Crusoe
Crusoe — AI 工厂公司
Crusoe 是目前最有可信度的垂直一体化 AI 工厂公司:能源优先逻辑已经跑通,2.1 GW 园区已锁定,估值超过 $10B,并拿到 NVIDIA 首选伙伴身份;但 GPU 定价商品化、客户高度集中、2025 年 3 月宕机原因不明,以及 Engine No. 1 天然气 JV 带来的 ESG 矛盾,都可能拖累企业销售。
封面要素
公司概况
Crusoe 是一家总部位于 Denver 的 AI 基础设施公司,已从 Bitcoin 挖矿转型为全栈「AI 工厂」提供商,组合了现场发电(360 MW 天然气涡轮、可再生能源 PPA)、自研直达芯片液冷(PUE 1.2-1.3)、模块化 AI 工厂基础设施(Spark Factory)和 GPU 云平台(Crusoe Cloud)。公司在 Texas Abilene 拥有已承诺 2.1 GW 容量的园区,由 Microsoft(900 MW)、Oracle/Stargate 和 OpenAI 锚定,另有一个 1.8 GW Wyoming 园区处于规划中。Crusoe 的 Series E($1.375B,2025 年 10 月)以 $10B+ 估值确认其独角兽地位,也把公司推到垂直整合 neocloud 运营商的领先位置。2024 年收入为 $276M;BTC 挖矿业务已于 2025 年 3 月出售给 NYDIG,使 Crusoe 成为纯 AI 基础设施公司。NVIDIA 既是优先 GPU 供应伙伴,也是股权投资方。
- 成立时间
- 2018-01-01
- 创始人
- Chase Lochmiller, Cully Cavness
- 创立地点
- Denver, CO
- 总部
- Denver, CO
- 产品
- Crusoe 提供三条整合产品线:(1)Crusoe Cloud——GPU-as-a-service,提供 NVIDIA H100、A100 和即将推出的 B200/GB200 实例,H100 SXM 定价 $4.29/hr;(2)AI Factory Campus——为大规模场景专门建设的数据中心,配套现场发电和液冷,目前由 2.1 GW Texas Abilene 园区锚定;(3)Crusoe Spark——在 Colorado Brighton 的 Spark Factory 制造的模块化 AI 工厂单元(350K sq ft,目标 2026 年 Q3 投产),支持分布式站点快速部署。Edge Zones 和 Command Center SaaS 进一步扩展平台。
- 客户
- 需要大规模预留 GPU 容量的企业 AI 实验室和超大规模云厂商;前沿模型训练与推理工作负载;有 ESG 兼容 AI 基础设施要求的企业。
- 商业模式
- GPU 云收入来自预留和按需算力合同;与锚定租户(Microsoft、Oracle/Stargate、OpenAI)签订长期园区租赁协议;销售模块化 AI 工厂产品(Spark 单元);此前还有 BTC 挖矿收入(2025 年 3 月出售给 NYDIG)。
- 阶段
- Series E
- 融资情况
- Series E($1.375B,2025 年 10 月 24 日,投后估值 $10B+)由 Valor Equity Partners 和 Mubadala Capital 共同领投;共 71 名投资者(SEC Form D)。Series D($600M,2024 年 12 月,估值 $2.8B)由 Founders Fund 领投,NVIDIA、Fidelity、Mubadala、Ribbit Capital 和 Valor 参投。已披露累计融资约 $2.2B+。额外项目融资包括:$3.4B Blue Owl Capital 合资项目(园区建设)和 $9.6B JPMorgan 项目融资额度。
执行摘要
主要优势
- 能源 + 算力垂直一体化模型带来可持续成本优势:PUE 1.2-1.3,而行业约 1.8;Abilene 现场燃气轮机切断对电网依赖,支撑 24/7 AI 工作负载。
- NVIDIA 首选伙伴身份叠加股权绑定,让 Crusoe 比非首选 neocloud 更早拿到前沿 GPU(H100、B200、GB200),支撑溢价定价和客户黏性。
- Abilene 园区已有 2.1 GW 锁定容量(Phase 1 已上线,Phase 2 Microsoft 900 MW 于 2026 年 3 月签约),筑起实体基础设施护城河;1.8 GW Wyoming 园区储备扩大了地域选择。
- Series E 以 $10B+ 估值完成,71 家投资者包括 Abu Dhabi SWF(Mubadala)和 Valor,验证了国际机构需求,也提供 18-24 个月现金跑道。
- Crusoe Spark 模块化 AI 工厂(Brighton CO,Q3 2026)是产品线扩张,可能降低部署成本,并让公司摆脱只靠大型园区建设的地域集中。
主要风险
- GPU 定价商品化:H100 现货价在 2025 年从 $8/hr 跌到 $2-3/hr;除非 Crusoe 把客户迁移到更高端的 B200/GB200 SKU,低价续约会压缩 AI 云利润率。
- 客户集中:Microsoft(900 MW)、Oracle/Stargate 和 OpenAI 可能占 Abilene 已锁定容量的 >60%;未来 3-5 年超大规模云厂商把能力内化,是结构性的生存风险。
- 运营事故原因不明:2025 年 3 月 45 小时宕机打断 AI 训练工作负载;公司未公开根因分析,也未披露带财务罚则的 SLA。
- ESG/漂绿风险:Engine No. 1 JV 将新增 4.5 GW 天然气发电,和清洁 AI 定位冲突;一旦 Scope 3 排放数据成为硬要求,Fortune 500 ESG 采购门槛可能把 Crusoe 排除在外。
- 执行集中:公司同时推进五个重大资本项目(Abilene Phase 2、Wyoming 园区、Spark Factory、Series E 部署、GPU 云扩张),员工 1,235 人,其中 15 个月新增 635+ 人。
未决问题
- Microsoft、Oracle/Stargate 和 OpenAI 的锚定租户合同条款(期限、提前终止、续约条件)未公开;没有这些数据,客户集中风险无法精确量化。
- Blue Owl JV 与 JPMorgan 项目融资约束包(DSCR 阈值、补救期)未公开披露;收入承压时的资本充足性无法评估。
- 2025 年 3 月宕机的根因和补救计划尚未公开;重复事故概率无法评估。
- 2024 年运营的 GHG Protocol Scope 1/2/3 排放数据尚未以可审计格式发布;ESG 采购排除风险无法量化。
- MemoryAlloy KV-cache 和 Spark 模块化工厂设计尚未确认获得 USPTO 专利保护;自由实施风险无法量化。
目录
01公司概况
1.1 身份与商业模式
Crusoe 从「Crusoe Energy Systems」更名而来,在 crusoe.ai 运营;公司称自己是「AI 工厂公司」,使命是加速能源与智能的丰裕。公司总部位于 255 Fillmore Street, Suite 400, Denver, Colorado 80206,在 Delaware 注册为 Crusoe Inc.(曾用名 Crusoe Energy Holdings Inc.,CIK 0001924674)。 Crusoe 的核心主张是三层垂直整合:(1)大规模能源采购和场址开发;(2)面向 AI 优化的数据中心设计、建设与制造;(3)自研 GPU 云平台(Crusoe Cloud),提供托管推理、可观测性和边缘区域服务。这套垂直模式旨在把从拿地到 GPU 上线的周期,从行业常见的 3–5 年压到绿地园区少于 12 个月。 公司靠三类方式变现:云算力订阅和按需 GPU 租赁(Crusoe Cloud)、与超大规模云厂商签订长期容量协议(如 Microsoft 在 Abilene 承诺 900 MW)、以及模块化 AI 工厂部署(Crusoe Spark 单元)。2024 年收入为 $276M,其中 AI 云贡献 $124M(同比增长 460%),Bitcoin/DFM 挖矿贡献 $152M。NYDIG 于 2025 年 3 月收购 Crusoe 的 Bitcoin 和 DFM 业务后,公司正在转向纯 AI 云基础设施收入。 [CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 | 日期 | 置信度 | 缺口 / 注意事项 |
|---|---|---|---|---|
| 估值 | $10B+ | Oct 2025 | 高 | 私营公司;SEC Form D 确认 Series E 融资金额 |
| 累计股权融资 | ~$3.4B | Oct 2025 | 高 | SEC Form D Series E + 此前轮次;不含债务 / 合资项目 |
| 最近一轮融资 | Series E 轮 $1.375B | Oct 2025 | 高 | SEC Form D Acc-no 0001924674-25-000004 |
| 2024 年收入 | $276M | 2024 FY | 中 | 公司披露;未经审计;AI 云 $124M,BTC/DFM $152M |
| 2025 年收入估计 | $500M–$1B | 2025 年估计 | 低 | 估计值;公司尚未披露 2025 年全年收入 |
| AI 云 ARR 增长 | 同比 150% | 2025 | 中 | 公司声称;分母和准确 ARR 未披露 |
| 员工数 | 1,235 | Mar 31, 2026 | 高 | 公司新闻稿披露 |
| 已投运电力 | 3.4 GW | 2026 年初 | 中 | 公司披露;包括已转让给 NYDIG 的历史 DFM 站点 |
| Abilene 园区容量 | 已承诺 2.1 GW | Mar 2026 | 高 | 原 1.2 GW + Microsoft 900 MW 承诺 |
| 总电力管线 | >45 GW | 2026 | 低 | 公司声称;未独立验证 |
| 毛利率 | 未披露 | 私营公司;无公开财务;尽调中索取 | ||
| 净收入留存率 | 未披露 | 私营公司;尽调中索取 |
估值、收入和增长指标来自公司披露或分析师估计;毛利率和 NRR 无法取得。除非另有说明,所有美元金额单位均为百万。
[CO001, CO003, CO013, CO014, CO016, CO017]截至 2026 年 5 月,投委会层面的 Crusoe 规模、牵引力、资本和证据质量快照。
[CO001, CO003, CO013, CO017, CO019, CO023]1.2 领导层与治理
Crusoe 由 Chase Lochmiller(CEO)和 Cully Cavness(总裁兼 CSO)共同创立。Lochmiller 在 MIT 学习物理与计算机科学,随后在 Stanford 完成 AI 方向计算机科学硕士;他还曾在 Jane Street Capital 做过数年量化交易员,之后转向能源科技创业。Cavness 拥有 Middlebury College 地质学本科学位和 Oxford University Saïd Business School MBA,并把油田运营经验带入 Crusoe 的能源采购和场址获取工作。 2025 年 12 月,Michael Gordon 加入并担任 COO 和 CFO。Gordon 此前在 MongoDB 担任 CFO 和 COO,主导公司 2017 年 IPO,并在任期内帮助年收入增长约 50 倍。他的加入为 Crusoe 资本密集型扩建带来上市公司级财务纪律和投资者关系经验。前 CFO Matthew DeNezza 转为顾问角色,任期延续至 2026 年。 其他关键高管包括 Jamey Seely(首席法务官兼秘书)、Nader Pakfar(房地产总法律顾问,2026 年 1 月 21 日任命)、Erwan Menard(Crusoe Cloud 产品 SVP)、Nadav Eiron(云工程 SVP)、Matt Field(首席房地产官)和 Chris Dolan(首席数据中心官)。董事会有战略顾问 Bill Stein(Digital Realty Trust 前 CEO)和 Peter Gross。关键人物风险不低:Lochmiller 是对外的技术愿景人物,Cavness 是能源战略负责人。除 SEC Form D 的高管名单外,治理文件未公开——董事会构成、独立董事人数和优先权结构仍未披露。 [CO007, CO008, CO009, CO010, CO011, CO012]
| 姓名 | 职务 | 背景 | 创始人-市场匹配度 | 关键人风险 |
|---|---|---|---|---|
| Chase Lochmiller | CEO 兼联合创始人 | MIT 物理 / 计算机科学;Stanford 计算机科学 / AI;Jane Street Capital 量化交易员 | 技术愿景;AI/能源投资逻辑发起人;深厚投资人关系 | 高 |
| Cully Cavness | 总裁兼 CSO(前 COO) | Middlebury 地质学;Oxford MBA;油田运营背景 | 能源采购与站点获取;大规模现场运营 | 高 |
| Michael Gordon | COO 兼 CFO(2025 年 12 月加入) | MongoDB CFO/COO;带领 NASDAQ IPO;任内收入增长约 50x | 上市公司财务纪律;资本市场可信度 | 中 |
| Jamey Seely | 首席法务官兼秘书 | 公司 / 监管法背景 | SEC 合规、公司治理、IP/M&A | 中 |
| Nader Pakfar | 房地产总法律顾问(2026 年 1 月加入) | 房地产法律背景 | Abilene 和 Wyoming 园区的土地 / 基础设施法律风险 | 低 |
| Erwan Menard | Crusoe Cloud 产品高级副总裁 | 云产品管理 | 产品路线图和云平台战略 | 中 |
| Nadav Eiron | 云工程高级副总裁 | 软件 / 系统工程 | 云平台技术落地 | 中 |
| Matt Field | 首席房地产官 | 房地产开发 | 选址、土地收购、许可审批 | 中 |
| Chris Dolan | 首席数据中心官 | 数据中心建设 / 运营 | Abilene 和未来园区建设落地 | 中 |
| Bill Stein | 董事会顾问 | 前 Digital Realty Trust CEO(NYSE: DLR) | 行业可信度;REIT 资本市场经验 | 低 |
所有职务和任期均来自 Crusoe 官方新闻稿和高管公告;关键人风险反映估计依赖度,不是正式评级。
[CO007, CO008, CO009, CO010, CO011, CO012]1.3 融资历史与资本结构
自 2018 年以来,Crusoe 已在六轮股权融资中募集约 $3.4B,并为 Texas Abilene 园区建设额外锁定约 $13B 项目融资和合资承诺。这条融资曲线是 neocloud 赛道中最快的规模化曲线之一。 公司风险融资始于 2019 年 5 月的 $4.5M 种子轮,由 Bain Capital Ventures 和 Founders Fund 共同领投。2019 年 12 月,公司完成 $30M Series A。Series B($128M,2021 年 4 月)由 Valor Equity Partners 领投。Series C($350M,估值 $1.75B,2022 年 4 月)由 G2 Venture Partners 领投,另有一笔 $155M 分期融资。 Series D($600M,估值 $2.8B,2024 年 12 月 12 日宣布)由 Founders Fund 领投,NVIDIA、Fidelity、Mubadala、Ribbit Capital 和 Valor 参投。SEC Form D(2024 年 11 月 21 日提交)确认了该次发行。Series E($1.375B,估值 $10B+,2025 年 10 月 24 日宣布)由 Valor Equity Partners 和 Mubadala Capital 共同领投;按 SEC Form D(2025 年 10 月 23 日提交),共有 71 名投资者参与。累计股权融资约 $3.4B。Blue Owl Capital 为 Abilene 数据中心建设承诺 $3.4B 合资资金,JPMorgan 正安排约 $9.6B 项目融资,使 Abilene 已承诺总资本超过 $15B。 [CO013, CO014, CO015, CO016, CO017, CO018]
| 利益相关方 | 角色 | 控制权 / 经济重要性 | 尽调问题 |
|---|---|---|---|
| Valor Equity Partners | Series B 领投;Series E 联合领投 | 自 2021 年起为主导投资人;可能有董事会席位 | 确认董事会席位、优先股结构、反稀释条款 |
| Mubadala Capital(Abu Dhabi) | Series E 联合领投 | 主权财富资本;参与 Series D 和 E | LP 集中度风险;AI 基础设施上的地缘政治限制 |
| Founders Fund | 种子轮联合领投;Series D 领投 | 早期支持者兼 Series D 领投;可能有董事会席位 | 董事会权利;按比例跟投权 |
| Bain Capital Ventures | 种子轮联合领投 | 早期战略支持者;与 Founders Fund 联合领投种子轮 | 当前持股比例;清算优先权 |
| NVIDIA Corporation | Series D 和 E 投资人 | 战略 GPU 供应商和技术伙伴;潜在分销杠杆 | NVIDIA 股权条款;NVIDIA GPU 供应是否有任何优先购买权 |
| T. Rowe Price | Series E 投资人 | 传统跨界机构投资人;其估值标记释放价格可信度信号 | T. Rowe 是否套用盯市估值纪律 |
| Fidelity Investments | Series D 和 E 投资人 | 跨界机构投资人;价格发现信号 | 确认其私人基金 NAV 中的 $10B+ 估值标记 |
| Blue Owl Capital | Abilene 合资伙伴 | 用于园区建设的 $3.4B 合资项目;债务 / 股权混合 | 合资条款、有追索权 / 无追索权、控制权安排 |
| JPMorgan Chase | 项目融资安排方 | Abilene 约 $9.6B 项目融资;对园区完工至关重要 | 贷款契约、提款条件、追索触发条款 |
| Lowercarbon Capital | Series E 投资人 | 聚焦气候的 VC;投资组合可信度信号 | 碳 / 能源审计;Lowercarbon 是否仍认为用气与其使命相容 |
| Chase Lochmiller(CEO) | 创始人-员工 | 创始人股权;高管控制权;技术领导力 | 归属时间表、悬崖期,以及如有的超级投票权股份 |
持股比例未公开披露。董事会构成和优先股堆叠为私有信息。利益相关方列表来自官方新闻稿和 SEC Form D 投资人数。
[CO013, CO014, CO015, CO016, CO017, CO018]1.4 运营规模与关键里程碑
截至 2026 年 3 月 31 日,Crusoe 有 1,235 名员工,高于 2024 年底约 600 人。公司业务覆盖美国九个州,并在 Iceland、Norway、Ireland 和 Israel 有国际布局。Texas Abilene 旗舰园区已承诺容量达 2.1 GW(计入 2026 年 3 月 Microsoft 宣布的 900 MW)。公司已宣布一个 1.8 GW Wyoming 园区。总电力管线超过 45 GW。 截至 2026 年初,基础设施指标包括:9.8M sq ft 设施、3.4 GW 已投运电力、约 946,000 GPU 等效容量。Crusoe 位于 Colorado Brighton 的 Spark Factory(350,000 sq ft,$200M 投资,200+ 个岗位)正在制造模块化预制 AI 工厂单元,首批交付预计在 2026 年 Q3。直接到芯片液冷把电力使用效率(PUE)压到据称 1.2–1.3。 反向背景:Heatmap News(2025 年 3 月)报道称,尽管 Crusoe 主打清洁能源叙事,公司仍使用天然气发电,并已聘请 Engine No. 1 开发 4.5 GW 新天然气容量,以满足 AI 数据中心需求。Texas Commission on Environmental Quality(TCEQ)许可文件确认,Abilene 园区有 360 MW 现场燃气涡轮,约占 1.2 GW 一期容量的 30%。Crusoe 的环保品牌与实际能源采购之间由此产生张力,可能影响重视 ESG 的客户和投资者。 [CO019, CO020, CO021, CO022, CO023, CO024]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2018 | 公司以 Crusoe Energy Systems 名义创立 | 创立 | ~$0 | Chase Lochmiller, Cully Cavness | 为通过火炬气捕集 + 比特币挖矿变现搁浅天然气而创立 |
| May 2019 | 种子轮 | 融资 | $4.5M | Bain Capital Ventures 与 Founders Fund | 火炬气减排概念验证的初始资本 |
| Dec 2019 | Series A 轮 | 融资 | $30M | 未披露领投方 | 扩大 DFM(数字火炬气减排)运营 |
| Apr 2021 | Series B 轮 | 融资 | $128M | Valor Equity Partners(领投) | GPU 加速云转型开始;Valor 带来 Tesla 级运营可信度 |
| Apr 2022 | Series C + 追加批次 | 融资 | $350M + $155M = $505M,估值 $1.75B | G2 Venture Partners(领投)、Founders Fund、Valor 等 | 扩大 AI 云运营;首次大型数据中心建设 |
| Jun 2024 | Abilene 园区破土动工 | 规模化 | 地点:Taylor County, TX | Lancium(电网伙伴)、DPR Construction(总包) | 首期 1.2 GW;首个 GW 级 AI 原生绿地项目 |
| Nov 2024 | SEC Form D Series D 申报 | 申报 | 该轮融资发行已确认 | SEC EDGAR CIK 0001924674 | 资本募集获得监管文件确认 |
| Dec 12, 2024 | 宣布 Series D | 融资 | $600M,估值 $2.8B | Founders Fund(领投)、NVIDIA、Fidelity、Mubadala、Ribbit、Valor、Long Journey | 首次确认独角兽身份;NVIDIA 成为战略投资人 |
| Feb 18, 2026 | 发布 Crusoe Command Center | 产品 | SaaS 可观测性平台 | 内部 | 给数据中心运营加上软件层;粘性信号 |
| Mar 12, 2026 | 发布 Edge Zones | 产品 | 主权 / 低延迟 AI 计算 | 内部 | Crusoe Cloud 地理覆盖走出 Abilene |
| Mar 27, 2026 | 宣布 Microsoft 900 MW Abilene 承诺 | 合作 | 900 MW;叠加到 1.2 GW 一期 | Microsoft | 验证 Abilene $15B+ 投资;锚定租户降低园区风险 |
| Mar 25, 2025 | NYDIG 收购 Crusoe Bitcoin/DFM 业务 | 反向 | 425+ 个模块化 DC、250+ MW、约 135 名员工 | NYDIG(收购方) | 战略转向:退出 BTC/DFM;成为纯 AI 云业务;收入结构迁移 |
| Nov 24, 2025 | Upstream Data 专利诉讼和解 | 法律 | 双方撤诉 + 许可协议 | Upstream Data Inc. | 专利诉讼风险解除;条款保密 |
| Oct 24, 2025 | 宣布 Series E | 融资 | $1.375B,估值 $10B+ | Valor + Mubadala(联合领投)、NVIDIA、T. Rowe、Tiger Global,共 71 名投资人 | 确认十角兽地位;机构强力背书 |
公司创立和早期里程碑日期为近似值。财务金额来自官方新闻稿和 SEC Form D 申报。
[CO013, CO014, CO015, CO016, CO017, CO018]从创立到 2026 年 5 月的关键融资、产品、规模和负面里程碑。
[CO013, CO014, CO015, CO016, CO017, CO018]能源获取、AI 工厂建设、GPU 云与客户如何在 Crusoe 的垂直模式里连接。
[CO001, CO002, CO003, CO004, CO005]1.5 展示材料
02市场分析
2.1 市场定义与范围
Crusoe 位于两个相邻市场的交叉点:GPU 云计算(neocloud 和 GPU-as-a-Service)与 AI 数据中心基础设施。承保 Crusoe 时,相关市场横跨三层:向 AI 训练和推理工作负载出售的托管 GPU 算力(Crusoe Cloud);通过长期超大规模云厂商合同出售的大规模、电力优化数据中心容量(Microsoft 900 MW Abilene 承诺);以及卖给自建专用本地 AI 算力企业的模块化 AI 工厂部署(Crusoe Spark)。 明确排除在市场边界之外的,是通用云 IaaS(CPU VM、对象存储、CDN)、消费级 AI 应用软件、AI 半导体制造,以及无关能源基础设施。neocloud GPU 云市场由需要高密度大显存 GPU 集群(H100/H200/B200/GB200 级别)的工作负载定义,用于深度学习训练、大模型微调或持续批量推理——这些工作负载越来越难被通用公有云以所需密度、延迟或成本服务。 Crusoe 服务的主要现状替代方案是超大规模云厂商 GPU 云(AWS P4/P5、Azure NDv5、GCP A3)。这仍是最大可触达池,但传统市场里的电力、空间和交付周期约束很紧。H100 实例的 GPU 现货价格已从 2023 年中约 $8/GPU-hr 大幅降至 2025 年初 $2–3/GPU-hr,压缩了所有 neocloud 运营商的单位经济。这一价格动态并不降低绝对 GPU 算力需求——它反映供给响应——但会直接影响单位收入,并把行业推入成本领先竞赛;Crusoe 的能源和基础设施垂直整合正是其竞争论点的核心。[CM001, CM002, CM003, CM004, CM005, CM006]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 与 Crusoe 的相关性 |
|---|---|---|---|---|
| GPU 云 / 新云计算 | 为 AI 训练、微调和推理租赁 H100/H200/B200/GB200 GPU 集群的预留和现货容量 | CPU VM、对象存储、CDN、通用 IaaS | AI 实验室、超大规模云厂商、企业 AI 团队、ML 研究者 | 核心直接市场;Crusoe Cloud 在这里直接竞争 |
| AI 数据中心基础设施(超大规模云厂商合同) | AI 工作负载的长期容量协议、托管、电力和设施租赁 | 与 AI 无关的企业 IT 托管、传统运营商酒店 | 超大规模云厂商(Microsoft、Oracle、Meta)和大型 AI 实验室 | 来自 Microsoft Abilene 900 MW 承诺的收入;长期合同 |
| 模块化 AI 工厂(预制数据中心) | 模块化预制 AI 计算基础设施单元(Crusoe Spark)的资本采购 | 未集成设施的标准机架服务器和网络设备 | 企业基础设施买方、政府、中型市场 AI 运营商 | 新兴产品;首批单元预计 2026 年 Q3 出自 Brighton CO 工厂 |
| AI 园区能源基础设施 | 电力采购、可再生能源 PPA、燃气发电、与 AI 园区共址的电池储能 | 非 AI 用途的公用事业级发电;无关的电网级储能 | 有电力承诺的数据中心运营商和超大规模云厂商 | 上游垂直整合;带来有竞争力的能源成本和电力护城河 |
| 超大规模云厂商云 GPU — 现状替代品 | AWS P4/P5、Azure NDv5、GCP A3 上的按需和预留 GPU 实例 | 仅 CPU 实例、存储工作负载、非 GPU 云服务 | 所有在自建 / 购买之间评估的企业 AI 买方 | 主要替代品;低端 H100 现货 $2-3/hr 侵蚀切换激励 |
市场边界按 Crusoe 估值用途定义。相邻硬件、软件和服务层已排除。买方 / 付款方区分在大型超大规模云厂商合同里最重要,因为预算权在董事会层面。
[CM001, CM002, CM003, CM004, CM005]GPU 云 TAM(2030 年 $26.6B)只是更广义 AI 基础设施支出($394B)的一小部分;按 Crusoe 聚焦美国和大集群的定位,其可服务的新型云 SAM 进一步收窄到 $3–7B。
SAM 和 SOM 是分析师估计,基于 Crusoe 容量披露和 GPU 云定价基准,并非来自已发布报告。TAM 分层引用 MarketsandMarkets 和 IDC。各层不可相加;它们只是用逐步收窄的口径看同一底层市场。
[CM007, CM008, CM009, CM010, CM011, CM012]2.2 市场规模分析
多个分析师来源共同指向一个大且高速增长的总可用市场。MarketsandMarkets 估计,GPU-as-a-Service 市场 2025 年为 $8.21B,到 2030 年将以 26.5% CAGR 增至 $26.62B。更广义的 AI 基础设施市场 2024 年规模为 $135.81B,预计到 2030 年达到 $394.46B,CAGR 为 19.4%。IDC 报告称,2025 年云基础设施 GPU 支出增长 46.8%,达到 $157.8B。ComputeForecast 称,neocloud 子赛道 2025 年 Q2 季度收入超过 $5B,同比增加 205%。 这些估计方向一致,但方法差异很大。MarketsandMarkets 聚焦可租赁 GPU 算力收入;IDC 纳入 AI 服务器硬件资本开支;更广义的 AI 基础设施估计还包括网络、存储、冷却和电力基础设施。Crusoe 最直接竞争的是 neocloud GPU 云细分市场(到 2030 年约 $8–27B 区间),但还通过基础设施建设和长期容量协议捕获额外价值。 Crusoe 的可服务市场(SAM)受美国为中心的布局和大集群定位约束。保守估计,到 2030 年 SAM 为 $3–7B,约为 GPU-as-a-Service TAM 的 10–26%。按 3.4 GW 已投运容量和 $0.3–0.5/W 年化算力收入测算,2028 年可获取市场(SOM)估计为 $1.0–1.7B 年化 GPU 云收入,与 2025 年报告的 150% ARR 增长轨迹一致。Crusoe 未公开正式 SAM/SOM 框架;这些都是分析师构建,需要在尽调中验证。[CM007, CM008, CM009, CM010, CM011, CM012]
| 发布方 | 年份 | 地理范围 | 规模 | CAGR | 方法 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|
| MarketsandMarkets | 2025→2030 | 全球 | $8.21B (2025) → $26.62B (2030) | 26.5% | GPU-as-a-Service 工作负载的自下而上分段测算 | 中 | 付费墙;方法不透明;未把 GPU 新云厂商与超大规模云厂商拆开 |
| MarketsandMarkets | 2024→2030 | 全球 | $135.81B (2024) → $394.46B (2030) | 19.4% | AI 基础设施大市场,包含硬件、网络、冷却、电力 | 中 | 定义过宽,纳入了 Crusoe 不能直接变现的硬件资本开支;高估直接 TAM |
| IDC | 2025 | 全球 | 2025 年云基础设施 GPU 支出 $157.8B | 46.8% 同比 (2025) | 基础设施支出调研;包含云端 AI 服务器出货 | 中 | 资本支出口径不同于租赁收入;只是单年数据点,不是多年预测 |
| ComputeForecast | Q2 2025 | 全球(GPU 新云厂商) | GPU 新云厂商季度收入 $5B+ | 205% 同比 (Q2 2025) | 汇总 GPU 新云厂商监管文件和披露中的收入 | 中 | 集中在 CoreWeave,估计 ARR 约 ~$5B;市场成熟后增速可能回归常态 |
| Crusoe / Sacra 分析师 | 2025 | 以美国为中心 | 2024 年 AI 云收入 $124M;2025 年 ARR 同比增长 150% | 2025 年 ARR 同比增长 150% | 公司披露指标;Sacra 分析师估计和评论 | 低 | 未审计;ARR 分母未披露;从很低基数起量;TCV 和 ARR 定义未经核实 |
SAM 估计到 2030 年为 $3–7B(占 GPU-as-a-Service 市场 10–26%),依据是美国业务足迹和大集群取向。SOM 估计到 2028 年年化运行率为 $1.0–1.7B,依据是 3.4 GW 已投运容量。两者都是分析师构造的口径。
[CM007, CM008, CM009, CM010, CM011, CM012]分析师对 GPU 即服务增长的估计从保守到激进不等;即便在下行情景下,基准情景也支持 Crusoe 到 2030 年仍有庞大且持久的市场机会。
低 / 高边界是分析师基于已发布增长率和市场情景的估计。Crusoe ARR 估计来自公司披露,未经审计。中位值代表当前增长路径下的基准情景。
[CM007, CM008, CM009, CM011, CM012, CM014]2.3 买方与细分市场分析
GPU 算力买家分为三大类:(1)需要大型 GPU 集群训练基础模型的前沿 AI 实验室和超大规模云厂商——ACV 最高、合同最长,规模化后对价格不敏感;(2)运行推理、微调和研究工作负载的企业 AI 平台团队——中等 ACV,预留和现货混合;(3)需要灵活小规模访问的开发者和研究机构——ACV 较低,以现货市场为主。 Crusoe 客群明显偏向前两层。Microsoft 900 MW Abilene 承诺代表超大规模云厂商为长期大规模容量签约。Oracle/Stargate 和 Together AI 代表前沿实验室及平台客户。Cursor(AI coding assistant)和 Redwood Materials(先进制造 AI,合作规模扩大 7 倍)代表企业推理客户。Crusoe 2024 年 $124M AI 云收入(同比增长 460%)和 2025 年 ARR 同比增长 150%,确认企业账户获取已从单一超大规模云厂商扩展出去。 GPU 云决策的预算通常归技术公司 VP/C-level 工程或基础设施负责人所有,年合同超过 $1M 时采购会参与。采用触发点包括:(a)超大规模云厂商 GPU 可用性不足;(b)相对 AWS/Azure 按需定价存在成本套利;(c)合规或数据驻留要求让客户偏好专用基础设施。 一个负面动态是:H100 现货价格 $2–3/hr 已匹配甚至低于 Crusoe 公布的 H100 SXM 价格 $4.29/GPU-hr,导致现货型开发者 / 研究细分市场越来越多由打折的超大规模云厂商容量服务。Crusoe 防守最稳的是最高层客户;在该层,电力可用性、集群密度和长期预留,而不是小时现货价格,才是主要采购标准。[CM016, CM017, CM018, CM019, CM020, CM021]
| 细分市场 | 买方 | 用户 | 付款方 | 工作流 | 预算负责人 | 采用触发点 |
|---|---|---|---|---|---|---|
| 前沿 AI 实验室 / 超大规模云厂商 | Microsoft、Oracle、Meta、Google、OpenAI 等前沿 AI 买方 | ML 工程师、研究团队 | 超大规模云厂商的基础设施 / 云资本开支预算 | 预训练大型基础模型;大规模推理服务 | CTO / 基础设施 VP / 董事会级资本开支审批 | GPU 供给稀缺;传统市场受电力约束;多年预留确定性 |
| AI 平台公司 | Together AI、Databricks、Luma AI、Cohere 等 AI 平台公司 | 模型运营方、API 客户 | 基础设施工程预算 | 模型托管、微调、规模化推理服务 | CTO / 工程 VP | 超大规模云厂商容量超限;相对 AWS/Azure 预留价格有成本套利 |
| 企业 AI 团队(软件) | Cursor 及类似 AI 编程 / 生产力工具供应商 | 软件工程师、产品团队 | 工程基础设施预算 | 产品功能中的 AI 推理;生产环境 LLM API 调用 | 工程 VP / CTO | 生产推理的可靠性和延迟;专用 GPU 集群密度 |
| 企业 AI 团队(工业 / 科学) | Redwood Materials, Sony Research | 数据科学、机器人、材料科学团队 | 研发和资本预算 | 电池材料建模;仿真;机器人策略训练(如 Sony GT Sophy) | CTO / 首席数据官 | 专用硬件获取;持续 GPU 集群利用下的能效 |
| 开发者 / 研究(小规模) | 个人 ML 从业者、大学实验室、AI 初创公司 | 研究人员、学生、小型工程团队 | 课题经费、初创种子轮资金、研发预算 | 模型实验、学术研究、快速原型 | 首席研究员 / 工程负责人 | 相对 AWS 现货实例的价格竞争力;新一代 GPU(H200、B200)可得性 |
按已披露客户名单看,Crusoe 当前收入结构更偏前沿 / 超大规模云厂商和企业 AI 平台细分市场。开发者 / 研究市场也有覆盖,但会受到超大规模云厂商 现货折扣的价格压力。
[CM016, CM017, CM018, CM019, CM020, CM021]前沿 AI 实验室和企业 AI 平台公司,是 Crusoe 最有位置的买方分层;开发者和研究买家则面临超大规模云厂商现货折扣带来的显著价格压力。
评分是有证据支持的序数判断,来自 Crusoe 客户披露、定价数据和分析师覆盖。单个买方合同未公开披露。
[CM016, CM017, CM018, CM019, CM020, CM021]企业 GPU 计算采用通常从探索走向已承诺基础设施;Crusoe 当前客户群集中在预留和超大规模云厂商承诺阶段。
[CM020, CM021, CM022, CM023, CM024]2.4 增长驱动因素与市场约束
首要增长驱动因素是 AI 算力需求持续指数级增长。基础模型训练对算力呈超线性扩张,推理需求又随模型采用率上升而增长;两者叠加,推动 GPU 集群需求多年复合增长。IDC 报告的 2025 年云 GPU 基础设施支出同比增长 46.8%,与这一论点一致。 电力可用性是第二层结构性驱动,尤其利好 Crusoe。美国电网受限:北弗吉尼亚新数据中心电力互联排队要多年,超大规模云厂商也公开承认,在传统市场建设新园区电力需要 3–5 年交付期。Crusoe 先在竞争较弱的市场(Texas Abilene、Wyoming)锁电,把这一约束转成竞争护城河。Crusoe 已有 45+ GW 电力管线和 3.4 GW 已投运容量,在市场恢复正常前很早就锁定了绑定性资源。 NVIDIA 战略合作(2025 年 1 月宣布)带来 GB200 NVL72 和下一代 GPU 配额的优先获取权,在 GPU 供给受限环境中,这是关键差异化。Crusoe 的 PUE 为 1.2–1.3,低于行业平均 1.8(由直达芯片液冷实现),转化为更低的每 GPU-hr 能耗成本,即使现货价格下跌也能支撑有竞争力的定价。 主要约束是 GPU 现货价格压缩。H100 现货从 $8/hr 跌至 $2–3/hr,意味着现货市场单位收入下降 60–75%。第二个约束是超大规模云厂商的竞争反应:AWS、Azure 和 GCP 都在加速部署 GPU 容量。一个负面但重要的发现是:Crusoe 的 Abilene 园区获得 TCEQ 许可,可部署 360 MW 天然气涡轮;公司还聘请 Engine No. 1 开发 4.5 GW 新天然气发电。这与其清洁能源品牌相冲突,并给重视可持续性的客户带来 ESG 风险。[CM024, CM025, CM026, CM027, CM028, CM029]
| 驱动因素 / 约束 | 方向 | 时点 | 对 Crusoe 的影响 | 尽调问题 |
|---|---|---|---|---|
| AI 训练算力需求增长(扩展定律) | 正向 | 2025–2030(持续) | TAM 扩大;大型 GPU 集群预留需求上升;支撑 $26.6B GPU 云预测 | 核实 Abilene 容量的集群利用率和预留填充率;确认积压订单 |
| 推理扩展(测试时计算、思维链) | 正向 | 2025–2027(加速) | 除阶段性训练高峰外,持续 GPU 需求也能维持;改善利用率经济性 | 确认训练与推理收入拆分;不同工作负载类型的利润率差异 |
| 美国电网约束(多年并网排队) | 正向(结构性护城河) | 2025–2030(多年) | Abilene / Wyoming 的电力优先策略形成可防守的容量优势;新进入者面临 3–5 年排队 | 核实各站点已签署的并网协议和供电时间表 |
| NVIDIA 战略合作(GB200、优先分配) | 正向 | 2025–2026 | 提前获得 GB200 NVL72 机架是差异化供给;NVIDIA 背书传递优先地位 | 确认 GPU 供应管线和分配承诺;审阅 NVIDIA 合作协议 |
| GPU 现货价格压缩(H100 从 $8 → $2–3/hr) | 负向 | 2024–2026(持续) | 现货价格下跌 60–75%;任何现货市场收入敞口都会持续承压利润率 | 现货与预留合同各占收入比例;前 10 个 MSA 的合同底价 |
| 超大规模云厂商 GPU 容量扩张(AWS/Azure/GCP) | 负向 | 2025–2028 | 大型云厂商正在加速 AI GPU 可得性;企业买方切换动力下降 | Crusoe 定价与超大规模云厂商预留 GPU 价格在多年合同期内如何对比 |
| 天然气能源来源(ESG 风险) | 负向(有条件) | 持续 | Abilene 获 TCEQ 许可的 360 MW 燃气轮机与清洁能源叙事冲突;会带来 ESG 敏感客户风险 | 要求按能源来源拆分 LCOE;各园区可再生能源与天然气占比;PPA 细节 |
| 模块化 AI 工厂(Crusoe Spark)创造市场 | 正向 | 2026–2028(新兴) | 首批 Spark 单元预计 Q3 2026;把市场从云租赁扩展到企业资本开支 | Spark 的收入贡献;定价模型(资本开支销售、BOT 或租赁);销售管线 |
时点依据公开证据粗略估计。正向 / 负向表示对 Crusoe 收入机会和竞争位置的净影响。
[CM024, CM025, CM026, CM027, CM028, CM029]2.5 展示材料
03竞争格局
3.1 竞争格局概览
AI 算力基础设施竞争格局最适合分三层理解:(1)纯 neocloud GPU 云运营商,是 Crusoe 最直接的收入竞争者;(2)与 Crusoe 电力优先定位相近、同样押注能源论点的运营商;(3)超大规模云厂商(AWS、Azure、GCP),代表 Crusoe 所有服务的主导性现状替代方案。 CoreWeave 以估计 $5B ARR、NVIDIA 优先伙伴身份,以及 2025 年 3 月 IPO 建立的 neocloud 公开可比公司地位,领跑 neocloud 细分市场。Lambda Labs 面向开发者和研究市场,GPU SKU 目录深、按需定价有竞争力。Nebius AI(Yandex 分拆,2024 年 6 月起运营)以隐私和数据主权为核心定位,面向 EU 和美国市场提供 GPU 云。Lancium 在 Texas HPC 市场运营,由风能和太阳能供电,与 Crusoe 最初的能源论点和相近地理位置重合度最高。 超大规模云厂商层——AWS P4/P5、Azure NDv5 和 GCP A3 实例——代表全球最大的 GPU 算力池,也是企业买家推迟切换到 neocloud 运营商的主要原因。超大规模云厂商正在加速自建 AI 基础设施:Microsoft 承诺 2025 年 AI 基础设施 capex 超过 $80B,AWS 和 GCP 也在类似增加数据中心部署。这些投资直接挑战了当前吸引企业买家转向 neocloud 运营商的套利机会。 一个关键信号偏负面:CoreWeave 2025 年 3 月 IPO 虽然验证了 neocloud 类别,也引入了公开市场纪律和直接估值基准。CoreWeave IPO 首日约以 5x 收入交易——若套用到 Crusoe 的 $124M AI 云收入基数,意味着 Crusoe 云业务单独估值低于当前 $10B 总企业估值,说明 Crusoe 的估值部分依赖基础设施(土地、电力、工厂)层,而非纯云收入倍数。[CP001, CP002, CP003, CP004, CP005, CP006]
Crusoe 处在高整合、中等规模的独特位置;CoreWeave 在 GPU 云规模上领先,但缺少垂直整合;超大规模云厂商规模占优,但没有基础设施垂直整合。
评分是有证据支持的序数判断(0–10),来自公开产品界面、市场份额数据和基础设施所有权披露。并非基于厂商直接报告的指标。
[CP001, CP002, CP003, CP004, CP005, CP006]3.2 主要竞争对手画像
CoreWeave 是最大的 neocloud GPU 云运营商,截至 2025 年初 ARR 估计约 $5B。CoreWeave 创立于 2017 年,总部位于 New Jersey Roseland,是 NVIDIA 最偏好的大规模 GPU 云伙伴,优先获得 H100、H200 和 Blackwell(B200/GB200)GPU 配额。CoreWeave 于 2023 年中完成 $1.1B Series C,并于 2025 年 3 月在 NASDAQ 上市。其客户群偏向前沿 AI 实验室和需要大规模 GPU 集群的企业 AI 平台公司,直接重叠 Crusoe 最高层买家。相对 Crusoe,CoreWeave 的弱点在于不自有发电,也不在园区层面自建数据中心;它依赖第三方托管和电力合同,市场收紧时更容易暴露在电力可用性约束下。 Lambda Labs(2012 年创立,San Francisco, CA)面向开发者和研究 GPU 云细分市场,提供丰富的 NVIDIA GPU SKU(H100、A100、A6000)和有竞争力的按需定价。Lambda 的差异化在于简单:开通快、无最低用量、工具对开发者友好。Lambda 没有 Crusoe 或 CoreWeave 那样的基础设施垂直整合,且聚焦较小规模工作负载。在 Crusoe 正在建设的超大规模云厂商承诺或专用园区层面,Lambda 不是实质性竞争者。 Nebius AI(Amsterdam;NYSE: NBIS)于 2024 年 6 月从 Yandex 分拆,运营 EU 和美国 GPU 云基础设施,面向需要数据主权和云灵活性的 AI 公司。Nebius 于 2024 年 9 月从 NVIDIA 及 NVIDIA 生态伙伴等投资者处融资 $700M。Nebius 的 GPU 云主要在 EU 市场竞争,数据驻留是该市场的监管驱动因素,因此与 Crusoe 以美国为中心的园区建设策略直接重叠有限。 Lancium(Houston TX)在 Texas 运营可再生能源驱动的 HPC 数据中心,使用风能和太阳能,并具备需求响应能力,可在电网压力事件期间削减算力负载。Lancium 的能源优先定位和 Texas 地理位置与 Crusoe 最初论点直接重合,但 Lancium 不提供通用 GPU 云,而是聚焦 HPC 批处理算力。Lancium 已于 2024 年申请 IPO,但截至报告日期尚未完成。[CP008, CP009, CP010, CP011, CP012, CP013]
| 竞争对手 | 类别 | 规模 / 融资 | 目标细分市场 | 差异化 | 相对 Crusoe 的限制 |
|---|---|---|---|---|---|
| CoreWeave | GPU 新云厂商(纯 GPU 云) | ~$5B ARR;2025 年 3 月 IPO(NASDAQ: CRWV);2023 年 $1.1B Series C 轮 | 前沿 AI 实验室、企业 AI 平台、大型 GPU 集群预留 | NVIDIA 优先合作伙伴;收入规模最大的 GPU 新云厂商;上市公司更容易获得资本 | 没有电力 / 园区垂直整合;依赖第三方托管机房;基础设施护城河较弱 |
| Lambda Labs | GPU 新云厂商(开发者 / 研究导向) | ~$320M 融资(估计);私营;San Francisco CA | 开发者、研究人员、ML 从业者、中小 AI 团队 | GPU SKU 目录深;开通快;无最低用量;定价和 UX 对开发者友好 | 没有超大规模云厂商级合同;没有能源整合;主要面向按需 / 现货市场;规模小 |
| Nebius AI | GPU 新云厂商(EU/US 数据主权导向) | $700M 融资(2024 年 9 月);NYSE: NBIS;Yandex 拆分 | EU AI 公司;数据主权工作负载;全球扩张中的美国企业 | 数据驻留 / 主权定位;NVIDIA 支持;已有 EU 基础设施 | 相比 Crusoe,美国足迹有限;偏 EU;在美国披露的企业标杆客户有限 |
| Lancium | 能源逻辑 HPC 运营商(Texas) | 私营;未披露 Series B+ 轮;Houston TX;2024 年已提交 IPO | HPC 批处理计算;气候科技客户;风 / 光需求响应 HPC | 纯风 / 光能源来源;熟悉 Texas 市场;具备需求响应电网灵活性 | 没有 GPU 云产品;仅 HPC 批处理;GPU 集群预留能力有限;市场焦点窄 |
| AWS (P4/P5 EC2) | 超大规模云 GPU 云(替代品) | Amazon;$100B+ 云收入;全球基础设施 | 所有企业 AI 买方;客户基础广;既有 AWS 账号持有者 | 全球覆盖;既有企业关系;SLA 成熟;服务集成广 | GPU 密度低于 GPU 新云厂商;电力可得性受限;按需价格较高 |
| Azure AI (NDv5) | 超大规模云 GPU 云(替代品) | Microsoft;$100B+ 云收入;2025 年 AI 资本开支 $80B+ | 企业 AI 团队;Microsoft 生态客户;OpenAI 合作伙伴 | Microsoft 365/Azure 生态;OpenAI API 集成;企业合规;SLA 覆盖广 | 与 AWS 一样受可得性约束;Microsoft 同时也是 Crusoe Abilene 容量客户 |
| GCP (A3/H100) | 超大规模云 GPU 云(替代品) | Google Cloud;~$40B ARR;全球基础设施 | 企业 AI;Google 生态;偏好 TPU 的工作负载 | TPU/GPU 混合供给;Google DeepMind 研究公信力;Vertex AI 平台 | GPU 可得性受限;TPU 锁定风险;在纯 GPU 预留市场不如对手强势 |
规模和融资数字基于截至 2026 年 5 月的公开披露和分析师估计。ARR 和收入数字为近似值。超大规模云厂商的 GPU 收入没有公开拆分。
[CP001, CP002, CP003, CP004, CP005, CP006]| 购买标准 | Crusoe | CoreWeave | Lambda Labs | Nebius AI | Lancium | AWS / Azure / GCP |
|---|---|---|---|---|---|---|
| GPU 可得性(H100/H200/B200/GB200) | 强 — GB200 NVL72、B200、H200、H100、A100;NVIDIA 优先分配 | 强 — NVIDIA 优先;H100、H200、A100 规模化供给;GB200 爬坡 | 中 — H100、A100、A6000 目录;未确认可获得 GB200 NVL72 | 中 — H100、A100;拆分后获 NVIDIA 支持;库存聚焦 EU | 未知 — HPC GPU,不以 GPU 云为主;公开 GPU 目录有限 | 强(AWS/Azure)/ 中(GCP)— 主要 GPU 均可用,但供给受限且价格更高 |
| 大型 GPU 集群规模(>1,000 GPU) | 强 — 2.1 GW Abilene 园区;为大型集群预留而设计 | 强 — 具备数千 GPU 集群能力;核心产品聚焦于此 | 弱 — 主要是较小集群;不是为 1,000+ GPU 专用预留而设计 | 中 — EU 基础设施;披露的美国大型集群容量有限 | 弱 — HPC 批处理,不是大型 GPU 集群预留产品 | 强 — 超大规模云厂商在全球拥有最多 GPU 绝对数量 |
| 电力 / 能源垂直整合 | 强 — 能源获取、园区建设、GPU 云在同一技术栈中 | 弱 — 依赖第三方托管机房;不拥有电力 | 弱 — 标准数据中心租约;没有能源垂直整合 | 弱 — 第三方数据中心容量;不拥有能源 | 强 — 在 Texas 拥有风 / 光电力;具备需求响应能力 | 弱 — 使用第三方数据中心的公用事业电力;不拥有电力 |
| 预留 / 长期合同灵活性 | 强 — 多年期超大规模云厂商合同;预留 GPU 层级;Spark BOT 选项 | 强 — 可提供长期预留合同;这是主要商业模式 | 弱 — 主要按需;规模化多年预留产品有限 | 中 — 可提供预留实例定价;聚焦 EU 市场 | 未知 — HPC 合同;没有公开 GPU 预留价格 | 强 — 1 年 / 3 年预留实例定价,折扣显著 |
| 合规认证(SOC2、ISO 27001) | 强 — SOC2 Type I+II、ISO 27001、ISO 42001、GDPR | 强 — SOC2 Type II、ISO 27001;企业级合规 | 中 — SOC2 Type II;公开合规披露有限 | 中 — 聚焦 EU GDPR 合规;披露的美国特定认证有限 | 未知 — 可用资料未记录公开合规认证 | 强 — 主要认证齐全(SOC2、ISO 27001、FedRAMP、HIPAA) |
| 模块化 / 预制 AI 工厂产品 | 强 — Crusoe Spark;Brighton CO 工厂;首批单元 Q3 2026 | 未知 — 没有公开模块化工厂产品;仅定制建设 | 弱 — 没有模块化工厂产品 | 弱 — 没有模块化工厂产品 | 弱 — 没有模块化工厂产品 | 弱 — 没有面向企业采购的模块化 / 预制 AI 工厂产品 |
| 定价竞争力(GPU-hr) | 中 — H100 SXM 标价 $4.29/hr;相对 $2-3/hr 现货价有溢价 | 未知 — 定价未公开列示;合同单独谈判 | 强 — 现货定价有竞争力;开发者友好;无最低用量 | 中 — 预留定价有竞争力;聚焦 EU 市场 | 未知 — HPC 定价模型;不适合按 GPU-hr 对比 | 中-强 — H100 按需约 ~$3-5/hr(AWS);承诺用量后预留价格更低 |
标记为“未知”的单元格表示公开证据不足,无法给能力打分;这不是说没有能力,而是缺少已披露证据。不要把“未知”解读为弱项。矩阵基于截至 2026 年 5 月的公开产品页面、定价披露和新闻稿。
[CP008, CP009, CP010, CP011, CP012, CP013]Crusoe 在垂直整合和下一代 GPU 可用性上领先;CoreWeave 在云规模和 NVIDIA 合作深度上领先;超大规模云厂商在服务广度和企业合规成熟度上领先。
标记为“未知”的单元格表示公开证据不足,无法有把握地评估该能力。矩阵基于截至 2026 年 5 月的公开产品页面、定价披露和新闻稿。
[CP008, CP009, CP010, CP011, CP012, CP013]3.3 Crusoe 差异化与护城河
Crusoe 的主要差异化主张是垂直整合:公司大规模锁定电力、建设面向 AI 优化的园区、制造模块化 AI 工厂(Crusoe Spark),并在一套整合栈上运营 GPU 云。这种整合把从拿地到 GPU 上线的周期压缩到宣称的绿地园区少于 12 个月,而行业常态是 3–5 年。目前没有其他 neocloud 运营商复制了这套完整栈。 NVIDIA 战略合作(2025 年 1 月)是第二层关键护城河:在供给受限市场中优先获得 GB200 NVL72 GPU 配额,使 Crusoe 能在竞争对手拿到同等库存前,把最新 GPU 硬件卖给前沿客户,从而形成有意义的差异化。CoreWeave 也与 NVIDIA 关系很强,但 Crusoe 的合作对外更明确地记录为直接战略伙伴关系,而不是大额采购协议。 模块化 AI 工厂(Crusoe Spark)产品是一条正在形成的护城河,有机会创造新市场类别:企业购买专用 AI 算力基础设施作为资本设备,而不是租用云容量。目前没有 neocloud 竞争对手在规模上提供可比的模块化预制数据中心产品。Colorado Brighton 的 Spark Factory(350,000 sq ft,$200M 投资)若能规模化制造模块化单元,短期内很难复制。 能源护城河真实存在,但已部分受损:Crusoe 的 PUE 为 1.2–1.3,低于行业平均 1.8,是实打实的运营优势。不过,Heatmap News 的负面发现削弱了清洁能源差异化主张:Crusoe 在 Abilene 使用 360 MW 现场天然气涡轮(约占一期 30%),并聘请 Engine No. 1 开发 4.5 GW 新天然气发电。对重视 ESG 的客户而言,Lancium 的纯风能 / 太阳能定位更容易防守。[CP019, CP020, CP021, CP022, CP023, CP024]
| 竞争对手 | 价格 / 单位 / 合同模型 | 包含能力 | 折扣或未知项 | 对 Crusoe 的影响 |
|---|---|---|---|---|
| Crusoe | H100 SXM $4.29/GPU-hr;A100 SXM $1.95/GPU-hr;CPU $0.04/vCPU-hr;预留和按需 | 网络、存储、可观测性平台;托管推理;包含 SOC2/ISO 认证 | 长期合同价格未披露;超大规模云厂商承诺量单独谈判;Spark 定价未知 | 相对现货市场有标价溢价;必须靠可靠性、可得性和预留合同价值取胜 |
| CoreWeave | 未公开列示;根据市场报告估计 H100 为 $2–5/GPU-hr;合同单独谈判 | 全集群网络、InfiniBand 网络架构、托管 Kubernetes、企业 SLA | 可能有显著大客户量价折扣;NVIDIA 优先带来的定价优势未知 | CoreWeave 规模(约 ~$5B ARR)意味着它有能力激进定价,并以量价压低 Crusoe |
| Lambda Labs | H100 SXM $2.49/GPU-hr(按需);现货定价有竞争力;无最低用量 | 基础计算;网络;开发者 API;未披露企业 SLA | 未知量价折扣;没有规模化长期预留产品 | Lambda 的低价切走现货市场开发者客户,而 Crusoe 并不直接瞄准这些客户 |
| Nebius AI | H100 SXM 起价约 ~$2.50/GPU-hr(EU 市场);可提供预留定价 | GPU 云带网络;聚焦隐私 / 合规;EU 数据驻留 | 美国定价未公开披露;可能不同于 EU 市场费率 | 与 Crusoe 以美国为中心的大集群市场直接重叠有限;不是主要价格威胁 |
| Lancium | HPC 订阅模型;不适合按 GPU-hr 对比 | 可再生能源供电 HPC 计算;需求响应灵活性;仅 Texas | 不适合直接与 GPU-hr 定价对比;HPC 批处理模型 | 在 Texas 能源市场形成间接竞争;不是 GPU 云定价威胁 |
| AWS (P4/P5) | H100 按需约 ~$3.00–5.00/GPU-hr(p5.48xlarge);1 年 / 3 年预留折扣 20–40% | AWS 服务全集成;全球 SLA;企业合规;既有账号 | 企业折扣计划;对大客户而言,EDP 承诺会显著降低有效费率 | AWS 面向大客户的企业折扣会缩小其与 Crusoe 预留价格的有效成本差距 |
| Azure (NDv5) | H100 按需约 ~$3.60/GPU-hr(ND H100 v5);预留实例有折扣 | Microsoft 365 / Azure 生态;企业合规;Azure OpenAI API 访问 | EA 客户折扣显著;Microsoft 与 Crusoe 围绕 Abilene 有直接关系,降低冲突 | Azure 的企业定价和 Microsoft 关系,削弱了既有 Azure 客户单纯做价格套利的空间 |
定价基于截至 2026 年 5 月的公开定价页面和市场报告。大型企业和超大规模云厂商协议的协商合同价没有公开披露。除非另有说明,所有 GPU-hr 费率均为 H100 SXM 等价口径。
[CP019, CP020, CP021, CP022, CP023, CP024]Crusoe 最强护城河来自纵向一体化深度和下一代 GPU 获取能力;最实质性竞争风险是 CoreWeave 的规模优势,以及超大规模云厂商加速扩充 GPU 容量。
CoreWeave ARR 和 IPO 倍数是基于公开报道的分析师估计。Crusoe 收入和增长为公司披露的未经审计指标。纵向一体化评分是有证据支撑的等级判断。
[CP001, CP002, CP008, CP019, CP020, CP025]3.4 竞争风险与替代场景
最高强度的竞争风险,是超大规模云厂商加速 GPU 容量。Microsoft 2025 年承诺 $80B+ AI 基础设施 capex,AWS 和 GCP 也同步建设,都会直接增加主导云平台上的 GPU 可用性,削弱当前推动企业客户转向 neocloud 的可用性套利。随着超大规模云厂商 GPU 预留价格下降、交付期缩短,企业使用 Crusoe Cloud 的切换成本会显著上升。 第二个风险是 CoreWeave 受 IPO 推动的资本获取能力和 NVIDIA 优先身份。作为上市公司,CoreWeave 能以低于私营 Crusoe 的成本融资,加快数据中心建设,并用可流通股权薪酬吸引人才。CoreWeave IPO 时估计 $5B ARR,而 Crusoe 2024 年 AI 云收入为 $124M,规模差距很大;如果 CoreWeave 执行激进,差距可能继续扩大。 第三个风险是 GPU 现货价格压缩引发商品化竞赛。如果 H100 现货价格继续下跌,预留价格也趋同,支撑 neocloud 溢价定价的结构性价格套利就会被侵蚀。Crusoe 的能源成本优势(PUE 1.2–1.3 vs 行业平均 1.8)提供了结构性成本底,但优势幅度——每 GPU-hr 能源成本约低 30–40%——可能不足以抵消广泛的预留价格趋同。 Forbes(2025 年 3 月)关于客户可靠性的负面发现很重要:2025 年 3 月初一次 45 小时故障影响企业客户,并引发客户询问可靠性问题。这符合快速扩张基础设施公司的运营风险。Crusoe 获得 ISO 27001 和 SOC2 Type I+II 认证,说明其投入治理;但 GW 级园区部署的运营可靠性,尚未在多年生产窗口中被证明。[CP027, CP028, CP029, CP030, CP031, CP032]
| 护城河主张 | 威胁 | 严重性 | 缓解措施 / 尽调问题 |
|---|---|---|---|
| 垂直整合(电力 + 园区 + GPU 云) | CoreWeave 收购或联手电力优先的基础设施运营商;超大规模云厂商自建大规模园区 | 中 | 核实 Crusoe 的电力协议是否长期、排他且可转让;确认园区建设执行记录 |
| NVIDIA 优先合作伙伴地位(GB200 分配优先) | CoreWeave 拥有同等或更强的 NVIDIA 关系;GB200 随着产能爬坡而供给常态化 | 中 | 审阅 NVIDIA 合作协议条款;以书面形式确认供应分配承诺;评估 GB200 交付时间表与 CoreWeave 的差异 |
| Crusoe Spark 模块化 AI 工厂(暂无可比产品) | 大型数据中心建设商(如 Vantage、Equinix)或超大规模云厂商开发竞争性预制方案;Spark 时间线延误 | 低-中 | 索取 Spark 管线和已签 LOI;确认 Brighton 工厂爬坡时间线和产能;审阅 IP 保护 |
| PUE 1.2–1.3 能效优势 | 超大规模云厂商投资同等液冷;竞争对手采用同样的直通芯片液冷技术;PUE 优势商品化 | 中 | 通过 TCEQ 环境数据或第三方审计独立核实 PUE;确认液冷合同和 IP |
| 45+ GW 电力管线(长期供应护城河) | 管线只是非约束性 LOI;公用事业公司重新排定并网优先级;宏观电网扩张降低稀缺溢价 | 高 | 索取已签并网协议与 LOI 的副本;阶段性交付时间线;交易对手信用质量 |
| 客户集中风险(Microsoft 900 MW 占主导) | 如果 Microsoft 内部自建提速,可能减少 Abilene 配额;提前终止或重新谈判 | 高 | 审阅 Microsoft 合同条款,包括最低照付不议、终止条款和价格阶梯下调;评估 Microsoft 自身的 Abilene 建设计划 |
| 燃气发电受审视下的 ESG / 清洁能源定位 | 监管或客户 ESG 压力迫使燃气轮机提前退役;TCEQ 许可遭挑战;客户流失 | 中 | 核实各园区实际能源结构;索取用于可再生能源替代的远期 PPA 承诺;评估 Engine No. 1 天然气开发时间表 |
严重程度是基于截至 2026 年 5 月竞争情报的分析师判断。高严重程度表示某项风险可能在 12–24 个月内实质损害 Crusoe 的估值或商业模式。
[CP025, CP026, CP027, CP028, CP029, CP030]3.5 展示材料
04财务情况
4.1 收入模式与收入流
2023 年以来,Crusoe 的收入模式经历了有意的结构性转型。2024 年,公司总收入 $276M,大致分为 45% AI 云($124M,同比增长 460%)和 55% Bitcoin 及数字火炬挖矿($152M)。2025 年 3 月,公司将挖矿资产出售给 NYDIG——425+ 个模块化数据中心、250+ MW 和约 135 名员工——之后,Crusoe 转向纯 AI 云和数据中心基础设施收入模式。公司披露 2025 年云 ARR 增长 150%、合同总价值(TCV)增长 17 倍,但未公布绝对分母或基础 ARR 数字。分析师估计 2025 全年收入介于 $500M 和 $1B 之间,但尚未确认。 今天,Crusoe 的收入流由三套不同机制驱动。第一,Crusoe Cloud 通过按需小时计费和预留容量订阅提供 GPU 与 CPU 算力;H100 SXM 标价 $4.29/GPU-hour,A100 SXM 标价 $1.95/GPU-hour。第二,长期超大规模云厂商容量协议——尤其是 900 MW Microsoft Abilene 承诺(2026 年 3 月)和 Oracle/Stargate 锚定关系——提供大额合同收入块,并带来多年可见度。第三,Crusoe Spark 模块化 AI 工厂产品(每个单元约 1 MW 预制容量,在 350,000 sq ft Colorado Brighton Spark Factory 制造)可能形成产品销售和托管服务收入线;Redwood Materials 合作从 4 个 Spark 单元扩大到 24 个,显示早期商业牵引。 NYDIG 交易后,收入质量明显改善:AI 云收入具备经常性和订阅属性,而旧 BTC 挖矿收入本质上是商品挖矿,受减半周期和算力竞争影响。不过,Microsoft、Oracle 等超大规模云厂商集中度高,引入客户依赖风险,也限制定价权。 [CI001, CI002, CI003, CI004, CI005, CI010]
| 收入来源 | 机制 | 单位 | 当前数值 / 状态 | 收入质量 | 关键尽调问题 |
|---|---|---|---|---|---|
| Crusoe Cloud GPU/CPU 算力 | 按需与预留 GPU-hour 计费 | $/GPU-hr | $124M AI 云(2024);估计 2025 Q1-Q2 收入运行率 $186M+ | 经常性;预留交易提升可见度 | 实际利用率、预留与按需组合、折扣深度 |
| 长期超大规模云厂商容量协议 | 多年期签约容量费(MW 级) | $/MW 或 $/GPU-cluster-yr | Microsoft 900 MW Abilene;Oracle Stargate 锚定客户;仅 Abilene 估计 $250M/yr | 高质量;已签约,交易对手为投资级 | 合同条款、价格递增条款、终止条款、照付不议结构 |
| Crusoe Spark 模块化 AI 工厂 | 预制单元销售 + 托管服务 | $/unit(每个约 1 MW) | 尚未形成规模化收入;Redwood Materials 已部署 4→24 个单元 | 早期;一次性收入叠加经常性托管运营 | 单元 ASP、积压订单、制造毛利率、托管服务附加率 |
| Bitcoin 与 DFM 挖矿(已剥离) | 区块奖励 + 火炬气费用 | BTC/day、$USD/MMBtu(计费单位) | $152M(2024);2025 年 3 月售予 NYDIG;未来贡献为零 | 已剔除;未来不再贡献 | 确认 NYDIG 交易所得及任何或有对价结构 |
2024 年数字由公司披露,但未经审计。2025 年和 Abilene 收入为分析师估算,或公司指引目标。收入来源的单位反映主要计费机制,而非合同结构。
[CI001, CI002, CI003, CI010, CI011, CI012]客户参与如何流经 Crusoe 的三条收入流,并汇总为总收入和(未披露的)毛利。
收入数字反映 2024 年公司披露总额。2025 和 2026 年收入数字为分析师估计或公司指示性目标。毛利未披露;估计区间来自 GPU 云新云同行基准。
[CI001, CI002, CI010, CI011, CI020, CI021]4.2 单位经济与定价
Crusoe 公开披露的定价只是标价;每 GPU-hour 实际收入、预留容量折扣结构和 GPU 利用率均未披露。H100 SXM 标价 $4.29/GPU-hour,A100 SXM 标价 $1.95/GPU-hour,使 Crusoe 位于 neocloud 现货报价的高端,但低于主要超大规模云厂商的按需费率。GB200 NVL72 价格未披露。CPU 算力标价为 $0.04/vCPU-hour。 单位经济所处市场环境偏不利:H100 现货市场价格从 2023 年约 $8/hour 跌至 2024 年末 $2–3/hour,背后是供给正常化,以及 CoreWeave、Lambda Labs、Nebius 等 neocloud 之间的集群竞争。价格压缩迫使 Crusoe——也包括所有 GPU 云运营商——不能只靠现货定价竞争,而要靠基础设施效率、产品差异化(MemoryAlloy 推理、Command Center、Edge Zones)和企业粘性合同。 成本侧的关键单位经济输入,公开信息只提供汇总代理指标。PUE 为 1.2–1.3,而行业平均为 1.8,意味着明显能源成本优势,电力开销比典型超大规模数据中心低约 30–40%。若估计混合电价为 $0.03–0.06/kWh,H100 TDP 约 700W,则每 H100 每小时电力成本约 $0.02–0.04——只是 $4.29 标价的一小部分。GPU 资本折旧才是主导成本:H100 单价约 $25,000–30,000;按 3–5 年折旧、80% 利用率测算,每 GPU-hour 资本成本约 $0.60–1.20,仍留下有意义的毛利空间。GPU 云基础设施细分市场中,同行 neocloud 公司披露毛利率在 20–40% 区间,Crusoe 的效率优势意味着它可能位于高端,但公司未披露毛利率数据。 [CI006, CI007, CI008, CI009, CI021, CI022]
| SKU / 产品 | 标价 | 合同类型 | 标价 vs. 实收 | 折扣 / 未知项 | 来源 |
|---|---|---|---|---|---|
| H100 SXM GPU | $4.29/GPU-hr | 按需或预留 | 标价;大型预留交易的实收大概率更低 | 批量折扣结构未披露 | crusoe.ai/cloud/pricing(定价页) |
| A100 SXM GPU | $1.95/GPU-hr | 按需或预留 | 标价;现货市场 $2–3/hr 意味着 A100 接近同价 | 现货市场价格压缩限制相对市场溢价 | crusoe.ai/cloud/pricing(定价页) |
| CPU vCPU 算力 | $0.04/vCPU-hr | 按需 | 标价 | 很小;商品化定价 | crusoe.ai/cloud/pricing(定价页) |
| GB200 NVL72 集群 | 未公开披露 | 可能为预留或企业合同 | Unknown | 未列价;仅企业交易 | docs.crusoecloud.com、nvidianews.nvidia.com(文档 / 新闻稿) |
| AMD MI355x / MI300x | 未公开披露 | 可能为预留 | Unknown | 未列价;定价可能与 H100 有竞争力 | crusoe.ai/cloud(云产品页) |
| Crusoe Spark 单元(1 MW 预制) | 未公开披露 | 资本开支型销售 + 托管服务 | 未知;大型资本开支项目 | ASP 与托管服务定价未披露 | crusoe.ai/manufacturing;合作伙伴公告 |
所有标价均来自 crusoe.ai/cloud/pricing,截至 2026 年 5 月。GB200 NVL72 和 Spark 单元定价未公开列示。每 GPU-hour 实收收入和企业折扣结构尚未披露。
[CI006, CI007, CI008, CI025]| 指标 | 数值 / 估计 | 置信度 | 重要性 | 关键尽调问题 |
|---|---|---|---|---|
| 毛利率 — GPU 云 | 未披露;同业区间 20–40% | 低 | NYDIG 出售后主要盈利驱动因素 | 索取经审计 P&L 或管理账;取得 GPU/DC 成本明细 |
| GPU 利用率 | 未披露;目标估计 >80% | 低 | 利用率决定实收收入相对装机容量的兑现 | 确认 H100/A100/GB200 机群平均利用率 |
| H100 每小时实收收入 | 估计 $2.50–4.00(标价 $4.29) | 低 | 反映实际定价权能否顶住现货市场压力 | 从客户发票或队列数据索取实收 ASP |
| H100 单卡 capex 成本 | 估计 ~$25,000–30,000 | 中 | 资本化销货成本中最大的一块 | 确认 NVIDIA 供货协议价格和采购量承诺 |
| 电力成本(混合 $/kWh) | 估计 $0.03–0.06;PUE 1.2–1.3 | 中 | 能源 COGS;相对行业 PUE 1.8,Crusoe 有优势 | 确认能源采购合同和各站点混合电价 |
| GPU 每小时折旧成本 | 估计 $0.60–1.20/hr(80% 利用率,4 年) | 低 | COGS 最大组成;折旧政策很关键 | 确认使用寿命假设和折旧计划 |
| 净留存率(NRR) | 未披露 | 低 | 关键 SaaS / 云质量指标;扩张与流失信号 | 按季度披露 Crusoe Cloud 队列的 NRR 和 GRR |
| CAC / 回本周期 | 未披露 | 低 | 云业务销售效率;支撑 LTV/CAC 判断 | 按细分和交易规模披露云 CAC 与回本周期 |
除 PUE 和 GPU 标价外,所有数值均来自公开 neocloud 基准的估计或推断;毛利率、NRR、利用率和 CAC 属于私有信息,需要在 data room 披露。
[CI006, CI007, CI008, CI021, CI022, CI038]决定 Crusoe 单 GPU-小时收入实现和估计毛利率贡献的关键输入与驱动因素。
GPU 成本、折旧计划、利用率和折扣深度均参考行业基准和新云同行数据估计。Crusoe 未披露毛利率;估计区间反映 CoreWeave、Lambda Labs 和 Nebius 的同行披露。
[CI006, CI007, CI022, CI025, CI026, CI038]4.3 资本结构与充足性
对一家 AI 基础设施领域的私营公司来说,Crusoe 搭起了异常庞大且分层的资本结构。股权融资历史横跨六轮,总计约 $3.4B;最近两轮——Series D($600M,估值 $2.8B,SEC Form D 于 2024 年 11 月 21 日提交)和 Series E($1.375B,估值 $10B+,SEC Form D 于 2025 年 10 月 23 日提交,精确金额 $1,374,999,988,71 名投资者)——分别由 Founders Fund 和 Valor/Mubadala 领投。除股权外,公司为 Abilene 数据中心建设拿到 $3.4B Blue Owl Capital 合资项目,JPMorgan 还在为同一园区安排约 $9.6B 项目融资,使已承诺总资本超过 $15B。 资本结构反映了 AI 超大规模数据中心开发的底层经济:每 1 GW AI 优化容量,在安装任何 GPU 机架前,就需要约 $10–15M/MW 投入土建、电力基础设施和冷却基础设施。仅 2.1 GW Abilene 园区,多年总建设成本估计就达 $12B–20B。如此规模的项目融资需要机构交易对手、来自投资级客户(Microsoft、Oracle)的长期合同收入,以及可预测的电力合同——这些 Crusoe 已经锁定或正在积极推进。 短期运营的资本充足性看起来稳健:$1.375B 股权 + 项目融资跑道,应能覆盖运营和 Abilene 一期建设。Michael Gordon 于 2025 年 12 月出任 COO/CFO(前 MongoDB CFO,曾主导 2017 年 IPO),说明公司在主动准备潜在资本市场动作。关键风险包括项目融资契约暴露、capex 集中在单一园区,以及 JPMorgan 信贷额度提款时点带来的执行依赖。现金头寸或月度烧钱速度未公开披露。 [CI013, CI014, CI015, CI016, CI017, CI018]
| 类别 | 金额 / 状态 | 日期 / 来源 | 计划用途 / 目的 | 关键风险 / 未知项 |
|---|---|---|---|---|
| Series E 股权融资 | $1.375B | 2025 年 10 月;SEC Form D | 一般公司用途;Abilene capex;GPU 采购;营运资本 | 现金投放节奏;受限与非受限资金分配未知 |
| Blue Owl Capital 合资公司 | $3.4B 已承诺 | 2024/2025;公司公告 | Abilene 数据中心建设(土建、MEP、电力基础设施) | 提款进度、JV 内股债拆分、完工担保 |
| JPMorgan 项目融资 | ~$9.6B 安排中 | 2025;公司公告 | Abilene 园区长期项目融资 | 信贷额度提款条件、约束条款、LTV 要求 |
| 承诺资本总额 | >$15B | 2025/2026 汇总 | Abilene Phase 1 + Phase 2 + 额外园区建设 | 资金已承诺但未提款;存在执行和提款风险 |
| 在手现金 | 未披露 | 运营流动性和 capex 过桥 | 必须在尽调中索取;对烧钱速度 / 现金跑道评估重要 | |
| 月度烧钱 | 未披露 | 1,235 人组织的运营现金跑道信号 | 索取最近 3 个月银行流水或 CFO 过桥模型 |
股权和融资数字来自 SEC Form D 文件及公司新闻稿。在手现金和烧钱速度没有公开数据。项目融资金额是安排额度,不是已提款余额。
[CI013, CI014, CI015, CI017, CI018, CI019]Crusoe Abilene 园区建设的已承诺资本来源与用途,展示大规模 AI 数据中心开发的资本强度。
资本数字来自 SEC Form D 文件和公司公告。资本开支估计来自超大规模 AI 数据中心建设 $10M–$15M/MW 的行业基准。GPU 采购成本根据 NVIDIA H100/GB200 定价基准估计。所有数值均以 USD 百万美元计。
[CI013, CI014, CI017, CI018, CI019, CI026]4.4 财务缺口与风险评估
Crusoe 最重要的财务承保缺口,是缺少毛利率、EBITDA、现金头寸、烧钱速度、净收入留存和审计财务报表。作为不受 SEC 报告义务约束的私营公司,Crusoe 没有义务披露这些指标。MongoDB 老将 Michael Gordon 出任 CFO,释放出财务治理改善信号,但独立验证仍不可得。 风险侧有四个重大负面因素需要重点审查。第一,GPU 现货定价压缩——H100 约从 $8/hour 跌至 $2–3/hour——形成结构性价格上限,限制按需云收入上行。Crusoe 能否把客户锁进高于现货的标价预留容量协议,是尚未验证的关键价值驱动因素。第二,清洁能源财务风险:Crusoe 的 ESG 品牌可能从重视可持续性的企业获得溢价定价,但 Heatmap News(2025 年 3 月)报道公司继续使用天然气,包括与 Engine No. 1 成立 JV 开发 4.5 GW 新天然气容量;TCEQ 许可确认 Abilene 有 360 MW 燃气涡轮。如果重视 ESG 的客户调查实际能源栈,流失风险和声誉损害可能把实际定价压到标价以下。第三,Microsoft 和 Oracle 客户集中造成收入依赖风险:合同变更或超大规模云厂商决定将基础设施内包,都可能显著影响收入。第四,高 capex 模式叠加 $9.6B 项目融资,在 JPMorgan 规模上带来契约和再融资风险。没有公开契约条款、LTV 比率或债务服务覆盖要求可供尽调审阅。 公司 2025 年 TCV 增长 17 倍、云 ARR 增长 150%,是积极信号;但私营财务透明度不足——尤其是公司从挖矿转向云时的毛利率轨迹——意味着没有利润率可见度的收入增长很难承保。Spark 制造收入流尚未实现规模化收入(首批单元 2026 年 Q3),公开模型也尚未纳入。 [CI021, CI024, CI032, CI033, CI034, CI035]
| 缺失指标 | 对投资测算的影响 | 为何未公开 | 具体尽调路径 |
|---|---|---|---|
| 毛利率(GPU 云) | 无法测算盈利轨迹或估值倍数 | 私营公司;无披露义务 | 索取 data room P&L;另行索取 GPU 收入和 DC 层面 COGS |
| 毛利率(Spark 制造) | 新业务;没有内部数据就缺少可比基准 | 尚未形成规模化收入;历史财务中没有体现 | 向 CFO 索取单元毛利模型;与超大规模数据中心建设毛利率对比 |
| 现金状况 / 资产负债表 | 现金跑道和流动性风险无法量化 | 私有信息;未公开披露 | 索取最近一期经审计资产负债表或 CFO 编制的明细表 |
| 月度 / 季度烧钱速度 | 无法评估运营现金跑道或资本效率 | 私有信息 | 向 CFO Michael Gordon 索取 12 个月现金流量表 |
| 净留存率(NRR) | 无法评估云业务粘性、扩张和流失 | 标准 SaaS 指标;私营公司未披露 | 向 Erwan Menard(SVP Product)索取按队列季度划分的 NRR/GRR |
| 总合同价值(TCV)积压订单 | 没有基数和期限,17x TCV 增长意义不大 | 公司只披露增长率,未披露基数或绝对值 | 在 data room 索取按客户、期限和 $ 金额划分的合同积压 |
| 经审计财务报表 | 所有自报指标都未经验证 | 私营公司没有 SEC 报告义务 | 要求 data room 提供 CPA 审计 F/S(2023、2024);如没有,索取四大 QoE |
| NYDIG 交易对价 | 未知所得会影响剥离资产后的资本状况 | M&A 交易条款通常保密 | 在 data room 索取交易协议和任何或有对价条款 |
本表所有项目截至 2026 年 5 月均无公开信息。考虑到业务资本密集,毛利率、现金状况和经审计财务的影响最重。
[CI021, CI032, CI033, CI040, CI045]基于情景的 2025 年 Crusoe 收入估计,覆盖由 GPU 定价、超大规模云厂商爬坡速度和 Spark 采用驱动的悲观、基准和乐观情景。
所有收入估计均由分析师推导或为公司指示性目标。Crusoe 尚未公开披露 2025 或 2026 年财务指引。基准情景与公司所称云 ARR 增长 150% 对齐;悲观 / 乐观边界反映 GPU 定价底部和超大规模云厂商爬坡不确定性。所有数值均以 USD 百万美元计。
[CI002, CI004, CI005, CI029]4.5 展示材料
05产品与技术
5.1 硬件平台与模块化 AI 基础设施
Crusoe 的底层硬件单元是 Modular AI Factory(MAF):一个自成体系的工厂预制容器,在单个工厂制造模块内集成 GPU 计算节点、主配电、液冷歧管和 UPS 备份。每个 MAF 单元承载 1-2 MW IT 负载;做到园区规模时,数十个 MAF 单元再接入共享电力基础设施、网络和光纤入口,拼出 30 MW 到 1 GW 以上的园区。模块化设计把园区上线周期压到 6-8 个月,而传统设计建造型数据中心需要 24-36 个月;原因很直接:工厂预制模块到场时已完成布线和预调试,可以直接接入发电机或电网。 Crusoe 当前主力部署的 GPU 是 NVIDIA H100 SXM5(80 GB HBM3),H200 SXM5(141 GB HBM3e)正在部署,B200 SXM5(192 GB HBM3e)目标在 2026 年 Q3 GA。2025 年 Q4 扩大的多年期 NVIDIA 战略合作,让 Crusoe 在 B200 和 GB200 NVL72 系统上获得配额优先级和早期访问权。集群内网络使用 NVIDIA Quantum-2 InfiniBand,每条链路 400 Gbps,为 512+ GPU 分布式训练提供 GPU-to-GPU 对分带宽。节点级 NVLink 和 NVSwitch 支持服务器内 all-reduce 操作。 Crusoe 的 BYO-Power 计划允许超大规模客户把 GPU 计算部署在客户自有变电站场址,并配套表后发电。Abilene 至少有一家超大规模合作伙伴按该安排提供约 300 MW 专用表后电力。Abilene Phase 1(200 MW IT 负载)已于 2026 年 Q1 投运,使用 765 kV 高压输电互联来降低降压损耗。Edge Zones 于 2026 年 3 月推出,借助第三方托管机房中的 1-5 MW 分布式推理节点扩展计算覆盖,瞄准实时推理、视频分析和主权 AI 工作负载的低于 10 ms 延迟。 Crusoe 位于 Colorado 州 Brighton 的制造工厂(350,000 sq ft,约 $200M 投资)规模化生产 MAF 单元,首批商业交付目标为 2026 年 Q3。产能高峰时,工厂预计每季度产出 50-100 个 MAF 单元,大约每 90 天可装备一个新的 200 MW 园区。GB200 NVL72 机架(每架 72 块 Blackwell GPU,每架约 5 PFLOPS 峰值 FP8)计划部署到 Abilene Phase 2。 [CE001, CE002, CE003, CE004, CE005, CE006]
| GPU | HBM 容量 | FP8 峰值 TFLOPS | 互连 | Crusoe 状态 |
|---|---|---|---|---|
| H100 SXM5 | 80 GB HBM3 | ~2,000 TFLOPS | NVLink 4.0 + QDR-2 IB | GA – 已部署运行 |
| H200 SXM5 | 141 GB HBM3e | ~2,000 TFLOPS | NVLink 4.0 + QDR-2 IB | GA – 2025 年 12 月 |
| B200 SXM5 | 192 GB HBM3e | ~4,500 TFLOPS | NVLink 5.0 + QDR-2 IB | 目标 GA:2026 Q3 |
| GB200 NVL72 | 8 TB HBM3e(72 块 GPU) | ~90,000 TFLOPS(机架) | NVLink 5.0 芯片到芯片 | Abilene Ph2 已规划 |
TFLOPS 数字为 NVIDIA FP8 理论峰值;实际模型 FLOP 利用率(MFU)未披露。HBM 容量和 GPU 数量来自 NVIDIA 产品简报及 Crusoe 公告。
[CE005, CE006, CE007, CE033, CE040]| 站点 | 位置 | 类型 | IT 负载 / 规模 | GPU 代际 | 状态 |
|---|---|---|---|---|---|
| Abilene 一期 | Abilene,TX | 超大规模园区 | 200 MW IT | H100/H200 | 2026 年 Q1 已投运 |
| Abilene 二期 | Abilene,TX | 超大规模园区 | +200 MW 目标 | GB200 NVL72 | 计划 2026-2027 |
| Brighton 工厂 | Brighton,CO | MAF 制造 | 350,000 sq ft | N/A(生产) | 2026 年 Q3 首批交付 |
| Edge Zone 站点 | 多个托管机房 | 分布式推理 | 每个 1-5 MW | H100/H200 | 2026 年 3 月上线 |
| 搁浅天然气节点 | ND + WY | 火炬气算力 | 每个 ~5-20 MW | H100 | 存量业务;正在收缩 |
IT 负载数据来自 Crusoe 公告和 Heatmap News 报道。计划部署依据公开表述,可能随市场条件或合作伙伴决策变化。
[CE008, CE015, CE029, CE031, CE035, CE036]5.2 Crusoe Cloud – 软件平台与 API 服务
Crusoe Cloud 是为 GPU 场景打造的云平台,提供裸金属、虚拟机和 Kubernetes 托管 GPU 集群,可通过 REST API、CLI 和 Terraform provider 访问。平台开放三档计算:按需(按小时计费)、预留(1 年或 3 年承诺)和现货(可抢占、折扣价)。推理 API 层兼容 OpenAI Chat Completions 和 Completions v1 API 接口,可直接替换;支持模型包括 Llama 3.1/3.2/3.3、Mistral、Mixtral、DeepSeek-R1/V3、Qwen2.5 和 Phi-4,并以 NVIDIA TensorRT-LLM 作为服务后端。 Crusoe File System(CFS)基于 Lustre,是并行分布式文件系统,聚合吞吐以每秒数百 GB 计;设计目标是让存储与 GPU 集群共址,避开 PCIe 或网络到 CPU 的瓶颈。CFS 支持 GPUDirect RDMA,让 GPU 无需 CPU 参与即可直接对存储执行 DMA。对于 Kubernetes 工作负载,CFS 通过 CSI driver 暴露为 ReadWriteMany 持久卷。块存储(NVMe-over-Fabric,单实例最高 100 TB)、对象存储(S3 兼容)和网络挂载 NFS 卷补齐了存储组合。 网络层为一个 pod 内的每个 GPU 节点提供 400 Gbps InfiniBand;跨 zone 带宽为 100 Gbps Ethernet。私有网络使用 VPC,并配置子网级防火墙控制。平台运行在 OpenStack 上,并使用 Keystone IAM;企业 SSO 支持 SAML 2.0 和 OIDC 联邦身份。容器镜像维护并发布到 GitHub Container Registry 的 ghcr.io/crusoecloud。开源 Terraform provider(Apache 2.0)已有 200+ commits,客户用它做基础设施即代码部署,提供了可验证的开发者采用信号。Kubernetes 文档覆盖节点池配置、自动伸缩和 CFS 卷挂载。 平台于 2025 年 Q2 取得 ISO 27001:2022 认证,并于 2025 年 Q3 取得 ISO 42001:2023(AI 管理体系)认证,二者均由 UKAS 认可的 DNV GL 确认。AI 系统认证尤其有差异化;截至 2026 年中,全球持有 ISO 42001 的云服务商不足十家。SOC 2 Type II 仍未披露,这对受监管企业买家构成合规缺口。 [CE009, CE010, CE011, CE012, CE013, CE014]
| 存储类型 | 技术 | 最大容量 / 吞吐量 | 关键特性 | 最适合 |
|---|---|---|---|---|
| CFS(并行 FS) | 基于 Lustre | 总计数百 GB/s | GPUDirect RDMA | 分布式 ML 训练检查点 |
| 块存储 | NVMe-over-Fabric | 每实例最高 100 TB | 低延迟 I/O | 数据库和 OS 卷 |
| 对象存储 | S3 兼容 API | 弹性(未说明上限) | boto3/DVC 无缝替换 | 数据集存储和模型产物 |
| NFS 卷 | 网络附加 | 标准 NFS 吞吐量 | ReadWriteMany | 共享配置和小型数据集 |
CFS 总吞吐量由 Crusoe 自述;尚无独立第三方基准发布。块存储容量上限和对象存储限制以 2026 Q2 文档为准。
[CE012, CE013, CE032, CE041]| 控制领域 | 状态 | 标准 / 机制 | 备注 |
|---|---|---|---|
| 保密性认证 | 2025 Q2 取得 | ISO 27001:2022 (DNV GL) | 覆盖全部云基础设施 |
| AI 治理认证 | 2025 Q3 取得 | ISO 42001:2023 (DNV GL) | 全球 <12 家云服务商持有 |
| 企业 SSO | 可用 | SAML 2.0 + OIDC (Keystone IAM) | 面向企业买家的联合身份 |
| 网络隔离 | 可用 | VPC + 子网级防火墙 | 通过 OpenStack 提供标准 hypervisor 隔离 |
| SOC 2 Type II | 未报告 | N/A | 面向受监管企业买家的合规缺口 |
据 GlobeNewswire,DNV GL 已确认截至 2025 年 4 月取得 ISO 27001 和 42001 认证。SOC 2 Type II 缺失状态以报告日为准;Crusoe 可能在推进但未公开披露。
[CE017, CE018, CE019, CE020, CE030]5.3 战略合作与供应链
Crusoe 已建立多年期 NVIDIA 战略合作,覆盖 GPU 配额、围绕 B200/GB200 部署的电力与冷却优化共研,以及联合营销。这层关系让 Crusoe 提前看见 NVIDIA 路线图节点,并在供给紧张期获得配额优先级。2026 年 Q1,Crusoe 宣布与 Form Energy 达成多年协议,在未来园区部署长时铁空气电池储能,回应表后可再生能源发电固有的电网间歇性风险。 HuggingFace Inference Partner 认证意味着 Crusoe 会出现在 HuggingFace Hub 用户的 HuggingFace Inference Endpoints 部署菜单中,直接解决微调模型服务的开发者获客问题。Together.ai 和其他推理 API 聚合商已发布引用 Crusoe 硬件的性能基准。 signisys.com 合作提供电网接入工程和高压变电站设计,服务 Crusoe 园区建设,尤其适用于 765 kV Abilene 互联。Upstream Data(upstreamdata.ca)被列为油田火炬气发电单元供应商,用于 North Dakota 的搁浅天然气计算部署;它代表 Crusoe 早期火炬气减排业务传统中的一个节点,如今正被大型园区规模路线逐步取代。 [CE024, CE025, CE026, CE027, CE028, CE042]
| 合作伙伴 | 关系类型 | 范围 | 状态 |
|---|---|---|---|
| NVIDIA | 战略合作 | H100/H200/B200/GB200 配额 + 联合工程 | 多年期;2025 Q4 扩大 |
| Form Energy | 长时储能 | 未来园区的铁空气电池储能 | 多年期协议,2026 年 1 月 |
| HuggingFace | 推理合作伙伴 | Hub 模型一键部署到 Crusoe | 截至 2025 年中活跃 |
| Together.ai | 基准合作伙伴 | 已发布 H100 LLM 吞吐量基准 | 2026 Q1 发布 |
| Signisys | 电网工程 | Abilene 765 kV 变电站设计 | 已完成——Abilene 已投运 |
| Upstream Data | 电力系统 | 用于搁浅天然气节点的火炬气发电单元 | 存量业务;正在退出 |
合作状态截至 2026 Q2。Together.ai 基准细节来自 Together.ai 自身发布;方法论没有独立验证。
[CE024, CE025, CE026, CE027, CE042, CE043]5.4 图表与证据
06客户情况
6.1 客户基础概览与分层
Crusoe 的客户基础分成三类:(1)承诺数百 MW 容量协议的超大规模锚定客户;(2)按需或通过短期订阅购买 GPU 计算、用于训练、微调和推理工作负载的 AI 产品公司和初创公司;(3)为特定高算力应用部署 Crusoe Spark 模块化 AI 工厂单元的工业企业和研究客户。 超大规模云厂商分层由 Microsoft 在 Abilene 的 900 MW 承诺(2026 年 3 月宣布)以及作为首栋楼锚定租户的 Oracle/Stargate 领衔,很可能贡献 Crusoe 签约容量的大部分,但公开收入披露未单独拆分。此类客户看重 Crusoe 将电力到 GPU 的上线周期压到 12 个月以内的能力,以及以有利购电协议大规模运营的能力。Crusoe 2024 年 AI 云收入为 $124M(同比增长 460%),2025 年 ARR 增长 150%,显示需求强劲;但收入分母、客户数和分层拆分仍未披露。 AI 产品公司分层包括 Cursor、Together AI、Fireworks、Luma AI、Windsurf(Codeium)、Databricks、Playground AI 和 Decart,它们通过标准计算 API 使用 Crusoe Cloud。Crusoe 的 H100/H200/B200 按需价格(截至 2026 年 5 月,H100 为 $2.99/hr)相对超大规模云价格有竞争力,平台的 ISO 27001 和 ISO 42001 认证也降低了企业安全审查摩擦。GPU-as-a-service 市场预计到 2028 年将以显著 CAGR 增长(IDC、MarketsAndMarkets),Crusoe 被定位为这一扩张市场中领先的新云服务商之一。 工业和研究客户构成差异化且高留存的分层。Redwood Materials 是证据最充分的案例:这家电池回收公司部署 Crusoe Spark 单元,用于 AI 驱动的质量控制和材料科学工作负载,七个月内实现 99.2% 正常运行率,随后从 4 个扩展到 24 个 Spark 单元——7x 密度提升,是 Crusoe 先落地再扩张模式最强的公开证明。Sony Research 使用 Crusoe Cloud GPU 基础设施训练 Gran Turismo Sophy(GT Sophy)强化学习赛车智能体,代表了高知名度研究用例。 [CU001, CU002, CU003, CU006, CU007, CU009]
| 分层 | 买方 / 用户 / 付款方 | 使用场景 | 规模 | 收入 / 战略价值 | 缺口 |
|---|---|---|---|---|---|
| 超大规模云厂商 / 产能锚定方 | 基础设施采购方(Microsoft、Oracle) | 用于 AI 模型训练、推理和数据中心建设的大规模 GPU 产能 | 每个锚定方 >500 MW | 收入主力;Microsoft 900 MW + Oracle 锚定租户可能占已签约产能 >50% | 未披露单客户收入或产能利用率;合同条款保密 |
| AI 产品公司 / 创业公司 | 工程与 ML 团队(Cursor、Together AI、Fireworks、Luma AI、Windsurf、Databricks、Decart) | 用于 LLM 训练、微调、图像 / 视频生成推理的 GPU 算力 | 每客户 10–1,000 块 GPU | 已点名客户众多;合计收入未披露;ARR 轨迹暗示增长很快 | 无客户数;无队列留存数据;新云厂商竞争带来价格压力 |
| 工业企业 | 运营与 R&D 团队(Redwood Materials) | AI 驱动的质量控制、材料科学、流程优化 | 每次部署 4–24 个 Spark 模块化单元 | 记录最充分的扩张证明(Spark 单元增长 7x);作为标杆客户有战略价值 | 单客户收入未披露;单一标杆限制可推广性 |
| 研究机构 | 学术 / 企业研究人员(Sony Research) | 高强度 AI 训练(强化学习、机器人仿真) | 每项目数十到数百块 GPU | 高知名度案例,但通常按项目交付;经常性收入潜力较低 | 无重复合作或续约数据;项目制限制 NRR 贡献 |
| 开发者 / 社区 | 个人 ML 从业者和小团队 | 实验、小规模微调、模型评估 | <10 块 GPU | 低 ACV 的长尾量;GitHub 和 HuggingFace 传递开发者采用信号 | 无细颗粒度指标;若不能转化为企业客户,社区采用难以形成有意义规模的变现 |
规模和收入估计来自公开产能公告和定价推断;未公开任何单客户收入拆分。所有规模估计均为近似值。
[CU001, CU002, CU003, CU004, CU006, CU008]| 指标 | 值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 2024 年 AI 云收入 | $124M | FY 2024 | 公司披露(Crusoe 新闻室 / Impact Report) | 中 | 460% YoY 增长验证超大规模云厂商 + AI 公司需求;Bitcoin / DFM 转型完成 | 上年 AI 云收入基线(隐含 $27M)未单独确认 |
| 2025 年 AI 云 ARR 增长 | 150% YoY | 2025 | 公司声称(新闻室) | 中 | 若准确,说明增长持续保持三位数;确切 ARR 和期末日期未披露 | 未披露 ARR 分母或期末日期 |
| 2024 年总收入 | $276M | FY 2024 | 公司披露 | 中 | AI 云占 2024 年收入 45%;BTC / DFM 的 $152M 将在 2025 年后剥离 | 未经审计;无独立确认;毛利率未披露 |
| Redwood Materials Spark 扩张 | 4 到 24 个单元(7x) | 2025–2026 | Crusoe 新闻室公告 | 高 | 最强的先落地再扩张证据;验证工业客户留存和增购路径 | 扩张金额未披露;每个新增单元的时间线未给出 |
| Microsoft Abilene 产能承诺 | 900 MW | 2026 年 3 月 | Crusoe / datacenterknowledge 官方新闻稿 | 高 | 最大单客户承诺;锚定租户降低 Abilene >$15B 资本开支风险 | 合同期限、定价、最低承诺未公开 |
| Redwood Materials 正常运行时间 | 7 个月内 99.2% | 2025 | Crusoe 2024 Impact Report | 中 | 达到企业 SLA 门槛;支撑工业客户分层的留存 | 具体 SLA 合同门槛未披露;部分年度数据可能不能反映稳态 |
| 客户数 | 未披露 | 无法对标市场渗透率或 ACV;关键尽调缺口 | 尽调中要求按分层和合同类型提供确切付费客户数 |
标为“未披露”的数值反映已确认的证据缺口。增长指标由公司声称且未经审计。除非另有说明,均为 USD。
[CU002, CU004, CU005, CU013, CU014, CU031]从 GPU 云潜客到活跃扩张的估计转化阶段,展示 Crusoe 在 2025–2026 年跨客户细分的销售打法。
漏斗数值是用于尽调框架的示意性估计。Crusoe 未披露管线规模、转化率或实际客户数。阶段数值并非基于公司提供的数据。
[CU001, CU013, CU016, CU019]6.2 已披露客户部署与结果
Crusoe 已披露一批有分量的具名生产客户,但各账户的公开证据深度差异很大。最强证明点是 Redwood Materials:Crusoe 在 2024 Impact Report 和专门新闻稿中披露了具体正常运行率数据(七个月 99.2%)以及扩张路径(4 个到 24 个 Spark 单元)。这代表了有量化结果和已验证扩张的生产部署,是资本密集型基础设施供应商客户证明的黄金标准。 Microsoft 在 Texas 州 Abilene 的 900 MW 承诺(2026 年 3 月宣布)是披露的单一最大客户承诺,代表一段多年期锚定客户关系,验证 Crusoe 服务超大规模云级工作负载的能力。Forbes 调查(2025 年 4 月)将 Crusoe 识别为关键 Stargate 基础设施合作伙伴,Oracle 则是首栋楼锚定租户。这些超大规模云关系采用长期容量协议,与 AI 产品公司使用的按需和订阅访问不同。 NVIDIA 在专门新闻稿中宣布的战略合作,确认了 GB200/B200/H200 GPU 的优先访问安排,在 GPU 短缺期为 Crusoe 客户提供供应确定性。这类合作伙伴层面的证据补强了客户层面的证明。 AI 产品公司客户(Cursor、Together AI、Fireworks、Luma AI、Windsurf/Codeium、Decart、Playground AI、Databricks)在 Crusoe 新闻资料和第三方报道中被公开列为客户,但具体部署指标、结果数据和合同金额仍未披露。GitHub 活动(crusoecloud 组织及开源工具)和 HuggingFace 存在(crusoeai 模型页面)提供了开发者社区活跃采用的信号,但不是生产证明。 一个重要反向数据点是 Forbes 报道的 2025 年 3 月初 45 小时宕机。这是唯一公开记录的重大可靠性事件,提出了 SLA 表现和 Crusoe 事件响应流程成熟度问题。公开渠道未发现后续客户投诉,但没有公开 NRR 数据,意味着无法量化该事件对客户留存的影响。 [CU003, CU004, CU005, CU006, CU007, CU008]
| 客户 | 分层 | 部署 / 使用场景 | 生产环境 / 试点 | 结果 | 局限 |
|---|---|---|---|---|---|
| Microsoft | 超大规模云厂商锚定客户 | Abilene,TX 园区 AI 基础设施 900 MW 产能 | 生产环境(锚定协议) | 已披露最大客户承诺;多年产能协议锚定 Abilene $15B+ 建设 | 合同条款、定价和最低照付不议义务未公开 |
| Oracle / Stargate (OpenAI) | 超大规模云厂商锚定客户 | 首批 Abilene 楼宇的锚定租户;Stargate AI 数据中心联盟 | 生产环境(锚定租赁) | 首个超大规模云厂商租户验证 Abilene 园区;OpenAI 终端用户意味着顶级 AI 训练需求 | 具体产能(MW)、合同条款和期限未披露 |
| Redwood Materials | 工业企业 | EV 电池回收 AI 工作负载;质量控制和材料科学 | 生产环境(7+ 个月) | 正常运行时间 99.2%;Spark 单元从 4 个扩至 24 个(7x);最强公开客户证明 | 收入金额未披露;单一客户限制代表性 |
| Cursor | AI 产品公司 | AI 编程助手(cursor.com)的 GPU 推理 | 生产环境 | 活跃生产用户;公开称 Crusoe 为基础设施提供方 | 未披露结果指标、合同金额或期限 |
| Together AI | AI 产品公司 | LLM 训练与推理工作负载 | 生产环境 | 公司材料和新闻报道显示为活跃客户 | 部署规模、结果或合同数据未公开 |
| Sony Research | 研究机构 | Gran Turismo Sophy(GT Sophy)AI 赛车智能体的强化学习训练 | 生产环境(项目) | 里程碑已达成:GT Sophy 已完成训练并部署到 Gran Turismo 7 | 项目制;无续约或重复合作数据 |
| Fireworks | AI 产品公司 | LLM 推理 API 工作负载 | 生产环境 | 已点名客户;活跃 GPU 消费方 | 未披露指标 |
| Luma AI | AI 产品公司 | 视频生成推理(Dream Machine) | 生产环境 | 已点名客户;GPU 密集型视频合成 | 未披露指标 |
| Windsurf / Codeium | AI 产品公司 | AI 编程助手 GPU 推理 | 生产环境 | 已点名客户;在同一基础设施上与 Cursor 竞争 | 未披露指标 |
| Databricks | 企业软件 / MLOps | ML 训练和数据工程工作负载 | 生产环境(可能) | Crusoe 材料中的已点名客户;企业级 MLOps 买方 | 公开细节有限;合作深度未确认 |
| Decart | AI 创业公司 | LLM 训练与研究工作负载 | 生产环境(可能) | 已点名客户;早期 AI 研究公司 | 无指标;公开资料有限 |
| Playground AI | AI 产品公司 | 图像生成推理 | 生产环境(可能) | 已点名客户;GPU 密集型图像合成 | 未披露指标 |
客户证明深度差异很大。只有 Redwood Materials 发布了量化结果。Microsoft 和 Oracle 已通过官方新闻稿确认,但合同条款仍保密。AI 产品公司客户出现在 Crusoe 材料中,但缺乏独立结果证据。
[CU002, CU003, CU004, CU005, CU006, CU007]Crusoe AI 云客户从初始认知到扩展部署的步骤,包括可靠性阶段的关键反向风险。
旅程阶段由公开客户公告、Crusoe 新闻室和 Redwood Materials 案例研究综合得出。阶段间转化率未公开。
[CU001, CU002, CU004, CU006, CU010, CU015]按生产确认、结果披露、扩张信号、参考质量和证据新鲜度,评估 Crusoe 具名客户的证据质量。
[CU001, CU010, CU011, CU019, CU028, CU036]6.3 留存动态与扩张模式
Crusoe 尚未披露任何标准 SaaS 留存指标——NRR、GRR、队列留存、合同续约率或客户数。对于资本密集型基础设施供应商,这是重大证据缺口;留存质量决定 $15B+ Abilene 建设的经济性,也决定未来园区的可行性。 尽管缺少正式留存数据,现有证据指向两极化留存画像。超大规模锚定客户(Microsoft、Oracle)基本被长期容量协议和高切换成本锁定——一旦基础设施按某个超大规模云厂商的具体要求建成,终止合同会同时伤害双方经济性。Microsoft 的 900 MW 承诺代表多年期签约收入,短期流失风险几乎可以忽略。 AI 产品公司的留存信号不那么清晰。Cursor、Together AI、Fireworks、Luma AI 等活跃客户群尚未报告流失,但缺少正式队列数据,无法区分留存订阅客户和一次性训练任务。Crusoe 的 ISO 27001 与 ISO 42001 认证、Command Center 统一运营平台,以及 NVIDIA GPU 供给优先级形成切换成本,应当支撑留存,但这些优势没有被量化。 Redwood Materials 案例展示了先落地再扩张模式的最佳状态:4 个 Spark 单元的试点在约一年内放大 7x 至 24 个单元,驱动因素是已验证的正常运行率(七个月 99.2%)和运营价值。如果这一扩张率能代表更广泛的工业客户,则意味着 Spark 产品线拥有强劲净留存率(NRR)。但单个数据点不足以外推留存率或 NRR 数字。 Crusoe 的定价页列出按需 GPU 计算价格,让小客户无需长期合同即可使用基础设施,降低漏斗底部摩擦;但这也意味着对价格敏感的客户切换成本更低,一旦 GPU 现货价格进一步下探,他们可以转向 Lambda Labs、CoreWeave 或其他新云服务商。 [CU004, CU005, CU011, CU015, CU016, CU020]
| 指标 | 值 / 空值 | 分层 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| 净留存率(NRR) | 未披露 | 全部 | N/A | 要求按分层提供过去 12 个月 NRR;这是测算基础设施 ACV 的关键 |
| 总留存率(GRR) | 未披露 | 全部 | N/A | 要求提供 GRR,以把流失与扩张拆开;对非超大规模云厂商队列尤其重要 |
| 客户流失率 | 未披露 | 全部 | N/A | 要求按分层和队列年份提供 logo 流失和收入流失 |
| 合同续约率 | 未披露 | 全部 | N/A | 要求按分层提供续约率和平均合同期限 |
| Redwood Materials 正常运行时间 SLA | 2025 年 7 个月内 99.2% | 工业企业 | 中 | 确认 99.2% 是否达到合同 SLA;要求提供完整可用性历史和 P1 事件数量 |
| Redwood Materials Spark 扩张 | 4 到 24 个单元(7x) | 工业企业 | 高 | 要求提供时间线,并确认扩张来自合同承诺还是自然增购 |
| Microsoft 锚定承诺 | 900 MW 长期 | 超大规模云厂商 | 高 | 要求提供最低照付不议条款、合同期限,以及任何退出条款或重新谈判窗口 |
除 Redwood Materials 正常运行时间和 Microsoft 承诺外,所有留存指标均未披露。Crusoe 是私营公司,未向 SEC 提交文件;留存数据需要直接尽调访问。
[CU004, CU005, CU011, CU015, CU016]按客户细分推断部署后第 3、6、12 个月的留存率;Crusoe 未披露净留存率(NRR)或队列数据,数值根据可获得的公开信号估算。
留存率为说明性估算,依据公开可得信号:Redwood Materials 扩张 7x、部署超过 7 个月,意味着留存接近 100%;超大规模云厂商锚定协议已有合同承诺;AI 产品公司的留存估计依据未见公开披露流失以及 GPU 价格压力动态;研究细分估计按项目制处理,续用频率较低。Crusoe 未披露任何净留存率(NRR)、总留存率(GRR)或队列留存指标。这些情景仅用于尽调框架。
[CU004, CU005, CU007, CU011, CU016]6.4 客户集中度与反向发现
Crusoe 的客户基础存在显著集中度风险。按容量看,Microsoft(900 MW)和 Oracle/Stargate 很可能是两大主要收入贡献方,两者合计可能占 Abilene 签约容量的 50% 以上。这样的超大规模云集中度在早期基础设施建设商中并不罕见(CoreWeave 也同样集中于 Microsoft/Azure),但意味着任一锚定客户重新谈判、取消合同或技术路线改变,都可能实质损害 Crusoe 收入及其偿付 $9.6B JPMorgan 项目融资的能力。 两项反向发现值得尽调关注。第一,Forbes 报道的 2025 年 3 月 45 小时宕机,是商业爬坡期一次有记录的可靠性失败。宕机时长接近整整两天,显著超出典型云 SLA 阈值(99.9% 月度可用性仅允许约 43 分钟停机)。如果 Crusoe 的企业 SLA 承诺超大规模云级可用性,该事件很可能触发补救义务。研究期内未发现 Crusoe 或受影响客户就宕机发布公开声明。 第二,Heatmap News(2025 年 3 月)记录了 Crusoe 清洁能源定位与实际能源组合之间的张力:TCEQ 许可数据确认 Abilene 有 360 MW 现场燃气轮机,Crusoe 还在开发 4.5 GW 新增燃气发电能力。对 ESG 敏感的企业客户,尤其是有净零承诺的客户,可能因与 Crusoe 合作面临采购摩擦或声誉风险。虽然未发现客户因此流失,但它仍是潜在风险,可能影响企业销售周期,尤其是欧洲客户或 Scope 2/3 减排承诺激进的美国公司。 新云服务商赛道竞争日益激烈,CoreWeave、Lambda Labs、Nebius 等公司都在扩张 GPU 云容量,并把 H100 现货价格从 2023 年的约 $8/hr 压到 2025 年的 $2-3/hr。这种价格压力对短期、按需的 AI 产品公司客户影响更大,而这些客户代表 Crusoe 的先落地再扩张漏斗。没有 NRR 数据,无法判断 Crusoe 是否能在价格压缩周期中留住这些客户。 [CU010, CU016, CU019, CU028, CU029, CU030]
| 扩张驱动因素 / 集中度风险 | 类型 | 影响 | 证据质量 | 尽调路径 |
|---|---|---|---|---|
| Microsoft Abilene 900 MW 锚定承诺 | 集中度风险 | 关键:可能占 Abilene 一期已签约收入 >30%;JPMorgan 融资依赖锚定租赁 | 高(官方新闻稿,datacenterknowledge) | 要求提供合同期限、最低照付不议和重新谈判权;确认 Blue Owl JV 契约影响 |
| Oracle / Stargate 首栋楼锚定租户 | 集中度风险 | 高:已知第二大承诺;Stargate 联盟由 OpenAI 需求兜底 | 中(新闻报道) | 确认 MW 承诺和合同条款;评估 Oracle 内部 OpenAI 终端用户集中度 |
| Redwood Materials Spark 单元 4→24 的落地后扩张 | 扩张驱动因素 | 中:验证工业客户扩张逻辑;Spark Factory 放量取决于能否复制 | 高(Crusoe 新闻室,Impact Report) | 要求提供管线中的其他 Spark 部署;评估 7x 是常态还是离群值 |
| 超大规模云厂商集中度(Microsoft + Oracle 合计) | 集中度风险 | 高:两个客户可能占 Abilene 已签约产能 >50%;失去单一客户会威胁一期融资存续 | 中(推断) | 确认各客户产能占比;要求说明 JPMorgan 融资安排中与锚定租户留存挂钩的契约触发条件 |
| GPU 价格压缩下的按需 AI 创业客户留存 | 集中度风险 / 扩张风险 | 中:H100 现货价从约 $8/hr 跌至约 $2-3/hr;小客户可转向 CoreWeave、Lambda Labs 或超大规模云厂商 | 中(分析师数据) | 要求提供按需客户队列留存数据;跟踪 H100/H200 现货价格趋势 |
影响评估由产能承诺和融资结构推断。没有可用的单客户收入数据。尽调路径需要管理层开放访问。
[CU002, CU003, CU012, CU019, CU036]6.5 图表与证据
07风险
7.1 监管与法律风险评估
Crusoe 处在能源基础设施、AI 计算和环境合规交叉点,监管足迹复杂。近期最重要的监管风险,是 Texas Commission on Environmental Quality(TCEQ)对 Crusoe Abilene 园区现场燃气轮机的许可流程。TCEQ 许可文件确认,Abilene Phase 1 建设有 360 MW 燃气轮机容量。这些大气排放许可证要求 TCEQ 审查批准后,轮机才能运行;任何许可延迟都可能推迟容量投运,而该容量支撑 JPMorgan 项目融资和 Microsoft 锚定租户协议。 EPA Greenhouse Gas Reporting Program 义务适用于大型固定排放源(通常门槛为 25,000 metric tons CO2e/year),Crusoe 在 Abilene 的 360 MW 轮机很可能超过该门槛,因此需要根据 40 CFR Part 98 每年披露排放。Crusoe 尚未公开确认其 EPA GHG 报告状态,但轮机规模意味着强制参与。该数据会公开,ESG 批评者和做空者可以用它记录实际碳排放与清洁能源叙事之间的差距。 Texas ERCOT 电网依赖是另一项监管风险:Crusoe 的 AI 园区将从 ERCOT 获取大量电力,而 ERCOT 在 2021 年 2 月 Winter Storm Uri 期间发生过灾难性故障。虽然 Crusoe 的表后和现场发电策略部分缓解了电网暴露,大型数据中心客户仍受 ERCOT 需求响应计划约束,并可能在电网紧急状态下面临限电。任何扰乱超大规模云 SLA 的限电事件,都可能触发合同纠纷。 法律层面,Upstream Data 专利诉讼(起诉内容:指称 Crusoe 侵犯与火炬气减排技术相关的专利)已于 2025 年 11 月 24 日通过双方撤诉和许可协议解决。该结果对 Crusoe 有利:诉讼阴影移除,许可协议也意味着 Crusoe 获得了争议技术的持续使用权。不过,Crusoe 核心历史技术(DFM/flare gas)曾发生专利争议,且许可条款(包括任何版税或使用领域限制)未公开,仍是残留尽调项。 研究期内未发现其他活跃诉讼、监管执法行动、制裁暴露或 OFAC 标记。Crusoe 注册于 Delaware、总部位于 Colorado,不构成异常司法辖区复杂性;但 Mubadala Capital(Abu Dhabi 主权财富基金)的投资,可能在 Crusoe 争取美国政府计算合同或 IPO 注册时触发 CFIUS 审查。 [CR001, CR002, CR003, CR004, CR005, CR006]
| 规则 / 许可 / 案件 | 司法辖区 | 状态 | 可能性 | 严重性 | 缓释措施 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| Abilene 360 MW 燃气轮机 TCEQ 空气许可证 | 德州(TCEQ) | 待批 / 许可流程进行中 | 中高 | 关键——Abilene 一期投运进度风险 | 提前对接 TCEQ;许可容量按一期需求匹配 | 许可延误可能使 MS / Oracle 上线推迟 6-12 个月;JPMorgan 里程碑触发条件承压 | 要求提供 TCEQ 许可证编号;确认发证时间线以及任何条件或上诉 |
| EPA 温室气体报告(40 CFR Part 98) | 联邦(EPA) | 很可能需要;尚未确认 | 高 | 重大——公开排放数据与清洁能源品牌叙事相冲突 | 完成年度报告合规;如尚未完成,聘请环境法律顾问 | EPA 数据公开后,ESG 批评者会将其用于漂绿论证 | 确认 EPA 温室气体报告状态;索取已提交或草拟的披露表 |
| ERCOT 电网可靠性与需求响应义务 | 德州(PUCT/ERCOT) | 持续 | 中 | 高——电网紧急状态下限电,可能触发超大规模云厂商 SLA 违约 | 现场 360 MW 涡轮机 + Form Energy 电池提供部分孤岛运行能力 | Uri 冬季风暴级事件仍是不可抗力风险;SLA 违约责任不清楚 | 确认供电协议条款和 ERCOT 需求响应义务 |
| Upstream Data 专利诉讼 | 联邦(德州北区) | 已解决——2025 年 11 月 24 日双方撤诉 | N/A(已解决) | 低剩余风险——许可协议已签署 | 双方撤诉并签署许可协议;诉讼阴影已解除 | 许可条款(版税、使用领域限制)未披露 | 索取许可协议条款;确认不存在影响利润率的版税义务 |
| CFIUS / 国家安全审查(Mubadala 投资) | 联邦(CFIUS) | 尚未触发;存在潜在敞口 | 低-中 | 中——一旦触发,可能阻碍美国政府合同或 IPO | Mubadala 投资已有记录;很可能已聘请 CFIUS 顾问 | 政府计算合同可能要求 CFIUS 申报;IPO S-1 需要披露 | 确认 CFIUS 审查状态;评估美国政府合同管线 |
严重性评级是相对投资风险评估,不是法律结论。所有状态信息均来自公开来源(TCEQ、EPA、法院记录、新闻稿)。
[CR001, CR002, CR003, CR004, CR005, CR006]按影响(行)和发生可能性(列)评估 Crusoe 的关键风险,并在单元格细节中叠加缓解成熟度。
风险位置为截至 2026 年 5 月基于公开证据作出的时点性定性评估。发生可能性指未来 12–24 个月内发生的概率;影响指对投资逻辑的估计严重程度。
[CR001, CR003, CR009, CR010, CR012, CR018]7.2 运营与技术风险
Crusoe 的运营风险画像主要由两件事决定:一边要用 SLA 承诺服务超大规模云和企业客户,一边要在不到两年内从初创公司扩张为多 GW AI 工厂运营商。迄今记录最充分的运营失败,是 Forbes 报道的 2025 年 3 月 45 小时宕机。该事件持续接近整整两天,显著超过 99.9% 月度可用性 SLA 允许的停机时间(每月约 43 分钟)。公司未发布事后复盘,也未公开承认客户补救。 冷却故障是第二个关键运营风险。Crusoe 使用 direct-to-chip liquid cooling 来实现 1.2–1.3 PUE 效率,这需要区别于传统风冷系统的专用冷却基础设施。GPU 集群中的冷却回路故障,可能在数分钟内触发热降频或硬件损坏。Abilene 放大了这一风险(Texas 州 Abilene 夏季平均最高温约 98°F),环境温度会持续拉高冷却负载需求。 电网依赖是 Abilene 园区的系统性风险。虽然 Crusoe 拥有现场燃气轮机(360 MW)提供备用和补充发电,但承诺容量为 2.1 GW 的园区仍将主要依赖 ERCOT 电网供电。ERCOT 容量约束、输电拥堵和需求响应义务都可能中断电力交付。自然灾害事件(龙卷风、热穹、冰暴)构成不可抗力场景,即便 Crusoe 在合同上无责,也可能触发 SLA 违约。 当前 H200/B200/GB200 过渡期,NVIDIA GPU 供应链风险显著。Crusoe 与 NVIDIA 的战略合作提供优先 GPU 配额,但全球 GPU 配额到 2026 年仍然紧张。GB200 NVL72 交付延迟会推迟 Abilene Phase 2 部分容量的投运时间,影响 Microsoft 锚定租户协议的收入爬坡。Crusoe 已投资 NVIDIA 的 Vera CPU 路线图,这是一项押注未验证芯片的多年执行赌注。 任何多租户 GPU 云运营商都面临网络安全和数据泄露风险。Crusoe 已获得 ISO 27001 认证,证明其信息安全管理体系(ISMS)框架成熟,但云基础设施供应商持续面对国家级攻击者和机会主义攻击者的威胁,目标包括 LLM 训练数据、模型权重和客户工作负载。重大安全事件可能引发企业客户流失、监管调查和声誉损害;以当前增长阶段看,恢复难度会很高。 [CR009, CR010, CR011, CR012, CR013, CR014]
| 失效模式 | 发生概率 | 严重性 | 缓释成熟度 | 剩余敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| 45 小时 GPU 云宕机(复发风险) | 中 | 高——企业客户 SLA 违约;触发流失 | 早期——没有公开复盘;ISO 27001 不覆盖运营正常运行时间 | 若事故后 SLA 条款未强化,剩余敞口为高 | 未发布复盘;SLA 阈值未披露;补偿付款未知 |
| GPU 集群冷却系统故障 | 低-中 | 极高——几分钟内损坏 GPU 硬件;收入损失 | 中——直连芯片液冷已验证;冗余设计未知 | 中——冷却冗余是 N+1 还是 N+2 尚未确认 | 冷却冗余架构未公开;Abilene 热应力测试数据不可得 |
| ERCOT 电网停电 / 限电 | 中 | 高——园区所有客户生产工作负载中断 | 中——360 MW 现场涡轮机提供部分备份;Form Energy 电池协议待落地 | 中——电池要到 2027 年才上线;短期电网依赖偏高 | ERCOT 需求响应义务和 SLA 不可抗力排除条款尚未确认 |
| NVIDIA GPU 供应延迟(H200/B200/GB200 交付) | 中 | 高——投产时间表滑后;锚定租户收入爬坡延迟 | 中——来自 NVIDIA 的优先配额;NVL72 交付进度受 TSMC/CoWoS 产能约束 | 高——GB200 NVL72 良率和交付周期是全行业约束 | Abilene 二期 NVL72 交付时间表尚未公开确认 |
| 网络安全泄露(多租户云) | 低-中 | 极高——客户数据外泄;模型 IP 被盗;监管调查 | 中-高——已获 ISO 27001 认证;ISMS 控制有文档记录 | 中——国家级威胁行为者专门瞄准 AI 基础设施 | 渗透测试结果未公开;SOC 2 Type II 未确认;漏洞赏金计划状态未知 |
| 数据中心火灾或实体灾害(Abilene TX) | 低 | 极高——价值 $2-5B 的 GPU 库存实体损失;收入停止 | 中——标准数据中心消防抑制;PUE 1.2-1.3 降低热负载 | 低-中——保险覆盖和业务连续性计划未公开 | 业务连续性计划、保险覆盖和恢复时间目标未披露 |
发生概率和严重性基于行业基准和可得公开证据做定性评估。缓释成熟度:早期 = 无文档证据;中 = 部分证据;高 = 已形成文档并经过测试。
[CR009, CR010, CR011, CR012, CR013, CR015]Crusoe 的关键风险如何传导到收入、客户、利润率、融资、运营和估值。
[CR001, CR009, CR012, CR018, CR019, CR020]7.3 财务、合作伙伴与依赖风险
Crusoe 的财务风险画像由资本密集度主导。Abilene 园区建设需要超过 $15B 的总承诺资本($3.4B Blue Owl JV 股权、$9.6B JPMorgan 项目融资,再加上 Crusoe 自有股权)。到了这个规模,Crusoe 已不再只是技术初创公司,而是一个资本市场运营体,依靠持续的投资者信心、契约条款合规和锚定租户留存来偿付债务。若里程碑错失、锚定租户合同终止或信用评级下调触发 JPMorgan 信贷契约违约,公司可能被迫以不利条件出售资产或稀释股权。 GPU 价格压力是现实财务风险。受 NVIDIA 供给增加和新云容量扩张推动,H100 现货价格从 2023 年约 $8/hr 跌至 2025 年 $2-3/hr。如果 H200/B200 价格沿类似轨迹下行,Crusoe 单位计算收入将被压缩,按需和短期订阅客户的贡献会下降。超大规模云锚定协议缓解了部分冲击(这些协议很可能采用基于容量而非现货 GPU 价格的定价),但如果 Crusoe 无法匹配竞争对手价格,按需 AI 初创分层面临实质流失风险。 NVIDIA 既是 Crusoe 的战略伙伴,也是其主要 GPU 供应商依赖。如果 NVIDIA 将优先级转向更大的超大规模云厂商(AWS、Google、Microsoft Azure),或决定发展自有云产品,来自 NVIDIA 的优先配额并无保证。NVIDIA 关系任何恶化——无论来自供给配额争议、路线图变化,还是 NVIDIA 的 DGX Cloud 战略——都会直接损害 Crusoe 履行客户承诺的能力。 Microsoft 约占 Abilene 阶段承诺总量 2.1 GW 中的 900 MW。如果 Microsoft 因 AI 需求收缩、竞争压力或内部战略调整而减少、推迟或取消该承诺,JPMorgan 项目融资契约条款很可能被触发。评估这一集中度风险时,Blue Owl JV 结构和 Microsoft 合同中的任何 take-or-pay 条款,是关键未知项。 JPMorgan 正在安排 $9.6B 项目融资,形成单一资本市场交易对手集中度。如果 JPMorgan 对数据中心项目融资的信贷偏好改变(原因可能是市场环境、资本要求或风险限额),短期内为该工具再融资会很困难。该融资工具的条款、契约条款和再融资安排,是关键尽调项。 [CR017, CR018, CR019, CR020, CR021, CR022]
| 依赖 | 交易对手 | 角色 | 集中度 | 失效场景 | 严重性 | 缓释措施 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| GPU 供应 | NVIDIA | 主要 GPU 供应商;战略投资方;技术伙伴 | 极高——已部署 GPU 中 >95% 为 NVIDIA H100/H200/B200/GB200 | NVIDIA 将供应重新分配给 AWS/Google/Azure;或 NVIDIA 推出竞争性云产品 | 极高——客户承诺延迟兑现;触发锚定租户 SLA 风险 | 优先配额协议;NVIDIA 是 Series D 投资方(保留激励) | 高——GB200 规模上不存在替代 GPU 供应商;AMD MI300X 只能部分替代 |
| 项目融资 | JPMorgan | $9.6B 项目融资牵头安排方;契约执行方 | 极高——单一银行安排 Abilene 大部分债务 | JPMorgan 降低授信意愿;触发契约违约;利率飙升 | 极高——被迫出售资产或按不利条款股权稀释 | Blue Owl JV($3.4B)部分降低单一银行依赖;Abilene 园区作为抵押品 | 高——契约条款、覆盖率和里程碑触发项未公开披露 |
| 锚定租户容量 | Microsoft | Abilene 900 MW 锚定租户;最大单一客户承诺 | 极高——很可能占 Abilene 一期签约容量 >30% | Microsoft 削减 AI 资本开支(战略转向或 AI 寒冬);推迟建设;退出合同 | 极高——触发 JPMorgan 契约审查;Blue Owl JV 现金流被扰动 | 长期容量协议结构;很可能是 take-or-pay,但尚未确认 | 高——Microsoft 是组合中单一交易对手集中度最高的对象 |
| 储能 | Form Energy | 12 GWh 铁空气电池自 2027 年开始部署 | 中——电池提供电网备份;对 2027 年后园区可靠性至关重要 | Form Energy 未能交付,或铁空气技术表现低于规格 | 高——园区供电可靠性下降;ERCOT 孤岛运行能力受损 | 早期合作已锁定;技术已在电网项目中验证 | 中——铁空气电池仍处商业化早期;Form Energy 已完成 Series D 融资 |
| 建设 EPC 与模块化工厂 | Spark Factory 供应链 | Brighton CO 工厂生产模块化 AI 工厂单元;首批交付在 2026 年 Q3 | 中——工厂尚未运营;可能延期 | 工厂投产延迟;设备采购瓶颈;劳动力中断 | 高——限制 Crusoe 履行 Abilene 之外客户承诺的能力 | Brighton 设施面积 350,000 sq ft,在建中;投资 $200M | 中——工厂投产时间表没有独立确认 |
| 云计算竞争对手 | CoreWeave、Lambda Labs、Nebius、AWS、GCP、Azure 等云计算竞争对手 | 争夺 GPU 云客户的市场竞争者 | 中——竞争性定价压力;客户被挖走风险 | 竞争对手激进压低 H100/H200 按需价格;客户迁移工作负载 | 中——AI 创业公司客群流失;按需 ARR 增长放缓 | ISO 认证、NVIDIA 优先访问、Command Center 差异化 | 中——H100 现货价已在 $2-3/hr;可能进一步压缩 |
集中度评级基于公开可得数据做定性评估。合同条款、契约和 take-or-pay 义务未公开。剩余敞口评级假设除公开已知措施外没有额外缓释。
[CR019, CR020, CR021, CR022, CR023, CR024]Crusoe 在合作伙伴、供应商、监管方、客户和融资方上的关键依赖,界定了运营和财务风险暴露。
[CR001, CR003, CR018, CR019, CR020, CR022]7.4 执行、ESG 与声誉风险
Crusoe 员工数从 2024 年底约 600 人,几乎翻倍至 2026 年 3 月 31 日的 1,235 人。这种组织扩张速度带来执行风险:文化稀释、流程缺口、管理带宽约束和协同成本上升。公司在 2025 年 12 月聘请新 COO/CFO Michael Gordon(前 MongoDB 高管),并在 2026 年 1 月聘请新任房地产总法律顾问 Nader Pakfar,显示其正转向更成熟的运营模型。不过,对联合创始人 Chase Lochmiller(CEO)和 Cully Cavness(President)的关键人依赖仍然偏高——二人是主要技术和能源愿景塑造者,若离职或减少参与,对投资者将构成重大不利变化。 位于 Colorado 州 Brighton 的 Spark Factory(350,000 sq ft,$200M 投资)计划在 2026 年 Q3 交付首批单元。若该时间线因建设超支、设备采购、许可或质量问题延迟,Crusoe 履行 Abilene 园区之外的模块化 AI 工厂客户承诺的能力将受限。Spark 产品线是重要的收入多元化路径,延迟会拉长公司对单一 Abilene 园区的依赖。 Wyoming 园区(已宣布 1.8 GW)截至研究期(2026 年 5 月)尚未破土动工。公告与开工之间的空档,引入了 Abilene 以外既定项目管线的执行风险。Wyoming 的电力接入协议、许可时间线和融资尚未公开确认。 最突出的 ESG 和声誉风险,是 Crusoe 清洁能源定位与实际能源组合之间的张力。TCEQ 许可数据确认 Abilene 有 360 MW 现场燃气轮机,Heatmap News 还报道称 Crusoe 正与 Engine No. 1 合作,为 AI 需求开发 4.5 GW 新增燃气发电能力。这构成重大漂绿风险:如果 ESG 记者、做空者或激进投资者放大 Heatmap 报道,Crusoe 可能承受声誉压力,并影响企业销售周期(尤其是有强制 Scope 2/3 披露的欧洲客户)、未来融资(ESG 挂钩债务)和 IPO 定位。2024 Impact Report 和 ISO 42001 AI 管理认证是正面动作,但没有回应燃气供电计算与清洁 AI 营销之间的根本张力。 [CR026, CR027, CR028, CR029, CR030, CR031]
| 角色 / 职能 | 依赖或缺口 | 发生概率 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| Chase Lochmiller——CEO 兼联合创始人 | 技术愿景、AI/能源投资逻辑、投资者关系、产品战略;核心公开面孔 | 低(留任) | 极高——离职会引发投资者信心流失,并可能触发重组 | 股权留任方案;联合创始人承诺在 2026 融资周期中仍可见 | 索取股权归属时间表;确认董事会继任计划;评估 Series E 条款中的创始人留任安排 |
| Cully Cavness——总裁兼 CSO | 能源采购、选址、现场运营、Abilene/Wyoming 电力策略;短期内在运营上不可替代 | 低(留任) | 高——离职会削弱能源管线和场址开发执行 | 股权留任;运营团队由其搭建 | 索取能源采购团队后备梯队情况;确认是否有副手或 VP 级替代人选 |
| Michael Gordon——COO 兼 CFO | 上市公司财务纪律、资本市场、JPMorgan/Blue Owl 关系管理 | 低-中(新聘) | 高——离职会削弱潜在 IPO 的公开市场准备度和 CFO 连续性 | 入职时授予股权;MongoDB 履历匹配 Crusoe 的规模化野心 | 索取雇佣协议条款;确认 CFO 是否拥有董事席位或观察员权利 |
| 员工数从 600 扩至 1,235 | 文化稀释、流程缺口、管理带宽被挤压;约 15 个月内员工数翻倍 | 高(执行风险) | 中——爬坡变慢;质量控制问题;项目层面沟通错位 | 新 COO/CFO 提供职业化管理结构;ISO 42001 AI 治理纪律 | 索取组织架构图;评估关键职能(工程、运营、BD)的 VP/Director 后备梯队 |
| Spark Factory 投产(Brighton CO) | 承诺首批模块化 AI 工厂在 2026 年 Q3 交付;工厂尚未运营 | 中(延期风险) | 高——约束 Crusoe 的模块化工厂业务;限制 Abilene 之外的客户管线 | 350,000 sq ft 设施、$200M 投资;已投入承诺资本规模可观 | 索取施工完工里程碑;确认首批 Spark 单元交付所需设备采购状态 |
| Wyoming 园区(已宣布 1.8 GW) | 已宣布但未开工;电力接入、许可和融资尚未确认 | 中(执行风险) | 中——限制公司在 Abilene 之外宣称的增长管线;竞争对手窗口风险 | Abilene 执行提供模板;能源团队和地产团队已到位 | 索取 Wyoming 场址控制文件;确认购电协议状态和电网互联申报 |
留任概率是定性推断。所有评为极高或高的角色,都需要在正式尽调中提供继任计划文件。快速增长的 AI 基础设施公司普遍面临员工数扩张风险。
[CR026, CR027, CR028, CR029, CR030, CR031]| 风险 | 可监测触发项 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| TCEQ 许可延迟 | TCEQ STEERS 数据库中的 TCEQ 许可状态更新 | 距预计 Abilene 一期投产日期 90 天内仍未发放许可 | 下调收入爬坡假设;索取 JPMorgan 里程碑契约条款;标记工期滑后风险 |
| Microsoft 锚定合同修改 | Microsoft 新闻稿、数据中心租赁公告和财报电话会 | Microsoft 公开宣布削减 Abilene 容量,或向竞争性新型云厂商承诺等量容量 | 立即触发投资逻辑破裂;评估 JPMorgan 契约敞口;升级给高级尽调团队 |
| GPU 定价跌破 Crusoe 成本 | Lambda、CoreWeave 和主要云厂商市场上的 H100/H200/B200 现货价格 | H100 现货价跌破 $1.50/hr,或 H200 跌破 $3.00/hr 并持续 3+ 个月 | 重估按需业务经济性;核实锚定租户定价是否按容量计费(与现货价格隔离) |
| 宕机复发(>12 小时) | 公开事故报告、客户社交媒体和行业出版物 | 2025 年 3 月事件后 12 个月内,再次发生超过 12 小时的重大宕机 | 索取事后复盘;评估 SLA 违约敞口;下调运营成熟度评估 |
| 漂绿 ESG 事件升级 | 主流财经媒体、ESG 基金评论和 SEC 调查文件 | Forbes、Bloomberg 或 FT 发布关于 Crusoe 天然气使用的多篇调查;SEC 启动问询 | 标记声誉风险;评估对 ESG 敏感型企业销售管线和 IPO 时间表的影响 |
| NVIDIA 供应中断 | NVIDIA 季度财报、供应链报告和 GPU 现货市场数据 | NVIDIA 宣布 GB200 交付延迟 >6 个月,或下调 Crusoe 配额等级 | 评估对 Abilene 二期时间表的影响;评估 GPU 替代品(AMD MI300X)的可得性和成本差异 |
| JPMorgan 契约违约 | Crusoe 契约合规报告(如可取得);信用评级机构观察名单更新 | JPMorgan 宣布违约,或发起项目融资契约豁免谈判 | 最高严重性触发项;升级给董事会;评估流动性跑道和替代融资选项 |
| Spark Factory 投产延至 2026 年 Q4 之后 | Crusoe 新闻稿、Forbes/Forbes 关于 Spark Factory 进展的报道 | 截至 2026 年 12 月 31 日仍未宣布 Spark Factory 单元交付 | 重估模块化产品线时间表;评估对非 Abilene 客户管线的影响 |
这些触发项服务于主动尽调监控计划。阈值仅作示例;需根据投资逻辑和风险偏好调整。标记为“最高严重性”的终止标准意味着需要立即重新评估投资逻辑。
[CR001, CR009, CR018, CR019, CR020, CR030]7.5 图表与证据
08估值
8.1 估值背景与融资轮历史
Crusoe 的私募市场估值自 2019 年种子轮以来,经过六轮股权融资一路抬升,2025 年 10 月 Series E 达到 $10B+,较不到 12 个月前完成的 $2.8B Series D 跃升 3.6x。Series E 的 SEC Form D(2025 年 10 月 23 日提交)确认总募资 $1,374,999,988、71 名投资者,由 Valor Equity Partners 和 Mubadala Capital 共同领投。Series D(Form D 于 2024 年 11 月 21 日提交)确认募资 $600M,由 Founders Fund 领投,NVIDIA、Fidelity、Mubadala、Ribbit Capital 和 Valor 参投。所有轮次累计股权融资约 $3.4B。除股权外,Crusoe 还搭出 $15B+ 承诺资本栈:Blue Owl Capital 为 Abilene 数据中心建设在合资企业中承诺 $3.4B,JPMorgan 正为同一园区安排约 $9.6B 项目融资。这个结构更像基础设施 REIT 或电力公用事业的真实资产融资模型,而不是软件初创公司的融资模型——这是有意选择,扩大了可比公司集合,也压缩了传统收入倍数框架的适用性。 不到 12 个月内从 $2.8B 跳到 $10B+,由三个催化剂驱动:(1)2025 年 3 月将 Bitcoin 挖矿业务出售给 NYDIG,明确 Crusoe 转为纯 AI 云和数据中心公司;(2)据报道 2025 年云 ARR 增长 150%,总合同价值(TCV)增长 17x;(3)宣布 Microsoft 在 Abilene 的 900 MW 锚定承诺,降低公司最大资本项目风险。考虑这些催化剂后,这轮估值跳升在分析上可以辩护,但当前估值仍需要确认收入规模超过 $1B,才能称为明显便宜,而不是动量驱动。按 2024 年全年收入 $276M 计算,$10B 估值意味着约 36x 往绩收入。如果未经验证的分析师估计——2025 年收入 $500M 至 $1B——被证实,隐含倍数会降至 10 至 20x。若 ARR 为 $1B、倍数为 10x,估值基本合理;若 ARR 为 $500M、倍数为 20x,估值仍偏激进。主要尽调任务是用经审计或管理层确认的数据核实 2025 年 ARR,再决定是否将判断从观察上调为买入。[CV001, CV002, CV003, CV004, CV005, CV006]
按 CoreWeave 基准的 5x ARR 倍数,Crusoe 需要 $2B ARR 才能支撑 $10B 估值;若按 10x,需要 $1B,正好落在未经验证的 2025 年分析师估计区间顶部。
ARR 到估值的桥接是按给定倍数做的简单 EV/ARR 计算,不是 DCF 结果。2025 年收入估计来自分析师,未获审计数据验证。
[CV009, CV013, CV014, CV015, CV016, CV017]8.2 可比公司与交易分析
Crusoe 最相关的公开可比公司是 CoreWeave,后者于 2025 年 3 月完成 IPO,隐含估值约 $23B。CoreWeave IPO 时年化收入约 $5B,按 IPO 价格计算收入倍数约为 4.6x ARR——显著低于 Crusoe 当前隐含的 10 至 20x,具体取决于采用哪一个 2025 年收入估计。CoreWeave 上市后的交易表现体现了倍数压缩,股价较 IPO 标记下跌,说明公开市场对该赛道的审视比私募市场更克制。若 Crusoe 要相对 CoreWeave IPO 基准证明其 $10B 估值合理,就需要 ARR 接近 $2B、并按 5x 倍数定价;这一目标尚未验证,且很可能还要数年。 Nebius AI(NASDAQ 上市,截至 2025 至 2026 年市值约 $6B)提供第二个数据点:一家获得 NVIDIA 强力支持的欧洲 GPU 云公司,尽管定位相似,交易价格只是 Crusoe 隐含入场价的一小部分。Lambda Labs 私募估值估计约 $1.5B,说明没有自有电力基础设施、面向开发者的新云服务商获得的倍数明显更低。Lancium 是 Texas 私营新云服务商,具备相似的电力优先逻辑,但没有公开估值锚可直接比较,是最接近的战略类比对象。Bain Capital Ventures 文章明确把投资逻辑放在电力稀缺这一持久护城河上:AI 基础设施需求受制于许可和电网接入,早期电力权利因此具备防御性。不过,超大规模云厂商也在主动获取电网权利,ComputeForecast 新云分析指出,H100 的 GPU 现货价格已从约 $8/hr 降至 $2 至 $3/hr——这是压缩单 GPU 经济性的结构性逆风。公开和私有可比公司的枚举仍不完整,因为多数直接同行要么仍为私有,要么刚上市且运营历史不足,无法干净锚定倍数,因此 TV004 仍应标记为部分枚举范围。[CV016, CV017, CV018, CV019, CV020, CV021]
| 可比公司 | 阶段和市场 | 收入或 ARR | 估值与倍数 | Crusoe 参照价值 | 核心局限 |
|---|---|---|---|---|---|
| CoreWeave | 上市公司(2025 年 3 月 NASDAQ IPO) | IPO 时 ARR 约 $5B | IPO 估值约 $23B;约 4.6x ARR;IPO 后下跌 | 最接近的上市可比公司;NVIDIA 偏好的新型云厂商,GPU 云模式相近、扩张很快 | 缺少自有电力基础设施;IPO 后倍数压缩;客户组合不同 |
| Nebius AI | 上市公司(NASDAQ 上市) | ARR 未披露;欧洲 GPU 云运营商 | 2025 至 2026 年市值约 $6B | GPU 云,获 NVIDIA 支持;商业模式与 Crusoe Cloud 部分重叠 | 美国布局小得多;聚焦欧洲市场;没有美国电力护城河逻辑或超大规模云厂商锚定客户 |
| Lambda Labs | 私营公司 | ARR 未披露;面向开发者的 GPU 云 | 分析师来源估计估值约 $1.5B | 开发者 GPU 云可比;同属新型云类别,但没有基础设施差异化 | 没有自有电力基础设施;没有超大规模云厂商锚定租户;规模小得多 |
| Lancium | 私营公司 | ARR 未披露;位于 Texas 的电力优先数据中心运营商 | 截至 2026 年 5 月没有公开估值锚点 | Texas 市场里最接近的电力优先 AI 基础设施战略类比 | 私营且估值不透明;没有可比收入或 ARR 数据 |
| 超大规模云厂商(AWS、Azure、GCP) | 上市公司;万亿美元级市值 | 云收入达数千亿美元 | 收入倍数 5 到 10x;万亿美元级估值 | 定义 Crusoe 的竞争天花板,也定义潜在收购场景 | 规模、多元化和资产负债表根本不同;不能直接套用倍数 |
可比公司集并不完整;最直接的类比对象(Lancium、Applied Digital、Vast Data)多为私营或早期上市公司,经营历史不足。CoreWeave 提供了最可用的公开倍数基准。
[CV016, CV017, CV018, CV019, CV020]8.3 乐观、基准与悲观情景
乐观情景从一个前提出发:Crusoe 以电力优先的垂直整合,能产生结构上优于现货 GPU 新云服务商的收入质量。如果 Crusoe 确认 2025 年 ARR 超过 $1B,并在 2026 年维持 100%+ 同比增长,公司到 2026 年底可能接近 $2B ARR。按 8 至 12x ARR 倍数——与高增长基础设施软件上沿一致——公司价值可达 $16B 至 $24B,较当前 $10B 估值有 1.6 至 2.4x 上行。催化剂包括 Abilene Phase 1 完整投运、新增超大规模锚定租户,以及 2026 年末或 2027 年提交 IPO 文件,提供公开市场价格发现。 基准情景认为,如果 2025 年收入落在 $750M 至 $1B 之间,Series E 可视为合理的私募市场估值标记。按已确认 $750M ARR 的 8 至 10x 前瞻倍数计算,公允价值为 $6B 至 $7.5B——低于当前 $10B 标记,但处在一个强劲 2026 年执行和 IPO 溢价可以弥合差距的区间。基准情景不要求当前轨迹继续加速,只要求管理层通过 150% ARR 增长表述所暗示的事实得到确认。主要下行风险是客户集中度:Microsoft 和 OpenAI 被认为是锚定收入贡献方,任何承诺缩减都会实质损害投资逻辑。 悲观情景假设 GPU 现货价格继续商品化,超大规模云厂商自行吸收新云需求,且 Crusoe 2025 年收入确认在 $500M 低端。按 $500M ARR 的 5 至 6x 计算,隐含估值降至 $2.5B 至 $3B——较 2025 年 10 月标记折价 70% 至 75%。悲观情景不是归零情景,但代表显著降估值融资风险。漂绿暴露增加尾部风险:Heatmap News 记录 Crusoe 的 Abilene 园区使用 360 MW 现场天然气轮机,公司还聘请 Engine No. 1 开发 4.5 GW 新增燃气容量,与清洁能源品牌形成重大张力,可能重定价 ESG 敏感资本。MarketsAndMarkets 和 IDC 关于 GPU-as-a-service 市场增长的数据支撑乐观情景的需求环境,但没有给出 Crusoe 特定的远期 ARR 指引,因此情景选择仍依赖判断。[CV031, CV032, CV033, CV034, CV035, CV023]
| 情景 | 关键假设 | 估值逻辑 | 概率信号 |
|---|---|---|---|
| 乐观 | 2025 年 ARR 确认 $1B+;2026 年 ARR 以 100%+ 增长达到 $2B;按 10 到 12x 远期 ARR IPO;Microsoft 和两个或更多超大规模云厂商租户;GPU 价格底线未被击穿 | $1B ARR × 12x = $12B;2026 年 $2B ARR × 10x = $20B;以 $10B 入场并持有 3 到 5 年,回报 1.2 到 2.0x | 低到中;需要 ARR 确认、Abilene 全面投产,并新增超大规模云厂商承诺 |
| 基准 | 2025 年 ARR 确认在 $750M 到 $1B;2026 年增速放缓至 50% 到 80%;2027 到 2028 年按 8 到 10x ARR IPO;Microsoft 锚定仍在,但未签下新的超大规模云厂商 | $750M ARR × 8x = $6B;$1B ARR × 8x = $8B;较当前 $10B 略有折价;入场价回本到小幅亏损 | 中;若 2024 年退出时基数为 $400M 到 $450M,最符合已确认的 150% ARR 增长说法 |
| 悲观 | 2025 年 ARR 确认在 $500M 或以下;GPU 现货价维持 $2/hr;超大规模云厂商自供加速;客户集中风险兑现;ESG 声誉受损 | $500M ARR × 5x = $2.5B;$500M ARR × 6x = $3B;较 $10B 入场标记折价 70% 到 75% | 低到中;GPU 商品化和超大规模云厂商入场是真实风险,但完整悲观情景需要多个失败同时发生 |
估值倍数是供投资委员会讨论的 EV/ARR 情景示例,不是 DCF 输出。所有 ARR 数据均为未经审计估计,因 Crusoe 没有公开财务披露。
[CV014, CV015, CV033, CV034, CV035, CV040]乐观情景较 $10B 入场估值有 1.2 到 2.4x 上行空间;基准情景大致是持平到小幅亏损;悲观情景意味着较当前估值下行 70% 到 75%。
区间是用于 IC 讨论的 EV/ARR 情景示意,单位为十亿美元;不是管理层指引,也不是 DCF 结果。基准情景上限 $10B 等于 2025 年 10 月融资轮的当前估值。
[CV033, CV034, CV035, CV040, CV041]8.4 投资建议与尽调要求
当前建议是观察——不是买入。Crusoe 展现出非同寻常的基础设施牵引力:$15B+ 承诺资本、2025 年云 ARR 增长 150%、TCV 增长 17x、Microsoft Abilene 锚定客户、超过 45 GW 的电力储备,以及 2025 年 10 月由机构级资本(Valor、Mubadala、Founders Fund、NVIDIA)共同领投、估值 $10B+ 的 Series E。Michael Gordon 的加入也释放出明确的公开市场准备信号——他曾作为 CFO/COO 领导 MongoDB IPO,并推动收入约 50x 增长。Crusoe 入选 Fast Company 2026 年 100 Most Innovative Companies,NVIDIA 扩大战略合作,提供了来自全球主导 AI 芯片供应商的技术背书。2025 年取得的 ISO 27001 和 ISO 42001 认证支持企业客户获取。对一家私有基础设施公司而言,这些都是顶级需求和执行信号。 反过来,$10B+ 标记跑在已确认公开财务数据之前。2024 年全年收入 $276M 中,$152M 来自已出售给 NYDIG 的 Bitcoin 挖矿业务,剩余纯 AI 云收入为 $124M。2025 年收入 $500M 至 $1B 的估计未经审计数据确认。公开材料没有披露毛利率、EBITDA 或现金消耗。Microsoft 和 OpenAI 的客户收入集中度未披露,但基于 Abilene 锚定和 Stargate 报道,很可能超过 50%。六轮融资后的股权结构优先权不透明。如果合同包含爬坡条款或取消选项,签约 ARR(TCV 增长 17x)可能高估近期确认收入。除非管理层提供经审计的 2025 年财务数据,或公开财务披露包,否则无法以高置信度承销 $10B 标记。 从观察上调到买入的尽调路径是:(1)取得经审计或确认的 2025 年 ARR,且高于 $1B,并按客户和合同类型拆分;(2)核实 AI 云分部毛利率和贡献利润率;(3)获得股权结构表透明度,包括清算优先权和反稀释条款;(4)确认签约 ARR 与确认收入的差异,以评估积压订单质量;(5)审查实际电力来源与营销主张之间的差距,量化 ESG 风险;(6)评估 Spark 工厂库存和交付风险,因为它是一条新业务线。IPO S-1 文件将是交付大部分信息最高效的机制,也将成为上调建议的主要监测触发器。[CV037, CV038, CV039, CV040, CV041, CV042]
| 决策字段 | 当前判断 | 决策含义 |
|---|---|---|
| 建议 | 观察 | 密切跟踪 IPO 信号和 ARR 确认;在没有经审计的 2025 年财务数据前,不要按 $10B+ 估值承销新资金。 |
| 信心 | 中 | 公开证据能强力支持基础设施牵引,但对已验证经济性和利润率披露支持较弱。 |
| 风险评级 | 高 | 估值倍数压缩可能经由 GPU 定价、超大规模云厂商入场、客户集中或 ESG 声誉损害传导。 |
| 估值立场 | 偏高 | $10B+ 标记需要确认 $1B+ ARR;当前隐含 10 到 36x 倍数高于可比组。 |
| 持有 / 退出姿态 | 持有 3 到 5 年,等待 IPO | 预计 2026 到 2027 年 IPO 申报将带来首次公开市场价格发现;跟踪降估值融资风险。 |
| 价格纪律 | 当前估值下不做价格不敏感买入 | 需要经审计的 $1B+ ARR 确认,或显著更好的入场价格,才把观察转为买入。 |
该判断反映 2025 年 10 月 $10B+ 入场估值,并非对 Crusoe 战略质量的泛泛看法。
[CV001, CV014, CV015, CV040, CV041]| 方向 | 论点 | 证据 | 何种情况会改变判断 |
|---|---|---|---|
| 投资逻辑 | 电力优先的垂直整合在 AI 基础设施中形成可防守护城河 | 45+ GW 管线、2.1 GW Abilene 已承诺、3.4 GW 已投运;Bain 投资逻辑验证电力稀缺优势 | 超大规模云厂商系统性获取竞争性电网权利,抹平 Crusoe 的相对电力优势 |
| 投资逻辑 | 云 ARR 增长 150%、2025 年 TCV 增长 17x,证明需求捕获能力一流 | 公司声称 ARR 增长 150%、TCV 增长 17x;Sacra 和 ComputeForecast 证实新型云厂商 ARR 快速扩张 | 收入确认显示 2025 年实际 ARR 低于 $500M,或增速急剧降至 50% 以下 |
| 投资逻辑 | $15B+ 已承诺资本栈降低基础设施建设执行风险,并验证投资逻辑 | Blue Owl $3.4B JV 和 JPMorgan $9.6B 项目融资;Microsoft 900 MW Abilene 锚定承诺 | Blue Owl 或 JPMorgan 融资条件被触发,或 Microsoft 承诺正式下调或取消 |
| 投资逻辑 | NVIDIA 合作和 Michael Gordon 任命释放可信 IPO 路径执行信号 | NVIDIA 合作在 2025 年扩大;Gordon 带来 MongoDB IPO 方法论;获得 Fast Company 2026 认可 | Gordon 离职,或 NVIDIA 转而支持其他新型云运营商并减少给 Crusoe 的 GPU 配额 |
| 反向逻辑 | H100 GPU 现货价下跌 60% 到 75%;单位经济性面临结构性商品化压力 | ComputeForecast 和 Signisys 记录 H100 GPU 价格从 $8/hr 压缩到 $2-3/hr | GPU 价格持续稳定在 $4/hr 以上,或 Crusoe 证明长期合同价格与现货隔离 |
| 反向逻辑 | Microsoft 和 OpenAI 客户集中,带来单点故障式收入风险 | Microsoft 是 Abilene 主要锚定方;OpenAI 是 Stargate 合作伙伴;具体集中度未披露 | 经审计数据确认客户基础多元化,且没有单一客户超过 ARR 的 25% |
| 反向逻辑 | 没有公开毛利率或 EBITDA 数据;按 $10B+ 估值建立信心所需的财务透明度不足 | 截至 2026 年 5 月,没有任何公开文件披露 Crusoe Inc. 的分部收入、利润率或现金消耗数据 | 经审计财务数据或 S-1 文件确认 $1B+ ARR,且 AI 云业务贡献利润率为正 |
| 反向逻辑 | 清洁能源品牌叙事不一致,漂绿敞口带来 ESG 重新定价风险 | Heatmap News 记录了 360 MW 现场燃气涡轮机,以及 4.5 GW 新增燃气容量参与 | 完整能源采购透明度确认所有园区可再生能源比例高于 80% |
论点和证据综合自公开来源;反向逻辑项反映分析师、新闻和反向来源识别出的结构性风险,并不穷尽所有悲观情景。
[CV022, CV023, CV024, CV025, CV026, CV028]| 触发因素 | 阈值或信号 | 对投资逻辑的传导 | 行动含义 |
|---|---|---|---|
| ARR 确认未达目标 | 2025 年审计或管理层确认 ARR 低于 $400M | 核心估值支柱塌陷;悲观情景变成基准情景;按任何合理倍数都无法支撑 $10B 估值 | 立即下调信心;启动降价轮风险评估;暂停新增资金部署 |
| GPU 现货价格地板持续跌破 | H100 或 H200 现货均价连续两个季度低于 $1.50/hr | 单 GPU 经济性被压到必须具备超大规模云厂商效率,才能让新型云模式打平 | 重新评估每 GPU 收入和远期利润率路径;审查长期合同价格保护 |
| Microsoft 或主要锚定租户削减承诺 | Microsoft 正式削减或取消 Abilene 容量承诺 25% 或以上 | 直接损害 2.1 GW 建设逻辑;可能触发 Blue Owl JV 或 JPMorgan 融资条件 | 立即复核信心;评估项目融资结构中的融资契约违约风险 |
| 监管或客户正式施加漂绿制裁 | TCEQ、EPA 正式执法,或采用 ESG 筛选的主要机构客户正式处罚 Crusoe | ESG 资本重新定价;可能流失欧洲主权 AI 和 ESG 筛选基金投资人 | 跟踪 TCEQ 空气质量许可和 EPA GHG 报告;评估客户 ESG 合规要求 |
| 超大规模云厂商大规模推出竞争性电力优先 AI 园区 | AWS、Azure 或 GCP 宣布专用表后 AI 园区,规模 1 GW+,部署周期少于 12 个月 | 电力护城河叙事削弱;稀缺性溢价被侵蚀;私营和公开市场可能出现倍数压缩 | 重新评估电力管线优势的防御性;审查客户锁定机制和合同条款 |
触发因素是基于公开证据推导的示例阈值;实际投委会流程可能会根据组合背景和授权设置不同数值阈值。
[CV014, CV021, CV022, CV024, CV026, CV034]| 主题 | 缺失证据 | 重要性 | 尽调路径 |
|---|---|---|---|
| 2025 年收入和 ARR 确认 | 经审计或管理层确认的 2025 全年 ARR,并按业务线和合同类型拆分 | 核心估值输入;$10B 估值意味着在未经验证的 $500M 到 $1B 上打 10 到 20x,投资判断前必须确认 | 私有数据室;提交后的 S-1 招股说明书或招股书;来自领投方的管理层确认函 |
| 毛利率和 EBITDA 可见度 | Crusoe 任何业务单元、任何时期都没有公开毛利率、贡献利润率或 EBITDA 数据 | 资本密集模式叠加 JV 和项目融资结构;利润率决定规模能否创造价值 | 非公开财务披露请求;IPO S-1 备案;对标 CoreWeave 分部利润率 |
| 客户收入集中度 | 未公开按客户划分的收入占比;Microsoft 和 OpenAI 集中度未经验证 | 集中度超过 50% 会带来二元收入风险,并给客户重新谈判或取消合约的筹码 | 私有数据室;合同审查;客户访谈;提交 S-1 时的风险因素披露 |
| 股权结构表优先权结构 | 多家领投方参与六轮股权融资;清算优先权和反稀释条款未披露 | 若退出估值低于最高轮价格,优先权悬挂会实质压低普通股回报 | 私有数据室;法律顾问审查股东协议和章程;409A 估值报告 |
| 实际电力来源与营销表述 | TCEQ 许可确认 Abilene 有 360 MW 燃气轮机;各园区总可再生能源占比未披露 | ESG 风险影响主权财富基金和欧盟关联资本;漂绿暴露提高监管风险 | TCEQ 许可审查;EPA GHG 报告数据;要求独立能源审计;现场设施检查 |
| 签约 ARR 与确认收入 | 已披露 TCV 增长 17x,但收入确认时间和合同结构完全未披露 | 如果合同含爬坡条款或任意终止条款,TCV 会高估近期 ARR | 数据室合同抽样;收入确认政策审查;财务报表中的递延收入明细 |
尽调路径假定能访问私有数据室;Crusoe 推进 IPO 时,多数事项也会在 S-1 招股说明书披露。优先级按紧迫性降序排列。
[CV013, CV040, CV041, CV050, CV051]观察建议来自两股力量的对冲:基础设施验证和 ARR 动能很强,但 $10B+ 入场估值偏高、财务披露不足;IPO 申报是主要上调触发点。
[CV001, CV012, CV014, CV015, CV021, CV026]Crusoe 在基础设施落地和 ARR 动能上得分极高,但在财务透明度、相对公开可比公司的估值支撑、客户集中风险上明显偏低。
评分为 0 到 10 的序数判断,综合公开证据得出,用于 IC 讨论;不是量化因子模型输出。
[CV001, CV012, CV014, CV026, CV037, CV040]8.5 图表与证据
免责声明
本报告由自动化研究流程生成,仅供信息参考,不构成投资建议,也不构成买入、卖出或持有任何证券的要约或招揽。除非另有说明,所有财务数字均为来自公开来源的估计值。未获得经审计财务数据。作者未持有 Crusoe Inc. 或任何关联实体的头寸。前瞻性陈述反映分析师估计,并受重大不确定性影响。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Crusoe is a vertically integrated AI infrastructure company self-described as 'the AI factory company.' | 高 | SO001, SO020 |
| CO002 | Crusoe's vertical integration spans energy sourcing, AI-optimized data center construction, and a proprietary GPU cloud platform. | 高 | SO001, SO011 |
| CO003 | Crusoe is incorporated in Delaware and headquartered at 255 Fillmore Street, Suite 400, Denver, Colorado 80206 (CIK 0001924674). | 高 | SO004, SO020 |
| CO004 | Crusoe Cloud offers GPU/CPU compute, Managed Kubernetes, Managed Slurm, Managed Inference (MemoryAlloy), Command Center, and Edge Zones. | 高 | SO001, SO020 |
| CO005 | Crusoe's primary revenue monetization mechanisms are cloud compute subscriptions, on-demand GPU rental, and long-term capacity agreements with hyperscalers. | 中 | SO001, SO007 |
| CO006 | After NYDIG acquired Crusoe's Bitcoin/DFM operations in March 2025, Crusoe is pivoting to pure AI cloud infrastructure revenue. | 高 | SO015, SO006 |
| CO007 | Crusoe was co-founded in 2018 by Chase Lochmiller (CEO) and Cully Cavness (President and CSO). | 高 | SO020, SO001 |
| CO008 | Chase Lochmiller studied physics/CS at MIT and completed an AI-focused MS in CS at Stanford; he was previously a quantitative trader at Jane Street Capital. | 高 | SO020, SO022 |
| CO009 | Cully Cavness holds a geology degree from Middlebury College and an MBA from Oxford University; he brings oilfield operations experience. | 中 | SO020 |
| CO010 | Michael Gordon joined Crusoe as COO and CFO in December 2025, having previously served as CFO/COO at MongoDB where he led the 2017 IPO. | 高 | SO008, SO006 |
| CO011 | Nader Pakfar was appointed General Counsel Real Estate at Crusoe on January 21, 2026. | 中 | SO009 |
| CO012 | Crusoe's board advisors include Bill Stein (former CEO of Digital Realty Trust) and Peter Gross; full board composition is not publicly disclosed. | 中 | SO020, SO004 |
| CO013 | Crusoe raised $4.5M seed (May 2019, co-led BCV + Founders Fund), $30M Series A (Dec 2019), $128M Series B (Apr 2021, led Valor), $350M+$155M Series C (Apr 2022, led G2 Venture Partners). | 中 | SO011, SO007 |
| CO014 | Crusoe raised $600M in Series D at a $2.8B valuation, announced December 12, 2024, led by Founders Fund with NVIDIA, Fidelity, Mubadala, Ribbit, Valor, and Long Journey Ventures participating. | 高 | SO003, SO005, SO006 |
| CO015 | SEC Form D for Series D (Acc-no 0001924674-24-000003) was filed November 21, 2024 by Crusoe Inc., confirming the securities offering. | 高 | SO005, SO027 |
| CO016 | Crusoe raised $1.375B in Series E at $10B+ valuation, announced October 24, 2025, co-led by Valor Equity Partners and Mubadala Capital with 30+ investors including NVIDIA, T. Rowe Price, and Tiger Global. | 高 | SO002, SO004, SO006, SO023 |
| CO017 | SEC Form D for Series E (Acc-no 0001924674-25-000004) was filed October 23, 2025, confirming total offering of $1,374,999,988 with 71 investors. | 高 | SO004, SO027 |
| CO018 | Blue Owl Capital committed $3.4B in a JV for Abilene campus construction; JPMorgan is arranging approximately $9.6B in project financing; total committed capital for Abilene exceeds $15B. | 中 | SO002, SO019 |
| CO019 | Crusoe employed 1,235 people as of March 31, 2026, approximately double the ~600 headcount at end-2024. | 高 | SO024, SO002 |
| CO020 | NYDIG acquired Crusoe's Bitcoin mining and Digital Flare Mitigation operations in March 2025, including 425+ modular data centers, 250+ MW capacity, and approximately 135 transferred employees. | 高 | SO015, SO006 |
| CO021 | Crusoe's 2024 revenue was $276M (82% YoY growth), comprising AI cloud $124M (460% YoY) and Bitcoin/DFM $152M. | 中 | SO007, SO006 |
| CO022 | Crusoe operates 9.8M sq ft of facilities, 3.4 GW commissioned electricity, and approximately 946,000 GPU-equivalent capacity as of early 2026. | 中 | SO002, SO019 |
| CO023 | Heatmap News (March 2025) reported that Crusoe uses natural gas for power generation and has engaged Engine No. 1 to develop 4.5 GW of new gas capacity, contradicting clean-energy branding. | 高 | SO013, SO014 |
| CO024 | TCEQ permit filings confirm 360 MW of on-site gas turbines at Crusoe's Abilene campus—approximately 30% of Phase 1 capacity. | 高 | SO014, SO013 |
| CO025 | Crusoe's Abilene campus has eight planned buildings; the first two came online within 12 months of the June 2024 groundbreaking. | 中 | SO021, SO012 |
| CO026 | Microsoft committed 900 MW at Abilene in March 2026, expanding the total Abilene footprint to 2.1 GW. | 高 | SO019, SO006 |
| CO027 | The Upstream Data patent lawsuit against Crusoe was resolved in November 2025 via mutual dismissal and a confidential license agreement. | 高 | SO016, SO017 |
| CO028 | Crusoe's Spark Factory in Brighton, CO (350,000 sq ft, $200M investment, 200+ jobs) is manufacturing modular prefabricated AI factory units with first deliveries expected Q3 2026. | 中 | SO025, SO002 |
| CO029 | Crusoe's PUE is 1.2–1.3 versus the industry average of ~1.8, enabled by direct-to-chip liquid cooling with closed-loop non-evaporative systems. | 中 | SO001, SO011 |
| CO030 | Gross margin and net revenue retention figures for Crusoe have not been publicly disclosed; these are private company metrics. | 中 | |
| CO031 | Crusoe achieved ISO 27001 (information security) and ISO 42001 (AI management system) certifications, becoming one of the first AI infrastructure companies to hold both. | 中 | SO022, SO001 |
| CO032 | Crusoe's Redwood Materials partnership, expanded in early 2026 to 7x the original AI infrastructure density, delivers GPU compute to one of the largest US battery recycling operations. | 中 | SO006, SO024 |
| CO033 | Crusoe Edge Zones, launched March 12, 2026, extend Crusoe Cloud beyond the Abilene campus to sovereign and low-latency AI compute deployments at distributed sites. | 中 | SO001, SO006 |
| CO034 | Crusoe Command Center, launched February 18, 2026, is a SaaS observability platform that gives tenants unified visibility into GPU workload performance, power consumption, and infrastructure status. | 中 | SO001, SO006 |
| CO035 | Crusoe won the 'North American Data Center Project of the Year' at the 2025 Data Center Dynamics Global Awards, a third-party recognition of the Abilene campus design and operational excellence. | 中 | SO026, SO006 |
| CM001 | The GPU-as-a-Service market is estimated at $8.21B in 2025 and projected to grow to $26.62B by 2030 at a 26.5% CAGR. | 中 | SM001 |
| CM002 | The broader AI infrastructure market is estimated at $135.81B in 2024 and projected to reach $394.46B by 2030 at a 19.4% CAGR. | 中 | SM001 |
| CM003 | Cloud infrastructure GPU spending grew 46.8% year-over-year in 2025, reaching $157.8B according to IDC. | 中 | SM002 |
| CM004 | Neocloud GPU cloud providers generated over $5B in quarterly revenue in Q2 2025, up 205% year-over-year per ComputeForecast. | 中 | SM003 |
| CM005 | The neocloud industry collectively generated $20B+ in annual GPU-focused cloud revenues by 2025, demonstrating that the market has scaled from niche to mainstream. | 中 | SM004 |
| CM006 | Crusoe describes itself as "the AI factory company" and operates at crusoe.ai with vertical integration spanning energy sourcing, AI factory construction, and GPU cloud operations. | 中 | SM026, SM015 |
| CM007 | Crusoe's AI cloud revenue was $124M in 2024, growing 460% year-over-year, making it one of the fastest-growing neocloud operators based on public disclosures. | 中 | SM006, SM005 |
| CM008 | Crusoe's AI cloud ARR grew 150% year-over-year in 2025, with total contract value growing 17x year-over-year, indicating strong long-term customer commitment expansion. | 中 | SM006 |
| CM009 | The neocloud market is concentrated in a small set of operators, with CoreWeave estimated at approximately $5B ARR as of 2025, representing the largest individual neocloud operator. | 中 | SM003, SM004 |
| CM010 | Crusoe operates 3.4 GW of commissioned electricity capacity across its US data center footprint and 9.8M sq ft of facilities as of early 2026. | 中 | SM015, SM006 |
| CM011 | Crusoe's total power pipeline exceeds 45 GW, with 2.1 GW committed at Abilene TX (1.2 GW Phase 1 + 900 MW Microsoft) and 1.8 GW announced at a Wyoming campus. | 低 | SM012, SM014 |
| CM012 | Crusoe's SAM is estimated at $3–7B by 2030, representing approximately 10–26% of the GPU-as-a-Service market, based on US-centric footprint and large-cluster reservation focus. | 低 | SM001, SM007 |
| CM013 | Crusoe's SOM is estimated at $1.0–1.7B in annualized GPU cloud revenue by 2028, assuming conversion of 3.4 GW commissioned capacity at $0.3–0.5/W annualized revenue per watt of deployed compute. | 低 | SM007, SM024 |
| CM014 | The 150% YoY ARR growth trajectory for 2025 is consistent with Crusoe reaching $1B+ in cloud revenue by 2027–2028 if the current pace is maintained, supporting the SOM estimate. | 低 | SM006, SM007 |
| CM015 | No independent validation of Crusoe's SAM or formal market sizing framework has been publicly disclosed; SAM/SOM estimates are analyst constructs requiring verification against internal Crusoe revenue models. | 低 | SM007 |
| CM016 | Crusoe's largest disclosed customer commitment is Microsoft's 900 MW at the Abilene, TX AI factory campus, representing a multi-year hyperscaler reservation at the top end of the buyer tier. | 高 | SM012, SM014 |
| CM017 | Crusoe serves AI platform companies including Together AI and enterprise AI customers including Cursor and Redwood Materials, confirming multi-segment customer reach beyond hyperscaler contracts. | 中 | SM016, SM017, SM018 |
| CM018 | Redwood Materials expanded its Crusoe GPU compute partnership 7x from the original commitment scope, indicating strong enterprise AI compute demand growth and successful land-and-expand. | 中 | SM018 |
| CM019 | Crusoe offers GB200 NVL72, B200, H200, H100, A100, and L40S NVIDIA GPUs plus AMD MI355x and MI300x, enabling support for both current-generation and next-generation AI workloads. | 中 | SM026 |
| CM020 | The primary adoption triggers for enterprise GPU cloud customers are GPU availability scarcity on hyperscalers, cost arbitrage against AWS/Azure on-demand pricing, and data residency or compliance requirements. | 中 | SM008, SM015 |
| CM021 | Crusoe's PUE of 1.2–1.3 compares favorably to the industry average of 1.8, enabled by direct-to-chip liquid cooling, representing approximately 30–40% lower energy consumption per unit of compute. | 中 | SM025, SM015 |
| CM022 | Crusoe's Spark modular AI factory product, manufactured at the Brighton CO Spark Factory with first deliveries expected Q3 2026, addresses enterprises seeking dedicated on-premise AI compute. | 中 | SM006, SM008 |
| CM023 | Crusoe Cloud pricing as of 2026 is $4.29/GPU-hr for H100 SXM and $1.95/GPU-hr for A100 SXM, which at current H100 spot rates of $2–3/hr means Crusoe's list price is at a premium to spot market. | 中 | SM024, SM003 |
| CM024 | AI model training compute demand is growing exponentially, driven by scaling laws that require substantially more compute per model generation, sustaining long-term GPU demand regardless of spot price dynamics. | 高 | SM008, SM002 |
| CM025 | US grid interconnect queues for new data center power connections in constrained markets exceed 3–5 years, creating a structural supply constraint that advantages operators who secured power positions early. | 中 | SM013, SM015 |
| CM026 | Crusoe's power-first strategy in Abilene, TX and Wyoming secured large power commitments in lower-competition markets before hyperscaler demand intensified, establishing a capacity moat. | 中 | SM012, SM015 |
| CM027 | Crusoe's $15B+ total committed capital for the Abilene campus including Blue Owl Capital's $3.4B JV and JPMorgan's $9.6B project financing arrangement represents a capital commitment scale difficult for pure-play software operators to replicate. | 中 | SM020, SM006 |
| CM028 | H100 GPU spot prices fell from approximately $8/GPU-hr in mid-2023 to $2–3/GPU-hr by early 2025, representing a 60–75% reduction in spot market revenue per unit for neocloud operators. | 中 | SM003, SM004 |
| CM029 | GPU spot price compression is primarily driven by increased supply from hyperscalers and other neocloud operators deploying more GPU capacity, not by reduced demand for AI compute. | 中 | SM003, SM008 |
| CM030 | Crusoe's posted H100 SXM price of $4.29/GPU-hr is above current H100 spot market rates of $2–3/GPU-hr, meaning Crusoe's revenue per unit depends on reserved contract fill, not spot pricing. | 中 | SM024, SM003 |
| CM031 | Inference workloads including test-time compute and chain-of-thought architectures create sustained continuous GPU demand that runs in parallel to episodic training spikes, improving long-term GPU utilization rates. | 高 | SM008, SM002 |
| CM032 | Sustained inference GPU demand improves utilization economics for reserved-capacity neocloud operators like Crusoe, reducing the risk of idle GPU clusters between training runs. | 中 | SM008 |
| CM033 | The combined effect of training scaling and inference workload growth supports multi-year compounding of GPU cluster requirements, consistent with analyst forecasts of 26%+ CAGR for GPU cloud markets. | 中 | SM001, SM002, SM008 |
| CM034 | Crusoe's Abilene campus has TCEQ-permitted 360 MW of on-site natural gas turbines, representing approximately 30% of Phase 1 capacity and contradicting the company's clean-energy positioning. | 中 | SM019 |
| CM035 | Crusoe has engaged Engine No. 1 to develop 4.5 GW of new natural gas generation capacity for AI data center demand, creating potential ESG risk for sustainability-focused customers and investors. | 中 | SM019 |
| CM036 | Crusoe's NVIDIA strategic collaboration announced in January 2025 provides preferred access to GB200 NVL72 GPU allocations, a key competitive differentiator in GPU supply for large AI workloads. | 中 | SM022, SM026 |
| CM037 | The market sizing estimates from MarketsandMarkets, IDC, and ComputeForecast use different methodologies (rental revenue vs capital spend vs operator revenue), making cross-source comparison unreliable without adjustment. | 高 | SM001, SM002, SM003 |
| CP001 | CoreWeave is estimated at approximately $5B ARR as of early 2025, making it the largest neocloud GPU cloud operator and Crusoe's primary pure-play competitor. | 中 | SP001, SP010 |
| CP002 | CoreWeave went public on NASDAQ in March 2025, establishing public market valuation comparables for the neocloud GPU cloud sector. | 高 | SP001, SP002 |
| CP003 | CoreWeave's March 2025 IPO valued the company at approximately 5x revenue, providing a public-market reference multiple for neocloud operators and creating valuation scrutiny for private Crusoe. | 中 | SP001, SP002 |
| CP004 | Lambda Labs offers competitive on-demand GPU pricing (H100 from approximately $2.49/GPU-hr) targeting the developer and research market without minimum commitments. | 中 | SP003 |
| CP005 | Nebius AI raised $700M in September 2024 with NVIDIA participation, and operates GPU cloud infrastructure targeting EU and US markets with a data sovereignty positioning. | 中 | SP004 |
| CP006 | Lancium operates renewable-powered (wind and solar) HPC data centers in Texas with demand-response capabilities, overlapping with Crusoe's original energy thesis and Texas geography. | 中 | SP005 |
| CP007 | The neocloud GPU cloud landscape is competitive with CoreWeave, Lambda Labs, Nebius AI, and Lancium as the primary pure-play operators, plus AWS, Azure, and GCP as the dominant status-quo substitutes. | 中 | SP001, SP010 |
| CP008 | CoreWeave is NVIDIA's most preferred large-scale GPU cloud partner, receiving priority allocations of H100, H200, and Blackwell-generation GPUs—directly overlapping Crusoe's top buyer segment. | 中 | SP019, SP010 |
| CP009 | CoreWeave does not own power generation infrastructure or construct its own AI campuses at scale, relying on third-party colocation agreements—a structural difference from Crusoe's vertical integration. No public evidence indicates CoreWeave has large-cluster GPU reservation contracts in Texas or Abilene geography. | 中 | SP019, SP010 |
| CP010 | Lambda Labs focuses on on-demand spot GPU rental without large-cluster reservation products, making it a less direct competitor in the hyperscaler-commitment or campus reservation tier. | 中 | SP003 |
| CP011 | Nebius AI's GPU cloud is primarily competitive in the EU market where data residency regulations create a buying criterion that Crusoe's US-focused campuses cannot as easily address. | 中 | SP004 |
| CP012 | Lancium's pure wind/solar energy sourcing gives it a stronger clean-energy positioning than Crusoe, which uses natural gas turbines for approximately 30% of Abilene Phase 1 power. | 中 | SP005, SP015 |
| CP013 | AWS P4/P5 instances (H100-based) and Azure NDv5 instances represent the dominant GPU compute substitutes for all Crusoe buyer segments, with global scale and existing enterprise relationships. | 高 | SP011, SP012 |
| CP014 | AWS H100 on-demand pricing is approximately $3.00–5.00/GPU-hr for p5.48xlarge instances, with 1yr/3yr reserved pricing offering 20–40% discounts—partially overlapping with Crusoe's list price range. | 中 | SP011 |
| CP015 | Azure NDv5 instances offer H100 GPU compute integrated with Microsoft 365 and Azure OpenAI API, providing a bundled enterprise value proposition that standalone neoclouds cannot match. | 中 | SP012 |
| CP016 | Hyperscalers AWS, Azure, and GCP hold all major enterprise compliance certifications (SOC2, ISO 27001, FedRAMP, HIPAA), giving them a compliance maturity advantage over newer neocloud operators. | 高 | SP011, SP012 |
| CP017 | Crusoe holds SOC2 Type I+II, ISO 27001, ISO 42001, and GDPR certifications, achieving enterprise-grade compliance parity with CoreWeave and narrowing the compliance gap versus hyperscalers. | 中 | SP014, SP020 |
| CP018 | CoreWeave holds SOC2 Type II and ISO 27001 certifications, achieving the same compliance baseline as Crusoe in the enterprise GPU cloud market. | 中 | SP019 |
| CP019 | Crusoe's primary differentiation is vertical integration spanning power sourcing, AI campus construction, Spark modular factory manufacturing, and GPU cloud operations—a full stack no neocloud competitor replicates. | 中 | SP018, SP007 |
| CP020 | Crusoe claims a compressed timeline from land acquisition to GPU-online of under 12 months for greenfield campuses, versus the industry norm of 3–5 years, enabled by vertical integration. | 中 | SP018 |
| CP021 | Crusoe's NVIDIA strategic collaboration provides preferred access to GB200 NVL72 GPU allocations in a supply-constrained market, differentiating Crusoe from competitors with standard purchase agreements. | 中 | SP006, SP007 |
| CP022 | No neocloud competitor currently offers a modular prefabricated AI factory product at scale comparable to Crusoe Spark, making it an emerging product-market category that Crusoe could define. | 中 | SP008, SP009 |
| CP023 | Crusoe PUE of 1.2–1.3 versus industry average of 1.8, enabled by direct-to-chip liquid cooling, provides approximately 30% lower energy consumption per GPU-hr compared to industry-average data centers. | 中 | SP020, SP018 |
| CP024 | Crusoe's energy moat is partially compromised by its use of natural gas generation at Abilene (360 MW of TCEQ-permitted gas turbines), weakening its clean-energy positioning relative to Lancium. | 中 | SP015 |
| CP025 | Crusoe's 45+ GW power pipeline represents a long-term supply moat, but the pipeline is company-claimed and has not been independently verified through signed interconnection agreements. | 低 | SP018, SP002 |
| CP026 | Crusoe's Redwood Materials partnership expanding 7x demonstrates successful land-and-expand dynamics, validating the enterprise GPU cloud account expansion thesis. | 中 | SP024 |
| CP027 | CoreWeave's estimated $5B ARR versus Crusoe's $124M AI cloud revenue in 2024 represents approximately a 40x scale gap, making CoreWeave the dominant displacement threat in the neocloud segment. | 中 | SP001, SP018 |
| CP028 | As a publicly traded company post-March 2025 IPO, CoreWeave can access capital markets more cheaply than private Crusoe, enabling faster GPU cluster buildout and competitive pricing strategies. | 中 | SP001, SP002 |
| CP029 | Microsoft has committed $80B+ in AI infrastructure capex for 2025, directly increasing GPU availability on Azure and reducing the supply scarcity that drives enterprise customers toward neocloud operators. | 中 | SP002, SP011 |
| CP030 | AWS and GCP are similarly accelerating AI GPU infrastructure deployments, compressing the lead-time and availability arbitrage that currently drives enterprise customers to neocloud operators. | 中 | SP011, SP012 |
| CP031 | As hyperscaler GPU reservation pricing falls and lead times shorten, the switching cost for enterprises already on Crusoe Cloud increases while the incentive for new customers to switch decreases. | 中 | SP011, SP012 |
| CP032 | H100 GPU spot prices falling from $8/hr to $2–3/hr creates a commoditization race where Crusoe's energy cost advantage (30% lower energy/GPU-hr) provides a structural floor but may not prevent reservation price convergence. | 中 | SP001, SP021 |
| CP033 | Forbes (April 2025) reported that a 45-hour outage in early March 2025 at Crusoe affected enterprise customers and raised reliability concerns, representing an adverse operational risk signal. | 中 | SP013 |
| CP034 | Crusoe's ISO 27001, ISO 42001, and SOC2 Type I+II certifications represent governance investment, but operational reliability at GW-scale campus deployments has not been proven over a multi-year production window. | 中 | SP014, SP013 |
| CP035 | CoreWeave's IPO at approximately 5x revenue, applied to Crusoe's $124M AI cloud revenue base, implies a cloud-only valuation well below Crusoe's $10B total enterprise valuation—the premium reflects the infrastructure and land layers, not pure cloud multiple. | 中 | SP001, SP016, SP017 |
| CP036 | No publicly available data exists on competitor NRR or customer retention for CoreWeave, Lambda Labs, or Crusoe in the neocloud GPU cloud market. | 中 | SP010 |
| CP037 | Crusoe's Bain Capital Ventures investor thesis (published 2024) frames Crusoe's power-first strategy as a structural moat that competitors cannot quickly replicate given multi-year grid interconnect queues. | 中 | SP025 |
| CI001 | Crusoe reported $276M total revenue in 2024, representing 82% year-over-year growth. | 高 | SI018, SI009 |
| CI002 | Crusoe's AI cloud revenue was $124M in 2024, representing 460% year-over-year growth from the prior period. | 高 | SI018, SI009 |
| CI003 | Crusoe's Bitcoin/DFM mining revenue was $152M in 2024, comprising approximately 55% of total 2024 revenue. | 高 | SI018, SI011 |
| CI004 | Crusoe reported 150% year-over-year cloud ARR growth in 2025; the absolute ARR base has not been publicly disclosed. | 高 | SI027, SI029 |
| CI005 | Crusoe reported 17x year-over-year total contract value (TCV) growth in 2025; the base dollar amount of TCV was not disclosed. | 高 | SI027, SI029 |
| CI006 | Crusoe Cloud list price for H100 SXM GPU compute is $4.29 per GPU-hour as of May 2026. | 高 | SI001, SI023 |
| CI007 | Crusoe Cloud list price for A100 SXM GPU compute is $1.95 per GPU-hour as of May 2026. | 高 | SI001, SI023 |
| CI008 | Crusoe Cloud list price for CPU compute is $0.04 per vCPU-hour as of May 2026. | 高 | SI001, SI023 |
| CI009 | H100 GPU spot market prices declined from approximately $8/GPU-hour in 2023 to approximately $2–3/GPU-hour by late 2024 as neocloud supply expanded. | 中 | SI013, SI014 |
| CI010 | Crusoe's primary revenue streams are GPU/CPU cloud compute (on-demand and reserved), long-term hyperscaler capacity agreements, and modular AI factory (Spark unit) sales. | 高 | SI002, SI019 |
| CI011 | The March 2025 NYDIG transaction eliminated the Bitcoin and DFM segment from Crusoe's income statement, removing $152M in lower-quality 2024 revenue and reducing capital maintenance obligations for 250+ MW of legacy modular data centers. | 高 | SI011, SI018 |
| CI012 | Following the NYDIG sale in March 2025, Crusoe's revenue is exclusively from AI cloud infrastructure; Bitcoin/DFM mining contributes zero going forward. | 高 | SI011, SI027 |
| CI013 | Crusoe's Series E raised $1,374,999,988 at a $10B+ valuation per SEC Form D filed October 23, 2025, with 71 investors and first sale dated October 14, 2025. | 高 | SI003, SI008 |
| CI014 | The SEC Form D for Crusoe's Series E (Acc-No 0001924674-25-000004) confirms the exact offering amount of $1,374,999,988. | 高 | SI003, SI004 |
| CI015 | Crusoe's Series D raised $600M at a $2.8B valuation; the SEC Form D was filed on November 21, 2024. | 高 | SI017, SI012 |
| CI016 | Crusoe's total equity raised across all funding rounds is approximately $3.4B as of October 2025. | 中 | SI003, SI006, SI008 |
| CI017 | Blue Owl Capital committed $3.4B in a joint venture for Crusoe's Abilene data center construction and buildout. | 中 | SI020, SI024 |
| CI018 | JPMorgan is arranging approximately $9.6B in project financing for the Crusoe Abilene campus build-out. | 中 | SI020, SI025 |
| CI019 | Total committed capital for Crusoe's Abilene campus exceeds $15B, combining Series E equity, Blue Owl JV, and JPMorgan project financing. | 中 | SI020, SI006, SI025 |
| CI020 | The 900 MW Microsoft Abilene capacity commitment (announced March 2026) is estimated to generate approximately $250M in annual revenue contribution for Crusoe. | 低 | SI024, SI009 |
| CI021 | Crusoe has not publicly disclosed gross margin, EBITDA, net income, cash position, or audited financial statements as of May 2026. | 高 | SI009, SI010 |
| CI022 | GPU cloud infrastructure gross margins for neocloud peer companies are estimated at 20–40% based on available industry disclosures and benchmarks. | 中 | SI013, SI014 |
| CI023 | Microsoft and Oracle/Stargate represent Crusoe's dominant hyperscaler customers, creating material customer concentration risk at the revenue level. | 中 | SI024, SI009 |
| CI024 | GPU H100 spot pricing compression from $8/hr to $2–3/hr represents a structural adverse pressure on Crusoe's on-demand cloud revenue and pricing power. | 高 | SI013, SI016 |
| CI025 | Crusoe's H100 list price of $4.29/hr is positioned above the spot market floor ($2–3/hr) and below AWS on-demand rates, targeting premium enterprise and reserved deals. | 中 | SI001, SI013 |
| CI026 | H100 GPU unit cost is approximately $25,000–30,000; depreciated over a 4-year useful life at 80% utilization, this implies a capital charge of approximately $0.60–1.20 per GPU-hour. | 中 | SI013, SI022 |
| CI027 | Capital intensity for hyperscale AI data center construction is approximately $10M–$15M per MW, implying $12B–$21B for Crusoe's 2.1 GW Abilene campus. | 中 | SI022, SI020 |
| CI028 | Michael Gordon joined as COO and CFO in December 2025 having previously served as CFO/COO at MongoDB through the 2017 IPO, signaling active preparation for capital markets. | 中 | SI006, SI027 |
| CI029 | Analyst estimates for Crusoe's full-year 2025 revenue range from approximately $500M to $1B, though these are unconfirmed by the company. | 低 | SI009, SI013 |
| CI030 | Post-NYDIG sale, Crusoe's revenue mix shifted from approximately 55% BTC mining / 45% AI cloud (2024) to an estimated 95%+ AI cloud-only (2026), dramatically improving revenue quality. | 中 | SI011, SI018 |
| CI031 | Crusoe disclosed 17x TCV growth in 2025 without disclosing the absolute dollar base of TCV, making the metric unquantifiable without a denominator. | 高 | SI027, SI029 |
| CI032 | Crusoe has not publicly disclosed net revenue retention (NRR) or gross revenue retention (GRR) metrics as of May 2026. | 高 | SI009, SI010 |
| CI033 | Crusoe is a private company with no SEC reporting obligations; no audited financial statements are publicly available as of May 2026. | 高 | SI004, SI005 |
| CI034 | Heatmap News (March 2025) reported that Crusoe uses natural gas generation and has an Engine No. 1 JV to develop 4.5 GW of new gas-fired AI data center capacity, contradicting its clean-energy brand. | 中 | SI016, SI026 |
| CI035 | Crusoe Spark modular AI factory units are approximately 1 MW each and represent a potential product-sale and managed-service revenue stream; the Spark Factory in Brighton CO begins first deliveries in Q3 2026. | 中 | SI019, SI002 |
| CI036 | Each Crusoe Spark modular AI factory unit is approximately 1 MW in capacity and is designed for 3-month deployment versus the industry norm of 3–5 years for greenfield data centers. | 中 | SI002, SI019 |
| CI037 | Redwood Materials expanded its Crusoe Spark deployment from 4 units to 24 units (a 7x increase in AI infrastructure density), providing the earliest commercial validation of Spark unit revenue at scale. | 中 | SI021, SI019 |
| CI038 | Crusoe's PUE of 1.2–1.3 versus the industry average of approximately 1.8 provides approximately 30–40% lower power overhead cost per unit of compute delivered. | 中 | SI018, SI020 |
| CI039 | Blended energy cost for AI data center operators is estimated at $0.03–0.06/kWh; H100 GPU TDP is approximately 700W, implying energy cost of $0.02–0.04 per GPU-hour at Crusoe's PUE. | 中 | SI028, SI022 |
| CI040 | Crusoe has not publicly disclosed its cash on hand, balance sheet, or monthly cash burn rate as of May 2026. | 高 | SI009, SI010 |
| CI041 | Post-Series E, Crusoe's operational runway appears multi-year based on the $1.375B equity raise, but the exact runway in months and capex draw schedule have not been publicly disclosed. | 低 | SI003, SI008 |
| CI042 | H100 GPUs are typically depreciated over 3–5 years; a 4-year depreciation schedule at 80% utilization yields an estimated capital charge of approximately $0.60–1.20 per GPU-hour. | 中 | SI013, SI022 |
| CI043 | Crusoe's premium ESG branding could erode pricing power if enterprise customers discover its ongoing natural gas energy use, creating an adverse pricing compression risk. | 中 | SI016, SI026 |
| CI044 | Major hyperscalers (AWS, Azure, GCP) compete with Crusoe on GPU cloud and have structural cross-subsidization advantages through bundled services, posing long-term pricing pressure. | 中 | SI013, SI015 |
| CI045 | Crusoe's reserved-capacity discount structure for large GPU deals has not been publicly disclosed; list pricing may not reflect realized ASP for hyperscaler and enterprise contracts. | 高 | SI001, SI009 |
| CI046 | The NYDIG transaction consideration for Crusoe's mining asset sale is not publicly disclosed; any earnout or contingent consideration would affect Crusoe's capital position. | 中 | SI011, SI009 |
| CI047 | JPMorgan's $9.6B project financing for Abilene is the largest single AI infrastructure financing arrangement disclosed for Crusoe; covenant terms and draw conditions have not been published. | 中 | SI020, SI025 |
| CI048 | Crusoe's primary go-to-market motion is direct enterprise sales to AI/ML engineering teams and hyperscaler procurement, supplemented by self-serve access via docs.crusoecloud.com and GitHub; specific CAC and sales cycle length remain undisclosed. | 中 | SI023, SI002 |
| CE001 | Modular AI Factory (MAF) units are self-contained, prefabricated containers integrating GPU compute, primary power distribution, liquid cooling, and UPS within a single factory-built module operating at 1-2 MW IT load. | 高 | SE023, SE003 |
| CE002 | Campus stand-up time with MAF units is 6-8 months versus 24-36 months for traditional data centers, a consequence of factory pre-commissioning that eliminates on-site commissioning work. | 高 | SE023, SE017 |
| CE003 | Crusoe's Brighton, Colorado manufacturing facility is 350,000 sq ft and represents approximately $200M in investment, with commercial deliveries targeted for Q3 2026. | 中 | SE017, SE023 |
| CE004 | At peak production the Brighton facility is expected to produce 50-100 MAF units per quarter, equipping a new 200 MW campus roughly every 90 days. | 中 | SE017, SE023 |
| CE005 | The primary deployed GPU is the NVIDIA H100 SXM5 (80 GB HBM3); H200 SXM5 (141 GB HBM3e) achieved general availability on Crusoe Cloud in December 2025. | 高 | SE020, SE003 |
| CE006 | B200 SXM5 instances (192 GB HBM3e) are targeted for general availability in Q3 2026; GB200 NVL72 racks (72 Blackwell GPUs, approximately 5 PFLOPS peak FP8 per rack) are planned for Abilene Phase 2. | 高 | SE024, SE011 |
| CE007 | Intra-cluster networking uses NVIDIA Quantum-2 InfiniBand at 400 Gbps per link, providing GPU-to-GPU bisection bandwidth suitable for 512+ GPU distributed training workloads. | 高 | SE003, SE004 |
| CE008 | Abilene Phase 1 achieved 200 MW IT load with a 765 kV high-voltage transmission interconnect, becoming operational in Q1 2026. | 高 | SE016, SE012 |
| CE009 | Crusoe Cloud exposes on-demand, reserved (1- or 3-year), and spot compute tiers accessible via REST API, CLI, and Terraform provider. | 高 | SE003, SE004 |
| CE010 | The inference API layer implements the OpenAI Chat Completions and Completions v1 API surface, enabling drop-in substitution for applications built against the OpenAI API. | 高 | SE021, SE003 |
| CE011 | Supported inference models include Llama 3.1/3.2/3.3, Mistral, Mixtral, DeepSeek-R1/V3, Qwen2.5, and Phi-4, served via NVIDIA TensorRT-LLM as the backend. | 高 | SE021, SE003 |
| CE012 | Crusoe File System (CFS) is a Lustre-based parallel distributed file system supporting GPUDirect RDMA, with ReadWriteMany Kubernetes persistent volumes via a CSI driver. | 高 | SE006, SE005 |
| CE013 | Block storage uses NVMe-over-Fabric at up to 100 TB per instance; S3-compatible object storage and NFS volumes complete the Crusoe Cloud storage portfolio. | 高 | SE026, SE004 |
| CE014 | Inter-zone bandwidth is 100 Gbps Ethernet; intra-pod GPU-to-GPU bandwidth is 400 Gbps InfiniBand. | 高 | SE025, SE007 |
| CE015 | Edge Zones, launched March 2026, extend the Crusoe platform to third-party colocation sites with 1-5 MW distributed inference nodes targeting sub-10 ms inference latency. | 高 | SE009, SE003 |
| CE016 | Container images are maintained at ghcr.io/crusoecloud on GitHub Container Registry; the open-source Terraform provider (Apache 2.0) has 200+ commits indicating active maintenance. | 中 | SE001, SE005 |
| CE017 | Crusoe achieved ISO 27001:2022 certification in Q2 2025 and ISO 42001:2023 (AI management system) certification in Q3 2025, both validated by UKAS-accredited DNV GL. | 高 | SE014, SE022 |
| CE018 | Fewer than a dozen cloud providers globally held ISO 42001 certification as of mid-2026, making Crusoe's dual ISO certification a differentiated compliance credential. | 中 | SE022, SE014 |
| CE019 | SOC 2 Type II compliance is not reported by Crusoe as of Q2 2026, representing a compliance gap for regulated enterprise buyers in financial services, healthcare, and government. | 高 | SE022, SE004 |
| CE020 | VPCs with subnet-level firewall rules and SAML 2.0 plus OIDC federated identity via Keystone IAM support enterprise SSO integration for multi-tenant deployments. | 高 | SE025, SE004 |
| CE021 | Crusoe is listed as an HuggingFace Inference Endpoints partner, enabling one-click deployment of Hub models on Crusoe Cloud directly from the HuggingFace interface. | 高 | SE002, SE021 |
| CE022 | Together.ai published benchmarks showing Crusoe H100 clusters achieve competitive tokens-per-second on Llama 3.1 405B versus major cloud providers. | 中 | SE015, SE003 |
| CE023 | Crusoe Command Center provides unified dashboard visibility into GPU fleet health, power utilization, thermal status, and workload scheduling across all campuses. | 高 | SE008, SE003 |
| CE024 | Crusoe's multi-year NVIDIA strategic collaboration covers H100, H200, B200, and GB200 GPU allocations with co-engineering on power and cooling optimization. | 高 | SE019, SE011 |
| CE025 | In Q1 2026 Crusoe announced a multi-year agreement with Form Energy for iron-air long-duration battery storage at future Crusoe campuses. | 高 | SE010, SE003 |
| CE026 | Signisys provided grid connection engineering and high-voltage substation design for the Abilene 765 kV interconnect that powers Abilene Phase 1. | 高 | SE018, SE016 |
| CE027 | Upstream Data supplies oil-field flare-gas power generation units for stranded-gas compute deployments; this represents Crusoe's earlier heritage now being phased out in favor of large-campus scale. | 中 | SE013, SE012 |
| CE028 | BYO-Power program allows hyperscale customers to co-locate GPU compute at customer-owned substation sites with behind-the-meter generation assets. | 高 | SE007, SE012 |
| CE029 | At least one hyperscale partner at Abilene provides approximately 300 MW of dedicated behind-the-meter power under the BYO-Power arrangement. | 中 | SE016, SE007 |
| CE030 | The Crusoe Cloud platform runs on OpenStack with Keystone IAM, providing proven hypervisor isolation and multi-tenant compute separation. | 高 | SE025, SE004 |
| CE031 | MAF units enable campus-scale AI factories from 30 MW to over 1 GW through modular aggregation of individual 1-2 MW compute containers. | 高 | SE012, SE023 |
| CE032 | Kubernetes node pools on Crusoe Cloud support autoscaling and CFS ReadWriteMany persistent volumes via CSI driver for ML training checkpointing workflows. | 高 | SE005, SE006 |
| CE033 | Node-level NVLink and NVSwitch support intra-server all-reduce operations in multi-GPU H100, H200, and B200 training nodes. | 高 | SE011, SE003 |
| CE034 | NVIDIA GTC 2026 session S72619 documents the expanded Crusoe-NVIDIA collaboration and joint B200/GB200 roadmap co-presentation. | 高 | SE011, SE019 |
| CE035 | Abilene uses 765 kV high-voltage transmission interconnection specifically to minimize step-down transformer losses from generation to compute load. | 高 | SE018, SE016 |
| CE036 | Crusoe campuses range from 30 MW to over 1 GW; the BYO-Power model and behind-the-meter generation are key differentiators for cost-efficient large-scale AI deployments versus traditional cloud providers. | 高 | SE012, SE007 |
| CE037 | The Crusoe Terraform provider is open-source under Apache 2.0 with 200+ commits, used by customers for infrastructure-as-code deployments indicating active community adoption. | 中 | SE001, SE004 |
| CE038 | The Crusoe inference API uses NVIDIA TensorRT-LLM backend for hardware-optimized throughput, providing efficiency gains over naive PyTorch serving for large language model inference. | 高 | SE021, SE011 |
| CE039 | Crusoe Cloud container images are stored and served via GitHub Container Registry (ghcr.io/crusoecloud), reducing dependency on third-party container registries. | 中 | SE001, SE003 |
| CE040 | Each GB200 NVL72 rack delivers approximately 5 PFLOPS peak FP8 throughput across 72 Blackwell GPUs, a significant compute density improvement over H100 SXM5 per rack. | 中 | SE011, SE024 |
| CE041 | Crusoe's S3-compatible object storage enables drop-in compatibility with existing ML pipeline tooling (boto3, DVC, etc.) without API changes. | 高 | SE026, SE004 |
| CE042 | HuggingFace Inference Partner designation directly addresses developer acquisition by surfacing Crusoe in the HuggingFace Hub deployment menu for fine-tuned model serving. | 高 | SE002, SE021 |
| CE043 | Form Energy iron-air batteries provide 100+ hour discharge duration, addressing grid-intermittency risk for renewable-powered campuses better than lithium-ion alternatives. | 中 | SE010, SE022 |
| CE044 | Edge Zones are designed for real-time inference, video analytics, and sovereign-AI workloads where central cloud latency is unacceptable. | 高 | SE009, SE003 |
| CE045 | Crusoe's NVIDIA collaboration provides forward visibility into GPU roadmap milestones and allocation priority during supply-constrained periods, a competitive advantage over non-strategic NVIDIA customers. | 高 | SE019, SE011 |
| CE046 | MAF liquid cooling manifolds are factory-installed and pre-tested, eliminating the most failure-prone commissioning step in traditional data center builds and reducing on-site labor requirements. | 中 | SE023, SE017 |
| CE047 | Crusoe's entire GPU fleet is NVIDIA-based with no disclosed AMD MI300X or Intel Gaudi 3 GPU roadmap, creating a single-vendor dependency risk across the entire compute stack. | 高 | SE019, SE024 |
| CE048 | No third-party benchmark or independent audit for CFS aggregate throughput or cluster-level model FLOP utilization (MFU) has been published as of Q2 2026. | 中 | SE006, SE015 |
| CU001 | Crusoe's GPU cloud customer base spans hyperscaler anchor tenants, AI product companies, industrial enterprises, and research organizations as of early 2026. | 高 | SU001, SU003, SU008, SU019 |
| CU002 | Microsoft committed 900 MW of AI factory capacity at Abilene, Texas to Crusoe in March 2026 as a long-term anchor tenant agreement. | 高 | SU003, SU029 |
| CU003 | Oracle/Stargate is the anchor tenant in the first Crusoe Abilene buildings, connecting Crusoe's infrastructure to OpenAI's Stargate AI data center consortium. | 中 | SU008 |
| CU004 | Redwood Materials' Crusoe Spark deployment achieved 99.2% uptime over seven months of production operation. | 高 | SU019, SU004 |
| CU005 | Redwood Materials expanded its Crusoe Spark deployment from 4 to 24 units, representing a 7x increase in AI infrastructure density. | 高 | SU002, SU019 |
| CU006 | Cursor is a production Crusoe Cloud customer, using GPU infrastructure for its AI coding assistant product. | 中 | SU005, SU001 |
| CU007 | Together AI is a production Crusoe Cloud customer using the platform for LLM training and inference workloads. | 中 | SU006, SU001 |
| CU008 | Sony Research used Crusoe Cloud GPU infrastructure to train the Gran Turismo Sophy (GT Sophy) reinforcement learning racing agent. | 中 | SU019 |
| CU009 | Crusoe's named AI product company customers include Fireworks, Luma AI, Windsurf (Codeium), Databricks, Decart, and Playground AI in addition to Cursor and Together AI. | 中 | SU001, SU019, SU010 |
| CU010 | A 45-hour outage in early March 2025 affected Crusoe Cloud customers and raised reliability concerns about the platform's production readiness. | 中 | SU007 |
| CU011 | Crusoe has not disclosed NRR, GRR, customer churn rate, or cohort retention metrics for its GPU cloud business. | 高 | SU010, SU019 |
| CU012 | Crusoe Cloud lists on-demand H100 GPU pricing publicly at $2.99/hr as of May 2026, down from peak market rates of approximately $8/hr in 2023. | 高 | SU026, SU011 |
| CU013 | Crusoe's AI cloud revenue reached $124M in 2024, representing 460% year-over-year growth. | 中 | SU019, SU010 |
| CU014 | Crusoe reported 150% year-over-year ARR growth for its AI cloud business in 2025. | 中 | SU021, SU010 |
| CU015 | No public customer complaints, churn events, or formal dissatisfaction reports were found for Crusoe Cloud customers beyond the March 2025 outage incident. | 中 | SU007, SU020 |
| CU016 | Crusoe has not disclosed its total paying customer count or per-customer revenue breakdown. | 高 | SU010, SU019 |
| CU017 | Crusoe's GitHub organization (crusoecloud) contains open-source tools including a Terraform provider and client SDKs, indicating active developer adoption. | 中 | SU014, SU013 |
| CU018 | Crusoe AI's HuggingFace page (crusoeai) hosts model resources indicating engagement with the ML research and developer community. | 中 | SU015 |
| CU019 | Microsoft and Oracle together likely represent more than 50% of Crusoe's contracted Abilene campus capacity, creating hyperscaler customer concentration risk. | 中 | SU003, SU008, SU011 |
| CU020 | Crusoe achieved ISO 27001 (information security management) and ISO 42001 (AI management) certifications, reducing enterprise customer security review friction. | 高 | SU017, SU001 |
| CU021 | Crusoe's Command Center platform provides unified operations and observability for GPU cluster workloads, creating switching costs and supporting customer retention. | 高 | SU025, SU001 |
| CU022 | Crusoe Cloud serves AI training, batch inference, fine-tuning, production inference, reinforcement learning, and video generation workloads across customer segments. | 高 | SU001, SU013, SU019 |
| CU023 | Sacra estimates Crusoe's GPU cloud ARR trajectory is consistent with the company's stated 150% YoY growth claims for 2025. | 中 | SU010 |
| CU024 | The GPU-as-a-service market is projected to grow at significant CAGR through 2028 per MarketsAndMarkets and IDC forecasts, validating Crusoe's addressable market expansion. | 中 | SU012, SU027 |
| CU025 | NVIDIA's strategic collaboration with Crusoe gives Crusoe customers preferred access to H200, B200, and GB200 GPU supply during periods of hardware scarcity. | 高 | SU016, SU013, SU031 |
| CU026 | The 45-hour March 2025 outage at Crusoe Cloud exceeded standard 99.9% monthly SLA thresholds and may have triggered enterprise SLA remediation obligations. | 中 | SU007 |
| CU027 | Crusoe's vertical integration model enables more cost-effective GPU compute pricing than hyperscaler alternatives at scale, per company positioning and analyst commentary. | 中 | SU001, SU018, SU030 |
| CU028 | No public reviews, Gartner Peer Insights entries, or G2 ratings were found for Crusoe Cloud, limiting independent customer satisfaction data. | 中 | SU010, SU011 |
| CU029 | Heatmap News reported that Crusoe's use of natural gas generation at Abilene undermines its clean-energy positioning and may create procurement friction for ESG-sensitive enterprise customers. | 中 | SU020 |
| CU030 | PitchBook confirms Crusoe has raised $3.4B+ across multiple equity rounds, demonstrating investor-validated commercial credibility for enterprise customer procurement decisions. | 中 | SU024, SU021 |
| CU031 | Crusoe's 2024 Impact Report publicly documented Redwood Materials' 99.2% uptime result over seven months of Spark deployment. | 高 | SU019, SU002 |
| CU032 | Crusoe's public pricing page lists on-demand GPU compute enabling smaller customers to access infrastructure without long-term contracts, lowering switching costs for price-sensitive buyers. | 高 | SU026, SU011 |
| CU033 | IDC forecasts strong AI infrastructure spending growth through 2028, benefiting Crusoe's large enterprise and hyperscaler customer segment expansion. | 中 | SU027 |
| CU034 | Crusoe was named to Fast Company's list of the World's Most Innovative Companies of 2026, supporting enterprise customer brand recognition. | 中 | SU021, SU032 |
| CU035 | Crusoe's Spark modular AI factory units enable customer deployments at industrial sites and edge locations beyond traditional data center formats. | 高 | SU009, SU002, SU035 |
| CU036 | The neocloud sector has attracted $20B+ in GPU-focused capital, with Crusoe, CoreWeave, Lambda Labs, and Nebius competing for AI workload customers in 2026. | 中 | SU028, SU011, SU033, SU034 |
| CU037 | Form Energy is a Crusoe strategic supply partner providing 12 GWh of iron-air batteries starting 2027 and is not a GPU cloud customer. | 高 | SU009, SU003 |
| CU038 | Crusoe and Form Energy announced a formal agreement for 12 GWh of iron-air long-duration battery storage to support Crusoe AI data centers, with deliveries starting in 2027. | 高 | SU036, SU035 |
| CU039 | Crusoe launched Edge Zones in March 2026 to extend compute capacity to distributed inference locations, broadening the addressable customer base beyond centralized AI factory tenants. | 中 | SU037 |
| CU040 | Crusoe and NVIDIA presented jointly at GTC 2026 on AI factory infrastructure, reinforcing the partnership and highlighting Crusoe's customer-facing integrated stack. | 中 | SU039 |
| CU041 | Upstream Data filed a patent lawsuit against Crusoe related to flare-gas capture technology in 2024; the case was resolved November 24, 2025 via mutual dismissal and license agreement. | 高 | SU040, SU004 |
| CU042 | Crusoe's LinkedIn profile (as Crusoe Energy Systems) lists 1,235 employees as of March 31, 2026, reflecting significant workforce growth aligned with rapid infrastructure scaling. | 中 | SU038, SU041 |
| CU043 | Anthropic, Mistral AI, and OpenAI represent the archetype of frontier AI model companies that constitute Crusoe's highest-value on-demand and contract GPU cloud customer segment. | 中 | SU042, SU043, SU044 |
| CU044 | The a16z AI portfolio and broader VC-backed AI startup ecosystem represents a structural demand driver for GPU cloud services; Crusoe's on-demand product targets this developer-led segment. | 中 | SU045, SU046 |
| CU045 | Lancium operates renewable-powered HPC and AI compute campuses in West Texas with behind-the-meter power, validating the market for clean-energy GPU infrastructure and the same customer segment Crusoe targets. | 中 | SU047 |
| CU046 | Crusoe's hiring for cloud account executives and enterprise customer success roles targeting Fortune 500 and AI lab accounts indicates a structured enterprise go-to-market motion beyond self-serve developer acquisition. | 中 | SU048 |
| CR001 | TCEQ permit filings confirmed 360 MW of on-site gas turbines at the Abilene campus, representing a material regulatory and environmental compliance risk. | 高 | SR001, SR006 |
| CR002 | Crusoe's Abilene gas turbines likely exceed the EPA Greenhouse Gas Reporting Program threshold of 25,000 metric tons CO2e/year, triggering mandatory annual public emissions reporting under 40 CFR Part 98. | 中 | SR003, SR006 |
| CR003 | The Abilene campus relies primarily on ERCOT grid power, which experienced catastrophic failure during Winter Storm Uri in February 2021; repeat curtailment events could interrupt hyperscaler SLAs. | 中 | SR002, SR022, SR037 |
| CR004 | The Upstream Data patent lawsuit was resolved on November 24, 2025 via mutual dismissal and a license agreement, removing the litigation overhang from Crusoe's DFM technology. | 高 | SR004, SR005 |
| CR005 | The terms of the Upstream Data license agreement—including royalties, field-of-use restrictions, and duration—have not been publicly disclosed. | 高 | SR004, SR005 |
| CR006 | Crusoe's Mubadala Capital (Abu Dhabi sovereign wealth) investment may require CFIUS review if Crusoe pursues US government AI compute contracts or files an IPO S-1. | 中 | SR023, SR034 |
| CR007 | Heatmap News reported that Crusoe is working with Engine No. 1 to develop 4.5 GW of new natural gas generation capacity for AI data center demand, contradicting its clean-energy positioning. | 中 | SR006 |
| CR008 | No regulatory enforcement actions, sanctions notices, export control inquiries, or OFAC flags were found for Crusoe in the research period. | 中 | SR001, SR003, SR023 |
| CR009 | A 45-hour outage in early March 2025 affected Crusoe Cloud customers; Forbes reported the event but no post-mortem, SLA terms, or customer remediation details were publicly disclosed. | 中 | SR008 |
| CR010 | The March 2025 45-hour outage likely exceeded standard 99.9% monthly uptime SLA thresholds, which permit only approximately 43 minutes of downtime per month. | 中 | SR008, SR025 |
| CR011 | Crusoe deploys direct-to-chip liquid cooling at Abilene, which achieves PUE 1.2–1.3 but requires specialized redundancy infrastructure; Abilene's summer temperatures (~98°F peak) create sustained cooling load stress. | 中 | SR007, SR021 |
| CR012 | NVIDIA is Crusoe's primary GPU supplier for H100, H200, B200, and GB200 hardware; preferred allocation is documented via strategic collaboration agreement but is not contractually guaranteed. | 高 | SR009, SR010 |
| CR013 | Crusoe achieved ISO 27001 (information security management) and ISO 42001 (AI management system) certifications, providing documented cybersecurity controls but not eliminating cloud security risk. | 高 | SR024, SR007 |
| CR014 | Crusoe transitioned from Bitcoin and DFM operations (divested to NYDIG in March 2025) to pure AI cloud, removing $152M in 2024 DFM/BTC revenue and creating a ramp-dependent financial profile. | 高 | SR013, SR014 |
| CR015 | H100 spot GPU prices fell from approximately $8/hr in 2023 to $2-3/hr in 2025, compressing on-demand GPU cloud margins industry-wide and creating pricing pressure on Crusoe's non-anchor AI startup customers. | 高 | SR015, SR016, SR035 |
| CR016 | Crusoe's multi-tenant GPU cloud faces cybersecurity risks from nation-state actors targeting AI model weights and training data; no major breach was reported in the research period. | 中 | SR024, SR025 |
| CR017 | Crusoe's Abilene build-out requires more than $15B in total committed capital: $3.4B Blue Owl JV equity, $9.6B JPMorgan project financing, and Crusoe's own equity contributions. | 高 | SR013, SR034, SR036 |
| CR018 | The JPMorgan $9.6B project financing for Abilene contains covenants and milestone triggers that are not publicly disclosed; a breach could force asset sales or dilutive equity issuance. | 中 | SR013, SR014, SR042 |
| CR019 | NVIDIA represents a critical single-supplier GPU concentration risk; no viable alternative at GB200 scale exists, with AMD MI300X offering only a partial substitute. | 中 | SR009, SR010, SR015 |
| CR020 | Microsoft's 900 MW Abilene commitment likely represents more than 30% of Phase 1 contracted capacity and is Crusoe's highest single-counterparty concentration risk. | 中 | SR011, SR012, SR035, SR040 |
| CR021 | CoreWeave, Lambda Labs, Nebius, and Lancium offer competing GPU cloud capacity at H100/H200 pricing that tracks or undercuts Crusoe's on-demand rates, creating churn risk for price-sensitive AI startup customers. | 中 | SR026, SR027, SR038, SR039 |
| CR022 | JPMorgan is the single arranger for $9.6B in Abilene project financing, creating capital-markets counterparty concentration risk if JPMorgan's credit appetite changes. | 中 | SR013, SR014, SR042 |
| CR023 | Blue Owl Capital committed $3.4B in a JV for Abilene data center construction; JV covenant terms and risk-sharing provisions are not publicly disclosed. | 中 | SR013, SR034, SR036 |
| CR024 | Form Energy's iron-air battery technology is early commercial stage; the 12 GWh deployment starting 2027 carries execution risk from technology underperformance or delivery delay. | 中 | SR028, SR029 |
| CR025 | Crusoe's Spark Factory supply chain for modular AI factory units is not yet operational as of early 2026; first deliveries are targeted for Q3 2026 from the Brighton CO facility. | 高 | SR019, SR020 |
| CR026 | Crusoe's headcount nearly doubled from approximately 600 at end-2024 to 1,235 on March 31, 2026, creating organizational scaling execution risk including culture dilution and management bandwidth constraints. | 高 | SR017, SR018 |
| CR027 | Co-founders Chase Lochmiller (CEO) and Cully Cavness (President) represent elevated key-person risk; their departure would trigger investor confidence loss and potentially impair the energy sourcing pipeline. | 中 | SR017, SR011 |
| CR028 | Michael Gordon joined as COO and CFO in December 2025 with MongoDB IPO experience, bringing public-market financial discipline but representing a near-term key-person risk as a recent hire. | 高 | SR017, SR018 |
| CR029 | The Wyoming campus (1.8 GW announced) has not begun construction as of May 2026; power access, permitting, and financing remain unconfirmed. | 中 | SR021, SR022 |
| CR030 | The Spark Factory in Brighton, Colorado targets first modular AI factory unit deliveries in Q3 2026; delays in commissioning or equipment procurement would constrain Crusoe's non-Abilene customer pipeline. | 高 | SR019, SR020 |
| CR031 | Crusoe's SEC Form D filings (Series D and Series E) are publicly available on EDGAR and disclose offering size, entity type, and officer names, but not revenue, margins, or covenant terms. | 高 | SR023, SR034 |
| CR032 | Crusoe's ESG greenwashing risk is the most significant reputational risk: the company markets clean AI infrastructure while operating 360 MW of gas turbines and developing 4.5 GW of new gas capacity. | 高 | SR006, SR007, SR031, SR041 |
| CR033 | Crusoe's 2024 Impact Report acknowledges its energy mix but does not provide a per-GPU or per-MWh carbon intensity figure, making independent ESG assessment difficult. | 中 | SR007 |
| CR034 | Crusoe's transition from Bitcoin/DFM revenue creates a ramp-dependent AI cloud financial profile where the $15B+ Abilene build-out must be serviced primarily by AI cloud revenue by 2026-2027. | 中 | SR013, SR014, SR033 |
| CR035 | GPU pricing pressure (H100 $8→$2-3/hr), hyperscaler anchor concentration, and JPMorgan covenant triggers collectively represent the highest-severity financial risk cluster for the Crusoe investment thesis. | 中 | SR015, SR016, SR011, SR013 |
| CR036 | Physical disaster risk (tornado, ice storm, heat dome) at the Abilene TX campus is low probability but critical severity given geographic GPU asset concentration worth $2-5B. | 中 | SR002, SR021 |
| CR037 | Crusoe's SEC Form D for Series E (filed October 2025, CIK 0001924674) confirms 71 investors and $1.375B raise; no material misstatement or disclosure failure was identified in the research period. | 高 | SR023, SR034 |
| CR038 | The Crusoe-NVIDIA collaboration includes access to GB200 NVL72 systems and participation in NVIDIA's Vera CPU early adopter program, creating a multi-year technology bet on unproven silicon. | 中 | SR009, SR010 |
| CR039 | No SOC 2 Type II audit report or formal penetration testing disclosure was found publicly for Crusoe Cloud, representing an open cybersecurity diligence gap. | 中 | SR024, SR025 |
| CR040 | Crusoe's natural gas use and Engine No. 1 gas development partnership create a Scope 1 emissions exposure that ESG-linked debt markets and IPO investors will scrutinize under climate disclosure frameworks. | 中 | SR006, SR031, SR032 |
| CR041 | AWS P5 (H100) and Google Cloud A3 instances offer comparable GPU compute with hyperscaler brand trust and higher SLA maturity, representing a risk to Crusoe's enterprise and mid-market customer retention. | 中 | SR038, SR039 |
| CR042 | Lancium operates sustainable HPC computing in Texas that competes with Crusoe for AI workloads and power procurement in the same ERCOT market, representing a direct Texas-geography competitive risk. | 中 | SR037, SR002 |
| CR043 | The $9.6B JPMorgan project financing raised Crusoe's total capital stack to more than $15B; Series E documentation confirmed the financing terms and Blue Owl JV structure in October 2025. | 高 | SR042, SR013 |
| CR044 | ERCOT, the Texas electricity grid operator, manages approximately 90% of Texas load; Crusoe's Abilene campus in Taylor County is subject to ERCOT wholesale price volatility and grid reliability constraints that directly affect electricity cost risk. | 高 | SR043, SR006 |
| CR045 | Lambda Labs publicly offers H100 SXM5 on-demand cloud GPU instances at rates comparable to or below Crusoe's AI factory pricing, representing a direct competitive pricing risk for Crusoe's cloud GPU products. | 中 | SR044, SR011 |
| CV001 | Crusoe Inc. SEC Form D (filed October 23, 2025) confirms total Series E proceeds of $1,374,999,988 at a $10B+ post-money valuation, co-led by Valor Equity Partners and Mubadala Capital with 71 total investors. | 高 | SV001, SV004, SV019 |
| CV002 | Crusoe Inc. SEC Form D (filed November 21, 2024) confirms the Series D at $600M led by Founders Fund with NVIDIA, Fidelity, Mubadala, Ribbit Capital, and Valor participating. | 高 | SV002, SV018, SV020 |
| CV003 | EDGAR confirms Crusoe Inc. (CIK 0001924674) filed Form D for both the Series D (November 2024) and Series E (October 2025) exempt offerings. | 高 | SV003, SV001, SV002 |
| CV004 | The Crusoe valuation step-up from Series D ($2.8B, December 2024) to Series E ($10B+, October 2025) represents a 3.6x increase in approximately ten months. | 高 | SV001, SV002, SV004 |
| CV005 | Crusoe has raised approximately $3.4B in total equity financing across six rounds since its 2019 seed, including the $1.375B Series E confirmed by SEC Form D filing. | 中 | SV001, SV002, SV008 |
| CV006 | Blue Owl Capital committed $3.4B in a joint venture for the Abilene, Texas data center campus construction, separate from Crusoe's equity capital raises. | 中 | SV007, SV009 |
| CV007 | JPMorgan is arranging approximately $9.6B in project financing for the Abilene campus, bringing total committed capital for the Abilene buildout to over $15B. | 中 | SV007, SV017 |
| CV008 | Total committed capital for Crusoe exceeds $15B combining the $3.4B equity, $3.4B Blue Owl JV, and $9.6B JPMorgan project financing for the Abilene campus. | 中 | SV007, SV017, SV009 |
| CV009 | Crusoe reported 2024 full-year revenue of $276M representing 82% year-over-year growth, with AI cloud contributing $124M (460% YoY) and Bitcoin mining contributing $152M. | 中 | SV009, SV022 |
| CV010 | The March 2025 sale of Crusoe's Bitcoin mining operations to NYDIG was completed, making AI cloud the sole business line and clarifying the investment thesis as pure AI infrastructure. | 中 | SV009, SV016 |
| CV011 | Crusoe reported 150% year-over-year cloud ARR growth in 2025 and 17x total contract value (TCV) growth; both figures are company-claimed and not yet verified by audited financial statements. | 中 | SV019, SV022, SV024 |
| CV012 | Analyst estimates for Crusoe's 2025 full-year revenue range from approximately $500M to $1B; these figures are unconfirmed and have not been verified by audited financial statements. | 低 | SV009, SV012 |
| CV013 | At the $10B Series E valuation with $500M estimated 2025 revenue, the implied revenue multiple is approximately 20x; at $1B estimated 2025 revenue the implied multiple is approximately 10x. | 中 | SV001, SV009 |
| CV014 | At 2024 full-year revenue of $276M including Bitcoin mining, the $10B Series E valuation implies approximately 36x trailing revenue — stretched relative to public infrastructure comparables. | 中 | SV001, SV009, SV022 |
| CV015 | CoreWeave completed its IPO in March 2025 at an implied valuation of approximately $23B based on IPO price per share times total diluted shares outstanding. | 中 | SV010, SV011, SV017 |
| CV056 | TechCrunch, Fortune, and CNBC coverage of the Crusoe Series E and CoreWeave IPO corroborate the $10B+ valuation mark and the approximately 4.6x ARR CoreWeave public benchmark as of October 2025. | 中 | SV033, SV034, SV035 |
| CV057 | Applied Digital operates power-first AI data center infrastructure with a model similar to Crusoe's but trades at a public market valuation substantially below the $10B Crusoe private mark, confirming that public market discipline applies stricter multiples than recent private round pricing. | 低 | SV036, SV037 |
| CV058 | Blue Owl Capital's joint venture and JPMorgan's project financing commitment to Crusoe represent institutional-quality validation of the Abilene campus investment thesis and provide off-balance-sheet leverage that expands Crusoe's effective capital base beyond the $3.4B in equity raised. | 中 | SV038, SV039 |
| CV059 | Grand View Research and CB Insights AI infrastructure market data confirm long-run demand tailwinds for GPU cloud operators but neither source provides Crusoe-specific financial guidance or ARR forecasts. | 中 | SV040, SV037 |
| CV016 | CoreWeave was generating approximately $5B in annualized recurring revenue at IPO, implying a revenue multiple of approximately 4.6x ARR at the IPO entry price — far below Crusoe's current implied multiple. | 中 | SV010, SV011 |
| CV017 | CoreWeave shares declined from the IPO valuation mark in post-IPO trading, indicating the public market applied a more disciplined multiple than the private financing price suggested. | 中 | SV007, SV010 |
| CV018 | Nebius AI is listed on NASDAQ with a market capitalization of approximately $6B as of 2025 to 2026, representing a European GPU cloud positioned as a sector comparable to Crusoe. | 中 | SV027, SV012 |
| CV019 | Lambda Labs is estimated at approximately $1.5B in private market valuation by analyst sources; the company operates a developer-focused GPU cloud without proprietary power infrastructure. | 低 | SV028, SV012 |
| CV020 | Lancium operates a power-first AI data center business in Texas representing the closest strategic analog to Crusoe; no public valuation anchor exists for Lancium as of May 2026. | 低 | SV012 |
| CV021 | ComputeForecast and Signisys neocloud analyses document H100 GPU spot prices declining from approximately $8/hr to $2 to $3/hr through 2024 to 2025, a 60 to 75% price compression. | 中 | SV010, SV012 |
| CV022 | GPU spot price commoditization creates structural headwinds for neocloud revenue per GPU; operators without proprietary power or long-term contracted supply face potential unit economics compression. | 中 | SV010, SV012 |
| CV023 | Heatmap News reported that Crusoe's Abilene campus uses on-site natural gas generation and that the company engaged Engine No. 1 to develop 4.5 GW of new natural gas capacity for AI data centers. | 中 | SV014 |
| CV024 | TCEQ air quality permits confirm 360 MW of on-site gas turbines at the Abilene campus representing approximately 30% of Phase 1 capacity, creating tension with Crusoe's clean-energy branding. | 中 | SV014, SV022 |
| CV025 | Microsoft is the primary anchor tenant at the Crusoe Abilene 900 MW campus announced in March 2026, making it the largest single capacity and revenue commitment in Crusoe's history. | 中 | SV021, SV032 |
| CV026 | Customer revenue concentration is believed to be elevated with Microsoft and OpenAI likely accounting for a majority of Crusoe's AI cloud ARR; exact concentration is not publicly disclosed. | 中 | SV021, SV031, SV009 |
| CV027 | The Upstream Data patent lawsuit filed against Crusoe was resolved in November 2025, removing a legal overhang from the company's flare-gas intellectual property portfolio. | 中 | SV023 |
| CV028 | Crusoe's total power pipeline exceeds 45 GW globally with 3.4 GW commissioned electricity and 9.8 million square feet of facilities as of early 2026 per company reporting. | 中 | SV022, SV024, SV023 |
| CV029 | The Abilene, Texas campus has 2.1 GW of committed capacity following the March 2026 Microsoft announcement, with Phase 1 already underway and approximately 1.2 GW initially planned. | 中 | SV021, SV032 |
| CV030 | Crusoe reports a PUE of 1.2 to 1.3 enabled by direct-to-chip liquid cooling, providing efficiency advantages over air-cooled data centers and supporting the efficiency narrative. | 中 | SV022 |
| CV031 | MarketsAndMarkets projects significant compound annual growth in the GPU-as-a-service market through 2030, providing a macro tailwind for Crusoe's AI cloud revenue growth thesis. | 中 | SV011, SV013 |
| CV032 | IDC research on AI infrastructure buildout confirms accelerating enterprise spending on GPU cloud capacity, providing market context for Crusoe's positioning as a power-optimized AI cloud provider. | 中 | SV013 |
| CV033 | The bull case for Crusoe implies $1B+ ARR confirmation in 2025 followed by $2B ARR in 2026; at 10 to 12x forward ARR this implies $20B to $24B valuation, representing 2 to 2.4x from $10B entry. | 中 | SV009, SV019 |
| CV034 | The base case for Crusoe implies 2025 ARR confirmation at $750M to $1B; at 8 to 10x ARR multiple, fair value is approximately $6B to $10B — roughly in-line with or a modest discount to current mark. | 中 | SV009, SV012 |
| CV035 | The bear case for Crusoe implies 2025 revenue confirmation at $500M or below; at 5 to 6x ARR multiple, fair value falls to approximately $2.5B to $3B — a 70 to 75% discount to the $10B mark. | 中 | SV009, SV010 |
| CV036 | Crusoe's power-first vertical integration spanning energy sourcing, site development, construction, modular manufacturing, and cloud platform justifies a premium over pure GPU rental peers lacking proprietary infrastructure. | 中 | SV024, SV029 |
| CV037 | Michael Gordon joined Crusoe as COO and CFO in December 2025 with a track record of leading MongoDB's 2017 IPO and approximately 50x revenue growth, signaling formal IPO-path preparation. | 中 | SV023, SV019 |
| CV038 | NVIDIA expanded its strategic collaboration with Crusoe in 2025, providing preferred GPU supply access, engineering support, and technical validation from the dominant AI chip supplier. | 中 | SV025, SV026 |
| CV039 | Crusoe was named to Fast Company's list of the 100 Most Innovative Companies of 2026, providing brand recognition consistent with IPO-path public profile building. | 中 | SV023 |
| CV040 | The current evidence supports a track recommendation with medium confidence and high risk: extraordinary infrastructure traction is real but the $10B+ mark requires $1B+ ARR confirmation before underwriting. | 中 | SV001, SV009, SV010, SV022 |
| CV041 | No public filing, press release, or verified third-party source discloses Crusoe's gross margin, contribution margin, or EBITDA for any period; financial transparency for external investors is zero. | 中 | SV008, SV009, SV022 |
| CV042 | Crusoe employs approximately 1,235 people as of March 2026, up from approximately 600 at end-2024, representing a doubling of headcount alongside rapid infrastructure scaling. | 中 | SV023, SV024 |
| CV043 | Crusoe and Form Energy announced an agreement for 12 GWh of iron-air battery storage for AI data centers, representing a differentiated clean-energy storage approach to strengthen ESG positioning. | 中 | SV023 |
| CV044 | Crusoe launched a manufacturing facility in Brighton, Colorado to produce modular prefabricated AI factory units (Spark), creating a new revenue vector and associated inventory and supply-chain risk. | 中 | SV023, SV024 |
| CV045 | Bain Capital Ventures published an investment thesis framing power scarcity as the binding constraint for AI infrastructure and characterizing Crusoe's early grid rights as a durable competitive moat. | 中 | SV029 |
| CV046 | Crusoe confirmed 150% cloud ARR growth and 17x TCV growth in 2025 in Series E announcement materials; these are company-stated metrics consistent with rapid scaling from Series D to Series E. | 中 | SV019, SV022 |
| CV047 | Crusoe achieved ISO 27001 and ISO 42001 certifications in 2025, demonstrating enterprise-grade security and AI management system compliance that supports hyperscaler customer acquisition. | 中 | SV023 |
| CV048 | Forbes and Data Center Dynamics reporting on the Crusoe Abilene 900 MW Microsoft campus confirms Crusoe's strategic positioning as an anchor AI infrastructure partner for hyperscale customers. | 中 | SV031, SV032 |
| CV049 | The Series D was initially announced in September 2024 per Axios reporting and formally closed and confirmed by SEC Form D in November 2024, consistent with a multi-tranche close structure. | 中 | SV002, SV018 |
| CV050 | Contracted ARR (TCV growth 17x) substantially exceeds implied revenue recognition, creating potential deferred revenue or ramp-period dynamics that must be verified in private diligence to assess backlog quality. | 中 | SV019, SV009 |
| CV051 | Crusoe's cap-table structure after six rounds with multiple lead investors likely includes layered liquidation preferences that could impair common equity returns below the $10B mark in a moderate-exit or down-round scenario. | 中 | SV001, SV002, SV008 |
| CV052 | The Forbes IPO article (March 2025) referenced Crusoe's wind and solar branding while Heatmap News simultaneously documented natural gas use, creating an ESG credibility gap that is a material reputational risk with institutional investors. | 中 | SV015, SV014 |
| CV053 | A Wyoming campus of 1.8 GW has been announced in addition to the 2.1 GW Abilene commitment, expanding Crusoe's geographic diversification and power sourcing options beyond Texas. | 中 | SV024, SV023 |
| CV054 | The Signisys neocloud report characterizes the $20B+ GPU-focused provider market as undergoing consolidation, with power-integrated operators likely to gain share over pure GPU rental competitors. | 低 | SV012 |
| CV055 | Forbes profiled Crusoe's Stargate construction role in April 2025, confirming the OpenAI infrastructure relationship and simultaneously corroborating customer concentration concerns. | 中 | SV031 |
| CV062 | Statista projects the global cloud computing market to grow from approximately $0.7 trillion in 2024 to over $1.2 trillion by 2030, a CAGR that underpins the long-runway addressable market supporting Crusoe's $10B valuation. | 中 | SV041 |
| CV063 | CB Insights' AI Cloud Infrastructure report identifies GPU-as-a-service and power-integrated neocloud operators as the fastest-growing segment of the cloud market, corroborating Crusoe's strategic positioning and premium-to-market multiple thesis. | 中 | SV042 |
| CV064 | S&P Capital IQ comparable company databases show AI infrastructure and data center REIT peers trading at EV/Revenue multiples of 8x–22x for high-growth operators, a band within which Crusoe's 10x–20x implied multiple at $10B valuation is defensible. | 低 | SV043 |
| CV065 | DigitalBridge's portfolio of digital infrastructure assets provides a private equity comparable set for Crusoe; infrastructure-focused PE transactions in the data center sector have occurred at 14x–22x EV/EBITDA in 2023–2025, supporting a premium multiple for power-integrated AI factory operators. | 低 | SV044 |
| CV060 | SemiAnalysis estimates that next-generation GPU (Blackwell B200 cluster) all-in costs exceed $30M per MW of capacity, implying that Crusoe's 3.9 GW committed pipeline represents over $100B in potential infrastructure value at replacement cost, providing an asset-based valuation floor argument. | 低 | SV045 |
| CV061 | GPU cloud spot pricing on platforms such as CloudPrice shows H100 SXM5 instance rates ranging from $1.80–$3.50/GPU-hour in 2025, a market rate that, applied to Crusoe's disclosed 100,000-GPU capacity, implies maximum annual revenue capacity of $1.6B–$3.1B, consistent with the $500M–$1B revenue estimate at current utilization rates. | 低 | SV046 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Crusoe | Crusoe Homepage — The AI Factory Company | The AI factory company. We accelerate the abundance of energy and intelligence. |
| SO002 | Crusoe | Crusoe Newsroom — Series E Funding Announcement | Crusoe has raised $1.375 billion in its Series E financing, reaching a valuation above $10 billion. |
| SO003 | Crusoe | Crusoe Newsroom — Series D Funding Announcement | Crusoe has raised $600 million in a Series D financing at a valuation of $2.8 billion. |
| SO004 | U.S. Securities and Exchange Commission | SEC Form D — Crusoe Inc. Series E (Acc-no 0001924674-25-000004) | Total offering amount: $1,374,999,988; 71 investors; item 06b; filed 2025-10-23 |
| SO005 | U.S. Securities and Exchange Commission | SEC Form D — Crusoe Inc. Series D (Acc-no 0001924674-24-000003) | Notice of Exempt Offering; item 06c; filed 2024-11-21; Denver CO; Crusoe Inc. |
| SO006 | Data Center Dynamics | Crusoe Raises $1.375bn in Latest Funding Round | Crusoe, the neocloud infrastructure company formerly known as Crusoe Energy Systems, has raised $1.375bn in a Series E funding round at a $10bn+ valuation. |
| SO007 | Sacra | Crusoe Revenue, Valuation, and Growth (2024) | Crusoe generated $276M in revenue in 2024, with AI cloud revenue growing 460% to $124M. |
| SO008 | Crusoe | Crusoe Newsroom — Michael Gordon Appointed COO and CFO | Michael Gordon joins Crusoe as COO and CFO; previously served in both roles at MongoDB. |
| SO009 | Crusoe | Crusoe Newsroom — Nader Pakfar General Counsel Real Estate | Crusoe appoints Nader Pakfar as General Counsel Real Estate to support growing infrastructure development. |
| SO010 | Forbes | Crusoe's AI Factory Model Rewrites the Data Center Playbook | Crusoe has positioned itself as the only company that owns the full stack from power generation to GPU cloud. |
| SO011 | Bain Capital Ventures | Bain Capital Ventures: Crusoe Investment Thesis | Crusoe is pioneering the concept of the AI factory — clean, purpose-built infrastructure for the age of AI. |
| SO012 | Data Center Knowledge | Crusoe's Abilene Campus: Inside the World's Largest AI Factory | The Abilene campus will cover 1.2 GW of capacity, with first buildings online within 12 months of groundbreaking. |
| SO013 | Heatmap News | Crusoe's Clean Energy Claim Is Complicated by Its Natural Gas Footprint | Despite Crusoe's clean-energy branding, the company relies on natural gas and is developing 4.5 GW of new gas capacity through an Engine No. 1 joint venture. |
| SO014 | Texas Commission on Environmental Quality | TCEQ Air Permit Applications — Crusoe Abilene Campus | TCEQ permit database confirms 360 MW of on-site gas turbines permitted at Crusoe's Abilene campus. |
| SO015 | Crusoe | Crusoe Newsroom — NYDIG Acquires Crusoe Bitcoin Mining and DFM Operations | Crusoe has completed the sale of its Bitcoin mining and Digital Flare Mitigation operations to NYDIG. |
| SO016 | Upstream Data | Crusoe and Upstream Data Resolve Lawsuit (Press Release) | Crusoe Energy Systems LLC and Upstream Data Inc. have agreed to resolve all pending lawsuits related to patent infringement. Both parties have agreed to dismiss the lawsuits and enter into a license agreement on confidential terms. |
| SO017 | Lowercarbon Capital | Lowercarbon Capital Portfolio — Crusoe | Emissions-slashing compute clusters. Crusoe. |
| SO018 | Founders Fund | Founders Fund Portfolio — Crusoe | Crusoe listed in Founders Fund portfolio alongside SpaceX, Palantir, Anduril, Stripe. |
| SO019 | Crusoe | Crusoe Newsroom — Microsoft 900 MW Abilene Campus | Crusoe announces a new 900 MW campus for Microsoft in Abilene, Texas, complementing the existing 1.2 GW development. |
| SO020 | Crusoe | Crusoe About and Leadership Page | Crusoe is an AI factory company headquartered in Denver, Colorado. Founded in 2018 by Chase Lochmiller and Cully Cavness. |
| SO021 | Data Center Dynamics | Crusoe Breaks Ground on Abilene Texas AI Campus | Crusoe has broken ground on a 1.2 GW AI data center campus in Abilene, Taylor County, Texas. |
| SO022 | Forbes | Crusoe CEO Chase Lochmiller on Building the AI Factory | Lochmiller says Crusoe's model of owning the full stack from energy to GPU has positioned the company to win long-term data center contracts that pure-play neoclouds cannot. |
| SO023 | Carbon Credits | Crusoe Energy AI Infrastructure Series E Round | Crusoe raised $1.375B in Series E led by Valor Equity and Mubadala, backed by 30+ investors including NVIDIA and T. Rowe Price. |
| SO024 | Crusoe | Crusoe Newsroom — 1,235 Employees March 2026 Milestone | Crusoe employed 1,235 people as of March 31, 2026. |
| SO025 | Crusoe | Crusoe Manufacturing — Spark Factory Brighton CO | The Crusoe Spark Factory in Brighton, Colorado is a 350,000 sq ft facility with a $200M investment and 200+ jobs, producing modular AI factory units. |
| SO026 | Data Center Dynamics | Crusoe Fast Company World Changing Idea Award 2026 | Crusoe was recognized by Fast Company as a World Changing Idea for its vertically integrated AI infrastructure approach. |
| SO027 | U.S. Securities and Exchange Commission | SEC EDGAR — Crusoe Inc. Filing History | Crusoe Inc. (CIK 0001924674) has filed Form D for Series E (2025-10-23) and Series D (2024-11-21) with the SEC. |
| SM001 | MarketsandMarkets | GPU as a Service Market Size, Share & Industry Trends Analysis Report | The GPU as a Service Market size is estimated to be USD 8.21 billion in 2025 and is expected to reach USD 26.62 billion by 2030, at a CAGR of 26.5%. |
| SM002 | IDC | IDC Market Note — AI Infrastructure Spending 2025 | Cloud infrastructure GPU market grew 46.8% in 2025, reaching $157.8 billion. |
| SM003 | ComputeForecast | The Neocloud GPU Service Industry Report 2025 | Neocloud revenue exceeded $5 billion in Q2 2025, up 205% year-over-year. |
| SM004 | Signisys | The Neocloud Revolution: How $20B in GPU-Focused Providers Are Reshaping the Cloud Market | |
| SM005 | Axios | Crusoe Energy raises $600M to power AI data centers | |
| SM006 | Crusoe | Crusoe Announces Series E Funding — $1.375B at $10B Valuation | Crusoe's AI cloud ARR has grown 150% year-over-year, and total contract value has grown 17x. |
| SM007 | Sacra | Crusoe Energy Systems Company Profile and Revenue Estimates | |
| SM008 | Crusoe | The New Equation: What AI Leaders Need to Know About Infrastructure in 2026 | |
| SM009 | PitchBook | Crusoe Energy Systems — Company Profile and Funding History | |
| SM010 | Data Center Dynamics | Crusoe Energy closes $600M funding round | |
| SM011 | Forbes | Crusoe's Wind and Solar Powered AI Data Centers Could Lead to IPO | |
| SM012 | Crusoe | Crusoe Announces New 900 MW AI Factory Campus in Abilene Texas to Support Microsoft AI Infrastructure | Crusoe announces a new 900 MW AI factory campus in Abilene, Texas to support Microsoft AI infrastructure needs. |
| SM013 | EIA | Electric Power Monthly — US Electricity Generation and Capacity Data | |
| SM014 | Data Center Knowledge | Crusoe Expands Abilene AI Campus with New 900MW AI Factory for Microsoft | |
| SM015 | Crusoe | Crusoe Data Centers Overview | |
| SM016 | Together AI | Together AI — AI Platform Company | |
| SM017 | Cursor | Cursor — AI Coding Assistant | |
| SM018 | Redwood Materials | Redwood Materials — Advanced Battery Materials Manufacturer | |
| SM019 | Heatmap News | Crusoe and Stargate — Clean Energy Claims vs Gas Reality | Despite clean-energy messaging, Crusoe uses natural gas generation and has engaged Engine No. 1 to develop 4.5 GW of new gas capacity for AI data center demand. |
| SM020 | GlobeNewswire | Crusoe Raises $1.375 Billion, Reaches $10 Billion Valuation | |
| SM021 | SEC EDGAR | Crusoe Inc. Form D — Series E Equity Offering (filed 2025-10-23) | |
| SM022 | NVIDIA | NVIDIA Announces Strategic Collaboration with Crusoe AI | |
| SM023 | Forbes | Meet the Tiny Startup Building Stargate, OpenAI's $500 Billion Data Center Moonshot | |
| SM024 | Crusoe | Crusoe Cloud Pricing Page | H100 SXM: $4.29/GPU-hr; A100 SXM: $1.95/GPU-hr |
| SM025 | Crusoe | Crusoe 2024 Impact Report | |
| SM026 | Crusoe | Crusoe Cloud — Official GPU Cloud Platform | |
| SP001 | ComputeForecast | The Neocloud GPU Service Industry Report 2025 — CoreWeave and Market Leaders | Neocloud revenue exceeded $5 billion in Q2 2025, up 205% year-over-year. |
| SP002 | Data Center Dynamics | Crusoe raises $1.375BN in latest funding round | |
| SP003 | Lambda Labs | Lambda Labs GPU Cloud | |
| SP004 | Nebius AI | Nebius AI GPU Cloud Platform | |
| SP005 | Lancium | Lancium Renewable-Powered HPC Data Centers | |
| SP006 | NVIDIA | NVIDIA Announces Strategic Collaboration with Crusoe AI | |
| SP007 | Crusoe | Crusoe Expands NVIDIA Collaboration | |
| SP008 | Crusoe | Crusoe Announces New Manufacturing Facility to Produce Modular AI Factories | |
| SP009 | Forbes | From Gigawatts to Grab-and-Go: Crusoe Leans Into Modular AI Data Centers | |
| SP010 | Signisys | The Neocloud Revolution: How $20B in GPU-Focused Providers Are Reshaping the Cloud Market | |
| SP011 | AWS | AWS EC2 P4 and P5 GPU Instances | |
| SP012 | Oracle | Oracle SuperCluster GPU Infrastructure | |
| SP013 | Forbes | Meet the Tiny Startup Building Stargate, OpenAI's $500 Billion Data Center Moonshot | A 45-hour outage in early March 2025 affected enterprise customers and raised reliability concerns. |
| SP014 | Crusoe | Crusoe Achieves ISO 27001 and ISO 42001 Certifications | |
| SP015 | Heatmap News | Crusoe and Stargate — Clean Energy Claims vs Gas Reality | Despite clean-energy messaging, Crusoe's Abilene campus has TCEQ-permitted 360 MW of natural gas turbines representing approximately 30% of Phase 1 capacity. |
| SP016 | GlobeNewswire | Crusoe Raises $1.375 Billion, Reaches $10 Billion Valuation | |
| SP017 | SEC EDGAR | Crusoe Inc. Form D — Series E Equity Offering (2025) | |
| SP018 | Crusoe | Crusoe Announces Series E Funding | |
| SP019 | CoreWeave | CoreWeave — GPU Cloud Infrastructure | |
| SP020 | Crusoe | Crusoe Cloud Official Platform Page | |
| SP021 | Crusoe | Crusoe Cloud Pricing Page | H100 SXM: $4.29/GPU-hr; A100 SXM: $1.95/GPU-hr |
| SP022 | Crusoe | Crusoe Named to Fast Company's Most Innovative Companies of 2026 | |
| SP023 | Crusoe | Crusoe Wins North American Data Center Project of the Year at 2025 Data Center Dynamics Global Awards | |
| SP024 | Crusoe | Crusoe and Redwood Materials Expand Strategic Partnership Scaling to 7x the Original AI Infrastructure Density | |
| SP025 | Bain Capital Ventures | Crusoe Climb — Betting on Power Before AI Was Cool | |
| SI001 | Crusoe | Crusoe Cloud Pricing Page | H100 SXM: $4.29/GPU-hr. A100 SXM: $1.95/GPU-hr. CPU: $0.04/vCPU-hr. |
| SI002 | Crusoe | Crusoe Cloud — GPU and AI Infrastructure Platform | Crusoe Cloud offers GPU compute including H100, A100, GB200 NVL72, B200, H200, L40S, AMD MI355x, and MI300x instances. |
| SI003 | U.S. Securities and Exchange Commission | SEC Form D — Crusoe Inc. Series E (2025-10-23) | Total offering amount: $1,374,999,988; total number of investors: 71; date of first sale: 2025-10-14. |
| SI004 | U.S. Securities and Exchange Commission | SEC EDGAR Full-Text Search — Crusoe Energy Form D filings | |
| SI005 | U.S. Securities and Exchange Commission | SEC EDGAR — Crusoe Company Search (Form D) | |
| SI006 | Axios | Crusoe Energy raises $1.375B in Series E | Crusoe raised $1.375 billion in its Series E at a valuation of more than $10 billion, co-led by Valor Equity Partners and Mubadala Capital. |
| SI007 | Axios | Crusoe Energy closes $600M Series D funding round | Crusoe Energy Systems has closed a $600 million funding round as it pivots to AI data centers. |
| SI008 | GlobeNewswire | Crusoe Raises $1.375 Billion, Reaches $10 Billion Valuation | Crusoe has raised $1.375 billion and reached a $10 billion valuation in its Series E financing led by Valor Equity Partners and Mubadala Capital. |
| SI009 | Sacra | Crusoe Revenue and Business Model Analysis | Crusoe's 2024 AI cloud revenue of $124M represents 460% year-over-year growth from the prior period. |
| SI010 | PitchBook | Crusoe Energy Systems — Company Profile and Funding | |
| SI011 | Data Center Dynamics | Crusoe exits crypto operations to focus on AI, sells to NYDIG | Crusoe is selling its Bitcoin mining and digital flare mitigation operations to NYDIG, exiting crypto to focus entirely on AI. |
| SI012 | Crusoe | Crusoe Closes Series D Funding — Newsroom | Crusoe closed a $600M Series D round led by Founders Fund at a $2.8B valuation. |
| SI013 | Compute Forecast | Neocloud GPU Service Market — Long Read | GPU cloud neocloud operators have disclosed gross margins in the 20–40% range as GPU spot prices normalized from $8/hr peaks. |
| SI014 | Signisys | The Neocloud Revolution: $20 Billion in GPU-Focused Providers Reshaping Cloud | Neocloud GPU providers have collectively raised over $20 billion as AI compute demand drove triple-digit growth rates in 2023–2024. |
| SI015 | Markets and Markets | GPU-as-a-Service Market Report | |
| SI016 | Heatmap News | Crusoe and the Stargate Problem: Clean Energy Claims vs. Natural Gas Reality | Despite Crusoe's clean-energy marketing, the company uses natural gas generators and has engaged Engine No. 1 to develop 4.5 GW of new gas-fired capacity for AI data centers. |
| SI017 | U.S. Securities and Exchange Commission | SEC Form D — Crusoe Inc. Series D (2024-11-21) | Crusoe Energy Holdings Inc. Series D Form D filed November 21, 2024. |
| SI018 | Crusoe | Crusoe 2024 Impact Report | Crusoe generated $276M in total revenue in 2024 with AI cloud contributing $124M (460% YoY growth). |
| SI019 | Crusoe | Crusoe Blog — The New Equation: What AI Leaders Need to Know About Infrastructure in 2026 | Crusoe's vertically integrated model compresses AI factory deployment from years to months, enabling superior economics for hyperscale customers. |
| SI020 | Crusoe | Crusoe Blog — Welcome to the Era of BYO Power | Crusoe is pioneering the bring-your-own-power model for AI data centers, enabling customers to access purpose-built AI infrastructure without grid dependency risk. |
| SI021 | NVIDIA | NVIDIA Announces Strategic Collaboration with Crusoe AI | NVIDIA has entered a strategic collaboration with Crusoe AI, with Crusoe serving as a launch partner for the GB200 NVL72 system. |
| SI022 | IDC | IDC Market Report: GPU Cloud and AI Infrastructure 2025 | |
| SI023 | Crusoe | Crusoe Cloud Documentation | |
| SI024 | Crusoe | Crusoe Announces 900 MW AI Factory Campus in Abilene Texas for Microsoft | Crusoe announced a new 900 MW AI factory campus in Abilene, Texas to support Microsoft's AI infrastructure needs. |
| SI025 | Data Center Dynamics | Crusoe Raises $1.375B in Latest Funding Round | Crusoe has raised $1.375 billion in its Series E at a $10 billion-plus valuation to fund its Abilene, Texas AI data center campus. |
| SI026 | Forbes | Crusoe's Wind and Solar Powered AI Data Centers Could Lead to IPO | Crusoe's renewable energy approach and rapid revenue growth position the company for a potential IPO in the coming years. |
| SI027 | Crusoe | Crusoe Announces Series E Funding — Newsroom | Crusoe achieved 150% year-over-year cloud ARR growth and 17x total contract value growth in 2025. |
| SI028 | U.S. Energy Information Administration | EIA Electricity Data Browser — Commercial Electricity Prices | |
| SI029 | Axios | Crusoe Energy Series E Funding — The AI Factory Company | Crusoe's 17x total contract value growth and 150% cloud ARR growth in 2025 underscore its rapid transition from Bitcoin mining to AI infrastructure. |
| SI030 | U.S. Environmental Protection Agency | EPA Greenhouse Gas Reporting Program — Data Sets | |
| SI031 | Form Energy | Form Energy — Long-Duration Iron-Air Battery Storage | |
| SE001 | Crusoe | Crusoe GitHub – Open-Source Tooling Organization | Terraform provider, Kubernetes CSI driver, and container images published under Apache 2.0 and MIT licenses; 200+ commits on Terraform provider. |
| SE002 | HuggingFace | HuggingFace – Crusoe AI Inference Partner Profile | Crusoe is listed as an HuggingFace Inference Endpoints partner, enabling one-click deployment of Hub models on Crusoe Cloud. |
| SE003 | Crusoe | Crusoe Cloud – Product Overview | On-demand, reserved, and spot GPU compute with bare-metal, VM, and Kubernetes tiers; H100, H200, B200, GB200 NVL72 instances. |
| SE004 | Crusoe | Crusoe Docs – Platform Documentation Root | Full API reference, CLI guide, and platform architecture documentation for Crusoe Cloud. |
| SE005 | Crusoe | Crusoe Docs – Kubernetes Getting Started | Create a Kubernetes cluster with node pools, configure autoscaling, and attach CFS persistent volumes as ReadWriteMany PVCs. |
| SE006 | Crusoe | Crusoe Docs – CFS Overview | Crusoe File System (CFS) delivers high-throughput parallel storage with GPUDirect RDMA and ReadWriteMany Kubernetes PVs. |
| SE007 | Crusoe | Crusoe Blog – BYO-Power Era | Hyperscale customers can co-locate compute at their own generation assets under the BYO-Power model. |
| SE008 | Crusoe | Crusoe Newsroom – Command Center Launch | Command Center provides a unified dashboard for GPU fleet health, power utilization, and workload scheduling. |
| SE009 | Crusoe | Crusoe Newsroom – Edge Zones Launch | Edge Zones extend the Crusoe platform to third-party colo sites for sub-10 ms inference latency. |
| SE010 | Crusoe | Crusoe Newsroom – Form Energy Agreement | Multi-year agreement for Form Energy iron-air long-duration battery storage at future Crusoe campuses. |
| SE011 | NVIDIA | NVIDIA GTC 2026 – Crusoe B200 Deployment Session (S72619) | Crusoe and NVIDIA co-present B200/GB200 NVL72 deployment roadmap and expanded strategic collaboration. |
| SE012 | Crusoe | Crusoe Energy – Power Infrastructure Overview | Crusoe builds campuses from 30 MW to over 1 GW anchored by co-located generation assets. |
| SE013 | Upstream Data | Upstream Data – Flare-Gas Power Systems | Upstream Data supplies oil-field flare-gas power generation units for stranded-gas compute deployments. |
| SE014 | GlobeNewswire | GlobeNewswire – Crusoe ISO 27001 and ISO 42001 Certification | Crusoe achieves ISO 27001:2022 and ISO 42001:2023 dual certification, validated by DNV GL. |
| SE015 | Together.ai | Together.ai – Crusoe Cloud GPU Benchmark | H100 cluster on Crusoe Cloud achieves competitive tokens-per-second on Llama 3.1 405B versus major cloud providers. |
| SE016 | Heatmap News | Heatmap News – Abilene Campus Profile | Abilene Phase 1 reaches 200 MW IT load with 765 kV interconnect; Phase 2 adds GB200 NVL72 racks. |
| SE017 | Data Center Dynamics | Data Center Dynamics – Neocloud Review 2025: AI-Focused Providers | Crusoe ranks among the top AI-focused neocloud providers in 2025, distinguished by its modular AI factory design, renewable energy sourcing, and NVIDIA GPU density. |
| SE018 | Signisys | Signisys – Abilene Substation Engineering Case Study | Signisys provided high-voltage substation design and commissioning for the Abilene 765 kV interconnect. |
| SE019 | NVIDIA Newsroom | NVIDIA Newsroom – Crusoe Strategic Collaboration | NVIDIA and Crusoe expand multi-year strategic collaboration covering H200, B200, and GB200 GPU allocations. |
| SE020 | Crusoe | Crusoe Blog – H200 General Availability | H200 SXM5 bare-metal instances are now generally available with 141 GB HBM3e and NVLink 4.0. |
| SE021 | Crusoe | Crusoe Blog – OpenAI-Compatible Inference API | Crusoe Inference API implements OpenAI Chat Completions v1 and Completions v1 endpoints for drop-in model serving. |
| SE022 | Crusoe | Crusoe Blog – ISO 27001 and ISO 42001 Certification Details | ISO 27001:2022 covers all cloud infrastructure; ISO 42001:2023 covers the AI platform and inference services. |
| SE023 | Crusoe | Crusoe Blog – Modular AI Factory Architecture Deep Dive | Each MAF unit is a 1-2 MW prefabricated container with integrated GPU compute, power, and liquid cooling. |
| SE024 | Crusoe | Crusoe Blog – B200 and GB200 Roadmap | B200 SXM5 instances targeting GA in Q3 2026; GB200 NVL72 racks planned for Abilene Phase 2. |
| SE025 | Crusoe | Crusoe Docs – VPC Networking Overview | VPCs with subnet-level firewall rules, SAML 2.0 and OIDC federated identity via Keystone IAM. |
| SE026 | Crusoe | Crusoe Docs – Block and Object Storage Overview | NVMe-over-Fabric block storage up to 100 TB per instance; S3-compatible object storage. |
| SU001 | Crusoe AI | Crusoe Cloud – GPU Cloud Platform | |
| SU002 | Crusoe AI | Crusoe and Redwood Materials Expand Strategic Partnership, Scaling to 7x the Original AI Infrastructure Density | Crusoe and Redwood Materials are scaling to 7x the original AI infrastructure density |
| SU003 | Crusoe AI | Crusoe Announces New 900 MW AI Factory Campus in Abilene, Texas to Support Microsoft AI Infrastructure | Crusoe announces a new 900 MW AI factory campus in Abilene, Texas to support Microsoft AI infrastructure |
| SU004 | Redwood Materials | Redwood Materials – Battery Recycling and AI Operations | |
| SU005 | Cursor | Cursor – The AI Code Editor | |
| SU006 | Together AI | Together AI – LLM Infrastructure and Services | |
| SU007 | Forbes | Crusoe's Wind and Solar-Powered AI Data Centers Could Lead to IPO | A 45-hour outage in early March 2025 raised reliability concerns about Crusoe's cloud infrastructure |
| SU008 | Forbes | Meet the Tiny Startup Building Stargate, OpenAI's $500 Billion Data Center Moonshot | |
| SU009 | Forbes | From Gigawatts to Grab-and-Go: Crusoe Leans Into Modular AI Data Centers | |
| SU010 | Sacra | Crusoe Revenue, Growth, and Valuation | |
| SU011 | Compute Forecast | NeoCloud GPU Service: Market Analysis and Provider Landscape | |
| SU012 | MarketsAndMarkets | GPU-as-a-Service Market – Global Forecast to 2028 | |
| SU013 | Crusoe Cloud | Crusoe Cloud Documentation | |
| SU014 | Crusoe (GitHub) | Crusoe Cloud GitHub Organization | |
| SU015 | Crusoe AI (HuggingFace) | Crusoe AI on HuggingFace | |
| SU016 | NVIDIA Newsroom | NVIDIA Announces Strategic Collaboration with Crusoe AI | |
| SU017 | Crusoe AI | Crusoe Achieves ISO 27001 and ISO 42001 Certifications | |
| SU018 | Crusoe AI | The New Equation: What AI Leaders Need to Know About Infrastructure in 2026 | |
| SU019 | Crusoe AI | Crusoe 2024 Impact Report | Redwood Materials has maintained 99.2% uptime across its Crusoe Spark deployment over seven months |
| SU020 | Heatmap News | Crusoe, Stargate, and the Natural Gas Contradiction | Crusoe's clean energy branding is undermined by TCEQ data confirming 360 MW of on-site gas turbines at Abilene |
| SU021 | GlobeNewswire | Crusoe Raises $1.375 Billion, Reaches $10 Billion Valuation | |
| SU022 | Axios | Crusoe Energy raises $600M for AI data centers | |
| SU023 | Axios | Crusoe raises $1.375M Series E | |
| SU024 | PitchBook | Crusoe Energy Systems – Company Profile and Funding Data | |
| SU025 | Crusoe AI | Crusoe Launches Command Center, a Unified Operations Platform for High-Performance AI Workloads | |
| SU026 | Crusoe AI | Crusoe Cloud Pricing | |
| SU027 | IDC | AI Infrastructure Spending Forecast 2025–2028 | |
| SU028 | Signisys | The NeoCloud Revolution: How $20 Billion in GPU-Focused Providers are Reshaping the Cloud Market | |
| SU029 | Data Center Dynamics | Crusoe Energy closes $600M funding round | |
| SU030 | Bain Capital Ventures | Crusoe Climb: Betting on Power Before AI Was Cool | |
| SU031 | Crusoe AI | Crusoe Expands NVIDIA Collaboration | |
| SU032 | Crusoe AI | Crusoe Wins North American Data Center Project of the Year at 2025 Data Center Dynamics Global Awards | |
| SU033 | CoreWeave | CoreWeave – GPU Cloud for AI | |
| SU034 | Lambda Labs | Lambda Labs – GPU Cloud for Deep Learning | |
| SU035 | Form Energy | Form Energy – Iron-Air Battery Storage | |
| SU036 | Crusoe | Crusoe and Form Energy Announce Agreement for 12 Gigawatt-Hours of Iron-Air Batteries for AI Data Centers | Crusoe and Form Energy announced an agreement for 12 GWh of iron-air long-duration battery storage to back Crusoe AI data center power beginning in 2027. |
| SU037 | Crusoe | Crusoe Unveils Crusoe Edge Zones | |
| SU038 | Crusoe Energy Systems – LinkedIn Company Page | ||
| SU039 | NVIDIA | Crusoe and NVIDIA GTC 2026 Session – AI Factory Infrastructure | |
| SU040 | Upstream Data | Upstream Data (Canada) – Flare Gas Bitcoin Mining Company | |
| SU041 | Crusoe | Crusoe Energy – Official Website | |
| SU042 | Anthropic | Anthropic – AI Safety Company | |
| SU043 | Mistral AI | Mistral AI – Open-Source Language Models | |
| SU044 | OpenAI | OpenAI – AI Research and Products | |
| SU045 | Andreessen Horowitz | Andreessen Horowitz (a16z) – AI Portfolio | |
| SU046 | SemiAnalysis | SemiAnalysis – GPU and Semiconductor Market Intelligence | AI cloud compute demand forecasts show sustained GPU infrastructure requirements through 2028, driven by foundation model training and enterprise AI adoption. |
| SU047 | Lancium | Lancium – Renewable-Powered HPC and AI Campuses | Lancium operates renewable-powered HPC and AI compute campuses in West Texas, validating the market for clean-energy, behind-the-meter AI infrastructure. |
| SU048 | Crusoe | Crusoe – Careers and Enterprise Customer Focus | Crusoe's open roles include cloud account executives, enterprise customer success managers, and solution architects targeting Fortune 500 and AI lab accounts. |
| SR001 | Texas Commission on Environmental Quality (TCEQ) | TCEQ Air Permit Database and Permitting Framework | |
| SR002 | U.S. Energy Information Administration | EIA Electric Power Data Browser | |
| SR003 | U.S. Environmental Protection Agency | EPA Greenhouse Gas Reporting Program Data Sets | |
| SR004 | Upstream Data | Crusoe, Upstream Data to Resolve Lawsuit | Crusoe and Upstream Data have agreed to resolve their patent dispute through mutual dismissal and a license agreement |
| SR005 | Crusoe AI | Crusoe AI Newsroom | |
| SR006 | Heatmap News | Crusoe, Stargate, and the Natural Gas Contradiction | Crusoe's clean energy branding is contradicted by TCEQ data confirming 360 MW of on-site gas turbines and plans for 4.5 GW of new gas generation |
| SR007 | Crusoe AI | Crusoe 2024 Impact Report | |
| SR008 | Forbes | Crusoe's Wind and Solar-Powered AI Data Centers Could Lead to IPO | A 45-hour outage in early March 2025 raised reliability concerns about Crusoe's cloud infrastructure |
| SR009 | NVIDIA Newsroom | NVIDIA Announces Strategic Collaboration with Crusoe AI | |
| SR010 | NVIDIA | GTC 2026 – Crusoe AI Infrastructure Session | |
| SR011 | Crusoe AI | Crusoe Announces New 900 MW AI Factory Campus in Abilene, Texas to Support Microsoft AI Infrastructure | |
| SR012 | Data Center Knowledge | Crusoe Expands Abilene AI Campus with New 900MW AI Factory for Microsoft | |
| SR013 | Axios | Crusoe raises $1.375M Series E | |
| SR014 | GlobeNewswire | Crusoe Raises $1.375 Billion, Reaches $10 Billion Valuation | |
| SR015 | Compute Forecast | NeoCloud GPU Service: Market Analysis and Provider Landscape | |
| SR016 | Signisys | The NeoCloud Revolution: How $20 Billion in GPU-Focused Providers are Reshaping the Cloud Market | |
| SR017 | Crusoe AI | Crusoe Appoints Former MongoDB Executive Michael Gordon as COO and CFO | |
| SR018 | Crusoe AI | Crusoe Appoints Nader Pakfar as General Counsel Real Estate | |
| SR019 | Forbes | From Gigawatts to Grab-and-Go: Crusoe Leans Into Modular AI Data Centers | |
| SR020 | Crusoe AI | Crusoe Announces New Manufacturing Facility to Produce Modular AI Factories | |
| SR021 | Crusoe AI | Crusoe AI Data Centers | |
| SR022 | Crusoe AI | Welcome to the Era of BYO Power | |
| SR023 | SEC EDGAR | Crusoe Inc. Form D Filings – SEC EDGAR | |
| SR024 | Crusoe AI | Crusoe Achieves ISO 27001 and ISO 42001 Certifications | |
| SR025 | Crusoe Cloud | Crusoe Cloud Documentation | |
| SR026 | CoreWeave | CoreWeave – GPU Cloud for AI | |
| SR027 | Lambda Labs | Lambda Labs – GPU Cloud for Deep Learning | |
| SR028 | Form Energy | Form Energy – Iron-Air Battery Storage | |
| SR029 | Crusoe AI | Form Energy and Crusoe Announce Agreement for 12 Gigawatt-Hours of Iron-Air Batteries for AI Data Centers | |
| SR030 | Crusoe AI | Crusoe Energy – Corporate Website | |
| SR031 | Crusoe AI | Crusoe AI Energy | |
| SR032 | Axios | Crusoe Energy series E funding round | |
| SR033 | Data Center Dynamics | Crusoe Energy Closes $600M Funding Round | |
| SR034 | Crusoe AI | Crusoe Announces Series E Funding | |
| SR035 | Forbes | Meet the Tiny Startup Building Stargate, OpenAI's $500 Billion Data Center Moonshot | |
| SR036 | Blue Owl Capital | Blue Owl Capital – Real Estate and Credit Strategies | |
| SR037 | U.S. Energy Information Administration | EIA Annual Energy Outlook — U.S. Energy Market and Grid Projections | |
| SR038 | Data Center Dynamics | Neocloud Market Review 2025 — GPU Cloud Competitive and Risk Analysis | |
| SR039 | Google Cloud | Google Cloud GPU Pricing — H100 and A100 On-Demand and Spot Rates | |
| SR040 | U.S. Securities and Exchange Commission | SEC EDGAR — Crusoe Inc. All Filings (CIK 0001924674) | |
| SR041 | PitchBook | CoreWeave IPO Filing — Neocloud Competition and Valuation Risk | |
| SR042 | GlobeNewsWire | GlobeNewsWire — Crusoe and Industry Press Releases | |
| SR043 | ERCOT | About ERCOT — Electric Reliability Council of Texas | |
| SR044 | Lambda Labs | Lambda Cloud GPU Instances — On-Demand and Reserved Pricing | |
| SV001 | U.S. Securities and Exchange Commission | Crusoe Inc. Form D — Series E Exempt Offering (filed October 23, 2025) | Form D confirms total offering of $1,374,999,988 with 71 investors; co-led by Valor Equity Partners and Mubadala Capital; filed October 23, 2025. |
| SV002 | U.S. Securities and Exchange Commission | Crusoe Inc. Form D — Series D Exempt Offering (filed November 21, 2024) | Form D filed November 21, 2024 confirms $600M Series D led by Founders Fund with NVIDIA, Fidelity, Mubadala, Ribbit Capital, and Valor participating. |
| SV003 | U.S. Securities and Exchange Commission — EDGAR | Crusoe Inc. EDGAR CIK 0001924674 Form D Filing History | |
| SV004 | GlobeNewsWire | Crusoe Raises $1.375 Billion, Reaches $10 Billion Valuation | Crusoe has raised $1.375 billion in Series E funding, reaching a $10 billion valuation. |
| SV005 | Axios | Crusoe Energy's Series E funding round | |
| SV006 | Axios | Crusoe raises $1.375B in Series E round | |
| SV007 | Data Center Dynamics | Crusoe raises $1.375bn in latest funding round | |
| SV008 | PitchBook | Crusoe Energy Systems — Company Profile and Funding Data | |
| SV009 | Sacra | Crusoe — AI Infrastructure Company Research | |
| SV010 | ComputeForecast | Neocloud GPU Service Long Read | |
| SV011 | MarketsAndMarkets | GPU as a Service Market — Global Forecast to 2030 | |
| SV012 | Signisys | The Neocloud Revolution: How $20 Billion in GPU-Focused Providers Are Reshaping the Cloud Market | |
| SV013 | IDC | IDC AI Infrastructure Market Study | |
| SV014 | Heatmap News | Crusoe and Stargate: The Natural Gas Question | Despite Crusoe's clean-energy marketing, the company engaged Engine No. 1 to develop 4.5 GW of new natural gas capacity, and TCEQ permits confirm 360 MW of on-site gas turbines at Abilene. |
| SV015 | Forbes | Crusoe's Wind and Solar-Powered AI Data Centers Could Lead to IPO | |
| SV016 | Carboncredits.com | Crusoe Energy's $600M Raise Fuels AI Revolution With Clean Energy Data Centers | |
| SV017 | Data Center Dynamics | Crusoe Energy closes $600M funding round | |
| SV018 | Axios | Crusoe Energy raises $600M for AI data centers | |
| SV019 | Crusoe | Crusoe Announces Series E Funding | |
| SV020 | Crusoe | Crusoe Closes Series D Funding | |
| SV021 | Crusoe | Crusoe Announces New 900 MW AI Factory Campus in Abilene Texas to Support Microsoft AI Infrastructure | |
| SV022 | Crusoe | Crusoe 2024 Impact Report | |
| SV023 | Crusoe | Crusoe Newsroom — Announcements and Milestones | |
| SV024 | Crusoe | The New Equation: What AI Leaders Need to Know About Infrastructure in 2026 | |
| SV025 | NVIDIA | NVIDIA Announces Strategic Collaboration with Crusoe AI | |
| SV026 | Crusoe | Crusoe Expands NVIDIA Collaboration | |
| SV027 | Nebius | Nebius — AI Cloud Infrastructure Platform | |
| SV028 | Lambda Labs | Lambda Labs — GPU Cloud for AI and Deep Learning | |
| SV029 | Bain Capital Ventures | Crusoe Climb: Betting on Power Before AI Was Cool | Bain Capital Ventures frames power scarcity as the binding constraint in AI infrastructure and characterizes Crusoe's early grid rights as a durable competitive moat that was secured before AI demand made the thesis obvious to competitors. |
| SV030 | Founders Fund | Founders Fund Portfolio — AI and ML | |
| SV031 | Forbes | Meet the Tiny Startup Building Stargate, OpenAI's $500 Billion Data Center Moonshot | |
| SV032 | Data Center Knowledge | Crusoe Expands Abilene AI Campus With New 900MW AI Factory for Microsoft | |
| SV033 | Axios Pro Tech Finance | Crusoe Energy Raises $1.375 Billion Series E at $10B+ Valuation | |
| SV034 | U.S. Securities and Exchange Commission | SEC EDGAR — CoreWeave Inc. S-1 Registration Statement Filings | |
| SV035 | Crusoe | Crusoe Blog — AI Infrastructure and Industry Insights | |
| SV036 | Lancium | Lancium — Renewable Energy Powered HPC and AI Data Centers | |
| SV037 | ComputeForecast | ComputeForecast — AI Infrastructure Market Research and Analysis | |
| SV038 | Together AI | Together AI Blog — AI Model Serving and GPU Cloud | |
| SV039 | Crusoe | Crusoe Careers — Open Roles and Headcount Data | |
| SV040 | PitchBook | CoreWeave IPO Filing — Neocloud Valuation and Market Analysis | |
| SV041 | Statista | Global Cloud Computing Market Size 2024–2030 | |
| SV042 | CB Insights | AI Cloud Infrastructure Report — Market Sizing and Competitive Dynamics | |
| SV043 | S&P Capital IQ | Capital IQ — Financial Data and Comparable Company Analysis | |
| SV044 | DigitalBridge | DigitalBridge Portfolio — Digital Infrastructure Private Equity Investments | |
| SV045 | SemiAnalysis | SemiAnalysis — Semiconductor and AI Infrastructure Research | |
| SV046 | CloudPrice | CloudPrice — GPU and Cloud Compute Price Comparison Platform |