化架构可以缩短部分集成步骤。并网时间、通电场地库存和现场发电计划。
合规面扩张快于公开证明BIS、EU AI Act、CISA、DoD 和 NDAA 来源都指向更多治理和控制义务。即便产品有需求,合规债也会拖住部署。合规最重要的地方,Armada 定位最强。出口管制工作流、ATO 证据和按部署划分的认证。
工厂吞吐量仍未公开Forge One 面积和岗位公开,但单位经济性和吞吐量没有公开。在产出质量和节奏可见前,执行风险仍高。Johnson Controls 部分降低了热管理和制造能力风险。吞吐量、良率、交期和现场服务指标。

反向日志记录的是今天可由公开证据支持的约束;这不是在声称这些风险已经演变成失败。

[CR006, CR027, CR032, CR033, CR038, CR039]
Chapter 08

08估值

8.1 价格已披露;经济分母没有披露

Armada 2026 年 5 月官方公告讲清楚了两件事,也留下了一件不清楚的事。清楚的是:公司融资 $230 million,称本轮超额认购,并把价格标为 $2 billion 投前估值。这意味着约 $2.23 billion 的投后进入点,以及新资金约 10.3% 的投后持股。也清楚的是:管理层用本轮资金讲的是围绕 Galleon Forge One、Leviathan 和 Johnson Controls 框架协议的制造与部署故事,而不是在推一家低资本开支软件公司。 仍不清楚的是收入分母;没有它,外部人无法用常规倍数判断价格便宜、公允还是昂贵。同一组来源给了订单额增长、累计融资和工厂计划,但没有给已确认收入、ARR、毛利率、待交付订单转化,也没有给硬件、部署服务和经常性软件之间的结构。因此,估值首先要按融资语义和里程碑承销来框定,而不是按披露经济性上的干净公开倍数。区分估值、投后测算、累计融资、订单额和收入不是记账琐事;这是本章的核心分析约束。[CV001, CV002, CV003, CV004, CV005, CV006]

建议摘要表
维度当前判断证据基础重要性决策含义
建议观察 / 继续研究战略定位强,但公开经济性不完整仅凭公开证据,无法称这个价格有吸引力没有非公开尽调前,不要按买入承销
确信度许多事实锚点公开,但核心收入和股权结构输入仍是非公开信息公允价值区间仍宽用里程碑闸门,而不是单点确信
风险评级工厂爬坡、电力就绪、营运资本和订单额转化主导下行执行风险可能迅速重估这一轮密切跟踪工厂和转化数据
估值立场偏高隐含投后估值在品类内可以理解,但没有已披露收入和利润率支撑经济性公开前,市场已经在为战略期权价值买单把上行视为有条件,不要默认成立
入场纪律等待证明或更好价格需要收入—订单额桥、利润率分层和优先权条款这些事实会真正改变承销讨论只有完成针对性尽调或出现实质证明后再推进

本表只用公开证据把本章转化为投资立场;它不能替代非公开财务尽调。

[CV004, CV005, CV007, CV037, CV043, CV044]
估值语义与公开已知事项表
事项公开数值 / 状态它是什么它不是什么含义
Series B 规模$230M2026 年 5 月新发行融资不是公司估值只显示轮次规模和隐含的新资金股权占比
投前估值$2.0B新资金进入前的官方定价不是投后估值,也不是收入摊薄测算锚点
隐含投后估值~$2.23B投前估值加新资金不是已披露收入倍数新投资人入场价的最佳代理指标
已披露累计融资近半十亿美元 / $465M累计已募资本不是当前现金余额,也不是盈利能力只显示资本支持,不支撑估值
订单额增长FY25-26 为 540%;Q1 FY27 同比 2000%公司披露的需求信号不是确认收入,也不是毛利率支撑兴趣,但不是经济性证明
收入 / ARR / 毛利率未公开披露关键经济分母仍未公开不能拿订单额或融资标题来推断阻碍精确倍数测算

之所以放这张区分表,是因为在 Armada 2026 年 5 月这一轮里,估值、投后测算、累计融资、订单额和收入是彼此独立的分析对象。

[CV001, CV002, CV003, CV004, CV005, CV006]
FV001: 建议逻辑

这一建议来自强战略验证、仍不透明的经济性,以及高度依赖执行的规模化之间的碰撞。

[CV006, CV007, CV008, CV009, CV035, CV043]

8.2 公开市场给出估值区间,而不是单一干净可比公司

公开可比公司给不出一个整齐答案,但能划出 2026 年 5 月市场愿意为相邻商业模式支付的范围。成熟数字基础设施和混合型公司,从 HPE 约 1.4x 过去十二个月销售额,到 Nutanix 的 4.7x、Pure Storage 的 7.9x,以及 Digital Realty、Equinix 和 Vertiv 约 10.9x 至 11.6x。Nebius 处在更投机的 62.6x 过去十二个月销售额,CoreWeave 即便已经上市并披露收入,仍约 9.24x 交易。这个跨度比任何单一公司更重要,因为 Armada 同时跨越几个类别。 Armada 有实体部署、热管理和电力暴露,使 Vertiv、HPE 和 Digital Realty 具备方向性参考价值。它也有软件和控制平面雄心,使 Pure 与 Nutanix 在上行情景中有用。但市场上没有直接对应的公开孪生公司:一家主权 AI、加固模块化工厂业务,有公开订单额证明,却不披露收入。因此,公开市场给出的正确教训是:当投资者能看见收入质量和结构时,基础设施可以拿到溢价倍数。Armada 要求投资者在这些披露出现之前付钱,这就是为什么本轮价格看起来可以定价,但还谈不上有明确支撑。[CV011, CV012, CV013, CV014, CV015, CV016]

可比估值表
可比对象状态当前市值 / EV 背景倍数 / 基准Armada 参照价值局限
Vertiv上市 / 数字基础设施设备$125.8B 市值11.60x TTM 销售额最接近成熟「硬件 + 服务」数字基础设施参照,覆盖电力与热管理披露质量和规模都远超 Armada
Equinix上市 / 数字基础设施平台$106.5B 市值11.18x TTM 销售额展示利用率和收入质量可见时,经常性数字基础设施能拿到什么定价Armada 目前还没有 REIT 经济性和经常性主机托管合同
Digital Realty上市 / 数据中心 REIT$68.7B 市值10.88x TTM 销售额可参照其资产支撑型资本开支,尤其是场址、电力和部署强度房地产持有模式在结构上不同
Pure Storage上市 / 混合硬件-软件$29.0B 市值7.91x TTM 销售额可参照混合产品叠加经常性软件 / 支持的模式存储业务经济性更清晰,披露也比模块化 AI 工厂充分
Nutanix上市 / 软件主导基础设施$12.7B 市值4.73x TTM 销售额展示没有重硬件资本开支时,软件主导基础设施能如何定价比 Armada 更偏软件
HPE上市 / 多元化基础设施$49.9B 市值1.40x TTM 销售额可作为多元化、低倍数基础设施的下行地板业务过于多元、增长过慢,不能直接可比
Nebius上市 / AI 云离群样本$55.0B 市值62.64x TTM 销售额展示公开市场对 AI 原生资产的热情可以走到多极端离群估值不应锚定基准情景
CoreWeave上市 / GPU 云$57.6B 市值9.24x TTM 销售额最接近且有收入披露的公开 AI 基础设施溢价参照GPU 云经济性和上市公司身份仍与 Armada 不同

倍数是 Stock Analysis 在 2026 年 5 月下旬给出的 TTM 销售额快照;公司描述来自 Macrotrends 和监管文件。表格刻意只放公开公司,因此证据缺口记录缺失的私有公司和分业务参照样本。

[CV011, CV012, CV013, CV014, CV015, CV016]
FV004: 投资 KPI

紧凑的 IC 式评分解释了为什么公司可以有吸引力,但当前公开估值仍缺少干净支撑。

评分只是方向性的投资判断辅助,不是机械投资公式。

[CV006, CV007, CV025, CV035, CV043, CV044]

8.3 私有市场类比支撑品类,但也抬高周期风险

私有市场类比确认,Armada 是在受益于一个活跃资本周期,而不是从零发明品类。Modular 在 2025 年宣布 $1.6 billion 估值,Crusoe 在垂直整合 AI 工厂基础设施规模化时披露了超过 $10 billion 的 Series E。S&P、Reuters 和 Colliers 都指向同一个宏观事实:巨量资本正流入 AI 基础设施、数据中心和与电力绑定的建设。Armada 的绝对金额估值因此在品类内有合理性,尤其是其产品和伙伴策略更接近基础设施,而不是纯企业软件。 同一组来源也给出了反向逻辑。当一个市场正在吸收创纪录的基础设施募资、快速上升的建设成本、大量私募信贷使用,以及关于不透明融资结构的警告时,投资者应担心轮次规模和估值标签会跑在耐久经济性前面。Armada 因而受益于当前私有 AI 基础设施买盘,也暴露于该买盘任何重估下行。公司的估值高于 Modular 披露值、远低于 Crusoe;这让本轮在语境中说得通,但不能让它自我证明。[CV028, CV029, CV030, CV031, CV032, CV033]

投资逻辑 / 反向逻辑表
视角乐观投资逻辑反向逻辑哪些证据会改变判断
品类匹配Armada 正处在真实的主权 AI 与模块化 AI 工厂建设周期中私有市场资本周期火热,可能在可持续经济性出现前推高估值复购部署能产生可持续收入和毛利率的证明
商业模式硬件加控制平面软件,可能值得更高的混合估值待遇公开证据看起来仍更像资本密集型部署,而不是经常性软件可见的搭售率和软件主导的毛利润贡献
私有市场类比对象Armada 高于 Modular、远低于 Crusoe;如果执行兑现能持续累积,仍有上行空间这些类比对象证明品类热度,不证明 Armada 已经拿到类似经济性Armada 正在超越标杆部署、进入规模化的证据
公开可比公司高溢价的公开基础设施公司说明,披露质量够好时,投资者愿意为战略基础设施付费Armada 缺少这些公开可比公司提供的收入、利润率和利用率披露已披露收入,或能缩窄倍数不确定性的非公开尽调
资本周期大量基础设施资本可以继续给工厂和部署供血融资不透明、建设成本通胀和电力瓶颈会压缩整个赛道不只是融资成功,还要看到资本效率和锁定电力的管线证据

估值案例对证据敏感:推高私募轮的同一股品类动能,一旦执行或融资透明度令人失望,也可能反转。

[CV028, CV029, CV030, CV031, CV034, CV035]
FV002: 估值 / 回报区间

方向性的价值区间解释了为什么 Armada 只有在基准情形下大致定价合理,只有在牛市情形下才明显有吸引力。

这些是按里程碑分段的估值区间,不是来自未披露收入模型的输出。

[CV004, CV036, CV037, CV038, CV039, CV040]

8.4 用里程碑分档情景框架,比假造收入模型更诚实

因为 Armada 当前收入不公开,精确 DCF 或当前收入倍数模型都是伪精确。更可防守的替代方案是里程碑分档。基准情景假设:订单额开始在可容忍时间内转为已确认收入;Forge One 达到连续生产,没有可见良率冲击或营运资本冲击;Bridge 或 Atlas 证明自己是真正的经常性附加层,而不只是战略包装。在这种状态下,Armada 有可能长进当前投后标记,甚至小幅超过。 悲观情景更简单,也更危险。如果订单额始终和已确认收入脱节,如果工厂爬坡或场址电力把部署时间向右推,或硬件营运资本比预期更快吃掉新融资,投资者可能迅速把 Armada 重估到多元化或硬件重型基础设施先例的低端。乐观情景需要的不只是更多客户标识:它需要重复的主权或受监管部署、可观察的软件控制平面货币化,以及可信证据证明 Armada 正成为运营层,而不只是箱体供应商。因此,敏感性更多落在执行、附加销售和融资质量上,而不是某个隐藏收入数字。[CV037, CV038, CV039, CV040]

乐观 / 基准 / 悲观情景表
情景当前公允价值区间核心假设必须成立的条件概率信号相对 $2.23B 投后估值
悲观$1.2B-$1.7B订单额转化慢,Forge One 爬坡不均衡,电力或场址就绪拖慢部署,公开市场倍数压缩经常性软件证据出现前,硬件营运资金先主导经济性如果 2026 年工厂节奏或转化开始滑坡,该情景就有分量当前估值标记的 ~0.5x-0.8x
基准$1.8B-$2.5B订单额转化可接受,Forge One 开始连续生产,Bridge 或 Atlas 展现可信的经常性附加收入,但还不足以获得完整软件重估Armada 变成可复制的混合基础设施平台,而不是一次性项目供应商当前最站得住脚的公开证据情景当前估值标记的 ~0.8x-1.1x
乐观$3.0B-$4.5B主权部署可重复,软件控制平面变现清楚,客户更分散,资本市场对 AI 基础设施的胃口持续Armada 更像主权 AI 工厂的规模化运营层,而不是硬件供应商需要尚未公开的私有证据当前估值标记的 ~1.3x-2.0x

区间按里程碑分带,并非来自未披露收入;它们把公开参照样本、资本强度和执行证据转成方向性价值区间。

[CV037, CV038, CV039, CV040, CV043, CV044]
FV003: 估值敏感性

围绕基准情形,最大驱动因素是软件附加、工厂执行,以及全市场基础设施估值下修。

敏感性数值是围绕基准情形中点的方向性估值变化,不是经审计预测的输出。

[CV033, CV034, CV038, CV039, CV040, CV045]

8.5 公开证据支持观察或继续研究,但必须设明确断点触发器

仅看公开证据,正确结论不是买入。按可比公开和私有基础设施公司的体量看,绝对金额估值并不荒谬;但证据质量太薄,不能称价格有吸引力。关键经济和股权结构披露仍缺失,最重要的承销驱动因素现在落在工厂吞吐、订单额转化、软件附加销售和融资纪律上。这一组合足以让 Armada 作为公司保持可投资性,但不足以从本轮融资中移除价格敏感性。 实际含义是,把 2026 年 5 月轮次当作带里程碑门槛的观察或继续研究情形。如果收入转化、利润率分层和 Forge One 执行得到证明,公司可以长进甚至超过当前标记。如果不能,下行路径很直接,因为这个品类已经有低倍数的混合型和硬件重型类比。因此,本章正确的最终输出是:估值立场偏高,置信度中等,执行风险高,尽调计划聚焦少数真正能改变承销判断的私有事实。[CV041, CV042, CV043, CV044, CV045]

投资逻辑破裂与叫停触发器表
触发因素阈值如何冲击投资逻辑监测节奏行动含义
订单额无法转化多个季度后,订单额与确认收入之间仍没有可信桥梁增长叙事转成订单质量风险季度 / 尽调更新转为回避,或要求更低估值
工厂爬坡滑坡Forge One 错过连续生产时间点,或出现明显良率 / 供应链问题削弱规模化故事,并消耗资本月度运营复盘暂停投资测算,直到吞吐量证据恢复
电力或场址就绪停滞客户出现重大延误,原因指向并网、公用事业押金或建设瓶颈固定成本堆起来,部署时间被推后按场址跟踪管线将价值区间下调至接近悲观情景
软件附加仍弱Bridge / Atlas 仍只是战略包装,而非已变现的经常性收入层Armada 仍落在硬件主导估值桶里合同与产品组合复盘不要按溢价软件倍数做投资测算
融资条款不透明优先权堆叠或结构让新资金承担不利的下行不对称即使业务跑成,也会压低股权上行空间条款清单审阅重新定价或退出

在完整运营历史公开之前,这些是 2026 年 5 月估值逻辑最容易被观察到的破裂方式。

[CV039, CV040, CV041, CV042, CV045]
最终尽调索取清单表
主题缺失证据重要性负责人 / 尽调路径决策影响
收入—订单额桥梁按季度确认收入,对照订单额和积压订单判断需求是否变成真实经济性财务资料室 + 董事会材料最高
按层拆分毛利率硬件、部署、支持、Bridge 和 Atlas 的毛利率结构决定应落入哪个公开可比公司估值桶财务资料室 + 产品 P&L最高
营运资金和资本开支库存、应收款、客户定金、工厂工装和公用事业押金检验 Series B 是在资助规模化,还是只填到下一轮融资前的缺口财务 + 运营最高
按金额看客户集中度头部客户、头部行业,以及续约 / 扩张依赖判断收入韧性和下行集中度销售运营 + 财务
工厂吞吐量Leviathan 产线速率、供应商就绪度、QA 指标和现场服务闭环验证 Forge One 是催化剂还是新瓶颈运营尽调
Series B 优先权条款清算优先权、参与权、反稀释,以及任何结构化附函改变新资金的实际回报分布法务 + 融资文件最高

这些私有事实最可能把本轮从偏高推向合理,或从观察推向回避。

[CV041, CV042, CV045]

免责声明

本报告基于公开来源,不构成投资建议。Armada 是私营公司;融资和估值数字来自已披露轮次条款;截至报告生成日,收入、利润率和运营指标大多仍未披露。

证据索引

结论
编号陈述可信度来源
CO001 Armada was founded in late 2022 by Dan Wright and Jon Runyan. SO019, SO025, SO026
CO002 Armada emerged from stealth in December 2023 with more than $55 million of early funding disclosed. SO025, SO027
CO003 Armada is headquartered in San Francisco. SO018, SO019, SO026
CO004 As of May 2026 Armada remained a private company whose latest disclosed financing was a Series B round. SO009, SO018, SO024
CO005 Armada describes itself as a full-stack edge infrastructure company focused on bridging the digital divide. SO001, SO002, SO025
CO006 Armada's current platform lineup includes Atlas, Galleon, Marketplace, and Bridge. SO002, SO003, SO004, SO005
CO007 Galleon is Armada's ruggedized modular data-center family for remote and harsh-environment deployments. SO003, SO009, SO018
CO008 Atlas is Armada's monitoring and management layer for connected assets such as satellite terminals and edge systems. SO004, SO008, SO014
CO009 Marketplace is Armada's hub for deploying third-party hardware, software, and applications at the edge. SO002, SO010, SO015
CO010 Bridge is Armada's software for GPU orchestration and GPU-as-a-Service on customer-controlled infrastructure. SO005, SO028
CO011 Dan Wright is Armada's co-founder and CEO and previously led DataRobot and served as COO at AppDynamics. SO025, SO027
CO012 Jon Runyan is Armada's co-founder and COO and previously served as Okta's general counsel through its IPO. SO015, SO027
CO013 Pradeep Nair is Armada's founding CTO and previously held engineering leadership roles at VMware and Microsoft Azure. SO010, SO017, SO027
CO014 Armada's public materials and media coverage identify Prag Mishra as Chief AI Officer. SO015, SO027
CO015 Public-source visibility around Armada's board composition and control rights is limited compared with its disclosed executive bench. SO002, SO015, SO019
CO016 Armada's public leadership presentation creates material key-person dependency on Dan Wright for narrative, fundraising, and strategic partnerships. SO009, SO018, SO025
CO017 A July 2024 strategic round led by M12 raised $40 million and pushed Armada's disclosed funding above $100 million at the time. SO017, SO020
CO018 The 2024 M12 round also made Armada products purchasable through Azure Marketplace using pre-committed Azure spend. SO017, SO023
CO019 Armada's July 2025 strategic round raised $131 million and introduced Leviathan. SO010, SO020
CO020 Leviathan is a megawatt-scale, liquid-cooled modular data center in the Galleon family with roughly ten times the compute of Armada's next-largest form factor. SO010, SO020
CO021 Armada's May 2026 Series B raised $230 million at a $2 billion valuation. SO009, SO018, SO024
CO022 The Series B was co-led by Overmatch, BlackRock, and 8090 Industries, with Johnson Controls joining as a new strategic investor. SO009, SO018, SO028
CO023 CNBC's 2026 Disruptor 50 profile lists Armada's total funding at $465 million, consistent with official sources describing it as nearly half a billion dollars after the Series B. SO019, SO009, SO024
CO024 Armada and Johnson Controls signed a Global Framework Agreement tied to Galleon Forge One, a planned Arizona factory of up to 400,000 square feet and roughly 500 jobs. SO009, SO022, SO018
CO025 Armada reported 540% bookings growth from FY25 to FY26 and approximately 2,000% year-over-year bookings growth in Q1 FY27. SO009, SO018
CO026 Armada sells into defense, energy, industrial, and public-sector use cases rather than a general enterprise cloud market. SO001, SO018, SO025
CO027 Armada's named target industries include defense, oil and gas, manufacturing, mining, telecommunications, and state or local government. SO002, SO018, SO029
CO028 Alaska DOT&PF used Atlas and Galleon to move drone-imagery processing from roughly 28-hour-to-multi-day delays to same-day or real-time outputs. SO008, SO021
CO029 Washington DNR uses Atlas to centrally manage approximately 45 Starlinks supporting wildfire and remote-operations response. SO014
CO030 Armada's Galleon and Atlas were used during UNITAS 2025 from ashore and aboard a Navy warship to support multinational maritime operations. SO013, SO018
CO031 Aker BP agreed in March 2026 to deploy an Armada Galleon on the Norwegian Continental Shelf, with an initial reference installation intended to become a repeatable fleet blueprint. SO012, SO009
CO032 Aker BP's offshore deployment is aimed at local data processing, model execution, lower downtime, and more remote or autonomous operations. SO012
CO033 Armada and Microsoft launched a March 2026 sovereign-AI solution that combines Azure Local with Galleon and AEP. SO011, SO007, SO023
CO034 Armada says the Microsoft collaboration is already available and targeted at defense, government, and regulated-industry deployments. SO011, SO007
CO035 Armada had customer deployments or distribution across 43 countries by mid-2024. SO017, SO025
CO036 Investor and partner materials name customers or deployments including the U.S. Navy, Alaska DOT&PF, Aker BP, Tampnet, and other industrial operators. SO009, SO021, SO026
CO037 Dragon Global says Armada's customer base includes Targa Resources, Atlas Energy, SQM, Mars, Marriott, Vocus, Tampnet, the U.S. Navy, and Alaska's Department of Transportation. SO026
CO038 Carahsoft markets Armada's stack for emergency response, military missions, critical infrastructure monitoring, and secure citizen-data use cases in the public sector. SO029
CO039 Capital intensity is a material diligence risk because Armada is funding physical modular-data-center manufacturing and factory scale-up before public profitability metrics are disclosed. SO018, SO027
CO040 Forbes reported that in late 2023 Armada had no customers beyond a proof-of-concept, zero revenue, and a capital-intensive commercialization path. SO027
CO041 Forbes and CNBC both continue to reference Dan Wright's exit from DataRobot as a reputational diligence flag in Armada's founder narrative. SO019, SO027
CO042 The fetched public sources do not disclose Armada's current revenue, ARR, exact customer count, headcount, or board composition as of the 2026 run date. SO002, SO018, SO019
CO043 Armada's public partner ecosystem includes Microsoft, NVIDIA, Palantir, Dell Technologies, and public-sector distribution channels such as Carahsoft. SO006, SO009, SO029
CO044 Armada's product and partner stack is intended to keep data, models, and governance inside sovereign or disconnected operating environments. SO011, SO023, SO029
CO045 The fetched public source set surfaced no Armada-specific lawsuit or regulatory action, but the absence of docket-level review prevents a definitive clean bill of health. SO018, SO019, SO027
CM001 Armada's homepage presents Atlas, Galleon, Bridge, and Marketplace as one connected edge platform. SM001
CM002 Armada's included spend is rugged modular AI-ready infrastructure plus orchestration and connectivity control, not generic hyperscale or ordinary public cloud. SM001, SM002, SM019
CM003 Armada says Galleon is operational in weeks rather than years. SM001, SM002
CM004 Armada's homepage compares Galleon deployment at roughly 60 days versus about 24 months for traditional data centers. SM001
CM005 Armada's Galleon page says modular units can be deployed in days rather than months. SM002, SM013
CM006 Armada's Galleon page says some configurations can operate fully air-gapped with no exposure to outside networks. SM002, SM003
CM007 Armada and Microsoft both frame the joint offer around local control over data, models, and operations. SM003, SM004
CM008 Microsoft says the primary target sectors include defense, public safety, energy, and critical infrastructure. SM004, SM005
CM009 Microsoft and Armada say the Azure Local plus Galleon solution is built for intermittently connected, contested, or fully disconnected environments. SM003, SM004, SM005
CM010 Johnson Controls and Armada plan a dedicated Arizona factory of up to 400,000 square feet and 500 jobs. SM006
CM011 Johnson Controls says Leviathan is a megawatt-scale modular data center built for high-density AI training and inference workloads. SM006
CM012 Aker BP says offshore drilling operations need local processing because connectivity to shore and cloud infrastructure is not always guaranteed. SM007, SM008
CM013 Aker BP's offshore deployment is intended to reduce downtime and preserve continuity during connectivity disruptions. SM007, SM008
CM014 Armada says Alaska DOT&PF moved from roughly 28 hours to four hours and in some workflows to real-time intelligence using Atlas and Galleon. SM009, SM010
CM015 Armada explicitly markets to oil and gas, defense, state and local government, manufacturing, mining, and telecommunications. SM001, SM009
CM016 Carahsoft positions Galleon for federal, state, local, education, and healthcare buyers that need self-sufficient AI compute where the cloud cannot reach. SM012, SM013
CM017 Armada's NVIDIA AI Grid pitch spans service-provider data centers, AI factories, regional hubs, and edge locations. SM014, SM015
CM018 Mitsui says industrial customers use local AI to support remote operations, predictive maintenance, autonomy, and continuity at the point of data generation. SM016
CM019 Nscale and Armada target sovereign AI deployments at both hyperscale and edge for public-sector and enterprise customers. SM017
CM020 IDC says full-year 2025 AI infrastructure spending reached $318 billion. SM018
CM021 IDC projects AI infrastructure spending of about $487 billion in 2026. SM018
CM022 IDC projects the AI infrastructure market will exceed $1 trillion by 2029. SM018
CM023 JLL says roughly 100 GW of new data-center capacity will be added from 2026 to 2030 at a 14% CAGR. SM019
CM024 JLL says inference could overtake training in 2027. SM019
CM025 JLL says inference demand requires geographic distribution and embedded systems at the edge. SM019
CM026 JLL says the average wait time for grid connection in primary data-center markets exceeds four years. SM019
CM027 JLL forecasts average 2026 shell-and-core construction cost of $11.3 million per MW. SM019
CM028 JLL says AI fit-out can cost as much as $25 million per MW. SM019
CM029 Vertiv says AI and high-performance compute demand is structurally transforming power, thermal, and service requirements. SM020
CM030 Vertiv says prefabricated, OCP-aligned racks, power, and cooling are intended to accelerate high-density AI deployments. SM021
CM031 Deloitte says Europe's sovereignty drive could mobilize over €100 billion of public and private investment over five years. SM023
CM032 Deloitte says European programs are explicitly funding AI factories, gigafactories, and sovereign-cloud adaptations. SM023
CM033 Brookings says federal AI spending is rising quickly and moving toward multiyear contracts concentrated in the Department of Defense. SM024
CM034 CDO says the Pentagon requested $13.4 billion for AI and autonomy in FY2026. SM025, SM035
CM035 NDU says distributed military AI adoption still faces certification, assurance, and human-machine teaming constraints. SM034
CM036 Ericsson and NTT DATA say private 5G plus edge AI is production-targeted across manufacturing, mining, ports, airports, energy, transportation, and smart cities. SM026, SM028
CM037 Ericsson says private 5G is built for industrial use, keeps sensitive data on site, and integrates with IT and OT systems. SM027
CM038 Uptime says the sector faces rising costs and worsening power constraints. SM029
CM039 Uptime says staffing shortages and cautious early-stage AI adoption remain material. SM029
CM040 Uptime's 2026 predictions say AI infrastructure is concentrating among hyperscalers and other well-capitalized enterprises. SM030
CM041 Uptime's 2026 predictions say power availability and long deployment timelines will remain bottlenecks. SM030
CM042 Schneider Electric executives say AI data centers now need deeper grid interaction, on-site power, storage, liquid cooling, and higher-voltage architectures. SM031
CM043 Future Market Insights estimates the modular data-center market at $29.3 billion in 2026 and $106.7 billion in 2036. SM032
CM044 Research and Markets estimates the modular data-center market at $47.75 billion in 2026 and $104.98 billion in 2030. SM033
CM045 The accessible public 2026 modular-data-center estimates differ by more than $18 billion. SM032, SM033
CM046 Armada's actual opportunity is narrower than total modular or total AI infrastructure because it depends on rugged, disconnected, or sovereignty-sensitive deployments in selected verticals. SM001, SM002, SM004, SM019
CM047 A reasonable analytical 2026 SAM for Armada is about $4-8 billion, triangulating 13-17% of the published modular-market range and roughly 1-1.6% of IDC's 2026 AI-infrastructure spend. SM018, SM032, SM033
CM048 A directional three-year SOM of about $0.2-0.6 billion is plausible only if Armada converts pilot-like proofs into repeatable multi-site programs through manufacturing and channel leverage. SM006, SM012, SM013, SM017
CM049 Budget ownership for Armada deployments is fragmented across operations, OT, security, IT infrastructure, and digital transformation rather than one universal line item. SM004, SM016, SM026, SM027
CM050 Adoption usually starts with one high-urgency remote workflow and expands only after local-compute ROI is proven. SM007, SM010, SM012, SM013
CM051 The main growth drivers are edge inference, sovereignty pressure, remote-site ROI, and deployment speed, while the main constraints are power, capex, integration, and concentration. SM003, SM019, SM029, SM030, SM031
CP001 Armada describes Galleon as a modular, containerized, ruggedized data-center product for harsh environments including offshore energy, defense missions, and remote mining sites. SP001
CP002 Armada says Galleon arrives preloaded with compute, networking, storage, heating, and cooling and can move from delivery to full operation in days or weeks rather than months or years. SP001, SP005
CP003 Armada publicly shows a product range from suitcase-sized Beacon and 20-foot Cruiser to 40-foot Triton and megawatt-scale Leviathan. SP001
CP004 Armada positions Bridge as a software layer that provides GPU orchestration, scaling, billing, observability, hard multi-tenant isolation, and GPU-as-a-service on customer-owned or Armada-owned infrastructure. SP002
CP005 Armada says its partner ecosystem includes 20-plus pre-integrated partners and is meant to take a deployment from Galleon to a fully operational edge-AI stack in weeks rather than years. SP003
CP006 Armada and Microsoft jointly position Azure Local on Galleon plus AEP as a sovereign private-cloud and AI stack for disconnected, contested, or regulated environments, with joint go-to-market activity. SP004
CP007 Armada and Carahsoft say the Galleon Experience Center gives federal, state, local, education, and healthcare buyers a procurement and demonstration path through Carahsoft contract vehicles and reseller partners. SP005
CP008 Armada says AEP can operate across existing service-provider data centers, centralized AI factories, regional hubs, and edge locations, with Galleon added when new infrastructure is required. SP006
CP009 Armada says AEP provides a unified control plane, workload-aware orchestration, centralized monitoring, and secure multi-tenant platform services for distributed AI Grid deployments. SP006
CP010 AWS Outposts extends select AWS services such as EC2, EKS, ECS, EBS, S3, RDS, and IoT Greengrass into on-premises and colocation environments while connecting back to the AWS Region. SP007, SP008
CP011 AWS says Outposts racks are delivered fully assembled and installed by AWS and support defined rack, networking, and power configurations for latency-sensitive on-prem workloads. SP007, SP008
CP012 AWS Outposts rack pricing is structured around a three-year term with all-upfront, partial-upfront, or no-upfront payment options and requires AWS Enterprise Support. SP009
CP013 Azure Local is an Azure Arc-enabled distributed infrastructure product for virtual machines, containers, and selected Azure services on customer-owned infrastructure. SP010, SP011
CP014 Microsoft prices Azure Local on a per-physical-core per-month basis, offers a 60-day free trial, and includes AKS enabled by Azure Arc at no extra charge on recent releases. SP011
CP015 Microsoft says Azure Local can be bought on validated partner hardware or installed on eligible hardware, and fully disconnected operation with a locally hosted control plane is available through account-representative engagement. SP011
CP016 Google Distributed Cloud is a fully managed software and hardware solution for data centers and edge locations designed for regulatory, local-data-processing, survivability, and low-latency needs. SP012, SP013
CP017 Google says Gemini is available on GDC on-prem and that GDC can scale from one to thousands of locations with a Kubernetes-based workflow and partner ecosystem. SP012
CP018 Google publishes connected GDC pricing starting at $35 per vCPU per month with a 96-vCPU minimum per site, while air-gapped deployments require a sales quote. SP012
CP019 HPE Private Cloud AI is positioned as a turnkey, pre-configured, validated private AI platform co-engineered with NVIDIA and delivered through an HPE GreenLake cloud experience. SP014, SP015
CP020 HPE explicitly frames the on-prem private-AI buying choice as build-your-own versus reference architecture plus services versus turnkey. SP014
CP021 HPE and NVIDIA say their portfolio is sold through joint sales teams, channel partners, and global system integrators including Deloitte, HCLTech, Infosys, TCS, and Wipro. SP015
CP022 Dell AI Factory with NVIDIA is positioned as an end-to-end enterprise AI solution spanning desktop to data center to edge and cloud, with a modular architecture for scaling from pilot to production. SP016, SP017
CP023 Dell says more than 4,000 customers are deploying the Dell AI Factory with NVIDIA and that early adopters have seen up to 2.6x ROI in the first year. SP017
CP024 Dell's public materials combine liquid-cooled servers, rack-level power and cooling management, automation blueprints, professional services, and pay-as-you-go consumption options. SP016, SP017
CP025 NVIDIA's Dell manufacturing profile says one Dell U.S. factory can ship thousands of Blackwell GPUs in a week and supported a 100,000-GPU deployment in six weeks for a large customer. SP018
CP026 Nscale presents a full-stack AI platform that spans inference endpoints, fine-tuning, a prompt-engineering workbench, bare-metal nodes or VMs, managed Kubernetes or Slurm, and automated fleet operations. SP029, SP030
CP027 Nscale's infrastructure pitch is modular, chip-agnostic, sovereign, and multi-megawatt, aimed at enterprises, governments, and mission-critical workloads. SP030
CP028 Nscale's public buildout evidence centers on hub-scale campuses such as 30MW Glomfjord expandable to 60MW, 230MW Narvik with further expansion, and a roughly 240MW Texas site rather than forward-deployed field units. SP030
CP029 Lambda positions itself as managed AI infrastructure from one GPU to hundreds of thousands, with single-tenant shared-nothing architecture and production-grade compliance certifications. SP026
CP030 Lambda publishes transparent pricing including 1-Click Clusters from 16 to 256-plus GPUs and GPU prices such as $6.69 per B200 GPU hour, $3.99 per H100 GPU hour, and $2.79 per A100 GPU hour. SP027
CP031 Lambda's managed Kubernetes for Private Cloud keeps Kubernetes and fleet-management components local, exposes nothing to the internet by default, uses secure VPN access, and supports single-tenant private clusters. SP028
CP032 Crusoe Cloud offers managed Kubernetes, managed Slurm, managed inference, VPC isolation, observability, and a 99.98% uptime claim for AI workloads. SP031
CP033 Crusoe Spark is described as a turnkey prefabricated modular AI factory for low-latency edge, on-premise deployments, sovereign AI, and grouped training clusters. SP032
CP034 Crusoe says Spark modules can be deployed in as little as three months and grouped from hundreds of kilowatts to tens or hundreds of megawatts. SP032
CP035 Vertiv says its prefabricated modular solutions provide over 40% time savings versus conventional builds and include portable SmartMod and multi-megawatt AI-ready modular offerings. SP019
CP036 Vertiv says OneCore can reduce time-to-token by up to 50%, reduce space by up to 30%, lower TCO by up to 25%, and support densities up to 600 kW per rack. SP020
CP037 Eaton and Flexnode offer turnkey prefabricated AI-factory infrastructure for 3.5MW to 35MW data halls with 800 VDC power architecture and rapid modular deployment. SP021
CP038 Rittal says its OCP and NVIDIA-aligned infrastructure includes more than one megawatt of water-based cooling in compact space and compatibility with 415/480 VAC, ±400 VDC, and 800 VDC environments. SP022
CP039 Schneider describes modular data centers as portable, scalable, pre-engineered, pre-tested, and quickly deployable across varied locations and environmental conditions. SP023
CP040 Schneider's AI pod launch supports one-megawatt-plus prefabricated pods with liquid cooling, power busway, MGX-aligned racks, and pre-assembled rapid deployment. SP024
CP041 Carahsoft says NVIDIA public-sector solutions support on-prem, edge, and hybrid deployments and are delivered through a broad ecosystem of partners, resellers, and systems integrators. SP025
CP042 The lock-in sources argue that compute can be portable, but switching cost grows in proprietary data services, application integrations, IAM, infrastructure-as-code, training, and organizational workflows. SP033, SP034
CP043 Serious Insights says about half of planned U.S. data-center builds in 2026 are projected to be delayed or canceled due to power constraints, making power availability a real cap on AI-infrastructure rollout speed. SP035
CP044 AWS's EC2 G5 page shows that centralized public-cloud GPU instances remain a status-quo substitute for buyers who want accelerated compute without deploying local modular infrastructure. SP036
CP045 Across Galleon, Bridge, Azure Local collaboration, and AI Grid materials, Armada's own evidence points to a differentiation wedge that combines rugged deployment with orchestration, marketplace distribution, and operation across existing and new sites. SP002, SP004, SP006
CP046 The retained evidence supports a thin direct-peer set in which Crusoe Spark and Nscale are the closest public modular sovereign-AI peers, while Lambda is better treated as a private-cluster substitute and physical vendors as hardware-layer pressure. SP026, SP029, SP030, SP032, SP019, SP023
CP047 Incumbents and large OEMs hold stronger installed-base and channel positions than Armada because AWS, Microsoft, Google, HPE, and Dell bundle software, hardware, services, and established account coverage, while Carahsoft and NVIDIA help shape who reaches government and enterprise buyers fastest. SP011, SP015, SP016, SP017, SP025
CP048 Because many vendors now offer AI-ready modular power, cooling, and rack systems, Armada's most durable moat increasingly depends on software orchestration, channel access, and execution in rugged disconnected deployments rather than on modular hardware alone. SP002, SP006, SP019, SP020, SP021, SP022, SP023, SP024
CP049 Armada's public proof set highlights named deployments and channel wins, but the retained sources do not disclose renewal rates, win-loss data, or conversion from showcase sites into standardized fleet rollouts. SP004, SP005, SP006
CP050 The retained channel evidence shows that partner leverage is real for Armada, but it does not quantify what share of demand, bookings, or renewals Armada sources directly versus through Microsoft, Carahsoft, NVIDIA, or other intermediaries. SP005, SP015, SP025
CI001 Armada's public product stack includes Galleon hardware, Bridge GPU orchestration software, Atlas management software, and a Marketplace for partner hardware and software. SI002, SI005, SI006
CI002 Armada markets Bridge as software to manage, scale, and monetize GPU clusters across data center, cloud, and edge environments. SI002, SI004
CI003 Armada says Bridge pricing is based on active GPU usage and structured as GPU/year or GPU/hour. SI003
CI004 Bridge can be deployed on existing customer infrastructure or paired with Galleon, so its software layer does not require a new Armada hardware purchase in every deployment. SI002, SI004
CI005 Bridge and Marketplace are explicitly framed as ways for operators to launch GPU-as-a-Service or Model-as-a-Service offerings and create new revenue streams. SI002, SI004
CI006 The Galleon family spans from smaller field units to the megawatt-scale Leviathan, implying a broad hardware ASP and deployment-cost ladder rather than a single appliance price point. SI005, SI010
CI007 Public deployment messaging consistently describes Galleon as preconfigured infrastructure that can go from delivery to operation in days or weeks, implying hardware-heavy revenue recognition closer to delivery and acceptance than to long-term software usage. SI005, SI008
CI008 Armada's 2024 Business Wire release says all Armada products were available in Azure Marketplace and could be purchased using pre-committed Azure spend. SI007
CI009 Armada says Armada Edge Platform is available through named Carahsoft contract vehicles including SEWP V, ITES-SW2, NASPO, TIPS, OMNIA, and Quilt. SI008
CI010 The clearest public traction metric is bookings growth rather than revenue or ARR. SI001, SI011
CI011 Armada said bookings grew 540% from FY25 to FY26. SI001, SI011
CI012 Armada said Q1 FY27 bookings grew 2,000% year over year. SI001, SI011
CI013 CNBC's 2026 Disruptor 50 profile reported total disclosed funding of $465 million as of May 2026. SI012
CI014 Armada's May 2026 Series B raised $230 million at a $2 billion valuation. SI001, SI011, SI013
CI015 Armada's July 2025 strategic round raised $131 million and coincided with the Leviathan launch. SI010, SI016
CI016 Armada's July 2024 round raised $40 million and brought total funding to over $100 million at that time. SI007
CI017 Armada and Johnson Controls disclosed Galleon Forge One in Arizona at up to 400,000 square feet and about 500 jobs. SI001, SI009
CI018 Johnson Controls is both a strategic investor in Armada and the manufacturing counterparty under a Global Framework Agreement for modular data center systems. SI001, SI009
CI019 Johnson Controls said continuous production at Forge One is planned to begin with Leviathan, anchoring manufacturing risk in a megawatt-scale product rather than only in small edge nodes. SI009, SI010
CI020 No public source reviewed disclosed Armada revenue, ARR, gross margin, EBITDA, cash balance, monthly burn, runway, customer concentration, backlog, deferred revenue, or realized pricing. SI001, SI010, SI011, SI021, SI023
CI021 The public record does not disclose the actual revenue split among Galleon hardware, Bridge, Atlas, Marketplace, deployment work, and support. SI001, SI002, SI005, SI006
CI022 Because Armada publicly emphasizes bookings rather than recognized revenue, the disclosed growth figures cannot be translated into revenue without contract mix, delivery/acceptance terms, and recognition policy. SI019, SI021, SI023
CI023 Atlas is a real product with pooled data-plan and Azure-integration language, but Armada does not publicly disclose Atlas pricing or revenue contribution. SI006
CI024 Marketplace purchase and deployment flows are public, but Armada does not publicly disclose any take rate, referral fee, or GMV tied to Marketplace activity. SI002, SI003
CI025 Vertiv's FY2024 results show about $6.394 billion of product sales and $1.618 billion of services sales on roughly $8011.8 billion of total sales, putting services at about 20.2% of revenue. SI017
CI026 Using Vertiv's FY2024 results table, about $2934.2 billion of gross profit on $8011.8 billion of sales implies blended gross margin of about 36.6% for a hardware-heavy digital infrastructure vendor. SI017
CI027 Vertiv guided to roughly $275 million of 2025 capital expenditures, or about 3% of sales, illustrating that a scaled assembly-and-services model can become relatively asset-light after buildout. SI017
CI028 Pure Storage's FY2025 results show $1.699 billion of product revenue and $1.469 billion of subscription-services revenue, so subscription services already represent nearly half of revenue in a mature hybrid model. SI019
CI029 Pure Storage's FY2025 results imply about 66.1% product gross margin, 74.1% subscription-services gross margin, and 69.8% blended gross margin. SI019
CI030 Nutanix reported FY2024 ARR of $1.91 billion, Q4 FY2024 ACV billings of $338 million, Q4 FY2024 revenue of $548 million, and Q4 FY2024 GAAP gross margin of 85.2%. SI021
CI031 Nutanix defines ARR from subscription contracts irrespective of the periods in which it recognizes revenue, making it a clear public example of how contracted value and GAAP timing can diverge. SI021
CI032 Equinix reported FY2024 revenue of $8.748 billion and adjusted EBITDA margin of 47%. SI023
CI033 Equinix said two-thirds of recurring revenues come from customers deployed in more than 10 IBX data centers, and interconnection represented 19% of recurring revenue. SI023
CI034 Equinix guided to $3.222-$3.472 billion of 2025 total capex, including $2.985-$3.215 billion of non-recurring capex and $237-$257 million of recurring capex. SI023
CI035 Equinix explains that installation fees are generally paid in a lump sum but recognized ratably over contract term, showing how billed cash and recognized revenue can diverge in infrastructure businesses. SI023
CI036 Current SEC EDGAR search pages confirm the latest 10-K cycles for Vertiv, Pure Storage, Nutanix, Equinix, and Eaton as of 2026, grounding the benchmark set in current filing context. SI018, SI020, SI022, SI024, SI025
CI037 Data Center Knowledge and Moody's warn that accelerating AI data-center investment brings significant credit, overbuild, obsolescence, and capex-renewal risks, particularly for turnkey assets. SI026, SI027
CI038 The public evidence supports treating Armada as a hardware-plus-software/services company, but not as a pure SaaS business, so software-style metrics should not be forced without evidence. SI002, SI005, SI006
CI039 Factory construction, modular deployment, and megawatt-scale Leviathan imply real manufacturing, inventory, deployment, and receivables needs even though the exact cash profile is not public. SI005, SI009, SI010
CI040 Although $465 million of disclosed funding is substantial, the public record is not detailed enough to prove it fully covers factory capex, GPU procurement, working capital, and operating burn. SI012, SI017, SI023, SI027
CI041 The highest-priority financial diligence asks are absolute bookings and revenue, revenue mix, gross margin by layer, burn and runway, factory capex responsibility, GPU procurement commitments, and customer concentration/backlog. SI001, SI009, SI023, SI027
CI042 No public list price for Galleon, Leviathan, or deployment services appears in the reviewed sources. SI005, SI008
CE001 Armada Edge Platform is publicly described as four products: Atlas, Galleon, Bridge, and Marketplace. SE001, SE007
CE002 Armada's homepage compares Galleon at 60 days with traditional data centers at 24 months and frames the platform as operational in weeks, not years. SE001
CE003 Galleon is publicly positioned as a portable, modular, ruggedized, containerized edge data-center family for harsh environments. SE002, SE007
CE004 The reviewed public sources show the Galleon family running from Beacon through Cruiser and Triton to Leviathan. SE002, SE024
CE005 Armada says Beacon is a suitcase-sized Galleon for remote sites with limited space and connectivity. SE002
CE006 Armada says Cruiser is a 20-foot Galleon with three racks of compute. SE002
CE007 Armada says Triton is a 40-foot Galleon with five racks of compute. SE002
CE008 The reviewed public product pages do not publish per-SKU power ratings for Beacon, Cruiser, or Triton. SE001, SE002, SE003
CE009 Galleon is marketed as turnkey infrastructure preloaded with compute, networking, storage, heating, and cooling. SE002
CE010 Armada says local processing on Galleon reduces latency and bandwidth by sending only mission-critical information back to the cloud via Starlink. SE002
CE011 Armada says Cruiser and Triton can be configured to operate fully air-gapped. SE002
CE012 Leviathan is publicly described as a liquid-cooled, megawatt-scale member of the Galleon family. SE002, SE003, SE009, SE024
CE013 Armada says Leviathan has ten times the compute capacity of Triton or Armada's next-largest form factor. SE009, SE024
CE014 Armada says Leviathan can be colocated with stranded natural gas, solar, nuclear, or other alternative energy sources. SE009, SE024
CE015 Armada says Leviathan can be operational in weeks and relocated as customer requirements evolve. SE009
CE016 Atlas is publicly positioned as the operational interface for Starlink terminals, SD-WAN, drones, cameras, sensors, and other connected assets. SE001, SE004, SE007
CE017 Atlas publicly offers pooled data plans, predictive monitoring, and twelve months of usage history. SE004
CE018 Atlas publicly names SSO and RBAC as control features. SE004
CE019 Atlas publicly names audit logs as part of its control surface. SE004
CE020 Atlas publicly says the platform is SOC 2 and ISO 27001 certified. SE004
CE021 Bridge is publicly marketed as on-prem software that turns GPU clusters into managed GPU-as-a-Service with hard isolation, elastic allocation, monetization, and unified billing plus observability. SE005, SE011
CE022 Bridge documentation describes the product as combined IaaS and PaaS for enterprise AI clouds. SE011
CE023 Bridge publicly supports multi-tenant operation across Kubernetes, SLURM, and Jupyter-oriented workflows. SE011, SE014
CE024 Bridge's Kubernetes layer publicly supports bare-metal or VM compute, multiple Kubernetes distributions, and autoscaling based on GPU utilization. SE012
CE025 Bridge documentation says third-party schedulers such as SLURM and Run:AI can sit behind a common interface. SE012
CE026 Bridge cluster templates publicly include basic Kubernetes, Ray, JupyterHub with KAI Scheduler, and NVIDIA NIM. SE013
CE027 Bridge publicly lets tenants configure MIG profiles from the UI on supported NVIDIA GPUs. SE015
CE028 Bridge documentation says operators can monitor GPU metrics including temperature and power consumption. SE015
CE029 Marketplace publicly supports first-party OpsAI apps, partner software, and bring-your-own containerized applications. SE006
CE030 Marketplace publicly names partner applications from Aveva, Metaspectral Fusion, Halliburton, and Avathon. SE006
CE031 Armada's March 2026 collaboration with Microsoft combines Azure Local, Galleon modular data centers, and AEP for sovereign private-cloud deployments. SE008, SE016, SE020
CE032 Microsoft says the Azure Local on Galleon reference architecture supports managed clusters with multi-rack scalability. SE016
CE033 Microsoft says the Azure Local on Galleon reference architecture supports hyperconverged and SAN-backed storage. SE016
CE034 Armada and Microsoft both describe the edge connectivity stack as spanning satellite, LTE/5G, RF, and SD-WAN with support for disconnected operation. SE008, SE016
CE035 Microsoft says Foundry Local and Azure Local can run inference and analytics locally even when disconnected from the public cloud. SE016, SE008
CE036 AEP is publicly described as the unified control layer for orchestration, monitoring, and operational insight across distributed edge environments. SE008, SE019
CE037 Armada's NVIDIA AI Grid release says AEP can operate across existing service-provider data centers, centralized AI factories, regional hubs, and edge locations. SE019
CE038 Armada's NVIDIA AI Grid release says AI Grid sites can expose managed Kubernetes, managed SLURM, Jupyter notebooks, and ML workflows. SE019
CE039 Carahsoft says Armada's public-sector stack supports Azure Local on Galleons and Palantir Foundry plus AIP powered by Dell and NVIDIA without new data-center construction. SE018, SE025
CE040 Galleon Forge One is planned as up to 400,000 square feet in Arizona with roughly 500 jobs and continuous production beginning with Leviathan. SE010, SE017
CE041 Johnson Controls says the partnership contributes advanced thermal management, mission-critical building systems, and more than 40,000 field personnel to Armada's rollout. SE010, SE017
CE042 Independent and official sources both tie Forge One to repeatable Leviathan manufacturing rather than one-off field builds. SE010, SE017
CE043 The Garuda job post shows Armada is staffing for on-prem CaaS and GPUaaS on bare-metal Kubernetes using KVM, container runtimes, KubeVirt, vGPU, and a cloud-integrated marketplace. SE023
CE044 Reviewed public materials describe ruggedized rapid deployment and sovereign control but do not publish formal uptime SLAs or site-prerequisite matrices by Galleon SKU. SE002, SE003, SE008
CE045 TechCrunch reports that power conversion inside AI data centers currently wastes about 15% to 20% of energy. SE022
CE046 TechCrunch reports that power rather than compute is becoming the limiting factor in scaling AI data centers. SE022
CE047 TechCrunch reports that AI-linked gas power projects face turbine shortages, order queues into 2028, and delivery times of about six years. SE021
CE048 DCD independently reports that Leviathan uses liquid cooling and can be deployed in weeks on the Armada Edge Platform. SE024
CE049 NightDragon says Armada units ship as turnkey solutions with the operating system, orchestration layer, and software stack already embedded. SE026
CU001 Publicly named Armada customer proof clusters in public-sector, defense, and industrial edge operations rather than broad horizontal SaaS adoption. SU001, SU004, SU005, SU009, SU024, SU028
CU002 Alaska DOT&PF uses Armada for drone-imagery and geospatial response across landslides, avalanches, rockfalls, and flooding. SU001, SU002
CU003 Alaska said pre-Armada workflows could take more than 28 hours because imagery moved by memory card and distant cloud processing. SU002, SU003
CU004 Alaska says Armada moved imagery-to-decision workflows to near real time. SU001, SU002, SU003
CU005 Data Center Dynamics reported that Alaska DOT&PF now operates two Galleons, one in Anchorage and one in Fairbanks. SU003
CU006 Alaska standardized Starlink backhaul through Atlas as part of the deployed workflow. SU002
CU007 Washington DNR uses Atlas to manage connectivity for wildfire operations and other remote government missions. SU004
CU008 Washington DNR said it had 35 separate Starlink instances without a complete picture before adopting Armada. SU004
CU009 Washington DNR now manages approximately 45 Starlinks through Atlas. SU004
CU010 Washington DNR shows Armada solving governed connectivity procurement and asset management, not only edge compute. SU004, SU014, SU015
CU011 Armada deployed a Galleon and Atlas during UNITAS 2025 from ashore and aboard a Navy warship. SU005, SU006, SU007
CU012 During UNITAS the Galleon supported Microsoft Flankspeed Edge and Minotaur workloads in disconnected maritime conditions. SU005, SU006, SU007
CU013 Armada tied UNITAS to CRADA-related testing with NIWCLANT, implying a deeper defense relationship than a trade-show demo. SU005, SU007
CU014 The Navy’s own UNITAS page confirms the exercise was a large multinational operational event, but it does not independently name Armada. SU005, SU008
CU015 Aker BP signed an agreement to deploy an offshore Galleon on the Norwegian Continental Shelf for drilling and operational data processing. SU009, SU010
CU016 Aker BP’s rollout begins with a single reference Galleon on one rig that is intended as a blueprint for additional assets. SU009, SU010
CU017 The Aker BP use case is therefore a signed reference deployment, not yet a disclosed fleet-wide offshore rollout. SU009, SU010
CU018 Armada and Carahsoft opened a Galleon Experience Center in Reston aimed at federal, state, local, education, and healthcare buyers. SU011, SU012
CU019 The Experience Center is demo and channel infrastructure rather than disclosed proof of deployed ARR or repeat contracts. SU011, SU012, SU013
CU020 Carahsoft positions Armada across SEWP, ITES-SW2, NASPO, TIPS, OMNIA, and Quilt procurement vehicles. SU013, SU014
CU021 Armada’s 2024 Carahsoft post says government customers can procure Starlink and Commander Connect through NASPO ValuePoint. SU014, SU015
CU022 Armada and Microsoft say the Azure Local plus Galleon offer is available now and both companies are actively engaging customer deployments. SU016, SU017, SU018
CU023 Microsoft frames the Azure Local offer around defense, public safety, energy, and other regulated environments where public cloud is not feasible. SU016, SU017
CU024 The Azure Local collaboration expands Armada’s reach into sovereign private cloud buyers but still lacks disclosed customer counts or contract economics. SU016, SU017, SU018
CU025 Armada, Second Front, and Microsoft said Frontier successfully deployed on Azure Local inside an Armada Galleon. SU019, SU020
CU026 The Second Front proof shows mission-critical software portability on Armada infrastructure, but it is still a partner-led proof point rather than a named end-customer award. SU019, SU020
CU027 Armada and Skydio say their partnership targets federal, state, and local agencies that need real-time drone intelligence in disconnected or emergency environments. SU021, SU002
CU028 DOE’s Genesis Mission page lists Armada among partner organizations. SU022, SU023
CU029 The Genesis Mission evidence shows collaborator status and federal relevance, not a disclosed revenue-bearing customer deployment. SU022, SU023
CU030 WinDC and Armada announced 11 MW of modular AI infrastructure across renewable energy sites in Australia. SU024, SU025, SU026, SU027
CU031 WinDC and independent coverage say the first unit is already on Australian soil. SU024, SU027
CU032 WinDC broadens Armada’s sector expansion into renewable-powered AI factories, but public materials do not identify the end-demand customers behind the capacity. SU024, SU025, SU026, SU027
CU033 Armada, Aramco Digital, and Microsoft said they deployed Galleon edge data centers, Commander, and AI applications in Saudi Arabia as an industrial distributed cloud. SU028
CU034 The Aramco announcement suggests Armada’s industrial opportunity extends beyond a single offshore reference account into broader industrial automation settings. SU017, SU028
CU035 Armada does not publicly disclose NRR, GRR, renewal rates, customer satisfaction scores, or multi-year cohort data for named accounts. SU001, SU016, SU029, SU030
CU036 Armada also does not publish a public customer count or revenue breakdown by account, leaving installed-base breadth opaque. SU001, SU004, SU016, SU029, SU030
CU037 The public proof set is concentrated in a small number of named accounts and partner surfaces, so concentration risk remains material if any reference deployment stalls. SU001, SU004, SU005, SU009, SU024
CU038 KPMG says many enterprise AI efforts stall after pilot success because operating model, governance, data, and financial readiness do not scale with the initial proof. SU029
CU039 VentureBeat cites MIT research that 95% of enterprise AI initiatives fail to deliver measurable business value, reinforcing the risk that pilots do not automatically convert to scaled execution. SU030
CU040 Crowell says FY2026 NDAA acquisition rules are being reworked to make defense acquisition more agile, implying procurement remains policy-mediated rather than frictionless for commercial vendors. SU031
CU041 Armada’s strongest public evidence is operational detail and partner corroboration, not disclosed contract values or ARR from those deployments. SU002, SU004, SU011, SU016, SU024, SU029, SU030
CU042 Carahsoft broadens Armada’s public-sector reach into education and healthcare even before Armada discloses named deployed accounts in those sectors. SU011, SU012, SU013
CR001 Armada announced a $230 million Series B at a $2 billion pre-money valuation and said the round brought total funding to nearly half a billion dollars. SR001, SR002
CR002 The Series B announcement was paired with a Johnson Controls global framework agreement and a Johnson Controls investment in Armada. SR001, SR004
CR003 Forge One is planned to span up to 400,000 square feet and create roughly 500 jobs in Arizona. SR001, SR003, SR004
CR004 Continuous production at Forge One is planned to begin with Leviathan, Armada’s megawatt-scale modular data center. SR001, SR003, SR004
CR005 Armada publicly disclosed 540% bookings growth from FY25 to FY26 and roughly 2,000% year-over-year bookings growth in Q1 FY27. SR001, SR002
CR006 Armada’s strongest public traction metric is bookings rather than disclosed revenue, ARR, gross margin, or backlog conversion. SR001, SR002
CR007 Armada’s latest financing round added strategic investors tied to infrastructure, industrial, or regional distribution as well as financial capital. SR001, SR002, SR003
CR008 Armada’s Microsoft collaboration explicitly targets defense, government, and regulated-industry workloads. SR005, SR006, SR007
CR009 The Azure Local and Galleon architecture is described for intermittently connected, contested, and fully disconnected environments. SR005, SR006, SR007
CR010 Sovereign-AI deployments are being sold on local control, auditability, and compliance rather than only on compute performance. SR005, SR006, SR017, SR029
CR011 Armada and Microsoft say they are actively engaging joint customer deployments and go-to-market efforts. SR005, SR006
CR012 Carahsoft gives Armada access to multiple public-sector contract vehicles rather than a purely direct sales path. SR008, SR030
CR013 The Carahsoft experience center is aimed at federal, state, local, education, and healthcare buyers. SR008, SR030
CR014 Armada says a Galleon and Atlas were used ashore and aboard a Navy warship during UNITAS 2025. SR009, SR010
CR015 UNITAS 2025 public descriptions say Armada supported Microsoft Flankspeed Edge and Minotaur workloads with multiple government and industry partners. SR009, SR010
CR016 Washington DNR said it previously had 35 separate Starlink instances without a complete unified view. SR013
CR017 Washington DNR now manages approximately 45 Starlinks through Atlas. SR013
CR018 Armada’s Alaska case study says the customer reduced emergency latency from 28 hours to real time. SR014, SR008
CR019 Aker BP’s rollout begins with a single offshore reference Galleon rather than a disclosed fleet deployment. SR011, SR012
CR020 The Aker BP deployment is for offshore drilling on the Norwegian Continental Shelf in a connectivity-limited environment. SR011, SR012
CR021 Offshore and at-sea use cases expose Armada to harsher environmental, logistics, and service conditions than a standard enterprise on-prem deployment. SR009, SR010, SR011, SR012
CR022 BIS said in May 2025 that it was rescinding the AI Diffusion Rule while strengthening export controls on semiconductor-related technologies and overseas AI chips. SR015, SR016
CR023 National Law Review’s summary of BIS updates says advanced-computing controls now require stronger end-use screening and diversion red-flag monitoring, including for IaaS-related activity. SR016, SR015
CR024 The European Commission’s AI Act service desk says AI systems and general-purpose AI models may be subject to legal obligations and structured compliance steps. SR017, SR029
CR025 The EU AI Act summary says high-risk AI systems and powerful general-purpose AI models face transparency, documentation, risk management, and cybersecurity obligations. SR017, SR029
CR026 CRS says FY2026 NDAA cyber and AI provisions include governance, procurement requirements, testing standards, and energy-use considerations for military data centers and AI systems. SR018, SR026
CR027 GAO says continuing resolutions limit new starts or production increases and create delays, cost overruns, and contracting bottlenecks for defense programs. SR019
CR028 CISA says integrating AI into operational technology requires governance, continuous testing, oversight, and incident-response integration because AI adds new adversarial threat avenues. SR020, SR021
CR029 DoD’s AI cybersecurity guide says AI systems require authorization, lifecycle monitoring, supply-chain risk controls, and protection of data, models, and infrastructure layers. SR021, SR020
CR030 Belfer says AI-driven data-center energy demand is outpacing capacity in some regions and can force project delays, direct power contracting, or on-site generation. SR022, SR023
CR031 Deloitte says grid bottlenecks and turbine or transformer supply constraints are central AI infrastructure bottlenecks. SR023, SR024
CR032 JLL says power rather than location or cost is now the primary site-selection criterion, with multiyear grid waits and over half of projects delayed in 2025. SR024, SR027
CR033 CBRE says H1 2025 primary-market vacancy fell to 1.6% and 74.3% of under-construction capacity was already preleased amid power and land constraints. SR027, SR028
CR034 Data Center Frontier says AI infrastructure growth is now constrained by power availability, cost pressure, and heavier infrastructure investment requirements. SR028, SR024
CR035 CoreWeave’s 2025 10-K says a substantial portion of its revenue is driven by a limited number of customers, including 67% from Microsoft in 2025. SR025
CR036 CoreWeave’s 10-K says larger cloud competitors can use greater resources, broader offerings, and existing customer or distributor relationships to win business. SR025
CR037 CoreWeave’s 10-K says export controls, privacy regulation, energy restrictions, and AI regulation can impair growth and create investigations, fines, or enforcement exposure for AI infrastructure businesses. SR025
CR038 Armada’s public proof set is concentrated in a small number of named public-sector, defense, and industrial reference accounts. SR008, SR009, SR011, SR013, SR014, SR030
CR039 Several flagship Armada deployments are exercises or single-reference systems rather than disclosed fleet-scale rollouts. SR009, SR011, SR012
CR040 Microsoft, Carahsoft, and Johnson Controls reduce go-to-market friction but also make Armada’s scale path dependent on strategic partners. SR004, SR006, SR008, SR030
CR041 Forge One and sovereign-private-cloud expansion require Armada to scale manufacturing, thermal, compliance, field-service, and AI-platform talent at the same time. SR004, SR005, SR024
CR042 Public materials reviewed for this chapter do not disclose Armada revenue, ARR, gross margin, backlog, or revenue-recognition policy. SR001, SR002
CR043 Because bookings are disclosed without revenue-conversion data, the public record cannot show whether current demand becomes predictable recognized revenue. SR001, SR002
CR044 Dan Wright remains the dominant public spokesperson across Armada’s funding and sovereign-AI partnership announcements. SR001, SR005, SR002
CR045 Founder concentration increases key-person and execution-bandwidth risk during Armada’s simultaneous factory, product, and market expansion. SR001, SR005, SR004
CR046 No reviewed public source in this chapter surfaced a direct Armada enforcement action or disclosed material security incident, so the current legal and cyber thesis is exposure-based rather than event-driven. SR001, SR002, SR005, SR015, SR020
CR047 Johnson Controls’ thermal-management expertise and roughly 40,000 field personnel partially mitigate Armada’s factory and field-service risk but do not remove partner concentration. SR004, SR001, SR002
CR048 The highest residual risks in Armada’s public record are factory execution, bookings-to-revenue conversion, sovereign-AI compliance complexity, and concentration rather than a known direct legal event. SR003, SR023, SR024, SR025, SR027
CV001 Armada announced a $230 million oversubscribed Series B in May 2026 at a $2 billion pre-money valuation. SV001, SV002
CV002 Armada said the Series B brought its total funding to nearly half a billion dollars. SV001, SV002
CV003 CNBC's May 2026 coverage listed Armada's total funding at $465 million. SV003
CV004 Using the official $2.0 billion pre-money figure implies an approximately $2.23 billion post-money valuation. SV001
CV005 That post-money math implies the new $230 million capital purchased about 10.3% of post-money equity. SV001
CV006 Armada publicly disclosed 540% customer bookings growth from FY25-26 and 2000% year-over-year bookings growth in Q1 FY27. SV001
CV007 Reviewed official and top-tier public sources did not disclose Armada's current revenue, ARR, gross margin, or backlog-conversion schedule. SV001, SV002, SV003
CV008 Galleon Forge One is planned to span up to 400,000 square feet and create about 500 jobs in Arizona. SV001, SV004
CV009 Johnson Controls says it brings advanced thermal-management expertise and a global footprint that includes more than 40,000 field personnel. SV001, SV004
CV010 Round size and valuation are different concepts: $230 million is new capital raised, while the pre- and post-money figures are equity valuation marks. SV001
CV011 Stock Analysis showed Vertiv at about $125.78 billion market cap and 11.60x trailing sales in late May 2026. SV019, SV011
CV012 Stock Analysis showed Equinix at about $106.49 billion market cap and 11.18x trailing sales in late May 2026. SV020, SV012
CV013 Stock Analysis showed Digital Realty at about $68.69 billion market cap and 10.88x trailing sales in late May 2026. SV021, SV013
CV014 Stock Analysis showed Pure Storage at about $28.96 billion market cap and 7.91x trailing sales in late May 2026. SV022, SV014
CV015 Stock Analysis showed Nutanix at about $12.69 billion market cap and 4.73x trailing sales in late May 2026. SV023, SV015
CV016 Stock Analysis showed HPE at about $49.86 billion market cap and 1.40x trailing sales in late May 2026. SV024, SV016
CV017 Stock Analysis showed Nebius at about $54.99 billion market cap and 62.64x trailing sales in late May 2026. SV025, SV017
CV018 Stock Analysis showed CoreWeave at about $57.55 billion market cap and 9.24x trailing sales in late May 2026. SV026, SV018, SV030
CV019 Vertiv's 2024 10-K describes the company as providing digital infrastructure and continuity solutions across hardware, software, analytics, and ongoing services. SV027, SV011
CV020 Nutanix's 2024 10-K describes an enterprise cloud operating system that can be delivered as an appliance or as software only. SV028, SV015
CV021 Equinix's public company description emphasizes a recurring revenue model built on colocation, interconnection, and managed infrastructure services. SV012, SV020
CV022 Digital Realty's public description centers on owned technology real estate, colocation, and interconnection infrastructure. SV013, SV021
CV023 HPE's public description spans compute, HPC and AI, intelligent edge, software, and storage, making it a diversified infrastructure benchmark rather than a pure-play AI factory. SV016, SV024
CV024 Pure Storage combines hardware platforms with cloud software and evergreen support, making it a cleaner hybrid hardware-plus-software analog than a data-center REIT. SV014, SV022
CV025 Among mature hybrid and infrastructure public names, current trailing-sales benchmarks span roughly 1.40x HPE, 4.73x Nutanix, 7.91x Pure Storage, 10.88x Digital Realty, 11.18x Equinix, and 11.60x Vertiv. SV019, SV020, SV021, SV022, SV023, SV024
CV026 Nebius is a public AI-cloud outlier at 62.64x trailing sales, while CoreWeave still trades at a premium 9.24x despite already being public and revenue disclosing. SV025, SV026, SV030
CV027 Armada's implied $2.23 billion post-money valuation is small versus the absolute market caps of public analogs but still unsupported by any public revenue denominator. SV001, SV019, SV020, SV021, SV022, SV023, SV024
CV028 Modular announced a $250 million Series C in September 2025 at a $1.6 billion valuation. SV009
CV029 Crusoe announced an initial closing of a $1.375 billion Series E round in October 2025 at an expected valuation above $10 billion. SV010
CV030 Record private infrastructure fundraising and data-center deal activity show that very large pools of capital are currently chasing AI infrastructure. SV007, SV008
CV031 Reuters reported that companies spent $37 billion in global private investment on AI infrastructure in 2024 and cited a McKinsey estimate of $5.2 trillion of data-center investment needed by 2030. SV005
CV032 Colliers reported more than $580 billion of global data-center investment in 2025 and a 47% year-over-year increase in build costs. SV008
CV033 Colliers also said power overtook location as the primary driver of site selection and that 40% to 50% of total project costs can sit in power infrastructure. SV008
CV034 CNBC reported that private capital, private credit, and debt are increasingly funding AI data-center build-outs and that insurers now treat multi-billion-dollar campuses as market-capacity stress tests. SV006
CV035 Armada shares the capital-intensity and deployment-execution profile of AI-factory infrastructure companies more than that of pure SaaS vendors. SV001, SV004, SV008, SV010
CV036 On disclosed private-market waypoints, Armada sits above Modular's $1.6 billion valuation but far below Crusoe's more-than-$10 billion scale. SV001, SV009, SV010
CV037 A milestone-banded valuation method is more defensible than a faux revenue multiple because Armada's current recognized revenue and mix are undisclosed publicly. SV001, SV003, SV007
CV038 Base-case support depends on bookings converting into recognized revenue, Forge One starting continuous production on time, and Bridge or Atlas proving recurring attach beyond headline messaging. SV001, SV004
CV039 Bear-case downside is most likely if bookings stay ahead of revenue recognition, if Forge One or site power slips, or if hardware working capital absorbs the new round faster than expected. SV001, SV006, SV008
CV040 Bull-case upside requires evidence that Armada can scale sovereign deployments beyond reference wins while monetizing software and control-plane layers, not just boxes. SV001, SV004
CV041 No reviewed public source disclosed the Series B preference stack, liquidation terms, or cap-table overhang. SV001, SV002, SV003
CV042 No reviewed public source disclosed a bookings-to-revenue bridge, customer concentration by dollars, or hardware-versus-software gross-margin mix. SV001, SV002, SV003
CV043 Because those economic disclosures are missing, Armada's May 2026 valuation is better described as stretched than attractive on public evidence alone. SV001, SV003, SV019, SV020, SV024
CV044 The current price could still prove fair if hidden revenue conversion and recurring software mix resemble premium hybrid-infrastructure peers rather than low-multiple diversified vendors. SV019, SV020, SV022, SV023, SV024, SV026
CV045 The most decision-critical next diligence asks are the revenue-bookings bridge, gross margin by layer, working-capital and capex needs, customer concentration, and Series B preference terms. SV001, SV004, SV006, SV008
来源
编号出版方标题引文
SO001 Armada Armada: Solving Your Hardest Problems at the Edge Unlock the power to compute in real-time where it matters most - anywhere on Earth.
SO002 Armada About Armada: Innovative Edge Computing Solutions At Armada, we're on a mission to redefine the boundaries of edge computing.
SO003 Armada Edge Computing, Redefined | Armada Galleon Galleon is a ruggedized mobile data center powered by our proprietary software stack.
SO004 Armada Atlas: An operational insights platform for all your connected assets Atlas is our enterprise monitoring and management platform for connected assets.
SO005 Armada Bridge: Deploy GPU-as-a-Service on Your Infrastructure Bridge is our software that enables GPU-as-a-Service through unified management, orchestration, and scheduling of GPU clusters.
SO006 Armada Armada Partners
SO007 Armada Sovereign AI at the Edge, Powered by Armada and Microsoft Armada and Microsoft unite Azure Local with AEP to enable sovereign AI and private cloud in any environment.
SO008 Armada Alaska DOT&PF Customer Story | Real Time Intelligence with Armada This change reduced processing time from twenty-eight days to a few hours.
SO009 Armada Armada Announces $230M Series B and Johnson Controls Manufacturing Agreement Galleon Forge One will span up to 400,000 square feet, and is expected to create 500 jobs.
SO010 Armada Armada Announces $131M Strategic Funding Round, Launch of Megawatt-Scale Modular AI Data Centers Armada today announced a $131 million strategic funding round ... coincides with the launch of Leviathan.
SO011 Armada Armada and Microsoft Collaborate to Deliver Sovereign AI at the Edge This solution ... is available now, and both companies are actively engaging customer deployments.
SO012 Armada Aker BP and Armada Deploy Edge Offshore Data Center Deployment will begin with a single reference Galleon on one rig.
SO013 Armada Armada Successfully Concludes UNITAS 2025 Participation An Armada Deployable Data Center (DDC), known as a Galleon, and Atlas software were tested from ashore and aboard a Navy warship.
SO014 Armada Washington DNR wildfire connectivity case study DNR now manages approximately 45 Starlinks through Atlas.
SO015 Armada Armada Resources Prag Mishra, Chief AI Officer | Aug 11, 2025
SO016 M12 Founders Feature: Armada We’re already working with many states, government agencies, and a number of Global 2000 companies in industries like oil and gas, mining, manufacturing, logistics, and others.
SO017 Armada / Business Wire Armada Announces $40M in Strategic Funding Round Led by M12 Armada products are now available in the Microsoft Azure Marketplace, with Azure customers able to use pre-committed Azure spend on Armada products.
SO018 CNBC Modular data center builder Armada raises $230 million, to build Arizona factory with new investor Johnson Controls Armada has raised a $230 million Series B round from investors at a $2 billion valuation and signed a manufacturing deal with Johnson Controls.
SO019 CNBC 26. Armada - CNBC Disruptor 50 Founders: Dan Wright (CEO), Jon Runyan, Pradeep Nair ... Headquarters: San Francisco ... Funding: $465 million.
SO020 Data Center Dynamics Containerized Edge data center firm Armada raises $131m, launches MW-scale Leviathan Armada has deployments already planned with Fidelis New Energy and Bakken Energy ... and in 2025 with Tampnet, the US Navy, Aramco, and Newlab.
SO021 Data Center Dynamics Alaska drone program taps Armada Edge computing offering The department can now operate with a decision-making window of just minutes.
SO022 Johnson Controls Armada announces agreement with Johnson Controls for modular data center production Galleon Forge One will span up to 400,000 square feet, and is expected to create 500 jobs.
SO023 Microsoft Azure Blog Building sovereign AI at the edge: Microsoft and Armada collaborate to deliver Azure Local on Galleon modular datacenters Microsoft confirms a strategic collaboration with Armada to deliver Azure Local on Galleon modular datacenters.
SO024 Wilson Sonsini Wilson Sonsini Advises Armada on $230 Million Series B Armada ... raised $230 million in an oversubscribed Series B financing at a $2 billion valuation.
SO025 Pulse 2.0 Armada Profile: Dan Wright Interview Since emerging from stealth with $55 million in funding in December 2023, Armada has made significant progress.
SO026 Dragon Global Case Study: Armada Armada’s growing customer base includes Targa Resources, Atlas Energy, SQM, Mars, Marriott, Vocus, Tampnet, the U.S. Navy, and Alaska’s Department of Transportation.
SO027 Forbes Armada founders Dan Wright and Jon Runyan are looking to bring compute to remote places Armada has no customers beyond a proof-of-concept trial, meaning its revenue remains at zero so far.
SO028 SiliconANGLE Armada raises $230M at $2B valuation to build portable AI data centers Johnson Controls will help Armada build a 400,000-square-foot manufacturing facility dedicated to producing modular data centers.
SO029 Carahsoft Armada.ai - Edge Computing Platform for the Public Sector Armada also offers an edge computing solution that unites Microsoft Azure Local with Armada's Galleons, powered by AEP.
SM001 Armada Armada: Solving Your Hardest Problems at the Edge
SM002 Armada Edge Computing, Redefined | Armada Galleon
SM003 Armada Armada to Deliver Sovereign AI at the Edge with Microsoft Azure Local
SM004 Microsoft Azure Blog Build sovereign AI at the edge with Azure Local | Microsoft Azure Blog
SM005 Data Center Dynamics Microsoft and Armada bring Azure Local to Galleon modular data centers
SM006 Johnson Controls Armada announces agreement with Johnson Controls for modular data center production
SM007 Armada Aker BP and Armada Announce Agreement to Deploy Offshore Modular Data Center
SM008 World Oil Aker BP, Armada to deploy offshore modular data center for AI-driven drilling operations
SM009 Armada Alaska DOT&PF Customer Story | Real Time Intelligence with Armada
SM010 Armada From 28 Hours to Real Time: How Alaska Uses Armada to Turn Drone Data into Decisions
SM011 Data Center Dynamics Alaska drone program taps Armada Edge computing offering
SM012 Armada Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters
SM013 Carahsoft Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters | Carahsoft
SM014 Armada Armada Brings NVIDIA AI Grid Capabilities to Telcos
SM015 PR Newswire Armada Brings NVIDIA AI Grid Capabilities to Telcos
SM016 Mitsui & Co., Ltd. Investment in U.S.-based Armada, a Provider of AI Infrastructure Solutions for Industrial Sites
SM017 Nscale Armada and Nscale Sign Letter of Intent to Accelerate Sovereign AI Through Global Hyperscale and Edge Deployments
SM018 IDC AI Infrastructure Spending Caps Historic Year at ~$90 Billion in Q4 2025; 2029 Spending to Eclipse $1 Trillion
SM019 JLL 2026 Global Data Center Outlook
SM020 Vertiv Vertiv Holdings Co. - Financials - Annual Reports
SM021 PR Newswire Vertiv Accelerates AI Infrastructure Deployment with OCP-Compliant Power, Cooling, and Rack Ecosystem
SM022 Data Center Knowledge 2026 Predictions: AI Sparks Data Center Power Revolution
SM023 Deloitte Insights A new era of self-reliance: Navigating technology sovereignty
SM024 Brookings Institution Where does federal AI spending stand in 2026?
SM025 CDO Magazine Pentagon Seeks $13.4 bn for AI and Autonomy FY 2026 Budget Request
SM026 Ericsson NTT DATA and Ericsson to scale Private 5G and Physical AI
SM027 Ericsson Ericsson Private 5G - Private network for your industry
SM028 NTT DATA NTT DATA and Ericsson Team Up to Scale Private 5G and Physical AI
SM029 BusinessWire Uptime’s 15th Annual Global Data Center Survey Results Show Both Commitment and Hesitancy as Industry Plans for Wider AI Usage, Climate Change Reporting, and the NVIDIA Revolution to Come
SM030 HostingJournalist Uptime Institute Unveils Its Data Center Predictions for 2026
SM031 Data Center Frontier Schneider Electric Maps the AI Data Center’s Next Design Era
SM032 Future Market Insights Modular Data Center Market Size, Growth 2026-2036
SM033 Research and Markets Modular Data Center Market Report 2026 - Research and Markets
SM034 National Defense University Strategic Innovation in the DoD FY 2026 RDTE Budget: Leveraging Disruptive Technologies
SM035 GovInfo Budget FY 2026 - Executive Office of the President
SP001 Armada Edge Computing, Redefined | Armada Galleon
SP002 Armada Bridge: Deploy GPU-as-a-Service on Your Infrastructure
SP003 Armada Armada Partners
SP004 Armada Armada to Deliver Sovereign AI at the Edge with Microsoft Azure Local
SP005 Armada Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters
SP006 Armada Armada Brings NVIDIA AI Grid Capabilities to Telcos
SP007 Amazon Web Services On-Premises Private Cloud - AWS Outposts Family
SP008 Amazon Web Services AWS Outposts racks features
SP009 Amazon Web Services AWS Outposts racks pricing
SP010 Microsoft Learn Azure Local documentation
SP011 Microsoft Azure Azure Local Pricing
SP012 Google Cloud Google Distributed Cloud
SP013 Google Cloud Documentation Google Distributed Cloud Documentation
SP014 Hewlett Packard Enterprise HPE Private Cloud AI
SP015 NVIDIA Newsroom Hewlett Packard Enterprise and NVIDIA Announce NVIDIA AI Computing by HPE
SP016 Dell Technologies The Dell AI Factory with NVIDIA
SP017 Dell Technologies Dell AI Factory with NVIDIA Delivers Proven Path to Enterprise AI ROI
SP018 NVIDIA Blog How Dell Technologies Is Building the Engines of AI Factories With NVIDIA Blackwell
SP019 Vertiv Prefabricated modular solutions
SP020 Vertiv Vertiv industrializes AI deployment with digitally orchestrated infrastructure
SP021 Eaton Eaton expands modular data center offering for rapid deployment of AI factories
SP022 Rittal Rittal at the OCP EMEA Summit 2026: The AI era requires standardised infrastructure
SP023 Schneider Electric EcoStruxure Modular Data Center
SP024 Business Wire Schneider Electric Launches New Data Centre Solutions to Meet Challenges of High-Density AI and Accelerated Compute Applications
SP025 Carahsoft NVIDIA Government AI Computing Solutions
SP026 Lambda The Superintelligence Cloud
SP027 Lambda AI Cloud Pricing | GPU Compute & AI Infrastructure
SP028 Lambda Managed Kubernetes for Private Cloud
SP029 Nscale The engine of superintelligence
SP030 Nscale AI Infrastructure
SP031 Crusoe Crusoe Cloud | AI Platform & Services
SP032 Crusoe Crusoe Announces New Manufacturing Facility to Produce Modular AI Factories
SP033 Open Edge Cloud Vendor Lock-In Is Technical Debt You Can Measure
SP034 Francesca Tabor Cloud Ecosystem Lock-In: Platform Dependency Economics, Developer Network Effects, and Switching Costs in Enterprise IT
SP035 Serious Insights The Serious Insights State of AI 2026 April Update
SP036 Amazon Web Services Amazon EC2 G5 Instances
SI001 Armada Armada Announces $230M Series B and Johnson Controls Manufacturing Agreement From FY25-26 Armada recorded 540% customer bookings growth. Q1 FY27 alone saw a 2000% increase in bookings growth, compared to Q1 the previous year.
SI002 Armada Bridge: Deploy GPU-as-a-Service on Your Infrastructure Bridge turns your GPUs into sovereign AI factories. Software to manage, scale, and monetize GPUs in your datacenter, cloud and edge deployments.
SI003 Armada Five Ways Bridge Turns GPU Clusters into Productive Infrastructure Armada's pricing is based on active GPU usage only, structured as GPU/year or GPU/hour.
SI004 Armada / PR Newswire Armada Launches Bridge to Power the Next Generation of AI Infrastructure Bridge also empowers data centers, telecom operators, and research institutions to monetize GPU capacity by launching GPU-as-a-Service or Model-as-a-Service offerings.
SI005 Armada Edge Computing, Redefined | Armada Galleon Galleon is a ruggedized mobile data center powered by our proprietary software stack.
SI006 Armada Atlas: An operational insights platform for all your connected assets Share data across all terminals for optimized utilization, customizable packages, transparent usage monitoring, and cost-effective management.
SI007 Armada / Business Wire Armada Announces $40M in Strategic Funding Round Led by M12 All Armada products are now available in the Microsoft Azure Marketplace, with Azure customers able to use pre-committed Azure spend on Armada products.
SI008 Armada Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters Armada Edge Platform is available through Carahsoft's SEWP V contracts NNG15SC03B and NNG15SC27B, ITES-SW2 Contract W52P1J-20-D-0042, NASPO ValuePoint Master Agreement #AR2472, TIPS Contract #220105, OMNIA Partners Contract #R240303 and The Quilt Master Service Agreement Number MSA05012019-F.
SI009 Johnson Controls Armada announces agreement with Johnson Controls for modular data center production Galleon Forge One will span up to 400,000 square feet, and is expected to create 500 jobs. Continuous production is planned to begin in the summer and will start with Leviathan.
SI010 Armada Armada Announces $131M Strategic Funding Round, Launch of Megawatt-Scale Modular AI Data Centers With this new funding, Armada will scale production of Leviathan.
SI011 CNBC Modular data center builder Armada raises $230 million $230M at a $2B valuation; bookings grew 540% FY25-FY26 and 2,000% in Q1 FY27 year over year.
SI012 CNBC 26. Armada — CNBC Disruptor 50 Total funding: $465 million.
SI013 Wilson Sonsini Goodrich & Rosati Wilson Sonsini Advises Armada on $230 Million Series B Armada on $230 million Series B.
SI014 Microsoft Azure Blog Building sovereign AI at the edge with Azure Local Both companies are actively engaging customer deployments.
SI015 Dragon Global Case Study: Armada Named customers include Targa Resources, Atlas Energy, SQM, Mars, Marriott, Vocus, Tampnet, U.S. Navy, and Alaska DOT&PF.
SI016 Data Center Dynamics Containerized edge data center firm Armada raises $131m, launches MW-scale Leviathan Armada has deployments already planned with Fidelis New Energy and Bakken Energy... So far in 2025, Armada has deployed with Tampnet, the U.S. Navy, Aramco, and Newlab.
SI017 Vertiv Vertiv Reports Strong Fourth Quarter 2024 Results Net sales of $2,346 million... The capital expenditure forecast for 2025 is ~$275 million, ~3.0% of sales.
SI018 U.S. Securities and Exchange Commission SEC EDGAR search results — Vertiv 10-K filings 10-K ... Acc-no: 0001674101-26-000008 ... 2026-02-13.
SI019 Pure Storage / PR Newswire Pure Storage Announces Fiscal Fourth Quarter and Full Year 2025 Financial Results Full-year revenue $3.2 billion... Full-year subscription services revenue $1.5 billion... Full-year GAAP gross margin 69.8%.
SI020 U.S. Securities and Exchange Commission SEC EDGAR search results — Pure Storage 10-K filings 10-K ... Acc-no: 0001474432-26-000027 ... 2026-03-25.
SI021 Nutanix Nutanix Reports Fourth Quarter and Fiscal 2024 Financial Results ARR $1.91 billion... ACV Billings $338.0 million... GAAP Gross Margin 85.2%.
SI022 U.S. Securities and Exchange Commission SEC EDGAR search results — Nutanix 10-K filings 10-K ... Acc-no: 0001193125-25-213801 ... 2025-09-24.
SI023 Equinix Equinix Reports Strong Fourth-Quarter and Full-Year 2024 Results Revenues $8.748 billion... adjusted EBITDA margin of 47%... Total capital expenditures are expected to range between $3.222 and $3.472 billion.
SI024 U.S. Securities and Exchange Commission SEC EDGAR search results — Equinix 10-K filings 10-K ... Acc-no: 0001101239-26-000032 ... 2026-02-11.
SI025 U.S. Securities and Exchange Commission SEC EDGAR search results — Eaton 10-K filings 10-K ... Acc-no: 0001551182-26-000007 ... 2026-02-26.
SI026 Data Center Knowledge Analysts Warn of Overbuild Risks as AI Data Centers Reshape Industry The accelerating investment comes with significant credit risks.
SI027 Moody's Ratings Data centers: managing risk amid a market boom Overbuilding, technology pose renewal and capex risks, particularly for turnkey assets.
SE001 Armada Armada: Solving Your Hardest Problems at the Edge Armada Edge Platform (AEP) is a full-stack edge computing platform comprised of four products: Atlas, Galleon, Bridge and Marketplace.
SE002 Armada Edge Computing, Redefined | Armada Galleon From a suitcase-sized Beacon to a megawatt-scale Leviathan, Galleon meets your operation where it is today and grows with it tomorrow.
SE003 Armada Leviathan: Megawatt Modular Data Center for AI & HPC | Armada Deploy Leviathans anywhere much more quickly than a traditional data center.
SE004 Armada Atlas: An operational insights platform for all your connected assets Atlas lets you seamlessly monitor and manage your Starlink terminals, SDWAN, drones, cameras, and sensors from a single pane of glass.
SE005 Armada Bridge: Deploy GPU-as-a-Service on Your Infrastructure Bridge unifies GPU orchestration, scaling, and management, so you can run AI workloads anywhere with cloud-like efficiency and control.
SE006 Armada Marketplace | AI Apps, Edge Hardware & Starlink | Armada Have a containerized service or application already? Import it into Marketplace, set deployment parameters, and push to any Galleon in minutes—no rewrite required.
SE007 Armada Edge Computing, AI, Starlink Control, Data Centers | Armada Atlas is our operational insights product for all your connected assets.
SE008 Armada Armada to Deliver Sovereign AI at the Edge with Microsoft Azure Local Provide continuous mission execution through a hardened, multipath communications architecture including satellite, 5G, LTE, RF, and SDWAN.
SE009 Armada Armada Announces $131M Strategic Funding Round, Launch of Megawatt-Scale Modular AI Data Centers to Accelerate American Energy and AI Dominance Leviathan’s liquid-cooled, energy-agnostic modules deliver megawatt-scale training and inference in weeks.
SE010 Armada Armada Announces Agreement with Johnson Controls for Galleon Forge One; Raises $230M in Oversubscribed Series B with a Pre-Money Valuation of $2B to Accelerate Deployment of the U.S. AI Stack and Support Explosive Customer Demand Growth Across Industries Galleon Forge One will span up to 400,000 square feet, and is expected to create 500 jobs.
SE011 Armada Documentation Overview | Armada Documentation Bridge is a IaaS (Infrastructure-as-a-Service) and PaaS (Platform-as-a-Service) solution designed to simplify GPU infrastructure management, resource allocation, and multi-tenant operations.
SE012 Armada Documentation Kubernetes Cluster (PaaS) Overview | Armada Documentation Dynamic autoscaling of AI/ML workloads within and across clusters, based on GPU utilization.
SE013 Armada Documentation Cluster Templates Overview | Armada Documentation The template determines what software and capabilities are provisioned on the cluster.
SE014 Armada Documentation SLURM Overview | Armada Documentation Bridge provides an option to create and manage Slurm setups directly from the UI.
SE015 Armada Documentation GPU & MIG Configuration | Armada Documentation In Bridge, you can configure and manage MIG profiles directly from the UI, without manual CLI setup.
SE016 Microsoft Azure Blog Build sovereign AI at the edge with Azure Local Resilient multi network connectivity, spanning satellite, LTE/5G, RF, and SD-WAN.
SE017 Johnson Controls Armada announces agreement with Johnson Controls for modular data center production Johnson Controls brings deep expertise in advanced thermal management and mission-critical building systems.
SE018 Carahsoft Armada.ai - Edge Computing Platform for the Public Sector | Carahsoft Armada also offers an edge computing solution that unites Microsoft Azure Local with Armada's Galleons, powered by AEP.
SE019 PR Newswire / Armada Armada Brings NVIDIA AI Grid Capabilities to Telcos AEP provides a unified control plane across geographically distributed AI infrastructure including existing service provider data centers, centralized AI factories, regional hubs, and edge locations.
SE020 Data Center Dynamics Microsoft and Armada bring Azure Local to Galleon modular data centers Armada's Galleon line comprises portable, modular, and ruggedized data centers that come as a turnkey offering.
SE021 TechCrunch AI companies are building huge natural gas plants to power data centers. What could go wrong? Companies won’t be able to place new orders until 2028, and it’s taking six years to get turbines delivered.
SE022 TechCrunch As AI data centers hit power limits, Peak XV backs Indian startup C2i to fix the bottleneck Power, rather than compute, is fast becoming the limiting factor in scaling AI data centers.
SE023 Garuda Ventures Job Board Senior Software Engineer - AI Infrastructure You will be instrumental in building a robust and scalable infrastructure platform ... onto our on-premise bare metal Kubernetes clusters.
SE024 Data Center Dynamics Containerized Edge data center firm Armada raises $131m, launches MW-scale Leviathan Leviathan uses liquid-cooling.
SE025 Dell Dell and Palantir Introduce an On-Premises AI Operating System Palantir Rubix and Apollo provide a containerized Kubernetes environment with multi-tenant isolation, continuous audit logging and centralized fleet management across clusters and sites.
SE026 NightDragon Powering AI at the Edge: Why NightDragon Invested in Armada Every Armada unit ships as a turnkey solution, with the operating system, orchestration layer, and software stack already embedded.
SU001 Armada Alaska DOT&PF Customer Story | Real Time Intelligence with Armada
SU002 Armada From 28 Hours to Real Time: How Alaska Uses Armada to Turn Drone Data into Decisions
SU003 Data Center Dynamics Alaska drone program taps Armada Edge computing offering
SU004 Armada From P-Card Sprawl to Split-Second Response: How Washington DNR Manages Connectivity for Wildfire Operations with Atlas
SU005 Armada Armada Participates in U.S. Navy UNITAS Exercise, Delivering Power, Data, and Resilience in Disconnected Environments
SU006 Armada via PR Newswire Armada Participates in U.S. Navy UNITAS Exercise, Delivering Power, Data, and Resilience in Disconnected Environments
SU007 Data Center Dynamics Containerized Edge data center firm Armada takes part in US Navy UNITAS exercise
SU008 U.S. Navy UNITAS 2025 To Be Held Across Multiple Locations Along the East Coast of United States
SU009 Armada Aker BP and Armada Announce Agreement to Deploy Offshore Modular Data Center
SU010 World Oil Aker BP, Armada to deploy offshore modular data center for AI-driven drilling operations
SU011 Armada Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters
SU012 Carahsoft Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters | Carahsoft
SU013 Carahsoft Armada.ai - Edge Computing Platform for the Public Sector | Carahsoft
SU014 Carahsoft Armada Government Solutions | Carahsoft
SU015 Armada Delivering an Enterprise Starlink Experience to Government Seamlessly via Carahsoft
SU016 Armada Armada to Deliver Sovereign AI at the Edge with Microsoft Azure Local
SU017 Microsoft Azure Blog Build sovereign AI at the edge with Azure Local | Microsoft Azure Blog
SU018 Data Center Dynamics Microsoft and Armada bring Azure Local to Galleon modular data centers
SU019 Armada Armada, Second Front, and Microsoft Partner to Power Mission-Critical Applications at the Edge
SU020 Second Front Armada, Second Front, and Microsoft Partner to Power Mission-Critical Applications at the Edge
SU021 Armada Armada and Skydio Partner to Deliver Real-Time Drone Intelligence for National Security Missions
SU022 Armada Armada Agreement with United States Department of Energy to Accelerate Genesis Mission
SU023 U.S. Department of Energy Genesis Mission Collaboration
SU024 Armada WinDC and Armada Join Forces to Turn Australia’s Renewable Energy Advantage into a Global AI Hub
SU025 Data Center Dynamics Edge data center firm Armada secures 11MW deployment with WinDC in Australia
SU026 PV Magazine Australia Portable data centers target curtailment, grid constraints
SU027 W.Media WinDC partners Armada to deploy modular AI DCs at renewable energy sites
SU028 Armada Aramco, Armada, and Microsoft Collaborate to Deploy World’s First Industrial Distributed Cloud
SU029 KPMG Why Enterprise AI Maturity Stalls After Pilot Success
SU030 VentureBeat Why enterprise AI pilots fail — and how to move to scaled execution
SU031 Crowell & Moring The FY 2026 National Defense Authorization Act
SR001 Armada Armada Announces Agreement with Johnson Controls for Galleon Forge One; Raises $230M in Oversubscribed Series B with a Pre-Money Valuation of $2B to Accelerate Deployment of the U.S. AI Stack and Support Explosive Customer Demand Growth Across Industries From FY25-26 Armada recorded 540% customer bookings growth... Q1 FY27 alone saw a 2000% increase in bookings growth.
SR002 CNBC Modular data center builder Armada raises $230 million, to build Arizona factory with new investor Johnson Controls Customer bookings grew 540% between FY25 and FY26, with Q1 FY27 alone posting roughly 2,000% year-over-year growth.
SR003 Data Center Dynamics Edge data center firm Armada raises $230m in latest funding round
SR004 Johnson Controls Armada announces agreement with Johnson Controls for modular data center production Galleon Forge One will span up to 400,000 square feet, and is expected to create 500 jobs... Continuous production is planned to begin in the summer and will start with Leviathan.
SR005 Armada Armada to Deliver Sovereign AI at the Edge with Microsoft Azure Local
SR006 Microsoft Build sovereign AI at the edge with Azure Local This customer-controlled cloud environment delivers Azure’s operating model, security, and AI-ready capabilities where traditional cloud approaches are not feasible.
SR007 Data Center Dynamics Microsoft and Armada bring Azure Local to Galleon modular data centers
SR008 Carahsoft Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters
SR009 Armada Armada Participates in U.S. Navy UNITAS Exercise, Delivering Power, Data, and Resilience in Disconnected Environments
SR010 Data Center Dynamics Containerized Edge data center firm Armada takes part in US Navy UNITAS exercise
SR011 Armada Aker BP and Armada Announce Agreement to Deploy Offshore Modular Data Center
SR012 World Oil Aker BP, Armada to deploy offshore modular data center for AI-driven drilling operations Initial deployment will focus on a single offshore installation, serving as a reference system for potential replication across additional assets.
SR013 Armada From P-Card Sprawl to Split-Second Response: How Washington DNR Manages Connectivity for Wildfire Operations with Atlas
SR014 Armada Alaska DOT&PF Customer Story | Real Time Intelligence with Armada
SR015 Bureau of Industry and Security Department of Commerce Announces Rescission of Biden Era Artificial Intelligence Diffusion Rule, Strengthens Export Controls on Semiconductor-Related Technologies BIS today announced actions to strengthen export controls for overseas AI chips.
SR016 The National Law Review BIS Issues Four Key Updates on Advanced Computing and AI Export Controls This development underscores the importance of robust due diligence and end-use screening... especially to Infrastructure as a Service providers.
SR017 European Commission AI Act Service Desk - AI Act Single Information Platform
SR018 Congressional Research Service Cyber and Artificial Intelligence Provisions in the FY2026 National Defense Authorization Act (NDAA)
SR019 U.S. Government Accountability Office Defense Budget: Effects of Continuing Resolutions on Selected Activities and Programs Critical to DOD’s National Security Mission CRs include constraints that place limits on starting new programs or increasing production of weapon systems and munitions.
SR020 CISA New Joint Guide Advances Secure Integration of Artificial Intelligence in Operational Technology OT systems are the backbone of our nation’s critical infrastructure, and integrating AI into these environments demands a thoughtful, risk-informed approach.
SR021 DoD CIO U_AI_Cybersecurity Risk Manangment Tailoring Guide_14July2025_vF
SR022 Belfer Center AI, Data Centers, and the U.S. Electric Grid: A Watershed Moment AI-driven energy demand is outpacing available capacity, driving companies to delay projects, contract power directly from private producers, and/or install multiple... generators.
SR023 Deloitte Can US infrastructure keep up with the AI economy?
SR024 JLL JLL 2026 Global Data Center Outlook Power, not location or cost, will be the primary site selection criteria due to multiyear wait times for a grid connection.
SR025 CoreWeave crwv-20251231 We recognized an aggregate of approximately 67% of our revenue from our top customer, Microsoft, for the year ended December 31, 2025.
SR026 Greenberg Traurig A Comprehensive Analysis Of The Fiscal Year 2025 National Defense Authorization Act’s Impact On Federal Procurement Law
SR027 CBRE North America Data Center Trends H1 2025 Preleasing activity remained strong with 74.3% of all under-construction capacity already committed, driven by cloud and AI providers seeking to lock in future infrastructure amid power and land constraints.
SR028 Data Center Frontier New Reports Show How AI, Power, and Investment Trends Are Reshaping the Data Center Landscape
SR029 EUR-Lex Rules for trustworthy artificial intelligence in the EU
SR030 Armada Armada and Carahsoft Open First Galleon Experience Center at Carahsoft Headquarters
SV001 Armada Armada Announces Agreement with Johnson Controls for Galleon Forge One; Raises $230M in Oversubscribed Series B with a Pre-Money Valuation of $2B to Accelerate Deployment of the U.S. AI Stack and Support Explosive Customer Demand Growth Across Industries Armada announced a $230M oversubscribed Series B with a pre-money valuation of $2B and said bookings grew 540% from FY25-26 and 2000% in Q1 FY27 YoY.
SV002 Wilson Sonsini Wilson Sonsini Advises Armada on $230 Million Series B On May 19, 2026, Armada announced that it has raised $230 million in an oversubscribed Series B financing at a $2 billion valuation.
SV003 CNBC Modular data center builder Armada raises $230 million, to build Arizona factory with new investor Johnson Controls CNBC described Armada as a modular data center builder that raised $230 million and is using the round to build an Arizona factory with Johnson Controls.
SV004 Johnson Controls Armada announces agreement with Johnson Controls for modular data center production Galleon Forge One will span up to 400,000 square feet, is expected to create 500 jobs, and Johnson Controls says it has more than 40,000 field personnel across key regions.
SV005 Reuters Just how big is the AI investment wave? This unprecedented spending spree creates risks of a financial bubble and questions about the circular financing deals propping up skyrocketing valuations.
SV006 CNBC AI data center boom stress tests insurers as private capital floods in We are talking about trillions of dollars, and almost going back to the same cycle where there is almost no transparency about the financing structures.
SV007 S&P Global Market Intelligence Rising private infrastructure fundraising; data center investment boom Private infrastructure funds collected a record US$250.70 billion in investor capital commitments in 2025.
SV008 Colliers 2026 Data Center Marketplace Report 2025 saw more than $580B of global data center investment, build costs rose 47% YoY, and power overtook location as the primary driver of valuation and liquidity.
SV009 Modular Modular Raises $250M to scale AI's Unified Compute Layer Modular raised $250M and said the round valued the company at $1.6 billion.
SV010 Crusoe Crusoe, the AI factory company, raising $1.375 billion at a valuation above $10 billion to power the future of AI infrastructure Crusoe announced the initial closing of an anticipated $1.375 billion Series E round at an expected valuation above $10 billion.
SV011 Macrotrends Vertiv Holdings Market Cap 2017-2025 | VRT
SV012 Macrotrends Equinix Market Cap 2011-2025 | EQIX
SV013 Macrotrends Digital Realty Trust Market Cap 2010-2025 | DLR
SV014 Macrotrends Pure Storage Market Cap 2015-2026 | PSTG
SV015 Macrotrends Nutanix Market Cap 2014-2025 | NTNX
SV016 Macrotrends Hewlett Packard Enterprise Market Cap 2013-2025 | HPE
SV017 Macrotrends Nebius Group Market Cap 2012-2025 | NBIS
SV018 Macrotrends CoreWeave Revenue 2025-2025 | CRWV
SV019 Stock Analysis Vertiv Holdings Co (VRT) Statistics & Valuation
SV020 Stock Analysis Equinix (EQIX) Statistics & Valuation
SV021 Stock Analysis Digital Realty (DLR) Statistics & Valuation
SV022 Stock Analysis Pure Storage (PSTG) Statistics & Valuation
SV023 Stock Analysis Nutanix (NTNX) Statistics & Valuation
SV024 Stock Analysis HPE (HPE) Statistics & Valuation
SV025 Stock Analysis Nebius Group (NBIS) Statistics & Valuation
SV026 Stock Analysis CoreWeave (CRWV) Statistics & Valuation
SV027 Securities and Exchange Commission Vertiv Holdings Co Form 10-K for fiscal year ended December 31, 2024
SV028 Securities and Exchange Commission Nutanix Form 10-K for fiscal year ended July 31, 2024
SV029 Stock Analysis Vertiv Holdings Co (VRT) Revenue
SV030 Stock Analysis CoreWeave (CRWV) Revenue 2022-2026