初创公司尽调
尽调报告 AI infrastructure / GPU cloud / data centers Growth-stage private company 2026-10-20

Firmus Technologies

大额战略融资和客户承诺支撑了机会,但运营经济性和交付证据仍过于有限,尚不足以支撑估值判断。

Firmus 获得了可信的战略支持,也出现了需求信号,但其报告估值已经超过公开证据所能支撑的运营经济性、融资交付能力和持续利用率。

覆盖范围与披露说明

截至 2026-10-20。公开证据未披露经审计收入、利润率、利用率、现金跑道、客户集中度或完整资本结构。已宣布和已签约的产能不视为已投产或已产生收入的产能。

封面要素

据报道的投后估值 02
>10.5 USD billion [CO006, CV001]
公司披露的签约产能 03
>900 MW [CU001]
2026 年 10 月披露的运营站点 04
2 sites [CO008, CP009]

公司概况

Firmus Technologies 是一家在澳大利亚注册的 AI 基础设施平台,开发并运营垂直整合的 AI 工厂和 GPU 云服务。其公开方案在亚太扩张计划中整合计算、网络、存储、液冷、电力管理和电网编排。战略投资者及具名超大规模客户协议证明了资本和需求渠道,但有限的运营披露使利用率、单位经济性和项目转化情况仍未得到验证。

官网
firmus.co
创始人
Oliver Curtis, Tim Rosenfield
总部
Sydney, Australia
产品
模块化 AI 工厂和 GPU 云产能,整合计算、网络、存储、液冷、电力系统及感知电网的编排能力。
客户
超大规模模型开发商、主权及政府工作负载、教育和研究用户,以及需要区域性高密度 AI 计算的企业。
商业模式
Firmus 开发并运营 AI 工厂基础设施,销售专用或云接入 GPU 计算产能;公开来源未披露标准化定价或合同经济性。
阶段
Growth-stage private company
融资情况
公司宣布 2026 年 8 月完成一轮全额认购的 USD 2 billion 战略股权融资,按公司披露的投后估值计价超过 USD 10.5 billion;拟议的 ASX IPO 于 2026 年 10 月撤回。
[CO001, CO002, CO005, CO006, CO007, CU001, CU002, CU003]

执行摘要

主要优势

  • 垂直整合的 AI 工厂与 GPU 云栈覆盖计算、网络、存储、冷却、电力和电网编排。
  • 一轮 USD 2 billion 股权融资,以及据报道规模达 USD 10 billion 的债务额度,显示公司获得战略资本和机构资本的能力非同寻常。
  • 已点名的 Meta 和 OpenAI 协议,加上公司报告的超过 900 MW 合同容量,构成了有分量的需求信号。

主要风险

  • 多站点建设、许可、电网接入和调试延误,可能进一步放大平台的高资本强度。
  • 经审计收入、利用率、利润率、现金消耗、现金跑道和站点级单位经济性均未公开披露。
  • 庞大的融资需求和未成功的 IPO 进程,增加了再融资、稀释和资本可获得性风险。
  • 客户集中度、合同保障、与 NVIDIA 绑定的供应商集中度以及治理控制仍披露不足。

未决问题

  • 按运营站点取得经审计收入、现金流、利用率、单位经济性、现金跑道和客户集中度数据。
  • 将合同容量、已融资容量、在建容量、已调试容量和产收容量,与里程碑日期及验收证据逐项核对。
  • 审阅完全稀释后的股权结构表、优先股堆叠、投资人权利、项目层面债务、契约、到期日、抵押品和融资瀑布。
  • 核实客户合同期限、照付不议保障、续约和终止权利、定价、流失率以及客户层面的容量分配。
  • 取得完整的许可和电网审批登记册,以及独立的安全性、可靠性、能源和水资源绩效证据。

目录

Chapter 01

01公司概览

1.1 身份与运营版图

Firmus Grid Limited 以 Firmus Technologies 名义经营,自称是一家整合型 AI Factory Platform 公司。其运营方案覆盖实体 AI 工厂、GPU 云产能、冷却、电力管理和感知电网的软件,而不是传统的纯托管模式。公司称其于 2019 年在澳大利亚注册,2026 年 8 月公告将总部列为悉尼。10 月的独立报道描述了一个分布于澳大利亚、新加坡、印度尼西亚和马来西亚的七座工厂组合,但其中只有两座在运营,五座仍在开发。这一区分至关重要:已宣布或已签约产能不等于已投产、可产生收入的产能。本报告审阅的公开来源未披露当前收入、利用率、客户集中度、员工人数或逐地点完整运营排期,因此这些指标仍是明确的尽调缺口,而不是可推断的封面指标。[CO001, CO002, CO008, CO009]

KPI 快照与证据状态
指标数值或状态截至置信度缺口或限定
法律身份Firmus Grid Limited 以 Firmus Technologies 名义经营2026-08-07中公司披露;指定材料未收录公司注册档案
总部悉尼,澳大利亚2026-08-07中公司披露;未保留独立注册信息佐证
注册成立2019 年在澳大利亚未注明日期的公司历史中无法取得确切注册日期和注册摘录
商业模式AI 工厂加 GPU 云服务2026-10-20 访问中经济性和细分收入未公开
战略股权融资USD 2.0 billion 全额认购2026-08-07中条款、优先权和现金可用性未披露
投后估值高于 USD 10.5 billion2026-08-07中公司披露的交易估值
站点版图4 个国家 7 座工厂;2 座运营、5 座开发中2026-10-09中独立报道转述的组合状态;投产排期仍未验证
收入或运行率2026-10-20低指定证据中没有可支持的公开数字
客户、员工人数与利用率2026-10-20低指定证据中没有可支持的当前数量

数值保留来源中的货币、日期和状态;null 表示指定的预取证据不支持当前数量。

[CO001, CO002, CO005, CO006, CO008, CO009]

1.2 领导层、治理与利益相关方

Firmus 由联席首席执行官 Oliver Curtis 和 Tim Rosenfield 领导。现有证据显示创始人集中度较高,但未列出完整董事会、委员会架构、独立董事或继任机制。这一缺失值得关注:平台正进行资本密集型、多国扩张,同时需要反复获得大规模股权和债务资金。Curtis 此前因内幕交易被定罪并服刑一年,这是治理尽调中的重要事实,但不能证明当前存在不当行为;投资者应核实此后采用的控制措施、授权边界、关联方政策和董事会监督。战略支持力度很大:8 月融资包括 Coatue、NVIDIA、Blackstone 管理的基金和 Jane Street。它们的参与证明了公司可接触专业资本和生态关系,但投资者声望或融资规模都不能替代股权结构表条款、锁定期、清算优先权或项目层面融资披露。[CO003, CO004, CO005, CO006]

领导层与创始人表
人员当前职务公开背景职能覆盖关键人物依赖
Oliver Curtis联席首席执行官创始人;前投资银行家;曾因内幕交易被定罪并服刑一年公司领导与募资高:联席 CEO 兼重要对外代表;董事会控制和继任机制未披露
Tim Rosenfield联席首席执行官创始人;独立报道称其为 Curtis 的表亲公司领导与平台执行高:联席 CEO;职责分工、继任和授权边界未披露

领导层仅根据两篇 2026 年 10 月独立报道部分列出;指定材料中没有完整高管团队和董事会名单。

[CO003, CO004]
利益相关方或投资者图谱
利益相关方角色控制或经济重要性证据状态尽调要求
Oliver Curtis创始人兼联席 CEO领导影响力;持股比例未披露独立报道获取完全摊薄持股、投票权、锁定期和关联方披露
Tim Rosenfield创始人兼联席 CEO领导影响力;持股比例未披露独立报道获取完全摊薄持股、投票权和继任计划
NVIDIA后续战略投资者及技术生态参与者ABC 报道持股 7.2%;当前完全摊薄比例需要核实公司及独立报道确认当前持股、商业协议、治理权和稀释情况
Coatue后续投资者参与 8 月 USD 2 billion 融资公司披露确认融资分配、优先股堆叠、董事会或观察员权利及锁定期
Blackstone 管理的基金8 月轮新投资者参与 8 月 USD 2 billion 融资公司披露确认投资载体、优先股堆叠、项目债务暴露和治理权
Jane Street新战略股权参与者参与 8 月 USD 2 billion 融资公司披露确认分配、转让权和任何商业安排

这是保留来源中具名利益相关方的部分图谱;经济条款、当前持股和治理权仍属私有信息。

[CO003, CO005, CO008]

1.3 融资轨迹与里程碑记录

Firmus 宣布 2026 年 8 月完成全额认购的 USD 2 billion 战略股权融资,并称交易将投后估值推高至 USD 10.5 billion 以上。该融资之前,公司披露的进展包括早期 HyperCube 研究、可再生能源供电的数据大厅和一体化计算部署、新加坡主权 AI 项目启动以及 Project Southgate。2026 年 10 月撤回 IPO 是更尖锐的当前信号。Firmus 称拟议条款未反映其长期前景,并将转向私募资本;独立报道则提到需求疲弱、估值阻力和未来资金需求担忧。撤回并未取消项目,但撤掉了计划中的公开股权融资路径,使扩张依赖替代资本。因此,里程碑记录一方面显示真实的技术和募资进展,另一方面也显示验证需求正在扩大:管理层应在一份可审计的排期中核对运营产能、建设里程碑、已承诺融资和投产日期。[CO005, CO006, CO007, CO010, CO011, CO012, CO013, CO014, CO015, CO016]

FO001: 公司里程碑时间线

公司报告的技术里程碑最终促成了一轮大型私募融资;两个月后,公司撤回了拟议 IPO。

2026 年之前的历史日期沿用 Firmus 自有时间线中的期间标签;本资料池未对公司里程碑进行独立审计。

在澳大利亚注册成立 [CO009]

初始 HyperCube 研发完成 [CO010]

浸没式冷却数据大厅完成 [CO011]

集成式 AI 算力平台完成 [CO012]

新加坡主权 AI 平台上线 [CO013]

塔斯马尼亚电网集成部署 [CO014]

公司援引 GPU 云前三排名 [CO015]

Southgate 项目宣布 [CO016]

宣布 USD 2 billion 战略轮融资 [CO005, CO006]

撤回拟议 IPO [CO007]

[CO005, CO007, CO009, CO010, CO011, CO012, CO013, CO014, CO015, CO016]

1.4 展示项

Chapter 02

02市场分析

2.1 市场边界与替代方案

Firmus 应按垂直整合的 AI 基础设施运营商来评估:它把高密度 GPU 计算、冷却、网络、电力接入和云接入层组合在一起。除非支出明确转向加速器密集型训练或推理,否则这一边界不包括规模更大的普通公有云、企业托管和通用数据中心地产。现状不只是另一家新云厂商;成熟买家还可以预订超大规模云厂商的 GPU 实例、托管自有系统或内部建设。Firmus 更窄的切入口是主权化、区域部署的产能,在这里数据本地性、部署速度和能源效率更重要。其官网描述了模块化 AI 工厂和新加坡 3 MW 改造项目,但效率优势仍属于公司声明,尚无公开的每 token 成本、利用率或独立标准化电力数据。[CM001, CM002, CM003]

市场定义与替代边界
类别纳入的支出或工作流排除或相邻支出买方或付款方相关性
GPU 云与 AI 工厂产能加速器计算加一体化电力、冷却、网络和运营通用 CPU 云AI 实验室、企业和公共部门技术预算核心市场
主权 AI 基础设施境内或区域控制的 AI 工作负载和产能不要求本地性的普通境外云政府机构和受监管企业优先切入口
超大规模云 GPU 服务租用的替代加速器产能非 AI 超大规模云服务云和 AI 平台团队直接替代
托管与客户自有集群空间、电力、网络和自有加速器部署不承载 AI 工作负载的普通企业机架基础设施和数据中心预算自建或采购替代
电网、土地与可再生能源激活 AI 工厂产能所需的投入尚未与计算项目签约的发电量开发商资本开支与项目融资预算关键相邻环节与瓶颈

该边界根据 Firmus 的产品描述和报道中的部署模式划定;各类别不是可相加的市场规模。

[CM001, CM002, CM010]

2.2 受证据约束的市场规模

现有证据不足以支持干净的 TAM/SAM/SOM 计算。一篇低置信度市场文章引用 2026 年澳大利亚云支出 A$33.6 billion 和 AI 专用云基础设施支出 A$946 million,但未公开底层数据集或方法。公司相关报道则描述了一个计划中的澳大利亚五站点、1.6 GW 项目,到 2028 年成本最高可达 A$73.3 billion;其他报道重复了 1.6 GW 目标,能佐证已宣布的范围,但不能证明需求、利用率、收入或经济性。因此,这些视角是背景,而非可相加的市场估算。可投资的规模测算任务,是核对已签约、已获批、在建、已通电和已产生收入的产能,再套用现实的利用率和定价。公开来源没有提供这座桥梁。[CM004, CM005, CM006, CM007, CM008]

市场规模与部署视角
视角期间地理范围数值方法或依据置信度与局限
整体云支出2026澳大利亚A$33.6bn;同比增长约 18%Tech Insider 援引的行业跟踪数据低:底层数据集和分类边界未披露
AI 专用云基础设施2026澳大利亚A$946m;增长 128.4%Tech Insider 援引的行业跟踪数据低:没有可审计的方法,也没有 Firmus 可触达份额
Firmus 全国项目2028 年目标澳大利亚五个站点合计最高 A$73.3bn、1.6GWStartup Daily 报道的 Project Southgate 已宣布部署计划中:这是已宣布的投资和产能,不代表市场需求或运营产出
塔斯马尼亚与墨尔本初始范围预计 2026 年澳大利亚合计 150MW;墨尔本包括 18,500 枚 GB300 GPUStartup Daily 报道的项目范围中:尚未证明已投运及利用率
可见的塔斯马尼亚项目管线当前提案与审批状态塔斯马尼亚合计 444MW 的需求申请The Urban Developer 报道的 52MW Wesley Vale、104MW St Leonards 和 288MW Bell Bay 管线中:数据与全国项目重叠,且包含尚未运营的产能

各数值使用不同的分母和成熟度阶段,因此分别呈现,不能相加为 TAM、SAM 或 SOM。

[CM004, CM005, CM007, CM008, CM011, CM012]
FM001: 已宣布足迹中的报告容量标记

六个报告中的兆瓦标记展示了 Firmus 已宣布足迹的规模和不均衡成熟度;重叠数值不可相加,也不代表 TAM。

数值按来源报告原样展示,未作转换。多个标记相互重叠,项目成熟度从目标、提案到获批不等。

[CM005, CM011, CM012, CM014]

2.3 买方、用户、付款方与采用路径

报道中的目标用户覆盖企业、教育、政府、初创公司、成长型公司、AI 实验室和超大规模云厂商,但它们的采购方式不同。政府和受监管企业是最清晰的主权计算客群:技术负责人和采购团队可能看重本国工作负载控制,最终则由财务部门或机构预算支付产能费用。AI 实验室和超大规模云厂商更可能围绕已交付 GPU、电力、网络和投产里程碑谈判大额产能承诺。中小企业和研究用户可能消费 Firmus Cloud,而不是签下整座设施。采用取决于工作负载评估、安全与主权审查、基准验证、商业承诺、站点通电和生产迁移。公开证据列出了广泛用户群,但未披露预算归属、定价、最低承诺或各细分市场的合同转化。[CM009, CM010]

细分市场、买方、付费方与采用路径
细分市场买方用户付费方或预算负责人采用触发因素关键不确定性
政府与受监管行业机构 CIO、受监管企业 CTO、采购部门公共部门与受监管行业的 AI 团队机构或企业技术预算主权性、本地化、安全性与可用的国内产能未公开定价、采购授标或转化数据
AI 实验室与超大规模云厂商产能与基础设施高管模型训练与推理团队中央基础设施或计算预算大规模已投运 GPU 产能与电力区块合同承诺、取消权与客户集中度未披露
大型企业CIO、CTO、数据与 AI 负责人内部产品与数据科学团队云或转型预算性能、成本、可靠性与迁移准备度没有独立验证的工作负载经济性
教育与科研大学科研计算负责人研究人员与模型开发者科研基础设施、拨款或联盟资金获得本地高性能计算资源具名用户和经常性预算证据稀少
初创公司与规模化初创公司创始人、CTO 或平台负责人工程与 ML 团队产品基础设施预算无需自建集群即可更快获得 GPU价格、最低承诺与留存率未公开

细分市场和采用触发因素综合了报道中的目标用户与主权计算主张;付费方归属和合同机制仍是尽调缺口。

[CM009, CM010, CM014]

2.4 增长驱动、约束与估值相关性

加速器计算、主权部署和受限的区域电力需求是核心增长驱动。Firmus 还声称可降低电网负荷、用水、占地和建设成本,若能独立证明,可能改善买方 ROI。但执行证据更不宽容:拟议的 52 MW Wesley Vale 项目面临完整提案截止期,更广泛的塔斯马尼亚管线取决于对 444 MW 长期能源申请的评估,当地团体还提出了用水和噪音担忧。资本负担同样关键,因为全国已宣布投资额和运营产能并不等价。Australian Financial Review 报道的 10 月 6 日 IPO 计划,被 The Urban Developer 于 10 月 9 日报道的撤回所取代,说明融资渠道变化可能快于实体项目管线。因此,估值应跟踪已投产产能、签约收入、利用率和已落实资本开支,而不是项目总值或计划兆瓦数。[CM002, CM011, CM012, CM013, CM014, CM015, CM016]

2.5 展示项

Chapter 03

03竞争对手

3.1 竞争格局与买方替代方案

相关竞争集合不只是 GPU 云厂商名单。Tracxn 将 CoreWeave、Together AI、FluidStack、Crusoe 及若干分布式计算平台列入 Firmus 的同行;另一份市场画像则将 Submer、LiquidStack 和 KyotoCooling 列为专业基础设施替代方案。买方还可以从 AWS、Microsoft Azure、Google Cloud 或 Oracle 采购,使用传统数据中心主机并另行采购加速器,或自行建设。这些选择解决的是同一任务的不同部分:以可接受的电力、主权、部署速度和单位成本获得可靠 AI 计算。因此,Firmus 同时争夺云工作负载和实体基础设施层的控制权。Tracxn 报道的 363 家活跃竞争者凸显了市场碎片化,但其分类很宽泛,不能理解为 363 家直接可比的 AI 工厂运营商。[CP002, CP003, CP014]

竞争对手画像与替代方案图谱
替代方案类别规模或融资证据目标任务差异化指定证据中的局限
Firmus一体化 AI 工厂运营商 / 新型云厂商Tracxn:融资 US$3.05B;公司:US$10B 债务额度亚太主权 AI 计算与实体 AI 工厂在计算、网络、电力和冷却之间进行垂直协同设计未获得公开利用率、实际定价和站点层面经济数据
CoreWeave直接 GPU 云同业Tracxn:公开;融资 US$2.38B面向 AI 工作负载的云计算聚焦 AI 云的运营模式指定证据中没有定价或亚太交付对比
FluidStack直接托管 GPU 同业Tracxn:B 轮;融资 US$2.55B托管式高性能 GPU 基础设施托管集群专业化指定证据中没有合同、可靠性或利润率对比
CrusoeAI 数据中心与云同业Tracxn:F 轮;融资 US$6.67BAI 基础设施与云产能数据中心与云服务合并布局指定证据中没有按地理范围标准化的能力对比
AWS / Azure / Google Cloud / Oracle超大规模云既有厂商列为公有平台替代方案;没有可比融资指标广泛的企业云与加速器采购分销、生态与多区域可用性指定证据未统一比较主权控制或 AI 工厂效率
Submer / LiquidStack / KyotoCooling专业基础设施替代商指定证据未量化规模冷却或设施现代化,不包含完整云服务采用买方可以采购更窄的基础设施层没有独立基准化的性能或定价对比

竞争格局基于当前指定来源的部分信息;融资数值由 Tracxn 报道,不代表收入、可用现金,也不是运营规模的直接可比指标。

[CP001, CP002, CP003, CP004, CP014]

3.2 能力、分销与锁定

Firmus 声称的差异化在于跨计算、网络、电力、冷却和电网编排的垂直协同设计,以及 APAC 主权计算版图。相比单独拼接主机、冷却供应商、硬件供应商和编排层,这可能降低协调成本;当产能、液冷设计、电力接入和长期设施被捆绑时,也可能加深锁定,迁移不再只是云软件决策。但同一体化也增加了执行暴露。报道显示,计划中的 CDC 合作在只建成一座 42 MW 中心后终止,说明即使宣布了大型全国扩张,合作伙伴准入和交付安排仍会改变。超大规模云厂商保有分销和生态优势,而专业供应商让成熟买方只采购冷却或设施层,不必采用 Firmus 的全栈方案。[CP001, CP005, CP008, CP013, CP014]

功能与采购标准矩阵
采购标准Firmus直接新型云厂商超大规模云厂商专业基础设施供应商证据限定
全栈 AI 工厂设计公司自述:整合计算、网络、电力和冷却各同业的云或托管 GPU 范围不同广泛云栈;此处未确立设施整合更窄的冷却或设施组件一手指定证据只描述了 Firmus 的范围
GPU 云服务公司自述的 Firmus Cloud 与裸金属能力具名直接同业的核心产品通过广泛云平台提供的核心产品通常不是完整云服务同业分类来自市场画像来源
能源与冷却主张公司自述的液冷与降低电网负载指定证据未知指定证据未知专业供应商的核心主张未保留独立的同口径效率基准
亚太主权定位明确聚焦澳大利亚和新加坡,并扩展至更广泛的亚太地区各家不同;Tracxn 样本中的 Aethir 以新加坡为基地区域平台可用,但未比较主权适配度取决于具体部署,而非整个云平台驻留、控制与认证细节尚未统一
交付与合作伙伴依赖报道称有两个运营站点;计划开发多个站点指定证据未知从既有厂商类别推断其分销广泛,但此处未量化依赖买方或托管方集成CDC 拆分显示 Firmus 的交付配置可以变化

“未知”表示指定的预取证据未确立该单元格,并不意味着该能力不存在。公司主张不是独立基准。

[CP001, CP003, CP005, CP008, CP009, CP010, CP014]
FP001: 按证据编码的竞争定位

序数定位将集成广度与已观察到的交付证据区分开来;这是一张分析图,不是经过测量的市场排名。

X: 垂直集成广度
low: 窄组件
high: 端到端平台
description: Analyst-assigned 1-5 score based only on product scope described in retained sources.
scale: ordinal
Y: 已观察到的交付证据
low: 留存证据有限
high: 留存证据广泛
description: Analyst-assigned 1-5 score based on operational status and scale evidence in retained sources.
scale: ordinal
条目垂直集成广度 (1–5)已观察到的交付证据 (1–5)说明
1. Firmus 5 3 公司自称集成范围广;据报道已有两个运营站点,但交付模式仍在调整,多个站点仍处于规划阶段。
2. CoreWeave 2 4 Tracxn 收录的专注型 GPU 云公开可比公司;未留存完整的设施集成对比。
3. FluidStack 2 3 托管型 GPU 基础设施可比公司,据报道融资规模较大,但留存的运营细节有限。
4. Crusoe 3 3 同时覆盖 AI 数据中心和云业务;留存证据不足以支持更细致的交付对比。
5. 超大规模云厂商 4 5 具备分发优势的广泛公有云既有厂商;该点是类别性定位,并非针对某家公司的判断。
6. 冷却专业厂商 1 2 独立画像中提到的更窄组件替代方案;无法获得可比部署规模。

坐标是分析用序数分数,不是市场份额、基准测试结果或精确测量值。分数仅反映留存证据,并有意对缺失的交付数据施加扣分。

[CP002, CP003, CP008, CP009, CP014, CP016]

3.3 商业证明与护城河持久性

资本渠道是竞争资产,但不能证明利用率或持久利润率。Firmus 称一项 US$10 billion 债务融资将支持其到 2028 年部署最高 1.6 GW;Tracxn 则称,在其宽泛的竞争者群体中,Firmus 按累计融资额排名第二。但 Forbes 指出该品类的结构性不对称:设备供应商在硬件发货时收款,而运营商承担利用率和再融资风险。ABC 也提醒,在招股说明书提交前,IPO 流传数字仍属推测。指定证据中没有 Firmus 价目表、实际单位价格、迁移条款、服务级别救济或标准化的竞争合同对比。因此,在审阅合同、利用率队列和已交付站点经济性之前,投资判断应把能源效率、价格优势和客户锁定视为可验证假设,而不是已证明的护城河。[CP004, CP006, CP007, CP009, CP010, CP011, CP012, CP015, CP016]

定价与套餐对比
替代方案公开价格或计价单位合同模式包含的能力折扣或承诺证据投资判断含义
Firmus指定证据未披露报道称存在长期产能与基础设施合作关系;条款不可得描述了 AI 工厂、云、裸金属和存储未知要求提供每加速器小时或 token 的价格、照付不议条款和 SLA 赔偿
直接新型云厂商指定证据未收集未知GPU 云或托管集群未知获取匹配区域、加速器、预留期限与网络成本报价
超大规模云厂商指定证据未收集此处未比较按需模式与承诺模式广泛云服务及加速器服务未知对比完整工作负载成本、生态切换成本和主权控制溢价
内部建设 / 专业技术栈没有可比的单位价格自有或单独签约的基础设施买方自行组合设施、硬件、冷却和软件未知建模资本开支、电力接入、人员、部署延迟与残值风险

这是缺口表,不是价格基准:没有保留来源提供各替代方案之间标准化的标价或实际价格。

[CP003, CP014, CP015]

3.4 展示项

Chapter 04

04财务

4.1 收入模式可见,但变现不可见

Firmus 展示了两个相互关联的商业界面:Firmus Cloud 让客户访问 GPU 基础设施,Firmus AI Factories 则由公司设计并运营为模块化、垂直整合的基础设施。这支持一个涵盖云计算和专用基础设施的合理产能与服务收入模式,但保留的官方页面没有公布标价、合同计量单位、折扣、收入确认条款或实际费率。3 MW Media Hub 部署和 SEA-LION 工作负载指标说明实体产能和计算已经交付,但二者都不能证明收入或毛利。下方桥接图因此是机制图,不是量化预测。[CI001, CI002, CI003, CI009, CI014]

收入流与证据状态
收入流机制单位公开状态收入质量尽调要求
Firmus Cloud客户访问 GPU 基础设施GPU 小时或合同产能;未披露已披露产品与工作负载交付;收入不可得无法评估经常性收入占比或照付不议保护按产品划分的收入、订单、积压、合同期限和续约数据
专用 AI 工厂产能设计并运营模块化高密度设施MW、机架、GPU 或预留产能;未披露已披露设施和部署模式;合同经济性不可得可能是已签约的基础设施收入,但租赁条款属于非公开信息已签署客户合同、承诺 MW、开始日期和终止权
主权与企业基础设施区域或专用 AI 计算部署项目费或托管服务费;未披露有 SEA-LION 使用证据,但没有披露商业价值交付证明不了价格或利润率客户发票、项目范围、确认收入和直接服务成本
基础设施效率层冷却、电力、电网编排与一体化运营捆绑或单独收费;未披露官方材料描述能力,不代表变现可能改善成本位置,但不是单独验证的收入流定价架构、转移定价、附加率及 Firmus 留存的节省额

公开来源审阅截至 2026-10-20;单位和商业条款标记为未披露而非估算,部署指标不视为收入。

[CI001, CI002, CI003, CI009, CI014]
FI001: 从资本到毛利的收入模型桥接

一条未量化的运营链,展示公开证据在收入和毛利之前止步于何处。

节点 / 连接节点 / 起点终点说明
节点 1 股权与债务融资能力 [funding]为基础设施建设提供资金;已披露承诺不等于可自由支配现金。
节点 2 AI Factory 部署 [factory]土地、电力、冷却、网络和 GPU 系统都需要大量前期投入。
节点 3 可用算力容量 [capacity]Firmus Cloud 和专用设施向客户提供 GPU 与基础设施。
节点 4 客户工作负载活动 [workloads]实验、模型开发、训练和推理都会消耗容量。
节点 5 已确认收入 [revenue]价格、合同单位、计费基础和收入确认政策均未公开。
节点 6 毛利与现金生成 [gross-profit]取决于利用率、电力、硬件、融资和服务交付成本,而这些信息仍未披露。

该序列仅是有证据支持的机制图;其中不包含预测、转化率或假定货币价值。

[CI001, CI002, CI006, CI012, CI013]

4.2 资本市场势头超过运营披露

独立报道普遍指向 2026 年 8 月 $2 billion 股权融资,投后估值超过 $10.5 billion;Forbes 报道此前一年筹得股权资金超过 $3 billion。同一证据集还报告 2026 年 2 月有一项 $10 billion 债务融资。这些是融资事件,不是收入、ARR、已入账积压订单或可用现金。二手数据库对总资本和轮次也存在重大分歧:InforCapital 报告四轮合计 $12.8 billion,而 Silicon Valley Investclub 报告九条记录合计至少 $25.7 billion,并明显重复了若干融资。逐笔交易披露比任一总额更有决策价值。[CI004, CI005, CI006, CI007, CI008, CI016]

定价与变现披露
产品价格或合同单位标价与实际价格折扣或未知项投资判断含义
Firmus Cloud未公开披露未找到标价或实际价格期限、最低支出、利用率收费和折扣未知无法估算每 GPU 小时收入或比较价格实现
专用 AI 工厂未公开披露没有公开的站点级费率或产能合同期限、递增条款、照付不议、转嫁费用和终止权未知无法测试收入持久性与项目融资覆盖
主权或企业部署未公开披露没有公开项目费或托管服务价目表硬件转售、实施和经常性服务组合未知收入确认时点与毛利率组合仍不可得

保留的官方来源描述了产品和部署,但没有发布标价;定价为 null 是证据缺口,不是零值。

[CI001, CI002, CI009]

4.3 利用率与融资结构主导单位经济性

AI 工厂需要先投入土地、电网接入、冷却和稀缺 GPU 硬件。The Australian 指出回报取决于机器保持忙碌,而需求可能不均衡;Forbes 同样认为,设备供应商在硬件发货时收款,运营商却承担利用率和再融资风险。Firmus 声称建设成本和电网负荷更低,但公开证据未量化每 MW 资本开支、每 GPU 资本开支、交付电力成本、维护负担、毛利率或回本期。一项据报道的 $10 billion 融资额度可能提供资金能力,但若没有已提款余额、契约、到期日、利息成本、项目层面担保和已承诺租户经济性,就不能转化为可靠的现金跑道估算。[CI006, CI012, CI013, CI016]

单位经济性与资本充足度评分卡
指标公开数值置信度重要性尽调要求
收入与 ARR不可得高确定规模与经常性收入组合按产品、站点和客户提供经审计的月度收入桥接表
毛利率不可得高检验效率主张能否转化为经济性按站点和产品提供毛利率,并分摊电力、冷却、硬件折旧和支持成本
利用率不可得高核心运营杠杆与下行情景变量按客户群和站点提供可用、已签约与已计费 GPU 小时
每 MW 和每部署 GPU 的资本开支不可得高决定建设效率与融资需求已承诺和已发生的资本开支,包括电网、土地、冷却、网络和硬件
债务额度报道称有 US$10B 额度;已提取余额不可得中带来产能,也带来再融资和契约风险敞口额度协议、提款计划、定价、到期日、契约、抵押品和担保
现金、消耗与现金跑道不可得高决定下一轮融资依赖度与建设抗压能力最新现金余额、非受限现金、月度消耗、资金来源与用途,以及下行情景下的现金跑道

未估算不可得数值。债务数值是报道的额度规模,不是手头现金;置信度标签描述披露置信度,而非业务质量。

[CI006, CI009, CI012, CI013, CI016]

4.4 财务结论与尽调门槛

公开记录支持的是高资本、基础设施驱动的增长故事,而不是已证明的高质量收入故事。Firmus 有可信的已部署计算和机构资本渠道证据,但当前收入、毛利率、利用率、客户集中度、现金、烧钱和债务义务仍属私有信息。低质量二手页面增加了噪音:一处将 Firmus 估为 $5.9 billion,另一篇 AI 生成文章则报道了不同的 A$6 billion 融资快照。因此,投资判断应锚定管理层提供的合同、账单和利用率队列、站点级贡献利润以及资金来源与用途表。在这些材料将融资能力与运营现金创造对上之前,收入质量和资本充足性仍是重大缺口。[CI004, CI010, CI015, CI016]

4.5 展示项

Chapter 05

05产品与技术

5.1 产品定义与客户工作流

Firmus 销售的是加速计算访问,而不是 AI 模型:面向客户的界面是 Firmus Cloud,实体交付系统则是围绕密集 GPU 工作负载设计的模块化 AI Factory。公司将训练、微调、推理和主权工作负载描述为核心任务,基础设施拟把计算、网络、电力、冷却、存储和编排结合起来,不让客户自行拼装这些层。已宣布的 NVIDIA 安排把工作流延伸到面向 AI 原生企业和软件供应商客户的采购与产能获取。但这是一项垂直整合基础设施方案,指定证据中没有找到公开的服务目录、正常运行时间承诺、部署周期或工作负载级定价和性能排期。[CE001, CE002, CE003, CE005]

产品模块与资产矩阵
产品或模块买方或用户公开成熟度信号声称的差异化尽调缺口
Firmus CloudAI 原生企业与 ISV 用户公开营销的访问层在 Firmus 基础设施上提供 GPU 访问服务层级、SLA、定价和利用率
Firmus AI Factory主权、受监管和大规模 AI 工作负载报道称部署规模达到数千枚 GPU围绕电力和成本优化的模块化高密度基础设施独立的站点级验收与基准测试包
HyperCube基础设施运营商与托管计算客户在已宣布的 Batam 设计中被点名与 NVIDIA DSX 蓝图一致的液冷架构生产配置、可维护性与故障率数据
Model-to-Grid 编排Firmus 运营团队与并网设施Firmus 通过供应商发布材料描述的架构串联工作负载、热、GPU、价格与电网信号控制回路设计、遥测验证与实际节省

状态与差异化依据公司或合作伙伴描述;表中没有任何一行意味着已完成独立性能验证。

[CE001, CE002, CE004, CE005, CE008, CE009]
工作流与用例评估
用户任务现有约束Firmus 方案公开受益信号局限
训练、微调或运行推理密集 GPU 集群需要协调计算与数据吞吐基于一体化 AI 工厂基础设施的 Firmus CloudVAST 描述了在大规模 GPU 上支持从训练到推理指定证据中没有工作负载级基准或 SLA
部署主权或受监管 AI数据所在地与基础设施控制限制了对海外云的使用配备安全多租户数据基础设施的境内 AI 工厂合作伙伴材料确认覆盖公共部门与受监管工作负载没有公开的控制映射、审计报告或认证
扩大 AI 原生公司的加速器容量需求可能超过客户签约资本密集型基础设施的能力NVIDIA 支持的采购与云容量模式协议覆盖多代加速器及规划中的容量路线图和已承诺容量不等于实际运行利用率

受益是 Firmus 或合作伙伴提出的方向性主张;缺失的基准、SLA 与控制证据已明确列出。

[CE002, CE005, CE006, CE007, CE015]

5.2 架构与关键依赖

已披露架构将 Firmus 的 HyperCube 液冷平台接入 NVIDIA 的 DSX 蓝图和加速器路线图,同时选用 VAST AI OS 作为数据层。Firmus 还描述了一个 Model-to-Grid 控制框架,把模型行为、GPU 性能、热条件、能源价格和电网信号连接起来。该设计足以识别主要依赖:NVIDIA 负责加速器和参考架构,VAST 负责数据移动和多租户存储,本地电网负责电力,专业团队负责冷却运营。这种整合可能让各层独立演进,但证据主要来自公司和供应商描述;接口边界、故障切换设计、遥测准确性和端到端实测瓶颈都未公开记录。[CE005, CE007, CE008, CE009, CE010]

技术与运营架构
层或组件作用依赖主要风险
客户工作负载训练、微调与推理客户软件与需求利用率与工作负载集中度
Firmus Cloud访问与服务交付界面Firmus 运营与 NVIDIA 容量SLA 与服务经济性未披露
计算与网络加速执行与集群网络NVIDIA 加速器与参考架构供应商集中度与硬件快速迭代
VAST 数据层统一吞吐与多租户数据访问VAST AI OS对供应商与集成的依赖
HyperCube 冷却为密集硬件散热液冷设备与专业技术人员维护复杂度高、具备相关技能的操作人员有限
Model-to-Grid 控制协调工作负载、热、功率与电网信号可靠遥测、电力可用性与控制软件控制验证与节省数据未发布

架构反映公开描述;接口、冗余、恢复目标和已测瓶颈仍未披露。

[CE005, CE007, CE008, CE009, CE012, CE013]
FE001: Firmus 产品架构图

从客户工作负载到云访问、算力、数据、冷却和电网感知控制的公开披露分层。

客户工作负载
AI 训练、微调、推理、主权、企业和 ISV 工作负载。
条目
  • 训练
  • 微调
  • 推理
  • 受监管工作负载
Firmus Cloud
客户可获得 GPU 容量、可靠性、性能和成本效率。
条目
  • GPU 访问
  • 服务运营
  • 多租户交付
NVIDIA 算力与网络
符合 DSX 的基础设施,以及已宣布的 Grace-Blackwell、Vera-Rubin 和 Vera 加速器。
条目
  • 加速器
  • 集群网络
  • DSX 蓝图
VAST 数据基础
统一数据访问,旨在维持分离式 GPU 集群之间的吞吐量。
条目
  • 数据移动
  • 存储
  • 多租户访问
HyperCube 物理平台
面向高密度 AI 工作负载的模块化液冷基础设施。
条目
  • 液冷
  • 高密度模块
  • 配电
Model-to-Grid 控制
连接工作负载行为、热状态、能源价格和电网信号的优化框架。
条目
  • 遥测
  • 热控制
  • 电网响应

仅为概念层级图;不代表未披露的接口、性能或部署完整性。

[CE001, CE005, CE007, CE008, CE009, CE015]

5.3 成熟度、部署与路线图

该产品已超出概念阶段架构:Firmus 列出新加坡改造项目和早期平台里程碑,VAST 与 TechEDT 报告了千卡级 AI 工厂的活跃部署。2026 年 6 月 Batam 公告新增一座 360 MW 园区,并签署覆盖 2027 和 2028 年最高 170,000 个 NVIDIA 加速器的协议,但这些是已宣布的产能和路线图数字,不证明整个资产已安装、验收或投入使用。一篇信誉较低的独立文章还将 Firmus 描述为一家在加密货币挖矿起点之后仍处于相对早期的数据中心运营商。投资判断应因此把运营站点与已签约或计划产能分开,并要求逐站点提供投产、验收、利用率和事故数据。[CE004, CE006, CE016]

5.4 差异化、信任与技术风险

Firmus 的差异化主张是系统级效率:液冷、闭环用水设计、密集模块化基础设施、感知芯片的编排,以及对电网条件的响应。The Australian 独立解释了这些特征为何重要,但也指出运营复杂度、稀缺专业技术人员、建设延迟、供应约束和技术标准变化等执行风险。Forbes 补充称,即使供应商在交付时收款,运营商仍承担长期利用率和再融资风险。公开材料提到境内数据控制和安全的多租户工作负载,但审阅来源未披露认证、审计报告、隐私控制、正常运行时间历史、恢复目标或事故响应。技术论点合理且有差异化;控制环境和量化优势仍是尽调事项,而非已验证的强项。[CE011, CE012, CE013, CE014, CE015]

5.5 展示项

Chapter 06

06客户

6.1 客户细分与采购中心

Firmus 面向多个买方群体,而不是单一同质客户群。独立市场画像将企业、政府机构、教育机构、初创公司和成长型公司列为目标客户,同时列出 Meta、OpenAI、AI Singapore、Fireworks、HTX 和 Rafay 等重要关系。最强的公开证明集中在 AI 模型开发商以及主权或研究工作负载:Firmus 自己描述了与 Meta 和 OpenAI 的专用产能协议,官网则把 AI Singapore 的 SEA-LION 工作展示为已部署案例。现有材料能识别用户和用例,但不能识别各细分市场的签约实体、付款方、合同金额、采购周期或收入贡献。因此,细分经济性仍是尽调问题,而不是客户组合中已证明的特征。[CU002, CU003, CU005, CU009, CU010, CU016]

客户分层与证据质量
客户群买方 / 付款方用户与用例公开规模信号战略价值证据缺口
超大规模模型开发商模型开发商或基础设施采购团队;付款方未披露使用专用 GPU 计算的 AI 研究人员与平台团队已点名 Meta 和 OpenAI;超过 900 MW 覆盖全部客户,而非该客户群可能成为锚定租户,并验证高密度工作负载没有合同金额、利用率、期限、续约或客户级 MW 拆分
主权、政府与教育政府机构、国家 AI 项目或教育机构区域语言模型开发与主权 AI 工作负载AI Singapore 的 SEA-LION 案例报告了 200 次实验和 100 个候选模型支撑区域化与主权计算定位没有定价、经常性收入、留存或完整客户名单
企业、初创与成长型公司企业技术买方或初创公司基础设施团队训练、推理、裸机与云工作负载独立画像确认了该客户群,但未披露客户数量将需求扩展到少数已点名模型开发商之外没有已点名的客户群规模、生产状态、收入区间或满意度证据

客户群结合公司发布的部署证据与独立市场画像描述;商业指标为 null 表示未披露,而非估算值。

[CU002, CU003, CU005, CU006, CU007, CU009, CU010]

6.2 采用与部署证明

公开采用证据涵盖合同、建设里程碑、投产和已执行工作负载,但这些阶段不能混为一谈。Firmus 称总签约产能超过 900 MW,但这一总量没有披露其中多少已运营、已占用、已计费或归属于各客户。更具体的部署信号包括已投产并移交以运行客户模型的 Brooklyn GB300 集群、已移交给 HyperCube 安装的印度尼西亚数据大厅,以及正在运营的新加坡 3 MW Media Hub 站点。AI Singapore 案例还报告了 200 次实验、100 个候选模型和 99.072 PFLOP/s 总计算量,补充了工作负载证据。这些指标证明了活动,而不是经常性收入或客户留存。[CU001, CU004, CU005, CU006, CU007, CU008, CU014, CU015]

客户采用与部署轨迹
指标或里程碑报告值证据日期来源置信度含义缺失分母
全部客户的签约容量超过 900 MW页面访问于 2026-10-20Firmus LinkedIn中表明整个客户组合存在可观的签约需求运营中、已计费、已利用以及客户级容量均未披露
Brooklyn GB300 集群已安装、完成调试并交付运行客户模型页面访问于 2026-10-20Firmus LinkedIn中显示已交接至工作负载执行,而不只是停留在建设阶段集群规模、客户身份、利用率与收入均未披露
Media Hub Singapore3 MW GPU AI 工厂未标日期的当前网站Firmus About中显示新加坡已有运营中的部署界面客户入驻率、工作负载组合与商业条款均未披露
SEA-LION 工作负载200 次实验;100 个候选模型;99.072 PFLOP/s 总计算量未标日期的当前网站Firmus homepage中显示可量化的研究工作负载活动时间段、付费合同金额、重复使用与续约均未披露

数值由来源报告,使用的单位和阶段各不相同;它们不是转化漏斗,也不能证明收入、利用率或留存。

[CU001, CU004, CU006, CU007, CU008, CU014]
已点名客户证明表
客户客户群部署或用例生产环境还是试点已报告结果局限
Meta超大规模模型开发商东南亚 Firmus AI 工厂在 NVIDIA DSX 平台上的 GPU 计算容量已宣布协议;运营状态未披露未披露客户特定结果证据池中没有期限、容量拆分、利用率、收入或独立客户确认
OpenAI超大规模模型开发商来自马来西亚两座 AI Factory 站点的专用 AI 计算容量已宣布多年期协议;运营状态未披露计入公司报告的超过 900 MW 组合总量证据池中没有客户级 MW、起始日期、收入、利用率或客户直接声明
AI Singapore主权与研究项目用于 SEA-LION 区域语言模型开发的计算基础设施Firmus 案例研究报告了工作负载活动200 次实验、100 个候选模型和 99.072 PFLOP/s 总计算量证据池中没有合同金额、续约、满意度或独立客户撰写的证据

仅列出指定证据池中有明确部署或工作负载的已点名关系;不代表完整客户群体。

[CU001, CU002, CU003, CU005, CU006, CU007, CU008, CU010]
FU001: 公开可见的客户旅程与证据断点

公开证据追踪了从资格评估到工作负载执行的过程,但续约和扩张情况仍未披露。

阶段顺序仅用于分析;不代表转化率、各阶段客户群规模相等,也不代表每个点名客户都完成了所有阶段。

[CU001, CU004, CU005, CU009, CU012, CU015]

6.3 持久性、扩张与集中度

证据集不支持留存结论。没有来源提供净收入留存、毛收入留存、流失、续约率、合同条款、积压订单转化、客户级利用率、满意度或收入集中度。所有客户合计超过 900 MW 的披露签约产能具有商业意义,但其分母和客户分布缺失。Forbes 的分析直接指出运营商风险:AI 基础设施运营商承担长期利用率和再融资暴露,而供应商在设备发货时收款。一份独立公司画像同样报告没有公开收入或盈利数字。因此,投资者应将具名客户标识和产能协议视为需求验证,而不是持久单位经济性的证明,并在判断扩张前要求合同和队列数据。[CU001, CU011, CU012, CU013]

6.4 展示项

Chapter 07

07风险

7.1 风险排序与投资含义

最大的剩余暴露是资本和交付执行。2026 年 4 月的一篇报道将当时计划中的全国项目规模定为 A$73.3 billion;三份审阅来源则描述了一项 US$10 billion 债务融资。这些数字代表融资规模,不证明所有资本仍可用,也不证明修订后的项目组合已获得全额资金。设备供应商收款后,运营模式仍让 Firmus 承担利用率和再融资风险。这种不对称很重要,因为后续画像称,在部署约 42 MW 后,拟议的 CDC 关联 1.6 GW 扩张将不再推进。因此,投资判断应分别评估每个园区和已承诺客户,而不是把已宣布的总产能视为已去风险的组合。[CR001, CR002, CR003, CR008]

FR001: 剩余风险热力图

对投资论点影响最大的六项风险,从发生可能性、影响、缓释成熟度和剩余严重性四个维度进行定性矩阵分析。

各类别是基于所引事实的分析师定性判断,不是公司评级、概率或量化损失模型。

[CR001, CR003, CR004, CR007, CR009, CR010, CR011, CR012, CR013]

7.2 监管、电网与环境执行

站点交付暴露于审批顺序和电力可用性。审阅画像称 Bell Bay 的审批正在上诉,Wesley Vale 在重大项目决策前需要补充信息,Hydro Tasmania 尚未决定一项涉及塔斯马尼亚三个站点约 444 MW 的申请。这些是可识别的流程依赖,不是审批或供电必然失败的证据,但任何一处延迟都可能推迟通电、收入和融资排期。Firmus 已发布能源和用水承诺,包括干式冷却和未来报告。这些措施提高了缓解机制的成熟度,但指定证据没有提供运营和计划站点的独立审计结果。审批文件、已执行电网合同以及实测用水和用电数据仍是决定性尽调材料。[CR009, CR010, CR011, CR012, CR013, CR015, CR016]

监管 / 法律风险登记表
规则、审批或事项司法辖区当前状态可能性严重性缓释成熟度剩余风险尽调路径
Bell Bay 开发审批上诉塔斯马尼亚州,澳大利亚已附条件批准;据报道已有上诉提交至 TASCAT进行中高上诉解决前有限进度不确定性高获取上诉文书、时间表、条件及管理层关键路径分析
Wesley Vale 重大项目与开发审查塔斯马尼亚州,澳大利亚已提交申请;据报道州政府要求补充信息进行中高早期阶段进度与范围不确定性高审查申请、补充信息请求、回复材料及审批依赖
多站点电网供电与接入塔斯马尼亚州,澳大利亚Hydro Tasmania 据报道正在评估约 444 MW 的申请,尚无决定不确定高已披露部分签约并网与成本风险高按站点核实已签署的供电、接入、限电、定价及里程碑义务
能源与用水预期澳大利亚公司已发布政策,并承诺未来进行年度报告持续合规中政策阶段在实测表现完成审计前为中等获取监管机构往来文件,以及站点层面的经审计能源与用水结果

登记表仅基于截至 2026-10-20 的指定公开来源;可能性和严重性是分析师判断,证据池中没有监管机构或法院记录。

[CR009, CR010, CR011, CR012, CR013, CR015]

7.3 运营、财务与合作伙伴依赖

Firmus 的垂直整合雄心很大:公司称平台从冷却系统延伸到电网编排软件,而 NVIDIA 据报道既是投资者,也是主要芯片供应商。一份公司发布的联合公告描述了持续至 2034 年的合作、360 MW Batam 园区、收入分成和信用支持,并预计已承诺包销带来六年 US$25 billion 至 US$30 billion 收入。这些安排可以支持采购和需求,但也集中暴露了技术路线图、交易对手和合同履约风险。撤回计划中的 ASX 上市并转向私募融资,进一步提高了融资条款、契约余量和里程碑挂钩提款可用性的权重。已筹融资、债务能力或预计包销收入都不应视为不受限制的现金或已实现收入。[CR002, CR004, CR005, CR006, CR007, CR014]

运营 / 质量 / 安全风险登记表
失效模式可能性严重性缓释成熟度剩余风险未解决缺口
资本密集型建设中的施工与融资延期中高高据报道具备大额债务与股权融资能力,但条款未审阅高按站点列示的资金来源与用途、应急资金、提款条件及完工担保
容量利用不足或再融资压力不确定高已宣称有承购,但合同质量与集中度属于非公开信息高已签约与可用 MW、照付不议条款、利用率及偿债结构
冷却、电力或用水表现未达公司目标不确定高政策、干式运行设计与遥测发布均为公司表述中高独立的正常运行时间、电力使用、用水、热学与维护记录
硬件、冷却与电网编排之间的一体化平台失效不确定高公司报告了内部平台覆盖面与基准认可中高架构审查、事故历史、冗余测试、保修与保险

定性评估基于指定证据;没有公开事故日志、经审计的正常运行时间序列、安全认证材料包或站点级运营成本记录。

[CR001, CR002, CR003, CR006, CR012, CR013, CR014, CR015]
合作伙伴 / 依赖风险登记表
依赖项交易对手方角色集中度信号失效情景严重性缓释措施剩余风险
加速器供应与参考架构NVIDIA主要芯片供应商、投资方与平台合作伙伴战略集中度高分配、路线图、定价或架构变化拖延容量交付高据报道有长期合作关系与信用支持没有替代供应与迁移计划时为高
Batam 采购与需求支持NVIDIA 与 AI 原生客户收入分成、信用支持与已宣称的承购合同细节未披露客户违约或信用支持减弱,影响项目融资高公司报告了已承诺承购与一致的经济安排在审阅合同与担保前为高
电网能源可用性Hydro Tasmania 与其他能源供应商为规划站点提供供电与接入站点特定的实体依赖审批、接入或限电延迟送电高公司能源政策与部分签约对仍在评估中的申请而言为高
园区扩张合作伙伴的持续性CDC Data Centres此前与全国扩张计划相关联据报道发生重大战略变化替代站点或交付能力落后于原计划中高Firmus 正在推进自有项目及其他区域项目在替代计划时间表完成核对前为中高
成长资本获取私人贷款方与股权投资者建设与扩张融资大额债务融资安排加上已撤回的 IPO提款条件收紧或私募轮估值低于计划高据报道有多家机构支持者没有契约与流动性披露时为高

依赖登记表部分依据公开证据;融资承诺不视为可用现金,预计承购也不视为已确认收入。

[CR002, CR004, CR005, CR006, CR007, CR008, CR011]

7.4 监测指标、止损标准与尽调路径

以下是拟议的投资者控制项,不是公司发布的阈值。如果核心园区在签约交付日前没有最终且未被上诉的审批和有约束力的电力安排;如果已承诺资金和获准提款不足以覆盖剩余建设和预备金;或客户合同缺少具备偿付能力且可执行的照付不议保护,应暂停投资判断。将 NVIDIA 供应、信用支持或平台经济性的重大恶化视为依赖触发器。在接受运营缓解措施已经成熟之前,要求提供正常运行时间、安全事故、保险、用水和用电表现的独立证据。若站点通电、融资和签约需求同时延误,论点就会破裂,因为这种组合会把排期风险变成流动性和再融资风险,而不是可恢复的建设延迟。[CR003, CR004, CR005, CR007, CR009, CR010, CR011, CR012, CR013, CR015]

7.5 展示项

Chapter 08

08估值

8.1 融资与估值背景

指定材料中最新的独立报道称,Firmus 在 2026 年 8 月获得 $2 billion 股权融资承诺,投后估值超过 $10.5 billion。GlobalData 的 Verdict 报告还称,该交易使此前一年累计筹得的股权资金超过 $3 billion。Forbes 将 2026 年 4 月估值标为 $5.5 billion,意味着标题估值在四个月内大致翻倍。SetterVC 独立展示了 4 月和 8 月的估值,但明确警告其融资和估值数据属于估算。这些估值显示融资轨迹,而非内在价值:没有一项提供收入倍数、现金流收益率或经审计资产负债表桥接。[CV001, CV002, CV003, CV010]

建议摘要
维度评估证据依据决策含义
建议跟踪 / 继续研究融资信号强,但公开运营经济性不足不应给出无条件买入建议
置信度中低融资估值标记得到交叉印证;价值驱动因素与证券条款尚未确认明确将结论保持为有条件结论
风险评级高资本密集度、已报告债务、利用率风险与再融资敞口要求下行保护与里程碑门槛
估值立场公开信息不足以支撑高于 $10.5 billion 的估值判断审阅材料中没有经审计的收入、EBITDA、现金流、利用率或优先权数据完成私人尽调后再确定价值
入场纪律对价格与条款敏感4 月 $5.5 billion 的估值标记显示,在公开运营经济性尚未显现前倍数已快速扩张寻求折价、分阶段投入资本或保护性条款

评估使用截至 2026-10-20 的八个指定来源;运营与证券数据缺失被视为缺口,而非进行估算。

[CV001, CV002, CV004, CV008, CV011, CV012]

8.2 正论、反论与投资判断缺口

正论建立在大规模股权和债务资金渠道、已声明的 1.6 GW Southgate 雄心,以及 Firmus 声称能降低基础设施资源需求的垂直整合平台上。官网还列出两个新加坡部署项目,这比纯概念开发故事更有实质。当前价格下,反论更强:Forbes 强调设备供应商收款后,运营商仍承担利用率和再融资风险;审阅证据则报告了一项 $10 billion 债务融资。最重要的是,指定的八个来源没有提供经审计收入、EBITDA、现金流、利用率、客户集中度或优先权条款。缺少这个分母,使公开市场或项目融资估值无法站稳。[CV004, CV005, CV006, CV007, CV008, CV011]

正方论点与反方论点
立场论点证据哪些信息会改变判断
投资论点大额战略股权支持据报道,公司以高于 $10.5 billion 的估值完成了 $2 billion 轮融资核实已到账现金、投资人权利及剩余承诺
投资论点大规模开发目标据报道,Southgate 计划到 2028 年达到最高 1.6 GW分别核实已通电、已签约和处于建设阶段的容量
投资论点已有一定运营足迹Firmus 将 Loyang 和 Media Hub 列为新加坡部署项目获取站点级利用率、收入和服务水平历史
反投资论点估值上涨快于已披露的经营表现据报道,估值从 4 月的 $5.5 billion 上升至 2026 年 8 月的 $10.5 billion 以上经审计的增长和利润率证据可能支撑这次重估
反投资论点杠杆放大交付风险Forbes 和 SetterVC 报称存在 $10 billion 债务融资额度将契约和到期日与已签约现金流匹配
反投资论点核心估值分母缺失已审阅来源未提供经审计收入、现金流、利用率、客户集中度和优先权信息提供完整的财务、合同、债务及资本结构数据室

关于公司平台和部署的表述由公司发布;融资和风险观察均归因于所列第三方来源。

[CV001, CV002, CV004, CV005, CV006, CV007, CV008, CV011]

8.3 乐观、基准与悲观框架

因此,情景分析应使用里程碑,而不是制造虚假的精确度。乐观情景要求已签约需求转化为有收入支撑的占用率、正向单位经济性和偿债覆盖,并由交易所申报提供治理和经审计财务数据。基准情景假设资本渠道和建设继续,而经济性仍属私有信息;在这一条件下,没有公开证据支持在没有大幅折价、分阶段出资或保护性条款的情况下支付 8 月标题估值。悲观情景则结合利用率不足、再融资压力和倍数压缩。报道中的 4 月 $5.5 billion 估值可作为有用的下行参考,约比 $10.5 billion 低 48%,但它不是估值底线或内在价值估算。[CV002, CV004, CV008, CV011, CV013, CV014]

乐观、基准和悲观情景
情景假设估值 / 回报逻辑主要风险概率信号
乐观收入支撑的产能占用、正向单位经济性、债务偿付有保障,且 IPO 披露经审计只有这些里程碑实现后,高于据报道 $10.5 billion 的估值才具备可验证性建设成本或需求不及预期会抹去经营杠杆核实从已签约到已通电的转化,以及经审计现金生成
基准资本获取和建设继续推进,但经营经济性仍未公开要求相对 8 月估值有明显折价、分阶段投入资本或保护性条款融资进展掩盖了利用率低或客户集中的问题站点级利用率、利润率和客户数据仍不可得
悲观利用率不足、再融资压力和基础设施估值倍数压缩4 月的 $5.5 billion 估值可作为下行参考,约比 $10.5 billion 低 48%,但不是底线契约压力、稀释或商业化延迟占用率低、未达成通电里程碑或再融资条款不利

约 48% 的下行参考值根据据报道的 $5.5 billion 和 $10.5 billion 估值计算;由于经营输入缺失,未赋予情景数值概率或内在估值区间。

[CV002, CV004, CV008, CV011, CV013, CV014]

8.4 决策、退出准备度与尽调门槛

建议为 track/research-more,置信度中低、风险高。融资名单和部署雄心值得继续研究,但不能回答新投资者能否从高于 $10.5 billion 的入场估值获得可接受的风险调整回报。公开材料讨论了 IPO 雄心或时间表,但指定证据既没有交易所申报,也没有完成上市公告。只有在管理层提供经审计财务数据、站点级签约与通电产能、利用率和客户集中度、债务契约、完全稀释后的股权结构表、清算优先权,以及标准化园区里程碑和资本开支排期后,决策才应改变。无法提供这些材料,或出现占用率疲弱、再融资承压的证据,都属于论点破裂信号。[CV001, CV004, CV008, CV012, CV013]

FV001: 从规模目标到决策门槛的推荐逻辑

战略资本和开发目标支持持续跟踪,但杠杆和缺失的运营证据不支持无条件买入。

节点 / 连接节点 / 起点终点说明
节点 1 据报道的规模目标 [reported-scale]Southgate 的目标是到 2028 年达到最高 1.6 GW。
节点 2 资本验证 [capital-validation]$2 billion 轮融资意味着投后估值高于 $10.5 billion。
节点 3 杠杆敞口 [leverage]第三方来源报告存在 $10 billion 债务融资额度。
节点 4 缺失的估值分母 [evidence-gap]无法获得经审计的经营经济性、利用率、集中度和担保条款。
节点 5 尽调门槛 [diligence-gates]核实合同、站点经济性、契约、股权结构表、优先股堆叠和 IPO 申报文件。
节点 6 持续跟踪 / 继续研究 [recommendation]只有在价格可支撑且运营证据得到核实后,才重新评估。

该流程为定性判断,不为未披露的运营输入赋予概率、转化率或价值。

[CV001, CV004, CV005, CV008, CV012, CV013]

8.5 展示项

免责声明

本报告基于公开可得信息,不构成投资、法律、税务或财务建议。

证据索引

结论
编号陈述可信度来源
CO001 Firmus Grid Limited trades as Firmus Technologies and identified Sydney, Australia as its headquarters in August 2026. 中 SO001
CO002 Firmus presents its business as modular AI factories paired with GPU cloud services. 中 SO002, SO006
CO003 Oliver Curtis and Tim Rosenfield are Firmus's co-chief executives and founders. 中 SO003, SO007
CO004 Oliver Curtis previously served one year in prison for insider trading. 中 SO003, SO007
CO005 Firmus announced full commitments for a USD 2 billion strategic equity round involving Coatue, NVIDIA, Blackstone-managed funds, and Jane Street on 7 August 2026. 中 SO001, SO004
CO006 Firmus stated that the August 2026 round valued the company above USD 10.5 billion post-money. 中 SO001, SO004
CO007 Firmus withdrew its proposed ASX initial public offering on 9 October 2026 and said it would pursue private-capital alternatives. 中 SO003, SO004, SO007, SO008
CO008 ABC reported that Firmus's seven-site, four-country portfolio had two operating sites and five sites under development in October 2026. 中 SO003
CO009 Firmus says it was incorporated in Australia in 2019. 中 SO005
CO010 Firmus says its initial HyperCube research and development was completed in 2020. 中 SO005
CO011 Firmus says it functionally completed a scale-up immersion-cooled data hall powered by renewable energy in 2021. 中 SO005
CO012 Firmus says it completed a fully integrated AI compute platform in 2022. 中 SO005
CO013 Firmus says SMC by Firmus went live in Singapore for sovereign AI services in 2023. 中 SO005
CO014 Firmus says its first live grid-integration project with Eaton was deployed in Tasmania in the first quarter of 2024. 中 SO005
CO015 Firmus says SemiAnalysis ranked it among the top three global GPU clouds in the fourth quarter of 2024. 中 SO005
CO016 Firmus says it announced Project Southgate, its Tasmania AI campus program, in the third quarter of 2025. 中 SO005
CM001 Firmus defines its core offer as modular AI factories and GPU cloud capacity integrating compute, networking, power, and cooling rather than general-purpose cloud or conventional colocation. 中 SM002, SM004
CM002 Firmus claims its AI-factory design reduces power, water, space, construction cost, and grid draw, but the retained evidence does not independently quantify those benefits. 中 SM002
CM003 Firmus identifies a 3 MW Singapore GPU AI-factory retrofit in a former basement car park as part of its deployed footprint. 中 SM002, SM004
CM004 Startup Daily reported a Firmus plan for up to A$73.3 billion of investment and 1.6 GW of compute across five Australian sites by 2028. 中 SM008
CM005 Forbes separately reported that Project Southgate targets 1.6 GW of AI compute across five sites by 2028. 中 SM005
CM006 Australia Times reported the same 1.6 GW by 2028 capacity target while noting that grid-connection issues remain a consideration. 中 SM006
CM007 Tech Insider cited an estimated A$33.6 billion Australian cloud market in 2026 growing about 18% year over year without exposing the underlying methodology. 低 SM003
CM008 Tech Insider cited about A$946 million of Australian AI-specific cloud-infrastructure spending in 2026 and 128.4% growth without an auditable category definition. 低 SM003
CM009 Startup Daily identified enterprise, education, government, startup, and scaleup users as intended customers for capacity delivered through Firmus AI Cloud. 中 SM008
CM010 Firmus's sovereign-compute positioning is most relevant to government and regulated buyers that require regional workload control, although public procurement and budget evidence is limited. 中 SM002, SM003, SM004
CM011 The Urban Developer reported that the proposed A$2.1 billion, 52 MW Wesley Vale project faces a December 15 complete-proposal deadline or withdrawal. 中 SM007
CM012 The Urban Developer reported a Tasmanian pipeline comprising 52 MW at Wesley Vale, 104 MW at St Leonards, and 288 MW at Bell Bay, with a combined 444 MW long-term energy request under assessment. 中 SM007
CM013 The Wesley Vale proposal has drawn local opposition focused on water security and noise. 中 SM007
CM014 Firmus's announced project values and capacity targets imply high capital intensity and leave utilization and refinancing risk with the infrastructure operator. 中 SM005, SM008
CM015 On October 6, 2026, the Australian Financial Review reported that Firmus was targeting at least A$7 billion in an IPO at a A$43.7 billion market capitalization. 中 SM001
CM016 The Urban Developer subsequently reported that Firmus withdrew the planned IPO on October 9, citing market volatility and prevailing market conditions. 中 SM007
CP001 Firmus describes its offering as a modular AI-factory and cloud stack spanning compute, network, power, cooling, bare metal and storage. 中 SP005, SP006
CP002 Tracxn reports 363 active Firmus competitors and names CoreWeave, Tuya and Together AI as the top three in its broad taxonomy. 中 SP001
CP003 AI Market Watch identifies Submer, LiquidStack and KyotoCooling as Firmus competitors at the specialist cooling and infrastructure layer. 低 SP008
CP004 Tracxn ranks Firmus second by total funding and twenty-first overall within its broad competitor cohort. 中 SP001
CP005 Firmus claims vertical co-design and liquid cooling reduce cost, power, water and space requirements relative to conventional infrastructure. 中 SP005, SP006
CP006 Firmus says its US$10 billion debt facility will finance a Project Southgate rollout targeting up to 1.6 GW of infrastructure through 2028. 中 SP003
CP007 Forbes argues that AI-infrastructure operators retain utilization and refinancing risk after equipment suppliers are paid. 中 SP007
CP008 The Australian reports that Firmus and CDC ended their proposed Southgate development partnership after only a 42 MW Melbourne centre had been completed. 中 SP004
CP009 The Australian reports that Firmus operated one Melbourne data centre and one Singapore data centre as of October 2026. 中 SP004
CP010 ABC reports that Firmus planned seven AI data centres across four countries over the following 24 months. 中 SP002
CP011 ABC reports that the prospectus had not been lodged and characterized circulated IPO valuation figures as conjectural. 中 SP002
CP012 Australia Times notes that grid-connection and local-energy constraints remain relevant to high-density Australian AI infrastructure. 低 SP009
CP013 Firmus's integrated model can reduce vendor coordination while increasing exposure to site, power and delivery-partner dependencies. 中 SP004, SP006
CP014 A market profile lists AWS, Microsoft Azure, Google Cloud and Oracle Cloud Infrastructure as broad-platform alternatives to Firmus. 低 SP010
CP015 Firmus's reviewed public product pages describe capabilities but do not publish standardized unit pricing, discounts, migration terms or service-level remedies. 中 SP005, SP006
CP016 The retained evidence supports high vertical-integration breadth for Firmus but only moderate delivered-scale proof relative to its announced pipeline. 中 SP004, SP005, SP006
CI001 Firmus says it designs and operates modular, vertically integrated AI Factories that optimize cost and power across the infrastructure stack. 中 SI006
CI002 Firmus describes Firmus Cloud as a product providing access to GPU infrastructure for AI workloads. 中 SI006
CI003 Firmus identifies its Singapore Media Hub as a 3 MW GPU AI Factory. 高 SI003, SI006
CI004 Independent sources report that Firmus raised $2 billion of equity on August 7, 2026 at a post-money valuation above $10.5 billion. 高 SI001, SI007, SI008
CI005 Forbes reports that Firmus raised more than $3 billion of equity during the year preceding August 2026. 中 SI007
CI006 Independent sources report a $10 billion debt facility led by Blackstone in February 2026. 高 SI004, SI007, SI008
CI007 InforCapital reports $12.8 billion of total capital across four rounds, comprising $2.8 billion of equity and $10 billion of debt. 中 SI004
CI008 Silicon Valley Investclub reports at least $25.7 billion across nine entries, a total inflated by visibly duplicated financing rows and inconsistent with InforCapital. 低 SI008
CI009 The retained official pages publish no list price, contract unit, realized rate, revenue, gross margin, profitability, utilization, cash, burn, or runway. 高 SI003, SI006
CI010 A low-reputation secondary article described Firmus as a $5.9 billion company while centering the co-CEO's prior insider-trading conviction. 低 SI002
CI011 The Australian reported that Firmus announced about A$830 million of new equity funding during 2025. 中 SI009
CI012 The Australian reports that AI factories require heavy upfront investment and that returns depend on keeping costly machines utilized. 中 SI009
CI013 Forbes reports that AI-infrastructure operators bear utilization and refinancing risk while equipment suppliers are paid when hardware ships. 中 SI007
CI014 Firmus's SEA-LION case study reports 200 experiments, 100 candidate models, and 99.072 PFLOP/s of total compute without disclosing commercial value. 中 SI006
CI015 Australia Times reported an A$500 million equity raise at an approximately A$6 billion post-money valuation but disclosed that its article was AI-generated and may contain inaccuracies. 低 SI005
CI016 The disclosed equity rounds and debt-facility size cannot establish current runway because capital raised and financing capacity are not equivalent to unrestricted cash available. 中 SI001, SI004, SI007, SI008
CE001 Firmus describes its offering as Firmus Cloud plus modular, vertically integrated AI Factories. 中 SE007
CE002 Firmus markets its cloud as GPU access designed around reliability, performance, and cost efficiency. 中 SE007
CE003 The Australian characterizes Firmus as an operator of AI-focused data centers rather than a developer of AI software or models. 中 SE005
CE004 VAST and TechEdt report that Firmus is operating AI-factory deployments at the scale of thousands of GPUs. 中 SE001, SE004
CE005 Firmus announced that its HyperCube liquid-cooled platform would be integrated with NVIDIA DSX blueprints at the Batam campus. 中 SE003
CE006 Firmus announced a 360 MW Batam campus and an agreement covering up to 170,000 NVIDIA accelerators across 2027 and 2028. 中 SE003
CE007 VAST says Firmus selected VAST AI OS as the foundational data layer for its next generation of Asia-Pacific AI factories. 中 SE001, SE004
CE008 Firmus describes Model-to-Grid as an optimization framework connecting model behavior, GPU performance, thermal management, and grid conditions. 中 SE001, SE004
CE009 The disclosed AI-factory model coordinates compute, networking, cooling, storage, and orchestration as one system while allowing layers to scale independently. 中 SE001, SE004
CE010 Firmus says liquid cooling and custom infrastructure reduce energy and water use for high-density compute. 中 SE007
CE011 The Australian reports that Project Southgate plans closed-loop cooling intended to reduce water use relative to evaporative cooling. 中 SE005
CE012 The Australian identifies liquid-cooling maintenance complexity and a limited pool of experienced technicians as scale risks. 中 SE005
CE013 The Australian identifies construction delays, supply-chain constraints, and changing technology standards as risks to large AI-factory projects. 中 SE005
CE014 Forbes argues that AI-factory operators retain long-duration utilization and refinancing risk after equipment suppliers are paid. 中 SE008
CE015 The reviewed public material mentions local data control and secure multi-tenant workloads but does not disclose certifications, control audits, uptime history, recovery objectives, or incident response. 中 SE001, SE004, SE006, SE007
CE016 A low-reputation independent article describes Firmus as relatively early in its evolution as a data-center operator after pivoting from cryptocurrency mining. 低 SE002
CU001 Firmus says total contracted capacity across all customers exceeds 900 MW. 中 SU002
CU002 Firmus announced agreements with Meta for GPU compute capacity at Southeast Asian AI factories built on NVIDIA's DSX platform. 中 SU002
CU003 Firmus announced a multi-year OpenAI agreement for dedicated AI compute capacity from two Malaysian AI Factory sites. 中 SU002, SU008
CU004 Firmus reported that its Brooklyn GB300 cluster was installed, commissioned, and handed over to run customer models. 中 SU002
CU005 Firmus presents its infrastructure support for AI Singapore's SEA-LION model-development work as a customer case study. 中 SU006, SU008
CU006 The Firmus SEA-LION case study reports 200 experiments completed. 中 SU006
CU007 The Firmus SEA-LION case study reports 100 candidate models produced and evaluated. 中 SU006
CU008 The Firmus SEA-LION case study reports 99.072 PFLOP/s of total compute. 中 SU006
CU009 AI Market Watch identifies enterprises, government agencies, education institutions, startups, and scaleups as Firmus target customers. 低 SU008
CU010 AI Market Watch lists OpenAI, Meta, Fireworks, AI Singapore, HTX, and Rafay among Firmus's notable customers. 低 SU008
CU011 Dynamic Export reported a multi-year capacity agreement with an unnamed global hyperscale customer at Project Southgate. 中 SU001
CU012 Forbes assessed that AI-infrastructure operators carry long-duration utilization and refinancing risk while equipment suppliers are paid on delivery. 中 SU004
CU013 Silicon Valley Investclub states that public sources in its profile do not disclose Firmus revenue or profitability. 低 SU007
CU014 Firmus describes its Media Hub site in Singapore as a 3 MW GPU AI factory. 中 SU003
CU015 Firmus reported that an Indonesian data hall had been handed over so HyperCube installation could begin before compute delivery. 中 SU002
CU016 Australia Times reported Firmus management's view that its platform addresses accelerating demand for inference and training compute. 低 SU005
CR001 In April 2026, The Next Web reported a projected A$73.3 billion build cost for the then-planned national Project Southgate program. 中 SR001
CR002 Three assigned independent sources report that Firmus secured a US$10 billion debt facility led by Blackstone funds in February 2026. 中 SR001, SR002, SR008
CR003 Forbes argues that AI infrastructure operators retain utilization and refinancing risk after equipment suppliers are paid. 中 SR008
CR004 The Next Web reports that NVIDIA is both a Firmus investor and the primary chip supplier for its facilities. 中 SR001
CR005 A company-issued syndicated announcement says the NVIDIA partnership runs through 2034 and is anchored by a 360 MW Batam campus. 中 SR004
CR006 Firmus expects US$25 billion to US$30 billion from committed offtake agreements during the first six years of the NVIDIA partnership. 低 SR004
CR007 Certified Strategic reports that Firmus withdrew its planned ASX float on 9 October 2026 and is seeking private funding. 中 SR002
CR008 Certified Strategic reports that the announced CDC-linked 1.6 GW national rollout will not proceed after about 42 MW was deployed in Melbourne. 中 SR002
CR009 Certified Strategic reports that appeals concerning Bell Bay's approval and one approval condition are before the Tasmanian Civil and Administrative Tribunal. 中 SR002
CR010 Certified Strategic reports that Tasmania requested more information before deciding whether to declare Wesley Vale a major project. 中 SR002
CR011 Certified Strategic reports that Hydro Tasmania was assessing a request associated with about 444 MW across three Tasmanian sites and had made no decision. 中 SR002
CR012 Firmus says its Australian energy and water policies establish commitments covering efficiency, procurement, storage, demand response, cooling, and reporting. 中 SR003
CR013 Firmus expects Launceston's average annual cooling-water use to approximate the annual water use of 20 average households. 低 SR003
CR014 Firmus describes its AI Factory platform as spanning cooling systems through grid-orchestration software. 中 SR006
CR015 Firmus says it publishes real-time energy-usage, thermal, and compute-benchmark data. 中 SR007
CR016 The Australia Times article discloses that an artificial-intelligence system generated its content. 中 SR005
CV001 GlobalData's Verdict report and Forbes each reported a $2 billion equity round at a post-money valuation above $10.5 billion in August 2026. 高 SV005, SV007
CV002 Forbes reported that Firmus was valued at $5.5 billion in April 2026, making the August mark roughly double the April mark. 中 SV002, SV007
CV003 GlobalData's Verdict report said the August transaction brought new equity raised over the preceding year to more than $3 billion. 中 SV005
CV004 Forbes and SetterVC each reported a $10 billion debt facility announced or closed in February 2026. 中 SV002, SV007
CV005 Reviewed reporting describes Project Southgate as targeting up to 1.6 GW of AI compute across Australian sites by 2028. 中 SV001, SV007, SV008
CV006 Firmus states that its AI Factory platform is modular and designed to lower power, water, and space requirements. 中 SV003, SV006
CV007 Firmus's official pages identify Loyang and Media Hub in Singapore as existing AI-factory locations or retrofits. 中 SV003, SV006
CV008 None of the eight assigned, successfully fetched sources provides audited current revenue, EBITDA, cash flow, utilization, customer concentration, or cap-table preference terms. 高 SV001, SV002, SV003, SV004, SV005, SV006, SV007, SV008
CV009 Australia Times labels both retained Firmus articles as AI-generated and warns that they may contain accidental inaccuracies. 中 SV001, SV008
CV010 SetterVC warns that its company, funding, and valuation marks are estimates that may be incomplete, stale, erroneous, or revised. 中 SV002
CV011 Forbes argues that AI-infrastructure operators, unlike equipment suppliers, carry utilization and refinancing risk over the buildout period. 中 SV007
CV012 The combination of a greater-than-$10.5 billion equity mark, a reported $10 billion debt facility, and absent public operating economics supports a price-sensitive track/research-more stance rather than an unconditional buy. 中 SV002, SV005, SV007
CV013 Public materials indicate an IPO ambition or timetable, but the assigned evidence does not include an exchange filing or completed-listing notice. 中 SV001, SV002, SV004
CV014 The reported April $5.5 billion mark is a downside reference rather than an intrinsic-value estimate because public metrics needed for revenue or cash-flow valuation are absent. 中 SV002, SV007
来源
编号出版方标题引文
SO001 Firmus Technologies Firmus announces fully subscribed USD$2 billion strategic equity investment This transaction brings Firmus’ total new equity raised over the past year to more than USD$3 billion, with the post-money valuation now above USD$10.5 billion.
SO002 Firmus Technologies Newsroom The AI Factory. Engineered through years of R&D. Proven and deployed.
SO003 ABC News Data centre operator Firmus pulls blockbuster ASX float after lacklustre demand Many investors were also focused on Mr Curtis's criminal background — he is a former investment banker who served one year in prison a decade ago for insider trading.
SO004 CNBC Nvidia-backed Aussie AI firm Firmus withdraws historic IPO, citing market volatility The company said the proposed offering terms did not adequately reflect its business strength and long-term growth outlook, and it will instead pursue private funding options.
SO005 Firmus Technologies About - Firmus 2019 Incorporated in Australia
SO006 Firmus Technologies Creating the most efficient AI infrastructure We design and operate Firmus AI Factories - modular, verticalized systems that optimise cost and power at every layer of the stack.
SO007 The Guardian From boom to bust in five days: how Firmus's much-hyped Australian stock market listing imploded On Friday, Firmus said it had withdrawn its application to list on the ASX, citing recent “market volatility and prevailing market conditions”.
SO008 TechNode Global Firmus scraps A$7 billion Australian IPO Australian AI infrastructure company Firmus Technologies has withdrawn its application to list on the Australian Securities Exchange, abandoning a proposed initial public offering that was expected to raise about A$7 billion and value the company at approximately A$44 billion.
SM001 Australian Financial Review Everything you need to know about Firmus before it goes public Artificial intelligence infrastructure start-up Firmus is set to be the second-biggest float in Australian sharemarket history, targeting a market capitalisation of $43.7 billion after tapping investors for at least $7 billion.
SM002 Firmus Technologies Creating the most efficient AI infrastructure At the core of every Firmus deployment is the AI Factory: a modular infrastructure platform designed to run high-density AI with lower power, water, and space requirements.
SM003 Tech Insider Firmus Targets A$7B ASX IPO, Second-Largest Ever [2026] Australia’s overall cloud spend hit an estimated A$33.6 billion in 2026, up close to 18% year over year, according to industry tracking cited across recent coverage, while AI-specific cloud infrastructure spending in the country reportedly climbed 128.4% to around A$946 million.
SM004 Firmus Technologies About We’ve deployed a suite of fit-for-purpose platforms and sovereign projects, refined through R&D, to deliver efficient, AI Factories direct to customers in the country.
SM005 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company Its flagship program, Project Southgate, started in Tasmania and is expanding into Melbourne, Sydney, Canberra, and Perth, targeting 1.6 gigawatts of AI compute across five sites by 2028 in partnership with CDC Data Centers.
SM006 Australia Times Firmus Technologies Secures A$500 Million to Accelerate Australia’s AI Data-Centre Ambitions Firmus says the raise will fund site development, infrastructure deployment and energy-partnerships to deliver up to 1.6 gigawatts of AI-compute capacity by 2028.
SM007 The Urban Developer Tasmania Hands Firmus Deadline for Wesley Vale ‘AI Factory’ Proposal Firmus Technologies has until the middle of December to file a complete major project proposal for its planned $2.1-billion 52MW data centre in north-west Tasmania or risk having the application withdrawn.
SM008 Startup Daily Green-powered AI data centre plan goes national with $73.3 billion project across 4 Australian capitals The bigger picture now includes new sites in Perth, Adelaide, Canberra and Sydney as well, at a total cost of up to $73.3 billion, delivering 1.6GW of compute at five Firmus sites by 2028.
SP001 Tracxn Firmus The company has 363 active competitors, including 55 funded and 25 that have exited
SP002 ABC News AI firm's ASX listing may be second-largest behind Telstra An important caveat for many of the figures being touted in the financial press for Firmus's float is that they are based on conjecture and hype.
SP003 Firmus Technologies Firmus Secures US$10 Billion Financing led by Blackstone and Coatue to Scale Energy-Efficient AI Infrastructure The financing will accelerate AI Factory deployment, infrastructure manufacturing, and energy integration to serve hyperscale and AI-native customers around the world.
SP004 The Australian Firmus splits with CDC ahead of float Many of those strategies have been abandoned, with Firmus now pivoting to building AI factories in Asia.
SP005 Firmus Technologies About - Firmus Our mission is to build the most energy efficient AI infrastructure
SP006 Firmus Technologies Creating the most efficient AI infrastructure - Firmus We design and operate Firmus AI Factories - modular, verticalized systems that optimise cost and power at every layer of the stack.
SP007 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company The operators carry a decade of utilization risk and refinancing risk on borrowed money.
SP008 AI Market Watch Firmus Technologies - AI Startup Profile Firmus differentiates through energy-efficient AI Factories with liquid-cooling, seawater-cooled AI compute, and a focus on sustainability.
SP009 Australia Times Firmus Technologies Secures A$500 Million to Accelerate Australia's AI Data-Centre Ambitions Global demand for AI compute continues to surge, with underlying infrastructure considered a key bottleneck.
SP010 Silicon Valley Investclub Firmus Technologies - Above $10.5B, CEO Oliver Curtis Offers AI infrastructure, accelerators, and managed AI services that overlap with Firmus's infrastructure pitch.
SI001 Venture Capital Tracker Firmus Funding, Valuation & Investors Latest funding Strategic equity $2B · August 7, 2026 Latest known valuation >$10.5B post-money
SI002 NPI Fund / Custom Map Poster From Prison to AI Unicorn: Oliver Curtis' $5.9 Billion Firmus Journey Oliver Curtis, a former convict, is now at the helm of a $5.9 billion AI company, Firmus, which is gearing up for a potential record-breaking tech IPO in Australia.
SI003 Firmus Technologies About - Firmus A global developer and operator powering the green AI revolution.
SI004 InforCapital Firmus Technologies - Funding & Investors The company has raised over US$2.8 billion in equity and secured a US$10 billion debt financing facility from Blackstone and Coatue in February 2026
SI005 Australia Times Firmus Technologies Secures A$500 Million to Accelerate Australia's AI Data-Centre Ambitions Australia’s AI infrastructure ecosystem has received a robust injection of capital with Firmus Technologies announcing a further A$500 million equity raise, taking its post-money valuation to approximately A$6 billion.
SI006 Firmus Technologies Creating the most efficient AI infrastructure We design and operate Firmus AI Factories - modular, verticalized systems that optimise cost and power at every layer of the stack.
SI007 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company The operators carry a decade of utilization risk and refinancing risk on borrowed money.
SI008 Silicon Valley Investclub Firmus Technologies — Above $10.5B · CEO Oliver Curtis No publicly reported financial figures beyond funding totals are present in the provided results.
SI009 The Australian What is Firmus, the data centre business everyone's talking about? The economics are also unforgiving. AI factories require heavy upfront investment in land, power infrastructure, cooling equipment and scarce GPU hardware.
SE001 VAST Data Firmus Technologies Group Selects VAST AI Operating System to Power Sovereign, Energy-Efficient AI Factories in Asia Pacific Our Model-to-Grid architecture integrates model behaviour, GPU performance, thermal management, and grid conditions into one optimisation framework, making our AI factories both model-aware and grid-aware, with real-time responsiveness to energy pricing and broader grid signals.
SE002 Tech Insider Firmus Tech's $505M Raise Powers $5.5B AI Data Center IPO The company’s pivot from crypto mining to AI infrastructure – while strategically sound – means it is still relatively early in its evolution as a data center operator.
SE003 Firmus Technologies Firmus to Build 170,000 GPU AI Factory Campus with NVIDIA for Global AI-Natives At the core of the partnership is the integration of NVIDIA DSX™ - NVIDIA's full-stack AI factory platform - with Firmus' proprietary HyperCube platform, a liquid-cooled AI Factory architecture developed in Australia and co-designed to NVIDIA DSX blueprints.
SE004 TechEdt Firmus Technologies Group selects VAST AI Operating System to power sovereign AI factories in Asia Pacific Firmus’ infrastructure model is based on the idea that AI factories must be engineered as vertically integrated systems rather than as collections of loosely coupled components.
SE005 The Australian What is Firmus, the data centre business everyone's talking about? While liquid cooling is more efficient than air cooling, it introduces operational complexity, including maintenance challenges and a limited pool of technicians experienced in operating such systems at scale.
SE006 Firmus Technologies About Firmus Our AI Factory platform, from cooling systems to grid orchestration software, was built to stay ahead of AI’s infrastructure and energy realities.
SE007 Firmus Technologies Creating the most efficient AI infrastructure At the core of every Firmus deployment is the AI Factory: a modular infrastructure platform designed to run high-density AI with lower power, water, and space requirements.
SE008 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company The operators carry a decade of utilization risk and refinancing risk on borrowed money. Companies selling into this buildout are collecting certainty, while companies operating it are absorbing leverage.
SU001 Dynamic Export Who Is Firmus Technologies? The Australian AI Company Building the Future of Data Centres The company has also struck a partnership with CDC Data Centres to extend its platform onto the Australian mainland, and it has signed a multi-year capacity agreement with a global hyperscale customer at Southgate — a sign the facility isn’t just speculative build-out but has committed demand behind it.
SU002 Firmus Technologies Firmus Technologies | LinkedIn With this partnership, our total contracted capacity across all customers now stands at more than 900 MW.
SU003 Firmus Technologies About - Firmus Transforming a former basement car park into 3MW GPU AI factory.
SU004 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company The operators carry a decade of utilization risk and refinancing risk on borrowed money.
SU005 Australia Times Firmus Technologies Secures A$500 Million to Accelerate Australia's AI Data-Centre Ambitions Co-CEO Tim Rosenfeld added that the vertically-integrated AI-factory platform had been developed over years and now allows Firmus to meet an accelerating demand for inference and training compute aligned with Australia’s renewable-energy ambitions.
SU006 Firmus Technologies Creating the most efficient AI infrastructure - Firmus This partnership provided the reliable, high-performance compute infrastructure needed for SEA-LION: the first family of open-source LLMs designed for Southeast Asian languages and contexts.
SU007 Silicon Valley Investclub Firmus Technologies — Above $10.5B · CEO Oliver Curtis No publicly reported financial figures beyond funding totals are present in the provided results.
SU008 AI Market Watch Firmus Technologies - AI Startup Profile | AI Market Watch Enterprises, government agencies, education institutions, startups, and scaleups requiring AI infrastructure, particularly in Australia, Singapore, and globally
SR001 The Next Web Nvidia-backed Firmus targets $2bn ASX IPO after locking in $505m equity and $10bn Blackstone debt for its AI factory network The full programme, developed alongside Nvidia and listed data centre operator CDC Data Centres, carries a projected build cost of $73.3bn, a figure that reflects both the capital intensity of AI infrastructure and the ambition of the national rollout.
SR002 Certified Strategic Firmus: valuation, owners, withdrawn IPO and AI factories It withdrew its planned ASX float on 9 October 2026 and is seeking private funding.
SR003 Firmus Technologies Firmus launches energy and water policies to support Australia's renewable transition and responsible AI growth Together, the policies establish measurable commitments covering energy efficiency, renewable energy procurement, battery storage, demand response, network costs, cooling technology, and transparent reporting across Firmus' Australian operations.
SR004 Firmus Technologies (syndicated by TelecomTV) Firmus to build 170,000 GPU AI factory campus with NVIDIA for global AI-natives The agreement covers up to 170,000 NVIDIA AI accelerators across Grace-Blackwell, Vera-Rubin, and Vera platforms through 2027 and 2028, positioning the Batam campus among the largest AI infrastructure developments in Asia-Pacific.
SR005 Australia Times Firmus Technologies Secures A$500 Million to Accelerate Australia's AI Data-Centre Ambitions AI Disclaimer: An advanced artificial intelligence (AI) system generated the content of this page on its own.
SR006 Firmus Technologies About - Firmus Our AI Factory platform, from cooling systems to grid orchestration software, was built to stay ahead of AI's infrastructure and energy realities.
SR007 Firmus Technologies Creating the most efficient AI infrastructure - Firmus Trust is built by visibility. We publish real-time energy usage, thermal data, and compute benchmarks.
SR008 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company The operators carry a decade of utilization risk and refinancing risk on borrowed money.
SV001 Australia Times Firmus Raises A$500 Million to Accelerate Australia's Green AI Infrastructure In September the company closed a A$330 million equity placement, bringing the post-money valuation to about A$1.85 billion.
SV002 SetterVC Firmus Technologies Company data and valuation marks are estimates and may be incomplete, stale, erroneous, or revised.
SV003 Firmus Technologies About Firmus Our mission is to build the most energy efficient AI infrastructure
SV004 AI Market Watch Firmus Technologies - AI Startup Profile Post-money valuation above US$10.5 billion following the August 2026 US$2 billion strategic equity round; US$10 billion debt financing facility announced February 2026.
SV005 Verdict / GlobalData Firmus raises $2bn to accelerate Project Southgate rollout The transaction brings the company's total new equity raised over the past year to more than $3bn, and it said its post-money valuation is now above $10.5bn.
SV006 Firmus Technologies Creating the most efficient AI infrastructure At the core of every Firmus deployment is the AI Factory: a modular infrastructure platform designed to run high-density AI with lower power, water, and space requirements.
SV007 Forbes Former Bitcoin Miner Firmus Became A $10.5 Billion AI Infrastructure Company Firmus, which now builds AI data centers, raised $2 billion in equity this week at a valuation above $10.5 billion, roughly double where it stood in April.
SV008 Australia Times Firmus Technologies Secures A$500 Million to Accelerate Australia's AI Data-Centre Ambitions Australia's AI infrastructure ecosystem has received a robust injection of capital with Firmus Technologies announcing a further A$500 million equity raise, taking its post-money valuation to approximately A$6 billion.