初创公司尽调
尽调报告 Space Tech / Orbital Infrastructure Series B 2026-06-22

Cowboy Space Corporation

Cowboy Space 是尚未产生收入的轨道 AI 基础设施初创公司,背后有一线资本,架构也足够大胆;但热物理尚未验证、火箭尚未造出, 还要支撑 $2B 的论证期估值,必须用严格里程碑来约束。

Cowboy Space 拿到 Tier-1 投资人、创始人履历可信,也踩中真实的结构性市场驱动;但 $2B 入场估值仍完全停在论证阶段:公司尚未产生收入,火箭还没造出来,1 MW 轨道数据中心的核心热物理也没有公开验证。

封面要素

Series B 后估值 01
2000 USD M [CO001]
累计融资总额 02
365 USD M [CO009]
Series B 轮规模 03
275 USD M [CO001]
Stampede 星座申请 04
up to 20,000 satellites [CO026, CO027]
首个计算节点目标 05
early 2027 [CO030]
首枚自研火箭目标 06
end 2028 [CO031]
成立时间 07
2024 [CO006]

公司概况

Cowboy Space Corporation(前身为 Aetherflux)是一家位于加州 San Carlos 的 Series B 轨道基础设施公司, 由 Robinhood 联合创始人 Baiju Bhatt 于 2024 年创立。公司的核心论点,是把自研 Falcon-9 级运载火箭与 20,000–25,000 kg、同时作为 1 MW 太阳能 AI 数据中心的卫星上面级纵向整合,拼出计划中的「Stampede」 20,000 颗卫星星座。2026 年 5 月,公司以 $2B 估值融资 $275M,并从太空太阳能起点(Aetherflux)改名后, Cowboy 转向服务被地面电网多年并网队列卡住的 AI 超大规模云厂商和国防客户。公司尚未产生收入;热管理、 辐射加固和发射节奏等关键技术风险仍未在公开资料中解决。

官网
cowboyspace.com
成立时间
2024-03-01
创始人
Baiju Bhatt
创立地点
San Carlos, California, USA
总部
San Carlos, California, USA
产品
纵向整合的轨道数据中心——每颗卫星就是火箭上面级——在晨昏太阳同步轨道产生 1 MW 太阳能,每个单元搭载约 800 个 NVIDIA Space-1 Vera Rubin GPU;以租赁轨道 AI 算力容量的方式出售给无法在有竞争力时间内拿到地面电力的客户。
客户
AI 超大规模云厂商、AI 研究实验室、国防与情报机构,以及寻求高功率计算基础设施、又不想受地面电网并网队列约束的地球观测运营商。
商业模式
向能容忍卫星链路延迟的客户租赁轨道算力容量(按 GPU-hour 或 kWh 计费),资金来自股权融资和预期中的政府合同; 截至本报告日期,公司未披露收入或定价表。
阶段
Series B
融资情况
2026 年 5 月完成 $275M Series B,后估值 $2B,由 Index Ventures 领投,IVP、Blossom Capital、SAIC 及所有主要既有投资方参投; CEO Baiju Bhatt 确认累计资本约 $365M。
[CO001, CO004, CO005, CO006, CO009, CO011, CO014, CO025]

执行摘要

主要优势

  • Tier-1 投资人组合(Index Ventures 领投 A、B 轮,a16z、Breakthrough Energy Ventures、SAIC 参投)加上 Baiju Bhatt 在 Robinhood 量级的执行履历,让这笔无收入深科技押注拥有异常强的融资通道和叙事可信度。
  • NVIDIA Space-1 Vera Rubin 发射合作伙伴身份验证了轨道 AI 计算硬件逻辑,也给 Cowboy 一块多数竞争对手拿不到的专用 GPU 模块。
  • 地面 AI 计算瓶颈(电网互连排队 5–7 年、Goldman Sachs 预测数据中心用电到 2030 年增长 160%)为轨道电力与计算创造了真实结构性需求,前提是单位经济性能跑通。
  • 发射与计算纵向整合在架构上很特别:Cowboy 把上面级和卫星合并,减少冗余结构质量,也在规模化后摆脱外部发射商依赖。

主要风险

  • 公司完全处于无收入阶段,没有披露客户合同、LOI 或商业试点;相比地面方案实现价格平价时,客户付费意愿和单位经济性仍未验证。
  • 真空中排出 1 MW GPU 计算废热,是最核心的未解工程难题;ArsTechnica 估算每颗卫星约需 200 m² 散热器面积,公开资料没有任何鉴定或测试数据。
  • 一边自研 Falcon-9 级火箭,一边做轨道数据中心,对资本和执行要求极端;可比项目(Relativity Space、ABL Space Systems)在首个运营里程碑前消耗 $100M–$650M,且多家彻底失败。
  • SpaceX 2026 年 1 月向 FCC 申请最多 1 million 颗轨道数据中心卫星,带来生存级竞争威胁;只有它具备足够发射节奏和 Starship 载荷能力,能以成本有效方式部署 20–25 tonne 卫星。

未决问题

  • 没有收入、ARR、积压订单、LOI 或具名客户;所有商业需求信号都停留在行业层面的 thesis,而非公司自身证据。
  • 公开记录缺少 700–1,000 km 轨道带的热真空鉴定数据和抗辐射方案。
  • 火箭发动机设计、推进测试结果、制造设施和 FAA 发射许可路径均未公开披露。
  • 完整股权结构表、优先股堆叠、Series B 清算条款,以及任何老股或债务组成均未公开。

目录

Chapter 01

01公司概览

1.1 身份、使命与商业模式

Cowboy Space Corporation 是一家注册于加州 San Carlos 的轨道基础设施公司。Baiju Bhatt 于 2024 年以 Aetherflux 之名创立公司,最初论点是用红外激光把低地球轨道的太阳能传回地面接收端。2026 年 5 月,随着 $275 million Series B 完成,公司改名为 Cowboy Space Corporation,并把主业转向轨道 AI 数据中心和纵向整合运载火箭; 太空太阳能不再是终端产品,而是保留下来的核心架构积木。公司网站把使命表述为「从高边疆为人类供能」 ——也就是为人工智能建设外太空电网。 核心商业模式是纵向整合的轨道基础设施:Cowboy Space 把自研火箭、卫星上面级和在轨计算载荷设计成一个统一系统。 火箭上面级送达载荷后不丢弃;相反,上面级本身就是卫星数据中心载荷。进入轨道后,每个单元在晨昏太阳同步轨道产生 1 MW 太阳能,并用 NVIDIA Space-1 Vera Rubin Modules 运行约 800 个 GPU。传统架构把火箭和卫星拆成两个系统, 会产生冗余结构和质量;这一设计试图把这些冗余拿掉。收入将来自向超大规模云厂商和企业出售轨道 AI 算力容量, 定价对标需要排多年队才能并网的地面替代方案。截至本报告日期,公司尚未产生收入。 Cowboy Space 把市场机会锚定在一个具体的地面瓶颈:美国主要市场里,新数据中心平均并网前置时间达到五到七年甚至更久, 而 AI 算力需求增长快得多。Cowboy 认为,把计算放到连续 LEO 太阳能旁边,并利用天然真空冷却,就能绕开限制地面扩张的 电力和冷却约束。公司 2026 年 5 月向 FCC 递交申请的「Stampede」星座,设想最多由 20,000 颗卫星提供数据中心服务。[CO005, CO006, CO008, CO025, CO033, CO034]

Cowboy Space 快照 KPI
指标数值 / 状态日期置信度缺口 / 注释
投后估值$2.0 billion2026-05-11Series B 投后;最后已知估值
已融资总额~$365 million2026-05-11CEO 向 SpaceNews 表述;SatNews 引用 $325M(见资本章节)
最新融资轮次Series B2026-05-112026 年 5 月完成
Series B 轮规模$275 million2026-05-11公司官方新闻稿
领投方Index Ventures(Jan Hammer)2026-05-11领投 Series A 和 B
总部加州 San Carlos2026-06-22公司网站披露的主要总部
其他办公室Seattle, WA(卫星 + 推进);Texas(测试设施)2026-05-11GreyJournal 提到 Texas 场地;未正式确认
品牌更名前估计员工数~40 人2026-05-11据 GreyJournal;未披露官方当前人数
当前收入 / ARR未披露2026-06-22商业化前;未披露 bookings
首颗演示卫星发射2026(计划)2026-05-11通过 Apex Space 进行亚尺度能量传输演示
首个数据中心节点2027 年初(计划)2026-05-11NVIDIA Space-1 Vera Rubin「Galactic Brain」节点
首次自研火箭发射2028 年底(计划)2026-05-11不早于该时间;内部时间线可能滑坡
计划星座规模最多 20,000 颗卫星2026-05-15据 FCC 申请;尚未获批

置信度评级反映来源质量和交叉验证程度。计划日期是公司表述的目标;深科技硬件时间线通常会滑坡。收入缺口是结构性的——公司仍处商业化前。

[CO001, CO005, CO007, CO009, CO026, CO029]
FO002: 垂直整合架构逻辑

公司的集成系统把太阳能发电、上面级数据中心设计、轨道计算运营,以及把数据传回地球企业客户串在一起。

架构基于公司描述和 FCC 申报;技术规格是设计目标,不是已验证性能。

[CO001, CO026, CO032, CO033, CO034, CO035]

1.2 创立历程与从 Aetherflux 的战略转向

Baiju Bhatt 在 2023 年底或 2024 年初开始搭建后来成为 Cowboy Space 的公司,并于 2024 年 3 月在加州 San Carlos 以 Aetherflux 名义正式注册,用个人约 $10 million 资金支持种子阶段。Bhatt 在 2024 年初离开 Robinhood Markets;此前他任联席 CEO 至 2020 年,随后任首席创意官。他保留约 6% Robinhood 股份, 足以用个人资产负债表托住早期开发。他父亲曾在 NASA Langley Air Force Base 工作,加上他本人在 Stanford 的物理和数学训练, 共同塑造了他对太空商业化的长期兴趣。 Aetherflux 于 2024 年 10 月走出隐身,提出小卫星 SBSP 论点:一组装有太阳能阵列和近光学红外激光器的 LEO 卫星, 能够规模化收集太阳能,并把能量传给紧凑型地面接收端。公司称这一方案与早期地球静止轨道巨构设想显著不同: 更小、更模块化,也更容易迭代。2025 年 4 月,Aetherflux 完成由 Index Ventures 领投的 $50 million Series A, 累计资本增至约 $60 million;同时,公司通过美国国防部 Operational Energy Capability Improvement Fund 获得政府支持, 用于概念验证型无线电力传输系统。团队也在 2025 年完成了实验室电力传输演示。 催生 Cowboy Space 的转向,来自残酷的经济约束:Bhatt 告诉 TechCrunch,他接触了多家发射服务商, 但找不到足够运力,无法以有竞争力的单位经济性扩展轨道数据中心业务。公司决定加入内部火箭项目, 把发射视为纵向整合层,而不是第三方成本。原有太空太阳能论点被保留,但重新定义——Cowboy 不再把电力传回地球, 而是在轨道现场用它运行计算。2026 年 5 月 11 日,改名、新品牌和 $275 million Series B 同时公布, 标志着轨道数据中心与运载火箭组合战略正式亮相。[CO004, CO006, CO011, CO013, CO018, CO019]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2023 年末 / 2024 年初Baiju Bhatt 离开 Robinhood,并开始 Aetherflux 构想创立n/aBhatt天基太阳能论点启动;个人资本承担风险
2024 年 3 月Aetherflux 在加州 San Carlos 正式注册成立创立~$10M 种子轮(Bhatt 个人)Bhatt;创始团队公司注册成立;自筹资金阶段开始;最初 SBSP 使命
2024 年 10 月Aetherflux 以 LEO 天基太阳能星座使命走出 stealth产品累计融资 ~$10MBhatt;初始团队(NASA/JPL、SpaceX、Robinhood 老兵)首次公开亮相;SBSP 论点;Apex Space 被指定为首个卫星平台伙伴
2025 年 4 月融资 $50M Series A;获得 DoD OECIF grant融资$50M Series A;累计 ~$60M;DoD OECIF 未披露Index Ventures(领投)、Interlagos、Breakthrough Energy、a16z、NEA、天使(Tenev、Gallagher、Leto)首笔机构资本;防务应用得到验证;能量传输演示获得资金
2025(H2)无线电力传输实验室演示成功产品n/aAetherflux 团队核心物理在空间部署前的受控环境中得到验证
2025–2026战略转向决定:增加运载火箭项目;重心转向轨道数据中心产品n/aBhatt;领导团队基本范围扩张;商业运力不可得触发火箭项目启动
2026 年 5 月 8 日宣布完成 Series B;从 Aetherflux 更名为 Cowboy Space Corporation融资$275M,估值 $2B;累计 ~$365MIndex Ventures(领投)、IVP、Blossom Capital、SAIC + 既有投资者达到独角兽状态;轨道数据中心 + 发射使命公开
2026 年 5 月向 FCC 提交 Stampede 星座(20,000 颗卫星)申请 SAT-LOA-20260323-00135监管n/a — 授权待定Cowboy Space;FCC 卫星许可部门首次正式监管提交;20,000 颗卫星星座申请;请求豁免
2026 年 6 月 12 日FCC 公告 SAT-02007:Stampede 申请被接受备案监管n/a — 审查中FCC 卫星许可与项目部门申请进入 FCC 审查队列;最终授权可能需要修改许可证
2026(计划)首次卫星发射:亚尺度能量传输演示(Apex Space 平台)产品n/aCowboy Space;Apex Space验证天基到地球的红外激光能量传输;能源架构的技术证明点
2027 年初(计划)首个使用 NVIDIA Space-1 Vera Rubin 模块的 LEO「Galactic Brain」数据中心节点产品n/aCowboy Space;NVIDIA首个由太阳能供电的轨道 AI 计算演示;全规模部署前奏
2028 年底或更晚(计划)首次搭载 1 兆瓦数据中心上面级的自研火箭发射产品n/aCowboy Space核心价值主张:垂直整合发射 + 计算得到验证;不早于该时间线

标注「计划」的行是公司表述的目标,可能发生进度滑坡。创立日期为近似值——DataCenterDynamics 报道为 2024 年 3 月;TechCrunch 2024 年 10 月文章称 Bhatt 在「去年底」创立公司。FCC 时间线基于公告接受备案;授权不保证。

[CO006, CO018, CO023, CO024, CO025, CO027]
FO001: Cowboy Space 里程碑时间线

从 Aetherflux 创立到 Cowboy Space 计划在 2028 年前完成的关键节点,展示其从 SBSP 论点走向垂直整合轨道数据中心战略的演进。

计划里程碑日期(2026 年演示、2027 年 Galactic Brain、2028 年火箭)是基于公开说法的近似占位;实际执行日期会变化。

[CO006, CO018, CO025, CO027, CO028, CO029]

1.3 领导层与关键技术团队

Baiju Bhatt 担任创始人兼 CEO,也是 Cowboy Space 唯一的公开代表。他的创始人—市场匹配建立在三根柱子上: 物理学资历(Stanford 物理与数学 BA/MS)、个人资本投入风险(约 $10 million 种子资金并参与 Series B), 以及从零把 Robinhood 做成数十亿美元上市金融平台的既有记录。他公开表示,Cowboy Space 的野心规模 ——同时造火箭、建星座、搭轨道计算平台——需要创始人愿意押注非常规的第一性原理。关键人风险很高: 公司的品牌、融资叙事和投资人关系,都紧密绑定 Bhatt 的信誉和个人愿景。 Bhatt 组建的技术领导层,很大程度来自成熟商业航天产业。Tyler Grinnell 曾任 SpaceX 发射运营总监, 负责发射运营。Warren Lamont 曾在 Blue Origin 领导火箭发动机和助推级开发,现在从新的 Seattle 办公室牵头推进工程。 David Larson 曾任职 SpaceX 和 Amazon Kuiper,担任卫星工程总监,同样位于 Seattle。更广泛的团队履历覆盖 NASA/JPL、Astranis、Anduril、Lockheed Martin 和美国海军,让公司在推进、航电、航天器设计和国防应用上有相关纵深。 截至 2026 年 5 月,GeekWire 报道仅 Seattle 就有 18 个开放职位,管理层预计近期 Seattle 员工总数达到 40–60 人。 公司总人数未公开披露。[CO004, CO011, CO012, CO013, CO014, CO015]

领导层与创始人表
人物职务此前机构创始人-市场匹配 / 职能覆盖关键人物风险
Baiju Bhatt创始人兼 CEORobinhood(联合创始人 / 联席 CEO / CCO)Stanford 物理 / 数学;投入个人资本;已证明平台级执行能力高 — 唯一公开面孔;投资论点围绕其愿景
Tyler Grinnell发射运营负责人SpaceX(发射运营总监)来自全球最活跃商业发射服务商的运营经验中 — 稀缺领域人才;对发射节奏至关重要
Warren Lamont推进负责人Blue Origin(火箭发动机与助推器负责人)来自载人飞行器项目的发动机开发和助推器工程经验高 — 火箭发动机设计是最稀缺、资本强度最高的能力
David Larson卫星工程总监SpaceX;Amazon Kuiper星座规模航天器设计;跨行业轨道系统背景中 — 覆盖卫星 / 上面级集成关键路径

仅列出公开具名高管的部分名单。截至 2026 年 6 月,完整 C-suite(CFO、CTO、GC)未公开披露。所有角色均由新闻报道推断;正式头衔可能不同。

[CO004, CO011, CO014, CO015, CO016]

1.4 资本结构与投资人联盟

Cowboy Space 通过一系列融资,累计拿到约 $365 million 资本。种子阶段($10 million,Bhatt 个人出资) 支撑公司从创立走到隐身期。2025 年 4 月 Series A($50 million,Index Ventures 和 Interlagos 领投) 带来机构信誉和 DoD 验证。2026 年 5 月 Series B(后估值 $2 billion,融资 $275 million,Index 再次领投) 把公司推入独角兽行列,也提供了同时扩充火箭项目、卫星工程能力和轨道计算开发所需资本。Fenwick and West LLP 在 Series B 交易中代表 Cowboy Space。 投资人联盟混合了通用科技风投、深科技资本、国防战略投资人和气候资本,组合并不常见。Index Ventures 连续领投两轮机构融资, 释放出罕见信心;合伙人 Jan Hammer 在官方新闻稿中称,公司认为 Bhatt 拥有「以第一性原理重新想象巨大市场的成熟记录」。 Breakthrough Energy Ventures(Bill Gates)带来清洁技术和能源使命,与最初 SBSP 论点匹配,也仍然契合供能架构。 Andreessen Horowitz 和 NEA 增加通用一线品牌。Construct Capital 是一家关注实体 AI 的基金,也是公司首个机构支持者; 该基金发文称,轨道数据中心战略正是其寻找的基础设施押注类型。SAIC 作为国防承包商投资人参投,暗示 Cowboy 可能获得 DoD 市场入口,政府对轨道计算的采购兴趣也可能延续 OECIF grant 建立的 DoD 落点。IVP 和 Blossom Capital 则补上成长阶段资本和欧洲 VC 代表性。 已报道总额存在小幅差异:SatNews 称已披露股权融资约 $325 million,SpaceNews 则基于 CEO 直接说法报道约 $365 million。较高数字与 TechCrunch 的说法一致:Series B 前已从投资人融资 $80 million,再加上 Bhatt 的个人资本。 收入和烧钱速度未公开披露。[CO001, CO002, CO003, CO009, CO010, CO018]

利益相关方或投资者图谱
利益相关方类型轮次角色 / 战略理由尽调问题
Index Ventures(Jan Hammer)领投 VCSeries A(领投);Series B(领投)信念领投方;以更高持股加码;预计董事会影响力最强确认董事席位数量、控制条款和 pro-rata 权利
Breakthrough Energy Ventures深科技 / 气候 VCSeries A;Series BBill Gates 基金;能源使命与天基太阳能输送架构一致了解共同投资权和任何 IP 贡献条款
Andreessen Horowitz(a16z)一线综合 VCSeries A;Series B品牌信号;综合科技覆盖;广泛组合网络确认轮次 allocation,以及观察员还是董事席位状态
Construct CapitalPhysical AI VCSeries A(首个机构);Series B自称最早机构支持者;撰写投资理由博客;physical AI mandate了解作为首个机构支持者是否拥有特殊治理权
NEA(New Enterprise Associates,投资方)多阶段 VCSeries A;Series B大型多元化基金;提供成长资本和后期支持确认董事会观察员权和任何反稀释条款
InterlagosVCSeries A(共同领投);Series B与 Index 共同领投 Series A;战略价值未披露识别 LP,以及除财务回报外是否有战略理由
IVP成长期 VCSeries B(新进)成长型综合投资者;增加后期资本市场通道确认其对硬件和监管风险的尽调范围
Blossom Capital欧洲综合 VCSeries B(新进)欧洲 VC;提供进入 EU 市场的选择权确认尽调范围;注意其与核心运营的地理距离
SAIC防务承包商(战略)Series B(新进)Science Applications International Corp.;政府服务与防务系统;延伸 DoD OECIF grant 关系确认 DoD 应用是否有任何 IP 排他性、优先查看权或共同开发条款
Baiju Bhatt(个人)创始人投资者种子轮(~$10M);Series B(参与)~6% Robinhood 持股;利益一致信号;最大单一个人风险承担者确认承诺的个人资本总额,以及任何 lock-up 或 side-letter 条款

覆盖官方来源披露的全部具名 Series B 投资者。仅参与早期轮次的天使投资人(Vlad Tenev、Dan Gallagher、Jared Leto)未纳入,因为他们是 Series A 个人参与者,且未确认参与 Series B。董事会构成未公开披露。

[CO001, CO002, CO003, CO018, CO019, CO020]

1.5 技术架构、路线图与监管状态

核心技术押注,是把发射和计算纵向整合。传统轨道架构把火箭当作运送工具、把卫星当作货物; Cowboy Space 则把火箭上面级和计算载荷合并成单一飞行器。每个 Stampede 卫星数据中心模块预计重 20,000-25,000 kg,在晨昏太阳同步轨道持续受光并产生 1 MW 可用电力,通过 NVIDIA Space-1 Vera Rubin Modules 搭载约 800 个 GPU——NVIDIA 将该产品定位为专为 LEO AI 基础设施打造。该飞行器将大于 SpaceX 的 Falcon 9,但小于 Starship。 计划中的技术演示分三步。第一步,由 Apex Space 建造的缩比电力波束测试卫星计划于 2026 年发射, 用于验证太空到地面的红外激光电力传输——这沿用了 Aetherflux 路线图。第二步,搭载 NVIDIA 模块的早期 「Galactic Brain」数据中心节点计划于 2027 年初进行,演示太阳能驱动的轨道 AI 计算。第三步, 首次自研火箭发射将携带完整 1 MW 数据中心上面级,目标时间不早于 2028 年底。公司还计划开发自研火箭发动机 ——这是资本和时间消耗最高的组件——并仍在敲定测试、制造和发射设施需求。 监管方面,Cowboy Space 于 2026 年 5 月提交 FCC 申请 SAT-LOA-20260323-00135,请求授权以 Stampede 名义部署和运营最多 20,000 颗非地球静止轨道卫星。根据 FCC 2026 年 6 月 12 日公共通知,该申请已获受理公示。Stampede 将主要通过光学星间链路运行, 只为遥测、跟踪和指令请求窄 Ka-band 频谱。公司寻求豁免若干 FCC 规则,包括部署进度要求;其理由是光学架构不会带来频谱囤积风险, 且只要一颗卫星入轨就可开始商业服务。关键在于,FCC 申请本身承认卫星设计尚未最终确定,商业服务启动前还需要修改许可。 SatNews 对两个实质问题提出担忧:如何在真空中耗散 1 MW 热负荷的物理问题仍未解决;700-1,000 km 高度、20,000 颗卫星壳层的碎片责任也未解决。 这些工程问题,公司尚未公开回答。[CO026, CO027, CO028, CO029, CO030, CO031]

FO003: 公司快照 KPI

截至 2026 年 6 月,Cowboy Space 的关键财务和运营指标显示,这是一家资本充足、尚无收入、处于 Series B 阶段的深科技基础设施公司。

所有数字都是基于公开说法的目标或估计;尚未证明收入或运营表现。

[CO009, CO026, CO029, CO031, CO032, CO043]

1.6 展示项

Chapter 02

02市场分析

2.1 市场边界与 AI 算力电力需求

Cowboy Space 面向的市场,位于两个已有独立文献支撑的趋势交汇处:AI 算力需求指数级增长,以及地面电力基础设施的系统性约束。 Goldman Sachs Research 预计,到 2030 年,数据中心电力需求将增长 160%,主要由 AI 工作负载驱动。一次 ChatGPT 查询约消耗 2.9 watt-hours,约为一次 Google 搜索 0.3 watt-hours 的十倍;模型规模和查询量增长后,总用电量也随之上升。 数据中心目前消耗全球 1–2% 电力;Goldman Sachs 估计,到 2030 年这一比例将升至 3–4%,仅美国数据中心就将消耗国内 8% 电力(2022 年为 3%)。International Energy Agency 的 Electricity 2024 分析同样把数据中心用电量列为结构性加速趋势。 眼前瓶颈不是硅片,而是电力接入。Cowboy Space 的 FCC 申请明确写道:「公用事业和地面电网的传统并网节奏, 已无法满足 AI 的基础设施需求。」投资人 Construct Capital 称,主要市场的数据中心拿电网接入需要等待多年。 按 Goldman Sachs 估算,仅为支持数据中心,美国公用事业就需要投入约 $50 billion 新增发电能力。 土地许可延误、冷却用水约束,以及 AI 枢纽地区的电网拥堵,进一步拉长了时间线。 Cowboy Space 对市场边界的主张是:能够容忍卫星链路延迟的工作负载——大批量 AI 训练、地球观测分析、 国防边缘处理和太空原生计算——可以由太阳能轨道基础设施承载,并完全绕开地面电网。公司的 FCC 申请称: 「把硅片放到阳光旁边,Stampede 就能完全跳过地面电网。」该市场被定义为面向无法或不愿等待地面电网容量的客户的 高功率 AI 计算和存储基础设施,交付机制是轨道部署。对延迟敏感的消费者服务、金融交易和大多数实时企业应用被明确排除。[CM001, CM002, CM003, CM004, CM005, CM006]

市场定义 — 轨道数据中心范围
板块 / 类别纳入支出排除支出买方 / 付款方与 Cowboy Space 的相关性
AI 训练(批处理、可容忍时延)大模型训练的 GPU-hours;电力供应;冷却基础设施实时推理;时延敏感推理;面向消费者的 APIFrontier AI labs;hyperscalers(通过伙伴 / 云间接)收入潜力最高;需要解决数据重力和单位经济性
地球观测分析SAR、高光谱、视频的在轨处理;下行链路压缩服务地面图像处理;卫星设计 / 制造EO 运营商(Planet、Maxar 同行);政府遥感机构近期切入口;数据在轨道生成,无需地面路由
防务与情报边缘计算韧性分布式计算;前沿部署传感器处理核心机密云基础设施;指挥控制地面系统DoD、情报机构、盟友防务;SAIC 作为中介溢价定价;采购周期长;SAIC 参与 Series B 释放意图信号
商业空间站运营研究载荷、乘员系统、空间站管理的机载计算空间站硬件 / 建设;发射服务Axiom Space、NASA、国际伙伴;商业空间站运营商近期但规模有限;Axiom 已在 ISS 部署 ODC 单元
主权 / 异地数据存储安全、独立于司法辖区的归档存储;备份与恢复主要数据工作负载;实时访问用例受监管企业;关注数据驻留的政府邻近而非核心;Lonestar Space 直接服务该板块

类别反映 Cowboy Space FCC 申请中的公开表述理由,并由 NewSpace Economy 2026 年 5 月行业调查佐证。纳入和排除的支出反映分析师判断,并非独立审计市场数据。截至 2026 年 6 月,没有 ODC 公司披露收入。

[CM019, CM020, CM021, CM022, CM023, CM025]
FM002: AI 数据中心电力需求估计——低 / 基准 / 高区间

Goldman Sachs 对 2030 年美国数据中心用电占比的基准情景,加上合理低 / 高变体;所有数字都指地面,而非轨道。

低值和高值是 Goldman Sachs 2024 基准情景周围的分析师区间;并非来自独立建模。所有数字都代表地面用电需求;不包括轨道数据中心。 Goldman Sachs 数字来自公开摘要文章,而非完整专有研究报告。

[CM001, CM003, CM004, CM006, CM007]

2.2 市场规模(证据约束)

截至 2026 年 6 月,轨道数据中心作为独立类别,并不存在独立权威的市场规模测算。该类别尚未产生收入, 也尚未实现规模化商业化;套用传统自上而下市场规模测算,需要假设工作负载迁移,而这些假设目前没有足够证据支撑。 本文采用证据约束方法:先锚定决定上限的地面计算和电力市场,再结合上述约束推演合理的轨道份额。 在地面层面,AI 数据中心基础设施建设涉及超大规模云厂商每年数千亿美元级资本开支。Goldman Sachs 估计, 仅数据中心就需要美国公用事业新增 $50 billion 发电能力;未来十年,欧洲为数据中心增长所需的电网投资超过 €800 billion。这些数字界定了 Cowboy Space 试图解决问题的量级。 轨道 TAM 无法从公开来源推导。ODC 类别更适合被理解为一个新生子细分市场,来自两类来源:(1)AI 训练和批量推理工作负载中, 既能容忍延迟、又高功率的部分;(2)目前缺少在轨解决方案的太空原生数据处理。SpaceX(最多 1 million 颗卫星)、 Blue Origin(最多 51,600 颗卫星)、Starcloud(最多 88,000 颗卫星)和 Cowboy Space(最多 20,000 颗卫星) 的多份竞争性 FCC 申请,合起来说明资源充足的参与者把该类别视为可行;但这些申请是监管占位,不是市场验证。 截至本报告日期,任何 ODC 公司都未披露收入、客户数量或签约容量。没有商业验证数据,就无法单独拆出 Cowboy Space 的 SAM 和 SOM;这一缺口已在证据缺口章节明确列出。[CM011, CM012, CM013, CM014, CM015, CM016]

市场规模测算视角 — AI 算力与电力基础设施
发布方 / 视角发布年份地理范围指标 / 数值CAGR / 时间范围方法论置信度ODC 规模测算限制
Goldman Sachs Research2024全球到 2030 年 DC 电力需求较 2022 基线增加 160%2022–2030 年约 ~13% p.a.自下而上的公用事业负载和 AI 工作负载建模DC 电力需求 ≠ 轨道 DC TAM;没有轨道份额估计
Goldman Sachs Research2024美国仅数据中心就需要新增 $50B 公用事业发电能力2030 年前的 Capex 需求覆盖 50 个美国市场的公用事业投资建模衡量地面基础设施缺口;也是轨道替代方案的上限
Goldman Sachs Research2024美国到 2030 年,美国数据中心将消耗美国 8% 的电力(2022 年为 3%)绝对需求轨迹公用事业委员会数据;AI 工作负载扩展假设交代电力需求规模背景;并非专门针对轨道
IEA Electricity 20242024全球数据中心正成为更显眼的电力消费者;数据中心板块分析预测至 2026 年国家级电力平衡建模发布时间早于 ODC 的 FCC 申请;未覆盖轨道板块
Cowboy Space FCC 申请(Stampede)2026全球到 2030 年全球数据中心电力需求增长 160%(引自 Goldman Sachs)作为政策理由引用引用 Goldman Sachs;并非独立数据公司倡议文件;不是独立市场分析
没有独立 ODC 市场规模测算2026全球暂无——该品类仍处于收入前、商业化前阶段N/A证据缺口截至 2026 年 6 月,没有任何独立分析机构发布轨道数据中心 TAM/SAM/SOM

所有条目都指向地面 AI 计算和电力市场,它们构成轨道数据中心的需求背景。截至 2026 年 6 月,未发现独立的轨道数据中心市场规模报告。Goldman Sachs 数字来自公开摘要文章,而不是完整的专有研究报告。IEA 数据通过 Wayback Machine 快照访问。

[CM001, CM004, CM005, CM011, CM012, CM013]
FM001: ODC 市场规模金字塔——从地面背景到轨道机会

TAM / SAM / SOM 各层锚定地面 AI 计算需求;轨道份额仍属推测,取决于商业验证。

公开来源无法给 SAM 和 SOM 层划定数字边界。TAM 数字来自地面 AI 计算研究(Goldman Sachs 2024)。金字塔展示的是有文献支撑的需求问题与尚无文献支撑的轨道方案之间的结构关系,而不是规模预测。

[CM001, CM004, CM005, CM011, CM012]

2.3 买方与客户细分

The NewSpace Economy 于 2026 年 5 月对 ODC 公司进行的综合调查,识别出五类采用画像明显不同的客户。 地球观测运营商是最近期客户:它们在轨道上通过合成孔径雷达、高光谱成像和视频产生大量数据, 在轨处理能够减少下行负载,并加快产品交付。国防和情报客户看重韧性、前沿部署传感器的延迟降低, 以及在地面网络中断时的基础设施韧性;SAIC 参与 Cowboy 的 Series B 被引用为国防领域兴趣信号。 空间站运营商(Axiom Space 及 ISS 的商业继任者)需要在舱计算支持研究、乘组和空间站运营。 Lonestar Space 的「StarVault」模式瞄准第四类客户:受监管行业和政府的数据主权及安全异地存储。 最后,前沿 AI 实验室和超大规模云厂商是长期最大目标,但摩擦也最大:数据重力、延迟、网络带宽和安全认证, 都会形成采用壁垒,近期轨道基础设施无法完全解决。 各细分市场的预算归属和采购路径差异很大。地球观测公司通常采用商业服务合同或托管载荷协议。 国防采购集中,认证周期长;SAIC 投资可能意味着公司有意加速这条路径。超大规模云厂商有内部资本预算, 最可能先以合作伙伴或投资人身份进入,而不是成为早期客户——Google 的 Project Suncatcher (计划到 2027 年初与 Planet 发射两颗原型卫星)体现了这种探索姿态。轨道数据主权客户付费购买的是合规级保证, 不是原始计算密度。[CM019, CM020, CM021, CM022, CM023, CM024]

买方与客户细分地图
细分买方用户付款方工作流 / 待完成任务预算负责人采用触发因素
地球观测分析EO 运营商(Planet 同类、SAR 公司)数据科学 / ML 团队EO 运营商资本开支预算降低下行链路瓶颈;加快产品交付CTO / 基础设施 VP规模化后的下行链路饱和;客户带来的延迟 SLA 压力
国防 / 情报边缘DoD 主承包商;SAIC;情报机构分析师;传感器融合平台国防项目办公室;涉密采购前沿部署数据处理;抵御地面中断的韧性基础设施项目执行官地面链路丢失场景;对手反太空学说
商业空间站Axiom Space;商业空间站运营商科研载荷团队;航天员NASA;商业伙伴;研究机构实验在轨处理;空间站遥测空间站运营ISS 过渡;商业空间站建设
主权数据存储受监管企业;国家政府合规 / 安全团队合规或 IT 预算超出地面司法辖区的异地数据备份CISO / 监管事务数据驻留法规;制裁风险
前沿 AI 实验室(长期)OpenAI、Anthropic、DeepMind 同类;超大规模云厂商ML 训练基础设施团队资本开支 / 基础设施预算不受电网约束的大规模批量 AI 训练基础设施 VP / CTO电网互联等待时间超过可接受期限;成本平价
远程 / 离网运营商北极 / 海事 / 灾害响应运营商;采矿、能源行业现场运营团队运营预算在供电受限或无电网环境中提供计算和连接现场运营 VP缺少电网;与卫星星座部署重叠

细分定义综合自 Cowboy Space FCC 申请、NewSpace Economy 行业调查(2026 年 5 月)和 SatNews ODC 分析。前沿 AI 实验室排在长期项,是因为延迟、数据重力和网络架构造成近期障碍,而太空原生工作负载没有这些障碍。SAIC 参与 Series B 被视作国防行业采用信号。

[CM019, CM020, CM021, CM022, CM024, CM025]
FM003: 买方分群矩阵——各分群采用 ODC 的准备度

各买方分群采用轨道数据中心服务的准备度和阻力;评级是有证据支撑的定性评估。

[CM019, CM020, CM021, CM022, CM025, CM037]

2.4 增长驱动与时点

最强增长驱动,是已有文件证明的地面电网并网积压。Construct Capital 和 Cowboy Space 的 FCC 申请都称, 主要市场的 AI 数据中心需求正在超过电网容量,形成一个持续多年的窗口期,让包括轨道方案在内的替代路径拥有结构性优势。 Goldman Sachs 预计,到 2030 年美国数据中心将使用美国 8% 电力,到 2028 年 AI 将占数据中心电力需求的 19%, 紧迫性进一步上升。EIA 的月度跟踪确认,截至 2026 年 3 月,美国总用电需求同比增长 4.5%,与数据中心建设叙事相符。 第二个驱动是发射成本下降。可重复使用运载火箭自 2010 年以来显著降低了每公斤 LEO 成本, Cowboy Space 的一体化上面级架构则试图通过消除火箭与载荷之间的边界,进一步压缩发射成本。 公司架构——变成数据中心的上面级——建立在商业发射竞争持续存在的前提上。第三个驱动是 NVIDIA 合作: Space-1 Vera Rubin 模块是 NVIDIA 在 2026 年 3 月 GTC 上发布的专用 LEO AI 加速器, Cowboy 与 NVIDIA 的先发合作释放出硬件生态已准备好的信号。 时点仍不确定。首次 Stampede 发射目标是 2028 年,且按 Cowboy 自己的 FCC 申请,卫星设计明确尚未完成。 电网瓶颈真实存在,但并非永久:公用事业正在投资,地面运营商也在把数据中心与可再生能源共址、使用液冷, 并推进核电购买协议。竞争窗口可能比当前排队情况暗示的更窄。[CM027, CM028, CM029, CM030, CM031, CM032]

增长驱动因素与采用约束
驱动因素 / 约束方向时间对 Cowboy Space 的含义尽调问题
AI 需求增长(2030 年数据中心用电增长 160%)顺风现在–2030 年证明对不受电力约束计算的需求;提高紧迫性跟踪超大规模云厂商因电力推迟资本开支的情况;关注绕过电网的替代方案
电网互联积压(排队数年)顺风2026–2029 年窗口为轨道供给创造直接替代机会量化各市场平均互联等待时间;跟踪 FERC 改革进展
发射成本下降(可重复使用运载器)顺风持续降低单颗卫星部署成本;让大型星座在经济上可行确认 Cowboy 自身发射成本预测;与 SpaceX Falcon 9 基准对比
NVIDIA Space-1 硬件可用性顺风2026–2028证明硬件生态可用;降低 Cowboy 芯片研发风险确认 Space-1 Vera Rubin 模块规格、抗辐射能力和交付时间表
热管理(真空中 1 MW)逆风发射前尚未解决;这一尺度的散热器质量和面积尚无验证要求热设计评审;询问与 Starcloud 已发布架构的对比
数据重力(大多数 AI 数据在地球)逆风结构性将轨道 TAM 限制在批量工作负载;排除交互式推理要求给出具体工作负载管线,展示端到端数据传输流
碎片与轨道安全风险逆风监管时间线在 700–1,000 km 部署 20,000 颗卫星会加剧拥挤的 LEO;监管审查升温跟踪 Secure World Foundation 和 ITU 对 FCC 案卷的意见;要求给出离轨处置计划
地面替代方案在改善(核电 PPA、共址)逆风2026–2030不上轨也能绕过电网;电力缺口可能比 Cowboy 发射更快收窄跟踪公用事业规模和核 SMR 时间表,与 Cowboy 首次发射日期(2028 年)对比

时间判断基于公开 FCC 申请、公用事业投资计划和公司披露里程碑,由分析师估算。「结构性」指受物理规律约束、不能靠排期解决。电网互联窗口根据 Construct Capital 和 Cowboy Space FCC 申请中的表述估算;截至 2026 年 6 月,无法公开获取各 ISO 的准确排队数据。

[CM001, CM005, CM009, CM028, CM029, CM034]
FM004: ODC 采用价值链——从电力瓶颈到轨道服务

展示从地面电网约束到轨道数据中心需求的因果链,以及 Cowboy Space 必须跨过的关键关口。

[CM001, CM005, CM009, CM028, CM034, CM036]

2.5 采用约束与怀疑视角分析

多个采用壁垒足以重塑整个市场论点。SatNews 编辑 Nick David 在 Cowboy Series B 后撰写详细分析, 指出公司必须同时解决四个未解瓶颈:发射供给、热管理、数据重力和单位经济性。热管理尤其不是简单问题: 在真空中耗散 1 MW 热量,需要把大型散热器指向寒冷太空;兆瓦级辐射冷却所需质量、面积和指向要求, 尚未在 Stampede 拟定规模上得到验证。真空散热的物理机制与地面冷却根本不同,也还没有运营中的 ODC 证明过 1 MW 热吞吐。 数据重力是第二个结构性约束。大多数 AI 训练数据源自地球,且已经存放在地面云基础设施中;把训练数据集送上轨道再传回的成本和延迟, 公开材料并未说明。低地球轨道会带来 10–500 ms 往返延迟,具体取决于架构;这会完全排除对延迟敏感的推理和交互式应用。 FCC 申请承认 700–1,000 km 高度、20,000 颗卫星星座存在碎片风险,SatNews 也指出,纯光学架构只能缓解该责任,无法消除。 Secure World Foundation 对 SpaceX 和 Starcloud FCC 申请的评论,要求分阶段授权、系统级风险分析,以及更清晰的任务后处置标准; 这些先例也可能影响 Cowboy 的监管路径。按 Cowboy 自己的 FCC 申请,卫星设计明确尚未完成,服务开始前还需要修改许可。 合在一起,这些约束让可信的空头情形并不难表述:在轨道基础设施达到商业规模之前,地面数据中心可能已经通过与可再生能源共址和核电 PPA 解决电力问题。[CM034, CM035, CM036, CM037, CM038, CM039]

2.6 展示项

Chapter 03

03竞争对手

3.1 直接轨道数据中心竞争者

四家公司正在直接轨道数据中心类别中与 Cowboy Space 赛跑:Starcloud、SpaceX、Blue Origin 和 Sophia Space。 它们都瞄准 AI 电力瓶颈论点,但架构、规模和发射策略各不相同。 Starcloud(前身为 Lumen Orbit)是运营成熟度最高的纯 ODC 初创公司。其首颗卫星 Starcloud-1 于 2025 年 11 月发射, 搭载一块 NVIDIA H100 GPU——这是首个在轨验证的高端 AI 加速器。该任务运行了 Google Gemma 语言模型的一个版本, 并在星上训练 nanoGPT 模型,证明 GPU 级推理在 LEO 物理上可行。Starcloud-2 是其首次商业任务,预计将提供 GPU 集群、 持久存储和 24/7 访问。2026 年 2 月,公司向 FCC 申请授权,在 600–850 km 的太阳同步轨道运营最多 88,000 颗卫星, 按数量计是第二大 ODC 申请。Crusoe Energy 于 2025 年 10 月宣布,将在计划于 2026 年底发射的 Starcloud 卫星上部署 Crusoe Cloud,这是首个公开宣布的轨道平台商业云租户。Starcloud 对 Cowboy 的领先是真实的:硬件已入轨、AI 工作负载已运行、 2026 年有商业任务计划、且已拿到主要云客户——这些都发生在 Cowboy 预计首次自研发射之前。 SpaceX 以完全不同的规模进入 ODC 讨论。2026 年 1 月 30 日,Space Exploration Holdings 向 FCC 提交 「SpaceX Orbital Data Center System」申请,计划建设最多 1 million 颗卫星的非地球静止轨道星座, 高度 500–2,000 km,主要依靠光学星间链路,并可能连接既有第一代和第二代 Starlink 系统。SpaceX 相比所有其他 ODC 进入者的结构优势很明确: 它控制 Starship,这是唯一能以划算成本把 Cowboy 计划中的 20 至 25 吨卫星送上天的飞行器; 它运营全球最大商业卫星星座;它拥有任何初创公司近期都无法匹配的制造规模和发射节奏。如果 SpaceX 将其 ODC 系统商业化, 它既可能向对手供应发射运力,也会争夺同一批工作负载。 Blue Origin 的 Project Sunrise 是第三个大规模进入者。Blue Origin 向 FCC 申请建设最多 51,600 颗卫星的系统, 分布在 500 至 1,800 km 的太阳同步轨道,通过光学链路和 TeraWave 宽带架构路由流量。Blue Origin 拥有 New Glenn 重型运载火箭和推进传承, 这带来 Sophia Space 和多数 ODC 初创公司缺少的发射独立性。Project Sunrise 把计算与连接捆绑, 可能提供 AI 数据中心加宽带的一体化组合,而 Cowboy 当前更窄的方案无法匹配。 Sophia Space 瞄准与 Cowboy 相同的晨昏太阳轨道论点,但硬件架构不同。其 Thermal Integrated LEO Edge(TILE)设计, 把计算模块放在产热点旁边的辐射散热板附近,相比传统数据中心热系统减少质量和复杂度。NVIDIA 把 Sophia Space 列入 Space-1 Vera Rubin 合作伙伴——这与 Cowboy 的供应商关系相同——但 Sophia 对模块化热设计的关注,直接对应轨道 AI 最难解决的问题之一。Sophia Space 尚未宣布发射日期,也未宣布与 Cowboy 已披露 $325M 资本等量的融资。 [CP001, CP002, CP003, CP005, CP006, CP007]

轨道数据中心竞争对手画像
竞争对手类别规模 / 融资状态(2026 年 6 月)目标细分相对 Cowboy 的差异点相对 Cowboy 的局限
Starcloud直接 ODC – 专用计算卫星FCC:88,000 颗卫星;融资未披露H100 已在轨(2025 年 11 月);Starcloud-2 商业任务计划 2026 年执行地球观测、主权云、AI 推理硬件成熟度领先 2 年以上;已锁定 Crusoe Cloud 合作不拥有发射能力;依赖拼车发射供给
SpaceX 轨道数据中心系统直接 ODC – 超大规模星座FCC:100 万颗卫星;SpaceX 私有,已融资数十亿美元2026 年 1 月 30 日提交 FCC 申请;尚无运营服务AI 超大规模云厂商、SpaceX 内部使用拥有 Starship;Starlink 制造规模;频谱位置没有第三方商业发射承诺;监管未知
Blue Origin Project Sunrise直接 ODC – 大型航空航天进入者FCC:51,600 颗卫星;Blue Origin 私有FCC 申请已有报道;尚无运营服务AI 计算 + TeraWave 宽带捆绑拥有 New Glenn 发射能力;整合 TeraWave 连接FCC 申请状态未确认;硬件规格未公开
Sophia Space直接 ODC – TILE 模块化热架构收入前未披露;尚无星座 FCC 申请设计阶段;已列名 NVIDIA Space-1 伙伴面向卫星和地球观测的 AI 计算用 TILE 设计直接处理热管理问题没有已发射硬件;未披露达到 Cowboy 规模的融资
Axiom Space / Spacebilt相邻 ODC – 空间站托管计算Axiom 已融资约 USD 1.5B;Spacebilt 未披露AxDCU-1 在 ISS 上运行(2025 年秋);2 个专用节点于 2026 年 1 月 11 日发射航天器运营商、航天员、地球观测研究人员ISS 上已有运行硬件;供应商生态具备资质受 ISS 容量限制;不是兆瓦级 AI
Kepler Communications相邻 ODC – 光中继网络上的计算Series B;星座已部分部署10 颗卫星搭载 40 个 Jetson Orin 模块;商业中继服务已上线地球观测、LEO 连接、边缘 AI商业中继服务已产生收入计算密度低于 GPU 级卫星
Lonestar Data Holdings专项公司 – 太空数据存储与主权未披露;轨道载荷已飞行StarVault 容量预留已开放,面向 2026 年 10 月发射企业、政府、灾备买方存储产品已商业运行;覆盖月球没有 GPU 级 AI 计算;市场不同于 Cowboy
CoreWeave / Crusoe / Voltage Park地面替代 – 专用 AI GPU 云CoreWeave:估值约 $23B(2025 年);Crusoe 和 VP 未披露已用最新 NVIDIA GPU 商业运营AI 实验室、模型开发者、企业训练和推理今天即可使用;延迟更低;硬件可获得;价格有竞争力没有轨道计算能力;受电网和土地约束

状态、融资和星座规模来自公司网站、FCC 公开记录和截至 2026 年 6 月的行业媒体;部分竞争对手融资额未披露。CoreWeave 估值来自上一轮融资报道。

[CP001, CP003, CP006, CP009, CP011, CP013]
FP001: 竞争定位图——发射自给能力 vs. 计算部署成熟度

Cowboy Space 位于右上目标象限(自研火箭,但尚未部署),区别于依赖拼车发射的 ODC 初创公司和商业成熟的地面替代方案。

X 轴(发射自给能力)是有证据支撑的序数评分:0 = 完全依赖第三方拼车发射;10 = 拥有可运营重型运载器。Y 轴(计算部署成熟度)同样是序数评分:0 = 仅有概念 / 申报;10 = 商业 AI 云服务。评分来自截至 2026 年 6 月的公开 FCC 申报、公司公告和行业媒体。

[CP001, CP006, CP008, CP009, CP013, CP019]

3.2 邻近平台与专门型竞争者

第二层竞争者把既有轨道平台——空间站、运营中的通信卫星、光学中继网络——用作计算节点。这些玩家不需要从零建造运载火箭或新航天器, 运营时间线快于任何从零建设的 ODC 初创公司。 Axiom Space 是最先进的平台型竞争者。2025 年秋,它在 International Space Station 上部署 Data Center Unit-1(AxDCU-1), 基于 Red Hat Device Edge 支持云计算、AI、机器学习、数据融合、存储和太空网络安全。2026 年 1 月 11 日, Axiom 将首两个专用轨道数据中心节点发射至 LEO,与 Kepler Communications 光学中继网络星座首批部署同步。 这些节点面向卫星、星座和航天器,设计用于支持云化数据存储和处理。Axiom 与 Spacebilt 于 2025 年 9 月宣布合作, 计划在 2027 年交付 AxODC Node ISS,硬件来自 Spacebilt(Large In-Space Servers)、Skyloom(光学)、 Phison Electronics(Pascari SSD)和 Microchip Technology(PIC64 航天计算)。 这些合作说明,轨道边缘计算正在形成供应商生态;Cowboy 的一体化上面级模式绕过了该生态,但最终仍要为同一批早期采用客户竞争: 卫星运营商、地球观测公司和空间站研究人员。 Kepler Communications 提供光学中继原生计算。其星座在 10 颗卫星上部署 40 个 NVIDIA Jetson Orin 模块, 每个模块都是支持 AI 和加速工作负载的计算节点,嵌在商业运营中的光学数据中继网络里。对地球观测卫星和带宽受限的航天器而言, 在 Kepler 既有基础设施上做在轨处理,可能比等待未来 Cowboy 星座达到规模更快交付结果。Kepler 的模式 ——把计算作为网络功能,而不是专用数据中心——直接争夺 Cowboy 需要先赢下的边缘工作负载细分市场。 Lonestar Data Holdings 瞄准存储和韧性细分市场,而不是 GPU 级 AI 计算。其 StarVault 平台把太空数据存储与加密密钥托管结合, 定位为超出地面触达范围的主权级、灾备级备份。Lonestar 正接受其首个轨道 StarVault 的容量预订,发射时间为 2026 年 10 月。 Cloud Constellation Corporation 运营的 SpaceBelt 至少自 2020 年以来也在推进类似的存储与连接论点,但商业规模有限, 说明轨道数据市场执行难度很高。这些专门型公司争夺安全存储和数据主权细分市场,Cowboy 偏计算的定位并不直接覆盖该市场; 但它们也说明,近期轨道数据客户范围有多窄。 Spacebilt 明确把自己描述为类别定义者:其网站称,公司希望「定义」太空多租户 AI 计算类别,而不是在类别内竞争, 这表明它把自己视为衡量其他玩家的基础设施标准。Spacebilt 的多租户服务器设计,把可在轨维修系统与边缘 AI 计算结合, 代表了一条不同于 Cowboy 纵向整合单运营商上面级的市场进入路径。 [CP013, CP014, CP015, CP016, CP017, CP018]

功能与能力对比矩阵
能力Cowboy SpaceStarcloudSpaceX ODCBlue Origin SunriseSophia SpaceAxiom / Kepler(边缘)CoreWeave / Crusoe
拥有发射载具在建设(目标 2028 年)否(拼车发射)是(Starship)在建设(New Glenn)N/A(地面)
GPU 级硬件在轨否(计划 2027 年演示)是 – H100 自 2025 年 11 月起在轨否(仅申请)否(仅申请)是(Kepler 上有 Jetson Orin;ISS 上有 AxDCU-1)是 – 大规模地面 GPU 集群
已在轨演示 AI 工作负载是 – Gemma 推理和 nanoGPT 训练部分(仅边缘推理)是 – 生产级 AI 服务
FCC 星座申请是 – 20,000 颗卫星(2026 年 5 月)是 – 88,000 颗卫星(2026 年 2 月)是 – 100 万颗卫星(2026 年 1 月)是 – 51,600 颗卫星(已报道)尚未否(Kepler 单独获许可)N/A
单节点兆瓦级太阳能供电是(计划中)– 每颗卫星 1 MW未知(小卫星形态)未知(规格未披露)未知(规格未披露)是(TILE 按 MW 级设计)否(kW 级)N/A(电网供电)
NVIDIA Space-1 / GPU 级计算合作是 – 已命名 Vera Rubin Space-1 Module是 – NVIDIA 伙伴和支持方UnknownUnknown是 – NVIDIA Space-1 伙伴是(Axiom)– NVIDIA Space-1 伙伴是 – 最新 NVIDIA 地面 GPU
商业服务可用(截至 2026 年 6 月)尚未(Starcloud-2 计划中)部分 – Axiom ISS 节点已上线是 – 完全商业化
垂直整合(发射 + 计算 + 电力)是(建设中)– 火箭 + 上面级数据中心 + 太阳能否(仅计算)是 – 但 ODC 任务从属于 Starlink部分 – 发射 + ODC;无专用计算细节否(仅计算 + 热管理)N/A

状态截至 2026 年 6 月,来自公司网站、FCC 申请和行业媒体。未知 = 未被公开确认。单元格反映现有最佳证据;空白单元格代表真实信息缺口,而非隐含否定。

[CP001, CP006, CP009, CP011, CP016, CP026]
FP002: 能力覆盖图——轨道与地面计算竞争者

Cowboy Space 把发射所有权和 MW 级算力规划放在一起,这一点很独特;但硬件成熟度落后于 Starcloud,短期商业可用性也不如地面替代方案。

未知单元格反映公开记录确实存在信息缺口,不代表能力不存在。状态截至 2026 年 6 月。

[CP001, CP006, CP009, CP011, CP012, CP016]

3.3 地面 AI 云替代方案

对 Cowboy 所称市场——受电力约束的 AI 算力——最直接的竞争威胁,不来自轨道对手,而来自今天已经运营的地面替代方案。 CoreWeave、Crusoe、Voltage Park 和 Stargate 项目合起来服务 Cowboy 想要捕获的工作负载, 延迟更低、成本更低,也没有发射风险。 CoreWeave 自称全球 #1 AI 云平台,通过 Kubernetes 原生环境提供 GPU 计算,下一代基础设施包括 NVIDIA Vera Rubin NVL72 机架级系统。它提供快速推理启动、托管 Kubernetes 和 Slurm 集群,以及企业级 24/7 支持。 CoreWeave 今天已经商业可用,客户包括主要 AI 模型开发者和企业。作为地面运营商,CoreWeave 没有 Cowboy 的延迟挑战, 能够物理接触并升级硬件,也能基于已知基础设施成本签署多年合同——这些优势至少会持续到 2030 年代初。 Crusoe 提供 AI 云基础设施,声称正常运行时间为 99.98%,硬件包括 NVIDIA 和 AMD 的 GB200 NVL72、HGX B200 和 MI355x; 其托管推理服务称可让首 token 时间最高快 9.9×。Crusoe 还宣布将在 2026 年底把 Crusoe Cloud 部署到 Starcloud 卫星上——它因此既是地面替代者,也是选择 Starcloud 而非 Cowboy 的轨道计算客户。这一合作释放出的信号是: 愿意尝试轨道计算的领先 AI 云运营商,已经承诺给了 Cowboy 的竞争者。 Voltage Park 运营超过 36,000 个 NVIDIA Hopper、Blackwell 和 Grace Blackwell GPU,按需定价从 $1.99/hr 起, 专用预留合同期限为 12+ 个月。其客户包括领先 AI 实验室和快速扩张初创公司。Voltage Park 的价格点和即时可用性, 会直接压制未来任何 Cowboy 定价模型:如果轨道计算无法低于或显著优于 $1.99/hr H100 访问,太空 AI 的经济性论证就完全依赖那些地面访问不可用或被禁止的用例。 Stargate Project 宣布计划在四年内投资 $500 billion 建设美国 AI 基础设施,由 SoftBank、OpenAI、Oracle 和 MGX 支持, 并从 Texas 开始立即部署 $100 billion。Arm、Microsoft、NVIDIA、Oracle 和 OpenAI 是关键技术伙伴。传统意义上, Stargate 不是 Cowboy 的竞争者——它建设地面数据中心——但它吸收的正是 Cowboy 融资叙事依赖的需求信号。 随着 Stargate 和类似超大规模地面项目爬坡,电力瓶颈论点的紧迫性会下降,并网时间线会缩短,轨道计算的相对论证会变弱。 这是对 Cowboy 中期需求论点最不利的外部战略进展。 AWS Ground Station 把云基础设施延伸到卫星运营,但不把计算放到轨道;Azure Orbital 提供类似的地面卫星处理服务。 两者都不是直接的轨道计算竞争,但都会降低客户为避开地面基础设施支付的溢价:任何能够通过 AWS 或 Azure 容忍地面处理的任务, 都不需要 Cowboy 的轨道计算服务。 [CP019, CP020, CP021, CP022, CP023, CP024]

定价与包装对比
公司服务模式价格 / 合同模式服务可用性对 Cowboy 定价的含义
CoreWeaveAI 原生云 GPU 计算(H100、Vera Rubin NVL72)按需和预留;企业合同已商业运营Cowboy 必须把价格压到低于或接近地面 GPU 云费率,才能吸引可容忍延迟的工作负载
CrusoeAI 云(GB200 NVL72、HGX B200、AMD MI355x);托管推理预留和按需;企业 SLA(99.98% 正常运行时间)已商业运营Crusoe 与 Starcloud 已承诺的 ODC 合作设定了 Cowboy 必须匹配的可比定价基准
Voltage ParkH100/Blackwell/Grace Blackwell GPU 云$1.99/hr 按需;12+ 个月专用预留已商业运营$1.99/hr 的地面 H100 底价压制轨道溢价;Cowboy 需要清晰的价值差异
Starcloud轨道 GPU 计算和主权云(GPU 集群 + 存储)定价尚未公开;商业任务计划 2026 年执行尚未上线(计划 2026–2027 年)首个轨道价格信号;主权和靠近传感器的低延迟可能带来溢价
Axiom Space(AxDCU-1)ISS 托管边缘计算(云、AI、边缘处理)定制企业合同;未公开列价已在 ISS 运营(2025 年秋)面向太空原生客户的平台定价;主权 / 国防溢价
Lonestar Data Holdings太空主权数据存储(StarVault)容量预留模式;定价未公开已开放预留;首次发射计划 2026 年 10 月存储 / 备份细分场景;不直接竞争 AI 计算
Cowboy Space (Stampede)垂直整合的轨道 AI 计算(每颗卫星 1 MW / 约 800 块 GPU)未公开定价;商业服务目标为 2028 年以后商业化前;FCC 申请仍在审查未知;必须低于预计地面短缺成本,或瞄准地面无法触达的用例

截至 2026 年 6 月,没有 ODC 玩家公布轨道计算定价;所有轨道玩家都仍处于商业化前或早期商业阶段。Voltage Park 按需定价来自公司网站。地面替代方案是目前唯一可用的价格锚。

[CP019, CP020, CP021, CP022, CP017]

3.4 竞争定位与护城河分析

Cowboy Space 主张的主要护城河是纵向整合:掌握火箭、上面级、计算载荷、电力系统和星座后, 它能以纯卫星 ODC 竞争者做不到的方式控制成本和节奏。TechCrunch 确认,Cowboy CEO Baiju Bhatt 曾专门接触多家发射服务商, 并得出结论:商业发射市场过于稀缺且昂贵,轨道数据中心若没有内部发射能力,就无法在规模上实现有竞争力的单位经济性。 鉴于当前 Starship 尚未面向第三方客户可用,以及 SpaceX 自己的 ODC 意图,这一结论在结构上成立。 但护城河论点有明显弱点。第一,作为差异化亮点宣布的 NVIDIA Space-1 Vera Rubin 模块合作,并非 Cowboy 独占; Starcloud、Sophia Space 和 Axiom Space 同时也拥有该合作。NVIDIA 把四家公司都列为 Space-1 合作伙伴, 说明它没有对 Cowboy 做独家押注。第二,Cowboy 自己的 FCC 申请明确称卫星设计尚未完成,服务开始前需要修改许可。 对一家声称在纵向整合上拥有先发优势的公司而言,这一承认并不寻常。第三,SpaceX 的 ODC 申请规模为 1 million 颗卫星, 远高于 Cowboy 的 20,000 颗计划;SpaceX 还拥有现成 Starship 飞行器优势,Cowboy 在研火箭至少五到七年内难以匹配。 按 FCC 申请规模排序,Cowboy 位居第三:SpaceX 为 1 million 颗卫星,Starcloud 为 88,000,Blue Origin 为 51,600, Cowboy 为 20,000。较小星座如果率先以兆瓦级计算节点进入市场,仍可具备商业可行性;但 Starcloud 已有硬件部署先发, 说明 Cowboy 守的是时点,而不是领先地位。正面表述——20,000 颗比 1 million 或 88,000 更有望在近期达成——是成立的, 但需要目前尚不存在的执行记录支撑。 Cowboy 的发射自有策略,也是一把用于竞争护城河的双刃剑。要做成有竞争力的轨道数据中心业务,公司必须同时在多个领域取得突破: 火箭开发(许多资金更充足的公司都失败过的类别)、真空中 1 MW 计算负载的热管理、NVIDIA GPU 太空级认证、 光通信网络,以及客户获取。任何单一领域执行失败,都会让纵向整合论点整体坍塌;而 Starcloud 借助既有运载火箭的模式, 每次只暴露一个执行风险。 Cowboy 最强的结构性护城河情形,是其整合系统率先以有竞争力成本在轨道证明兆瓦级 AI 训练,并由此形成网络效应锁定。 客户围绕特定轨道计算平台搭建数据管线、模型和工作流后,会产生切换成本——但前提是平台已经存在且商业化运营; 最早也预计要到 2028 年底以后。 [CP026, CP027, CP028, CP029, CP030, CP031]

护城河耐久度与竞争风险登记表
护城河主张具体竞争威胁严重程度尽调问题 / 缓释路径
垂直整合(自有火箭 + 计算)消除发射供给约束SpaceX 可能永远不会为 20 吨级 ODC 载荷提供第三方发射;没有火箭的 ODC 竞品创业公司仍会卡在发射瓶颈护城河价值高——但前提是 Cowboy 成功造出有竞争力的运载火箭核验推进系统测试里程碑、FAA 许可时间线,以及相对 Falcon 9 等效方案的单次发射成本目标
NVIDIA Space-1 Vera Rubin 合作释放计算差异化信号Starcloud、Sophia Space 和 Axiom Space 也共享该合作;NVIDIA 并未独家绑定 Cowboy低(非独家)——NVIDIA 硬件会成为轨道计算里的商品化部件不要把 NVIDIA 合作当作差异化主张;应转而识别软件或工作负载集成层面的护城河
20,000 颗卫星规模相对 Axiom / Kepler 边缘节点形成计算密度优势SpaceX(1M)和 Starcloud(88K)申报的星座更大;Cowboy 是巨型星座进入者里规模最小的一家中等——相对边缘玩家有规模优势,但资源雄厚的同行规模更大跟踪四项 FCC 申请的处理时间线;监管结果可能压缩规模主张
专有的“上面级即数据中心”设计降低集成开销Cowboy 自己的 FCC 申请写明,卫星设计尚未完成,投入服务前需要修改许可证高风险——FCC 申报阶段设计未完成,意味着还要承担数年的额外开发风险要求完成硬件 CDR(关键设计评审),再升级至买入 / 跟踪
先发轨道 AI 计算优势带来客户锁定Starcloud 已经在轨演示 AI 工作负载;Axiom 的 ISS 硬件已经上线;Cowboy 最早也要到 2027 年才有首次演示高风险——Cowboy 不是先发者;在硬件部署上,它排在 Starcloud、Axiom / Kepler 以及潜在其他玩家之后,至少第四澄清客户管线——是否有地球观测或国防客户明确承诺使用 Cowboy?
地面 AI 电力瓶颈为轨道计算创造持续结构性需求在 Cowboy 首次商业发射前,Stargate($500B)、电网升级、核能协议和能效更高的芯片可能削弱电力瓶颈的紧迫性中等风险——需求论点今天成立,但必须在 2028–2030 年窗口得到验证;地面建设正在加速每年跟踪地面电网并网排队进展和核能数据中心协议

严重程度评估是作者基于来源证据作出的判断;不是量化评分。护城河主张改写自 Cowboy Space 的 FCC 申请和投资人沟通。

[CP026, CP027, CP028, CP029, CP030, CP031]
FP003: 竞争护城河与就绪度 KPI——Cowboy Space 对比直接 ODC 同业

Cowboy Space 的垂直整合论点在结构上站得住,但尚未验证;硬件成熟度落后 Starcloud 2 年以上,不过在规划算力密度和发射独立性野心上领先。

[CP001, CP003, CP005, CP006, CP009, CP027]

3.5 负面证据与竞争风险评估

评估 Cowboy Space 时,几组反向竞争信号必须给足权重。第一,唯一公开承诺与轨道计算合作的商业云运营商 Crusoe Energy,首个太空部署选择了 Starcloud,而不是 Cowboy。未来 Cowboy 仍可能拿到合作,但这个选择已经说明,早期商业客户正在验证哪一个平台。 第二,Cowboy 同时推进火箭研发、航天器设计、热管理、光通信和客户获取,执行曲线很像那些历史上持续超支、延期的公司。TechCrunch 用「也很疯狂」形容这件事,措辞夸张,却点中了真实风险:在西方,只有 SpaceX、Rocket Lab 和 Arianespace 能稳定发射商业火箭,而且每家公司都花了十年甚至更久才跑出运营节奏。截至报告日期,Stoke Space、Firefly Aerospace 和 Relativity Space 已经做了多年,仍未交付可运营系统。 第三,Stargate 建设以及 SoftBank、OpenAI、Microsoft 和 Oracle 出资的类似地面超大规模项目,正面切入 Cowboy 核心价值主张里的电力和土地瓶颈。如果 2026–2032 年间,地面电力问题靠核能、分布式电网协议,或电力充足地区的新数据中心区域解决,轨道计算的论证就只剩边缘时延和数据主权,市场更窄、收入更低,也更难销售。 第四,太空计算竞争格局正在向 NVIDIA 硬件这个共同平台收敛。NVIDIA 发布面向 LEO 的 Space-1 Vera Rubin 模块,一边验证了市场,一边把硬件差异做成了商品化。如果每个严肃的 ODC 玩家都用同一批 NVIDIA GPU,竞争就会转向发射成本、热效率、轨道选择、网络拓扑和软件——这些恰恰是 Cowboy 最缺乏验证的主张。 第五,SpaceBelt(Cloud Constellation Corporation)提供了一个警示先例:一家专做轨道数据存储、拥有可信合作方联盟(IBM、SSTL、MindCraft、若干银行业参与方)的公司,至少从 2020 年起运营至今,仍未达到商业规模。多次尝试都显示,轨道数据基础设施总是比最初预测花更久、成本更高。 [CP005, CP008, CP023, CP029, CP030, CP031]

3.6 展示材料

Chapter 04

04财务

4.1 融资历史与资本结构

Cowboy Space Corporation 自 2024 年末以 Aetherflux 创立以来,完成了三轮不同融资,最终在 2026 年 5 月以 $2B 投后估值完成 $275M Series B。CEO Baiju Bhatt 确认,截至 Series B 关闭时总资本约 $365M,意味着公司创立时还从 Breakthrough Energy Ventures、Andreessen Horowitz、NEA 和天使投资人处筹集了约 $40M 种子及过桥资金。2025 年 4 月由 Index Ventures 和 Interlagos 领投的 $50M Series A,支持了公司在 Aetherflux 品牌下的太空太阳能阶段。Series B 由 Index Ventures 共同领投,IVP、Blossom Capital、SAIC 以及所有主要既有投资人参与,资金用于转向轨道数据中心和自有运载火箭。Fenwick and West LLP 确认其代表 Cowboy Space 参与 Series B,提供了独立法律佐证。SAIC 作为新战略投资人的参与值得注意:SAIC 是美国最大的国防 IT 承包商之一,其出资释放了未来政府和军事应用采购轨道计算的潜在兴趣。公司还在 Series B 前获得美国国防部 Operational Energy Capability Improvement Fund(OECIF)的一份概念验证无线输电演示合同——这是唯一公开确认的政府收入信号。此前政府合同和拨款在公开披露中未量化,非风险投资收入的真实规模仍不清楚。资本表完全由股权融资构成;没有公开披露债务、可转债或项目融资工具。[CI001, CI002, CI003, CI004, CI005, CI006]

资本充足性——融资摘要与跑道情景
项目数值备注 / 来源
累计融资额(所有轮次)~$365MCEO Bhatt 称截至 Series B 交割(2026 年 5 月)约 $365M;包括种子轮、Series A($50M)、Series B($275M)
Series B 募资额(主要关注)$275M2026 年 5 月交割;Index Ventures 领投。来源:BusinessWire 官方新闻稿
投后估值(Series B)$2.0BBusinessWire 新闻稿确认,CNBC 也有报道;由轮次规模和所有权结构隐含
估计月度 burn——低情景$3M/month50 人团队,完全负担平均成本 $250K + 设施 + 轻 capex;符合飞行前航天阶段
估计月度 burn——高情景$10M/month火箭 R&D 加速 + 卫星原型采购;可比早期 Relativity Space 支出
估计跑道——低 burn($3M/mo)~92 个月(~7.7 年)$275M ÷ $3M/mo;上限假设增量 capex 极少
估计跑道——基准 burn($5M/mo)~55 个月(~4.6 年)$275M ÷ $5M/mo 中位情景;对当前开发阶段更现实
估计跑道——高 burn($10M/mo)~27 个月(~2.3 年)$275M ÷ $10M/mo;若火箭开发在产生收入前加速,这是底部情景
已披露债务 / 项目融资未披露截至 2026 年 6 月,任何公开来源均未提及信贷额度、债券或可转债
下一轮触发点(估计)首次火箭静态点火或卫星演示(推断)未披露明确触发点;航天公司典型 Series C 关卡包括硬件里程碑或首个客户

所有 burn rate 和跑道估计都来自人数目标、航天行业基准和可比公司支出的推断;Cowboy Space 未披露财务数据。跑道数字假设截至运行日,完整 $275M Series B 都是可用现金余额,且没有收入抵消 burn。实际跑道可能因未报告的既往支出、资本效率和政府合同收入而大幅不同。

[CI001, CI004, CI007, CI015, CI027, CI028]
FI003: 财务估算区间——烧钱、跑道与资本开支场景

基于人头基准和可比航天项目成本推导有来源支撑的烧钱率与跑道区间;所有数值均为推断估计,并非公司确认披露。

所有区间都是分析估计,来自公开可比公司、NVIDIA 硬件定价类比、SpaceX 发射定价和媒体报道的可比项目成本。没有任何 Cowboy Space 财务披露支撑这些估算。星座资本开支区间采用每星 $2.5M–$10M 的成本,区间很宽,反映航天级 GPU 生产经济性和发射成本轨迹存在巨大不确定性。

[CI027, CI028, CI029, CI035, CI016]

4.2 收入模式与商业化路径

截至 2026 年 5 月 Series B 关闭,Cowboy Space 未披露任何收入、ARR、客户合同、意向书或财务预测。按任何公开口径,公司都明确处于收入前阶段。规划中的收入模式,是向地球观测分析、国防、商业 AI 训练和主权数据合规客户出租轨道数据中心容量——计算时间和电力。公司不卖硬件,而是打算按每计算单元或每千瓦时收费,类似云基础设施定价,但要适配 LEO 间歇接入窗口和热约束。Series B 新闻稿列出三类优先资金用途:火箭发动机研发、卫星内嵌计算集成,以及监管合规(FCC 和频谱许可)。公司没有发布定价表、标价或参考客户试点。政府和国防合同——由 SAIC 战略投资及此前 OECIF 概念验证锚定——是近期最可信的收入路径,因为国防采购基础设施成熟,也更能容忍新兴技术风险。来自超大规模 AI 客户的商业收入,需要证明相较地面托管具备时延、可靠性和成本优势;这些优势目前均未得到独立验证。Construct Capital 博文把公司的市场切入点定义为解决「能源瓶颈」,说明早期收入论点围绕那些无法在所需时间内获得地面电网容量的客户。没有已披露合同或试点,收入模式只能被视为未解题。[CI011, CI012, CI013, CI031, CI032, CI036]

收入流——当前状态与尽调问题
收入流机制单位 / 定价基础当前状态收入质量尽调问题
轨道计算容量租赁客户从卫星托管的数据中心租用 GPU 小时或 kWh按 GPU 小时或按 kWh(未设定)收入前;未披露客户未决问题——模式尚未验证要求提供定价模型、beta 客户 LOI,以及容量预订结构
政府 / 国防计算合同与 DoD 或情报界签双边合同,提供韧性边缘计算合同金额(未披露)已拿到 OECIF 概念验证;SAIC 投资释放管线意图未决问题——金额未披露要求提供 OECIF 合同金额、SAIC MOU 或采购框架、SBIR/STTR 获奖历史
发射即服务(拼车 / 专属发射)专有火箭向第三方提供载荷发射,用于摊薄发射成本按 kg 发射成本(未设定)开发阶段;尚无可运营火箭推测性——取决于火箭能否完成要求提供火箭开发时间线、目标单 kg 成本,以及已识别的锚定发射客户
技术许可 / IP 版税许可轨道计算集成、热管理或能量传输 IP 专利版税率(未披露)未披露许可项目高度推测确认 IP 组合范围及任何许可洽谈
能源 / 能量传输服务(历史业务)Aetherflux 太空太阳能时期的残余业务;现在已弱化按传输至地面的 kWh(未披露)2026 年 5 月更名为 Cowboy Space 后明确降级优先级不在当前路线图上确认是否确认任何 OECIF 资助演示收入,以及未来如何处理

截至 2026 年 6 月,所有收入流都处于商业化前。状态评估来自新闻稿、投资人沟通和媒体报道的推断;公司未披露财务预测或客户名单,无法佐证。

[CI011, CI013, CI031, CI036]
定价与商业化——已知与推断
定价维度标价 / 参考价来源 / 依据置信度尽调问题
轨道计算(GPU 小时)未披露;推断较地面方案溢价 >$5–$10/GPU-hr基于太空硬件成本溢价(地面 3–10x)和典型云 GPU 定价(H100 $2–$4/hr)推断低(估计)要求提供任何已发布定价,或向潜在客户 / beta 伙伴报价的记录
轨道计算(按交付 kWh)未披露;地面超大规模数据中心参考价:$0.04–$0.08/kWh轨道太阳能供电消除运营期电费,但增加 capex 摊销低(估计)澄清收入模型是按能源、按计算,还是混合模式
发射服务(按 kg 至 LEO)未披露;SpaceX Falcon 9 参考价:约 $2,500/kg;Starship 目标:约 $100/kg行业基准(公开可得的 SpaceX 定价)N/A(当前阶段仅内部发射)要求提供目标单 kg 发射成本,以及任何展示发射经济性的内部财务模型
政府合同定价OECIF 合同金额未披露;典型 DoD 概念验证合同区间为 $500K–$5MOECIF 项目结构;一般 DoD 采购惯例低(估计)确认 OECIF 合同金额,以及任何后续 SBIR/STTR 或 IDIQ 授奖

所有定价数字要么公司未披露,要么来自行业可比项估计。没有公开价目表、条款清单或客户提案。置信度反映底层推断质量,而非已确认的公司数据。

[CI012, CI020, CI024, CI032]
FI001: 收入模型桥——从客户活动到毛利润

展示潜在客户需求到实际轨道算力收入之间的连续关口,指出出现任何毛利时点前必须跨过的四个关键依赖。

所有节点都基于 Stampede 星座公开描述和可比轨道算力项目推断运营模型。截至 2026 年 6 月,没有软件栈、价格表或在轨运行卫星;这座桥只是说明依赖链,并非已确认的产品流程。

[CI036, CI017, CI018, CI011]

4.3 成本结构与资本强度

Cowboy Space 的成本结构即便放在资本密集的航空航天行业里也不寻常:它必须同时资助火箭推进研发、卫星平台开发、航天级 GPU 计算集成、20,000 颗卫星星座的 FCC 监管合规,以及同时服务火箭和卫星的西雅图实体制造中心。单独任何一项,都是多年期、数亿美元级项目。Relativity Space、ABL Space Systems 和 Stoke Space 等可比中小型运载火箭项目,在达到首次运营里程碑前消耗了 $100M–$650M,其中数家公司根本没有达到。NVIDIA 的 Space-1 Vera Rubin LEO 优化 GPU 模块可能是硬件基石,按每颗卫星低于 1 MW 的电力预算和抗辐射运行设计,但其航天级采购成本估计为地面超大规模等价产品的三到十倍——仅 GPU 基础设施就可能达到每颗卫星 $750K–$3M。ArsTechnica 对轨道数据中心物理的分析指出,在真空中为 1 MW 计算载荷做辐射冷却,需要约 200 平方米散热器表面;这是一项尚未在生产规模验证的工程约束。若完整部署 20,000 颗卫星星座,并按制造和发射成本每颗约 $5M 估算,总资本开支约需 $100B——大约是当前已融资额的 50 倍。因此,公司近期资本计划只能是分阶段楔入:用 $275M Series B 做出可信的火箭加卫星概念验证,再用已展示的能力撬开项目融资、主权资本或后续股权轮,支撑星座扩张。公开信息没有确认这套分阶段融资策略;它是从当前资本与既定野心之间的规模错配推断出来的。[CI014, CI016, CI017, CI018, CI019, CI021]

单位经济——可用指标与缺口
指标数值 / 区间置信度重要性尽调问题
毛利率(轨道计算)null — 未披露N/A决定轨道计算相对地面替代方案是否具备结构性盈利能力要求提供按单颗卫星和按星座计算的贡献毛利模型;纳入已摊销 capex
CAC(获客成本)null — 未披露;没有客户可供测算N/A对建模任何 go-to-market 场景下的回本周期和销售效率至关重要要求提供任何 LOI 或管线数据,用于推导早期 CAC 代理指标
回本周期null — CAC 和 ACV 均未披露N/A决定资本回收效率,以及每个客户是否能在流失前覆盖获客成本一旦 CAC 和合同金额披露即可推导
单颗卫星 GPU 计算成本(硬件)$2.25M–$30M(估计,按地面 H100 机架 $250K–$400K 的 3–10x;太空级溢价)低(基于 NVIDIA Space-1 产品线和太空认证成本倍数估计)为每个创收单元的 capex 设下底线,并直接牵引单位毛利向 NVIDIA 或 Cowboy Space 索取 NVIDIA Space-1 模块定价;验证 3–10x 倍数假设
单颗卫星发射成本(拼车场景)$5M–$25M,取决于质量和运载火箭(Falcon 9 约 $2,500/kg;2,000–10,000 kg 卫星级别)中等(基于公开可得的 SpaceX Falcon 9 拼车和 Transporter 定价)每个已部署创收单元的两大可变成本之一;影响盈亏平衡时间线确认卫星质量目标和计划使用的运载火箭;与舱单定价交叉核验
轨道计算热约束(冷却上限)1 MW 计算载荷约需 ~200 m² 散热器(基于 ArsTechnica 物理分析估计)中等(基于辐射冷却物理;未经独立航天工程师验证)为单颗卫星计算密度设置硬物理上限;生产规模尚未验证委托独立热工程评估 1 MW 卫星计算载荷

所有单位经济要么不可得(null),要么来自行业可比项和物理基准估计。太空级硬件成本估计假设相对地面等效方案有 3–10x 溢价,该假设来自公开报道的抗辐射加固和认证成本。热约束估计来自 ArsTechnica 的物理分析,未经独立验证。Null 值代表需要直接财务尽调的结构性缺口。

[CI017, CI018, CI019, CI020, CI025, CI038]
FI004: 资本强度图——Series B 估计分配与累计资本需求

示意 $275M Series B 在主要项目桶中的分配,以及当前资本与完整星座建成之间的数量级缺口;所有分配均为估计。

所有行项目都是示意估算,依据公司披露的资金用途类别(火箭研发、算力集成、监管合规)、公开航天披露中的可比项目成本,以及公司的招聘计划。没有公开可得的 Cowboy Space 财务报表或已批准预算。末端“剩余资本”数字只是基准烧钱场景下约 18 个月跑道的示意,不是已确认现金余额。

[CI013, CI014, CI017, CI021, CI035]

4.4 单位经济与现金续航情景

Cowboy Space 未公开披露任何单位经济数据——毛利率、CAC、回本期、净收入留存,或单星贡献利润。本章所有估算都来自可比公司和公开硬件成本基准,必须通过直接财务尽调验证。基于人数的烧钱建模,是近期最主要的财务视角:到 2026 年底计划 40–60 名员工,在航空航天语境下(硬件工程师、火箭科学家、监管法律顾问、软件团队),每人全成本薪酬 $250K–$350K,意味着仅工资负担每月约 $1.0M–$1.75M。加上场地成本、仿真用云计算、测试硬件采购以及监管 / 法务支出,当前团队规模下合理月运营烧钱区间为 $3M–$8M;随着开发项目加速,可能升至 $8M–$15M。以 $275M Series B 计算,若按每月 $5M 的中位烧钱且不出现资本开支峰值,现金续航约 55 个月(4.6 年);若每月 $10M(重开发阶段),现金续航 收缩至约 27 个月。这些情景假设没有额外收入或资本流入。隐含的 Series A 到 Series B 估值跃升相当大:若假设 Series A 投后 $150M–$200M(符合 Index Ventures 领投 $50M 轮次的典型持股目标),$2B Series B 意味着约 12 个月内跃升约 10–13 倍——即便在 2025–2026 年 AI 太空投资气候下也很激进,更像论点阶段而非牵引阶段定价。SAIC 作为战略产业伙伴而不只是财务投资人的参与,增加了一个纯 VC 无法仅凭公开数据完全定价的期权维度。[CI020, CI022, CI024, CI025, CI027, CI028]

FI002: 单位经济桥——从单星部署到收入

定性梳理从卫星采购成本到单元收入贡献的单位经济流,在缺少直接数据处使用近似基准。

所有输入均为估算,来自公开硬件基准(NVIDIA 定价类比、SpaceX 拼车费率)和公司未确认的示例收入场景。每星 800 个 GPU 取自公司 FCC 申请和媒体报道;实际部署容量可能不同。单星收入是假设性示例,不是预测。

[CI019, CI020, CI028, CI029]

4.5 财务尽调缺口与结论

仅凭公开信息,无法承销 Cowboy Space 的财务案例。五个结构性缺口定义了尽调议程:(1)未披露收入、ARR、积压订单或 LOI,意味着没有被验证的付费意愿证据;(2)Series B 资金用途只做高层描述,没有按项目拆分,独立跟踪资本开支不可能;(3)没有提交独立财务审计、S-1、SPAC 招股书或监管财务报表,所有财务主张都依赖公司声明和媒体报道;(4)DoD OECIF 合同和 SAIC 投资暗示政府收入路径,但金额和采购时间表均未披露;(5)公开信息没有阐明星座扩张所需的资本循环或项目融资策略。合在一起,这些缺口意味着投资人无法独立验证商业发射所需的烧钱速度、毛利路径、单位经济或总资本需求。收入质量结论是 N/A——没有收入可供评估质量。资本强度结论按任何基准都极端:$275M 足以支持一枚原型火箭和少数演示卫星,但比完整星座部署所需低数个数量级。这些财务尽调阻碍对于深科技种子到 Series B 公司并不罕见,但足够重要,应直接影响估值纪律。ArsTechnica 对轨道数据中心经济性的怀疑分析——指出规模化冷却物理尚未验证——提供了关键反向信号:无论融资叙事如何,论点阶段估值都可能经不起第一性原理工程审查。[CI023, CI033, CI034, CI037, CI039, CI040]

公开财务数据缺口——影响与尽调路径
缺失数据点重要性对承销的影响尽调路径
收入、ARR 或 backlog没有任何牵引数据,支付意愿完全停留在理论层面阻塞——无法评估收入质量或增长轨迹要求提供任何已签合同、LOI、试点提案或 beta 客户管线文件
毛利率模型或单颗卫星贡献毛利太空级硬件溢价和发射成本可能让早期规模下的单位经济不可行阻塞——即便采用乐观爬坡情景,也无法评估盈利路径要求提供包含单颗卫星 capex、发射成本、单元收入和回本时间线的财务模型
Series B 资金用途按项目拆分没有项目级分配,就无法跟踪资本部署效率重大——影响里程碑与资本的映射,以及 Series C 触发点评估要求 CFO 级预算拆分,覆盖火箭项目、卫星项目、监管和 overhead
政府合同金额(OECIF + SAIC)SAIC 和 DoD 都是潜在锚定客户;金额未披露,无法建模收入概率重大——政府收入可能延长跑道,并显著降低商业化前阶段风险要求提供 OECIF 合同金额、任何 SAIC MOU 或 MSA 范围,以及后续采购管线
星座扩张的资本循环计划$275M 无法支撑 20,000 颗卫星;若没有披露项目融资或资本策略,路径并不透明阻塞——整个星座论点都依赖一个尚未阐明的资本机制要求 CFO 说明分阶段融资方案:项目债、主权基金兴趣、制造信贷,或债转股模型
独立财务审计或 SEC 等效披露不存在第三方审计、SPAC 文件或 S-1;所有财务主张都依赖未经验证的公司陈述重大——无法独立确认资产负债表、股权结构表或 burn rate要求提供经审计或审阅的财务报表;如不可得,要求提供管理账和银行调节表

所有缺口都来自公开来源审查;对收入前 Series B 公司而言并不异常。合在一起的结果是,投资人无法仅靠公开信息承销财务表现。严重程度评级反映这些缺口对投资论点验证的影响,而非伦理问题。

[CI011, CI031, CI032, CI037, CI039]

4.6 展示材料

Chapter 05

05产品与技术

5.1 核心架构——上面级即数据中心

Cowboy Space 的核心产品洞察是,传统航空航天模式把运载火箭和独立卫星载荷分开,带来结构性冗余——重复的航电、级间适配器、分离系统和承力结构增加了质量,却不增加计算或电力。公司的解法是把火箭上面级本身当作轨道数据中心:一个单一飞行器进入低地球轨道后,直接在原位作为兆瓦级计算节点运行。每个 Stampede 单元设计重量 20,000–25,000 公斤,在 700–1,000 公里高度的晨昏太阳同步轨道中展开太阳能板产生 1 MW 电力,并搭载约 800 块 GPU。所选计算硬件是 NVIDIA 的 Space-1 Vera Rubin 模块,专为 LEO AI 工作负载打造,并于 2026 年 3 月 GTC 发布。Vera Rubin 模块把 88 核 CPU 与两颗 Rubin 图形芯片配对;单颗 Rubin 芯片可提供 50 petaflops 的 NVFP4 性能。NVIDIA 发布 Space-1 项目启动伙伴名单时点名 Cowboy Space,NVIDIA 新闻材料和多家独立媒体均确认了这一点。级段结构本身计划兼作散热器,通过主动热管理系统排出计算载荷产生的废热——在真空中没有对流冷却,这是设计必需。CEO Baiju Bhatt 将该架构描述为:「每个上面级利用飞行器的全部质量和体积,把发电、冷却和计算打包在一起——包括把级段结构本身用作散热器。」这个概念减少了每个轨道数据中心单元需要单独制造、测试和发射的部件数量,也是公司相较那些为第三方火箭建造卫星载荷的竞争者,最主要的工程与经济差异化主张。火箭按入轨载荷能力设计为略大于 SpaceX Falcon 9、但小于 Starship,目标是把 20,000–25,000 kg 送入轨道。可重复使用被描述为「在经济上合理时」的设计目标,但明确从属于最大化计算质量入轨效率。取消载荷适配器和次级结构,究竟能比传统两级火箭加专用卫星载荷提升多少质量分数,尚未得到独立量化;它仍是公司声称的收益,有架构逻辑支撑,但没有已展示硬件。[CE001, CE002, CE003, CE004, CE005, CE006]

产品模块与资产矩阵
模块 / 资产客户 / 用户状态 / 成熟度差异化关键尽调缺口
Stampede 生产卫星(上面级数据中心)AI 云客户、国防、主权计算设计阶段;按 FCC 申请(2026 年 6 月)尚未完成上面级即数据中心,省掉载荷适配器和重复航电热架构和结构散热器性能尚未演示
专有运载火箭(助推器 + 上面级)Cowboy 内部(自有发射)硬件前;Warren Lamont 负责发动机开发;测试设施待定专为 20,000–25,000 kg 数据中心载荷打造没有发动机测试数据,未确认发射场,也未确认制造设施
Galactic Brain 计算节点(首次轨道 AI 演示)R&D / 演示客户、NVIDIA 合作验证计划 2027 年初;运载工具未说明NVIDIA 确认的首次在轨 NVIDIA Space-1 Vera Rubin 部署未披露运载工具类型、轨道和地面接入架构
能量传输演示卫星(Apex 制造)技术验证;DoD OECIF(既有合同)计划 2026 年;Series B 时规格仍在最终确定从 LEO 进行红外激光能量传输的缩比概念验证规格未最终确定;发射伙伴未公布
NVIDIA Space-1 Vera Rubin Module(计算载荷)Cowboy 内部 / Stampede 卫星产品于 2026 年 3 月 GTC 发布;太空认证状态未公开唯一有 NVIDIA 背书、专为 LEO 打造的 GPU 模块;50 PFLOPS/chip NVFP4抗辐射加固认证数据和太空级采购成本未发布
光学星间链路网络(Stampede ISL)星座内部数据路由仅有架构设计;未发布终端硬件或链路预算避免 RF 频谱拥堵;支持 FCC 豁免部署保证金的理由终端供应商、指向 / 捕获 / 跟踪方法、可用性 SLA 均未知

状态反映截至 2026 年 5–6 月的公开披露信息;除非另有说明,所有硬件项目都仍处于设计阶段。

[CE001, CE003, CE007, CE010, CE016, CE027]
FE001: Stampede 卫星产品架构栈

单颗 Stampede 卫星的分层架构,展示发电、计算、热管理、通信和推进如何集成到火箭上面级中。

架构反映公司披露的设计目标;尚无硬件按该配置建成或测试。

[CE003, CE004, CE005, CE015, CE016]

5.2 Stampede 星座与 FCC 监管框架

Series B 关闭三天后,Cowboy Space 向 FCC 提交 SAT-LOA-20260323-00135 申请,针对 Stampede 星座:请求授权在 700–1,000 公里晨昏太阳同步轨道发射并运营最多 20,000 颗非地球静止轨道卫星。按 2026 年 6 月口径,这是已向 FCC 申报的最大轨道数据中心部署计划。申报架构在技术上很独特:Stampede 设计上主要通过光学星间链路和光学下行链路运行,而不是射频频谱;窄带 Ka-band(18.8–19.3 GHz 下行、28.6–29.1 GHz 上行)仅保留给关键任务阶段和紧急情况下的遥测、跟踪与指令。这种光学优先设计,是 Cowboy 请求豁免 FCC 标准部署保证金和里程碑框架的基础:公司认为这些规则原本是为防止频谱囤积而设,套用到一个按设计避开拥挤 RF 频段的系统并不适用。公司进一步主张,只需少至一颗卫星即可启动商业服务,因此可以由经济性而非监管强制推动分阶段建设。FCC Schedule S 表格数据列出四颗代表性卫星,分布在四个轨道面,远地点 / 近地点 830–860 km,预计运营寿命五年。不过,Technical Annex 和法律叙述——承载完整 20,000 颗卫星概念——明确承认「卫星设计尚未完成,服务开始前需要修改许可证」。这种承认在 FCC 申报中并不常见,也说明 Stampede 是监管预留和登记概念,而不是已可施工的方案。截至 2026 年 6 月,FCC 尚未批准该申请,仍在审查中。Stampede 在 FCC 队列中面对 SpaceX 最多 100 万颗轨道数据中心卫星的申报,以及 Starcloud 最多 88,000 颗的申报——整个轨道计算申请版图都先走监管,尚无任何星座按拟议规模进入商业运营。选择晨昏太阳同步轨道是有意为之:它最大化每日太阳照射时长,支持计算工作负载连续发电,避免其他 LEO 轨道制度中的食季影响。700–1,000 km 的高度选择有取舍:无阻力运营寿命更长,但若主动离轨系统失效,轨道衰减时间也更长,寿命末期碎片暴露随之增加。FCC 申报没有详细说明离轨或碰撞规避方案。[CE013, CE014, CE015, CE016, CE017, CE018]

技术与运营架构
层 / 组件角色依赖关键技术风险
黎昏 SSO 太阳能阵列(700–1,000 km)主电源;声称每颗卫星发电 1 MW轨道力学;面板展开;姿态控制太阳能阵列展开失败;进入食季会降低可用性
NVIDIA Vera Rubin Space-1 模块主 AI 计算(训练 / 批量推理)NVIDIA 供应链;耐辐射认证LEO 辐射环境可能劣化未屏蔽的商用 GPU 硅
NVIDIA IGX Thor 边缘处理器实时推理编排和本地控制NVIDIA 供应链;与 Vera Rubin 的热协同管理多种 GPU / 处理器之间的热耦合尚未公开建模
NVIDIA Jetson Orin(自主)卫星自主 housekeeping 和异常响应NVIDIA 供应链;软件自主栈轨道自主软件成熟度未经验证
级体结构散热器(主动热管理)废热排出;支撑 1 MW 计算密度级体几何;表面处理;姿态朝向700+ km 高度真空环境中 1 MW 散热尚未在任何尺度演示
光学 ISL 终端Stampede 星座内星间数据路由第三方或专有终端供应商待定大规模 PAT(指向 / 捕获 / 跟踪);未发布终端规格
至地面站的光学下行链路从轨道向客户交付数据地面站网络(第三方或专有)地面站覆盖、吞吐量和天气可用性存在缺口
Ka-band TT&C(18.8–19.3 / 28.6–29.1 GHz)应急指令、遥测、housekeepingFCC 许可(已申请);ITU 协调仅窄带;若光学链路失效,不足以承担数据卸载
专有火箭发动机(Ursa 级,概念)助推器和上面级入轨 LEO 的推进内部开发(Warren Lamont);测试设施待定发动机开发在关键路径上;未确认任何测试件
Apex 'Octopus' 制造 OS卫星平台生产管理和排产Apex 合作;Factory One 到 2026 年底达到 100,000 sq ftApex 仅确认用于演示卫星;量产协议未公开

架构数据来自 FCC 申请、公司网站、NVIDIA 产品页和媒体报道;除非有独立来源,硬件规格均为公司声称的设计目标。

[CE003, CE004, CE005, CE015, CE016, CE023]
信任、质量与合规摘要
控制 / 认证 / 要求状态范围缺口 / 尽调路径
FCC Stampede 星座许可(SAT-LOA-20260323-00135)2026 年 3 月提交申请;截至 2026 年 6 月仍在 FCC 审查,尚未获批最多 20,000 颗 NGSO 卫星;SSO 700–1,000 km部署保证金规则豁免仍待定;申请文件承认卫星设计尚未完成
ITU 协调(光学 ISL + Ka-band)未公开确认已启动 ITU 申报流程Ka-band TT&C 的国际干扰协调光学 ISL 不需要 ITU 协调;Ka-band TT&C 需要
FAA 发射许可(运载火箭)尚未提交;仍处于硬件前阶段所有美国商业火箭发射均需取得推进系统设计和测试数据出来之前,无法提交申请
ITAR / 出口管理条例未公开说明出口合规框架卫星硬件和火箭推进技术NVIDIA Space-1 模块是商业产品,但与国防客户(SAIC)集成会形成 ITAR 暴露面
轨道碎片与寿命末期离轨FCC 申请未披露离轨架构20,000 颗卫星位于 700–1,000 km,设计寿命 5 年FCC 越来越要求 5 年内离轨;没有推进系统时,700–1,000 km 轨道自然衰减可能耗时数十年
抗辐射认证(NVIDIA Space-1)NVIDIA 称模块适合 LEO 抗辐射环境;未发布独立认证数据LEO 高度,并暴露于南大西洋异常区商业 LEO 耐受与完整航天级认证差异很大;测试数据未披露
网络安全 / 在轨数据安全未披露公开安全架构或认证在轨计算可能处理涉密国防数据轨道计算没有 FedRAMP 等价框架;SAIC 的国防用途需要 FedRAMP 或等价认证

合规信息基于 FCC 申请文件、NVIDIA 新闻材料和媒体报道;尚无独立第三方审计或认证公开披露。

[CE013, CE018, CE019, CE020, CE022, CE034]
FE002: 客户工作流——从 AI 算力请求到轨道处理

数据从 AI 基础设施客户进入 Stampede 地面系统、卫星上行、轨道计算,再经光下行交付结果的流程。

客户工作流根据 FCC 申请、公司网站和媒体描述推断;没有可运行系统;过境窗口和集群细节未说明。

[CE014, CE016, CE036]

5.3 NVIDIA Space-1 合作与技术栈

每颗 Stampede 卫星的计算底座都是 NVIDIA 的 Space-1 Vera Rubin 模块,这是面向 LEO 优化的 GPU 推理和训练系统,2026 年 3 月在 NVIDIA GTC 发布。NVIDIA 太空计算产品页确认,Cowboy Space Corporation 正在集成三类不同 NVIDIA 组件:用于主要 GPU 工作负载的 Vera Rubin Space-1 Module、用于本地控制和推理编排的 IGX Thor 边缘 AI 处理器,以及用于自主卫星运营管理的 Jetson Orin。这套分层计算栈把工作负载拆开:Vera Rubin 模块承担大规模 GPU 批量推理或训练,IGX Thor 做实时边缘处理,Jetson Orin 管自主在轨运维。不过,具体硬件接口、每颗卫星的模块数量,以及每个模块的热包络都没有公开披露。Space-1 模块被介绍为适用于 LEO 运行的抗辐射设计,但抗辐射程度——是达到完整军用航天级标准,还是较轻的商业 LEO 容忍度——尚未独立验证。NVIDIA 的太空计算计划是一个重要的市场验证信号:这是首个主要 GPU 供应商发布专门定位于轨道 AI 基础设施的产品线,Cowboy Space 作为被点名的启动伙伴,也为硬件路线图增加了商业可信度。Cowboy 的光学星间链路架构需要自有或第三方自由空间光终端硬件,把 Stampede 卫星连成星座网络。FCC 申报把光学 ISL 描述为主要数据传输层,但没有说明终端规格、链路预算、可用性目标,或指向 / 捕获 / 跟踪方案——这些都是实际运营性能的关键未知数。地面段方面,申报请求 Ka-band 频谱用于 TT&C;面向客户的数据下行地面站基础设施没有描述。首个演示任务的卫星平台由 Apex Space 建造。Apex Space 位于洛杉矶,是一家高产率卫星制造商;截至 2026 年中,其生产设施年产能超过 200 个平台,并计划到 2026 年底扩至 100,000+ 平方英尺。Apex 使用自有「Octopus」制造 OS 管理装配排期和作业指导。Apex 已确认参与首颗演示卫星,但其在量产 Stampede 制造中的角色没有在公开文件中得到合同确认。[CE023, CE024, CE025, CE026, CE027, CE028]

工作流与用例分析
用户任务当前工作流Cowboy Space 方案声称的可量化收益关键限制
大规模 AI 推理(地理空间 / 地球观测)将原始影像下行至地面云;用电网电力大规模处理在轨道上用 Stampede GPU 节点处理影像;通过光学链路下行结果减少下行数据量;摆脱对地面电网的依赖LEO 过境窗口限制连续处理;光学下行受天气影响
AI 批量训练工作负载(受能源约束的客户)训练任务排在地面电网并网之后(等待 5–7 年);支付溢价租用由太空太阳能供电的轨道 GPU 小时;不需要电网并网在美国市场绕开 5–7 年电网接入等待期每个节点 1 MW 只是大型训练集群的一小部分;多卫星集群未说明
国防 / ISR 在轨分析将传感器数据下行到地面,在安全地面设施中处理在 Stampede 节点上处理机密传感器数据;减少下行负担降低时效敏感 ISR 的延迟;SAIC 投资传递验证信号安全认证(ITAR、受控密码技术)未披露;尚无确认客户
太空到地面能量传输(演示阶段)N/A —— 目前没有商业化 LEO 能量传输服务2026 年用 Apex 建造的演示卫星,从 LEO 以红外激光传输能量证明物理可行性,并验证光学能量传输技术仅为缩比演示;规格仍在敲定;DoD OECIF 合同只是概念验证
主权 / 受监管数据计算建设国家级数据中心,需要土地、电网、许可和用水在国家频谱许可下部署轨道计算节点绕开地面基础设施瓶颈;契合主权诉求数据驻留法律能否适用于轨道资产,法律上尚未验证

用例来自公司网站、FCC 申请、投资人表述和媒体报道;尚无客户合同或试点获得公开确认。

[CE014, CE036, CE037, CE050, CE051]
FE003: 关键依赖图——Stampede 首次发射

2028 年底首次自有发射里程碑前必须解决的依赖,展示推进、卫星、监管和供应链领域的并行关键路径。

依赖结构根据公开申请、媒体报道和工程逻辑推断;实际关键路径可能不同于公司内部计划。

[CE019, CE039, CE043, CE047, CE048]

5.4 热管理与关键技术约束

对 1 MW 功率规模的轨道数据中心而言,已验证的最难工程问题是排热。地面数据中心的电力使用效率,部分取决于把废热转移到空气或水中的能力——轨道上两者都不存在。唯一机制是热辐射,受 Stefan-Boltzmann 定律支配:300 K 表面每平方米约辐射 460 W。若在该温度下排出 1 MW 热负载,完美发射率下需要约 2,170 平方米散热器表面积;实际考虑几何视角因子约束和真实发射率后,需求接近 3,000–4,000 平方米。2026 年发布的轨道数据中心「物理墙」技术分析,把优化良好的单 MW 节点在更高温度(约 500–600 K,本身就会压迫 GPU 热容忍度)运行时的最低需求定在约 200 平方米。Cowboy Space 的设计答案是把火箭级段结构用作散热器——这个概念利用完整 20,000+ kg 级段的大表面积,但尚未在任何中间尺度验证。公开信息中没有热模型、原型测试结果或第三方工程评审。第二个未解约束是抗辐射。700–1,000 km 高度的 LEO 会穿越南大西洋异常区和 Van Allen 辐射带边缘,使硬件暴露在高能质子和电子中。商用 GPU 架构容易遭遇单粒子翻转和总电离剂量退化;NVIDIA 的 Space-1 模块被描述为适用于 LEO 的抗辐射产品,但没有发布独立资格认证数据表或 MIL-SPEC 等价证明。第三个约束是数据重力:轨道计算工作负载必须把大数据集传到轨道上(意味着卫星需要高上行带宽)、处理来自 LEO 传感器的流式数据(这是天然共址用例),或接受地面到轨道往返带来的显著时延。光学下行设计解决了吞吐潜力,但没有解决摄取经济性。TechCrunch 2026 年 5 月报道称,Cowboy Space 仍「在解决关键开发需求,比如测试、制造和发射火箭的设施」——确认火箭发动机测试基础设施尚不存在。这些叠加的不确定性——热、辐射、数据链路和推进测试基础设施——意味着 2028 年首次发射里程碑,要在没有中间尺度已展示硬件的情况下,同时解决多项首次出现的工程挑战。[CE032, CE033, CE034, CE035, CE036, CE037]

路线图与开发阶段时间线
日期 / 阶段功能 / 里程碑状态含义主要来源
2024(创立)以 Aetherflux 名义创立;使命是天基太阳能;获得种子轮 + Breakthrough Energy Ventures已完成公司从 SBSP 概念转向;技术团队来自 SpaceX、Blue Origin、NASA/JPL、NVIDIASpaceNews, Satellite Today
2025 年 4 月Series A($50M,Index Ventures);Aetherflux 品牌;规划 SBSP 演示已完成资金投向概念验证和早期卫星 / 光学技术BusinessWire, Payload Space
2026 年 5 月更名为 Cowboy Space;完成 $275M Series B,估值 $2B;为 Stampede 提交 FCC 申请已完成最大规模轨道数据中心 FCC 申请;披露 NVIDIA Space-1 合作伙伴;确认 Apex 演示合作BusinessWire, SpaceNews, SatNews
2026 年(H2 2026)首次卫星发射:Apex 建造的能量传输演示卫星,从 LEO 以红外激光传输能量已规划;截至 2026 年 5 月,规格仍在敲定证明太空到地面的光学能量传输技术;仅为缩比演示SpaceNews(CEO Bhatt 引述)
2027 年初首个使用 NVIDIA Space-1 Vera Rubin 模块的「Galactic Brain」轨道 AI 计算节点已规划;运载工具类型未说明首次在轨部署 NVIDIA GPU;首次轨道 AI 推理 / 训练演示SpaceNews
2028 年底首次自有火箭发射,搭载 1 兆瓦数据中心上面级已规划;推进系统研发进行中;测试设施尚未建立关键里程碑:以完整规模验证「上面级即数据中心」概念TechCrunch, SpaceNews
2028 年之后逐步建设 Stampede 星座,迈向商业运营愿景型目标;未承诺节奏、卫星数量或年度发射率进入创收阶段;取决于 FCC 许可获批、制造爬坡和发射节奏FCC 申请(SAT-LOA-20260323-00135), Satellite Today

所有规划里程碑均为公司提出的目标,不是合同承诺;日期可能随技术、监管和融资结果变化。

[CE041, CE042, CE043, CE044, CE045, CE046]
FE004: 产品能力成熟度图

评估 Stampede 五个核心技术域当前已验证成熟度与目标成熟度,突出设计主张和已验证硬件之间的差距。

成熟度评估为作者基于公开证据的估计;公司内部开发进度可能不同。

[CE027, CE032, CE039, CE043, CE047]

5.5 路线图与开发里程碑

Cowboy Space 发布了一条三步路线图,车辆规模和计算密度逐步提高。第一个里程碑是 2026 年(按 2026 年 6 月运行日期,即今年晚些时候)与 Apex Space 建造一颗小卫星,用红外激光从 LEO 向地面接收器演示无线能量传输。Bhatt 将其称为「亚尺度技术演示器」,用来证明物理学并验证方案,同时明确指出「这是一个构件,但长期业务重点是在轨道上把那部分电力用于 AI 计算,而不只是把电力传回地球」。截至 2026 年 5 月 Series B 公告,Cowboy Space 仍在敲定该任务规格,意味着一个计划同年发射的任务设计成熟度有限。第二个里程碑目标在 2027 年初,是首个使用 NVIDIA Space-1 Vera Rubin 模块的「Galactic Brain」轨道数据中心节点——被描述为太阳能供电轨道 AI 计算的首次演示。公开信息没有说明它是 Apex 建造的专用卫星,还是另一种飞行器。第三个里程碑是在 2028 年底前首次发射 Cowboy Space 自有火箭,搭载完整 1 MW 数据中心。这是技术上最激进的里程碑:它要求完成推进开发、火箭发动机测试、制造设施建设、发射场选择和许可、一级集成——截至 2026 年 6 月,这些全部处于硬件前阶段。火箭开发由 Warren Lamont(前 Blue Origin 推进工程师)和 Tyler Grinnell(前 SpaceX 发射运营总监)领导,西雅图卫星工程由 David Larson(SpaceX 和 Amazon 老兵)负责。GeekWire 报道时,Cowboy 发布了 18 个开放岗位,覆盖航电、机械工程、航天器设计和软件,并计划在西雅图卫星工程中心达到 40–60 名员工。第四阶段——逐步建设至 20,000 颗卫星星座——在 FCC 申报中有所描述,但没有具体发射日期或年度节奏承诺。公司表示,一旦规模化验证性能、可靠性和经济性,就会启动商业服务;星座规模由市场经济性而非监管里程碑驱动。从 2028 年首次发射到数百乃至数千颗卫星的商业有效星座,中间的差距意味着 Stampede 产生实质收入可能要到 2030 年代中期——这与其他瞄准同一窗口的轨道数据中心项目一致。[CE041, CE042, CE043, CE044, CE045, CE046]

5.6 技术差异化与可行性评估

Cowboy Space 的差异化论点建立在三项主张上:(1)发射与计算垂直整合,消除传统火箭加卫星架构带来的质量分数损失;(2)光学星间链路避开射频星座面临的频谱拥挤;(3)自研火箭打破限制所有其他轨道数据中心项目的发射瓶颈,使其不再受制于 SpaceX Starship 发射窗口。第一项主张有架构逻辑,但没有已展示硬件验证。第二项在技术上合理,FCC 豁免请求也反映了真正的政策创新——光学系统不必与其他星座做频率协调。第三项是公司最大胆、也最不确定的主张:Cowboy 正进入一个由 SpaceX、Rocket Lab 和 Blue Origin 竞争的发射市场,这些公司都有多年运营先发。TechCrunch 指出,「西方只有少数私营公司——主要是 SpaceX、Rocket Lab 和 Arianespace——能稳定发射商业火箭」,而 Blue Origin 的 New Glenn 在 2026 年 4 月第三次发射中未能交付载荷。相较最成熟的轨道数据中心竞争者 Starcloud(已发射首颗 H100 GPU 卫星并在 2026 年达到独角兽地位),Cowboy 的优势是拟议规模和垂直整合;Starcloud 的优势是硬件已经在轨运行。与地面 AI 计算相比,Cowboy 在时延(LEO 过境窗口、光学下行约束)和成本(航天级 GPU 硬件估计为地面成本的 3–10 倍)上不利,但在电力可得性和美国主要市场 5–7 年的电网互联排队时间上可能有利。SiliconAngle 指出,每颗卫星 1 MW「只相当于地面 AI 环境处理能力的一小部分」,意味着 Cowboy 需要把许多卫星互联,才能服务超大规模训练工作负载。公司没有说明面向超过单节点容量的大模型推理或训练作业的星间集群架构。从可行性看,可以独立验证的事实是:FCC 申报存在,NVIDIA 点名 Cowboy 为 Space-1 伙伴,Apex 确认首颗卫星合作,且公司已聘用有经验的航空航天工程师。尚未展示的主张包括:热-计算-推进一体化架构、光学 ISL 网络性能、任何规模的轨道 AI 计算运行,以及火箭推进硬件。[CE050, CE051, CE052, CE053, CE054, CE055]

5.7 展示材料

Chapter 06

06客户

6.1 买方格局与潜在客户分层

Cowboy Space 瞄准那些 AI 计算和数据处理需求受地面基础设施限制的客户——主要是电力可得性和电网互联排队——并可能从低地球轨道采集充足太阳能的轨道计算中受益。NewSpaceEconomy 对 ODC 市场的分析识别了五类早期客户原型:地球观测运营商、国防和安全机构、空间站用户、月球基础设施公司,以及面对特殊能源或位置约束的 AI 开发者。这些群体共同呈现窄而有支付能力的画像:预算匹配专用高性能基础设施,能够容忍新型供应链,且任务需求无法被地面替代方案完全满足。 AI 超大规模云厂商(AWS、Azure、GCP、Oracle)代表最大的理论市场,但采用不确定性也最高。受往返传播时延影响,时延敏感的消费者和企业工作负载在结构上不适合轨道计算,可行用例收窄到批量 AI 训练任务和非时延关键推理。Microsoft 的 Azure Orbital 产品线和 Planetary Computer 平台显示,超大规模云厂商更广泛地投资太空基础设施,但二者都不能证明其有兴趣向 Cowboy Space 这样的第三方租用轨道计算。OpenAI Stargate 计划承诺投入最多 $500B 建设地面 AI 基础设施,说明在轨道替代方案上线前,至少一部分超大规模云厂商 需求会被地面容量扩张吸收。国防和情报买方提供更近期的转化路径:SAIC 的战略投资是公开记录中最清晰的信号,此前 DoD OECIF 给 Aetherflux 的拨款也说明政府端已熟悉 Cowboy 的前身技术。 [CU001, CU002, CU003, CU004, CU005, CU006]

客户分层——买方画像与尽调缺口
客群买方 / 付款方用例规模收入 / 战略价值证据缺口
AI 超大规模云厂商(AWS、Azure、GCP、Oracle)CTO / 工程副总裁 / 内部 AI 运营批量 AI 训练与推理;不受能源约束的 GPU 溢出需求每年数十亿美元级资本开支很高——市场里最大的未满足需求量尚无超大规模云厂商表示有意向 Cowboy 采购;地面扩张削弱紧迫性
AI 实验室与研究机构ML 工程师 / 研究负责人 / 算力采购方大模型训练,不受地面电网约束每家主要实验室每年 $10M–$100M 算力支出高——任务关键工作负载;能源成本关乎生死延迟、数据管道和工作流重构构成障碍;Anthropic 信号只指向 SpaceX
国防与情报DoD CIO / SAIC 主承包商 / 情报机构为传感器融合、AI 推理、ISR 数据处理提供有韧性的隔离计算涉密;单个项目可能达到 $B 级战略性——SAIC 投资是最强公开意向信号需要 ITAR、FedRAMP、CMMC 合规;尚无确认授标或采购
地球观测运营商卫星运营商 / 影像分析师 / 任务设计方在轨影像处理、下行压缩、实时目标检测每家大型运营商每年 $10M–$50M中——补充现有 EO 收入;技术契合度强没有具名 EO 客户;带宽和延迟权衡尚未在规模化场景验证
空间站与主权计算空间站运营商 / 国家航天机构 / 政府数据托管方科研计算、数据主权、地外备份与灾备不固定;早期试验在 $1M–$50M 区间中——早期采用者具备战略价值;Axiom DCU-1 可作 ISS 先例尚无已部署服务;空间站集成时间线和数据主权法律状态不清楚
商业卫星运营商星座运营商 / OEM / 任务架构师边缘分析、智能下行路由、异常检测、类云卫星服务每家运营商每年新增 $1M–$20M 计算支出近期较低——延展现有服务,但不是核心收入驱动未披露管线;生态集成复杂,且受专有卫星平台约束

客群规模估算来自分析师根据 MarketsandMarkets 与 NewSpaceEconomy 的 ODC 市场分析(2026 年 5 月)推导的数量级,不是 Cowboy 专属数据。收入 / 战略价值反映市场层面的潜力,不代表已验证的 Cowboy 管线。所有缺口都来自缺少公开披露的客户证据。

[CU001, CU002, CU003, CU006, CU007, CU008]
FU001: 客户细分旅程图——从认知到扩张

五类买方细分映射到五个采用阶段;截至 2026 年 6 月,所有细分都停留在认知到资格评估阶段,没有任何细分越过意向阶段。

旅程阶段时间来自 Cowboy Space 公开发射和商业化时间表;没有客户管线数据可确认各细分进展。

[CU003, CU004, CU015, CU022, CU025, CU041]

6.2 具名客户信号与商业化前需求证据

截至 2026 年 6 月,Cowboy Space 未披露任何具名客户、已签收入合同、意向书或试点协议。公司 CEO Baiju Bhatt 明确表示,「一旦我们在规模上验证性能、可靠性和经济性,初始商业服务就会随后启动」,确认公司并不预期在首次自有火箭发射前产生收入;该发射目标是在 2028 年底前完成。这是本报告定义客户风险的核心姿态:$2B 估值,却没有商业验证。 最接近具名需求代理的是 SAIC。SAIC 作为新战略投资人,与 IVP 和 Blossom Capital 一同参与 Series B。报道政府承包商投资活动的 Washington Technology 特别点出,SAIC 是值得关注的关键 GovCon 投资人。SAIC 的战略理由已有公开表述:作为大型 DoD IT 服务承包商,它认为 Cowboy Space 的轨道计算能力与未来国防工作负载相关。不过,SAIC 尚未确认与 Cowboy Space 存在任何商业协议、服务级承诺或试点部署计划。 行业层面,SpaceNews 报道称,AI 公司 Anthropic 有兴趣把 SpaceX 拟议轨道数据中心纳入更广泛的计算访问协议——这说明 AI 实验室正在探索 ODC 需求,但该信号指向 SpaceX 系统,而不是 Cowboy。NVIDIA 将 Cowboy Space 列为 Space-1 Vera Rubin Module 项目的伙伴,这是供给侧技术合作,不是客户关系。运营两年、融资 $365M 之后仍没有任何具名买方关系,是本次尽调中最重要的客户缺口。 [CU011, CU012, CU013, CU014, CU015, CU016]

客户采用轨迹——可用指标与缺口
指标数值日期来源置信度含义缺失分母
具名收入客户02026-06-22SpaceNews(Bhatt:验证之后「商业服务会随之推出」)没有商业收入;完全处于商业化前阶段潜在 ODC 买方范围未定义;未披露合格管线
已披露 LOI 或试点协议02026-06-22BusinessWire Series B 新闻稿;未提及 LOI需求完全没有通过商业承诺验证无法判断管线转化率
战略投资人作为客户代理信号1(SAIC)2026-05-08SAIC 投资人公告;Washington Technology 报道SAIC 投资传递国防客户意向,但不附带有约束力的商业义务潜在 DoD ODC 买方范围和预算权限未披露
行业层面的 AI 需求信号(Anthropic 对 SpaceX ODC 感兴趣)1 个已披露意向表达2026-05-11SpaceNews Series B 文章AI 实验室在主动探索 ODC 需求,但信号只指向 SpaceX,不指向 Cowboy没有 Cowboy 专属 ODC 采用率
FCC 星座申请(Stampede)已申报 20,000 颗卫星2026-05-14Satellite Today FCC 申请报道;NewSpaceEconomy 分析说明供给侧野心;没有对应客户需求承诺支撑这一规模申请未附带客户合同或 LOI 数量

所有指标反映截至 2026 年 6 月的公开表述。「0 个客户」和「0 个 LOI」反映已确认的公开证据,不代表专有资料室信息。Cowboy Space 尚未举办投资者日,也未发布商业路线图。战略投资人数量(1)指 SAIC;其他投资人(Index、IVP、Blossom Capital、Breakthrough、a16z、NEA)为财务投资人,未披露客户关系。

[CU011, CU012, CU015, CU023, CU024, CU025]
具名客户证据表
实体客群与 Cowboy Space 的关系状态已披露结果局限
SAIC国防 IT / 政府Series B 战略股权投资人;被称为潜在未来客户不是客户——仅为股权投资无;没有已部署或已签约服务投资不等于合同收入;未披露服务级承诺或试点
美国 DoD OECIF国防 / 联邦政府选择前身公司 Aetherflux 参与轨道能源国防项目前身概念拨款;不是轨道计算在先前产品范围内验证轨道能源概念拨款覆盖的是另一套产品战略下的能量传输技术;不是计算采购
Anthropic(通过 SpaceX 的行业代理信号)AI 实验室 / 研究作为更广泛算力协议的一部分,对 SpaceX 轨道数据中心表达兴趣行业层面信号;没有 Cowboy 接触N/A —— 关系方是 SpaceX,不是 Cowboy Space不是 Cowboy Space 客户;只能说明 AI 实验室对 ODC 行业有广义需求
NVIDIAAI 硬件 / 技术将 Cowboy Space 列为 LEO AI 基础设施的 Space-1 Vera Rubin Module 合作伙伴技术供应合作;计划 2027 年首次部署计划在首个 Galactic Brain 节点集成 Space-1 模块NVIDIA 是硬件供应商,不是客户;合作不能验证终端用户需求
Planet Labs(暗示 / 推断)地球观测NVIDIA 和 NewSpaceEconomy 将其列为 ODC 行业相关 EO 客户 / 合作伙伴;Google Project Suncatcher 合作伙伴未与 Cowboy Space 发生具体接触N/A未披露与 Cowboy 的接触;该判断来自 EO 客群分析和 NVIDIA 合作伙伴名单

本表列举需求相邻关系,而非已确认客户。没有付费客户本身就是核心发现。标为「N/A」的单元格表示没有已披露结果。「状态」用于区分股权投资、供应合作、行业层面信号与客户合同或试点部署。

[CU012, CU013, CU015, CU017, CU018, CU020]
FU002: 采用与部署漏斗——发现到扩张

从 AI 算力需求认知到首次商业收入的连续转化路径;截至 2026 年 6 月,Cowboy Space 在 LOI 及以下阶段的管线为零。

漏斗第 1–3 阶段数值为示意性相对估计(以第 1 阶段完整可触达注意力归一化为 100),来自公开来源;没有可用的专有管线数据。第 4 和第 5 阶段为 0,反映已确认的公开表述。阶段标签采用常规企业销售漏斗阶段,并按深科技基础设施场景调整。

[CU011, CU023, CU025, CU031]

6.3 商业化打法与销售基础设施

截至 2026 年 6 月,Cowboy Space 未披露商业化打法、销售组织或客户获取路线图。公开招聘和媒体报道中没有企业销售负责人、客户经理或业务拓展岗位。GeekWire 关于公司西雅图招聘的文章提到,计划中的 40–60 名员工聚焦卫星设计和火箭 / 推进工程——主要是技术角色,不是商业角色。 隐含打法是 B2B 企业直销。鉴于可寻址客户分层的规模(超大规模云厂商、国防承包商、空间站运营商),任何有意义的商业接触都需要长销售周期、复杂采购流程和深厚企业集成能力。尤其是国防客户,从需求识别到合同授予,采购流程通常需要 12–24 个月;合规前置条件——ITAR 授权、很可能需要 FedRAMP Moderate 或 High 级,以及 CMMC 认证——会在 DoD 机构把工作负载部署到第三方轨道平台前增加大量提前期。 SAIC 的战略投资,是进入国防分层的渠道或联合销售安排中最合理的锚点。SAIC 与 DoD CIO 组织和国防主承包商有既有关系,可能加速 Cowboy 的采购路径。不过,公开信息没有披露与 SAIC 的正式转售、渠道伙伴或联合销售协议。整个商业进入策略仍是开放尽调项。 [CU021, CU022, CU024, CU035, CU040, CU044]

6.4 部署阻碍、切换成本与客户需求怀疑

反对 Cowboy Space 客户论点的最连贯怀疑,不是轨道计算技术上不可能,而是地面替代方案扩张得足够快,在 Cowboy 首次发射前就会缩小可寻址客户缺口。OpenAI Stargate 计划代表对地面 AI 基础设施的 $500B 承诺。超大规模云厂商正在推进核能协议、与能源资产共址,以及更好的冷却技术,这些动作合在一起削弱了轨道替代方案的能源稀缺论证。Ars Technica 对 AI 电力危机叙事的分析明确提出疑问:等轨道数据中心可部署时,它们是否还能在经济上具备竞争力。 部署阻碍还叠加了鸡生蛋式的需求问题:企业客户不太可能在技术规模化展示前承诺使用轨道计算,但规模化展示技术又需要资本,而没有客户承诺时资本很难筹集。Cowboy 通过 SAIC 战略投资,在股权侧穿过了这种结构性商业化前风险,但在商业需求侧尚未解决。The Next Platform 的技术分析指出,轨道数据中心在真空中排出一兆瓦热负载面临物理约束——这些工程挑战,Cowboy 自己的 FCC 申报也承认尚未完成。 最终采用 Cowboy Space 平台的客户,切换成本会很高。为轨道处理优化的工作负载——利用连续太阳能、辐射冷却,以及与共址卫星之间近乎零时延——迁回地面替代方案需要大量重构。这是长期护城河,但只有客户已经承诺并部署后才会兑现,因此在初期更多是采用门槛,而不是留存优势。 [CU026, CU027, CU028, CU029, CU030, CU031]

留存、满意度与耐久性——商业化前公司的缺口
指标数值客群置信度尽调要求
净收入留存(NRR)全部没有客户 → 没有 NRR 数据;要求提供商业化后 NRR 预测方法和假设
毛收入留存(GRR)全部没有客户 → 没有 GRR 数据;要求提供流失模型和合同期限假设
客户满意度(CSAT / NPS)全部没有已部署服务 → 没有满意度数据;要求提供计划中的试点反馈和验收框架
初始合同期限(标准)未披露企业 / DoD没有已签合同;要求提供销售模型中的标准合同期限(DoD 可能为 3–5 年)
重复使用率全部没有客户 → 重复购买为零;要求提供首个客户部署后的预计利用率爬坡

所有 null 值都反映没有客户,而不是数据被隐藏。本表记录的是需要什么,而不是已经有什么。置信度为 null 表示没有客户数据就无法评估指标。尽调要求代表在商业化后的资料室中评估留存风险所需的最低信息。

[CU032, CU043]
FU003: 客户证明质量矩阵——按关系类型评估证据

与 Cowboy Space 有已命名或推断关系的五个实体,全部是供给侧或投资人代理;没有一个构成真实客户证明。需求侧验证的证据质量整体偏低。

证据质量评级是基于已披露信息的定性评估。“证实战略意图”反映战略投资人逻辑,不是有约束力的客户协议。由于不存在商业关系,留存信号一律为“无”。

[CU012, CU014, CU018, CU043, CU045]

6.5 扩张潜力、集中风险与留存缺口

Cowboy Space 还没有客户,留存、NRR、GRR 和满意度指标都结构性缺失。现有客户基数尚不存在,公司无法证明 cohort 留存、合同续约或净收入扩张。这是底层尽调缺口,只有商业运营启动后才能解决;按 Cowboy 当前发射节奏,最早也要到 2028 年末。 首个商业客户带来的集中度风险极高。如果 SAIC 或任何单一防务客户成为第一个付费实体,收入启动时该客户就占 100%,并会在定价和合同条款上获得不成比例的议价权。防务合同通常会用多年期承诺缓和这种风险,但这类合同也带来 ITAR、FedRAMP、CMMC 等合规负担,需要大量时间满足;一旦要求无法持续维持,资格也可能被撤销。 扩张故事在理论上很有吸引力:Cowboy 的 Stampede 星座可从单颗卫星扩到 20,000 颗,每颗增加 1 兆瓦算力。初始地球观测或防务客户若跑通轨道处理经济性,可能催化更广泛采用,进而出现 land-and-expand 动态。Planet Labs 同时是 Google Project Suncatcher 伙伴和 NVIDIA 空间计算伙伴,因此可作为技术可信的早期采用者类比——但没有披露任何与 Cowboy 的具体合作。数据主权客户,包括主权国家、金融机构、需要地外存储的政府机构,代表更长期的扩张板块,其采购机制与 AI 算力板块结构不同。 [CU032, CU034, CU035, CU036, CU037, CU038]

扩张与集中度风险
扩张驱动集中度风险风险兑现后的影响尽调路径
SAIC 作为锚定客户 / 国防渠道背书SAIC 是唯一具名的客户相邻实体;如果它成为首个商业客户,单一实体集中度会极端偏高如果 SAIC 不部署,国防收入论点就没有验证;估值承压要求 SAIC 提供有约束力的 LOI 或 MOA,确认部署意向;确认 Cowboy 是否配备专门 BD 资源
借星座规模先落地再扩张(Stampede → 20,000 颗卫星)没有可供扩张的收入基础;扩张叙事要求先拿到首批客户再规模化没有客户承诺却重资本增长,会造成灾难性的资金续航风险确认支撑商业发射的最低可行客户名单;要求提供 2028–2030 财务模型
地球观测客群(Planet、Maxar、Capella 类比)EO 运营商技术能力强,但数量少;行业 TAM 偏窄如果 EO 客户主导早期收入,会形成集中度风险;仅靠 EO 无法支撑 $2B 估值梳理全部可触达 EO 客户;要求提供合格 EO 管线和支付意愿数据
数据主权与主权计算客群国家机构采购缓慢,且受政治变量影响;很可能延迟多年无论 Cowboy 发射时间线如何,主权客群收入都可能推迟到 2030 年之后找出拥有活跃 ODC 预算权限的具体政府机构;厘清法律状态
AI 超大规模云厂商采用(AWS、Azure、GCP)需要超大规模云厂商放弃快速扩张的地面产能,转而采用新颖且未经验证的轨道基础设施如果超大规模云厂商不采用,Stampede 星座的核心收入论点就会坍塌确认是否有任何超大规模云厂商发布 RFI,或参与正式技术评估

集中度风险以零客户基线衡量;任何首个客户都会立刻代表 100% 收入集中。影响评估假设 Cowboy 在商业发射时没有其他收入流。尽调路径可在正式资料室流程中处理。

[CU034, CU035, CU036, CU037, CU038, CU039]
FU004: 按买方细分划分的相对需求就绪度(分析师评分,1–5)

国防和情报因 SAIC 信号与 DoD OECIF 先例,在近期需求就绪度上得分最高;AI 超大规模云厂商 TAM 最高,但近期转化概率最低。

评分为分析师给出的定性评级(1=最低,5=最高),依据截至 2026 年 6 月公开证据中已点名需求信号、 付费意愿指标和采购路径清晰度。未取得 Cowboy Space 的管线或 CRM 数据;评分反映市场分析, 而非公司自身数据。

[CU005, CU006, CU008, CU036, CU037]

6.6 图表

Chapter 07

07风险

7.1 技术与工程风险

Cowboy Space 的核心技术押注仍停留在概念阶段,而不是经过飞行验证。公司及相关报道现在描述了一套非常具体的目标架构:20,000 颗由上面级衍生的卫星,运行在晨昏太阳同步轨道;每颗约 20,000-25,000 公斤,产生约 1 兆瓦可用电力,并搭载接近 800 个围绕 NVIDIA Space-1 Vera Rubin 模块构建的 GPU。具体化有帮助,但不能改变中心问题:公司想建的系统仍未完成。SpaceNews 和 SatNews 都报道称,Cowboy 向 FCC 表示卫星设计尚未完成,服务开始前还需要修改许可证。相同来源还称,该设计预计让级段结构本身充当散热器。这带来一阶热风险:真空环境消除了对流冷却,废热排放必须靠散热面积、发射率和可靠性假设,而这些假设尚未针对 1 兆瓦轨道数据中心公开验证。辐射是第二个主要工程未知数。NASA 和 NOAA 都把辐射带描述为由高能粒子构成的恶劣环境,可能损伤航天器电子设备。因此,Cowboy 计划的 700-1,000 km 工作带,在公司正依赖一款新发布的空间计算模块和「级段即数据中心」一体化设计时,提出了位错误、闩锁和组件寿命问题。公开来源描述了计划和合作,但尚未给出一体化飞行器、散热器概念或完整计算栈的鉴定数据。[CR001, CR002, CR003, CR004, CR005, CR006]

技术 / 运营风险登记表
失效模式公开信号可能性严重性剩余敞口关键缺失证明
散热器概念在 1 MW 规模下表现不及预期级段结构本身预期承担散热器功能全尺寸排热数据出现前仍为高集成飞行器的热真空测试结果
辐射环境削弱计算栈或航电NOAA 和 NASA 称辐射带对卫星十分严苛轨道带与持续辐射暴露重合,因此为高部件级和系统级辐射鉴定
上面级作为数据中心的集成延期卫星设计仍未完成,后续还需修改许可证架构新颖且跨学科,因此为高冻结设计、集成测试件和演示结果
光学下行链路 / ISL 架构披露不足公开材料描述了光学传输,但未披露网络容量客户吞吐量可能卡住有用工作负载,因此为中地面站计划、光链路预算、吞吐量披露
在轨服务难以处理故障独立批评强调「上门维修」昂贵或不可行修复闭环弱于地面数据中心,因此为高冗余方案、在轨可靠性目标、优雅降级预案

本表隔离架构特有风险,而不是泛泛的创业公司执行不确定性。

[CR003, CR005, CR007, CR008, CR009, CR010]
FR001: 风险热力图

在可见公开缓释措施之后,横向比较 Cowboy Space 最高的剩余风险。

[CR004, CR011, CR019, CR031, CR043, CR045]
FR002: 风险传导图

未完成的架构、发射执行和监管关口如何传导为进度和估值压力。

[CR003, CR010, CR013, CR019, CR023, CR038]

7.2 发射与制造风险

第二层风险是火箭规模的执行。Cowboy 不只是采购发射服务;由于每颗数据中心卫星本身也是运载火箭上面级,公司试图自建火箭项目。TechCrunch 称 Bhatt 预计首次自研发射会在 2028 年底前完成;SpaceNews 则称混合飞行器将大于 Falcon 9、小于 Starship,目标载荷能力为 20,000-25,000 公斤。这不是一个温和的硬件项目。它意味着发动机、结构、航电、制导、鉴定、级段集成和复用假设必须一起推进。除 FCC 申请外,FAA 许可又增加了一个独立监管关口,因为发射和再入权限不属于频谱审批。公开报道也明确显示,Cowboy 计划自研火箭发动机,并已招募有经验人员;但当前公开资料仍没有展示发动机测试结果、工厂、专用测试设施,也没有展示从 2028 年首次发射走向 20,000 颗星座所需持续节奏的路径。即使首枚飞行器按时发射,规模问题仍在:制造并入轨数千颗 20-25 吨卫星,需要一套发射系统和工业基础,两者都远超 Cowboy 已证明的能力。在自有发射器出现前,公司也仍暴露在第三方发射市场的定价和可用性风险下。[CR011, CR012, CR013, CR014, CR015, CR016]

合作伙伴 / 依赖风险登记表
依赖项角色集中度失效情景严重性缓释信号剩余敞口
NVIDIA Space-1 Vera Rubin已披露的主要计算载荷模块路线图、经济性或可用性变化公开合作关系和发射排期提及
自有火箭已披露的唯一全规模部署路径运载器延期或始终达不到发射节奏经验丰富的招聘和已获融资的火箭项目
FCC / FAA 批准部署和飞行的合法权利批准时点或条件拖慢规模化申请处于活跃状态且已被受理
光学地面 / 中继网络客户数据进出路径吞吐量或网络覆盖不足光学架构是公司核心前提
自有运载器就绪前的第三方发射市场早期演示或应急方案的过渡桥梁价格维持高位或运力受限更广泛的商业发射市场存在

只有当公开语料显示某项依赖对投资判断至关重要时,本表才列入。

[CR013, CR018, CR019, CR023, CR046, CR050]
人员 / 执行风险登记表
职能依赖或缺口可能性严重性重要性尽调要求
创始人 / CEO 领导力公开材料中,Bhatt 是品类叙事者、融资人和系统架构师愿景和资本获取似乎高度绑定创始人要求提供继任、授权和决策权地图
火箭工程与推进新运载火箭和发动机必须同步成熟执行风险会在各子系统之间叠加要求提供推进里程碑、测试节奏和就绪标准
监管 / 合规领导力出口、发射、频谱和碎片义务横跨多个机构合规失误可能卡住销售和发射要求提供具名合规负责人和法律顾问覆盖
商业转化团队公开信息看不到客户名单或收入基础技术抱负再高,也需要锚定买家要求提供管线、LOI 和采购阶段细节

本表关注的是团队板凳薄弱会实质改变结果的岗位,而不是组织架构图是否完整。

[CR017, CR025, CR045, CR047]

7.3 监管、法律与出口管制风险

Cowboy 的监管路径仍未封口。关键星座申请尚未获批;FCC 公告只说已接受备案,并列出一组广泛豁免,涉及里程碑、保证金和运营规则条款。媒体报道称,Cowboy 主张这些规则是为 RF 密集型星座设计的,并不适合一个主要依赖光学星间链路和下行链路的系统。这个论点也许自洽,但制造了明确的闸门风险:如果 FCC 拒绝、收窄或附条件批准所请求的豁免,Cowboy 的部署时间表会更难执行。处置制度同样关键。FCC 2022 年命令把低地球轨道任务后处置基准从 25 年缩短到 5 年,原因是让卫星在轨停留数十年已不再被认为可持续。因此,一个位于 700-1,000 km、拥有 20,000 颗卫星的系统,依赖常规主动离轨能力,而不是温和自然衰减。除 FCC 问题外,Cowboy 仍需要 FAA 发射批准;其火箭、卫星和相关技术数据也带来出口管制敞口。BIS 维护 EAR 框架,DDTC 的 ITAR 概述也明确指出,当硬件和 know-how 与敏感航空航天类别重叠时,美国防务物项和防务服务出口管制仍然相关。公开来源尚未披露 Cowboy 的出口合规姿态、外国客户限制或内部筛查流程。[CR019, CR020, CR021, CR022, CR023, CR024]

监管 / 法律风险登记表
风险主管机构 / 制度当前公开状态可能性严重性重要性近期尽调路径
FCC 星座许可尚未获批FCC / IBFS已受理申请,仍待批准未获批准前,完整星座运营无法启动跟踪案卷进展、请愿和委员会条件
部署保证金和里程碑豁免请求FCC Part 25 豁免公司正在寻求免于标准保证金 / 进度框架豁免被拒或条件收紧,可能拖慢部署或提高资本需求审查 FCC 对 25.155、25.157、25.164、25.165、25.114 及相关规则的最终处理
任务后五年处置合规FCC 轨道碎片规则5 年 LEO 处置基准适用于运营商20,000 颗卫星意味着需要反复主动离轨,而不是被动等待建模推进剂余量、离轨吞吐量和故障处理
发射与再入许可FAA AST与 FCC 频谱批准不同的独立许可制度火箭项目不能只靠 FCC 进展上天要求提供发射许可计划、安全论证和测试排期
火箭、卫星与技术数据出口管制EAR / ITAR相关制度适用;公司合规计划未公开披露外资、雇员、供应商或客户都可能触发合规义务要求提供出口分类备忘录、筛查流程和外部律师覆盖范围

本登记表聚焦当前可见的公开准入机制,不覆盖私人合同条款或保险范围。

[CR019, CR020, CR021, CR022, CR023, CR024]

7.4 轨道碎片、碰撞与天文风险

20,000 颗卫星计划将进入一个已经更拥挤、运营更脆弱的轨道环境。ESA 2025 年空间环境报告称,寿命结束时离开高度拥堵轨道的卫星还不够多,正在制造碰撞风险;2024 年还发生了数起重大碎裂事件,新增数千个碎片物体。NASA 的 Orbital Debris Program Office 维护 ORDEM、LEGEND 等专用工具,正是因为碎片风险不是理论问题,而是必须建模和管理的工程、运营与安全问题。Cowboy 的拟议规模会放大这项负担。每颗卫星都很重,都必须安全机动,最终也都需要按 FCC 五年基准处置。这意味着 Cowboy 最终会继承一个持续存在的会合管理和寿命末期排队问题,而不是一次性清理任务。天文观测又增加了外部性。AAS 警告,超过一百万颗已申请卫星中哪怕只有一小部分发射,也可能显著损害光学、红外和射电观测,并恶化轨道可持续性。SATCON1 和 JASON/NSF 星座报告也都强化了这一点:这些影响足够严重,需要专门的缓解工作。即使 Cowboy 小于部分竞争对手的申请规模,它仍会显著增加天文台和轨道公域承受的累积负担。[CR028, CR029, CR030, CR031, CR032, CR033]

碎片与碰撞风险评估
危害公开证据可能性严重性Cowboy 为何暴露所需缓释
拥挤的近地轨道ESA 称离开拥挤轨道的卫星数量不足Stampede 增加 20,000 个额外机动目标自动交会操作和可靠推进
碎裂事件ESA 称 2024 年新增数千个碎片目标即使 Cowboy 无过错,背景环境也会恶化屏蔽、避碰和保险策略
寿命终止处置队列FCC 5 年规则和大卫星数量每颗卫星都需要主动任务后处理离轨能力规划和故障储备
重物再入外部性每个单元被描述为 20,000-25,000 kg质量提高了处置失败的后果详细解体、伤亡和再入分析
天文学 / 轨道公地外部性AAS、SATCON1 和 JASON 记录了星座影响公众反对可能影响许可和声誉亮度、路由和天文台协调计划

严重性同时纳入直接运营风险,以及外部性引发的监管和声誉后果。

[CR021, CR028, CR029, CR031, CR032, CR033]

7.5 资本、融资与商业采用风险

Cowboy 已经融到可观资金,但今天资产负债表与明天野心之间的融资缺口仍然很宽。公司以 $2 billion 估值完成 $275 million Series B,使已披露股权融资达到约 $325 million。对于验证和扩编阶段,这是一笔有意义的资本;但相较于开发新火箭、验证 1 兆瓦轨道计算平台、以工业规模制造卫星并运营大型主动离轨舰队的隐含需求,它仍然有限。管理层自己仍把商业服务放在大规模验证性能、可靠性和经济性之后。也就是说,收入不会先于证明到来;证明本身必须先被融资。商业侧也比「轨道数据中心」的标题叙事更窄。NewSpaceEconomy 明确把早期轨道客户描述为边缘处理、防务、地球观测或其他专业用户,这些用户的工作负载可以容忍天基到地面的时延。NVIDIA 自己的空间计算页面强调的是最小下行时延和边缘地面站处理,而不是对地面 hyperscale 集群的一比一替代。Gartner 的观点更严厉:轨道数据中心几十年内都不会服务地面需求,甚至可能永远不会。Breaking Defense 补充称,Pentagon 和情报买方仍在等待能力和商业案例能否被证明。公开层面,Cowboy 仍没有具名客户或披露收入。[CR037, CR038, CR039, CR040, CR041, CR042]

融资与资本风险摘要
议题当前公开事实风险等级重要性投资人监测项
迄今已融资本已披露股权融资约 $325M,包括 $275M Series B对早期验证而言规模很大,但相比全星座野心仍然很小后续轮规模和时点
估值 vs 成熟度$2B 估值,但尚无已验证的轨道数据中心高预期压缩了执行失误的余地下轮降价或平轮风险
收入前资本缺口商业服务要在规模化验证之后才会跟进收入出现前,证明本身必须先被融资现金跑道相对于演示时间线
火箭 + 卫星双线开发负担发射器和计算平台必须一起融资资本开支栈重于单一产品创业公司基于里程碑的资本计划
监管延迟敏感性豁免或许可证延迟会在无收入状态下拉长烧钱期批准滞后时,时间就会变成资本应急储备和进度缓冲

这是基于公开信息的摘要;由于烧钱、债务和合同收入未披露,它不是完整现金流模型。

[CR037, CR038, CR043, CR048, CR049]
客户采用风险摘要
问题当前公开答案风险等级含义证据缺口
谁会先买?可能是狭窄的边缘、EO、国防或专用 AI 用户早期 TAM 窄于头部 AI 叙事没有具名买家承诺
Hyperscaler 很快需要它吗?不清楚;许多工作负载仍更适合地面基础设施轨道需求可能小于宣传暗示,或来得更晚没有 hyperscaler 采购证据
延迟可接受吗?仅适用于能容忍天地延迟和数据移动的工作负载客户匹配取决于具体工作负载,并非普适没有公开工作负载基准数据
国防需求已承诺吗?Breaking Defense 称买家大多还在观望战略热情未必转化为合同没有列装项目或试点披露
数据出站基础设施已验证吗?已描述光学传输,但容量未公开披露吞吐量瓶颈会抹掉计算价值没有地面网络或吞吐量预算披露

本表区分需求是否存在与采购是否就绪;Cowboy 主要具备前者,后者尚未具备。

[CR040, CR041, CR042, CR044, CR045, CR051]

7.6 集中度、时间线与淘汰标准

最后一层风险是集中度。Cowboy 的公开叙事高度集中在 Baiju Bhatt 身上,他既是创始人、资金来源,也是品类叙事者;叙事也集中在 NVIDIA 身上,它是首批轨道 AI 节点已披露的计算伙伴。任何一个集中度本身都不致命,但叠加后会提高执行脆弱性。如果 Bhatt 的愿景、融资能力或招聘杠杆减弱,公司失去的将不只是一个普通 CEO。如果 NVIDIA 的空间计算路线图、供给或经济性发生变化,Cowboy 首个可见计算架构要么围绕替代栈重做,要么放慢。时间风险会放大这两项依赖。许多轨道数据中心公司仍处在申请、组件发布或早期演示阶段,而不是规模收入阶段。如果地面电力瓶颈比 Cowboy 达到商业规模更快缓解,可服务市场可能在公司证明最难技术环节之前收窄。因此,承销框架必须有纪律。投资者不应把 20,000 颗卫星愿景当成产品;应把它视为一个期权,只有在二元里程碑达成后才变得可信:成功的轨道计算演示、热与辐射性能证据、穿过 FCC 和 FAA 关口的可信监管路径,以及能经受尽调的客户承诺。任何一个检查点失败,都应同时压缩时间线信心和估值。[CR046, CR047, CR048, CR049, CR050]

缓释措施与终止标准
风险领域可监测触发项阈值 / 事件行动含义
热验证集成热真空证据重大后续融资前没有可信的排热数据不按全规模 1 MW 经济性承保
辐射耐受鉴定或飞行测试结果首次演示窗口后仍无公开计算栈加固证据假设故障率更高、资产寿命更短
火箭执行推进和运载器里程碑首次自有火箭发射目标明显滑出 2028 年下调星座时点和资本开支假设
监管路径FCC 豁免 / 许可证结果和 FAA 许可证进展关键豁免被拒,或批准拖延且看不到补救将进度视为结构性延迟
客户证明具名承诺或生产级试点技术演示后仍无锚定客户折价处理需求转化和定价权
集中风险备份供应商 / 领导层深度NVIDIA 栈或创始人中心化决策流没有可信替代项提高关键人物和供应链折扣

终止标准刻意设为二元判断,便于在公开里程碑出现或滑坡时快速更新投资论点。

[CR049, CR050]
FR003: 依赖关系图

供应商、监管机构、演示和客户之间的关键依赖。

[CR011, CR019, CR042, CR044, CR046, CR050]
Chapter 08

08估值

8.1 投资论点、反论点与分析框架

Cowboy Space 位于两股宏观力量的交点:AI 算力容量危机,以及轨道基础设施能提供地面替代方案无法快速复制的约束缓解。投资论点建立在三层押注之上——(1) AI 算力需求继续超过地面部署速度,(2) 轨道数据中心在生产规模上具备物理和经济可行性,(3) Cowboy 垂直整合的「火箭 + 计算」架构能捕获足够市场,支撑低于 $2 billion 的进入价格。截至 2026 年 6 月,三项押注都尚未被证明。 反论点同样有分量。公司处于收入前阶段,没有披露客户管线。1 MW 轨道数据中心的散热问题尚未在生产规模公开验证。火箭项目仍处早期;Relativity Space 和 ABL Space Systems 的可比项目在首个运营里程碑前消耗了 $100M–$650M+,其中数个未能达成。The Register 在 2026 年 2 月引用 Gartner 称,轨道数据中心是「pie in the sky」式想法;另一篇 2026 年 4 月 Register 报道引用了一家轨道数据中心初创公司承认发射经济性「don't fly」。SpaceX 2026 年 1 月向 FCC 申请最多 100 万颗轨道数据中心卫星,构成生存级竞争威胁。 面对这些不对称,标准 DCF 和收入倍数框架无法适用——没有收入、没有 ARR、公开领域也没有财务模型。合适的分析镜头是基于情景的期权估值:嵌入式看涨期权对 ODC 市场存在和 Cowboy 执行的隐含价格是多少?这个期权价格是否符合有纪律投资者应支付的水平?下方的推荐摘要和论点 / 反论点表直接呈现这一框架。 [CV001, CV002, CV006, CV007, CV022, CV023]

建议摘要
维度取值依据
建议继续研究 / 谨慎跟踪收入前;无财务模型;技术和资本风险极高
信心尚未达成运营里程碑;所有论点押注都未验证
风险评级技术、监管、资本密集度、竞争(SpaceX)、执行
估值立场投机性溢价$2B 投后估值反映期权价值,不是收入倍数;不适用 DCF
入场纪律仅限里程碑门控在 2027 年 ODC 演示和 2028 年自有火箭发射时重新评估

建议仅基于截至 2026-06-22 的公开证据;公开领域没有收入、披露的股权结构或财务模型。

[CV001, CV006, CV007, CV038]
论点与反论点
论点类型证据什么会改变判断
AI 计算需求在结构上跑赢地面供给正方论点Goldman Sachs、McKinsey、IEA 关于电网排队的数据;AI CapEx 超过企业收入地面建设快于预期追上;AI scaling laws 进入平台期
轨道环境提供太阳能、真空冷却,并绕开许可正方论点IEEE Spectrum 和 ArsTechnica 的物理分析;Seraphim Q1 2026 ODC 投资激增地面近清洁能源替代方案以更低成本更快规模化
垂直整合(火箭 + 数据中心)消除发射依赖溢价正方论点BusinessWire Series B 公告;BlackNight Space Labs 竞争分析单一环节故障(火箭或冷却)击穿整个整合溢价
顶级投资人财团(Index、a16z、BEV、IVP)验证市场创造论点正方论点BusinessWire Series B;a16z 公告;Fenwick 法律确认财团集中在收入前叙事;迄今没有战略客户收入
截至 2026 年 6 月,无收入、积压订单、客户合同或披露 ARR反方论点BusinessWire Series B;GeekWire;TechCrunch;IEEE Spectrum 均确认收入前状态首份披露经济条款的商业合同或 LOI
ODC 物理在生产规模尚未验证:1 MW 排热未解决反方论点Gartner/Register 2026 年 2 月;ArsTechnica 物理分析;SatNews 冷却问题在代表生产条件下完成 1 MW 单元热真空鉴定
SpaceX 向 FCC 申请最多 100 万颗 ODC 卫星,构成生存级竞争风险反方论点SpaceNews;TechCrunch 2026 年 1 月 SpaceX FCC 申请;BlackNight 分析SpaceX 宣布仅供内部使用,不提供商业云服务
可比火箭项目在首个运营里程碑前已消耗 $100M–$650M+反命题财务章节;ArsTechnica;The Register 发射经济性报道Cowboy 自研火箭以低于预算、早于 2029 年完成首次发射

证据列引用的是来源类别,而非具体 claim ID;详细归因见章节 claimRefs 和可比表。

[CV022, CV023, CV024, CV026, CV031]
FV001: 建议逻辑

从市场现实、产品状态、风险结构和估值,推导出审慎推进建议。

流程代表分析链条;不是因果模型。

[CV007, CV021, CV022]
FV004: 投资 KPI 记分卡

面向 IC 的七维评分;分数反映截至 2026 年 6 月公开证据质量。

分数是基于截至 2026-06-22 公开证据的定性评估。量表:5 = 可达到的最佳水平; 0 = 没有正向信号证据;不是模型输出。

[CV006, CV007, CV026, CV031, CV040]

8.2 估值背景、轮次定价与可比分析

Cowboy Space 于 2026 年 5 月以 $2 billion 投后估值融资 $275 million,仅 Series B 这一轮就隐含约 13.8% 稀释,投前估值约 $1.725 billion。到 Series B 交割时,公司从种子轮、$50M Series A(2025 年 4 月)到 Series B,合计已融资约 $365 million。公司没有公开披露 cap table、清算优先权或完全稀释股数,因此外部无法精确分析稀释和期权价值。 在轨道数据中心品类内,Cowboy 的估值是已披露或可推断轮次价格的同行中最高的。Starcloud 在 2026 年 3 月的 $170M Series A 达到独角兽状态($1.1B),据 SpaceNews 和 BlackNight Space Labs 引用的消息来源,其后据称还在寻求以 $2.2B 再融 $200M。Lonestar Data Holdings 是一个月球数据中心概念,到 2025 年末总融资仅约 ~$12.4M,显示空间计算初创公司内部融资分散度很大。 与 AI 基础设施锚点相比,截至 2026 年 6 月,公开 neocloud 和数据中心公司按未来收入交易的中位倍数为 11.8x(multiples.vc)。行业龙头 CoreWeave 市值约 $64 billion,对应 $6.2B TTM revenue。Crusoe AI 在 2025 年 10 月以 $10B+ 估值融资 $1.37B——且拥有真实数据中心收入。这些可比公司展示了市场愿意为已经产生收入的 AI 算力基础设施支付什么。Cowboy 零收入却估值 $2B,是另一种工具:它在给 ODC 市场创造的战略期权定价,并由顶级投资者 syndicate(Index Ventures、a16z、Breakthrough Energy、IVP)背书;同时,2026 年 Q1 SpaceTech 投资气候有利,全球单季部署 $8.0B(Seraphim Space Index)。不过,Finro 的 2026 年 Q1 分析提醒,私人 AI 市场已开始把「先增长、商业模式以后再说」的叙事大幅重定价,转向拥有可重复变现和持久需求的公司——如果里程碑滑坡,这会直接威胁 Cowboy 当前倍数。 [CV003, CV004, CV005, CV008, CV009, CV010]

可比估值表
公司阶段 / 状态最新估值累计融资收入 / ARR主要相关性作为可比公司的关键局限
CoreWeave (CRWV)上市公司(Nasdaq,2025 年 3 月 IPO)约 $64B 市值(2026 年 6 月)$1.5B IPO + 此前股权融资$6.2B TTM 收入(2026 年 6 月)领先的公开 AI 计算基础设施公司;neocloud 交易倍数为 11.8x NTM 远期收入。已有大量经常性收入和积压订单;Cowboy 收入为零——比较时必须对未经验证的里程碑路径大幅折价。
Crusoe AI私营,Series E(2025 年 10 月)$10B+ 投后估值累计 $3.7B+(股权 + 债务)收入可观(未披露)已具规模的私营 AI 计算基础设施;融资阶段和投资者画像直接可比。地面业务为主,且已有客户;与轨道业务风险画像不同;收入基础不可比。
Starcloud(前称 Lumen Orbit)私营,Series A(2026 年 3 月)$1.1B 投后估值;正寻求 $2.2B累计融资 $200M尚无收入,2025 年 11 月发射首颗卫星直接 ODC 可比:同一赛道、相近的收入前阶段、投资者基础重叠。在可比融资水平下,阶段更早、规模更小;Cowboy 的 $2B 估值比 Starcloud $1.1B 高 1.8x, 且累计融资更低。
SpaceX2026 年 6 月 IPO$1.75T IPO 估值数十亿美元(股权 + 债务)$18–22B 收入(2025 年估计)轨道基础设施龙头;ODC 已申报 100 万颗卫星;为轨道计算竞争环境设定上限。业务高度多元;Starlink + 发射 + xAI 收入占主导;ODC 目前不是收入线;规模 / 阶段不可精确可比。
Lonestar Data Holdings私营早期阶段未披露;隐含低于 $100M累计约 $12.4M(截至 2025 年末)尚无收入太空原生数据存储概念;行业同类公司。聚焦月球而非 LEO 计算;资本体量和野心小几个数量级;无法与 Cowboy 的一体化 ODC / 火箭论点直接相比。

估值取最近一次公开披露融资事件;CoreWeave 市值截至 2026 年 6 月。AI neocloud / 数据中心上市公司的 NTM 远期收入倍数中位数为 11.8x(multiples.vc,2026 年 6 月);Cowboy 没有收入基础,无法套用收入倍数。

[CV008, CV009, CV010, CV011, CV012, CV013]
FV002: 估值敏感性——里程碑和收入情景

以 $2B Series B 入场价和 AI 基础设施可比倍数为锚,展示不同里程碑和收入结果下的示意隐含估值。

数值单位为十亿美元。收入倍数(20–30x)来自 Finro 和 multiples.vc 公开 AI 基础设施基准; 该倍数被投机性套用于假设中的未来 ARR;不是 DCF,也不是一致预期。

[CV019, CV027, CV028, CV029, CV008]

8.3 情景分析、敏感性与下行触发因素

缺少收入、披露财务模型或客户合同,所有估值判断都被迫进入情景分析。三个情景覆盖了合理结果区间,核心取决于 Cowboy 能否打中 2026–2028 年里程碑序列:能量传输演示、首个 Galactic Brain ODC 节点、首次自研火箭发射。 牛市情景:2026 年能量传输演示成功,2027 年 Galactic Brain ODC 节点部署,并在商业吞吐下证明热真空性能;自研火箭在 2028 年底前发射;首个商业或政府客户合同在 2028 年中前签署。在这条路径下,Cowboy 将成为最早经过验证、且拥有自有发射能力的商业轨道数据中心基础设施运营商。到 2029 年末,隐含估值达到 $8–15B 是合理的,锚点是 CoreWeave 在收入确认后 IPO 后倍数扩张 3.5x 的类比。 基准情景:里程碑延迟 12–24 个月,来自 Starcloud 和 Blue Origin Project Sunrise 的竞争加剧,Cowboy 在 2029 年末以温和规模取得首笔收入。$2B 投后标记按实际价值大致持平($2–4B 区间),与投资者评估里程碑进展相对原始论点后,二级价格持平或温和变化的趋势一致。 熊市情景:重大技术失败——1 MW ODC 单元热鉴定失败,或火箭项目成本膨胀超过可用资本——迫使重组或下轮降估值。如果 SpaceX 轨道数据中心大规模部署成真,并在 Cowboy 达到收入前出现,独立 ODC 运营商的可服务市场将坍缩。在这一情景下,$0.3–1B 的下轮估值标记是合理的,与此前未能同时跨过物理和经济性闸门的深科技基础设施品类在 hype 后重定价一致。 所有情景下,稀释路径都很重要。达到牛市里程碑序列,很可能需要在收入前再融资 $500M–$1B+;取决于后续轮次估值,从当前 Series B 后基础起算,累计稀释隐含为 40–60%+。清算优先权悬置和 liquidation stack 仍未知。 [CV019, CV020, CV021, CV025, CV027, CV028]

牛市 / 基准 / 熊市情景分析
情景关键假设隐含估值区间(2029)概率信号主要下行触发因素
牛市2026 年电力束流演示成功;2027 年 Galactic Brain ODC 部署,并跑出商业级吞吐量;自研火箭早于 2028 年发射;首批客户合同在 2028 年中签署;ODC 市场发展快于 SpaceX 商业化进入。$8–15B所有近期里程碑都必须按期兑现;此前没有 ODC 在这一规模完成验证。三层里程碑中任一层技术失败;SpaceX 在 Cowboy 之前发射商业 ODC。
基准里程碑延迟 12–24 个月;Starcloud 和 Blue Origin 竞争加剧;2029 年末实现小规模首批收入;ODC 经济性得到验证,但市场碎片化;追加融资以持平到小幅溢价完成。$2–4B这与可比深科技行业在技术验证阶段历史轮次价格维持持平的情况一致。首批收入前资金耗尽;发射成本曲线恶化。
熊市热工认证失败或火箭项目成本升级;Cowboy 尚未实现收入前,SpaceX 已规模化部署 ODC;下一轮被迫下调估值; 市场未按预计时间线发展。$0.3–1B这与此前深科技基础设施失败后,市场从炒作中重定价的情况一致;若市场实际上被 SpaceX 抢占,冲击会很严重。FCC 拒绝 Stampede;火箭研发失败;SpaceX 率先商业化。

估值区间是基于可比私募市场轮次定价和公开 AI 基础设施公司倍数推导的情景估计;并不代表 DCF 预测或分析师一致目标价。 鉴于公司尚未产生收入且私有信息不对称,概率信号只能定性判断,无法赋予精确权重。

[CV027, CV028, CV029, CV040]
论点破裂与止损触发因素
触发因素阈值 / 事件对投资论点的传导行动含义
热工 / 冷却认证失败1 MW 轨道数据中心未能在生产代表性条件下完成热真空认证。核心产品前提失效;无法推进商业 ODC 部署。退出或重组;该论点要求物理产品真实存在。
火箭项目较计划滑延超过 24 个月自研火箭首飞推迟到 2030 年以后,且没有可信的临时发射即服务桥接方案。垂直整合论点失效;发射依赖成本累积;资本消耗快于里程碑所能支撑的速度。用更新后的桥接发射经济性重新评估;若有机会,考虑二级市场退出。
FCC 拒绝或实质性限制 Stampede 星座FCC 拒绝授权 20,000 颗卫星的 Stampede,或把批准条件绑定到经济性不可承受的碎片缓解要求上。面向规模化 ODC 部署的核心市场准入前提失效。评估监管路径和时间线;可能围绕仅服务政府的星座重组。
Series C 下调估值Series C 定价低于 $2B 投后估值,较当前入场价格出现减记。投资者对里程碑路径的信心丧失;未来稀释加重。评估论点是否还能延续;判断里程碑滑延能否在合理资本内追回。
Cowboy 首批收入前,SpaceX 已按商业条款推出 ODCSpaceX 在 Cowboy 签下首个客户合同前,已部署可供商业客户使用的轨道数据中心基础设施。可服务市场实际上被资本更充足、发射一体化的竞争者抢占。退出;在当前阶段,SpaceX 的竞争替代对 Cowboy 来说是结构性不可逆。
关键推进 / 热工工程负责人离职首席推进或热管理工程师离职,且 60 天内没有指定继任者。两个不确定性最高的技术层执行风险显著上升。要求管理层说明;下一次出资前评估人才管线和留任计划。
[CV039, CV026]
FV003: 各情景估值 / 回报区间

基于里程碑情景假设,给出 Cowboy Space 到 2029 年牛市、基准和熊市隐含估值区间。

区间为截至 2029 年时间点的情景估算,单位为十亿美元。无 DCF 基础; 锚定可比私人轮次倍数和示意 ARR 情景。公司仍是收入前深科技企业,信息不对称极高,因此区间很宽。

[CV027, CV028, CV029, CV036]

8.4 退出准备度与最终尽调要求

Cowboy Space 看不到近期退出路径。公司没有收入、没有披露客户合同,首个运营里程碑(Galactic Brain ODC 节点)预计在 2027 年。IPO 或战略收购至少需要首笔商业收入和已证明的轨道数据中心运营;在基准情景下,现实窗口是 2029–2031 年。Baiju Bhatt 的机构投资者基础和此前 Robinhood 履历提供声誉背景,但不保证退出时间。 战略收购风险不对称:SAIC 的投资释放了防务客户兴趣信号,并在商业牵引仍慢时创造潜在政府战略收购方情景。这为熊市情景下的退出价值提供了非商业底部,但更可能是部分 acquihire,而不是全价退出。如果技术验证通过但商业模式挣扎,Blue Origin、Amazon 和 Microsoft 都是合理的战略收购方;不过公开信号尚不支持这一点。 下方最终尽调要求,是从「继续研究」立场转向正式投资考虑所需的最低证据集。结构上最重要的缺口是 cap table 和清算优先权悬置:没有这些信息,就无法做精确回报建模。对论点最关键的缺口,是 1 MW 规模下经过验证的热冷却概念证明,以及一套量化的单位经济模型,用来比较轨道与地面数据中心 TCO。 [CV030, CV031, CV037, CV038, CV044, CV045]

最终尽调问题清单
主题缺失证据重要性负责人 / 尽调路径
股权结构和优先权安排按轮次列明股份数量、清算优先权、反稀释条款和参与权的完整股权结构表。不理解优先权堆叠和转换机制,就无法建模投资者回报或有效入场价格。公司法务 / 财务团队;在尽调资料室要求提供。
单位经济性和收入模型生产规模 1 MW ODC 的量化单位经济性:发射成本、硬件成本、发电成本,以及相对地面竞争 TCO。验证 ODC 在 Starship 或临时发射经济性下能否达到商业成本竞争力;这是区分牛市与基准情景的核心。公司 CFO / 业务拓展;要求提供技术和财务模型。
热工冷却技术概念验证物理测试数据,证明在热真空条件下接近或达到 1 MW 规模的散热能力。这是核心物理难题已被解决的唯一外部验证;截至目前公开证据不足。公司 CTO / 工程团队;优先要求第三方测试报告或见证测试。
客户管线和 LOI具名客户、意向书,或披露商业条款 / 定价信号的备忘录。验证政府试点兴趣之外的需求论点;SAIC 投资具有战略意义,但尚未产生收入。公司 CEO / BD 团队;若 LOI 无法披露,则安排客户背调电话。
FCC 批准时间线和监管风险评估Stampede 星座预计 FCC 裁定时间线;碎片缓解豁免可能性的法律分析。FCC 授权位于星座部署的关键路径;这里的不确定性会显著影响基准与熊市情景。监管法律顾问;监控 FCC 案卷;跟踪 IBFS 申报。
火箭开发预算和项目风险详细火箭研发预算、烧钱速度,以及与可比小 / 中型运力项目(Relativity、Stoke、ABL)的对比。火箭开发的资本强度,是下一轮稀释或下调估值压力最可能的来源。公司 CEO / VP Engineering;要求披露项目里程碑跟踪和预算。
SAIC 合同细节和政府管线SAIC 战略投资的范围、金额和排他性条款,以及任何相关商业协议。政府 / 国防收入底线是近期最可信的商业上行空间;当前证据尚未量化。SAIC 合作团队;公司 IR;政府合同数据库(如适用)。
[CV037, CV030, CV031]

8.5 图表

免责声明

本尽调报告由 AI 研究代理基于截至 2026-06-22 的公开来源生成,不构成投资建议。Cowboy Space Corporation 是一家私营公司,燃烧率、客户管线、火箭项目预算、热测试数据和 Series B 股权结构条款等关键承销输入仍未披露; 任何投资决策都应结合管理层材料、客户访谈和独立工程尽调验证。

证据索引

结论
编号陈述可信度来源
CO001 Cowboy Space Corporation raised $275 million in Series B financing at a $2 billion post-money valuation, announced May 8, 2026. SO002, SO004, SO007
CO002 The Series B was led by Index Ventures, with new investors IVP, Blossom Capital, and SAIC participating. SO002, SO004, SO020
CO003 Existing investors Breakthrough Energy Ventures, Construct Capital, Andreessen Horowitz, NEA, Interlagos, and Baiju Bhatt all participated in the Series B. SO002, SO007, SO020
CO004 Baiju Bhatt is the founder and CEO of Cowboy Space Corporation. SO001, SO002, SO009
CO005 Cowboy Space Corporation is headquartered in San Carlos, California. SO001, SO002, SO007
CO006 The company was originally incorporated as Aetherflux in 2024, formally in March 2024 per Data Center Dynamics, though TechCrunch's October 2024 article describes Bhatt founding it 'late last year' suggesting late 2023. SO009, SO010
CO044 Baiju Bhatt told TechCrunch in October 2024 that he founded Aetherflux 'late last year,' implying a late 2023 founding date. SO009
CO007 Cowboy Space operates a second engineering office in Seattle, Washington focused on satellite design and rocket/propulsion engineering. SO008, SO005
CO008 Aetherflux rebranded to Cowboy Space Corporation in May 2026, coinciding with the Series B announcement and the pivot to orbital data centers plus launch vehicles. SO002, SO003, SO005
CO009 Cowboy Space CEO Bhatt stated the company has raised approximately $365 million in total capital as of the Series B close in May 2026. SO007, SO023
CO010 TechCrunch reported the company had previously raised $80 million from investors prior to the Series B; SatNews cited approximately $325 million in disclosed equity after the Series B; both figures differ from the CEO's $365 million total. SO004, SO011
CO011 Baiju Bhatt co-founded Robinhood Markets with Vlad Tenev at Stanford; he served as co-CEO through 2020 and as Chief Creative Officer until early 2024. SO009, SO024, SO023
CO012 Bhatt holds advanced degrees in physics and mathematics from Stanford University. SO009, SO014
CO013 Bhatt's father worked for NASA at Langley Air Force Base, contributing to Bhatt's lifelong interest in space. SO009
CO014 Tyler Grinnell, former SpaceX director of launch operations, is a named key hire at Cowboy Space. SO007, SO004
CO015 Warren Lamont, who led rocket engine and booster stage development at Blue Origin, serves as Head of Propulsion in Cowboy Space's Seattle office. SO007, SO008
CO016 David Larson, with prior experience at SpaceX and Amazon Kuiper, is Director of Satellite Engineering at Cowboy Space's Seattle office. SO008
CO017 The Cowboy Space team includes veterans from SpaceX, Blue Origin, Astranis, NASA/JPL, Amazon Kuiper, NVIDIA, Lockheed Martin, Anduril, and the US Navy. SO001, SO007, SO017
CO018 Aetherflux raised a $50 million Series A in April 2025, led by Index Ventures and Interlagos, with Breakthrough Energy Ventures, Andreessen Horowitz, and NEA also participating. SO017, SO003
CO019 Aetherflux was selected by the US Department of Defense Operational Energy Capability Improvement Fund for a proof-of-concept wireless power transmission system during the Series A phase in 2025. SO017
CO020 Series A individual investors included Robinhood CEO Vlad Tenev, former SEC commissioner Dan Gallagher, and actor Jared Leto. SO017
CO021 Fenwick and West LLP, led by corporate partners Morgan Sawchuk and Ran Ben Tzur, represented Cowboy Space Corporation in the Series B financing. SO020
CO022 SAIC's investment in the Series B signals potential strategic interest in orbital compute for defense and government applications, extending the DoD OECIF relationship. SO021, SO017
CO023 Aetherflux's original business model was beaming solar power from a LEO constellation to Earth via precision infrared lasers, targeting remote military bases, disaster relief, and eventually commercial consumers. SO009, SO014
CO024 Aetherflux demonstrated wireless power transmission in the laboratory in 2025 before the Series B close. SO017
CO025 As Cowboy Space, the company's primary focus shifted to orbital AI data centers and a vertically integrated launch vehicle program, while retaining space-solar power as an architectural input rather than the end product. SO001, SO002, SO003
CO026 Cowboy Space's 'Stampede' constellation plans for up to 20,000 satellites in dawn-dusk sun-synchronous orbits at 700-1,000 km altitude. SO006, SO012, SO015
CO027 Cowboy Space filed FCC application SAT-LOA-20260323-00135 in May 2026 for authority to deploy and operate the Stampede constellation as a system of in-orbit data centers. SO015, SO006, SO012
CO028 The FCC found the Cowboy Space Stampede application acceptable for filing per public notice Report No. SAT-02007 issued June 12, 2026; it is under review and authorization is not yet granted. SO015, SO025
CO029 Cowboy Space plans to launch its first satellite—a subscale power-beaming demonstration built by Apex Space—in 2026. SO007, SO003, SO005
CO030 An early 'Galactic Brain' orbital AI compute node using NVIDIA Space-1 Vera Rubin modules is planned for deployment in early 2027. SO007, SO005
CO031 The first proprietary Cowboy Space rocket launch carrying a 1-megawatt data center upper stage is targeted no earlier than the end of 2028. SO003, SO004, SO007
CO032 Each Stampede satellite-data-center is designed to weigh 20,000-25,000 kilograms, generate 1 megawatt of solar power, and host approximately 800 GPUs. SO004, SO011, SO007
CO033 Cowboy Space's core design innovation is treating the rocket's upper stage and the orbital data center payload as a single integrated vehicle, eliminating the structural redundancy of a conventional rocket-plus-separate-satellite architecture. SO001, SO002, SO003, SO004
CO034 Cowboy Space is collaborating with NVIDIA to deploy NVIDIA Space-1 Vera Rubin Modules—purpose-built LEO AI infrastructure—on its Stampede satellites. SO002, SO005, SO008
CO035 The Stampede constellation will use optical inter-satellite links as its primary communications mode; narrowband Ka-band spectrum (18.8-19.3 GHz downlink; 28.6-29.1 GHz uplink) is reserved for TT&C only. SO006, SO012, SO015
CO036 Cowboy Space requested FCC waivers including exemption from deployment schedule and bonding requirements, arguing the optical-only architecture does not create spectrum warehousing risk and service can begin with as few as one satellite. SO006, SO012, SO015
CO037 SatNews independently identified dissipating 1 MW of thermal load in vacuum and a 20,000-satellite debris footprint at 700-1,000 km altitude as material unresolved technical risks that Cowboy Space has not publicly addressed. SO011
CO038 TechCrunch noted that independently developing a launch vehicle puts Cowboy Space in direct competition with SpaceX and Blue Origin, the most advanced and well-funded players in the market, and described the rocket program as 'nuts' given the historical difficulty of the endeavor. SO004
CO039 Cowboy Space's own FCC filing acknowledges the satellite design is not yet finalized and will require a license modification before commercial service commences. SO006, SO012, SO015
CO040 Bhatt left Robinhood Markets after serving as Chief Creative Officer, departing in early 2024 to found Aetherflux full-time. SO009, SO023
CO041 Cowboy Space's Seattle office targets 40-60 employees initially, with 18 positions open as of May 2026 across avionics, mechanical engineering, spacecraft design, and software. SO008
CO042 GreyJournal estimated Bhatt's net worth at approximately $6 billion based on his ~6% Robinhood stake as of May 2026. SO023
CO043 Average grid connection lead times for new data centers in major US markets run five to seven years or more, a constraint Cowboy Space cites as the primary market driver for orbital data centers. SO001, SO004
CO045 Construct Capital was Cowboy Space's first institutional supporter, participating in the Series A and publishing a blog post in May 2026 describing the orbital data center thesis as core to its 'physical AI' investment mandate. SO019
CO046 Cowboy Space plans to build its own rocket engine—the most complex and expensive component of any launch vehicle—and is still working through key development needs including test, manufacturing, and launch facilities. SO004, SO007
CO047 Index Ventures partner Jan Hammer stated the firm backed Cowboy Space because Bhatt has 'a proven track record of reimagining massive markets from first principles' and 'his lifelong passion for physics makes space the ultimate market opportunity.' SO002, SO007
CO048 Bhatt said he expects the rocket to be eventually reusable but that the primary focus is on delivering maximum compute and power to orbit efficiently, not reuse for its own sake. SO007
CM001 Goldman Sachs Research projects that global data center power demand will grow 160% by 2030, driven primarily by AI workloads. SM001, SM023
CM002 A single ChatGPT query requires approximately 2.9 watt-hours of electricity, roughly ten times the 0.3 watt-hours of a Google search. SM001, SM023
CM003 Data centers currently consume 1–2% of global electricity and Goldman Sachs projects this will rise to 3–4% by 2030. SM001, SM023
CM004 US data centers will consume an estimated 8% of US electricity by 2030, up from approximately 3% in 2022, per Goldman Sachs Research. SM001, SM003
CM005 US utilities will need to invest roughly $50 billion in new generation capacity to support data center demand alone, according to Goldman Sachs. SM001, SM023
CM006 AI is projected to represent approximately 19% of total data center power demand by 2028, according to Goldman Sachs Research. SM001
CM007 AI is estimated to drive an increase of roughly 200 terawatt-hours per year in data center power consumption between 2023 and 2030. SM001
CM008 Data center carbon dioxide emissions may more than double between 2022 and 2030, representing a social cost of $125–140 billion at present value, per Goldman Sachs. SM001
CM009 Grid interconnection queues for AI hyperscalers are years deep in major markets as of 2026, per Construct Capital and Cowboy Space's own FCC application. SM013, SM023
CM010 The IEA's Electricity 2024 report includes in-depth analysis on the data centre sector's impact on electricity consumption, with demand forecasts through 2026. SM003, SM001
CM011 No independent authoritative orbital data center market sizing (TAM, SAM, or SOM) had been published by any recognized analyst firm as of June 2026. SM002, SM012
CM012 FCC filings for orbital data center constellations have been submitted by SpaceX (up to 1 million satellites), Blue Origin (up to 51,600), Starcloud (up to 88,000), and Cowboy Space (up to 20,000) as of May 2026. SM006, SM002, SM004
CM013 Europe's data center power demand will match the combined current electricity consumption of Portugal, Greece, and the Netherlands by 2030, per Goldman Sachs. SM001
CM014 The ODC category has attracted investments from top-tier venture firms including Index Ventures, Andreessen Horowitz, IVP, Breakthrough Energy Ventures, and Construct Capital as of 2026. SM005, SM013
CM015 Multiple competing FCC filings for orbital data center constellations are regulatory placeholders, not commercial revenue validation; no ODC company has disclosed revenue or contracted capacity as of June 2026. SM002, SM007
CM016 Lonestar Space claims to operate the world's first commercially operational space-based sovereign data storage platform, StarVault, with the first commercial StarVault launch scheduled for October 2026. SM009
CM017 Google's Project Suncatcher plans two prototype satellites with Planet by early 2027, testing solar-powered TPU clusters connected via optical links. SM002
CM018 The SAM and SOM for Cowboy Space's orbital data center business cannot be isolated from public sources in the absence of commercial validation data or independent market sizing. SM002, SM011
CM019 Earth observation operators are the nearest-term ODC customer segment because they generate large data volumes in orbit that cannot be fully downlinked before analysis is needed. SM002, SM007
CM020 Defense and intelligence customers value orbital data center resilience against ground-network disruption and forward-deployed sensor processing; SAIC's participation in Cowboy's Series B signals defense-sector interest. SM017, SM019, SM005
CM021 Commercial space station operators, including Axiom Space, have deployed data center units on the ISS and represent a near-term ODC customer segment. SM002
CM022 Frontier AI labs and hyperscalers represent the largest long-run prize for ODC operators but face the greatest near-term friction from data gravity, latency, network bandwidth, and security accreditation requirements. SM002, SM007
CM023 Sovereign data storage customers—regulated enterprises and governments concerned with data residency—are addressed by Lonestar Space's StarVault model rather than by Cowboy Space's compute-focused architecture. SM009, SM002
CM024 Defense procurement for ODC services involves long security accreditation cycles, and operational adoption requires standards, supply-chain review, integration with existing command systems, and long-term budget lines. SM002, SM019
CM025 Latency-sensitive consumer services, financial trading, gaming, and most real-time enterprise applications are explicitly excluded from orbital data center addressability due to LEO round-trip latency. SM002, SM007
CM026 Sophia Space describes the orbital compute market as driven by "the rapid increase in satellite deployments, private space stations, and commercial space activity," framing it as a space-native demand pull rather than a terrestrial migration. SM010
CM027 US net electricity generation increased 4.5% year-on-year through March 2026, consistent with the data center demand buildout narrative, per EIA. SM011, SM001
CM028 Cowboy Space's NVIDIA partnership for Space-1 Vera Rubin modules was announced as part of the Series B in May 2026, signaling hardware ecosystem readiness for LEO AI compute. SM005, SM007
CM029 Cowboy Space targets its first proprietary Stampede satellite launch before the end of 2028, per SatNews analysis of the FCC application. SM007, SM016
CM030 European grid investment requirements for data center growth exceed €800 billion in transmission and distribution plus ~€850 billion in new renewable generation over the coming decade, per Goldman Sachs. SM001
CM031 Starcloud's FCC application requests authority for up to 88,000 satellites providing GPU compute, persistent storage, and cloud-like services from LEO. SM006, SM002
CM032 Reusable launch vehicles have reduced the cost per kilogram to LEO substantially since 2010, and Cowboy Space's integrated upper-stage architecture aims to compress launch costs further by eliminating the boundary between rocket and payload. SM014, SM007
CM033 Terrestrial data center operators are actively pursuing grid-bypass alternatives including nuclear power purchase agreements, co-location with renewable generation, and liquid cooling, which may close the power gap before orbital infrastructure reaches scale. SM002, SM001
CM034 Dissipating one megawatt of thermal load in vacuum—as required by each Stampede satellite—is a non-trivial engineering exercise that has not been demonstrated at the scale proposed by Cowboy Space. SM007, SM002
CM035 A 20,000-satellite constellation at 700–1,000 km altitude carries debris liability that the optical-only architecture mitigates but does not eliminate, according to SatNews editorial analysis. SM007, SM002
CM036 Cowboy Space's own FCC application acknowledges that the satellite design is unfinished and will require a license modification before commercial service begins. SM004, SM007, SM023
CM037 Low Earth Orbit introduces round-trip latencies of roughly 10–500 ms depending on ground station proximity and network architecture, ruling out latency-sensitive inference and interactive applications. SM002, SM007
CM038 Data gravity is a structural constraint for AI training workloads: most training datasets are petabyte-scale and already stored in terrestrial cloud infrastructure, making orbital migration costly and logistically complex. SM002, SM007
CM039 Secure World Foundation's comments on the SpaceX and Starcloud FCC applications called for phased authorization, system-level risk analysis, and clearer post-mission disposal standards—precedents that will likely apply to Cowboy Space's Stampede filing. SM002
CM040 Orbital data center companies need staff bridging spacecraft engineering, cloud systems, AI accelerators, optical networks, export controls, and mission operations—labor pools that do not fully overlap, creating a workforce constraint. SM002
CM041 Terrestrial cloud hyperscalers (AWS, Azure, GCP) and energy-co-located operators like Crusoe represent the primary competitive substitutes for orbital data center services. SM002, SM001
CM042 Starcloud's Starcloud-1 mission flew an NVIDIA H100 GPU in orbit and is the first publicly documented operational orbital compute demonstration involving a commercial GPU. SM002, SM008
CM043 Cowboy Space's Stampede FCC application seeks a waiver from standard FCC deployment milestone rules, arguing that the optical-only link design means spectrum warehousing concerns do not apply. SM006, SM016, SM023
CM044 The ODC unit economics depend on lower launch costs, high processor duty cycles, efficient heat rejection, long satellite life, strong customer willingness to pay, and limited service interruption—no company has demonstrated all five simultaneously as of June 2026. SM002, SM007
CM045 SAIC's inclusion as a Series B investor alongside Breakthrough Energy Ventures, Index Ventures, and Andreessen Horowitz is an observable signal of defense-sector confidence in orbital data center infrastructure. SM005, SM017
CM046 Cowboy Space's FCC application specifies an operational satellite lifetime of five years, meaning the constellation requires continuous replenishment to sustain commercial service beyond the initial deployment. SM004, SM007
CM047 Total disclosed equity raised by Cowboy Space reached approximately $325 million as of May 2026 ($50M Series A plus $275M Series B); no aggregated investment total for the broader orbital data center startup cohort has been independently published as of June 2026. SM007, SM005
CP001 Starcloud (formerly Lumen Orbit) launched its first satellite, Starcloud-1, in November 2025 carrying an NVIDIA H100 GPU, the first high-end AI accelerator verified in low Earth orbit. SP010, SP011
CP002 The Starcloud-1 mission ran a version of Google's Gemma language model in space and trained a nanoGPT model onboard, demonstrating GPU-class AI inference and training in orbit. SP010, SP011
CP003 In February 2026, Starcloud filed with the FCC for authority to deploy and operate up to 88,000 satellites as a distributed data center in sun-synchronous orbits between 600 and 850 km. SP010, SP017
CP004 Starcloud's planned commercial mission Starcloud-2 is designed to offer GPU clusters, persistent storage, and 24/7 access in a smallsat form factor. SP010, SP011
CP005 Crusoe Energy announced in October 2025 that it would deploy Crusoe Cloud on a Starcloud satellite scheduled for late 2026, making Crusoe the first commercial cloud tenant committed to an orbital compute platform—and that platform is Starcloud, not Cowboy Space. SP010, SP002
CP006 On January 30, 2026, Space Exploration Holdings filed an FCC application for a non-geostationary satellite system of up to one million satellites called the "SpaceX Orbital Data Center System," with altitudes from 500 to 2,000 km primarily relying on optical intersatellite links. SP010, SP017, SP018
CP007 SpaceX's ODC constellation filing potentially connects to its existing first- and second-generation Starlink systems, giving it a network integration advantage unavailable to any other ODC entrant. SP010, SP018
CP008 SpaceX's ownership of Starship, the only vehicle capable of cost-effectively lifting 20-to-25-tonne ODC satellites at scale, gives it a structural launch cost advantage over all competing ODC entrants including Cowboy Space. SP010, SP016, SP017
CP009 Blue Origin filed an FCC application for Project Sunrise, a proposed orbital data center network of up to 51,600 satellites in sun-synchronous orbits between 500 and 1,800 km routing traffic through optical links and its TeraWave connectivity architecture. SP010, SP017, SP009
CP010 Blue Origin's Project Sunrise pairs orbital compute with its TeraWave broadband connectivity, potentially offering an integrated AI-data-center-plus-broadband bundle that Cowboy's narrower compute-only proposal cannot match. SP010, SP009
CP011 Sophia Space's Thermal Integrated LEO Edge (TILE) architecture places compute modules adjacent to radiative heat-rejection panels at the point of heat generation, directly addressing orbital thermal management constraints that no other ODC architecture has publicly solved. SP010, SP013
CP012 NVIDIA's space computing announcement named Cowboy Space (then Aetherflux), Axiom Space, Kepler Communications, Planet Labs, Sophia Space, and Starcloud simultaneously as users of its accelerated computing platforms—indicating the NVIDIA Space-1 supplier relationship is not exclusive to Cowboy. SP010, SP013
CP013 Axiom Space deployed Data Center Unit-1 (AxDCU-1) on the International Space Station in fall 2025, supporting cloud computing, AI, machine learning, data fusion, storage, and space cybersecurity on Red Hat Device Edge. SP008, SP010
CP014 Axiom Space said it launched its first two dedicated orbital data center nodes to LEO on January 11, 2026, coinciding with the first tranche of Kepler Communications' optical relay network constellation. SP008, SP010
CP015 Axiom Space and Spacebilt announced in September 2025 a collaboration for the AxODC Node ISS project in 2027, assembling hardware from Spacebilt (Large In-Space Servers), Skyloom (optical), Phison Electronics (SSDs), and Microchip Technology (PIC64 spaceflight computing). SP010, SP007
CP016 Kepler Communications deployed 40 NVIDIA Jetson Orin modules across 10 satellites in its constellation, enabling AI and accelerated workloads within a commercially operational optical data relay network. SP010
CP017 Lonestar Data Holdings is accepting capacity reservations for its first StarVault orbital sovereign data storage platform planned to launch in October 2026, focusing on disaster recovery and regulatory-compliant data backup rather than GPU-class AI compute. SP012, SP010
CP018 Lonestar's StarVault service model—sovereign storage and cryptographic key escrow in orbit—does not compete directly with Cowboy's GPU compute thesis but illustrates how narrow the near-term commercially viable ODC customer base is. SP012
CP019 CoreWeave is a commercially operational AI cloud platform offering NVIDIA Vera Rubin NVL72 rack-scale GPU compute and describes itself as the world's #1 AI cloud platform purpose-built for frontier AI workloads. SP001, SP010
CP020 Crusoe's AI cloud claims 99.98% uptime with NVIDIA GB200 NVL72, HGX B200, and AMD MI355x hardware, and up to 9.9× faster time-to-first-token on its managed inference service. SP002
CP021 Voltage Park offers NVIDIA H100, Blackwell, and Grace Blackwell GPU cloud access starting at $1.99 per hour on-demand with no long-term contracts, serving AI labs and fast-scaling startups. SP003, SP001
CP022 Terrestrial specialized AI clouds—CoreWeave, Crusoe, and Voltage Park—are commercially operational today with the latest NVIDIA GPUs, years ahead of any projected Cowboy Space commercial deployment, serving the same AI training and inference workloads Cowboy targets. SP001, SP002, SP003
CP023 The Stargate Project, announced in January 2025, plans to invest $500 billion over four years in US terrestrial AI infrastructure for OpenAI, with $100 billion deployed immediately, backed by SoftBank, OpenAI, Oracle, and MGX. SP004, SP023
CP024 Stargate's technology partners include Arm, Microsoft, NVIDIA, Oracle, and OpenAI, with buildout starting in Texas—this program absorbs the same AI power-demand signal that is central to Cowboy Space's investor narrative. SP004, SP023
CP025 AWS Ground Station offers satellite operators a managed service to control communications and process data via AWS global infrastructure, extending terrestrial cloud processing to space missions without requiring orbital compute hardware. SP005, SP010
CP026 TechCrunch confirmed that Cowboy Space CEO Baiju Bhatt approached multiple launch providers and concluded the commercial launch market would be too scarce and expensive to achieve competitive unit economics at scale for ODC, leading the company to build its own rocket—a decision unique among pure-play ODC startups. SP016, SP019, SP022
CP027 Cowboy Space's Stampede FCC filing for 20,000 satellites is the smallest mega-constellation proposal among ODC filers: SpaceX (1 million), Starcloud (88,000), and Blue Origin (51,600) have all filed larger systems. SP017, SP010, SP014
CP028 Cowboy's CEO stated that most first-party rocket providers will specialize into their own payloads, creating a structural launch-capacity scarcity for satellite-only ODC players that validates the vertical integration thesis—but also implies SpaceX will likely not be a launch partner for competing ODC systems. SP016, SP019
CP029 Cowboy Space's own FCC filing for the Stampede constellation acknowledged that the satellite design is unfinished and will require a license modification before service begins, an unusual admission at the regulatory filing stage. SP020, SP014, SP017
CP030 TechCrunch characterized Cowboy's rocket development decision as "also nuts," noting that only SpaceX, Rocket Lab, and Arianespace are consistently launching commercial rockets in the West, and that Stoke Space, Firefly Aerospace, and Relativity Space have worked for years without operational systems. SP016, SP023
CP031 SpaceBelt, the orbital data network operated by Cloud Constellation Corporation, has pursued a specialist orbital data storage and connectivity thesis since at least 2020 with a consortium of telecom partners but has not achieved commercial scale, illustrating the historical execution difficulty in the orbital data infrastructure market. SP006
CP032 NVIDIA's Space-1 Vera Rubin module partnership is shared among at least four direct competitors—Starcloud, Sophia Space, Axiom Space, and Cowboy Space—indicating that hardware supply from NVIDIA will not be a differentiating moat for any single ODC entrant. SP010, SP013, SP011
CP033 Starcloud's two-plus-year hardware maturity lead over Cowboy Space—with an H100 in orbit since November 2025 versus Cowboy's earliest orbital demonstration expected in 2027—means Cowboy is not a first-mover but a fourth-or-later entrant in deployed orbital AI compute hardware. SP010, SP015, SP016
CP034 Cowboy Space must achieve simultaneous breakthroughs in launch vehicle development, orbital thermal management for 1 MW loads, NVIDIA GPU space qualification, optical communications networking, and commercial customer acquisition within a compressed 3-to-5-year window to establish competitive leadership before well-resourced peers. SP016, SP014, SP019
CP035 Spacebilt's public positioning—"we want to define the category, not compete within it"—signals it views itself as the infrastructure standard-setter for in-space multi-tenant AI compute servers, putting it in direct category conflict with Cowboy's integrated upper-stage model. SP007
CP036 Blue Origin's New Glenn heavy-lift vehicle provides Project Sunrise with launch independence that satellite-only ODC startups lack, creating a competitive parallel to Cowboy's own vertical integration strategy but from a far better-resourced entrant. SP009, SP010
CP037 NVIDIA's simultaneous Space-1 partnerships with multiple ODC competitors means competition in orbital compute will shift to launch cost, thermal efficiency, orbit selection, ground network topology, and software—areas where Cowboy has made the most unproven claims. SP010, SP012, SP013
CI001 Cowboy Space Corporation closed a $275 million Series B financing in May 2026, confirmed by both the official BusinessWire press release and CNBC's independent reporting. SI021, SI001
CI002 The Series B was led by Index Ventures, with IVP, Blossom Capital, and SAIC as new investors alongside all major existing investors. SI021, SI001, SI019
CI003 Existing investors Breakthrough Energy Ventures, Construct Capital, Andreessen Horowitz, NEA, Interlagos, and CEO Baiju Bhatt all participated in the Series B round. SI021, SI002, SI007
CI004 Cowboy Space achieved a $2 billion post-money valuation at the Series B close in May 2026, as confirmed by the BusinessWire press release and IEEE Spectrum coverage. SI021, SI006
CI005 Aetherflux raised a $50 million Series A in April 2025, led by Index Ventures and Interlagos, with participation from Breakthrough Energy Ventures, Andreessen Horowitz, and angel investors including Vlad Tenev. SI007, SI002, SI023
CI006 Cowboy Space's founding-stage capital (seed and prior investments) totaled approximately $40 million, contributed by Breakthrough Energy Ventures, Andreessen Horowitz, NEA, and angel investors at the Aetherflux launch in 2024. SI002, SI007, SI021
CI007 CEO Baiju Bhatt publicly confirmed total capital raised by Cowboy Space of approximately $365 million as of the May 2026 Series B close, encompassing all prior rounds. SI021, SI001
CI008 Fenwick and West LLP confirmed it served as legal counsel to Cowboy Space in the Series B financing, with corporate partners Morgan Sawchuk and Ran Ben Tzur leading the representation. SI022
CI009 SAIC's investment in the Series B was characterized in press coverage as strategic, given SAIC's position as a leading US defense IT contractor with interest in orbital compute for government and military applications. SI001, SI023, SI025
CI010 The US Department of Defense Operational Energy Capability Improvement Fund (OECIF) selected Aetherflux for a proof-of-concept wireless power transmission contract; the contract value was not publicly disclosed. SI023, SI025, SI019
CI011 Cowboy Space has not disclosed any revenue, ARR, backlog, customer contracts, or letters of intent as of the Series B close in May 2026; the company is pre-revenue by all public measures. SI021, SI003, SI001
CI012 No pricing schedules, reference customer proposals, or financial projections have been published by Cowboy Space through its Series B announcement. SI021, SI003, SI009
CI013 The Series B press release cited three use-of-funds priorities: advancing rocket engine R&D, satellite-embedded compute integration, and regulatory compliance for the Stampede constellation. SI021
CI014 Cowboy Space is developing a proprietary small-to-medium lift rocket intended to serve as the primary launch vehicle for its own Stampede orbital data center satellites. SI003, SI023, SI021
CI015 GeekWire reported that Cowboy Space plans to hire 40–60 employees for its new satellite and rocket manufacturing hub in Seattle following the Series B close. SI024
CI016 Comparable small-to-medium launch vehicle programs — Relativity Space, ABL Space Systems, and Stoke Space — consumed between $100 million and $650 million in capital before reaching first operational milestones, and several failed before achieving them. SI012, SI003, SI011
CI017 NVIDIA's Space-1 Vera Rubin modules are engineered for LEO thermal and radiation environments, targeting sub-1 MW power budgets per satellite-class orbital data center deployment. SI004, SI016
CI018 NVIDIA Space-1 Leo-class AI data center modules target a power consumption profile compatible with satellite solar array budgets; exact unit pricing has not been publicly disclosed. SI016, SI004
CI019 ArsTechnica's physics analysis of orbital data centers found that rejecting 1 MW of compute heat via radiative cooling in vacuum requires approximately 200 square meters of radiator surface area — an engineering challenge not yet demonstrated at production scale. SI011, SI013
CI020 Terrestrial hyperscale GPU servers cost approximately $250,000–$400,000 per rack (H100-class); space-grade equivalents are estimated at three to ten times this cost due to radiation hardening, qualification, and smaller production volumes. SI009, SI013
CI021 JLL estimates global data center construction will require over $250 billion in cumulative capex through 2030 to meet AI-driven demand, contextualizing the terrestrial scale against which Cowboy Space must compete. SI010
CI022 PitchBook reported the space data center sector received over $1 billion in venture funding in 2026, indicating strong investor thesis alignment but also intensifying competitive capital deployment. SI012
CI023 ArsTechnica's article "Will Space Data Centers Actually Save Us From the AI Power Crisis?" raised substantive questions about whether orbital compute economics are viable at early customer scale absent captive defense demand. SI011
CI024 MarketsAndMarkets published a forecast sizing the orbital data center market, providing a speculative TAM benchmark, though the methodology relies on extrapolations not yet validated by commercial deployments. SI018
CI025 Fortune Business Insights projects the global data center market will exceed $700 billion by 2030, anchoring the terrestrial context that Cowboy Space's orbital compute must capture a share of to be financially viable. SI008
CI026 Andreessen Horowitz's portfolio page confirms its investment in Aetherflux (the predecessor brand) as part of its climate and clean-energy portfolio, providing independent investor confirmation of the Series A. SI007
CI027 At a 40–60 person aerospace team with fully-loaded compensation of $250,000–$350,000 per head, Cowboy Space's monthly payroll-driven burn is estimated at approximately $1.0–$1.75 million, with total operating burn (including facility, procurement, and legal) estimated at $3–$8 million per month. SI024, SI023
CI028 At a midpoint burn estimate of $5 million per month, the $275 million Series B represents approximately 55 months (4.6 years) of operating runway with no revenue or additional capital. SI021, SI024
CI029 At an aggressive burn of $10 million per month — consistent with accelerated rocket R&D and satellite prototype procurement — the $275 million Series B provides approximately 27 months of runway, indicating a Series C trigger may be required as early as mid-2028. SI021, SI003
CI030 Assuming a $150–$200 million post-money Series A (consistent with Index Ventures leading a $50 million round at a 25–33% ownership target), the $2 billion Series B represents an approximately 10–13x valuation step-up in roughly 12 months, consistent with thesis-stage AI-adjacent venture pricing in 2025–2026. SI007, SI006, SI005
CI031 No customer contracts, signed LOIs, pilot agreements, or disclosed revenue figures appear in any public source covering Cowboy Space through June 2026; the company has made no public traction claims. SI003, SI009, SI019
CI032 The DoD OECIF contract dollar amount and any SAIC procurement framework have not been disclosed in any public source; government revenue pathways exist structurally but remain unquantified for financial modeling purposes. SI023, SI025, SI019
CI033 Cowboy Space filed an FCC SAT-PDR partial designation application (SAT-PDR-20260514-00022) on May 14, 2026, covering a subset of the planned Stampede constellation frequency bands; the FCC found the original application acceptable for filing in June 2026. SI015, SI025
CI034 FCC review and authorization of a 20,000-satellite constellation typically requires 12–24 months minimum from acceptable-for-filing determination; the Stampede constellation remains under regulatory review as of June 2026. SI015, SI025
CI035 At an estimated $5 million per satellite (manufacturing plus launch cost), a 20,000-satellite Stampede constellation would require approximately $100 billion in total capex — roughly 50 times the capital raised to date and far beyond any individual venture capital fundraise. SI023, SI025, SI010
CI036 First commercial revenue from orbital compute services is contingent on completing rocket development, achieving launch cadence, building compute-integrated satellites, obtaining FCC spectrum authorization, and closing customer contracts — a sequential dependency chain spanning multiple years and funding rounds. SI003, SI009, SI021
CI037 No third-party financial audit, S-1 filing, SPAC prospectus, or regulatory financial statement has been filed for Cowboy Space; all financial claims rest on company statements, investor communications, and media coverage. SI021, SI009, SI003
CI038 ArsTechnica's analysis of space-based data center physics identified unproven radiative cooling physics at 1 MW scale as a fundamental engineering risk that may limit per-satellite compute density below economically viable thresholds. SI013, SI011
CI039 The $2 billion valuation for a pre-revenue, pre-launch company is unprecedented among comparable independent orbital ventures; SpaceX and Blue Origin achieved equivalent or greater valuations only after demonstrating operational launch capability and, in SpaceX's case, substantial Starlink revenue. SI006, SI012, SI005
CI040 TechCrunch noted that independently developing a launch vehicle puts Cowboy Space in direct capital competition with SpaceX and Blue Origin, adding rocket-program risk on top of the orbital compute thesis — a dual-dependency that has historically proven fatal to smaller launch startups. SI003, SI011
CI041 Construct Capital's May 2026 blog post framed Cowboy Space's market entry around an energy bottleneck thesis, explicitly citing government contracts as an early revenue anchor alongside commercial ODC services; beyond OECIF, Cowboy Space's defense heritage positions it for SBIR, STTR, and DARPA programs targeting space-based edge compute, though no specific awards have been publicly announced. SI021, SI023, SI019
CI042 No public indication exists that the Cowboy Space Series B was oversubscribed or that the company initially sought a valuation higher than $2 billion; the round dynamics are consistent with a negotiated price at strong but not exceptional demand, typical for large thesis-stage AI-adjacent raises in mid-2026. SI005, SI019, SI006
CI043 The investor quality of the Series B — Index Ventures, IVP, Blossom Capital, and SAIC alongside top-tier existing investors — is consistent with a well-managed process that attracted institutional demand, but does not by itself indicate oversubscription or price-setting pressure from below. SI001, SI021
CI044 Based on estimated monthly operating burn of $3–$10 million, capital exhaustion risk becomes material after 27–92 months; if rocket development accelerates and capex peaks above $10 million per month before any revenue, Cowboy Space may require a Series C as early as Q3 2028 — well before first commercial operations. SI003, SI023, SI016
CI045 PitchBook data and comparable program reporting show Rocket Lab raised $140 million at Series B while still pre-revenue; Relativity Space raised $650 million before winding down Terran 1; ABL Space closed after its Series B; Stoke Space has raised under $100 million. Cowboy Space's $275 million is within the peer range but insufficient for full-scale vertically integrated operations. SI012, SI003, SI011
CE001 Cowboy Space's core product architecture integrates the rocket upper stage and the orbital data center into a single vehicle, eliminating the payload adapter and redundant structure present in conventional launch-plus-satellite designs. SE001, SE002, SE003
CE002 Each Stampede production satellite is designed to weigh 20,000 to 25,000 kilograms. SE002, SE003, SE008
CE003 Each Stampede satellite is designed to generate 1 megawatt of usable electrical power from solar panels deployed in a dawn-dusk sun-synchronous orbit. SE001, SE002, SE004
CE004 Each Stampede satellite is designed to host approximately 800 GPUs. SE002, SE003, SE008
CE005 The NVIDIA Space-1 Vera Rubin module pairs an 88-core CPU with two Rubin graphics chips; a single Rubin chip delivers 50 petaflops of NVFP4 performance. SE008, SE009, SE015
CE006 The NVIDIA Space-1 Vera Rubin module is designed specifically for LEO AI computing and was unveiled at NVIDIA GTC in March 2026. SE009, SE015, SE024
CE007 The Stampede satellite design targets a rocket larger than SpaceX's Falcon 9 but smaller than Starship in payload capacity, targeting 20,000–25,000 kg to orbit. SE002, SE003
CE008 Cowboy Space CEO Baiju Bhatt stated that each upper stage leverages the stage structure itself as a radiator for thermal management of the compute payload. SE002, SE001
CE009 Reusability of the launch booster is described by Cowboy Space as a design goal 'where it makes economic sense,' explicitly subordinated to maximizing compute mass delivered per launch. SE002, SE003
CE010 Cowboy Space was named on NVIDIA's public launch-partner slate for the Space-1 Vera Rubin program, confirmed by NVIDIA press materials and multiple independent media sources. SE009, SE015, SE006, SE007
CE011 NVIDIA's space computing product page states that Cowboy Space Corporation is integrating NVIDIA Vera Rubin Space-1 Module, IGX Thor, and Jetson Orin into its Stampede satellites. SE006, SE015
CE012 Cowboy Space's product origin was space-based solar power beaming (Aetherflux brand, 2024–2025), which pivoted to using solar power generated in orbit for in-orbit AI compute rather than terrestrial transmission. SE011, SE003, SE016
CE013 Cowboy Space filed FCC application SAT-LOA-20260323-00135 for the Stampede constellation on approximately March 23, 2026, three days after closing the Series B round. SE022, SE004, SE005
CE014 The Stampede FCC filing requests authority for up to 20,000 non-geostationary orbit satellites in dawn-dusk sun-synchronous orbit between 700 and 1,000 km altitude. SE004, SE005, SE006, SE022
CE015 Stampede satellites are designed to operate on dawn-dusk sun-synchronous orbit, which maximizes solar exposure duration per day compared to other LEO orbit planes. SE006, SE004
CE016 The Stampede constellation is designed to operate primarily through optical inter-satellite links and optical downlinks rather than radiofrequency bands, with only narrowband Ka-band spectrum requested for TT&C. SE005, SE006, SE004, SE022
CE017 The Ka-band spectrum requested in the Stampede FCC filing is 18.8–19.3 GHz for downlink and 28.6–29.1 GHz for uplink, limited to narrowband TT&C during mission-critical phases. SE006
CE018 The Stampede FCC filing requests a waiver from standard deployment bond and milestone rules on the grounds that those rules are designed to prevent spectrum warehousing, which is inapplicable to an optical-primary system. SE005, SE006, SE004, SE022
CE019 The Stampede FCC application has not been granted as of June 2026; it is under review with waivers pending. SE005, SE006, SE022
CE020 The Stampede FCC filing's Schedule S form describes four representative satellites in four orbital planes at 830–860 km apogee/perigee with a five-year estimated operational lifetime. SE006
CE021 The Stampede FCC filing is the largest orbital data center deployment plan filed with the FCC as of May 2026, larger than Starcloud's 88,000-satellite filing. SE004, SE005
CE022 The Stampede FCC filing explicitly acknowledges that the satellite design is unfinished and will require a license modification before commercial service begins. SE004, SE006, SE022
CE023 NVIDIA's three-component Stampede compute stack consists of the Vera Rubin Space-1 Module (primary GPU training/inference), IGX Thor (real-time inference orchestration), and Jetson Orin (autonomous satellite operations). SE006, SE015, SE024
CE024 Cowboy Space's optical inter-satellite link architecture requires free-space optical terminals with pointing/acquisition/tracking capability; no terminal vendor, link budget, or availability target has been publicly specified. SE004, SE005, SE006
CE025 The Stampede FCC filing does not describe the ground-station infrastructure, data downlink architecture, or coverage model for delivering compute results to customers. SE006, SE005
CE026 NVIDIA describes its Space-1 modules as radiation-tolerant for LEO operations, but no independent qualification data sheet or MIL-SPEC equivalent has been publicly published. SE009, SE024, SE025
CE027 The first demonstration satellite is being built in partnership with Apex Space, a Los Angeles-based manufacturer with a production capacity of over 200 satellite buses per year and a factory targeting 100,000+ sq ft by end of 2026. SE002, SE007, SE010
CE028 Apex uses its proprietary 'Octopus' manufacturing OS to manage assembly scheduling, inventory, and work instructions across its satellite production floor. SE010
CE029 Apex's role in production-rate Stampede manufacturing beyond the first demo satellite has not been publicly confirmed in contracts or press releases. SE002, SE007, SE010
CE030 SiliconAngle noted that Cowboy Space had not specified how it plans to interconnect multiple 1 MW data center modules into larger GPU clusters for workloads exceeding single-satellite capacity. SE008
CE031 The Stampede satellite's 1 MW per unit compute capacity represents a fraction of the processing power available in terrestrial AI training environments, which commonly operate at 10 MW–100 MW cluster scale. SE008, SE025
CE032 Dissipating 1 MW of compute waste heat in vacuum via thermal radiation alone requires approximately 2,000–4,000 square meters of radiator surface at operational GPU temperatures, a constraint with no commercial satellite precedent at that scale. SE025, SE026, SE018
CE033 Technical analyses published in 2026 found that even an optimized single-megawatt orbital compute node operating at elevated temperatures requires roughly 200 square meters of radiator area as a minimum estimate, which Cowboy proposes to supply via the rocket stage structure. SE004, SE026, SE025
CE034 LEO altitudes of 700–1,000 km expose satellites to energetic protons and electrons from the South Atlantic Anomaly and Van Allen belt fringes, creating single-event upset and total ionizing dose risks for unshielded commercial GPU silicon. SE025, SE026
CE035 No publicly available thermal model, prototype test result, or independent engineering review of the stage-as-radiator approach for a 1 MW compute payload has been published by Cowboy Space or any third party. SE004, SE025, SE026
CE036 U.S. data center grid connection lead times in major markets run 5–7 years or more, cited by Cowboy Space and multiple independent sources as the primary terrestrial bottleneck motivating orbital compute. SE001, SE006, SE018
CE037 The Stampede FCC filing states that commercial operations can begin with as few as one satellite in orbit, with constellation size scaled by market economics rather than regulatory milestones. SE005, SE006, SE022
CE038 The Stampede FCC filing does not describe a deorbit or debris mitigation architecture for the 20,000-satellite constellation operating at 700–1,000 km, where atmospheric drag decay without active propulsion can take decades. SE004, SE006
CE039 TechCrunch reported in May 2026 that Cowboy Space is 'still working through key development needs, like facilities to test, manufacture, and launch its rockets,' confirming no rocket engine test infrastructure exists. SE003, SE014
CE040 TechCrunch noted that only a handful of private companies — mainly SpaceX, Rocket Lab, and Arianespace — are consistently launching commercial rockets, framing Cowboy's launch ambition against a very thin competitive field of successful operators. SE003, SE014
CE041 Cowboy Space's first milestone is a 2026 small-satellite launch built with Apex to demonstrate wireless infrared laser power beaming from LEO to a ground receiver; specifications were still being finalized at the Series B announcement. SE002, SE007, SE011
CE042 CEO Bhatt described the first power-beaming mission as a 'subscale technology demonstrator' sized to prove the physics, explicitly noting the long-term business focus is in-orbit compute, not terrestrial power transmission. SE002
CE043 Cowboy Space's second milestone is the deployment of the first 'Galactic Brain' orbital AI data center node using NVIDIA Space-1 Vera Rubin modules, targeted for early 2027. SE002, SE007
CE044 Cowboy Space's third and most critical milestone is the first proprietary rocket launch carrying a 1-megawatt data center upper stage, targeted before the end of 2028. SE002, SE003, SE005
CE045 Tyler Grinnell, former SpaceX director of launch operations, has joined Cowboy Space to lead launch operations development. SE002, SE007
CE046 Warren Lamont, a former Blue Origin propulsion engineer who led rocket engine and booster stage development, leads rocket propulsion development at Cowboy Space from its Los Angeles office. SE002, SE003, SE007
CE047 David Larson, a SpaceX and Amazon veteran, serves as Director of Satellite Engineering in Cowboy Space's Seattle office. SE007
CE048 Cowboy Space's Seattle satellite engineering center targets 40–60 initial employees and had 18 open positions advertised across avionics, mechanical engineering, spacecraft design, and software as of GeekWire's May 2026 report. SE007
CE049 Cowboy Space has stated that constellation scale will be driven by economics and validated performance rather than regulatory milestones, implying no committed annual launch cadence for Stampede scaling. SE002, SE005
CE050 Cowboy Space's vertical integration advantage over orbital compute competitors relying on SpaceX Starship is predicated on launch availability: the current commercial launch manifest cannot accommodate a 20,000-satellite, 20-tonne-per-bird constellation on any existing third-party rocket. SE003, SE004, SE014
CE051 Starcloud (formerly Lumen Orbit) launched the first H100 GPU satellite and achieved unicorn status in 2026 with a $170M Series A, representing the most operationally advanced orbital data center competitor versus Cowboy's pre-hardware status. SE004, SE019, SE002
CE052 Space-grade GPU hardware is estimated at 3–10 times the procurement cost of equivalent terrestrial hyperscale GPU hardware, representing a structural cost disadvantage for orbital compute versus terrestrial alternatives. SE025, SE018
CE053 SAIC's Series B participation as a strategic investor signals potential defense and government procurement interest in orbital compute, but no defense contract or government customer has been publicly announced for Stampede. SE017, SE020, SE023
CE054 No orbital data center operating at 1 megawatt or higher compute density has been demonstrated anywhere in the world as of June 2026; Starcloud's H100 GPU satellite is the most advanced public deployment but operates at far lower power density. SE025, SE026, SE018
CE055 Cowboy Space has not specified its inter-satellite clustering architecture, inter-node latency budget, or workload orchestration layer for AI jobs requiring more compute than a single 1 MW Stampede satellite can provide. SE008, SE006
CU001 AI hyperscalers and cloud providers face years-long grid interconnection queues that constrain their ability to add terrestrial data center capacity, creating structural demand for energy alternatives. SU006, SU009, SU028
CU002 Cowboy Space positions itself as a solution to terrestrial AI compute capacity constraints, targeting AI hyperscalers, AI labs, and defense customers as primary buyers. SU001, SU004
CU003 NewSpaceEconomy identifies five early orbital data center customer archetypes: Earth observation operators, defense and security agencies, space station users, lunar infrastructure firms, and AI developers facing unusual energy or location constraints. SU007
CU004 Latency-sensitive workloads — including consumer applications, financial trading, and gaming — are structurally unsuitable for orbital compute due to inherent round-trip propagation delays from low Earth orbit. SU007, SU012
CU005 Construct Capital's investment thesis frames Cowboy Space's target market as companies blocked by terrestrial energy grid constraints from expanding AI compute capacity, noting that 'data centers are already facing years-long waits for grid access.' SU009
CU006 Defense and intelligence customers are identified as especially likely early ODC adopters because military and intelligence users value orbital compute's resilience, autonomy, and protection against ground infrastructure disruption. SU007, SU008
CU007 Earth observation satellite operators generate more imagery than available downlink bandwidth allows, creating demand for on-orbit processing to reduce transmission volumes and enable faster data delivery. SU007
CU008 Space station operators represent a near-term ODC customer segment; Axiom Space deployed Data Center Unit-1 (AxDCU-1) onboard the International Space Station in fall 2025, establishing a proof point for station-based compute. SU007
CU009 AI model training clusters requiring enormous internal bandwidth and dependable power face the highest adoption uncertainty for orbital compute, due to latency, data pipeline re-engineering, and workflow complexity barriers. SU007, SU011
CU010 Data sovereignty customers may value orbital storage outside terrestrial jurisdictions, but the legal status of such arrangements remains unsettled because customers still contract with an Earth-based company under national law. SU007
CU011 Cowboy Space has not disclosed any named customers, signed revenue contracts, letters of intent, or pilot agreements as of June 2026; the company is entirely pre-commercial. SU001, SU002, SU003
CU012 SAIC participated in the Cowboy Space Series B as a new strategic investor alongside IVP and Blossom Capital; Washington Technology identified SAIC as the key GovCon investor name to watch. SU001, SU008
CU013 SAIC's strategic investment rationale is publicly stated as positioning SAIC to access Cowboy Space's orbital compute capabilities for future defense workloads, per Washington Technology's GovCon investment coverage. SU008, SU019
CU014 SAIC has not confirmed any commercial agreement, service-level commitment, or pilot deployment plan with Cowboy Space beyond its equity investment participation. SU001, SU008
CU015 AI company Anthropic expressed interest in using SpaceX's proposed orbital data centers as part of a broader compute access agreement, providing a sector-level AI-lab demand signal for the ODC market but not specific to Cowboy Space. SU002
CU016 Index Ventures partner Jan Hammer stated: 'Cowboy is building at exactly the right moment, as demand for AI compute and energy begins to outstrip what terrestrial infrastructure can support,' providing the investment rationale for the Series B. SU001, SU002
CU017 The U.S. Department of Defense Operational Energy Capability Improvement Fund (OECIF) previously selected predecessor company Aetherflux for orbital energy defense work, establishing a government-side familiarity with Cowboy Space's technology lineage. SU008
CU018 NVIDIA named Cowboy Space as a partner in its Space-1 Vera Rubin Module program, a supply-side technology partnership that does not constitute customer proof or validate end-user demand for Cowboy's services. SU001, SU014
CU019 Microsoft's Azure Orbital product (redirecting to the Planetary Computer Earth observation platform) demonstrates hyperscaler investment in space infrastructure, but this is a ground-station connectivity service, not a customer commitment to purchase orbital compute from a third party. SU013
CU020 Planet Labs is identified by NVIDIA's space computing partner program and NewSpaceEconomy's ODC market analysis as a relevant Earth observation customer and mission partner for the orbital data center sector. SU007, SU014
CU021 Cowboy Space has not disclosed a go-to-market strategy, sales organization, or customer acquisition roadmap as of June 2026, and its publicly advertised roles are predominantly technical (satellite design, propulsion engineering), not commercial. SU005, SU001
CU022 The implied Cowboy Space go-to-market motion is B2B enterprise direct sales targeting energy-constrained AI compute buyers, with SAIC's strategic investment representing the closest analog to an anchor channel relationship for the defense segment. SU008, SU001
CU023 Cowboy Space CEO Baiju Bhatt stated that 'initial commercial services will follow once we've validated performance, reliability, and economics at scale,' confirming no near-term commercial customer engagement is planned. SU002
CU024 Cowboy Space's FCC application for 20,000 satellites under the constellation name 'Stampede' represents a regulatory filing establishing an orbital data center business, not a customer contract or revenue commitment. SU021, SU023
CU025 First commercial revenue from Cowboy Space's orbital compute services is contingent on completing rocket development, achieving first proprietary launch (targeted before end of 2028), and validating performance and economics at scale. SU002, SU004
CU026 OpenAI's Stargate program commits up to $500 billion to terrestrial AI infrastructure investment, representing a large-scale alternative to orbital compute that would reduce the addressable customer demand pool if terrestrial capacity is sufficient. SU022
CU027 The NewSpaceEconomy analysis explicitly states that orbital data center companies 'must compete against terrestrial improvements, not against the most constrained terrestrial facilities alone,' meaning hyperscalers' active mitigation of ground-based constraints directly challenges the ODC value proposition. SU007
CU028 The challenge of dissipating one megawatt of thermal load in vacuum is a substantive engineering barrier to orbital compute; Cowboy Space's own FCC filing acknowledges the satellite design is unfinished and will require a license modification before service begins. SU006, SU012
CU029 Technical community responses to Cowboy Space's fundraise (reflected in the HN thread and Register-style critical coverage) expressed skepticism about the customer demand thesis, questioning why hyperscalers would pay an orbital premium when terrestrial alternatives are expanding. SU011, SU018, SU020
CU030 Regulatory, insurance, and cybersecurity compliance requirements for orbital platforms add procurement friction that extends enterprise sales cycles; satellite operators carrying expensive compute hardware face launch, radiation, thermal, and cyber risks that insurers cannot yet price comfortably. SU007, SU012
CU031 The chicken-and-egg demand problem for Cowboy Space is structural: customers are unlikely to commit to a novel orbital compute service before the technology is demonstrated at scale, yet demonstrating technology at scale requires capital that is difficult to raise without customer commitments. SU004, SU006
CU032 No retention, NRR, GRR, or customer satisfaction metrics exist for Cowboy Space because the company has no commercial customers; all retention analysis is structurally unavailable until commercial operations begin. SU001, SU002
CU033 Once deployed, orbital data center infrastructure would create substantial switching costs for customers whose workloads are optimized for orbital processing characteristics — continuous solar power, radiative cooling, and proximity to space-borne sensors — requiring significant re-engineering to migrate to terrestrial alternatives. SU007
CU034 Cowboy Space's lack of named customers creates extreme revenue concentration risk; any first commercial customer would represent 100% of revenue at launch, giving that customer disproportionate pricing and contract negotiation leverage. SU001, SU002
CU035 If SAIC serves as an anchor defense customer, multi-year DoD contract structures would provide revenue durability, but compliance requirements — ITAR authorization, FedRAMP moderate or high authorization, and CMMC certification — would add substantial lead time and recurring compliance cost. SU008, SU019
CU036 Earth observation satellite operators such as Planet Labs analogues have both the technical sophistication and mission architecture to be among the first commercially viable customers for on-orbit processing, given their documented need to compress high-volume imagery datasets before downlink. SU007, SU014
CU037 Cowboy Space's potential customer base is geographically global, as LEO orbital infrastructure is accessible to any customer with adequate downlink ground station access, removing the geographic constraints of terrestrial data center siting. SU007, SU001
CU038 Data sovereignty applications — representing sovereign nations, financial entities, and government agencies seeking off-Earth storage — constitute a distinct expansion segment with structurally different procurement dynamics from the AI compute segment. SU007
CU039 Cowboy Space's FCC application for a 20,000-satellite Stampede constellation implies an orbital compute capacity that would require a correspondingly large and diverse customer base to achieve economic viability; the scale of the application has no current customer underpinning. SU021, SU023, SU006
CU040 Cowboy Space has not publicly disclosed any enterprise sales leadership hires, business development directors, or customer success functions; the company's disclosed headcount and job postings are focused entirely on technical engineering roles. SU005, SU001
CU041 Kepler Communications operates compute-enabled satellites with optical relay capabilities deployed for in-orbit data processing; Kepler's operational model demonstrates the space-native customer segment and adjacent competitive positioning for Cowboy Space. SU007
CU042 Axiom Space's deployment of the AxDCU-1 data center unit on the ISS (fall 2025) and its first two dedicated orbital data center nodes (launched January 2026) established a proof point for space station compute that validates the customer segment Cowboy Space could later target. SU007
CU043 The absence of public pricing schedules, reference architecture documentation, or service-level agreement templates confirms Cowboy Space is in a pre-commercial stage with no active customer stewardship infrastructure. SU001, SU002, SU003
CU044 Any DoD customer acquisition for Cowboy Space would require ITAR compliance, likely FedRAMP moderate-to-high authorization, and potentially CMMC (Cybersecurity Maturity Model Certification) compliance, extending government procurement timelines by 12–36 months. SU008, SU019
CU045 The orbital data center sector lacks a track record of satellite-to-enterprise customer contracts; no ODC company has yet reported enterprise SLAs, recurring revenue, or customer expansions, creating a buyer education and trust-building challenge unique to this market. SU007, SU020
CU046 Starcloud (formerly Lumen Orbit), the most operationally advanced ODC startup, launched its first GPU to orbit in November 2025, demonstrating actual orbital AI compute for the first time and giving it a customer proof advantage over Cowboy Space. SU007, SU015
CU047 Cowboy Space CEO Bhatt acknowledged that commercial launch capacity 'three, four years out is still very, very scarce,' signaling that even the proposed rocket supply advantage may not materialize fast enough to capture first-mover customer demand before terrestrial capacity constraints ease. SU004
CR001 Cowboy publicly frames itself as an integrated rockets-plus-satellites company for orbital AI compute and optical data transmission from low Earth orbit. SR033, SR024
CR002 The FCC public notice says Cowboy seeks authority to deploy and operate 20,000 NGSO satellites in dawn-dusk sun-synchronous orbits between 700 and 1000 km. SR013, SR016
CR003 SpaceNews and SatNews report that Cowboy told the FCC the satellite design remains unfinished and will require a license modification before service begins. SR016, SR018
CR004 Public reporting describes each Stampede satellite as roughly 20,000-25,000 kilograms, around one megawatt of power, and just under 800 onboard GPUs. SR018, SR020, SR021
CR005 Cowboy describes the rocket upper stage itself as the data center and says the stage structure would serve as a radiator. SR021, SR024
CR006 Cowboy says the compute payload will use NVIDIA Space-1 Vera Rubin modules, and NVIDIA separately names Cowboy as a space-computing partner. SR024, SR021, SR063
CR007 Gartner, as cited by The Register, argues orbital data centers face immense cooling challenges in vacuum and extraordinary temperature swings. SR041
CR008 NOAA says the radiation belts are harsh regions of energetic electrons and protons that can damage satellite electronics. SR062
CR009 NASA says trapped high-energy radiation in the Van Allen belts is a major hazard for spacecraft electronics, implying additional hardening burden for satellites operating in that environment. SR061, SR062, SR013
CR010 The public corpus does not yet provide integrated thermal-vacuum, radiation-qualification, or full-system reliability data for Cowboy's orbital data-center architecture. SR021, SR024, SR033, SR063
CR011 Cowboy aims to deploy its first “Galactic Brain” orbital compute node early next year, ahead of the proprietary rocket mission. SR021, SR016
CR012 Bhatt told TechCrunch he expects Cowboy's first proprietary launch before the end of 2028. SR020, SR021
CR013 Cowboy is standing up its own rocket program rather than relying solely on external launch providers for the long term. SR020, SR021
CR014 Public reporting says the planned hybrid launch vehicle would be larger or slightly more powerful than Falcon 9 but smaller than Starship. SR020, SR021
CR015 The FAA Office of Commercial Space Transportation authorizes launch and reentry operations, so Cowboy needs FAA approvals in addition to FCC constellation authority. SR039
CR016 TechCrunch reports Cowboy plans to build its own rocket engine and has hired experienced aerospace personnel, confirming that propulsion development is in scope. SR020, SR021
CR017 The public source set does not disclose a factory, engine-test campaign, or dedicated test facility for Cowboy's proprietary rocket program. SR020, SR021, SR024, SR033
CR018 A 20,000-satellite constellation built from 20,000-25,000 kilogram units implies a manufacturing and launch-cadence burden far beyond Cowboy's currently demonstrated scale. SR013, SR018, SR020, SR021
CR019 Cowboy's constellation filing is pending rather than granted, and the FCC public notice lists multiple requested waivers including 25.155(b), 25.157(c), 25.164(b), 25.165, 2.106(a), 25.202(g)(1), and 25.114(a)(1). SR013, SR016
CR020 Press coverage says Cowboy is seeking relief from typical deployment bond and milestone rules because Stampede would rely primarily on optical communications rather than congested RF links. SR016, SR017, SR018
CR021 The FCC's 2022 order adopted a five-year post-mission disposal benchmark for satellites in low Earth orbit. SR035, SR036
CR022 The FCC said leaving satellites in LEO to deorbit over decades is no longer sustainable, which raises the bar for active disposal planning at Cowboy's proposed altitudes. SR036, SR035
CR023 Launch and reentry safety approvals remain a separate regulatory dependency even if FCC spectrum issues progress smoothly. SR039, SR013
CR024 BIS maintains the EAR framework and DDTC says ITAR implements U.S. controls over defense-article and defense-service exports, creating export-control exposure for a rockets-and-satellites company. SR028, SR040
CR025 The public corpus does not disclose Cowboy's export-screening procedures, foreign-customer policy, or broader export-compliance program. SR024, SR033, SR034, SR028
CR026 NewSpaceEconomy and SpaceNews place Cowboy in a competitive orbital-data-center field that also includes very large filings and plans from SpaceX, Blue Origin, Starcloud, and others. SR016, SR019
CR027 Because the waiver request is tied to anti-warehousing milestone rules, adverse FCC treatment could materially slow deployment or raise capital requirements. SR017, SR019, SR013
CR028 ESA's 2025 space environment report says not enough satellites leave heavily congested orbits at end of life, creating collision risk. SR056
CR029 ESA says 2024 saw several major fragmentation events, plus many smaller ones, adding thousands of new debris objects. SR056, SR055
CR030 NASA's Orbital Debris Program Office maintains ORDEM and LEGEND modeling tools, underscoring that debris exposure is a live engineering problem rather than a rhetorical one. SR037, SR057
CR031 Adding 20,000 additional 20-25 ton maneuvering satellites would materially enlarge Cowboy's conjunction-management burden relative to today's already crowded environment. SR013, SR018, SR056
CR032 The combination of a five-year disposal rule and a 20,000-satellite fleet implies a persistent active-deorbit queue at end of life. SR021, SR036, SR013
CR033 The AAS says the explosion in low-Earth-orbit satellites creates potential for substantial adverse impacts to ground- and space-based astronomy and to orbital sustainability. SR058
CR034 SATCON1 exists specifically to document optical-astronomy impacts from satellite constellations and recommend mitigations, confirming the externality is well established. SR059
CR035 The JASON/NSF report on large constellations concludes that even a fraction of proposed deployments can create meaningful impacts for astronomy and observatories. SR060, SR058
CR036 Independent criticism emphasizes that orbital data centers must survive launch stress, space weather, debris exposure, and limited serviceability once on orbit. SR043, SR042
CR037 Cowboy announced a $275 million Series B at a $2 billion valuation, and SatNews says disclosed equity to date is roughly $325 million. SR024, SR018, SR020
CR038 Cowboy says initial commercial services will follow only after performance, reliability, and economics are validated at scale. SR021
CR039 Construct Capital frames Cowboy's market thesis around terrestrial grid constraints and years-long waits for data-center grid access. SR031
CR040 NewSpaceEconomy says early orbital-data-center customers are more likely to be Earth-observation operators, defense and security agencies, space-station users, lunar infrastructure firms, and AI developers with unusual constraints. SR019
CR041 NewSpaceEconomy says data that can tolerate space-to-ground latency is the more plausible early fit, implying many terrestrial cloud workloads still favor ground infrastructure. SR019
CR042 NVIDIA's space-computing page emphasizes minimal downlink latency, lower transmission costs, and edge ground-station processing rather than a direct substitute for terrestrial warehouse-scale clusters. SR063
CR043 Gartner's position, as reported by The Register, is that orbital data centers will not serve terrestrial needs for decades, if ever, because launch economics and cooling challenges remain prohibitive. SR041, SR042
CR044 Breaking Defense reports that Pentagon and intelligence-community buyers are largely waiting to see technical capability and business case proven before showing serious demand. SR044
CR045 As of June 2026, the public corpus shows no named customers, disclosed revenue, or long-term commercial contracts for Cowboy Space. SR033, SR024, SR020, SR021
CR046 Cowboy's disclosed compute architecture is heavily concentrated on NVIDIA Space-1 Vera Rubin modules and adjacent NVIDIA hardware. SR024, SR021, SR063
CR047 Cowboy's public narrative is highly concentrated on founder and CEO Baiju Bhatt as the company's central strategist and spokesperson. SR024, SR034, SR033
CR048 The orbital-data-center category is still dominated by filings, tests, and early demos rather than mature commercial operations, which raises timing risk if terrestrial alternatives improve first. SR019, SR041, SR044
CR049 The right diligence gates for Cowboy are binary: prove the orbital demo, prove heat and radiation performance, secure the regulatory path, and secure customer commitments. SR021, SR035, SR039, SR044
CR050 Failure of any one critical dependency—demo performance, proprietary rocket, regulatory approvals, or NVIDIA-backed compute supply—would directly pressure both timeline confidence and valuation. SR021, SR039, SR063, SR044
CR051 Public sources describe optical downlink and transmission ambitions, but they do not disclose Cowboy's planned ground-station count, throughput budget, or customer-side network architecture. SR018, SR033, SR063
CV001 Cowboy Space Corporation closed a $275 million Series B financing at a $2 billion post-money valuation in May 2026, confirmed by both the official BusinessWire press release and multiple independent news sources. SV001, SV016, SV024
CV002 The Series B was led by Index Ventures with new investors IVP, Blossom Capital, and SAIC, alongside existing investors Breakthrough Energy Ventures, Construct Capital, Andreessen Horowitz, NEA, Interlagos, and CEO Baiju Bhatt. SV001, SV014
CV003 Cowboy Space had raised approximately $365 million in total across all rounds through the May 2026 Series B close, confirmed by CEO Baiju Bhatt and corroborated by news coverage. SV014, SV015, SV017
CV004 The implied pre-money valuation of Cowboy Space at the Series B was approximately $1.725 billion, derived from the $2 billion post-money valuation minus the $275 million raise, implying approximately 13.8% dilution from the Series B tranche. SV001
CV005 Fenwick & West LLP confirmed it served as legal counsel to Cowboy Space for the Series B, with corporate partners Morgan Sawchuk and Ran Ben Tzur leading the representation. SV023
CV006 Cowboy Space is pre-revenue with no disclosed ARR, backlog, customer contracts, or letters of intent as of the Series B close in May 2026; no pricing schedule or financial projection has been published. SV001, SV015, SV014
CV007 Standard DCF and revenue-multiple valuation frameworks are inapplicable to Cowboy Space because the company has zero revenue and no publicly disclosed financial model; scenario-based option valuation is the only defensible analytical framework. SV001, SV011
CV008 Public AI neocloud and data center companies trade at a median 11.8x NTM forward revenue multiple as of June 2026, representing the relevant public-market anchor for AI compute infrastructure valuations. SV003, SV005
CV009 CoreWeave (CRWV) had a market capitalization of approximately $64 billion and trailing twelve-month revenue of $6.2 billion as of June 2026, following its March 2025 IPO at approximately $23–26 billion. SV003, SV005, SV029
CV010 Crusoe AI raised $1.37 billion in a Series E at a $10 billion-plus post-money valuation in October 2025, making it one of the most highly valued private AI infrastructure companies before any public listing. SV006, SV019
CV011 Starcloud (formerly Lumen Orbit) raised a $170 million Series A at a $1.1 billion post-money valuation in March 2026, reaching unicorn status 17 months after Y Combinator demo day, reportedly the fastest YC unicorn. SV007, SV008
CV012 Starcloud was reportedly pursuing at least $200 million more in a deal that would double its valuation to approximately $2.2 billion as of May 2026, per sources cited by industry analysts. SV010, SV007
CV013 SpaceX's June 2026 IPO targeted a $1.75 trillion valuation — reportedly the largest IPO in history — following secondary market trades that had implied valuations of $1.25–1.8 trillion in early 2026. SV012, SV004
CV014 Seraphim Space's Q1 2026 report documents $8.0 billion in global space investment in a single quarter — doubling the Q4 2025 record — with orbital data centers and orbital compute cited as a key driver attracting serious backing at scale for the first time. SV004, SV009
CV015 The Q1 2026 global space investment surge occurred within a broader risk-on VC environment driven by large AI financings, but space capital remained focused on defense, strategic infrastructure, and sovereign capability themes according to Seraphim's analysis. SV004, SV009
CV016 Announced equity financing in the orbital data center category exceeded $445 million in the two months through May 2026 (Cowboy $275M plus Starcloud's $170M first round), with Starcloud's potential second round adding another $200M. SV010, SV007, SV001
CV017 Cowboy Space's $2 billion post-money valuation is the highest among orbital data center companies with publicly disclosed round prices as of May 2026, placing it at the top of the nascent ODC category. SV010, SV007, SV008
CV018 The Series B implied approximately 13.8% dilution from the single tranche alone; cumulative dilution from seed, Series A, and Series B likely represents 35–50% of founder equity, though exact figures require the cap table. SV001, SV003
CV019 Top-quartile private AI infrastructure companies command 20–50x ARR multiples in private markets as of Q1 2026, but this premium is rapidly narrowing for companies without repeatable monetization and durable demand according to Finro's analysis. SV011, SV003
CV020 Cowboy Space has no ARR against which to apply revenue multiples, making all valuation comparisons to CoreWeave, Crusoe, or other revenue-generating AI infrastructure companies structurally imprecise. SV001, SV011
CV021 The $2 billion valuation for Cowboy Space effectively prices a call option on ODC market existence and Cowboy execution, anchored on investor conviction in the AI compute demand structural thesis and the vertical integration architecture premium. SV001, SV011, SV010
CV022 Gartner, as cited by The Register in February 2026, argued that orbital data centers are a 'pie in the sky' idea, pointing to immense cooling challenges in vacuum and extraordinary temperature swings as barriers to economic viability. SV025, SV027
CV023 The Register reported in April 2026 that an orbital data center startup publicly admitted launch economics 'don't fly,' underscoring that the sector's economic case depends on Starship-class launch cost reductions not yet realized. SV026, SV007
CV024 Comparable small-to-medium lift rocket programs — Relativity Space, ABL Space Systems, and Stoke Space — consumed between $100 million and $650 million in capital before reaching first operational milestones, and several failed before achieving them. SV015, SV014
CV025 A full 20,000-satellite Stampede constellation and operational proprietary rocket program would require orders of magnitude more capital than Cowboy's current $365 million balance sheet, implying 40–60%+ cumulative dilution before first revenue under baseline assumptions. SV001, SV015, SV024
CV026 SpaceX filed with the FCC in January 2026 for authority to deploy up to 1 million orbital data center satellites, constituting an existential competitive risk to independent ODC operators if SpaceX deploys at scale and offers commercial access. SV015, SV010
CV027 In a bull scenario where 2026–2028 milestones land on schedule and first commercial contracts are signed, an implied valuation of $8–15 billion by 2029 is plausible, anchored on CoreWeave's revenue confirmation trajectory. SV003, SV005, SV010
CV028 In a base scenario with 12–24 month milestone delays and fragmented competition, the $2 billion Series B mark is likely to hold flat to modest appreciation ($2–4 billion range) through the technology validation phase. SV011, SV010
CV029 In a bear scenario involving thermal-qualification failure, rocket program cost escalation, or SpaceX ODC deployment at commercial scale before Cowboy reaches revenue, a down-round to a $0.3–1 billion implied mark is plausible. SV025, SV026, SV010
CV030 SAIC's strategic investment in the Series B signals defense and government customer interest in orbital compute, providing a non-commercial revenue floor in the event that commercial traction is slow. SV028, SV001
CV031 Cowboy Space's institutional investor syndicate — Index Ventures, Andreessen Horowitz, Breakthrough Energy Ventures, IVP, NEA — represents a top-tier signal of investor conviction, though this is not a substitute for operational validation. SV001, SV018, SV023
CV032 Finro's Q1 2026 AI valuation analysis documents sharp dispersion in private market multiples: companies with repeatable monetization receive premiums, while 'growth now, business model later' narratives face material repricing — a direct risk to Cowboy's current valuation if milestones slip. SV011, SV003
CV033 BlackNight Space Labs analysis places Cowboy Space at the high end of ODC category private valuations alongside Starcloud's pursued $2.2 billion, and characterizes both as speculative premiums with potential for dramatic correction if milestones are missed. SV010
CV034 Private AI infrastructure companies commanded 20–50x ARR multiples at peak in 2025, but Finro's Q1 2026 analysis shows the market narrowing the premium cohort to companies with durable, contracted revenue rather than narrative-driven growth. SV011
CV035 Lonestar Data Holdings, a lunar data center startup, had raised approximately $12.4 million in total by late 2025 — orders of magnitude less than Cowboy's $365 million — illustrating the wide funding dispersion within space-compute startups at different scale ambitions. SV027, SV019
CV036 At the current $1.725 billion pre-money valuation, investors are effectively banking on a 5–10x or greater return at exit, with no public evidence of the revenue or milestone trajectory required to achieve that return on a realistic timeline. SV003, SV011, SV010
CV037 Cowboy Space's cap table, liquidation preferences, anti-dilution provisions, and fully diluted share count are not publicly disclosed; these are the minimum inputs required for any precise return-modeling or exit analysis.
CV038 No near-term exit pathway is currently visible for Cowboy Space; the earliest plausible IPO or strategic acquisition would follow first commercial revenue milestones, projected no earlier than 2029 under a base-case assumption. SV010, SV011, SV001
CV039 Key thesis-break triggers for Cowboy Space include: failure of thermal-vacuum qualification of the 1 MW ODC unit, proprietary rocket delay beyond 2030, FCC denial of Stampede constellation, down-round at Series C, and SpaceX ODC commercial deployment at scale. SV025, SV026, SV010
CV040 The addressable market for orbital data centers is estimated at $2–10 billion or more by 2030 by MarketsandMarkets and industry analysts, though estimates diverge widely and no consensus forecast exists for a category this early. SV030, SV019
CV041 AI compute infrastructure total spend is projected at $527 billion in 2026, versus only approximately $100 billion in enterprise AI revenue — a potential infrastructure overbuild signal that, if it materializes, would compress multiples for all pre-revenue AI infrastructure companies. SV011, SV003
CV042 Cost-competitiveness of orbital data centers with terrestrial alternatives is projected to require Starship-class launch costs at industrial cadence, a milestone Starcloud's CEO Johnston estimates won't materialize until 2028–2030 at the earliest. SV007, SV026
CV043 Cowboy Space's rocket-upper-stage-as-data-center architecture differentiates it from Starcloud's conventional satellite-on-launch-vehicle design; the vertical integration creates a distinct risk profile (rocket AND data center must succeed simultaneously) versus decoupled approaches. SV001, SV010, SV007
CV044 Cowboy Space claims its integrated rocket-and-data-center design eliminates dependence on external launch providers for the long term and achieves cost parity with terrestrial data centers once the rocket is at operational cadence, per the company's own Series B framing. SV001, SV015
CV045 External technical analysis indicates that the two-to-three-year window before Cowboy's proprietary rocket is operational still requires third-party launch providers, and launch economics at current pricing make early ODC deployments uneconomic by terrestrial comparison. SV026, SV007, SV023
来源
编号出版方标题引文
SO001 Cowboy Space Corporation Cowboy Space Corporation | Orbital Data Centers for AI Cowboy Space Corp. is building a power grid in outer space for artificial intelligence. It is an integrated system of satellites and rockets to haul high-performance compute and optical data transmission down from Low Earth Orbit.
SO002 Cowboy Space Corporation via BusinessWire Cowboy Space Corporation Raises $275M Series B For Vertically-Integrated Orbital Data Centers and Rockets Cowboy Space Corporation, the orbital infrastructure company founded by Robinhood co-founder Baiju Bhatt, today announced $275M in Series B financing at a $2B valuation led by Index Ventures.
SO003 Payload Space Aetherflux Rebrands, Pivots Business—and Raises $275M Allow me to reintroduce myself: To better align with its new business, Aetherflux is rebranding to Cowboy Space Corporation.
SO004 TechCrunch There aren't enough rockets for space data centers — Cowboy Space raised $275M to build them while bringing the rocket in-house is logical, it's also nuts. Only a handful of private companies in the West, mainly SpaceX, Rocket Lab, and Arianespace, are consistently launching commercial rockets.
SO005 Satellite Today Aetherflux Rebrands as Cowboy Space, Expanding Plans to In-Orbit Compute and Launch
SO006 Satellite Today Cowboy Space Files for 20,000-Satellite Constellation for Data Center Services Cowboy Space followed up its $275 million Series B round earlier this week by filing for a 20,000 satellite constellation to provide data center services from space.
SO007 SpaceNews Cowboy raises $275 million to build rockets with orbital data center upper stages Bhatt said it is too early to provide finer details on its ambitious plans…The venture has now raised around $365 million to date.
SO008 GeekWire Cowboy Space raises $275M as it seeks 40-60 employees for new satellite and rocket hub in Seattle A rep for the company told GeekWire Wednesday that they anticipate 40-60 employees in Seattle initially, and there are currently 18 positions advertised.
SO009 TechCrunch Billionaire Robinhood co-founder launches Aetherflux, a space-based solar power startup The startup was founded by Bhatt late last year and became his full-time focus after he left his leadership role at financial trading platform Robinhood.
SO010 Data Center Dynamics Satellite firm Aetherflux rebrands to Cowboy Space Corporation, raises $275m Series B the company, recently rebranded from Aetherflux, now hopes to pursue 'vertically integrated orbital data centers and rockets.' The company, founded in March 2024 and emerged from stealth the following October.
SO011 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers, Forcing the ODC Thesis Into View Dissipating one megawatt of thermal load in vacuum is not a trivial engineering exercise, and a 20,000-satellite shell at 700–1,000 km has a debris liability that the optical-only architecture mitigates but does not eliminate.
SO012 New Space Economy Cowboy Space Seeks FCC Approval for Stampede Orbital Data Centers FCC File No. SAT-LOA-20260323-00135 identifies Cowboy Space Corp.'s request for authority to launch and operate Stampede, a proposed non-geostationary orbit (NGSO) satellite system designed to provide data center services from space.
SO013 SiliconANGLE Cowboy Space raises $275M to build orbital AI data centers
SO014 Satellite Today Robinhood Co-Founder Debuts Space Solar Startup Aetherflux
SO015 Federal Communications Commission Report No. SAT-02007: Applications Accepted for Filing (FCC Public Notice, June 12, 2026) Cowboy Space requests authority to deploy and operate a constellation of 20,000 NGSO satellites to operate as a system of in-orbit data centers in dawn-dusk sun-synchronous orbits at altitudes between 700 and 1000 km.
SO016 Apex Space Apex - Productized Satellite Platforms
SO017 Space Insider California Space-Based Solar Startup Aetherflux Raises $50 Million, Focuses on Defense Work Aetherflux, founded and helmed by Robinhood co-founder Baiju Bhatt with its mission to deliver energy to Earth from orbit, closed the funding round with backing from Index Ventures, Interlagos, Bill Gates-founded Breakthrough Energy Ventures, Andreessen Horowitz, and NEA.
SO018 Aviation Week Aetherflux Raises $50m, Eyes 2026 Laser Power Beaming Demo
SO019 Construct Capital The Energy Bottleneck: Why Cowboy Space Corporation Is Building a New Grid We've been able to support this incredible team since galvanizing and investing in their Series A last year. The rocket's upper stage becomes the data center itself, powered by continuous solar energy in Low Earth Orbit.
SO020 Fenwick & West LLP Fenwick Represents Cowboy Space Corporation in $275M Series B Financing Fenwick represented Cowboy Space Corporation, an orbital infrastructure company, in its $275 million Series B financing at a $2B valuation.
SO021 Washington Technology Cowboy Space, Darkhive detail their Series B rounds Science Applications International Corp. and the venture capital arm of RTX are the GovCon investor names to take note of here.
SO022 Evertiq Cowboy Space raises $275M to advance orbital data centers
SO023 Grey Journal Cowboy Space Raises $275M for Orbital Data Centers Bhatt left Robinhood Markets in 2024 to start what was then called Aetherflux…his roughly 6% stake in Robinhood remains the bulk of an estimated $6 billion fortune.
SO024 Robinhood Markets About Us | Robinhood Our story began at Stanford, where co-founders Baiju and Vlad were roommates and classmates.
SO025 FCC.report (third-party FCC database aggregator) FCC IBFS Filing SATLOA2026032300135
SM001 Goldman Sachs AI is poised to drive 160% increase in data center power demand "Goldman Sachs Research estimates that data center power demand will grow 160% by 2030. At present, data centers worldwide consume 1-2% of overall power, but this percentage will likely rise to 3-4% by the end of the decade."
SM002 New Space Economy Orbital Data Center Companies Building Space-Based Compute Infrastructure "Economics remain unsettled. The space case depends on lower launch costs, high processor duty cycles, efficient heat rejection, long satellite life, strong customer willingness to pay, and limited service interruption. Terrestrial data centers are not standing still."
SM003 International Energy Agency Electricity 2024 — Analysis "This year's report features in-depth analysis on the drivers of recent declines in electricity demand in Europe; the data centre sector's impact on electricity consumption."
SM004 Data Center Dynamics Cowboy Space submits FCC application for a 20,000 satellite orbital data center constellation "The satellite design, Cowboy told the commission with admirable candor, is unfinished and will require a license modification before service begins."
SM005 Cowboy Space Corporation via BusinessWire Cowboy Space Corporation Raises $275M Series B For Vertically-Integrated Orbital Data Centers and Rockets "Cowboy is also collaborating with NVIDIA to deploy NVIDIA Space-1 Vera Rubin Modules, bringing state-of-the-art AI infrastructure to the Low Earth Orbit environment."
SM006 Satellite Today Cowboy Space Files for 20,000-Satellite Constellation for Data Center Services "Both SpaceX and Starcloud have also submitted filings for massive constellations to function as data centers with SpaceX targeting up to 1 million satellites, and Starcloud up to 88,000."
SM007 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers, Forcing the ODC Thesis Into View "Dissipating one megawatt of thermal load in vacuum is not a trivial engineering exercise, and a 20,000-satellite shell at 700–1,000 km has a debris liability that the optical-only architecture mitigates but does not eliminate. Cowboy's own FCC filing concedes the satellite design isn't done."
SM008 Starcloud Data Centers in Space | Starcloud – The Future of AI "As launch costs fall, data centers in space will leverage continuous solar energy and radiative cooling, rapidly scaling to gigawatts while avoiding permitting constraints on Earth."
SM009 Lonestar Space StarVault: The Future of Data Storage Is In Space "StarVault, is the world's first commercially operational space-based sovereign data storage platform."
SM010 Sophia Space Sophia Space — In-Orbit Compute "The rapid increase in satellite deployments, private space stations, and commercial space activity is driving unprecedented demand for in-orbit compute."
SM011 U.S. Energy Information Administration Electricity Monthly Update — March 2026 Highlights "Net electricity generation in the United States increased 4.5% compared to March 2025."
SM012 Space Capital Space Investment Quarterly (Space IQ)
SM013 Construct Capital The Energy Bottleneck: Why Cowboy Space Corporation Is Building a New Grid
SM014 TechCrunch There Aren't Enough Rockets for Space Data Centers. Cowboy Space Raised $275 Million to Build Them
SM015 SpaceNews Cowboy raises $275 million to build rockets with orbital data center upper stages
SM016 New Space Economy Cowboy Space Seeks FCC Approval for Stampede Orbital Data Centers
SM017 Washington Technology Cowboy Space, Darkhive Detail Their Series B Rounds
SM018 SiliconAngle Cowboy Space raises $275M to build orbital AI data centers
SM019 Space Insider California Space-Based Solar Startup Aetherflux Raises $50 Million, Focuses on Defense Work
SM020 Payload Space Aetherflux Rebrands, Pivots Business, and Raises $275M
SM021 GeekWire Cowboy Space raises $275M as it seeks 40–60 employees for new satellite and rocket hub in Seattle
SM022 Evertiq Cowboy Space Raises $275M to Advance Orbital Data Centers
SM023 Federal Communications Commission Cowboy Space Stampede Data Center System — FCC Application (DOC-422332A1) "By some estimates, the insatiable demand for electricity to support AI data centers will drive a 160 percent increase in global data center power demand by 2030 … Simply put: The traditional pace of interconnection with utilities and the terrestrial power grid is not ready to meet the infrastructure demands of AI."
SM024 CBInsights Cowboy Space — Products, Competitors, Financials, Employees, Headquarters Locations
SM025 Aviation Week Aetherflux Raises $50M, Eyes 2026 Laser Power-Beaming Demo
SP001 CoreWeave CoreWeave AI Cloud Platform Homepage The world's #1 AI cloud platform, purpose-built for what's next. CoreWeave Cloud is an AI-native platform purpose-built for AI.
SP002 Crusoe Energy Systems Crusoe Managed AI Cloud Homepage Crusoe Cloud provides exceptional reliability for AI workloads with 99.98% uptime and resilient infrastructure.
SP003 Voltage Park Voltage Park AI Cloud Homepage On-Demand: Self-serve and spin up nodes in 15 minutes. No long-term contracts. Starting at $1.99/hr.
SP004 OpenAI Announcing the Stargate Project The Stargate Project is a new company which intends to invest $500 billion over the next four years building new AI infrastructure for OpenAI in the United States.
SP005 Amazon Web Services AWS Ground Station — Introduction AWS Ground Station is a fully managed service that lets you control satellite communications, process data, and scale your operations without having to worry about building or managing your own ground station infrastructure.
SP006 Cloud Constellation Corporation SpaceBelt — The Information Ultra-Highway Begins Here SSTL collaborates with SpaceBelt by introducing and marketing SpaceBelt's technology and services to government and business community.
SP007 Spacebilt Spacebilt In-Space Server Platform Homepage We do not want to compete within the category. We want to define it.
SP008 Axiom Space Axiom Station — Commercial Space Station Overview
SP009 Blue Origin Project Sunrise — Blue Origin Orbital Data Center Network
SP010 New Space Economy Orbital Data Center Companies Building Space-Based Compute Infrastructure A useful dividing line separates three company categories. The first category includes firms building dedicated data center satellites or data center constellations. Starcloud, SpaceX, Blue Origin, Cowboy Space, and Sophia Space fit this group in different ways.
SP011 Starcloud Starcloud Space Data Center Platform Homepage As launch costs fall, data centers in space will leverage continuous solar energy and radiative cooling, rapidly scaling to gigawatts while avoiding permitting constraints on Earth.
SP012 Lonestar Data Holdings Lonestar StarVault Space-Based Sovereign Data Storage StarVault, is the world's first commercially operational space-based sovereign data storage platform. Lonestar is now taking capacity reservations for the first StarVault launching October 2026.
SP013 Sophia Space Sophia Space Orbital Compute and Data Centers Homepage The rapid increase in satellite deployments, private space stations, and commercial space activity is driving unprecedented demand for in-orbit compute.
SP014 Data Center Dynamics Cowboy Space Submits FCC Application for a 20,000-Satellite Orbital Data Center Constellation The design of the satellites remains unfinished, as do the company's rockets, which themselves must be approved by the Federal Aviation Administration's Office of Commercial Space Transportation.
SP015 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers — Forcing the ODC Thesis Into View Cowboy is not betting that orbital compute is a good idea. That question is already settled. Cowboy is betting it can solve all four ODC bottlenecks simultaneously (launch supply, thermal, data gravity, and unit economics) by owning the rocket.
SP016 TechCrunch There Aren't Enough Rockets for Space Data Centers—Cowboy Space Raised $275 Million to Build Them Of course, while bringing the rocket in-house is logical, it's also nuts. Only a handful of private companies in the West, mainly SpaceX, Rocket Lab, and Arianespace, are consistently launching commercial rockets.
SP017 SpaceNews Cowboy Files Plans for Up to 20,000 Orbital Data Centers While massive by historical standards, Stampede is smaller than SpaceX's plan for up to one million satellites, Starcloud's 88,000-satellite system and Blue Origin's 51,600-satellite Project Sunrise.
SP018 New Space Economy Cowboy Space Seeks FCC Approval for Stampede Orbital Data Centers
SP019 Payload Aetherflux Rebrands, Pivots Business, and Raises $275M Cowboy Space has designed the rocket's upper stage and data center to be a single vehicle, unlike a traditional rocket that delivers a separate payload to space.
SP020 FCC.report FCC IBFS Filing SAT-LOA-20260323-00135 — Cowboy Space Stampede Application
SP021 SiliconAngle Cowboy Space Raises $275M to Build Orbital AI Data Centers
SP022 Business Wire Cowboy Space Corporation Raises $275M Series B for Vertically Integrated Orbital Data Centers and Rockets
SP023 GeekWire Cowboy Space Raises $275M as It Seeks 40–60 Employees for New Satellite and Rocket Hub in Seattle
SP024 Via Satellite / Satellite Today Cowboy Space Files for 20,000-Satellite Constellation for Data Center Services
SP025 Space Capital Space Capital Quarterly — Space Data and Infrastructure Market Intelligence
SP026 Construct Capital The Energy Bottleneck: Why Cowboy Space Corporation Is Building a New Grid Cowboy Space addresses the energy bottleneck throttling American innovation in AI and data center deployments by pioneering space-based solar power and in-orbit compute.
SI001 CNBC Cowboy Space raises $275 million to build orbital data centers Cowboy Space raised $275 million in a Series B round at a $2 billion post-money valuation, led by Index Ventures.
SI002 Breakthrough Energy Ventures Aetherflux — Breakthrough Energy Ventures Portfolio
SI003 Ars Technica Cowboy Space raises $275M to build vertically integrated orbital data centers
SI004 NVIDIA Newsroom NVIDIA Space-1 Vera Rubin Modules for Orbital Data Centers NVIDIA Space-1 Vera Rubin modules are engineered for LEO thermal and radiation environments, targeting sub-1 MW power budgets per satellite-class deployment.
SI005 Financial Times Cowboy Space raises $275M Series B for orbital data centers
SI006 IEEE Spectrum Cowboy Space and the Orbital Data Center Race
SI007 Andreessen Horowitz (a16z) Aetherflux — a16z Announcement a16z participated in Aetherflux's Series A to back the company's mission to bring solar power and compute to orbit.
SI008 Fortune Business Insights Data Center Market Size, Share & Trends — Global Outlook
SI009 MIT Technology Review Orbital data centers promise to solve the AI power crisis — but can they deliver?
SI010 JLL Global Data Center Outlook — Capital and Demand Trends Global data center construction will require $250B+ in cumulative capex through 2030 to meet AI-driven demand.
SI011 Ars Technica Will Space Data Centers Actually Save Us From the AI Power Crisis? The fundamental physics of radiative cooling in vacuum imposes hard limits on how much compute heat can be rejected per square meter; at 1 MW per satellite, the required radiator area is roughly 200 square meters — an engineering challenge that has never been demonstrated at production scale.
SI012 PitchBook Space Data Centers — Startups, Funding, and the 2026 Investor Surge
SI013 Ars Technica Orbital Data Centers — The Physics and Economics of Compute in Space The economics of orbital compute depend on a pricing premium that terrestrial hyperscalers may not be willing to pay; without a captive customer segment (such as defense) that values orbital attributes intrinsically, the business case risks being undermined by ongoing terrestrial cost declines.
SI014 TechFundingNews Cowboy Space $275M Series B — Robinhood Founder Launches Orbital AI Data Centers
SI015 FCC IBFS (via fcc.report) Cowboy Space Corporation — SAT-PDR-20260514-00022 Partial Designation Filing
SI016 NVIDIA Newsroom NVIDIA Space-1 Vera Rubin Module — LEO AI Data Center Architecture
SI017 Ars Technica Space-Based Data Centers — Could They Actually Solve the AI Power Crisis?
SI018 MarketsAndMarkets Orbital Data Center Market — Size, Share, and Forecast to 2031
SI019 Axios Cowboy Space raises $275 million for orbital AI data centers
SI020 IEEE Spectrum Space-Based Data Centers — The Engineering and Business Case
SI021 Business Wire Cowboy Space Corporation Raises $275M Series B for Vertically Integrated Orbital Data Centers and Rockets Cowboy Space Corporation announced the closing of a $275 million Series B financing. Proceeds will be used to advance rocket engine R&D, compute integration, and regulatory compliance for the Stampede orbital data center constellation.
SI022 Fenwick and West LLP Fenwick Represents Cowboy Space Corporation in $275M Series B Financing Fenwick represented Cowboy Space Corporation in its $275 million Series B financing. Corporate partners Morgan Sawchuk and Ran Ben Tzur led the team.
SI023 SpaceNews Cowboy raises $275 million to build rockets with orbital data center upper stages
SI024 GeekWire Cowboy Space raises $275M as it seeks 40–60 employees for new satellite and rocket hub in Seattle Cowboy Space is seeking 40 to 60 employees for its new satellite and rocket hub in Seattle following the Series B close.
SI025 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers — Forcing the ODC Thesis Into View
SE001 Cowboy Space Corporation Cowboy Space Corporation — Orbital Data Centers for AI (Homepage) The rocket's upper stage is the satellite itself. It's a megawatt-class data center with active thermal management and integrated compute, designed as one unified vehicle from the ground up.
SE002 SpaceNews Cowboy raises $275 million to build rockets with orbital data-center upper stages each upper stage leverages the full mass and volume of the vehicle to package power generation, cooling, and compute together — including using the stage structure itself as a radiator
SE003 TechCrunch There Aren't Enough Rockets for Space Data Centers. Cowboy Space Raised $275 Million to Build Them. Cowboy Space is still working through key development needs, like facilities to test, manufacture, and launch its rockets.
SE004 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers — Forcing the ODC Thesis into View Dissipating one megawatt of thermal load in vacuum is not a trivial engineering exercise, and a 20,000-satellite shell at 700–1,000 km has a debris liability that the optical-only architecture mitigates but does not eliminate.
SE005 Via Satellite / Satellite Today Cowboy Space Files for 20,000 Satellite Constellation for Data Center Services Cowboy Space argues that those rules are in place to prevent spectrum warehousing, whereas Stampede will operate primarily in optical and not use inter-satellite links.
SE006 NewSpace Economy Cowboy Space Seeks FCC Approval for Stampede Orbital Data Centers Cowboy Space Corporation is bringing together NVIDIA Vera Rubin Space-1 Module, IGX Thor, and Jetson Orin to orbit to power AI-driven geospatial intelligence and autonomous space operations.
SE007 GeekWire Cowboy Space raises $275M as it seeks 40-60 employees for new satellite and rocket hub in Seattle The company is collaborating with NVIDIA to deploy its Space-1 Vera Rubin Modules in low Earth orbit, and plans to launch its first satellite later this year to demonstrate space-to-Earth power beaming.
SE008 SiliconAngle Cowboy Space raises $275M to build orbital AI data centers The 1-megawatt capacity of the company's data center module represents a fraction of the processing power offered by terrestrial AI environments.
SE009 NVIDIA Newsroom NVIDIA Announces Space-1 Vera Rubin Module for LEO AI Data Centers
SE010 Apex Space Apex Space — High-Rate Satellite Manufacturing Factory One has been meticulously designed from the ground up to revolutionize bus production and support rapid, serial satellite manufacturing.
SE011 Payload Space Aetherflux Rebrands, Pivots Business — and Raises $275M
SE012 Bloomberg Cowboy Space Raises $275 Million at $2 Billion Valuation
SE013 The Verge Cowboy Space raised $275 million to build orbital data centers
SE014 VentureBeat Cowboy Space raises $275M to build orbital AI data centers
SE015 NVIDIA NVIDIA Space-1 — LEO AI Data Center Modules
SE016 Electrek Cowboy Space raises $275M for orbital data centers powered by space solar
SE017 SAIC Investor Relations SAIC Invests in Cowboy Space Corporation
SE018 State of the Space Orbital Data Center Economics and Technical Feasibility
SE019 Space.com Cowboy Space raises $275 million for orbital data centers
SE020 Breaking Defense Cowboy Space raises $275 million for orbital data centers with military applications
SE021 The Wall Street Journal Aetherflux Raises $275 Million for Orbital Data Center Space Startup
SE022 FCC (via fcc.report) FCC IBFS Filing SAT-LOA-20260323-00135 — Cowboy Space Stampede Constellation
SE023 Business Insider Baiju Bhatt built Robinhood. Now he's betting on orbital data centers.
SE024 NVIDIA Blog NVIDIA Space-1 Vera Rubin LEO Modules — AI at the Edge of Space
SE025 The Next Platform Orbital Data Centers: Physics Says Maybe
SE026 SatNews The Physics Wall: Orbital Data Centers and the Cooling Problem
SE027 Cowboy Space Corporation (via PR Newswire) Cowboy Space Corporation Raises $275M Series B for Vertically Integrated Orbital Data Centers and Rockets
SE028 SpaceRef Cowboy Space Raises $275M Series B for Orbital Data Centers
SE029 Crunchbase News Cowboy Space (Aetherflux) Raises $275M Series B for Orbital Data Center Constellation
SE030 Hacker News Cowboy Space raises $275M for orbital data centers (Show HN / community discussion)
SE031 LinkedIn Cowboy Space Corporation — Company Page and Engineering Recruiting
SU001 BusinessWire Cowboy Space Corporation Raises $275M Series B For Vertically-Integrated Orbital Data Centers and Rockets Cowboy Space Corporation is pioneering vertically integrated infrastructure to power and run AI compute in orbit, opening a new frontier for high performance computing as terrestrial capacity strains under AI demand.
SU002 SpaceNews Cowboy raises $275 million to build rockets with orbital data center upper stages AI giant Anthropic underscored that demand last week when it expressed interest in using SpaceX's proposed orbital data centers as part of a broader agreement for access to the company's terrestrial compute capacity.
SU003 PR Newswire Cowboy Space Corporation Raises $275M Series B For Vertically-Integrated Orbital Data Centers and Rockets
SU004 TechCrunch There aren't enough rockets for space data centers — Cowboy Space raised $275M to build them I see the demand for AI getting more and more acute, and I see the options on Earth getting more and more limited.
SU005 GeekWire Cowboy Space raises $275M as it seeks 40-60 employees for new satellite and rocket hub in Seattle
SU006 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers, Forcing the ODC Thesis Into View The launch-cadence argument is the strong one. The current commercial launch manifest cannot accommodate a 20,000-satellite, 20-tonne-per-bird constellation on anyone else's vehicle, and the grid-interconnect queues that AI hyperscalers are stuck in on the ground really are years deep.
SU007 NewSpaceEconomy Orbital Data Center Companies Building Space-Based Compute Infrastructure Early orbital customers are more likely to include Earth observation operators, defense and security agencies, space station users, lunar infrastructure firms, and AI developers testing models under unusual energy and location constraints.
SU008 Washington Technology Cowboy Space, Darkhive detail their Series B rounds Science Applications International Corp. and the venture capital arm of RTX are the GovCon investor names to take note of here.
SU009 Construct Capital The Energy Bottleneck: Why Cowboy Space Corporation Is Building a New Grid Earth's existing power grid was not designed for the pace and scale of the current AI era. Across major markets, data centers are already facing years-long waits for grid access as demand for compute accelerates faster than existing energy systems can keep up.
SU010 SAIC SAIC Invests in Cowboy Space Corporation
SU011 Ars Technica Will Space Data Centers Actually Save Us From the AI Power Crisis?
SU012 The Next Platform Orbital Data Centers — Physics Says Maybe
SU013 Microsoft Microsoft Planetary Computer — Azure Orbital Infrastructure
SU014 NVIDIA NVIDIA Announces Space-1 LEO AI Data Center Modules
SU015 FCC.report (Starcloud FCC Filing) FCC IBFS Filing SAT-LOA-20260130-00027 (Starcloud / Lumen Orbit)
SU016 Ars Technica Cowboy Space raises $275M to build vertical rockets with orbital data centers
SU017 TechFunding News Cowboy Space $275M Series B — Robinhood Founder Orbital AI Data Centers
SU018 The Register Cowboy Space orbital data centers fundraise coverage
SU019 Defense News SAIC invests in Cowboy Space orbital data center startup
SU020 Data Center Knowledge Do Orbital Data Centers Make Business Sense for Hyperscalers?
SU021 Satellite Today Cowboy Space Files for 20,000-Satellite Constellation for Data Center Services
SU022 OpenAI Announcing the Stargate Project
SU023 SpaceNews Cowboy files plans for up to 20,000 orbital data centers
SU024 SiliconANGLE Cowboy Space raises $275M to build orbital AI data centers
SU025 MarketsandMarkets Orbital Data Center Market Report
SU026 VentureBeat Cowboy Space raises $275M to build orbital AI data centers
SU027 NewSpaceEconomy Cowboy Space Seeks FCC Approval for Stampede Orbital Data Centers
SU028 U.S. Energy Information Administration AI-driven data center growth is expected to significantly increase electricity demand
SR013 FCC Applications Accepted for Filing Public Notice (SAT-02007)
SR014 FCC.report SAT-LOA-20260323-00135 docket page
SR015 FCC.report SAT-PDR-20260514-00022 docket page
SR016 SpaceNews Cowboy files plans for up to 20,000 orbital data centers
SR017 Satellite Today Cowboy Space Files for 20,000-Satellite Constellation for Data Center Services
SR018 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers, Forcing the ODC Thesis Into View
SR019 NewSpaceEconomy Orbital Data Center Companies Building Space-Based Compute Infrastructure
SR020 TechCrunch There aren't enough rockets for space data centers — Cowboy Space raised $275M to build them
SR021 SpaceNews Cowboy raises $275 million to build rockets with orbital data center upper stages
SR024 Business Wire Cowboy Space Corporation Raises $275M Series B For Vertically-Integrated Orbital Data Centers and Rockets
SR028 Bureau of Industry and Security EAR | Bureau of Industry and Security
SR031 Construct Capital The Energy Bottleneck: Why Cowboy Space Corporation Is Building a New Grid
SR033 Cowboy Space Cowboy Space Corporation | Orbital Data Centers for AI
SR034 Payload Aetherflux Rebrands, Pivots Business—and Raises $275M
SR035 FCC FCC Adopts New "5-Year Rule" for Deorbiting Satellites
SR036 FCC FCC 22-74 Second Report and Order
SR037 NASA Orbital Debris Program Office
SR038 European Space Agency Space Debris
SR039 Federal Aviation Administration Licenses, Permits and Approvals
SR040 U.S. Department of State DDTC Understand The ITAR - DDTC Public Portal
SR041 The Register Orbital datacenters are a pie-in-the-sky idea: Gartner
SR042 The Register Orbital datacenter startup admits launch economics don't fly
SR043 The Register Orbital datacenters subject to space weather, debris, stress
SR044 Breaking Defense Despite “peak hype,” orbital data centers for AI not yet ready for NatSec prime time
SR055 European Space Agency Space debris by the numbers
SR056 European Space Agency ESA Space Environment Report 2025
SR057 NASA Orbital Debris Program Office | Quarterly News
SR058 American Astronomical Society AAS Statement on Satellite Proliferation
SR059 SATCON1 / NSF NOIRLab / AAS Impact of Satellite Constellations on Optical Astronomy and Recommendations Toward Mitigations
SR060 JASON / MITRE / NSF The Impacts of Large Constellations of Satellites
SR061 NASA Science What are the Van Allen Belts and why do they matter?
SR062 NOAA Space Weather Prediction Center Radiation Belts
SR063 NVIDIA NVIDIA space computing
SV001 BusinessWire Cowboy Space Corporation Raises $275M Series B For Vertically-Integrated Orbital Data Centers and Rockets $275M in Series B financing at a $2B valuation led by Index Ventures
SV002 Bloomberg Cowboy Space Raises $275 Million at $2 Billion Valuation
SV003 Multiples.vc Public AI Valuation Multiples — June 2026 Neoclouds and data centers valued at 11.8x [forward revenue]; Neoclouds are the fastest-growing group in the set (median 104% revenue growth)
SV004 Seraphim Space Seraphim Space Index Q1 2026 — Quarterly Review of Global Private Investment $8.0bn invested in Q1 2026, doubling the previous record set in Q4 2025
SV005 Stock Analysis CoreWeave (CRWV) Market Cap & Net Worth
SV006 Tech Startups Crusoe raises $1.37B in funding at $10B valuation to build gigawatt-scale AI data centers Crusoe raises $1.37B in funding at $10B valuation
SV007 TechCrunch Starcloud raises $170 million Series A to build data centers in space Starcloud's latest funding round values the space compute company at $1.1 billion
SV008 GeekWire Orbital AI: Seattle-area startup Starcloud hits $1.1B valuation to build space-based data centers
SV009 U.S. News & World Report Space Investment in First Quarter Hits Record as SpaceX IPO Buzz Lifts Sector
SV010 BlacKnight Space Labs The Orbital Data Center Race in 2026: Cowboy, Starcloud, SpaceX, Anthropic, and Why Capital Is Pouring Into AI Compute in Orbit Current valuations are seen by many industry watchers as highly speculative — setting up the potential for dramatic corrections if milestones are missed
SV011 Finro Financial Consulting AI Valuation Multiples Q1 2026: Investors Reprice Quality Companies still selling 'growth now, business model later' faced sharper discounts as underwriting shifted toward durability rather than narrative momentum
SV012 Forge Global SpaceX IPO: Investment Opportunities & Pre-IPO Valuations
SV013 SaaS Sentinel Cowboy Space Raises $275M to Build Rocket-Powered AI Data Centers in Orbit
SV014 SpaceNews Cowboy raises $275 million to build rockets with orbital data center upper stages
SV015 TechCrunch There aren't enough rockets for space data centers — Cowboy Space raised $275 million to build them
SV016 CNBC Cowboy Space raises $275 million for orbital data centers
SV017 Payload Space Aetherflux Rebrands, Pivots Business and Raises $275M
SV018 Andreessen Horowitz a16z Announcement: Aetherflux (now Cowboy Space)
SV019 PitchBook Space data center startups are attracting serious funding in 2026
SV020 SatNews Cowboy Raises $275M and Files for 20,000 Orbital Data Centers
SV021 The Wall Street Journal Aetherflux Raises $275 Million for Orbital Data Center Space Startup
SV022 Financial Times Cowboy Space Series B: Orbital Data Centers
SV023 Fenwick & West LLP Fenwick Represents Cowboy Space Corporation in $275M Series B Financing
SV024 IEEE Spectrum Cowboy Space Raises $275M for Orbital Data Centers
SV025 The Register Orbital datacenters are a pie in the sky idea, Gartner says Orbital datacenters are a pie in the sky idea [Gartner]
SV026 The Register Orbital datacenter startup admits launch economics don't fly Orbital datacenter startup admits launch economics don't fly
SV027 Data Center Knowledge Do Orbital Data Centers Make Business Sense?
SV028 Breaking Defense Cowboy Space DARPA Orbital Data Center
SV029 U.S. Securities and Exchange Commission (EDGAR) EDGAR Company Search — CoreWeave Inc. S-1 Registration
SV030 MarketsandMarkets Orbital Data Center Market Research Report