Field AI
具身 AI 尽调 — 面向工业机器人的 Field Foundation Models
FieldAI 汇集一流现场机器人团队、强投资人背书和差异化的硬件无关具身 AI 栈;但缺少经审计财务披露,具名客户证据也有限,仅靠公开证据很难承销 $2B 估值。
封面要素
公司概况
FieldAI 是一家私营具身 AI 软件公司,2023 年由前 NASA JPL 机器人负责人 Ali Agha 创立,技术联合创始人来自 DeepMind、NASA JPL 和 DARPA 自主项目。公司销售 Field Foundation Models,这是一个硬件无关的软件大脑,让第三方机器人在非结构化工业环境中安全作业,不依赖 GPS、预建图场景或持续云连接。FieldAI 公开牵引力最强的场景在建筑业,已具名部署到 Big-D Construction 和 DPR Construction;公司也瞄准能源、制造、巡检、城市配送和联邦用例。2025 年 8 月和 2026 年 2 月的融资披露 / 第三方融资记录显示,公司估值维持在 $2B,同时披露总融资超过 $500M;经审计收入、治理和当前规模仍是实质性未解问题。
- 成立时间
- 2023-01-01
- 创始人
- Ali Agha, Shayegan Omidshafiei, David Fan
- 创立地点
- Southern California, USA
- 总部
- Irvine, CA, USA
- 产品
- FieldAI 的产品是 Field Foundation Model 平台:物理优先、风险感知的具身 AI,运行在合作伙伴机器人硬件的端侧,让机器人在动态工业环境中完成感知、推理、规划和多机器人协同。
- 客户
- 建筑、能源、制造、巡检、城市配送和精选联邦应用中的大型工业运营商与承包商;这些客户需要机器人在危险或非结构化环境中自主运行。
- 商业模式
- B2B 企业软件授权加集成和部署服务,在客户或合作伙伴拥有的机器人硬件之上叠加经常性自治订阅。
- 阶段
- Private, post-Series A / early-stage VC
- 融资情况
- PremierAlts 报告称,截至 2026 年 2 月,FieldAI 已通过五轮融资累计获得 $506.1M,其中最新一轮早期 VC 融资为 $315M、估值 $2B;公司 2025 年 8 月公告披露,连续两轮融资合计 $405M,投资方包括 Bezos Expeditions、Temasek、NVentures、Intel Capital、Khosla Ventures 以及其他战略投资人。
执行摘要
主要优势
- 硬件无关的 Field Foundation Models 切中真实机器人瓶颈:在非结构化、安全关键环境中实现自主性。
- 创始人-市场匹配少见地强,NASA JPL、DARPA SubT、DeepMind 和高端现场机器人经验是核心支撑。
- 投资人质量一线,Bezos Expeditions、Temasek、NVentures、Intel Capital 和 Khosla 都传递了强私募市场信心。
- 公开案例显示,建筑场景部署有可量化价值,包括检查 / 文档时间减少 90%+,以及 Big-D 的多项目扩张。
主要风险
- 收入、利润率、留存和董事会治理大多未披露,估值承销高度不确定。
- 公开具名客户证据仍窄,集中在建筑领域,还未显示跨垂直行业的大规模生产化。
- 工业场景中具身 AI 的安全、认证和监管就绪度仍只有部分验证。
- 对 Ali Agha 的关键人依赖,以及对 NVIDIA、Boston Dynamics 等平台伙伴的深度依赖,增加集中度风险。
未决问题
- 没有公开可得的经审计 ARR、确认收入、毛利率或 NRR 披露。
- 董事会构成、优先股结构和详细治理权利未公开。
- 软件订阅、服务收入和硬件集成费之间的拆分尚未验证。
- 当前员工数和确切总部标识在公开来源中不一致。
- 可见建筑部署之外的客户集中度,无法通过公开证据衡量。
目录
01公司概览
1.1 公司身份与商业模式
FieldAI 是一家位于 California Irvine 的机器人 AI 软件公司,2023 年由 NASA Jet Propulsion Laboratory(JPL)老将 Ali Agha 创立。公司的核心判断很清楚:工业机器人采用的卡点不在硬件,而在智能——机器人缺少能在非结构化真实环境中稳定运行的自适应、风险感知自治能力。FieldAI 的解法是构建 Field Foundation Models(FFMs),这是一类“物理优先”的基础模型,训练数据来自物理部署中的传感器数据,而不是互联网文本或图像。FFMs 充当通用软件大脑,可装到多种机器人平台上——四足机器人、人形机器人、轮式漫游车、履带车、乘用级自动驾驶车辆——不需要按平台重新编程,也不需要预建图环境、GPS 或云连接。机器人完全在端侧实时决策。FieldAI 面向建筑、能源、采矿、制造、物流、城市配送、巡检和国防领域的企业工业客户销售平台。公司明确不制造机器人;它是纯 AI 软件供应商,与机器人 OEM 合作,把软件部署到客户自有或第三方机器人硬件上。这种硬件无关模式把 FieldAI 放在机器人技术栈的“大脑层”,类似企业计算里抽象硬件的 OS 或中间件层。Bill Gates 对公司描述得很直接:FieldAI“为其他公司的机器人开发 AI 软件,让机器人能够感知环境、在没有 GPS 的情况下导航,甚至彼此通信”。主要收入渠道是 B2B 软件授权和硬件集成服务;订阅价格据称从数万美元到每年 $500,000 不等,取决于部署规模。截至 2026 年 4 月,知情人士称公司已拥有超过 $100M 的已签约收入,并与大型工业客户建立了合作。 [CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 / 备注 |
|---|---|---|---|---|
| 创立 | 2023 | 2023 | 高 | 公司和多个独立来源均已确认 |
| 总部 | Irvine, California | 2025-09 | 高 | 已从 Mission Viejo, CA 迁出;早前关于 San Francisco 总部的说法未获验证 |
| 融资总额 | $405M | 2025-08-20 | 高 | 退出隐身模式时同时披露连续两轮融资 |
| 估值 | $2 billion | 2025-08-20 | 高 | 较上一轮 $500M 上升;Reuters、CNBC、TechCrunch 均有报道 |
| 融资时员工数 | ~130 | 2025-08 | 高 | Reuters 在公告时的报道口径;已从 2024 年底约 30 人增长 |
| 员工数(2026 年初) | 201–500(LinkedIn) | 2026-04 | 中 | LinkedIn 区间;公司未公开披露精确人数 |
| 已预订收入 | >$100M(合作伙伴关系与合同) | 2026-04 | 中 | OCBJ 报道援引知情人士;非公司披露 |
收入和 ARR 均为私人信息,未公开披露。员工数来自 LinkedIn 区间。已预订收入来自 OCBJ 引用的匿名消息源,并非公司直接披露。
[CO001, CO002, CO022, CO023, CO031, CO032]截至 2026 年 5 月尽调日,FieldAI 的关键财务和运营指标。
收入和 ARR 为私有且未披露。预订收入($100M+)来自 OCBJ 引述的匿名来源。员工数来自 2026 年 4 月 LinkedIn 报告的区间。
[CO021, CO022, CO023, CO031, CO032, CO033]1.2 创始人、领导层与治理
FieldAI 由 Ali Agha 创立并担任 CEO,他带来近二十年的机器人 AI 经验。创立 FieldAI 前,Agha 在 NASA Jet Propulsion Laboratory(JPL)工作约七年,担任多个美国高知名度自主项目的首席研究员,包括 DARPA Subterranean Challenge、DARPA RACER(自动驾驶越野车辆)、NASA Autonomous Mars Cave Exploration,以及 Coordinated Autonomy for Prototype Mars Helicopter-Rover。他还曾在 Qualcomm 任研究员,并在 MIT 做博士后。关键在于,Agha 带领 JPL-MIT-Caltech-KAIST-LTU CoSTAR 团队赢得 2020 DARPA Subterranean Challenge 的 Urban Circuit;这是现场机器人领域的标志性成果,也验证了如今 FieldAI 技术核心中的风险感知自治路线。Agha 在 Texas A&M University 获得计算机科学与工程博士。联合创始人包括 Dr. Shayegan Omidshafiei(首席科学官)和 Dr. David Fan(首席技术官)。Omidshafiei 曾在 DeepMind/Google 担任研究科学家超过五年,研究大规模 AI 模型和多智能体强化学习;他在 MIT 完成博士,并在那里首次与 Agha 合作。Fan 曾任多个 DARPA SubT 和 DARPA RACER 项目的首席技术专家,也是 NASA JPL 研究员。其他高级领导包括 Dr. Eric Krotkov,他负责公司的国防子公司 FieldAI Federal,曾任 DARPA Program Manager 和 Toyota Research Institute 的 CSO;以及 Sebastian Scherer,他是 Fieldable Embodied AI 负责人,也是 Carnegie Mellon Robotics Institute 副研究教授。创始团队共同的 JPL、DARPA 和 MIT 背景,让 FieldAI 在安全关键现场环境部署 AI 上拥有罕见的创始人 - 市场匹配。董事会构成和正式治理结构尚未公开披露。CEO Ali Agha 是公司的主要公开发言人与募资人,形成了有意义的关键人依赖。 [CO011, CO012, CO013, CO014, CO015, CO016]
| 人物 | 角色 | 背景 | 创始人-市场匹配 | 关键人物依赖 |
|---|---|---|---|---|
| Ali Agha | 创始人兼 CEO | Texas A&M CS&E 博士;MIT 博士后;Qualcomm Research;NASA JPL 首席研究员 7 年 | 领导 DARPA SubT CoSTAR 团队(2020 Urban 获胜);20 年野外机器人 AI 经验 | 高 — 主要公开发言人与募资负责人 |
| Shayegan Omidshafiei | 联合创始人兼首席科学官 | MIT 博士;在 DeepMind/Google 担任研究科学家 5 年以上;深度 RL 和多智能体 AI | 在 MIT 与 Agha 合作;具备博士级基础模型专长 | 高 — 负责 AI/ML 模型开发 |
| David Fan | 联合创始人兼 CTO | Georgia Tech 博士;DARPA SubT 和 DARPA RACER 首席技术专家;NASA JPL 研究员 | 有把 DARPA 级自主能力带到商业规模的验证记录 | 高 — 核心技术架构师 |
| Eric Krotkov | FieldAI Federal 负责人 | 前 DARPA 项目经理;Toyota Research Institute CSO;CMU 教授;PackBot/Talon 机器人奠基人 | 三次 DARPA Grand Challenges 领导经验;深厚联邦 / 国防关系 | 中 — 事业部负责人 |
| Sebastian Scherer | Fieldable 具身 AI 总监 | CMU Robotics Institute 副研究教授;专长于不确定环境中的可部署 AI | 学术研究管线;CMU 机器人关系 | 低 — 职能贡献者 |
董事会构成尚未公开披露。三位联合创始人(Agha、Omidshafiei、Fan)都深度承袭 JPL、DARPA 和 MIT 脉络。关键人物风险集中在 CEO。
[CO011, CO012, CO013, CO014, CO015, CO016]1.3 融资历史与投资人构成
FieldAI 从 2023 年创立到 2025 年 8 月 20 日一直保持隐身;当天公司同时披露连续两轮融资合计 $405M,并宣布估值达到 $2B——较前一轮确定的 $500M 估值翻了四倍。最近一轮融资 $314M,并且获得超额认购;共同投资方包括 Bezos Expeditions(Jeff Bezos 家族办公室)、Prysm Capital 和 Temasek。两轮融资中的其他投资方包括 BHP Ventures、Canaan Partners、Emerson Collective、Intel Capital、Khosla Ventures、NVentures(NVIDIA 的风险投资部门)等。Gates Frontier(Bill Gates 的投资机构)和 Samsung 被列为“过往投资人”,说明它们参与了前一轮、也就是确定 $500M 估值的那轮。按披露总额倒推,第一轮约融资 $91M。注意:部分尽调前资料把拆分写成 $150M(种子轮)加 $255M(Series A);该拆分与 Reuters 关于最新一轮 $314M 的报道不一致,也无法从一手来源确认。超额认购与投资人阵容——覆盖科技风投领袖(Khosla、NVentures、Intel Capital)、主权财富和机构资金(Temasek、BHP)、慈善资本(Gates Frontier、Emerson Collective),以及全球最富有个人的家族办公室(Bezos Expeditions)——反映出异常广泛的投资人信念。Vinod Khosla 表示:“FieldAI 站在通用机器人革命的前沿,它快速部署的能力将释放长期经济和社会价值。”Intel Capital 的公开背书正式确认了与 Intel 硬件栈的平台集成和战略关系。本轮资金将用于全球扩张、产品开发(移动、操作)以及员工扩张,目标是在 2025 年底前把团队规模翻倍。 [CO021, CO022, CO023, CO024, CO025, CO026]
| 利益相关方 | 角色 / 类型 | 轮次 | 控制权 / 经济重要性 | 尽调问题 |
|---|---|---|---|---|
| Bezos Expeditions | 领投方,家族办公室(Jeff Bezos) | Series A1 轮(2025) | 共同领投最新 $314M 融资;战略形象强 | 确认董事会席位和已谈妥的治理权利 |
| Prysm Capital | 领投方,风投 | Series A1 轮(2025) | 共同领投最新一轮;Jay Park 公开背书 | 确认持股比例和董事会参与 |
| Temasek | 机构投资人,主权财富基金(Singapore) | Series A1 轮(2025) | 重要机构资本;亚太扩张伙伴 | 确认其在 APAC 市场的战略增值 |
| Gates Frontier | 既有投资人,家族办公室(Bill Gates) | Series A 轮(约 2024) | 早期高确信投资人;Gates 本人公开背书 | 确认持仓规模和投后董事会权利 |
| Samsung | 既有投资人,战略方 | Series A 轮(约 2024) | 战略硬件伙伴;可能与 Samsung 机器人集成 | 确认 Samsung 关系的商业含义 |
| NVentures (NVIDIA) | 战略投资人,风投部门 | Series A1 轮(2025) | NVIDIA 推理硬件协同;Omniverse 和仿真集成 | 评估算力排他性或优选供应商条款 |
| Khosla Ventures | 风投 | Series A1 轮(2025) | 一线 VC;Vinod Khosla 公开支持 | 确认持股比例和董事会席位 |
| Intel Capital | 战略投资人,企业风投 | Series A1 轮(2025) | Intel 芯片关系;Intel Capital 公开公告 | 评估与 Intel 的优选硬件或集成条款 |
Emerson Collective、BHP Ventures 和 Canaan Partners 也已确认是投资人。董事会构成和治理权利尚未公开披露。
[CO024, CO025, CO026, CO027, CO028, CO029]1.4 规模、合作伙伴与里程碑时间线
自 2023 年创立以来,FieldAI 的增长曲线很陡。公司 2024 年底约有 30 名员工,到 2025 年 8 月融资公告时增至约 130 人;LinkedIn 在 2026 年初显示员工数为 201–500 人,与公司所称到 2025 年底把员工翻倍至约 260 人的计划一致。2025 年末,出于招聘和实验室扩张需要,公司从 Mission Viejo 约 13,000 平方英尺办公空间,搬到 Irvine Spectrum 的 3 Morgan、面积 41,000 平方英尺的研发总部。FieldAI 的生产部署已覆盖三大洲——北美、欧洲和亚太(包括日本)——数百个站点、数千次任务。公司吸引了建筑、能源、制造、城市配送、安防和国防领域数十家大型企业客户,并成立 FieldAI Federal 子公司服务政府和军事应用。2026 年初,公司宣布两项重要战略合作:2 月,Certis Group(新加坡领先的综合安防服务商)与 FieldAI 建立合作,在全球部署自主安防机器人;2026 年 3 月,Boston Dynamics 与 FieldAI 宣布合作,把 Spot 的移动能力与 FieldAI 的 FFMs 结合,用于建筑和其他动态环境。Boston Dynamics 创始人 Marc Raibert 专门提到双方在 2020 年 DARPA SubT 的胜利来背书公司:“自 Ali 和他的团队用 Spot 机器人赢得 DARPA Subterranean Challenge 的 Urban-circuit、同时为 NASA 处理类似挑战以来,我们就认识他们。”Construction Dive 提到的一个关键风险指标是:受访承包商中的活跃机器人部署比例从 2024 年的 65% 降至 2025 年的 46%,尽管正面评价达到 95%;这说明行业在意愿和实际采用之间存在普遍摩擦,也会成为 FieldAI 近期商业规模目标的逆风。 [CO031, CO032, CO033, CO034, CO035, CO036]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2020-01 | CoSTAR 团队赢得 DARPA SubT Urban Circuit | 创立 | 第一名;多机器人、无 GPS 城市场景自主能力 | Ali Agha、Shayegan Omidshafiei、David Fan(当时在 JPL/MIT) | 验证风险感知自主路线;直接孕育 FieldAI 技术 |
| 2023-01 | FieldAI 创立 | 创立 | 公司注册成立;Irvine/Mission Viejo, CA | Ali Agha、Shayegan Omidshafiei、David Fan 三位创始人 | DARPA 和 JPL 时代自主研究开始商业化 |
| 2023-2024 | 隐身模式运营和初始融资(Gates Frontier + Samsung) | 融资 | ~$91M;估值 ~$500M | Gates Frontier;Samsung | 最早的机构验证;融资直到 2025 年 8 月才披露 |
| 2024 | 上一轮以 $500M 估值完成 | 融资 | $500M 投前估值(Reuters 报道) | Gates Frontier、Samsung,以及可能的其他投资人 | 为 2025 年 8 月 Series A1 达成 4x 估值跃升设定基线 |
| 2025-08-20 | FieldAI 退出隐身模式;宣布累计融资 $405M | 融资 | 累计 $405M;估值 $2B | Bezos Expeditions、Prysm、Temasek、Khosla、Intel Capital、NVentures 等 | 晋级独角兽;2025 年最大机器人软件融资事件之一 |
| 2025-09 | FieldAI 开始租赁 Irvine Spectrum 3 Morgan 的 41,000 sqft 总部 | 扩张 | 从 Mission Viejo 约 13,000 sqft 扩大 | Irvine Company(房东) | 释放激进招聘和研发实验室扩建信号 |
| 2025-12 | 员工数翻倍,逼近约 260 人目标 | 扩张 | ~30(2024 年底)→ ~130(2025 年 8 月)→ 200+(2026 年初) | FieldAI | 快速扩张与数百万美元级商业合同积压相符 |
| 2026-02 | 与 Certis Group 合作开发自主安防机器人 | 合作 | 战略合作;全球安防部署 | FieldAI、Certis Group(Singapore) | 验证全球触达;切入 $30B+ 安防劳动力市场 |
| 2026-03-12 | 与 Boston Dynamics 合作,面向建筑和动态环境 | 合作 | Spot + FFM 集成;多站点车队部署 | Boston Dynamics、FieldAI | Marc Raibert 背书;在开发中的最大第三方四足机器人车队 |
| 2026-04 | OCBJ 参访:报道已预订收入 $100M+;LinkedIn 显示 201–500 名员工 | 扩张 | >$100M 已预订(据知情人士) | OC Business Journal 报道 | 首个公开商业规模收入信号;确认企业牵引 |
隐身期融资拆分($91M)由披露总额($405M)与最近一轮($314M,据 Reuters)之间的差额推导。部分预尽调资料称拆分为 $150M + $255M,但无法从一手来源确认。
[CO013, CO021, CO022, CO023, CO031, CO034]从 2020 年赢得 DARPA Subterranean Challenge 到 2026 年 3 月与 Boston Dynamics 合作,FieldAI 历史上的关键里程碑。
隐身期融资日期为近似值;公司在 2025 年 8 月同时披露了所有此前轮次。
[CO001, CO013, CO021, CO022, CO023, CO031]FieldAI 的 JPL/DARPA 研究传承、物理优先 AI 模型、硬件无关平台、客户部署和投资人基础,如何拼成一个相互咬合的商业系统。
[CO003, CO004, CO005, CO021, CO022, CO038]1.5 图表与要点
02市场分析
2.1 市场边界与定义
FieldAI 的可服务市场不是整个工业机器人市场——大多数已部署工业机器人都在结构化、重复性的制造环境中运行(汽车装配、半导体制造、消费电子),这些场景已经由 FANUC、ABB、Yaskawa 等传统固定路径自动化厂商充分服务。FieldAI 真正不同的市场,是那些需要自适应、风险感知自治能力的工业机器人部署:环境非结构化,布局每天变化,GPS 和预建图路径不可用,静态编程失效。具体包括活跃建筑工地(地形、障碍和工作流按小时变化)、地下与露天矿山(落石、瓦斯风险和狭窄空间让人员暴露危险)、油气地面设施与巡检环境(资产访问危险且偏远),以及不适合固定路径机器人的柔性或高混合制造设施。 市场边界排除以下支出:(a)结构化工厂自动化硬件和软件,例如固定装配线上的传统关节臂机器人;(b)在受控环境中运行的消费级或物流仓储机器人;(c)机器人硬件本身,因为 FieldAI 是纯软件供应商;(d)水下 ROV 和特种水下车辆,它们由另一套供应商生态服务。工业买家当前依赖的现状替代方案包括人工巡检和监控班组、需要大量预编程的传统固定路径或单任务机器人、无人机 / UAV 航测,以及配合人工采集数据的 BIM/CAD 建模。按单位现场智能成本看,这些替代方案更贵,实时性更弱;随着劳动力池收缩、安全要求收紧,它们也越来越难以维持。 FieldAI 的市场横跨四个主要垂直:(1)建筑与基础设施,(2)能源、采矿和采掘,(3)油气巡检与运营,(4)非结构化或高混合配置下的工业制造。公司把目标场景描述为“脏、枯燥、危险”的环境;自主机器人能在安全、文档和运营洞察上持续叠加价值,这一框架与四个服务垂直的结构特征高度一致。可作为未来扩张路径的相邻市场包括国防与周界安防、复杂环境中的物流,以及半结构化场景中的城市配送。更合适的市场规模测算方式,是按垂直行业引用分析师估算,而不是套用一个宽泛机器人 TAM;因为不同分析师报告的范围定义差异很大。 [CM001, CM002, CM003, CM004, CM005]
| 细分 / 类别 | 纳入支出 | 排除支出 | 主要买方 / 付款方 | 与 FieldAI 的相关性 |
|---|---|---|---|---|
| 建筑 AI 机器人 | 自主检测、测绘、进度跟踪、安全监测机器人软件;数字孪生平台费用;车队管理 SaaS | 建筑硬件(起重机、挖掘机)、CAD/BIM 软件许可、人工劳动力 | ENR 排名前列总承包商、房地产开发商、公共基础设施业主 | 当前最大垂直;Spot+FFM 已部署于 ENR top-10 企业;检测时间较人工缩短 90%+ |
| 采矿与采掘 AI 机器人 | 自主运输 AI、钻爆机器人智能、勘探 / 检测机器人软件、AI 车队管理 | 采矿机械硬件(矿用卡车、钻机)、散装物料运输、矿石处理系统 | 大型矿山运营商(BHP、Rio Tinto 级别)、运营 VP、HSE 预算负责人 | BHP Ventures 是 FieldAI 投资人;采矿面临严重劳动力短缺和由死亡事故驱动的监管压力 |
| 油气检测与 AI 机器人 | 地上 / 地表检测机器人、管道检测 AI、自主 UAV、预测性维护软件 | 水下 ROV(独立供应商生态)、钻井设备、生产硬件、大宗商品交易系统 | 上游运营商、NOC、炼厂、LNG 终端运营商、HSE 合规预算负责人 | 2026 年垂直规模 $3.8B;检测机器人细分 $0.93B;安全法规和深水扩张是需求驱动因素 |
| 工业制造(非结构化) | 面向高混合、柔性制造的通用机器人 AI;复杂工厂布局中的物流;人形机器人智能 | 面向固定装配线的结构化 SCARA/delta 机器人软件、PLC、SCADA 系统、工业控制硬件 | 工厂运营经理、制造 VP、工厂工程师 | FieldAI 当前足迹较小;人形机器人进入制造后,未来机会很大 |
| 国防与安防检测 | 军事基地 / 周界自主巡逻、自主侦察软件、搜索与救援 AI | 军事硬件采购、武器系统、涉密项目 | DoD 项目办公室、政府安全机构、企业安防公司(例如 Certis Group) | 邻近 FieldAI 的 NASA JPL/DARPA 传承;已建立 Certis Group 合作;商业条款未披露 |
垂直范围基于 FieldAI 已披露的客户垂直和投资人基础。纳入支出遵循 FieldAI 纯软件商业模式;硬件 capex 被排除。油气检测细分价值来自 Fortune Business Insights 2026 年报告;建筑和采矿价值为分析师估算,不同发布方的定义差异较大。由于政府采购条款不透明,国防未纳入 TAM 测算。
[CM001, CM002, CM003, CM004, CM019, CM029]自上而下的规模金字塔,从最宽的工业机器人 TAM 到 FieldAI 在非结构化工业垂直中的受限软件 SAM。所有数值均为 2026 年估计。层级之间的差距反映范围边界不确定;SAM 为作者估算,并非任何分析师发布。
金字塔各层数值整合了多家分析师口径不同的估计,不应相加。建筑机器人数字区间反映 $1.4B(GVR 窄口径)到 $7.79B(TBRC 宽口径);正文采用窄口径,因为它对仅机器人平台更有防御性。SAM 和 SOM 数字为作者估算,置信度低。
[CM006, CM010, CM012, CM013, CM017, CM019]2.2 市场规模测算口径
没有一个公开分析师数字能干净覆盖 FieldAI 的可服务可触达市场,因为多数报告要么定义过宽(纳入硬件或全部工业自动化),要么过窄(只覆盖一个垂直)。最相关的跨领域数字来自 Global Market Insights 对 AI 驱动工业机器人市场的估算:2026 年 $17.9B,按 7.1% CAGR 增长至 2035 年 $33.3B。该数字包含硬件,但能提示 FieldAI 所瞄准的 AI 增强机器人软件层规模。另一个口径是 The Business Research Company 对工业人工智能市场(全部软件、平台、解决方案)的估算:2026 年 $13.69B,并以 51.1% CAGR 指数式增长至 2030 年 $73.54B;这是覆盖制造、能源和物流的最宽用例集合,并不专指非结构化环境。 对 FieldAI 的四个核心垂直,垂直行业规模给出更清晰信号。建筑机器人:2026 年 $7.79B(TBRC)和 $1.4B(Grand View Research,建筑机器人窄口径);区间反映定义范围差异,较低数字对机器人平台本身可能更可辩护。采矿机器人:2026 年 $1.7B(PMR),按 9.8% CAGR 增长至 2033 年 $3.3B。油气机器人:2026 年 $3.8B(MarkWide Research),按 14.6% CAGR 增长至 2035 年 $12.96B;仅巡检子板块 2026 年估计为 $0.93B(Fortune Business Insights)。具身 AI 市场(跨垂直、所有物理 AI 系统)预计到 2030 年达到 $7.24B,CAGR 为 17.5%(Research and Markets)。 对 FieldAI 当前垂直中的纯软件、非结构化环境市场,保守自下而上 SAM 估算可这样推导:把 15–25% 的软件占比假设套到 2026 年建筑 + 采矿 + 油气机器人合计 $13.2B 的垂直硬件与服务市场,得到约 $2–3.3B 的 2026 年可触达软件支出区间。该估算天然不确定,置信度低;FieldAI 没有发布 SAM 数字,也没有分析师发布专门针对“非结构化环境中工业机器人的 AI 软件”的市场估算。IFR 的 World Robotics 2025 报告记录了 2024 年 542,000 台工业机器人安装量,以及全球 4.66M 台在运行机器人;其中有多少部署在 FieldAI FFMs 能创造价值的非结构化环境,并未单独量化。分析师估算分布很宽(见 TM002),相互矛盾的规模测算被保留为证据缺口。 [CM006, CM007, CM008, CM009, CM010, CM011]
| 发布方 | 年份(基准) | 地域 | 市场定义 | 价值(USD B) | CAGR | 方法论 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|---|
| Global Market Insights (GMI) 估算 | 2026 | 全球 | AI 驱动工业机器人(全部垂直,硬件 + 软件) | $17.9B | 到 2035 年 7.1% → $33.3B | 自上而下分析师模型,一手 / 二手研究 | 中 | 包含硬件;AI 标签使用宽泛;不专门针对非结构化环境 |
| The Business Research Company | 2026 | 全球 | 工业人工智能(全部软件、平台、服务、硬件) | $13.69B | 到 2030 年 51.1% → $73.54B | 自下而上市场模型 | 低-中 | 范围极宽;高 CAGR 反映早期基数;包含制造业 LLM/NLP |
| The Business Research Company / Research and Markets 估算 | 2026 | 全球 | 建筑机器人(宽口径) | ~$7.79B | 到 2030 年 18% → $15.39B | 自下而上市场模型;包含自动化技术 | 低-中 | 最宽建筑机器人定义;Grand View 更窄估算为 $1.4B(2024),CAGR 同为 18% |
| Grand View Research | 2024 | 全球 | 建筑机器人(窄口径,仅机器人硬件和软件) | $1.4B | 到 2030 年 18% → $3.66B | 一手和二手研究 | 中 | 定义窄于 TBRC;排除更广义自动化;基准年为 2024 而非 2026 |
| Persistence Market Research 估算 | 2026 | 全球 | 采矿机器人(专用采矿机器人硬件和 AI 软件) | $1.7B | 到 2033 年 9.8% → $3.3B | 一手研究、行业访谈 | 中 | 仅覆盖采矿机器人;排除相邻采矿自动化和车队管理支出 |
| MarkWide Research | 2026 | 全球 | 油气机器人(ROV、检测、施工机器人) | $3.8B | 到 2035/36 年 14.6% → $12.96B | 分析师模型 | 低-中 | 包含 FieldAI 不直接服务的水下 ROV;O&G 口径最宽 |
| Fortune Business Insights | 2026 | 全球(北美占 32%) | 油气检测机器人(仅检测机器人) | $0.93B | 到 2034 年 6.23% → $1.51B | 自下而上分析师模型 | 中 | O&G 口径最窄;排除资产管理平台层和非检测机器人功能 |
| Research and Markets 估算 | 2030 预测 | 全球 | 具身 AI 市场(全部物理 AI 系统,跨垂直) | 到 2030 年 $7.24B | 17.5% CAGR | 自上而下分析师模型 | 低-中 | 最宽具身 AI 口径;2030 年预测而非 2026 年;包含非工业具身 AI |
分析师估算的范围定义差异很大。数值不可相加(覆盖细分相互重叠)。没有任何分析师发布 FieldAI 非结构化环境 AI 软件的 SAM;章节正文中的 $2–3.3B 区间是内部推导,对 $13.2B 合计垂直市场套用 15–25% 软件占比假设,应视为推测。没有 FieldAI 收入披露,就无法支撑 SOM 估算。
[CM006, CM007, CM010, CM011, CM012, CM013]AI 驱动工业机器人市场 2026 年的低 / 基准 / 高估计区间,来自四家分析师来源。宽区间反映不同口径(含硬件 vs. 仅软件;宽口径 vs. 特定垂直)。所有数值均为 2026 年估计,单位为十亿美元。
这些数值不可相加 —— 它们覆盖重叠和嵌套范围。TBRC 与 GMI 使用不同定义和方法。作者推导的 SAM 数字带有推测性(15–25% 软件假设)。中值采用可用的公开点估计;低 / 高边界代表分析师承认的不确定性,或同一市场来源估计的区间。所有数字均为 2026 日历年。
[CM007, CM015, CM016, CM017]2.3 买方、用户与付款方分层
FieldAI 销售 B2B 企业软件,不卖硬件。公司的买方结构遵循典型企业工业技术模式:付款方是最高管理层或运营 / 财务 VP,技术买方是运营或 HSE(健康、安全、环境)负责人,日常用户是现场总监、现场工程师或项目经理。预算归属因垂直而落在两类池子里:(1)用于新自动化投资的资本开支(capex)预算,由 CFO、工程 VP 或运营 VP 控制;(2)用于经常性巡检、监控和现场智能合同的运营开支(opex),由工厂或项目经理在预先批准的服务供应商协议内控制。FieldAI 把产品定位成能在 opex 侧立刻产生 ROI(人工文档时间减少 90%+,更早识别返工),让初始销售可被包装为成本规避,而不是资本开支改造。 建筑板块最直接的购买力集中在最大型总承包商——例如 FieldAI 的 Boston Dynamics 案例研究中提到但未具名的 ENR 排名前列公司。这些公司拥有大量项目积压,承担成本超支风险,也有为自动化试点预留的内部创新或数字化转型预算。采用触发因素包括:无法快速生成准确竣工 BIM 数据,现场文档团队劳动力短缺,以及安全要求规定必须定期巡检危险区域。采矿和能源买家集中在 BHP、Rio Tinto 及同类大型运营商,HSE 与运营负责人受监管合规、事故减少和车队生产率指标驱动。BHP Ventures 投资 FieldAI,构成来自这一买方群体的直接市场验证。油气客户由巡检合规要求、危险环境访问降本和资产完整性管理要求驱动。所有板块的采用路径都从有限概念验证开始(通常是一个站点或资产集群),经 ROI 验证后,进入覆盖多站点或多区域的企业框架协议——这与 FieldAI 提到“全车队部署”和全球运营“企业级”扩张的模式一致。 [CM025, CM026, CM027, CM028, CM029, CM030]
| 细分 | 买方(决策者) | 用户(操作员) | 付款方 | 预算负责人 | 采用触发因素 | 预计年度合同规模 |
|---|---|---|---|---|---|---|
| 建筑(大型项目、基础设施) | ENR 头部总承包商的运营副总裁或首席数字官 | 现场总监、项目经理、BIM 协调员 | 总承包商或资产所有者的资本开支 / 创新预算 | 运营副总裁、CFO | 文档采集人手短缺;现场情报速度提升 90%+;危险区域监测的安全强制要求 | $50K–$500K / 项目 |
| 采矿(露天与地下) | 大型矿山运营商的采矿运营副总裁、HSE 总监 | 矿山运营经理、安全官、班次主管 | 矿山运营资本开支或 HSE 运营开支预算 | 运营副总裁、HSE/ESG 副总裁 | MSHA / 监管死亡事故压力;劳动力短缺(美国西部矿业劳动力 50%+ 到 2029 年退休);自主设备带来生产率提升 | $100K–$1M / 站点 |
| 油气巡检(地面 / 海底上部设施) | 资产 / 工程副总裁、资产完整性经理 | 现场检测工程师、NDT 技术员、HSE 官员 | 资产完整性或 HSE 合规运营开支预算 | HSE / 完整性预算负责人、工程副总裁 | 监管检测合规期限;自主检测降低 OPEX;不可达 / 高危资产暴露 | $100K–$500K / 资产集群 |
| 工业制造(柔性 / 多品种) | 制造副总裁、工厂经理 | 生产经理、机器人集成工程师 | 工厂运营或数字化转型资本开支 | 制造副总裁、CTO | 劳动力成本上涨;柔性要求;多品种产线不适合固定路径自动化 | $20K–$200K / 工厂 |
| 防务与安全 | 政府项目办公室、企业安全副总裁、防务主承包商项目经理 | 外场操作员、安保人员、自主系统操作员 | 政府合同(IDIQ、成本加成)或企业安全运营开支 | 政府合同授权方、企业安全负责人 | 威胁环境;自主周界 / 巡逻能力;已建立 Certis Group 合作关系 | 取决于合同;政府项目规模不一 |
合同规模为示意性估算,依据 FieldAI 披露的定价区间(公司来源称每部署每年 $10K–$500K)以及可比工业软件合同基准。FieldAI 尚未公开客户数或平均合同金额;估算置信度低,应在数据室尽调中核验。
[CM025, CM026, CM027, CM028, CM029, CM030]FieldAI AI 机器人平台的工业客户采用流程,映射从最初认知到企业级舰队部署的买方角色和关键决策关口。
采用路径根据 FieldAI 的 Boston Dynamics 案例研究描述和公司 PR 推断;未披露转化率或销售周期长度数据。「试点炼狱」约束(BuiltWorlds:活跃使用率从 65% 降至 46%)说明,ROI 验证关口是许多行业部署卡住的地方。
[CM022, CM023, CM028, CM039]2.4 增长驱动与采用约束
结构性增长驱动持续数十年,且证据充分。最直接的是劳动力短缺:美国西部超过 50% 的采矿劳动力预计到 2029 年退休,全球建筑业也长期缺熟练工。US Mine Safety and Health Administration(MSHA)在 2024 财年记录了 31 起采矿死亡事故,致死伤害率为每 200,000 工时 0.0110;高危任务转向自主替代方案的监管压力仍在。IFR 的 World Robotics 2025 数据显示,到 2028 年全球工业机器人年安装量将保持在 500,000 台以上;当硬件规格趋同,软件定义自治会越来越成为差异化因素。Deloitte 的 2026 TMT Prediction 确认,有别于消费级 LLM 的专用 AI 基础模型,可能是 2026–2030 年间推动机器人从命令 - 控制转向真正泛化自治的技术突破。 采用约束同样实质。McKinsey 的 Industrial Robotics Survey 显示,71% 的工业公司把硬件资本成本列为机器人采用的主要障碍,61% 则提到缺乏内部经验;装配、表面处理、焊接等复杂活动在中短期内也不太可能自动化。Deloitte 观察到,工业机器人年销量自 2021 年以来基本持平在约 500,000 台,并警告如果数据质量、集成复杂度和网络安全瓶颈不解决,市场很可能保持相对温和的年增长。最关键的是,BuiltWorlds 的 2025 Equipment and Robotics Benchmarking 报告——Construction Dive 在 FieldAI 融资报道中引用——发现报告活跃使用机器人的建筑公司比例从 2024 年的 65% 降至 2025 年的 46%,即便正面评价从 74% 跳升至 95%。热情与部署之间的背离,是近期市场的核心张力,也是 FieldAI 最直接的执行风险:把强试点表现转化为持久企业收入。高 capex 周期、与存量工厂系统的 OT/IT 集成摩擦、较长回本期要求,以及油气运营商所处资本环境波动,都是进一步的结构性逆风。既有市场规模与已部署产品之间的差距真实且重大。 [CM031, CM032, CM033, CM034, CM035, CM036]
| 驱动因素 / 约束 | 类型 | 方向 | 时间 | FieldAI 影响 | 尽调问题 |
|---|---|---|---|---|---|
| 建筑、采矿、能源普遍劳动力短缺 | 驱动 | 正向 | 当前 → 数十年结构性 | 人工替代方案变少,付费意愿和紧迫感上升;试点到合同周期缩短 | 核验各垂直行业劳动力短缺的量化口径,并对应到具体客户预算审批 |
| 安全监管(MSHA、OSHA、IEC 61511、OHS 框架) | 驱动 | 正向 | 当前 → 加速 | 监管强制要求推动采矿和 O&G 更快采用自主化;HSE 预算成为部署触发点 | 跟踪 MSHA 执法动作,并映射 FieldAI 目标客户的 MSHA 合规状态 |
| 自主化成熟(FFM、VLA 模型、边缘 AI) | 驱动 | 正向 | 2026–2028 拐点 | FieldAI 物理优先的 FFM 目前有差异化;竞争对手的 VLA 模型到 2028 年可能缩小差距 | 在真实非结构化部署中,对标 FFM 与竞品基础模型的任务表现 |
| 硬件商品化(Boston Dynamics Spot 价格、人形机器人) | 驱动 | 正向 | 2026–2030 趋势 | 机器人硬件商品化后,软件智能层吃到更多利润 | 监测 Spot 定价趋势,以及 FieldAI 与 OEM 伙伴的版税 / 分成模式 |
| 资本开支周期与 ROI 验证要求 | 约束 | 负向 | 持续 | 企业采购委员会在规模化前需要多年 ROI 证明,拖慢收入确认 | FieldAI 已将多少试点转成多年企业合同?平均销售周期多长? |
| 集成复杂度与遗留 OT/IT 系统 | 约束 | 负向 | 持续 | 运营方使用彼此不兼容的遗留数据格式(BIM、ERP、historian);集成开销推高部署成本 | 评估 FieldAI 与主流 BIM(Autodesk BIM 360)、ERP 和 OSIsoft PI 系统的集成深度 |
| “试点炼狱”(BuiltWorlds:尽管 95%+ 评价正面,活跃机器人使用率仍从 65% 降至 46%) | 约束 | 负向 | 近期(2025–2027) | 正面评价没有转化为活跃部署;全行业试点后流失风险高 | 要求 FieldAI 提供试点转企业合同转化率和流失数据;这是最关键的商业指标 |
| 地缘政治扰动与大宗商品价格波动 | 约束 | 负向 | 事件性 | 油价冲击、工业硬件关税和地缘政治冲突可能让 O&G 与采矿资本开支周期暂停 | 评估 FieldAI 在垂直行业和地域上的收入多元化,判断宏观韧性 |
“试点炼狱”数据来自 Construction Dive 引用的 BuiltWorlds 2025 Equipment & Robotics Benchmarking report。MSHA 死亡事故数据来自 PMR 对 MSHA FY2024 记录的引用。采矿劳动力短缺数字来自 PMR 引用的行业估计。McKinsey 障碍数据来自 2022 Global Industrial Robotics Survey。所有约束严重性评级均为分析师判断;FieldAI 未披露转化率或流失指标。
[CM031, CM033, CM034, CM035, CM036, CM037]FieldAI 企业机器人软件在重工业垂直中的示意部署漏斗。数值代表每个阶段可触达工业运营方的估计占比。FieldAI 未披露实际转化率;数字为示意性行业基准。
转化百分比是企业工业 AI 机器人部署的示意估计。BuiltWorlds 2025 数据点(尽管正面评估率超过 95%,活跃使用率仍从 65% 降至 46%)锚定了试点到活跃使用之间的缺口。FieldAI 未披露管线指标、赢率或转化数据。实际漏斗形态未知,需要数据室访问验证。
[CM020, CM022, CM039, CM041]2.5 图表与要点
03竞争对手
3.1 竞争格局与买方替代方案
到 2026 年,需要自主机器人作业的工业企业实质上有五条路径可选:(1)在现有硬件上部署 FieldAI 的 Field Foundation Models;(2)采用资金充足竞争者的具身 AI 基础模型平台(Intrinsic/Google、Physical Intelligence、Covariant);(3)使用带自有机器人栈的巡检专用平台(Gecko Robotics、Machina Labs);(4)依赖机器人 OEM 捆绑的智能软件(Boston Dynamics Orbit、KUKA、Fanuc 原生 AI);(5)维持现状——人工巡检班组、依赖传统 SLAM/GPS 的导航,或内部软件团队。多数大型企业买家会同时使用不止一种方案,因此多平台并用很常见,转换成本也比初看更软。FieldAI 瞄准最大的未满足切片:动态、拒止 GPS、非结构化现场环境,在这些场景里现状方案和 OEM 原生方案都会失效。公司的硬件无关软件模式避开了机器人制造的资本密集度,同时可覆盖比任何单一 OEM 更多的平台。截至 2026 年 5 月,下文画像中的八家直接和替代玩家合计代表的 VC 与企业资本大幅超过 FieldAI;这既验证了市场,也带来资源充足的竞争者形成稳固位置的实质风险。 [CP001, CP002, CP003, CP004, CP005]
| 竞争对手 | 类别 | 规模 / 融资 | 目标细分市场 | 差异化 | 相比 FieldAI 的限制 |
|---|---|---|---|---|---|
| FieldAI | 具身 AI 软件(横向) | 已融资 $405M;估值 $2B(2025 年 8 月) | 建筑、能源、采矿、物流、防务;异构机器人队伍 | 物理优先、风险感知 FFM;边缘运行、GPS 受限;多智能体机群;硬件无关 | 参照公司 |
| Intrinsic / Google | 机器人软件平台(开发工具 + AI) | Alphabet → Google(2026 年 2 月);Vicarious 收购价 ~$250M;未披露独立融资 | 制造、物流、电子;面向开发者;Foxconn / Fanuc OEM 渠道 | Flowstate 无代码 / 低代码开发环境;集成 Gemini AI;借 Fanuc 1.1M 台机器人形成“机器人 Android”规模 | 聚焦结构化工厂环境;非结构化地形自主能力有限;依赖云 |
| Physical Intelligence (Pi) | 具身 AI 基础模型(研究) | $70M 种子轮 + $400M Series A + $600M Series B + 正在寻求 $1B(合计 ~$2.07B);估值 $11B | 通用机器人;偏操作任务的实验室到生产迁移 | VLA 架构;涌现式泛化;用于精细操作的 RL token;学术背景强 | 截至 2026 年 3 月无商业产品或收入;偏操作任务,尚未在工业现场验证 |
| Covariant | 机器人 AI 平台(仓储 / 操作) | 已融资 $222M;估值 $625M;2024 年 8 月 Amazon 收购创始团队(非独家许可) | 仓储、物流、零售履约(单件拣选、分拣) | RFM-1 80 亿参数多模态基础模型;机群学习;通过 ABB 和 KNAPP 集成商铺渠道 | 收购团队后范围收窄;以仓储为中心;未在户外 / GPS 受限工业环境验证 |
| Gecko Robotics | 检测机器人 + AI 平台(垂直) | 已融资 $347M;估值 $1.25B(2025 年 6 月) | 能源、防务、制造、海军——关键基础设施检测 | TOKA 磁轮机器人 + Cantilever 数字孪生平台;$71M 海军 IDIQ(2026 年 3 月) | 垂直整合(自有机器人),限制 OEM / 硬件无关打法;聚焦检测 |
| Boston Dynamics (Hyundai) | 机器人 OEM + 机群管理软件 | Hyundai 持有(无独立 VC 融资);出售给 Hyundai 的 $14.5B 交易于 2021 年完成 | 工业检测、物流、制造;Atlas 人形机器人到 2028 年进入 Kia / Hyundai 工厂 | Orbit 软件;集成 Google Gemini / DeepMind;2026 年 3 月与 FieldAI 合作,覆盖未知地形 | Orbit 锁定 Boston Dynamics 硬件;不具备具身无关性;非结构化自主能力依赖 FieldAI |
| Machina Labs | AI 驱动制造机器人(垂直) | 已融资 $209M;$124M Series C(2026 年 2 月);Lockheed Martin、Toyota、Woven Capital 参投 | 防务、航空航天、先进出行——机器人钣金成形 | RoboCraftsman AI 成形平台;软件定义工厂;数字设计直达生产 | 应用专用(金属成形),不是通用自主大脑;只在防务机器人预算上重叠 |
| Viam | 开源机器人软件平台(开发者) | 已融资 $117M;2025 年 3 月 Series C | IoT、工业自动化、智能机器——开发者 / 工程团队 | 开源;模块化;多语言 SDK;云端机群管理;Universal Robots 合作伙伴 | SDK / 中间件层没有训练好的自主模型;要达到 FieldAI 同等能力,需要大量开发投入 |
| 现状 / 内部自建 | 人工检测 + 自研 SLAM/GPS 导航 | N/A — 现有人工或内部 R&D 预算 | 拥有内部工程团队的大型工业企业;风险规避型买家 | 无软件许可成本;数据完全可控;可定制 | 难以在 GPS 受限或非结构化环境中可靠运行;人工成本高;没有跨平台机群学习 |
规模 / 融资数据反映截至 2026 年 5 月最新公开报告数字。估值为最后已知投后估值。Boston Dynamics 融资 / 估值采用 2021 年 Hyundai 交易价,不代表当前独立 VC 估值。Physical Intelligence 合计包含已报道的 Series B 和公开提及的进行中融资;可能未包含完整 Series C 交割。
[CP001, CP002, CP006, CP007, CP008, CP009]基于证据的 0–10 序数评分,把 FieldAI 及主要对手放在部署就绪度(x 轴)和具身 AI 泛化广度(y 轴)上。FieldAI 在部署就绪度领先;Physical Intelligence 模型广度领先,但没有商业部署。
序数评分是有证据支撑的编辑估计,并非公开基准。x 轴(工业部署就绪度)反映已记录的生产部署、客户合同证据和多站点舰队规模。y 轴(具身 AI 泛化广度)反映模型架构范围、形态覆盖和任务多样性。Physical Intelligence 由于截至 2026 年初确认没有商业产品,在部署就绪度上评分较低。考虑到证据稀疏,所有评分误差约为 ±1–2 个序数单位。
[CP001, CP008, CP009, CP010, CP011, CP012]3.2 竞争者画像与能力对比
Intrinsic 已在 2026 年 2 月并入 Google,提供 Flowstate,这是面向工业机器人的网页开发与仿真环境。该平台跨 KUKA、Fanuc、Universal Robots 等硬件,且借助 Google Gemini AI 模型做视觉和推理。2026 年 5 月与 Fanuc 的合作让 Google 触达 1.1M 台已部署工业机器人——这是任何初创公司都无法匹敌的分销渠道。Intrinsic 在研发五年后,于 2023 年 1 月裁掉 20% 员工,也没有披露商业收入;但并入 Google 以及与 Foxconn 的合资(2025 年 10 月)显示其商业化意图正在加速。Physical Intelligence(Pi)是一家研究优先的基础模型公司,2024 年创立,截至 2026 年 3 月没有商业产品或收入。其 π0.7 模型(2026 年 4 月)在操作任务上展现出涌现式泛化,并据称正以 $11B 估值融资 $1B。Pi 的 VLA(Vision-Language-Action)架构瞄准实验室到生产的迁移。Covariant 是机器人基础模型早期玩家(RFM-1,8B 参数,2024 年 3 月推出),但其创始人和约 25% 员工在 2024 年 8 月被 Amazon 以团队收购方式吸收,交换条件是一份非排他技术许可。剩余公司由新 CEO Ted Stinson 领导,继续服务非 Amazon 物流客户(服装、杂货、医药),但面临公司层面的不确定性;竞争性数据积累者也让其护城河偏窄。Gecko Robotics 把自有 TOKA 巡检机器人与 Cantilever AI 平台结合,服务关键基础设施数字孪生;公司 2025 年 6 月达到独角兽状态(估值 $1.25B),并在 2026 年 3 月赢得一份 $71M、五年期 U.S. Navy IDIQ 合同。Gecko 的市场深扎能源、国防和制造巡检,但结构上比 FieldAI 的横向野心更窄。Boston Dynamics(Hyundai 旗下)提供 Spot、Stretch、Atlas 机器人,以及用于车队智能的 Orbit 软件平台。2026 年 3 月,Boston Dynamics 宣布与 FieldAI 直接合作,把 Spot 延伸到未知、动态环境——这让 Boston Dynamics 同时成为硬件伙伴和潜在软件竞争者,因为 Orbit 正在扩大与 Google Gemini 的 AI 集成。Machina Labs 专注 AI 驱动的机器人钣金成形(RoboCraftsman),服务国防和航空航天,并在 2026 年 2 月完成 $124M Series C。它不是 FieldAI 巡检与自治重点的直接竞争者,但会争夺国防预算和 OEM 机器人心智。Viam 是开源机器人软件平台(总融资 $117M),面向希望在 IoT、机器人和工业机器之间抽象硬件的开发者。Viam 争夺同一个“大脑层”叙事,但把自己定位为开发者 SDK 路线,而不是 FieldAI 的预训练 AI 模型。内部自研仍是既有方案:大型工业公司(能源、采矿、建筑)保有内部机器人和软件工程团队。自研单位经济很难量化;当没有商业平台满足所需安全认证或数据驻留约束时,自研就会成为默认选择。 [CP006, CP007, CP008, CP009, CP010, CP011]
| 采购标准 | FieldAI | Intrinsic / Google | Physical Intelligence | Covariant | Gecko Robotics | Boston Dynamics Orbit | Viam |
|---|---|---|---|---|---|---|---|
| GPS 受限 / 非结构化地形自主能力 | 是 — 核心能力;边缘运行,无需预建图 | 否 — 工厂结构化环境;依赖 GPS/SLAM | 未知 — 实验室操作任务场景;未确认现场部署 | 否 — 仅限结构化仓储环境 | 部分 — 室内 GPS 受限爬行;不是通用移动能力 | 部分 — 根据合作关系,未知地形需要 FieldAI FFM | 否 — 仅 SDK;用户需自备自主模型 |
| 硬件 / 具身形态无关 | 是 — 四足、人形、轮式、履带式、AV | 是 — Fanuc、KUKA、Universal Robots 等 | 部分 — 聚焦操作任务;未通过现场机器人验证 | 部分 — 机械臂 / 仓储操作机器人 | 否 — 专有 TOKA 机器人;仅自有硬件 | 否 — 仅 Spot/Stretch/Atlas | 是 — 开放 API 覆盖多种硬件类型 |
| 边缘推理(无需云) | 是 — <100ms 延迟;可完全断网隔离 | 否 — 依赖 Google Cloud | 未知 — 研究架构未披露边缘方案 | 未知 — 未记录边缘 / 离线运行能力 | 部分 — 实时传感器流;需要云分析 | 否 — Orbit 仪表盘和洞察需要连接 | 部分 — 有边缘 SDK;建议使用云管理 |
| 多智能体 / 机群协同 | 是 — Multiagent Foundation Model(MFM);机群级编排 | 部分 — 任务编排,但不是多机器人 AI 协同 | 未知 — 无产品文档 | 是 — 跨站点机群学习,但非实时多智能体 | 否 — 每站点单机器人工作流 | 是 — Orbit 集中管理多台 Spot/Stretch 机器人 | 是 — 机群管理,带监控和 OTA 更新 |
| 防务 / 政府资质准备度 | 是 — FieldAI Federal 子公司;DARPA 背景;DoD 管线 | 否 — 聚焦商业;未披露联邦子公司 | 否 — 研究阶段;未披露政府项目 | 否 — 聚焦商业物流 | 是 — 美国海军 IDIQ $71M;承接空军和海军工作 | 部分 — Hyundai 所有权限制单方面签美国 DoD 合同 | 否 — 聚焦商业 |
| 基础模型 / 跨任务 AI 泛化 | 是 — Field Foundation Models(FFM);物理优先、风险感知 | 是 — Flowstate + Gemini AI;靠无代码技能实现任务泛化 | 是 — π0.7 VLA;跨任务涌现式泛化(实验室) | 是 — RFM-1 80 亿参数;多模态;仓储泛化 | 部分 — AI 预测分析;不是通用机器人推理 | 部分 — Gemini 视觉语言;面向检测的推理 | 否 — SDK 平台;模型需用户自备或来自模型注册表 |
| 数字孪生 / 检测分析 | 部分 — 通过 Spot 合作集成 BIM / 数字孪生 | 部分 — 通过 Flowstate 仿真;非实时检测 | 否 — 未记录为产品能力 | 否 — 聚焦仓储操作 | 是 — Cantilever 数字孪生;3D 厚度图;AI 异常检测 | 是 — Orbit 历史站点目录;视觉、声学、热数据 | 部分 — 数据管道通向云端;不是完整数字孪生平台 |
单元格反映截至 2026 年 5 月的公开产品文档和新闻稿。所有“未知”单元格代表真实证据缺口,不是假定无能力。序数标签(是 / 否 / 部分 / 未知)为基于引用来源的编辑判断,不是厂商发布的基准。
[CP003, CP004, CP005, CP018, CP019, CP020]围绕七个竞争者的六项购买标准做序数能力评估。单元格含义:有 = 来源确认,部分 = 部分 / 有限,否 = 确认不存在,?= 未知 / 证据不足。
有 = 主要来源确认,部分 = 部分证据,否 = 确认不存在 / 不在范围,?= 公开证据不足。本图采用与 TP002 不同的证据视角(能力是否存在的二元判断),而不是按标准逐项文字描述。
[CP003, CP004, CP005, CP015, CP016, CP017]3.3 定价与包装对比
整个竞争领域的定价透明度都很低;多数平台面向企业采购销售,没有公开标价。FieldAI 的订阅据称按部署规模不同,从每年数万美元到 $500,000 不等;公司引用第三方估算称 2025 年 ARR 为 $140M,并称已签约收入达到 $100M 或以上。Intrinsic 不公布 Flowstate 定价;Google 的云优先模式暗示会在 Google Cloud 计算费之上叠加企业 SaaS 档位,而 Fanuc 合作意味着可能存在批量授权。Physical Intelligence 没有商业产品或定价。Gecko Robotics 把巡检服务、Cantilever 平台访问和工程咨询打包成多年合同;Navy IDIQ 展示的是 18 艘船、五年 $71M 的合同框架。Boston Dynamics Orbit 作为硬件之上的软件层出售;年费未披露,但 Michelin 和其他企业案例研究把它称为持续运营软件。Viam 提供对开发者友好的免费档,并按需提供企业定价;开源模式先推动开发者采用,再向托管基础设施追加销售。Machina Labs 按制造服务合同定价,而不是软件订阅。行业估计,美国海军单是现状人工巡检每年就花费 $13–20B;按 Boston Dynamics 合作公告,FieldAI 支持的系统可让建筑工地文档时间减少 90% 以上。 [CP021, CP022, CP023, CP024, CP025, CP026]
| 竞争对手 | 价格 / 单位 / 合同模式 | 包含能力 | 已知折扣或未知项 | 买方影响 |
|---|---|---|---|---|
| FieldAI | 数万美元至 ~$500K / 年 / 部署(第三方估计);$140M ARR(2025 年估计) | FFM 软件许可;传感器-计算载荷;边缘推理;云分析;联邦模型更新 | 企业合同;取决于部署规模;确切价格未公开披露 | 价格区间宽,暗示分层部署模式;小型试点可进入;企业级规模昂贵 |
| Intrinsic / Google | 未公开披露;预计为企业 SaaS + Google Cloud 计算 | Flowstate 开发环境;仿真;Gemini AI 模型;支持 KUKA / Fanuc / UR 硬件 | Fanuc 合作条款未披露;开发者访问可能为挑战赛参与者提供免费层 | Google 规模可能支撑激进定价,以抢 OEM 量;FieldAI 面临直接价格竞争风险 |
| Physical Intelligence | 无商业产品;定价未知 | 无 — 研究阶段 | N/A | 短期无定价威胁;若 Pi 在 2026–2027 年商业化,相关性会上升 |
| Covariant | 带性能保证的 SLA;多年企业合同;单位经济性与吞吐量挂钩(仓储) | Covariant Brain AI;机械臂集成;站点专属训练;机群学习 | 收购团队后的定价策略未公开更新;Amazon 关系可能压缩利润率 | SLA 模型成熟;仓储专用;不适用于 FieldAI 现场部署定价场景 |
| Gecko Robotics | 多年服务合同;5 年 $54M–$71M 海军 IDIQ;商业私有定价 | TOKA 机器人部署;Cantilever SaaS 平台;工程咨询;数字孪生 | 检测服务 + 软件打包;按资产覆盖范围定价,不按席位 | 服务模式毛利更高,但需要 Gecko 自有机器人;不是纯软件打法 |
| Boston Dynamics Orbit | 按机器人 / 站点的软件许可;价格未公开披露;与 Spot/Stretch 采购打包 | 机群仪表盘;视觉-声学-热检测;Gemini 视觉;任务调度;CMMS/WMS 集成 | 绑定硬件定价;企业支持包增加成本 | Spot 机群部署后切换成本高;Orbit 成本并入更大的 BD 硬件合同 |
| Viam | 开发者优先免费层;企业定价需询价;总融资 $117M(非披露收入) | 开源 SDK;多语言 API;云端机群管理;Viam Registry 提供驱动 / 模型 | 开源核心消除供应商锁定;企业层增加托管基础设施 | 入门成本低;吸引绿地部署;买方需自备自主模型 |
| 现状 / 内部自建 | 人工:熟练检测员 $50–$200 / 小时;内部机器人 R&D 的 CAPEX($1M–$50M 区间) | 完全控制;集成现有工作流;无软件许可 | 隐性成本:安全事故、停机、慢周期数据采集 | 高危或大规模环境中总成本最高;没有复利式模型改进 |
所有定价数字要么来自公开报告的第三方估计,要么明确标注为未披露。FieldAI ARR 估计来自 Getlatka / Sacra,未经 FieldAI 确认。政府合同金额来自新闻稿。内部自建成本区间为行业典型值,并非特定站点估算。
[CP021, CP022, CP023, CP024, CP025, CP026]3.4 差异化可持续性、护城河与反向证据
FieldAI 声称的护城河建立在四根柱子上:(1)用专有真实部署数据训练的物理优先、风险感知基础模型;(2)不依赖云或 GPS 的端侧部署,使其进入云优先竞争者无法进入的市场;(3)来自建筑、能源、采矿和国防等不断增长机队的数据飞轮;(4)由 Boston Dynamics、Certis Group 和 Carnegie Mellon Robotics Innovation Center 锚定的合作伙伴生态。现阶段独立评估这条护城河很难:部署数据量和模型性能基准没有公开披露;硬件无关定位是优势,也意味着竞争者可以追逐同样的机器人伙伴。几项反向信号值得关注。第一,Physical Intelligence 正以 $11B 估值融资——是 FieldAI 的 5.5×——即便它没有商业产品。如果 Pi 在 FieldAI 建立主导数据优势前实现商业化,更强资本基础可能压缩 FieldAI 的差异化窗口。第二,Google/Intrinsic/Fanuc 集成(1.1M 台已部署机器人获得 AI 升级)带来 FieldAI 无法靠自身复制的装机基础杠杆。第三,UpsideList 2026 年 3 月分析给出 50% 概率的熊市情景:NVIDIA Isaac 或 Boston Dynamics 内化自治层,通过降价轮抹掉 FieldAI 普通股价值。关键在于,FieldAI 最显眼的合作伙伴 Boston Dynamics 也运营 Orbit 软件平台,背后有 Google Gemini,且可以选择开发竞争性自治能力。第四,对 Covariant——一个仓库 AI 类比公司的独立分析,把其数据护城河评为“窄而非宽”,理由是 Physical Intelligence、Nimble 和 Amazon 都在竞争性积累数据;这一逻辑适用于任何具身 AI 平台。基础模型商品化风险真实存在:随着 VLA 模型通过开源发布(Pi 已开源早期模型)和超大规模云厂商 API 扩散,物理优先差异化可能被压缩成商业秘密,而非结构性护城河。FieldAI 的国防垂直(由前 DARPA PM Eric Krotkov 领导的 FieldAI Federal)以及拒止 GPS 的端侧部署要求,可能成为最可防守的细分,因为它们需要硬件认证和安全许可,能形成独立于模型质量的监管转换成本。 [CP027, CP028, CP029, CP030, CP031, CP032]
| 护城河主张 | 威胁 | 严重性 | 证据基础 | 缓解措施 / 尽调问题 |
|---|---|---|---|---|
| 物理优先 FFM 数据飞轮 | Physical Intelligence 和开源 VLA 资本更多,可能更快积累竞争性部署数据 | 高 | Pi 按 $11B 估值融资 $1B(thelec.net);开源 π 模型可用 | 核验 FieldAI 当前机群规模和数据量相对 Pi 的差距;确认模型更新节奏;评估数据独占性 |
| 边缘 GPS 受限部署(无云) | 如果 Gemini 模型压缩推进,Google Cloud / Intrinsic 云原生架构可能适配边缘 | 中 | Fanuc / Google 合作(thenextweb.com);NVIDIA Isaac 边缘计算路线图 | 评估边缘模型压缩轨迹;量化 FieldAI 目标环境的延迟要求 |
| 硬件无关的“通用机器人大脑” | 机器人 OEM(Boston Dynamics、Fanuc)可能更偏好捆绑内部 AI,或选择排他的 Google / NVIDIA AI 合作 | 高 | BD Orbit + Google Gemini 集成(bostondynamics.com);Fanuc / Intrinsic 合作(thenextweb.com) | 确认 Boston Dynamics 及其他 OEM 合作中是否存在合同排他性 |
| 多智能体机群协同 | Boston Dynamics Orbit 已能集中管理机群;NVIDIA Isaac ROS 支持多智能体仿真 | 中 | BD Orbit 多站点机群管理已确认(bostondynamics.com);Orbit 与 FieldAI 合作覆盖地形缺口 | 区分 FieldAI 实时多智能体 AI 推理与 Orbit 编排层;取得第三方基准 |
| 防务 / DoD 资质护城河 | Gecko Robotics 已有 $71M 海军 IDIQ;其他 DoD 主承包商(FLIR、Teledyne)可争夺 DoD AI 检测 | 中 | Gecko 海军合同(ubos.tech);FieldAI Federal 子公司已披露,但未宣布公开 DoD 合同 | 获取 FieldAI Federal 合同管线;对比其 DARPA 背景与 Gecko 现有海军认证 |
| 优先权堆栈风险(资本结构) | $506M 优先清算权压在 $2B 估值上;独立分析称,熊市情景下普通股可能被抹去 50% | 高 | UpsideList 2026 年 3 月:「熊市(50%)—NVIDIA Isaac/BD 内部化;降价轮;$506M 优先权抹去普通股」 | 获取完整股权结构表、优先权条款和清算瀑布;为普通股持有人建模下行情景 |
| 跨行业数据多样性广 | Amazon 借 Covariant 授权积累 750,000 台机器人规模的操作数据;Google 借 Intrinsic/Fanuc 积累数据 | 高 | Amazon Covariant 授权(techcrunch.com);Fanuc/Google 1.1M 台机器人合作(thenextweb.com) | 确认 FieldAI 在异构环境中的机队规模;量化其数据多样性与 Amazon/Google 规模的差距 |
严重性评级是基于证据强度与资本 / 渠道差距的编辑判断。「高」表示一旦威胁兑现,可能在 3 年内显著削弱 FieldAI 的竞争地位。「中」表示威胁真实存在,但推进更慢,或有缓释空间。
[CP027, CP028, CP029, CP030, CP031, CP032]截至 2026 年 5 月,FieldAI 相对威胁格局的关键竞争耐久性指标简表。
Physical Intelligence 估值溢价:$11B Pi / $2B FieldAI = 5.5×;两者均取最近报道轮次。优先权负担:据 UpsideList,$506M / $2B = 25.3%。所有其他数值来自引用的主要来源。
[CP028, CP029, CP030, CP031, CP032, CP035]3.5 图表与要点
04财务
4.1 收入来源与变现策略
FieldAI 的收入模式是 B2B 企业模式,围绕两条主线:(1)客户首次把 FieldAI 的传感 - 计算载荷和固件部署到现有第三方机器人上时,收取前置硬件集成费;(2)客户持续访问 Field Foundation Models(FFMs)、车队管理和分析服务时,收取经常性软件授权费。公司不制造机器人,也明确避开硬件 OEM 定位,使模式更接近企业软件订阅,而不是机器人硬件厂商。所有设备端推理都在端侧运行,延迟低于 100ms;车队数据会上传到 FieldAI 云平台,用于分析和联邦式模型改进。 标价未公开披露。第三方来源把订阅区间描述为“每年数万美元到 $500,000”,取决于部署规模;CEO Ali Agha 则引用美国、欧洲和亚洲的“数百万美元合同”证明大型企业交易规模。Construction Dive 和 Sacra 确认了 B2B 硬件集成加软件授权结构。没有证据显示公司存在面向消费者的收入流、市场平台模式或硬件 OEM 收入。面向国防的子公司 FieldAI Federal 已经运营,未来可能贡献政府合同收入,但尚未公开披露任何合同授予。 唯一公开收入信号来自 Orange County Business Journal 2026 年 4 月报道:匿名“知情人士”称公司已拥有超过 $100M 的已签约收入。GetLatka 的分析师平台估计,截至 2025 年 11 月,公司 ARR 为 $140M。两个数字都未经审计、未经公司确认,也没有按收入流拆分。因此,这些数据点只能作为方向性信号,而不是已确认财务指标。 [CI001, CI002, CI003, CI004, CI005, CI006]
| 来源 | 机制 | 单位 | 当前价值 / 状态 | 收入质量 | 尽调请求 |
|---|---|---|---|---|---|
| 软件授权(FFM) | 按站点或按机器人订阅,获得 FFM 访问、机队分析和持续模型更新 | 年度经常性合同 | 早期;预订收入 >$100M(匿名,2026 年 4 月);ARR 估计 $140M(GetLatka,2025 年 11 月) | 中 — 经常性结构是加分项;软件与服务的收入占比未核实 | 披露经审计的 ARR、ACV、NRR 和客户数量 |
| 硬件集成(一次性) | 前期把传感器 / 计算载荷部署到客户机器人,并完成固件集成 | 按部署收费 | 包含在预订收入中;相对经常性收入的占比未披露 | 低 — 一次性收费削弱收入可预测性;利润率可能低于软件 | 披露授权收入与服务收入拆分 |
| 边缘分析 / 云平台 | 机队数据汇总、分析仪表盘、联邦学习基础设施 | 包含在软件订阅中,或打包销售 | 未单独披露;可能打包进授权费 | 低 — 没有证据表明这是可单独变现的 SKU | 确认分析功能是单独加购,还是包含在基础授权中 |
| 政府 / 国防(FieldAI Federal) | 通过 FieldAI Federal 子公司为军事和高危侦察任务提供自主 AI | 政府合同(IDIQ、成本加成或固定价格) | 未披露公开中标合同;子公司活跃,并有 DARPA 背景 | 未知 — DoD 市场大,但销售周期为 18–36 个月 | 确认是否有 DoD 合同工具(OTA、SBIR)及收入贡献 |
收入数据来自第三方估计(GetLatka,2025 年 11 月)或匿名来源(OCBJ,2026 年 4 月);公开渠道没有经审计或公司确认的财务数据。「预订收入」不是 ARR,可能包括一次性集成费和管线承诺。
[CI001, CI002, CI003, CI005, CI006]| 数据点 | 描述 | 类型 | 来源 | 置信度 |
|---|---|---|---|---|
| 数万美元到 $500K/年 | 据称每个客户部署的订阅定价区间 | 标价区间(第三方报告) | Sacra;Construction Dive;跨行业分析 | 低 — 公司未确认;区间很宽 |
| 数百万美元级合同 | CEO Agha 在 Reuters/ET 采访中提到具体交易规模(2025 年 8 月) | CEO 陈述的合同规模 | Reuters / Economic Times(2025 年 8 月) | 中 — CEO 陈述;未经审计;可能只反映最大合同 |
| 预订收入 >$100M | OCBJ 匿名来源称,截至 2026 年 4 月,累计预订收入达到该规模 | 累计订单信号 | Orange County Business Journal(2026 年 4 月) | 低 — 匿名第三方来源;未经审计;「预订」与已确认收入边界不清 |
| $140M ARR(估计) | GetLatka 分析师平台截至 2025 年 11 月的估计 | 分析师估计 | GetLatka(2025 年 11 月) | 低 — 可能来自自报或模型估计;方法未披露;未经审计 |
FieldAI 所有定价和收入数据点都来自第三方报道、匿名来源或分析师估计。公开渠道没有公司发布的定价表、收入披露或经审计财务报表。所有数字都应视为方向性信号。
[CI004, CI005, CI006, CI007]基于已披露的模型结构,FieldAI 企业客户互动如何从初始接触流向经常性授权,并转化为估计毛利润。
所有美元数字均为估计或披露区间;没有经审计财务数据。节点数值仅表示方向。
[CI001, CI002, CI003, CI016]4.2 资本充足性与后续资金续航
FieldAI 的两轮融资历史已在公司概览章节展开。看资本充足性,关键输入是:截至 2025 年 8 月,公司累计募资 $405M;第二轮融资($314M)在 2025 年 8 月超额认购完成,由 Bezos Expeditions、Prysm Capital 和 Temasek 共同领投,Khosla Ventures、NVentures(NVIDIA)、Intel Capital、Canaan Partners、BHP Ventures 和 Emerson Collective 参投。第一轮融资(约 $91M)在 2024 年完成,估值 $500M;Gates Frontier 和 Samsung 被列为上一轮投资方。 公司未披露账上现金,也未披露月度烧钱速度。人员规模从 2024 年末约 30 人增至 2025 年 8 月公告时约 130 人,并明确计划到 2025 年末翻倍至约 260 人,说明固定成本基数正在快速拉升。若按 200–260 名员工、单人全包成本 $250K–$350K(与 Irvine, CA 及分布式工程团队匹配)估算,仅年化薪酬开支就达 $50–90M/年,对应月度人工 run-rate 约 $4–8M。再叠加全球部署费用、云基础设施、硬件载荷成本和 G&A,总月度 burn 估计为 $5–10M。低端情形下,从成立算起,累计募资 $405M、月 burn $5M 对应约 81 个月的毛 runway;若月 burn $10M,则约 40 个月。扣除假设的 $100M+ 累计收入抵消后,自 2025 年 8 月交割起的净 runway 估计为 24–48 个月;若收入不加速、成本不意外上行,下一轮融资触发点落在 2027 年末至 2029 年初。 FieldAI 尚未宣布任何债务融资、项目融资义务、收入分成融资或授信额度。公司资本结构看起来是纯股权。与纵向一体化机器人公司不同,FieldAI 不做硬件制造,这显著降低了相对于同业的营运资本和库存融资需求。 [CI009, CI010, CI011, CI012, CI013, CI014]
| 指标 | 数值 | 来源 / 依据 | 置信度 | 备注 |
|---|---|---|---|---|
| 已筹集总资本 | $405M | 官方(fieldai.com 公告,2025 年 8 月) | 高 | 官方公告和多家独立媒体均确认;两轮融资 |
| 投后估值 | $2B | Reuters / CNBC 知情人士(2025 年 8 月) | 高 | 高于上一轮的 ~$500M;消息源称,最新 $314M 融资对应 $2B 投后估值 |
| 资本效率比(估值 / 已融资) | 约 ~4x–5x($2B / $405M) | Premier Alts;推导 | 中 | Premier Alts 报告 3.95x;与高增长 AI 软件溢价一致,但财务数据未验证 |
| 估计月度现金消耗(区间) | $5–10M/month | 根据员工数和全球扩张范围估计 | 低 | >200 名员工,全成本 $250K–$350K + 基础设施 + 部署运营开支;未披露 |
| 从 2025 年 8 月交割推算的毛现金可支撑期 | 24–48 个月(2027 年末至 2029 年初) | 估计 | 低 | 假设月度现金消耗为 $5–10M,收入抵消会随时间增加;未披露账面现金 |
| 债务 / 项目融资义务 | 未公开披露 | N/A | Unknown | 未宣布信贷额度、收入分成融资或项目融资;结构看起来是纯股权 |
| 下一轮融资触发因素 | 未公开披露 | N/A | Unknown | 截至 2026 年 5 月,未宣布 IPO 时间表、Series B 计划或 M&A 活动 |
现金消耗率和资金可支撑期基于员工数数据及可比公司基准推导。账面现金、实际月度现金消耗和下一轮资本计划未公开披露。本表引用公司概览章节的完整融资时间线;本章中的论断均有独立来源。
[CI009, CI010, CI011, CI013, CI014, CI015]截至 2026 年中,FieldAI 关键财务估计的来源支持边界。所有数字均为估计;没有经审计数据。
收入下限($50M)采用较保守口径,把预订额与已确认 ARR 区分开;收入上限($140M)采用 GetLatka 分析师估算。烧钱测算来自员工数代理指标。现金续航期按收入抵消后的净烧钱额计算。
[CI005, CI006, CI013, CI014, CI022]FieldAI 已融资 $405M 在主要支出类别中的流向;区分低资本开支的软件模式与资本强度更高的运营优先项。
分配比例根据已披露重点(全球扩张、产品开发、招聘、伙伴关系)估算;公司没有正式披露资本分配。
[CI009, CI010, CI011, CI012]4.3 单位经济模型与成本结构
FieldAI 基本没有披露单位经济模型。公司未发布经审计财务报表,也未披露毛利率、客户获取成本(CAC)、净收入留存(NRR)、年度合同价值(ACV)或回本周期。以下分析基于可比 B2B 工业 AI 软件公司,以及 FieldAI 收入模式中可观察到的结构。 按典型企业 AI 软件基准估算,软件许可部分的毛利率约为 55–80%。硬件集成服务涉及实体传感器-计算载荷部署和现场支持,利润率可能较低(20–45%),会把混合毛利率拉到纯软件可比公司之下。规模化之后,软件许可收入应成为主导,并改善混合毛利。端侧架构不依赖云端推理,相比重云端的 AI 平台,每次部署的持续可变成本更低,这是有利的结构特征。 企业工业机器人销售效率天然承压:销售周期通常为 12–24 个月,需要现场演示和集成测试,还牵涉多方采购。CEO Agha 提到合同价值为数百万美元,说明单笔规模或可在中等流失率下支撑 1–3 年 CAC 回本——但没有披露 CAC 和 NRR 数据,无法验证。若 $140M ARR 估计和 150 人员工数准确,则人均收入约 $900K–$1M,对这一阶段的企业软件公司而言很强。FieldAI 的 GTM 路径聚焦直销企业、硬件伙伴关系(Boston Dynamics、OEM)和地域扩张伙伴(Certis Group、亚洲);这也说明如果只靠纯直营,可服务市场天花板会被压低。 [CI016, CI017, CI018, CI019, CI020, CI021]
| 指标 | 数值 / 估计 | 置信度 | 为何重要 | 尽调请求 |
|---|---|---|---|---|
| 毛利率(软件授权) | 估计 65–80%(企业 AI 软件行业基准) | 低 — 估计值;未披露 | 长期现金生成的核心驱动;必须高于边缘计算和支持的边际成本 | 要求管理账按收入来源拆分毛利瀑布 |
| 毛利率(硬件集成服务) | 估计 20–45%(现场服务与集成工作) | 低 — 估计值;未披露 | 服务毛利会稀释综合毛利;服务与授权的占比是关键 | 披露服务与软件收入拆分及各自毛利率 |
| 综合毛利率 | 估计 55–75%(取决于授权 / 服务组合) | 低 — 估计值;未披露 | 影响估值:软件级毛利支撑溢价倍数;混合毛利不支撑 | 要求提供最新 P&L 或贡献毛利数据 |
| 客户获取成本(CAC) | 未披露;考虑 12–24 个月销售周期,估计每个企业客户 $200K–$1M | Unknown | CAC 高,即便回本期短,也会带来现金流风险;回本期长则需要 NRR 足够粘 | 要求按客户细分披露综合 CAC 和回本期 |
| 净收入留存率(NRR) | 未披露;官方公告提到「扩展合同」 | Unknown | NRR >120% 将验证先落地再扩张模式,并显著抬高 LTV 估计 | 要求客户 cohort 留存数据;确认扩展合同是否抬高 ACV |
| 平均合同价值(ACV) | 基于数百万美元交易说法和定价区间,估计 $300K–$2M | 低 — 估计值 | ACV 驱动 LTV/CAC 分析;估计区间很宽,说明信息缺口大 | 披露 ACV 四分位数;确认 ACV 是否包含集成费 |
| 人均收入 | 估计 ~$900K–$1M(若 $140M ARR / 150 名员工) | 低 — 分子和分母都是估计值 | 若准确,该指标是强生产率信号;验证软件模式可扩展 | 同时确认员工数和 ARR,验证该比率 |
| 燃烧倍数(每新增 $1 ARR 消耗的资本) | 估计 1.5x–3x(与激进增长期 AI 同业一致) | 低 — 估计值;未披露 | 燃烧倍数高说明资本效率低;高效增长的目标应 <1.5x | 要求管理层提供滚动 12 个月现金消耗和新增 ARR |
所有单位经济模型数值都依据可比上市公司基准和可观察运营数据(员工数、定价区间、交易规模说法)估计。没有经审计数字。置信度整体偏低;这些估计应作为情景建模的方向性输入,而非确认指标。
[CI016, CI017, CI018, CI019, CI020, CI021]用定性方式呈现单位经济如何从赢单推进到年度合同价值、成本扣除和多年生命周期价值。由于缺少公开数据,所有输入均为估计。
FieldAI 未披露实际 CAC、ACV 或 NRR 数据。所有节点均根据行业基准和可观察交易规模主张估计。应视为情景结构,而非已确认指标。
[CI017, CI018, CI019, CI020]4.4 财务不透明与尽调阻点
FieldAI 是私营公司,没有公开财务报告义务。截至 2026 年 5 月,仍未披露或无法验证的关键财务指标包括:(1)经审计 ARR 或总收入;(2)按收入流拆分的毛利率;(3)CAC 和回本周期;(4)净收入留存率;(5)客户数量与集中度;(6)账上现金;(7)月度或季度 burn。 现有收入信号彼此冲突。GetLatka 估计 ARR 为 $140M(2025 年 11 月),而其他分析平台曾给出低至 $5M 的数字——相差 28 倍,反映了外部估算私营公司指标的难度。UpsideList 的社区模型标记 -2% 上行空间,并从二级市场定价角度把公司视为近似尚未产生收入,指向私募二级层面的流动性和估值不确定性。 若以 $140M ARR 估计为基准,$2B 估值对应 14x 收入倍数;若以 $100M 已签约收入信号为基准,则对应 20x。对一家未披露盈利能力、没有公开财务记录、且所在市场中至少 54% 建筑垂直机器人部署(公司披露的最大市场)仍停留在试点或搁置状态的公司而言,这两个倍数都很激进(据 BuiltWorlds 2025 数据)。只有在 ARR 快速增长且利润率很高时,这些倍数才站得住脚——但现有数据无法确认任一条件。FieldAI 未向 SEC 提交 Form D(经 EDGAR 全文搜索确认),意味着公司要么使用了无需 Form D 的豁免,要么使用了不同法律实体,要么尚未提交。加州实体 “Field Ai, Inc.”(文件 #6436098,2024 年 10 月 24 日注册,Delaware 设立,CEO Aliakbar Aghamohammadi)处于 active 状态。 二级市场分析师给出的负面估值信号和宽幅收入估计区间本身不构成否决项,但意味着要承销 $2B 估值,必须通过正式尽调拿到非公开财务信息。 [CI022, CI023, CI024, CI025, CI026, CI027]
| 缺失指标 | 重要性 | 对分析的影响 | 尽调路径 |
|---|---|---|---|
| 经审计年度收入 / ARR | 阻断 | 无法确认 $100M 订单或 $140M ARR 估计是否反映已确认收入;估值倍数无法验证 | 要求在 NDA 下提供管理账或经审计 P&L;可接受滚动 12 个月 ARR 数字 |
| 按收入来源拆分的毛利率 | 重大 | 无法判断综合毛利是否支撑 $2B 估值,也无法判断服务拖累是否限制软件溢价 | 要求按软件授权与服务分拆贡献毛利瀑布 |
| 客户获取成本和回本期 | 重大 | 无法验证 GTM 效率或每 1 美元新增 ARR 的资本强度;长企业销售周期意味着 CAC 高 | 要求提供销售漏斗数据、交易级成本分配,以及 2024–2026 年客户 cohort 的 CAC |
| 净收入留存率 | 重大 | 无法验证先落地再扩张说法;即便 $140M ARR 上只有 10% 流失,也会显著影响 LTV | 要求按产品线提供季度 cohort 留存;确认「扩展合同」是否抬高 ACV |
| 账面现金(资产负债表) | 阻断 | 无法确认资金可支撑期估计;现金消耗率可能与基于员工数的替代指标有重大差异 | 要求提供最新资产负债表,或最近月份账面现金 |
| 客户数量与集中度 | 重大 | 收入可能集中在 3–5 个早期采用者;客户流失会形成非对称风险 | 要求在 NDA 下提供客户明细;确认前 5 大客户收入集中度 |
六项缺口都是投资前尽调的标准要求。对一家处于该阶段的私营公司而言并不异常,但在用任何基本面锚定 $2B 估值前,六项都必须补齐。
[CI022, CI023, CI024, CI025, CI026, CI027]4.5 财务结论
FieldAI 的财务图景是一家资本充足、收入早期但增长很快的软件公司:24 个月内吸引 $405M 融资,但对外部承销仍过于不透明。$2B 估值完全由投资人信号、增长轨迹和团队履历托住,而不是由已验证的财务基本面支撑。若 $140M ARR 估计准确,且同比增长 50–100%,这一估值落在品类开创型 AI 软件公司的合理区间;若 ARR 明显更低或增长更慢,倍数就难以自圆其说。前瞻资本充足性看起来稳健:$405M 融资对应纯软件运营模型(无硬件制造 capex、无库存融资),再加上披露的 >$100M bookings,意味着下一轮融资前至少有 2–3 年运营 runway。最大尽调风险不是偿付能力,而是信息不对称——公开可验证信息与承销估值所需信息之间的缺口,对一家 $2B 公司而言异常大。 [CI030, CI031, CI032]
4.6 附录
05产品与技术
5.1 产品定义与模块地图
FieldAI 是一家纯 AI 软件公司,不制造机器人。它的商业产品是 Field Foundation Model(FFM)平台:不绑定具身形态的自主软件,安装在合作伙伴或客户自有机器人硬件上。FFM 充当通用“机器人大脑”,让机器人在非结构化工业环境中感知、推理和行动,无需预先绘制地图、GPS、云连接或按任务重新编程。该平台可跨四足机器人(如 Boston Dynamics Spot)、轮式机器人、人形平台和乘用级自动驾驶车辆使用,硬件无关。 FieldAI 的应用组合覆盖其官网 Solutions 页面列出的八类方案:现场测绘和文档化;设施与设备巡检;状态异常检测;大规模数据采集;无人化物料运输;安保和威胁检测;远程呈现和远程操作;搜索与 ISR。公司既自研应用,也通过伙伴载荷模式与第三方提供商合作。短期商业重点是巡检和数据采集(当前已产生收入),操作能力和通用作业则位于中长期路线图上。 一个有辨识度的架构原则是 “Context over Training”:FFM 不要求穷尽式场景预训练,而是利用上下文线索和学习到的物理先验,对未见过的情况做推断。CEO Ali Agha 将机器人的风险感知描述为它知道自己对某个动作“有多大把握”,客户可配置风险阈值,从而在安全关键区域调整行为。结果是一种部署模型:机器人可在新场地快速启用,并在不做大量逐场地定制的情况下扩展到企业级机队。 [CE001, CE003, CE006, CE007, CE008, CE009]
| 模块 / 应用 | 用户 / 买方 | 状态 / 成熟度 | 差异化 | 尽调缺口 |
|---|---|---|---|---|
| Field Foundation Model(FFM)— 核心平台 | 工业企业、国防主承包商 | 生产 — 全球数百个站点 | 物理优先的信念空间架构;不需要地图 / GPS / 云;跨形态 | 没有独立基准;性能规格未披露 |
| Dynamics Foundation Model(DFM) | 与 FFM 相同 — 子模块 | 生产 — 与 FFM 打包 | 将认知世界模型与机器人本体解耦,支持跨平台部署 | 没有独立文档;打包在 FFM 中,无法单独测试 |
| Multiagent Foundation Model(MFM) | 协调多台机器人的机队运营商 | 生产 — 已部署多机器人机队 | 机队级协调;在站点尺度共享环境推理 | 未发布协调延迟规格或独立机队规模基准 |
| 站点测绘与文档化 | 建筑、采矿、设施管理者 | 生产 — Big-D Construction 等 | 夜间自主 3D 扫描;与 BIM / 数字孪生对齐 | 具名客户仅 Big-D;未发布地图精度规格 |
| 设施与设备巡检 | 能源、油气、公用事业、制造 | 生产 — 日本、美国、欧洲的能源和工业站点 | 无需预编程路线即可检测危险、发现异常 | 没有独立准确率或误报率数据 |
| 无人物料运输 | 建筑、物流、制造 | 早期商业化 — 路线图中期 2026–2028 | 同一 FFM 大脑延伸到操作任务;避开地图依赖 | 操作能力尚未在生产环境证明;没有客户案例 |
| 安防与威胁检测 | 安保公司、公共基础设施运营商 | 生产 — Certis Group Singapore 合作 | 接入 Certis Mozart 指挥平台;机队级巡逻 | Certis 合作于 2026 年 2 月签署;独立来源未确认实地部署 |
| 远程临场 / 远程操作 | 远程操作员、设施管理 | 可用 — 已列在 Solutions 页面 | 人在回路,作为自主运行的兜底 | 未发布面向用户的文档或性能数据 |
| 搜索与 ISR | 联邦、国防、应急响应 | 生产 — FieldAI Federal 垂直业务 | DARPA/NASA 血统;GPS 拒止地下导航履历 | 国防合同和认证未公开披露 |
状态来自 FieldAI 官方材料、合作伙伴公告和新闻;部署数量没有独立审计。「生产」表示公司确认的实地部署;「早期商业化」表示分析师 / 投资人覆盖中的路线图阶段。空值单元格表示缺少公开披露,不表示能力不存在。
[CE001, CE005, CE008, CE009, CE010, CE012]FFM 平台从传感器输入到应用输出的分层模型架构。
层级边界和名称根据官方产品页、NVIDIA 合作新闻稿和 FAIRI 论文推断。DFM 和 MFM 有官方描述;内部模型权重和推理硬件未披露。
[CE001, CE002, CE004, CE005, CE016, CE030]5.2 架构与技术设计
FFM 平台由三层模型组件组织。核心 FFM 使用视觉、LiDAR、文本、音频等多模态传感器流,负责感知、世界建模和决策。Dynamics Foundation Model(DFM)是子模型,整合机器人的内在运动学和动力学属性,使同一个认知层能够驱动机械结构不同的机器人身体。Multiagent Foundation Model(MFM)是协同层,允许机队中的多台机器人共享环境表征,并在场地尺度上集体推理——把机器人集群变成分布式自主系统,而不是一组孤立工具。 架构上,FieldAI 运行在“信念空间”:系统持续跟踪环境状态的概率分布,而不是假定信息完美。这一设计脉络可追溯到 CEO Ali Agha 关于 FIRM(Feedback-based Information RoadMap)信念空间规划的学术工作,也包括赢得 DARPA SubT 挑战赛 Phase II 的 NeBula 系统。概率式方法意味着机器人可在无 GPS、低能见度、传感器退化的条件下工作。当 FFM 遇到陌生条件时,会明确降速或选择更保守路径——这是公司承认的吞吐成本,但被视为安全特性。真实世界数据飞轮意味着每次部署都会随着时间改善模型性能,并形成复利式数据优势。 NVIDIA 集成加深了工程栈。FieldAI 使用 NVIDIA Omniverse NuRec,将机器人正常作业顺带采集的传感器数据重建为高保真 3D 数字孪生。这些数字孪生会加载到 NVIDIA Isaac Sim 和 Isaac Lab,用于机器人策略训练与验证;FieldAI 还通过 Microsoft Azure 采用 NVIDIA Physical AI Data Factory Blueprint,使用 NVIDIA Cosmos 开放世界基础模型和 NVIDIA OSMO 自动化合成数据流水线。公司的 GitHub 组织(field-ai)维护着多个机器人基础设施库的活跃 fork——Spot SDK、rosbridge_suite、spconv、unitree_sdk2——最近提交时间到 2026 年 4 月,为平台集成面提供了开发者侧证据;但公司未公开任何专有模型权重或 SDK 文档。私有 Squarespace 发布说明页面进一步限制了外部对已交付能力的审计。 [CE002, CE004, CE005, CE021, CE022, CE014]
| 层 / 组件 | 作用 | 关键依赖 | 风险 |
|---|---|---|---|
| 多模态感知栈 | 将视觉、LiDAR、文本、音频、热成像、事件相机输入统一的信念状态表征 | 合作伙伴机器人上的传感器硬件(Spot、Unitree 等) | 新平台上传感器退化或不支持某些模态,会降低感知质量 |
| 信念空间世界模型(FFM 核心) | 概率式环境表征;管理未知场景中的不确定性 | 运营部署产生的真实世界训练数据 | 分布偏移:训练语料未覆盖的新环境会让系统更保守,牺牲吞吐 |
| Dynamics Foundation Model(DFM) | 集成机器人运动学 / 动力学模型,用于跨平台执行 | 机器人 URDF / 动力学模型;平台专属校准 | 每种新机器人形态都需要适配 DFM;集成工作深度未公开 |
| Multiagent Foundation Model(MFM) | 机队级协调;共享环境推理 | 机器人间通信(边缘 mesh 或本地网络) | 密集工业环境中的通信故障或带宽约束 |
| 边缘计算运行时 | FFM 推理完全在机器人端运行,不需要云连接 | 设备端加速硬件(推测为 NVIDIA Jetson 或同类) | 硬件采购与兼容性没有公开文档;功耗 / 散热限制会约束机器人平台选择 |
| NVIDIA Omniverse NuRec 管线 | 将运营传感器数据转换为高保真 3D 数字孪生,供客户和 sim-to-real 使用 | NVIDIA Omniverse 授权;Data Factory Blueprint 的 Azure 工具链 | 依赖 NVIDIA 商业栈;存在 Omniverse 平台连续性和授权风险 |
| Isaac Sim / Isaac Lab 训练流水线 | 在可用于仿真的重建环境中训练并验证策略 | 重建数字孪生;NVIDIA Isaac 计算基础设施 | 仿真到现实存在落差:机器人遇到仿真未覆盖的陌生现实特征时,行为会趋于保守 |
| 合成数据流水线(Cosmos + OSMO) | 用合成训练场景补充真实部署数据 | NVIDIA Cosmos 开放世界模型;NVIDIA OSMO;Microsoft Azure | NVIDIA、Microsoft、FieldAI 三方依赖抬高集成复杂度,也加深供应商锁定 |
架构层来自 FieldAI 官方产品页、NVIDIA 合作新闻稿,以及 FieldAI 的 NVIDIA Omniverse 博客。边缘计算硬件细节未公开;NVIDIA Jetson 是基于行业惯例和 NVIDIA 合作背景的推断。依赖项只反映已记录的集成点;未披露组件可能仍然存在。
[CE002, CE003, CE004, CE005, CE014, CE015]FFM 平台如何在客户站点启用,并持续交付运营价值。
[CE003, CE007, CE015, CE018, CE024]5.3 部署流程与使用场景
FieldAI 的部署模型是 B2B 软件:客户自有或租赁的机器人通过软件安装获得 FFM 大脑,FieldAI 提供集成服务和持续模型更新。客户不需要向 FieldAI 购买专用硬件。部署时间是关键差异化主张:机器人不需要预先绘制环境或 GPS 基础设施,因此场地启用被描述为很快;公司及其投资人称,过去“要几个月”的流程现在“只要几个小时”。一位未具名全球工业制造高管给出了 3.5 个月缩短到 12 小时的具体结果(超过 200 倍改善),使用的是 FieldAI 技术加 NVIDIA NuRec 数字孪生生成——但该主张未获独立第三方验证。 建筑是领先的商业垂直。Big-D Construction 使用 FieldAI 已超过两年,并在 2026 年 4 月宣布扩展到多个项目;据称其中一个项目通过早期缺陷检测节省了 $1.2 million。在安保垂直,Certis Group(新加坡)在公共基础设施、交通枢纽和工业设施中部署与其专有 Mozart 编排平台集成的 FFM。FieldAI 与 Boston Dynamics 的合作于 2026 年 3 月 12 日正式化,目标是建筑巡检;搭载 FFM 的 Spot 机器人可自主执行 3D 扫描、危险识别和夜间监控。公司还通过 FieldAI Federal 处理联邦和国防用例,该实体由前 DARPA Program Manager Dr. Eric Krotkov 领导。 地域上,FieldAI 已记录美国、日本、欧洲,以及通过 Certis 覆盖亚太的部署。客户通常不公开具名(TechCrunch 确认,公司在 2025 年 8 月融资公告时拒绝披露客户名称);截至 2026 年 5 月,Big-D Construction 是唯一按名称公开确认的建筑行业客户。 [CE011, CE012, CE018, CE023, CE024, CE032]
| 用户任务 | 当前 / FieldAI 前工作流 | FieldAI 方案 | 可衡量收益(声称) | 已记录限制 |
|---|---|---|---|---|
| 建筑进度监测 | 人工巡场、无人机航拍、3D 扫描项目(数周) | Spot 夜间自主巡逻 → 每日 3D 进度报告,并与 BIM 集成 | 巡检 / 文档时间减少 90%+(Boston Dynamics 新闻稿) | 说法未经独立第三方验证;来源仅单一供应商 |
| 工业设备巡检 | 受训技术员在危险区域人工巡检 | 机器人在设施内自主导航,无需预绘地图即可采集设备读数 | 降低人员暴露在危险环境中的时间;提前发现异常 | 未发布误报率或覆盖率数据 |
| 数字孪生生成 | 人工 3D 扫描项目 — 数月完成,很快过时 | 机器人每天生成重建数据作为运营副产物,再经 NuRec → Isaac Sim | 某未具名客户从 3.5 个月 → 12 小时(提速 >200×,公司说法) | 单一未具名客户;未独立验证 |
| 安保巡逻与事件响应 | 人工安保巡逻,覆盖一致性有限 | 自主机器人巡逻 → 实时事件检测 → 通过 Mozart 协同人机响应 | 运营效率和韧性提升(Certis 说法) | Certis 部署规模未披露;缺少独立评估 |
| 联邦 / ISR 行动 | GPS 拒止或危险区域的有人侦察 | FFM 驱动机器人在未测绘环境中导航,用于地下和野外 ISR | DARPA SubT 获胜验证了 GPS 拒止条件下的方法 | 国防合同细节保密;没有商业等效性能基准 |
| 多机器人站点协同 | 机器人顺序执行或人工调度,共享感知有限 | MFM 让机器人机队共享环境模型并集体推理 | 声称具备真正的分布式系统行为;全球最大四足机器人机队之一(计划) | 机队规模 MFM 部署细节稀少;没有独立机队性能数据 |
收益除非标注为独立验证,否则均来自公司声称或合作伙伴新闻稿报道。「可衡量收益」单元格反映的是声称;投资者和从业者应要求生产数据日志和第三方审计来验证。
[CE011, CE015, CE018, CE012, CE013, CE032]FieldAI 产品运行和迭代依赖的关键外部伙伴、平台与数据。
[CE010, CE013, CE014, CE015, CE016, CE017]5.4 信任、安全与合规
FieldAI 的安全架构基于概率式风险管理,而不是显式规则:FFM 在信念空间中运行,持续维护对环境的置信度估计;当不确定性超过客户配置的阈值时,系统会采取更保守动作。CEO Agha 将其描述为机器人知道自己“有多大把握”,并通过减速或改道避免不安全动作。该设计回应了传统 LLM 套进机器人时的幻觉风险,因为 FieldAI 的模型从一开始就为物理安全设计,而不是事后改造语言或视觉模型。 但截至 2026 年 5 月,公司未公开记录任何安全认证、监管批准或合规标准(如 ISO 10218、IEC 62443、ANSI/RIA、FedRAMP,或工业机器人同等标准)。Certis 合作协议提到,双方将“共同开展培训、安全验证和运营框架,以支持负责任部署”,这意味着正式安全验证框架仍在开发,而非已经完成。Boston Dynamics 合作同样聚焦实际部署成果,而不是认证。FieldAI 通过 FieldAI Federal 进入国防 / 联邦场景,暗示存在某种政府安全审查,但公司未发布具体 clearance 或 FedRAMP 状态。私有发布说明页面阻止外部审计已交付能力与路线图能力之间的差异,也未识别到正式漏洞赏金或安全披露政策。 [CE006, CE026, CE031, CE033, CE034]
| 控制项 / 认证 / 质量指标 | 状态(截至 2026 年 5 月) | 范围 | 缺口 / 尽调问题 |
|---|---|---|---|
| 安全风险阈值(客户可配置) | 已落地——用于生产部署 | 所有由 FFM 驱动的机器人;阈值可按部署配置 | 默认阈值、边界条件或覆盖策略均未见文档 |
| 保守回退行为(不确定时减速) | 已落地——官方博客已承认 | FFM 核心——模型遇到陌生场景时触发 | 未量化对吞吐的影响;触发回退的条件未公布 |
| ISO 10218 / ISO TS 15066(工业机器人安全) | 未公开披露 | 工业和协作机器人部署 | 要求提供合规文档或路线图;缺失会实质影响企业采购 |
| IEC 62443(工业控制系统安全) | 未公开披露 | 安防和能源垂直部署 | 未披露机器人到车队通信的网络安全认证 |
| FedRAMP / ITAR / DoD 安全许可 | 未公开披露 | FieldAI Federal(国防 / ISR 垂直) | 国防合同和许可级别未披露;考虑到 DARPA 资助背景,可假设已经过一定政府审查 |
| 安全验证框架(Certis 合作) | 开发中——合作方已承诺共建框架 | 与 Certis Mozart 平台结合的安防部署 | 框架未发布;合作公告未给出时间表 |
| 发布说明 / 更新日志 | 访问受限——私有 Squarespace 站点 | 所有产品版本 | 外部方无法审计已发布功能与路线图承诺是否一致;需申请访问权限或更新日志摘要 |
| 漏洞赏金 / 安全披露政策 | 公开来源未发现 | 产品安全 | 未发现公开的协调漏洞披露计划;尽调应索取内部政策 |
未确认事项的状态依据公开披露缺失判断。「未公开披露」不等于不合规;企业应直接索取认证文件。风险感知安全设计在架构层面已有充分记录,但公开记录中缺少正式第三方认证证据。
[CE026, CE031, CE027, CE034]5.5 路线图、成熟度与开发者生态
FieldAI 在巡检和数据采集上的产品成熟度最高:公司已有付费客户,并在数百个场地记录了生产部署。分析师报道和投资人材料中提出的三阶段路线图是:(1)近期至 2026 年——自主巡检、监控和数据采集,已商业化;(2)中期 2026–2028 年——干预和操作,从感知扩展到物理动作;(3)长期 2028+——跨所有平台的通用自主,包括人形机器人。这一路线图与 FieldAI 对 $405M 融资用途的表述一致:资金将用于机动能力和操作能力的开发。 CMU Robotics Innovation Center 入驻安排(2026 年 2 月宣布,Hazelwood Green 2,500 sq ft)正式化了公司的学术研究合作,让 FieldAI 可使用 1.5 英亩户外测试场和专门机器人实验室。FieldAI 赞助 CMU 的 AirLab 和 LeCAR lab 研究,并支持 VectorRobotics 学生团队的人形 loco-manipulation,显示公司正在投入中长期路线图上的能力缺口。Field AI Research Institute(FAIRI)发表同行评审研究;NeBula DARPA SubT 论文,以及 2024 年关于风险感知 AI 和 MPPI 控制的论文,为公司提供学术可信度。 开发者生态入口有限。FieldAI 的 GitHub 组织(field-ai)托管 10+ 个标准机器人库 fork,但没有发布任何专有 FFM 代码、模型权重或 SDK。未发现公开开发者门户、API 文档或社区论坛(Discord、Slack、Stack Overflow tag)。发布说明页面是私有 Squarespace 站点。这把开发者信号面限制在可观察的 GitHub 仓库活动,而非真正的从业者社区。与 NVIDIA 和 CMU 的合作为工程可信度提供了间接代理,但从公开来源还无法独立技术评估 FFM 能力。 [CE019, CE025, CE028, CE029, CE036, CE035]
| 日期 / 阶段 | 功能 / 里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2023 | 公司成立;FFM 概念来自 DARPA SubT / NASA 系谱并完成验证 | 已完成 | IP 和技术血统的基础 | Intel Capital 官方公告 |
| Aug 2025 | 完成 $405M 融资;FFM 已部署到全球数百个站点;资金支持运动和操作研发 | 已完成 | 资本验证产品市场匹配;操作能力是下一块主要扩张区 | TechCrunch、Intel Capital |
| Feb 2026 | 入驻 CMU Robotics Innovation Center;扩展 Pittsburgh 研发和测试基础设施 | 已完成 | 借助 CMU 户外测试场,加速操作和人形能力研发 | CMU 官方新闻稿 |
| Feb 2026 | Certis Group 战略合作——大规模自主安防运营 | 已签署;部署推进中 | 验证 FFM 可用于安防垂直;Certis 提供车队编排基础设施 | Certis 与 FieldAI 官方新闻稿 |
| Mar 2026 | Boston Dynamics 合作——在 Spot 上部署 FFM,用于施工巡检 | 已签署;车队扩张中 | 计划建设最大的第三方四足机器人车队之一;施工是近期收入引擎 | Boston Dynamics 官方公告 |
| Mar 2026 | 深化 NVIDIA Omniverse 合作——NuRec 数字孪生、Isaac Sim、Data Factory Blueprint | 集成进行中 | 缩小仿真到现实落差;可生成工业环境用于训练;NVIDIA GTC 2026 演示 | FieldAI NVIDIA Omniverse 博客、PR Newswire |
| Apr 2026 | Big-D Construction 宣布多项目扩张 | 生产中——继续扩张 | 具名客户证明多站点企业采用;需求来自组织各层级 | FieldAI 博客(官方) |
| 2026–2028(路线图) | 中期:介入和操作——开阀、清障、运输物料 | 路线图——未获资金保障的里程碑;研发推进中 | 显著扩大 TAM;关键技术风险是非结构化环境中的操作能力 | xmaquina.io 分析师分析;Intel Capital 投资人表述 |
| 2028+(路线图) | 长期:通用自主——跨行业多任务人形机器人 | 路线图——愿景型 | 需要在移动操作和长周期任务规划上突破;CMU 合作瞄准这一点 | xmaquina.io 分析师分析 |
Q1 2026 之后的路线图阶段基于分析师和投资人摘要,不是 FieldAI 官方产品公告。时间线是估计;操作和通用能力的实际可用性取决于技术进展和部署规模。已完成里程碑来自官方或合作方证明材料。
[CE010, CE012, CE014, CE019, CE024, CE025]FieldAI 核心能力在机器人形态、应用领域和部署就绪度上的相对成熟度。
成熟度评级根据官方公告、伙伴新闻稿、GitHub 活动和分析师报道推断。「高」= 已确认生产部署。「中」= 已可用或公司声称具备,但第三方证据有限。「低」= 路线图或研发阶段。任何成熟度评级都尚未获得独立验证并公开发布。
[CE008, CE009, CE020, CE029]5.6 附录
06客户
6.1 客户分层、买方 / 用户 / 付款方逻辑
FieldAI 向企业级工业组织销售具身 AI 软件和传感器-计算载荷,覆盖六个已具名垂直:建筑、能源 / 油气、矿业、制造、城市配送与巡检,以及联邦 / 国防。公司不制造机器人;产品是自主软件和载荷,用来改造买方已经拥有的硬件,或买方从 Boston Dynamics 等 OEM 伙伴采购的硬件。这种干净的软件许可模型意味着付款方始终是工业企业,买方通常是技术负责人、运营 VP 或创新总监,日常用户则是现场主管、安全经理或运营负责人,由他们部署和监控机器人。 据报道,订阅价格随部署规模不同,从数万美元到每年 $500,000 不等,首次部署还会叠加硬件集成服务。建筑垂直是 FieldAI 商业成熟度最高的板块——截至 2026 年中,它拥有唯一公开具名的客户案例(Big-D Construction、DPR Construction),也拥有唯一确认的多年生产部署。能源 / 油气和工业制造由专门产品页覆盖,内容包括设备异常检测、仪表和热成像监控、数字孪生创建,但公司未公开披露这些板块中的具名企业客户。安保运营通过 2026 年 2 月宣布的 Certis Group 战略合作进入可服务市场;联邦 / 国防是单独品牌化的板块(FieldAI Federal),前文已提到 Detroit Defense 合作。 买方 / 用户分离带来重要销售动态:采购决策中的 champion 往往是技术或创新总监,但生产采用取决于每天使用系统的现场主管和安全经理是否买账。在 Big-D,现场主管的自发需求强化了自上而下的投资决策——这是黏性的正面信号——但也意味着单个项目团队的抵触可能拖慢场地级扩张。 [CU001, CU002, CU003, CU004, CU005, CU006]
| 细分 | 买方画像 | 用户画像 | 付款逻辑 | 具名客户证明 | 尽调缺口 |
|---|---|---|---|---|---|
| 施工 | 技术 / 创新总监,VDC 负责人 | 现场总监,安全经理 | 企业承包商;项目级预算 | Big-D、DPR(具名);2 家未具名 ENR 公司 | 无客户数量;无单位经济性 |
| 能源 / 油气 | 运营副总裁,设施总监 | 巡检工程师,操作员 | 能源企业;维护预算 | 未公开具名 | 零个具名证明;无结果数据 |
| 采矿 | 矿山经理,自动化负责人 | 地下 / 露天矿操作员 | 矿业企业;运营支出 / 安全预算 | 未公开具名 | 零个具名证明;只有 BHP Ventures 投资人信号 |
| 制造 | 运营副总裁,工厂总监 | 工厂技术员,质量工程师 | 制造企业;运营预算 | 未具名「全球制造商」(匿名引述) | 公司名称未披露;结果无法验证 |
| 安防 / 设施 | 安防总监,运营负责人 | 安防巡逻员 | Certis Group 作为合作运营方;终端客户未具名 | Certis Group(合作运营方) | Certis 是渠道,不是直接销售;终端用户未具名 |
| 联邦 / 国防 | 项目经理,合同官 | 现场操作员,分析师 | 政府机构;国防预算 | Detroit Defense(合作方,按前文不是客户) | 未公开确认已授予的政府合同 |
具名客户证明列反映截至 2026 年 5 月已公开披露、并可独立佐证的部署。来源包括 FieldAI 官方页面、robotsinconstruction.com 追踪器(2026 年 4 月)、合作方公告。证明列为空表示该细分中没有任何独立公开确认的具名生产部署。
[CU001, CU006, CU007, CU008, CU009]展示 FieldAI 主要和次要客群中,典型企业工业客户从认知、生产部署到机队扩张的路径。
阶段顺序根据可获得客户证据(Big-D 2 年轨迹、DPR 从 PoC 到生产的路径)和公司材料推断。并非所有细分市场都有已确认生产部署;能源、采矿和联邦仅按公司定位纳入。
[CU002, CU003, CU004, CU015, CU016]6.2 采用轨迹与部署规模
FieldAI 2025 年 8 月融资公告称,公司已在日本、欧洲和美国“数百个复杂真实世界工业环境”中完成“成功测试和部署”,客户包括建筑、能源、制造、城市配送与巡检领域的“全球最大公司中的一些”。该轮融资被描述为在“快速客户采用和多份扩张合同”之后超额认购;到 2026 年 3 月,公司在 NVIDIA 合作公告中提到一个“快速增长的机队”,客户正在北美、欧洲和亚洲扩张。这些都是公司方说法,没有第三方验证活跃场地数量、客户数量或机队规模。 robotsinconstruction.com 截至 2026 年 4 月的独立跟踪只统计到四个确认的建筑部署:Big-D Construction(生产,2+ 年)、DPR Construction(生产,截至 2025 年 11 月约 1.5 年),以及两家未具名 top-10 ENR 总承包商。该平台指出 FieldAI 自 2024 年开始出货。“数百个环境”与四个确认建筑部署之间的差距,可能部分来自非建筑垂直、商业化前测试环境,以及未被独立来源跟踪的演示或评估部署。Big-D 和 DPR 是仅有的、带有客户具名结果的案例(具名高管和现场主管引述、具体平方英尺数据、已运行部署时长)。 Boston Dynamics 合作(2026 年 3 月)增加了一个可信的第三方验证者;Boston Dynamics 提到客户“从亚洲到欧洲再到北美”正在扩展到企业级部署。FieldAI 表示计划把 Spot 部署扩展成“全球最大的第三方四足机器人机队之一”。两个说法都没有独立场地级佐证。 [CU010, CU011, CU012, CU013, CU014, CU015]
| 指标 | 数值 / 区间 | 日期 | 来源类型 | 置信度 | 含义 / 缺失分母 |
|---|---|---|---|---|---|
| 总部署环境数(公司声称) | 数百个 | 2025-08-20 | 公司声称 | 低 | 无法验证;包括测试、试点和线上运营 |
| 已确认具名施工部署(独立) | 4 | 2026-04 | 独立追踪器 | 高 | 仅施工;其他垂直未追踪 |
| 地理覆盖 | 北美、欧洲、日本(亚洲) | 2025-08-20 | 公司声称 | 中 | 未披露各地理区域的站点级数量 |
| Big-D Construction 部署时长 | >2 年活跃 | 2026-04 | 客户引述,独立媒体 | 高 | 多个工地;无车队单位数量 |
| DPR Construction 部署时长 | 截至 2025 年 11 月约 1.5 年 | 2025-11 | 客户引述,独立媒体 | 高 | 仅确认单个数据中心站点;更广泛铺开未确认 |
来源置信度反映佐证层级:低 = 仅公司;中 = 由合作方或投资人确认;高 = 由具名客户或独立分析师确认。所有案例都缺分母:未披露机器车队数量、活跃站点数或客户账户总数。
[CU010, CU011, CU014, CU015, CU016]| 证据项 | 主张类型 | 日期 | 独立性 | 置信度 | 主要限制 |
|---|---|---|---|---|---|
| Big-D Construction——Shaun Orr 高管引述(「每个项目」) | 客户引述(公司赞助) | 2026-04 | 公司赞助案例研究 | 高 | 存在正向选择偏差;没有反方或已流失客户引述 |
| DPR Construction——Justin Schreiner 现场总监引述(「让我们变得更好」) | 客户引述(独立媒体) | 2025-11 | 第三方媒体(Interesting Engineering) | 高 | 单一站点、单一项目类型;多站点泛化未确认 |
| 未具名制造商——「快 200 倍」指标 | 公司声称(归因 CEO) | 2026-03 | 公司新闻稿 | 低 | 客户未具名;指标未独立验证;方法未披露 |
| Boston Dynamics 引述的「检查时间减少 >90%」 | 合作方引用 | 2026-03 | 合作方新闻稿(Boston Dynamics) | 中 | 基线方法不清晰;合作伙伴有动机证实正向结果 |
| “数百个工业环境”的机队说法 | 公司自称 | 2025-08 | 公司新闻稿 / 融资公告 | 低 | 无法核验的汇总口径;包含测试、试点和上线部署;缺少分母 |
| 超额认购融资 + 多个扩张合同 | 财务信号(投资人佐证) | 2025-08 | 融资公告(公司 + 投资人引述) | 中 | 未量化客户数、ARR 或哪些客户扩张 |
独立性列反映证据可追溯到的最接近独立交易关系。公司赞助 = 来自 FieldAI 自身新闻材料,且客户直接参与。第三方媒体 = 客户引述发表于编辑语境,FieldAI 未发起询问。没有外部佐证的公司自称事项,置信度下调。
[CU019, CU023, CU028, CU030, CU010, CU012]用可获得的量化锚点,呈现从市场认知到生产级机队部署的转化漏斗。
漏斗顶部数值是市场规模估算,不是 FieldAI 披露的销售管线指标。阶段边界由分析师定义。独立追踪数据中的 4 个部署只覆盖建筑行业;能源、采矿和制造业的生产部署数量未披露。
[CU010, CU014, CU019, CU023]6.3 具名客户证据
Big-D Construction 和 DPR Construction 是 FieldAI 两个完全具名、已生产部署且公开归因结果的客户。Big-D 是一家拥有 25+ 年商业建筑经验的美国承包商,截至 2026 年 4 月已活跃部署超过两年。C 级高管 Shaun Orr 描述了一个突破性时刻:机器人实时走遍项目现场,联接 BIM 模型和项目进度,识别漏做工作和安全问题,并把发现回连到公司的项目管理平台——这是单个人类团队无法独立复制的能力。Orr 预计“每个项目都会在某种程度上使用 FieldAI 工具”。Big-D 现场主管 Bronson Dupaix 描述了用机器人替代五小时人工现场巡检、手持相机和扫描仪的流程。Big-D 总监 Chantelle Menlove 还独立提到,一位在某个工地见过机器人的现场主管打电话要求在下一个工地使用它,说明存在自下而上的自然需求。 DPR Construction 是美国十大承包商之一,专长于数据中心和医疗项目;截至 2025 年 11 月,它在 California Santa Clara 的一个数据中心现场部署 FieldAI 机器人约 1.5 年。机器人采集了 45,000+ 张照片,行走 100+ 英里,扫描 500,000 平方英尺室内空间和 125,000 平方英尺屋面。现场主管 Justin Schreiner 表示,该系统通过去掉琐碎文档任务、让团队聚焦关键工作,“让我们把事情做得更好”。 Certis Group 是新加坡最大的安保和运营公司,2026 年 2 月宣布与 FieldAI 建立战略合作,在其全球运营中部署自主安保机器人。Certis 的 Mozart 编排平台与 FieldAI 的 FFM 集成,用于多场地安保环境中的自主巡逻、事件检测和人机协同。Certis 总裁 Ng Tian Beng 确认,该合作面向“实时、任务关键环境”。这不是直接客户销售,而是渠道 / OEM 式安排:Certis 既是伙伴,也是安保垂直的运营方客户。 一位未具名全球工业制造高管在 2026 年 3 月 FieldAI/NVIDIA 新闻稿中被引用称,某个过去“需要三个半月”的流程现在“只需十二小时”——提升 200 倍。客户未具名,指标也无法验证,但这段话由公司具名高管(CEO Ali Agha)在正式新闻稿中归因。robotsinconstruction.com 独立跟踪器中还出现了两家额外的未具名 top-10 ENR 公司,确认有生产部署,但未披露结果。 [CU019, CU020, CU021, CU022, CU023, CU024]
| 客户 / 合作方 | 细分 | 部署 / 用例 | 生产 / 试点 | 结果证据 | 限制 / 缺口 |
|---|---|---|---|---|---|
| Big-D Construction | 施工 | 工地巡检、进度记录、BIM 对比、安全监控 | 生产(2 年以上,多站点) | 高管引述:「每个项目都会配备 FieldAI 工具」;5 小时人工巡场被省掉;现场总监自下而上提出需求 | 结果仍是定性;未披露 ROI、正常运行率或返工减少百分比 |
| DPR Construction | 施工 | 进度记录、室内扫描、屋面勘察、危险检测、安防 | 生产(约 1.5 年,Santa Clara 数据中心) | 45K+ 张照片、行走 100+ mi、扫描 500K sqft;现场总监称「让我们变得更好」 | 公开只确认单站点;多站点扩张未确认;无生产率美元指标 |
| Certis Group | 安防 / 设施 | 自主巡逻、事件检测、人机协同 | 合作 / 早期部署(2026 年 2 月宣布) | 已签署战略合作;已确认 Mozart 编排集成;线上运营规模未确认 | 合作运营方模式;终端客户名称未披露;未分享部署 KPI |
| 未具名 ENR Top-10 公司 A | 施工 | 未说明的巡检、测绘、监控 | 生产(据独立追踪器,2025) | robotsinconstruction.com 确认部署(2026 年 4 月) | 客户身份因 NDA 未披露;无结果数据 |
| 未具名 ENR Top-10 公司 B | 施工 | 未说明的巡检、测绘、监控 | 生产(据独立追踪器) | robotsinconstruction.com 确认部署(2026 年 4 月) | 客户身份因 NDA 未披露;无结果数据 |
| 未具名全球工业制造商 | 制造 | 创建数字孪生、设施运营 | 未知(状态未确认) | 高管引述:流程从 3.5 个月缩短到 12 小时(提升 200×) | 客户身份未披露;结果指标未独立验证;部署状态未知 |
生产 / 试点分类基于部署时长、高管措辞和独立追踪器状态。「试点」指单次演示或探索性概念验证,没有确认的持续运营。结果证据列只包含公开归属的引述和独立确认的指标。200× 指标来自公司新闻稿,尚未独立验证。
[CU019, CU020, CU021, CU023, CU024, CU025]从结果证据质量(低到高)和生产部署成熟度(试点到规模化生产)两个维度,展示 FieldAI 已知客户和伙伴关系。
结果证据质量由分析师按证据独立性、具体程度和归因方式评级。成熟度位置反映公开披露的部署时长和规模,不代表公司内部数据。该矩阵只是局部视图——FieldAI 完整客户群未公开。
[CU019, CU023, CU026, CU028, CU029, CU030]6.4 留存、扩张与集中度风险
FieldAI 未披露任何留存指标——截至 2026 年 5 月,任何公开声明、新闻稿或投资人沟通中都没有出现 NRR、GRR、队列数据、流失率、合同续约率或客户满意度评分。公司仍处于未披露收入阶段且为私营公司;没有监管文件强制这些披露。 建筑板块的定性留存信号略偏正面。Big-D 的多年活跃部署,以及将 FieldAI 扩展到未来每个项目的明确计划,是目前最强的留存和扩张信号。DPR 的企业风投部门 WND Ventures 称 2025 年 8 月融资是“人们可以期待机器人领域出现更大融资轮次的信号”,并提到 DPR 工地帮助测试概念验证——这种措辞暗示关系仍在持续。超额认购融资轮明确归因于“多份扩张合同”,也从财务信号上佐证了复购,尽管具体合同价值和客户名称仍未披露。 结构性集中度风险较高。基于公开确认的生产部署,FieldAI 的建筑客户基础由四个账户构成:Big-D、DPR 和两家未具名 ENR 公司。任何一个具名锚定客户流失,都会拿走超过 50% 的可见证据基础。能源、矿业和制造垂直没有公开具名客户,意味着建筑之外的生产证据仍是缺口。Certis 带来了向安保的垂直多元化,但它是伙伴运营方,不是直接企业 SaaS 账户。 主力垂直还存在更广泛的宏观逆风:BuiltWorlds 2025 Equipment & Robotics Benchmarking report 发现,承包商的活跃建筑机器人使用率同比从 65% 降至 46%,即便正面情绪升至 95%。下降被归因于行业从实验性试点转向选择性、已验证的落地——这一动态可能拖慢 FieldAI 新客户获取,同时利好其现有生产账户。企业销售周期长、多方审批流程、NDA 限制客户具名,意味着 FieldAI 真实客户基础几乎肯定大于公开证据所示,但没有独立方法验证。 [CU031, CU032, CU033, CU034, CU035, CU036]
| 指标 | 数值 / 状态 | 细分 | 置信度 | 尽调问题 |
|---|---|---|---|---|
| NRR(净收入留存率) | 未披露 | 全部 | Unknown | 在下一次投资人尽调节点索取 NRR 或扩张收入占比 |
| GRR(总收入留存率) | 未披露 | 全部 | Unknown | 对照任何契约或董事会层级指标,核验最低留存 |
| 客户流失率 | 未披露 | 全部 | Unknown | 确认是否有任何具名客户已结束部署合同 |
| 合同期限 | 未披露(暗示为年度或多年订阅) | 全部 | 低 | 确认典型合同期限和自动续约条款 |
| 扩张合同 | 多次提及(来自融资轮超额认购措辞) | 全部 | 低 | 为尽调获取具名扩张账户清单和合同金额 |
由于 FieldAI 未披露任何留存或财务表现数据,所有指标值均为空。公司为私营企业,没有披露这些数字的监管义务。置信度评级反映底层来源质量;「低」表示扩张合同的存在仅由公司发布的融资措辞确认。
[CU031, CU032, CU033]| 因素 | 证据 | 严重性 | 影响 | 尽调路径 |
|---|---|---|---|---|
| 数据飞轮形成的转换成本 | 每次部署都会丰富 FieldAI 模型;联邦学习随时间叠加站点智能 | 正向 / 缓释因素 | 随时间提高客户黏性和转换成本 | 核验客户合同中的数据可携带权;确认联邦学习架构 |
| 施工垂直集中度 | 所有具名生产客户都在施工;4 个已确认站点中 4 个都是施工 | 高风险 | 失去 Big-D 或 DPR 会实质削弱公开证据基础,并带来信号风险 | 索取按垂直拆分的客户数量;确认能源 / 采矿生产部署确实存在 |
| 具名客户 NDA 集中度 | 大多数客户未具名;Big-D 和 DPR 是仅有的两个具名生产账户 | 中风险 | 无法独立验证「数百个环境」说法,增加尽调不确定性 | 安排与 3 个以上具名能源 / 制造客户进行参考电话 |
| 施工机器人采用逆风 | BuiltWorlds:活跃施工机器人使用率从 65%(2024)降至 46%(2025) | 中风险 | 全行业降速可能拖慢核心垂直的新客户获取 | 跟踪 BuiltWorlds 2026 基准报告;监测 FieldAI 管线数据 |
严重性评级是基于截至 2026 年 5 月公开证据的分析师判断。「正向 / 缓释因素」表示该因素降低风险,而不是抬高风险。扩张驱动因素来自结构性产品优势;集中度风险反映当前客户基础构成。披露经核验的完整客户名单后,应重新评估二者。
[CU034, CU035, CU037, CU040]6.5 附录
07风险
7.1 监管与法律风险
FieldAI 在安全关键工业环境中部署会自主做出物理决策的 AI 系统。这使公司正好落入多个持续演进的 2026 年 AI 监管框架范围。根据 EU AI Act,用作工业场景安全组件的机器人 AI 系统被归为高风险,需要风险管理系统、技术文档、人类监督设计、CE marking 和上市后监测。Baker Botts 指出,Annex III 高风险系统原定 2026 年 8 月 2 日截止日期,可能因 European Commission Digital Omnibus 提案延后至协调标准发布后六个月(最终兜底日期为 2027 年 12 月);但延期取决于 2026 年 8 月前是否通过,因此公司必须按现有期限并行准备合规。Colorado AI Act(SB 24-205)将于 2026 年 6 月 30 日生效,是美国首部针对高风险 AI 系统的综合州法,要求影响评估、消费者披露,以及采取合理注意防止算法歧视。California SB 53 于 2026 年 1 月生效,要求大型前沿 AI 模型开发者(训练算力 >10²⁶ FLOPS)发布风险框架,并在 15 天内报告关键安全事件;FieldAI 的训练算力是否跨过该阈值尚未披露。联邦方向仍未稳定:Trump administration 2025 年 12 月行政令表达了预先排除州级 AI 法的意图,但今天尚未实际排除,合规负担仍呈碎片化。 产品责任方面,法院正在围绕 AI 部署形成产品责任框架。KL Gates(2026 年 3 月)记录了越来越多的诉状,将已部署 AI 系统视为“产品”,适用设计缺陷、未警示和可预见误用等理论。欧盟修订版产品责任指令将严格责任沿供应链延伸到上游 AI 组件提供商;即使部署客户或机器人 OEM 拥有终端产品关系,FieldAI 也会被纳入。Garcia v. Character Technologies 树立了一个先例:在起诉状阶段,面向消费者的 AI 可被视作严格责任下的产品。对工业机器人 AI 而言,物理伤害因果链(挤压、撞击、意外运动)会比消费者聊天机器人案件更直接。 未发现涉及 FieldAI 的公开执法行动、监管调查、诉讼或 IP 争议。国防子公司 FieldAI Federal 若其自主系统进入军民两用技术,将受 ITAR/EAR 出口管制约束;未发现公开披露的出口授权。IP 保护主要依靠商业秘密和论文发布速度,而不是已授权专利:USPTO 专利申请检索找到一件已提交专利申请(US 2025/0252306),但截至 2026 年 5 月没有已授权专利;若人员离职,核心 FFM 技术可能面临被模仿的风险。 [CR001, CR002, CR003, CR004, CR005, CR006]
| 规则 / 许可 / 框架 | 司法辖区 | 2026 年 5 月状态 | FieldAI 触发可能性 | 触发后严重性 | 缓释措施 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| 欧盟 AI 法案(高风险 AI — 附件 III 安全部件) | 欧盟 / 全球 | 义务自 2026 年 8 月起适用(若 Digital Omnibus 通过,最终后备期限延至 2027 年 12 月) | 高 — FieldAI 在欧盟市场的工业自主部署可能符合高风险定义 | 关键 — 市场准入被阻断,罚款最高达 €35M 或全球收入的 7% | 准备合规计划;技术文档;CE 标志流程 | 实质性 — 暂无公开证据显示已准备就绪;欧盟部署的确切范围未披露 | 索取欧盟专项合规路线图、CE 标志状态、公告机构接洽文件 |
| Colorado AI Act(SB 24-205) | 美国 — 科罗拉多州 | 2026 年 6 月 30 日生效;首部美国州级综合性高风险 AI 法规 | 中 — 取决于 FieldAI 系统是否服务位于科罗拉多州的企业客户 | 中 — 影响评估处罚、消费者披露义务 | 监测州法适用门槛;聘请科罗拉多州法律顾问 | 低-中 — 创业公司规模若低于某些门槛,可能豁免 | 确认科罗拉多州客户清单及适用门槛分析 |
| California SB 53(前沿 AI 透明法案) | 美国 — 加利福尼亚州 | 2026 年 1 月生效;要求发布风险框架 + 15 天内报告事件 | 中 — 取决于训练算力;10²⁶ FLOP 门槛大概率尚未达到 | 中 — 每次违规最高 $1M;吹哨人保护启动 | 模型扩展后,监测训练算力报告义务 | 低-中 — 可能尚未触发,但模型增长后可能跨过门槛 | 获取训练算力估计,并索取 SB 53 适用性的法律意见 |
| AI 产品责任(欧盟 PLD / 美国州侵权法) | 美国 / 欧盟 | 判例法到 2026 年仍在演变;欧盟 PLD 须在 2026 年 12 月前转化为成员国法律 | 高 — 工业机器人场景里的身体伤害因果关系直接且可预见 | 关键 — 设计缺陷、未警示、供应链上游严格责任 | 产品责任保险;安全文档;系统护栏;客户赔偿条款 | 高 — 未发现已发布的安全认证、DPA 或独立审计 | 审查客户合同中的赔偿范围;获取产品责任保险凭证;索取安全测试报告 |
| ITAR / EAR 出口管制(FieldAI Federal) | 美国联邦 | ITAR/EAR 适用于军民两用军事技术;FieldAI Federal 是运营中的子公司 | 中 — FieldAI Federal 的自主军用机器人可能触发出口管制分类 | 高 — 刑事处罚、出口许可证被撤销、政府合同流失 | 聘请 ITAR 法律顾问;完成产品分类;按需取得美国国务院 / 商务部授权 | 实质性 — 未发现公开 ITAR 授权文件 | 索取 FieldAI Federal 的 ITAR/EAR 分类意见及出口授权状态 |
覆盖范围不完整;确切欧盟部署地理范围和美国州级客户足迹未公开披露。可能性评估依据公开产品描述、合作公告和监管文本推断;未经公司或律师确认。
[CR001, CR002, CR003, CR004, CR005, CR006]7.2 运营、安全与安保风险
FieldAI 的核心价值主张——自主机器人在非结构化、安全关键环境中运行——同时也是最高的运营风险面。搭载 FFM 的机器人完全在端侧决策,不依赖 GPS、地图或云连接。新环境中的一次模型推理错误(例如训练数据未覆盖的碎片构型、预授权机器人路径上突然出现行人),就可能在建筑、矿业或能源作业等高后果场景中造成身体伤害、财产损失或任务失败。2026 International AI Safety Report 将“loss-of-control scenarios”和“inadequate real-world reliability testing”列为快速演进自主系统中最紧迫的两类风险,直接适用于 FieldAI 的产品域。2025 年一宗被广泛报道的 Unitree H1 机器人在人类工人附近故障事件,以及 SRES 在 CES 2026 安全分析中提到“当机器人与人类共享开放工作空间时,现有工业安全协议往往不足”,说明更广泛同业群体——并非 FieldAI 本身——已经出现公开安全失败,这会塑造客户尽调要求、采购标准和潜在诉讼预期。 截至 2026 年 5 月,未有公开报道称 FieldAI 驱动的机器人发生安全事故。但公司没有发布性能基准、安全验证测试报告或第三方安全认证(工业机器人 ISO 10218、工业网络安全 IEC 62443,或数据处理 SOC 2),企业客户和投资人无法独立验证剩余风险水平。NIOSH Center for Occupational Robotics Research 和 OSHA 都记录了工业环境中严重机器人相关工伤的基线;FieldAI 部署在活跃建筑区和采矿作业中,恰恰进入了这些事故最常出现的场景。公司自己的部署叙事——三大洲数百个场地——意味着随着部署规模扩大,遇到首个不利事件的概率也会上升。 网络安全方面:FieldAI 机器人会把机队数据上传到云分析平台,并与 NVIDIA Omniverse 和 Isaac 基础设施通信。未发现公开 DPA、安全白皮书、渗透测试报告或 SOC 2 认证。若攻击者能够欺骗设备端传感器数据,或向机队管理平面注入命令,可能造成物理伤害。工业 IoT 和运营技术(OT)安全标准(IEC 62443)尚未确认实施。如果安全事件早于正式认证发生,这些缺口会形成实质剩余暴露。 [CR013, CR014, CR015, CR016, CR017, CR018]
| 失效模式 | 可能性 | 严重性 | 缓释成熟度 | 剩余敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| AI 模型在新环境中推理出错并造成身体伤害 | 中 — 规模扩大后风险叠加;非结构化部署不可避免会遇到新环境 | 关键 — 工人受伤、财产损失、诉讼、客户撤出 | 低-中 — 端侧不确定性量化是设计功能;未发布独立安全审计 | 高 — 未发现 ISO 10218 或 IEC 61508 认证 | 未发布性能基准、安全验证报告或第三方安全审计 |
| 边缘机队或云分析平台遭受网络攻击 | 低-中 — 工业 OT 环境越来越常成为攻击目标;机队联网扩大攻击面 | 高 — 对抗性传感器欺骗可能造成身体伤害;客户现场数据外泄 | 低 — 未发现 SOC 2、IEC 62443 或已发布的渗透测试 | 高 — 暂无安全认证证据 | 未披露公开 DPA、安全白皮书或 OT 专项网络安全计划 |
| 危险环境中的端侧硬件故障(边缘计算、传感器) | 中 — 采矿、建筑、能源部署的恶劣条件会加速硬件退化 | 中 — 任务失败、数据丢失、机器人回收可能带来危险 | 中 — Boston Dynamics Spot 面向工业工作周期设计;FieldAI 的计算载荷为定制 | 中 — 极端环境下载荷 MTBF 的公开数据有限 | FieldAI 专属载荷的平均故障间隔时间(MTBF)数据未公开 |
| 监管违规阻断欧盟或美国州级市场准入 | 中 — 2026 年 8 月欧盟 AI 法案期限;2026 年 6 月科罗拉多 AI 法案;合规计划未确认 | 高 — 市场准入受扰、罚款、企业客户声誉受损 | 低 — 公开披露中未发现合规计划证据 | 高 — 未发布准备就绪文件 | 确认欧盟合规计划状态;索取内部风险评估文件 |
| 数据主权和跨境数据传输违规(客户运营数据) | 低-中 — FieldAI 从三大洲客户现场收集传感器 / 视频数据 | 中 — GDPR 执法、客户合同违约、声誉受损 | 未知 — 未发现 DPA 或跨境传输机制文件 | 中 — 未确认存在标准合同条款或 DPA | 索取 DPA 模板、跨境传输机制和数据保留政策 |
可能性和严重性为基于证据的序数评估;公开来源没有定量故障树分析。缓释成熟度仅按公开披露评估。
7.3 伙伴与技术依赖风险
FieldAI 的 GTM 策略和技术栈集中在少数高价值伙伴和供应商上。2026 年 3 月 Boston Dynamics 合作是最可见的商业依赖:公司面向建筑的自主故事建立在 Spot 四足硬件之上,Marc Raibert 公开背书并提到 DARPA SubT 历史,把品牌和市场可信度都绑定到这段关系上。Boston Dynamics 由 Hyundai Motor Group 持有,而 Hyundai 又在 2026 年 2 月单独投资了 FieldAI——如果 Hyundai 决定为 Boston Dynamics 硬件投资竞争性的专有自主软件,或在 Hyundai 层面战略重组中调整合作,就可能出现错位。合作条款(排他条款、收入分成、最低承诺、终止权)未公开披露,无法评估合同保护。 NVIDIA 是更深、更广的依赖。NVentures(NVIDIA 的企业风投部门)是财务投资人,FieldAI 使用 NVIDIA Omniverse NuRec 做 3D 重建,用 Isaac Sim / Isaac Lab 做仿真训练,用 NVIDIA Cosmos 生成合成数据,并用 OSMO 自动化流水线。这形成了多层战略和商业锁定。投资人关系降低了突发供应中断的概率,但 NVIDIA 自身优先级(hyperscaler 分配、数据中心 GPU 供应瓶颈、竞争动态)可能压缩 FieldAI 的算力访问,或推高训练成本。VerticalData 报告称,高带宽内存(HBM)瓶颈在 2026 年仍然存在,最新型号的数据中心 GPU 交期延长至 9–12 个月;没有多年云承诺的小型 AI 公司暴露最大。 在 NVIDIA 和 Boston Dynamics 之外,Certis Group 安保合作(2026 年 2 月)增加了第三个集成依赖。Certis 使用其 Mozart 平台进行机队指挥控制,FieldAI 的 FFM 必须与这套专有基础设施互通。伙伴多元化是 FieldAI 硬件无关叙事中的明确组成部分——公司声称 FFM 兼容四足、人形、轮式机器人和 AV——但截至 2026 年 5 月,已具名商业合作深度集中在两个伙伴上:Boston Dynamics 面向建筑,Certis 面向安保;收入和声誉暴露也随之集中。 [CR023, CR024, CR025, CR026, CR027, CR028]
| 依赖项 | 对手方 | 角色 | 集中度 | 失效情景 | 严重性 | 缓释措施 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| Boston Dynamics Spot 硬件平台 | Boston Dynamics 与 Hyundai Motor Group | 建筑垂直领域的主要商业硬件平台;主要参考客户案例研究 | 高 — 建筑市场进入案例以 Spot 为中心 | 合作终止、Hyundai 战略转向、Spot 平台停产,或 Hyundai-FieldAI 投资人冲突 | 关键 — 主要商业案例研究流失、收入集中、市场叙事受损 | 硬件无关的 FFM 设计;正在推进更多 OEM 合作;Hyundai 财务投资让部分激励对齐 | 实质性 — 未披露排他或最低承诺条款;投资人关系不保证商业连续性 |
| NVIDIA Omniverse / Isaac / OSMO 计算与仿真栈 | NVIDIA(通过 NVentures 战略投资) | 仿真训练、合成数据生成、数字孪生重建;深度依赖模型管线 | 高 — 训练和部署管线多个环节使用 NVIDIA 产品 | 涨价、供给向超大规模云厂商倾斜、API 弃用,或战略合作伙伴优先级变化 | 高 — 模型训练成本和管线可靠性绑定 NVIDIA 可得性与定价 | NVentures 投资人关系;NVIDIA Omniverse 合作加深;边缘推理设计成可在多元硬件上运行 | 中 — 投资人利益一致降低突然断供风险;GPU 供应紧张是全市场约束 |
| Certis Group 机队编排(Mozart 平台) | Certis Group(安全垂直合作伙伴) | 安全运营的机队指挥与控制集成;互操作依赖 | 中 — 安全垂直商业管线依赖 Mozart 兼容性 | Certis 平台弃用、战略变化,或与 FieldAI 直销发生竞争冲突 | 中 — 若 Certis 路线分化,会限制安全垂直的可部署性 | FFM 设计为中间件层;可接入替代编排系统 | 低-中 — 依赖深度低于 NVIDIA 或 Boston Dynamics |
| NVIDIA 和云 GPU 算力(模型训练) | NVIDIA、Azure 与超大规模云厂商 | FFM 模型改进的训练基础设施;从部署数据做联邦学习 | 中 — 集中在 NVIDIA GPU 生态;Azure 用于管线工具链 | GPU 持续短缺(HBM 瓶颈、TSMC 冲击)推高训练成本或延误模型更新 | 中 — 拖慢模型改进速度;提高资本强度 | $405M 融资提供资本缓冲,可锁定长期云承诺 | 中 — 结构性 GPU 供应约束属于全市场问题,并非 FieldAI 特有 |
| 机器人 OEM SDK 和载荷访问(多平台) | Boston Dynamics、Unitree、其他 OEM | 每个硬件平台接入 FFM 都需要 SDK 访问;FieldAI 维护定制 Spot SDK 分支 | 中 — 每个 OEM 集成都是定制工程;OEM SDK 变更会打断集成 | OEM SDK 不兼容、许可证撤销,或 OEM 自研竞争性自主软件 | 中-高 — OEM 纵向整合 AI 栈是长期竞争威胁 | 硬件无关设计思路;多个活跃 OEM 集成降低单一 OEM 锁定 | 中 — Boston Dynamics 的 OEM 所有者(Hyundai)长期有战略动机掌握自主层 |
合作伙伴条款(排他、收入分成、最低承诺、终止权)未公开披露。集中度评估仅基于公开合作公告和产品文档。
截至 2026 年 5 月,FieldAI 运营生态中的关键技术、商业和资本依赖。箭头表示依赖方向;线条更粗代表集中度风险更高。
[CR023, CR024, CR025, CR026, CR027, CR028]7.4 人员、人才与执行风险
FieldAI 创始团队的行业可信度很强——Ali Agha 领导过赢得 DARPA SubT 的 CoSTAR 团队,也曾在 NASA JPL 担任首席研究员,这正是投资人和客户信心的核心来源。但风险也集中在一个人身上:Agha 同时是公司的公开门面、首席技术意见领袖,也是最醒目的商业关系负责人。外界尚未看到与他公众分量相当的 COO、工程负责人或并行商业领导者。David Fan(CTO)和 Shayegan Omidshafiei(CSO/总裁)提供了一定管理纵深,但公众能见度远低于 Agha。Agha 若非自愿离职、长期无法履职,或卷入声誉事件,募资、企业客户信心和合作伙伴关系都可能同时受冲击。 创始人之外,FieldAI 还面临高速扩张带来的用人风险,而它所在的技术劳动力市场竞争最激烈。公司据报从 2023 年创立扩至 2025 年 8 月的 130 多人,折算年化增速很高,企业文化维护、工程协同和产品交付纪律都难以持续守住。FieldAI 要和 Google Intrinsic(现已并入 Google AI)、Physical Intelligence(融资超过 $700M)、Amazon Robotics、Tesla Optimus 抢机器人 AI 工程师;这些公司盯着同一小群能规模化做具身 AI 的博士和资深工程师。尚未发现融资后关键技术人员离职的公开证据,但没有公开证据并不等于团队稳定。 执行风险也体现在商业端。FieldAI 到 2025 年 8 月才走出隐身模式,可验证的企业销售能力还不到一年。公司尚未公布客户名称、带有验证结果的部署案例或留存指标。从 NASA JPL、DARPA 传承的研究团队,转型为要对收入负责的企业软件供应商,是组织上的关键拐点;许多深科技衍生公司都没能足够快地跨过去,最终支撑不了估值曲线。 [CR033, CR034, CR035, CR036, CR037, CR038]
| 角色 / 职能 | 依赖或缺口 | 可能性 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| Ali Agha — 创始人兼 CEO | 核心技术思想领袖、商业关系负责人、主要公开信誉信号;没有具名的同等地位联合创始人或 COO | 低-中 — 没有公开迹象显示离职风险;早期创始人留任激励通常存在 | 关键 — 融资受扰、企业客户信心崩塌、合作伙伴关系破裂 | 预计有稳健股权归属和投资人留任协议(未确认);CTO/CSO 岗位有领导层深度 | 确认 Ali Agha 董事会席位、股权归属悬崖期安排,以及前 5 大商业关系的继任计划 |
| CTO 与 CSO / 总裁(David Fan、Shayegan Omidshafiei) | 核心技术执行;FFM 架构负责人;Agha 的继任深度 | 低 — 无离职信号;学术和产业背景强 | 高 — 任一人离开都会拖慢模型开发,并可能影响投资人信心 | 资深技术招聘;通过代码和文档沉淀 IP;继任规划 | 确认股权归属时间表;获取组织架构图,展示高管层以下工程梯队深度 |
| 企业销售与客户成功(销售副总裁:Patrick Purwin) | 公司直到 2025 年 8 月仍处于隐身状态;A 轮阶段的企业销售能力不到一年 | 中 — 销售团队扩张是深科技分拆创业公司的已知难题;无销售指标完成数据 | 高 — 若试点不能按所需速度转为合同,将削弱收入和下一轮叙事 | FieldAI Federal 子公司提供互补的政府销售渠道;销售副总裁来自商业企业市场 | 索取管线指标、试点转生产转化率和平均销售周期 |
| 工程人才留存(机器人 AI 资深工程师) | 人才池竞争激烈;Physical Intelligence、Google Intrinsic、Amazon Robotics、Tesla Optimus 争抢同一批候选人 | 中 — 高估值深科技公司融资后常见人员流失,股权价值只有流动性事件时才能兑现 | 高 — 掌握模型架构知识的资深工程师流失后,恢复困难且耗时 | 按 $2B 估值发放进取股权激励;NASA JPL 背景带来强技术文化和使命感 | 索取工程团队自愿流失率(过去 12 个月);确认资深技术员工的归属悬崖期 |
| 客户成功与现场部署运营 | 从早期试点扩到数百个企业现场,需要公开资料未记录的运营基础设施 | 中 — 2 年深科技公司常见运营扩张滞后 | 中 — 大规模部署效果差会导致客户流失,并反向影响客户背书质量 | Boston Dynamics、Certis 等合作伙伴在各自垂直领域提供联合部署支持 | 索取客户部署成功指标、投产周期数据和支持工单趋势 |
可能性和严重性基于同阶段深科技分拆创业公司的行业类比及公开领导层信息。暂无内部 HR 数据。
7.5 财务、估值与投资论证失效风险
FieldAI 的 $2B Series A 估值,是同阶段机器人 AI 公司历史最高,也为后续资本事件抬高了门槛。唯一独立收入信号是 OCBJ 2026 年 4 月匿名消息源提到的「已预订收入超过 $100M」;按 $2B 估值折算,收入倍数约 20x,和企业 AI 软件最高倍数组相当,但一旦增长放缓、AI 板块情绪转向,或公开市场可比公司重估,就会暴露在倍数压缩下。GetLatka 2025 年 11 月估算的 $140M ARR 未经审计,也未获公司确认。毛利率、NRR、CAC、账上现金和烧钱速度完全未披露,传统 SaaS 或深科技指标无法支撑估值承销。 资本强度是隐藏风险:工业机器人部署需要现场集成、定制载荷安装和持续模型微调,都会带来交付成本,可能把毛利率显著压到纯软件水平以下。如果毛利率明显低于 60–70%,即便收入加速增长,$2B 估值也很难支撑。第二个风险是烧钱集中:公司 130 多名员工,领导层来自高成本学术和科技机构,月度烧钱可能达到 $10–20M;按 $405M 融资计算,现金跑道约 2–4 年。若市场或宏观条件在盈亏平衡前打断增长,公司可能被迫融资降价。 投资论证失效触发点,落在会动摇核心投资假设的证据交叉处:(1)FieldAI 驱动的机器人发生重大安全事故,导致客户退出、监管行动或诉讼;(2)EU AI Act 执法阻断欧盟市场准入;(3)Boston Dynamics 合作终止,或 Hyundai 战略转向,移走主要商业案例;(4)Ali Agha 离职或发生声誉事件;(5)估值重置或降价轮,显示市场无法维持 $2B 入场估值。任一触发点单独发生,都足以实质性改变风险调整后回报。 [CR041, CR042, CR043, CR044, CR045, CR046]
| 风险 | 可监测触发项 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 安全事故(AI 造成身体伤害) | 公开事故报告、OSHA 申报、新闻报道、客户披露 | 任何 FieldAI 部署中首次确认因 FFM 失效导致机器人造成伤害或财产损失 | 暂停新增部署扩张;开展根因分析;恢复前取得独立安全审计;监测客户撤出公告 |
| 欧盟 AI 法案执法行动 | 欧盟市场监管机构公告、FieldAI 新闻稿、新闻报道、客户采购因合规冻结 | 欧盟市场监管机构正式调查、不合规认定,或客户因 AI 法案合规冻结采购 | 立即尽调欧盟合规计划;评估市场准入风险;认证完成前考虑暂停欧盟部署 |
| Boston Dynamics 合作终止或 Hyundai 战略转向 | 公开合作公告、Hyundai Motor Group 战略披露、FieldAI OEM 合作伙伴更新 | 任一方公开声明终止合作,或 Hyundai 宣布为 Boston Dynamics 硬件推出自有自主栈 | 在没有 Spot 集成的情况下重新评估建筑垂直商业案例;评估 OEM 多元化时间表;重校收入预测 |
| Ali Agha 离职或声誉事件 | LinkedIn、媒体报道、公司公告、投资人沟通 | 确认离开 CEO / 创始人角色、长期病假,或公开声誉争议 | 紧急评估管理层继任计划;开展投资人沟通;考虑陈述与保证保险触发项 |
| 融资估值重置或估值下调轮 | 未来融资公告、二级市场数据(Nasdaq Private Market、PremierAlts)、投资人沟通 | 任何已公告融资轮估值显著低于 $2B,或二级市场价格显示较上一轮价格折价 30%+ | 重估终局估值上行空间;审查优先权结构和清算优先权;评估相对原始入场假设的稀释敞口 |
| 商业化停滞(收入增长低于阈值) | OCBJ、Sacra、GetLatka 收入报告、客户数披露、招聘放缓信号 | 公开或可靠第三方迹象显示 ARR 同比增长低于 80%,或 2027 年前已签约收入停滞在 $150M 以下 | 在下一次资本部署决定前,深挖销售生产率指标、客户留存和管线质量 |
阈值是分析师定义、用于提高尽调强度的触发点,不是合同或法律约束标准。鉴于 FieldAI 是私营公司且披露有限,监测主要依赖公开信息。
截至 2026 年 5 月,按四档影响(致命、高、中、低)和三档可能性(低、中、高)对 FieldAI 主要风险做序数定位。位置由分析师根据公开证据评估。
可能性和影响是分析师基于公开证据与行业类比做出的序数评估;并非来自定量故障树或 Monte Carlo 模型。
[CR004, CR006, CR016, CR019, CR040, CR041]FieldAI 的主要风险事件如何向收入、客户、利润率、融资和估值结果传导。箭头表示传播路径。
[CR005, CR009, CR016, CR023, CR035, CR040]7.6 附录
08估值
8.1 估值背景与融资历史
FieldAI 最近一次披露的估值是投后 $2.0B;公司在 2025 年 8 月宣布连续两轮共融资 $405M 时确立了这一估值。最近一轮融资 $314M,获得超额认购,由 Bezos Expeditions、Prysm Capital 和 Temasek 共同领投;前一轮约融资 $91M,估值约 $500M。因此,公司约 12 个月内把估值推高了 4 倍。GeekWire 和 Reuters 消息源把两轮描述为「Series A and A1」。部分预尽调资料称拆分为 $150M(种子轮)加 $255M(Series A),但这与 Reuters 报道的最新一轮 $314M 不一致,也无法从一手来源确认;本章采用兄弟章节中的标准口径。 PremierAlts(2026 年 5 月访问)显示,公司五轮合计融资 $506.1M,最近一轮为 2026 年 2 月完成的 $315M,估值同为 $2.0B——这可能反映公告后二级市场 SPV 活动,或一轮新的一级融资。SEC 的 Form D 文件确认,投资工具「HII Field AI Series I, II, and III, a Series of HII Field AI, LLC」在 2025 年 11 月和 2026 年 4 月提交备案(CIK 0002095306、0002095305、0002126354),符合特殊目的集合投资基金常见结构,用于聚合高知名度私人公司的二级交易和 LP 权益。加州主体「Field Ai, Inc.」名下没有直接提交 Form D,说明公司可能依赖豁免,或使用了不同法律实体结构。 二级市场交易活跃:FieldAI 股份在 Nasdaq Private Market 和 Notice.co 交易(截至 2026 年 5 月约 $34.32/股),UpsideList 也收录公司但上行信号有限,说明二级定价接近或位于 $2B 标题估值附近。PitchBook(2025 年 6 月归档)显示公司 47 名员工,交易类型为「Early Stage VC」,这反映 PitchBook 私营公司覆盖常见的数据滞后。公司自身披露的 LinkedIn 员工数 201–500,以及 2026 年 4 月已预订收入 >$100M,说明自 PitchBook 截图以来公司已显著增长。 [CV001, CV002, CV003, CV004, CV005, CV006]
| 维度 | 评估 | 依据 | 决策含义 |
|---|---|---|---|
| 建议 | 跟踪 | 投资主线强,但估值不透明和收入未验证形成抵消 | 跟踪商业证明点;无审计财务前,不按当前价格承诺投资 |
| 置信度 | 中 | 投资人阵容顶尖;收入区间跨度 28x;治理未披露 | 若有 2 年以上经审计 ARR 数据并披露 NRR,置信度可升至高 |
| 风险评级 | 高 | 收入未验证;若行业倍数压缩或收入不及预期,存在估值下调轮风险 | 对冲敞口;相对组合保守控制仓位 |
| 估值立场 | 偏高 | $2B 估值对应 $100–140M ARR 为 14–20x;若 ARR 为 $5M 则约 400x;AI 原生公司中位数按已验证 ARR 为 21x | 入场价格要求收入按高端执行;外部视角安全边际很薄 |
| 上调触发项 | 经审计 ARR ≥$100M + GM ≥60% + NRR ≥110% | 这三项数据足以支持在当前 $2B 估值附近给出买入 | 任何承诺前先索取资料室访问权限 |
建议反映无法接触私有尽调资料的外部投资人立场。战略投资人和领投 VC(Bezos Expeditions、Khosla、NVentures)已开展非公开尽调,外部无法取得;他们的信念是正向信号,可补充但不能替代本评估。
[CV001, CV014, CV015, CV036]从证据质量、估值评估到最终 TRACK 建议的决策逻辑,并标出哪些证据升级会改变结论。
逻辑流是定性判断;节点权重基于可获得公开证据评估,不是定量评分。投资人质量信号是最强正面因素;收入证据缺口是升级到 BUY 的主要约束。
[CV009, CV010, CV011, CV015, CV036]8.2 投资论证与反论证
FieldAI 的投资论证建立在五个相互咬合的支柱上。第一,工业机器人硬件商品化后,具身 AI 软件层在结构上更容易捕获企业价值:一个不绑定硬件的软件大脑,叠加跨多样化部署的联邦学习网络效应,天然具备垄断吸引力。第二,创始团队——Ali Agha(NASA JPL / DARPA SubT 获胜者)、Shayegan Omidshafiei(DeepMind / MIT)和 David Fan(DARPA RACER / JPL)——有可信的先验验证:这一路线能在野外条件跑通,而不只是仿真里有效。第三,投资人组合(Bezos Expeditions、Khosla Ventures、NVentures/NVIDIA、Temasek、Intel Capital、Gates Frontier)可说是围绕一家机器人 AI 公司集结的最高确信度战略与财务投资人组合;他们做过外部无法获得的私下尽调。第四,公司走出隐身模式时,生产部署已在三大洲数百个场站上线,并在公告后九个月内报告 >$100M 已预订收入——商业牵引速度快过任何可比具身 AI 软件公司。第五,建筑、能源、采矿、物流和国防的 TAM 很大,渗透率仍低。 反论证逐一挑战这些支柱。如果机器人 OEM 纵向整合自有 AI 软件(Boston Dynamics、Figure AI、Agility Robotics),或超大规模云厂商(Google/Intrinsic、NVIDIA Cosmos)把基础模型层商品化,硬件无关优势可能被削弱。收入数字仍未经验证:$100M 已预订收入来自匿名消息源,「booked」可能包含销售管线或未承诺订单;$140M ARR 估算来自第三方分析平台,方法未披露;另一个估算把收入压到 $5M,差异达 28 倍。外部投资人看到的治理和财务透明度几乎为零:没有经审计财务、没有运营实体 Form D、没有披露毛利率、CAC 或 NRR。$2B 估值已经计入大量执行兑现:即便按 $140M ARR,14x 收入倍数也需要高毛利、高留存和持续高速增长,才足以支撑二级入场。资深投资人 Howard Morgan(First Round Capital / B Capital)公开警告,「高买、更高卖只在泡沫里有效」;Seattle 地区 VC 在 2025 年 12 月 GeekWire 圆桌上也观察到,种子轮和 Series A 私募市场估值已远远跑在基本面前面。2026–2027 年环境下,Series B+ AI 软件倍数压缩是真实风险。 [CV009, CV010, CV011, CV012, CV013, CV014]
| 论点 | 证据基础 | 什么会改变观点 |
|---|---|---|
| 正方:机器人硬件商品化后,硬件无关 AI 软件拿到近似垄断份额 | FFM 已部署到四足 / 人形 / 轮式 / 履带式平台;TechCrunch 和官方来源确认硬件无关架构 | OEM 纵向整合(Boston Dynamics、Figure)拿下“大脑层”;或超大规模云厂商(Google Intrinsic、NVIDIA Cosmos)让基础模型商品化 |
| 正方:创始团队拥有行业最高现场机器人信誉 | DARPA SubT 获胜、NASA JPL、MIT/DeepMind 履历有据可查;Marc Raibert / Vinod Khosla 公开背书 | 若 CEO 离开,出现关键人风险;或 CSO / CTO 被资金更充裕的竞争对手吸引 |
| 正方:顶尖投资人投入 $405M,说明其私有尽调优于公开数据 | 轮次超额认购;Bezos Expeditions、Temasek、Khosla、NVentures 均参与;投资人声明确认商业信念 | 若 B 轮出现估值下调轮,将推翻私有尽调假设;若投资人减记该头寸,信号反转 |
| 正方:公告后 9 个月已签约收入超过 $100M,商业牵引异常快 | OCBJ 2026 年 4 月匿名来源;GetLatka 估计 ARR 为 $140M;CEO 称合同为数百万美元级 | 已签约收入与管线混同;ARR 明显更低;流失暴露客户不满 |
| 反方:收入未验证,估计区间跨度达 28x | 估计从 $5M(CompWorth)到 $140M ARR(GetLatka)不等;无审计数字;OCBJ 使用匿名来源 | 公司发布经审计 ARR;若数字高于 $80M,基本可补上这一缺口 |
| 反方:对一家 $2B 公司而言,治理不透明程度极高 | 未披露董事会构成;Field Ai, Inc. 未提交 Form D;无审计财务;无股权结构表细节 | NDA 下正式开放资料室即可满足;IPO 前财务报告可关闭缺口 |
| 反方:AI 软件倍数压缩是 2026–2027 年结构性风险 | Howard Morgan(First Round Capital)警告 AI 估值过热;Madrona 合伙人称种子轮 / A 轮定价远超基本面 | 经过验证的收入推动全行业重新上调估值,则该风险失效;或 FieldAI 自身 B 轮提价确认 |
| 反方:竞争对手(Physical Intelligence、Skild AI)以相当或更弱收入拿到更高估值 | PI 以零收入洽谈 $11B 估值;Skild 以 $30M ARR 拿到 $14B 估值;行业可选性溢价推高全部估值 | 若行业倍数压缩,且 PI 或 Skild 以平轮 / 估值下调轮融资,FieldAI 的 $2B 可能显得昂贵 |
正方行代表多头 / 证实性解读;反方行代表空头 / 挑战性解读。两组论点截至 2026 年 5 月 runDate 均有证据支持。
[CV009, CV010, CV011, CV012, CV013, CV014]8.3 可比估值与市场背景
具身 AI 和工业机器人软件赛道提供了一组有用但并不完美的可比公司。在具身 AI 软件同业中,Physical Intelligence 正洽谈近乎零披露收入下的 $11B 估值(2026 年 3 月)——这是由研究声望和期权价值驱动的极端离群点,而不是当前商业牵引。Skild AI 2026 年 1 月以 $30M ARR 且指数级增长拿到 $14B 估值,对小基数隐含约 467x 远期倍数——同样是早期阶段的期权倍数,不是基本面倍数。这些可比项确认,投资人群体正在给具身 AI 软件平台很高的期权溢价;但从近零收入对应 $14B–$5.6B 的巨大区间,也说明终局价值假设高度不确定。 如果 $100–140M ARR 估算正确,FieldAI $2B 估值实际隐含的倍数更温和(14–20x),低于两个最接近的具身 AI 软件同业;只要收入真实,它在品类内看起来定价有吸引力。但如果真实收入只有 $5–20M,隐含倍数(100–400x)会跻身赛道内最激进估值之列。硬件一体化可比公司——Apptronik($5.3B,人形硬件 + 软件)和 Gecko Robotics($1.25B,关键基础设施检测)——商业模式结构不同,软件毛利更低,但收入更可验证,因此会把纯软件估值可比倍数往下拉。Houlihan Lokey 2026 年一季度关于 AI 垂直软件的报告发现,AI 原生软件公司在 VC 轮中的 EV/Revenue 中位数为 21.2x,而 AI 赋能软件为 8.5x、传统 SaaS 为 5.5x。若 FieldAI 有 $100M ARR 且按 21x 计,估值可支撑约 $2.1B,基本与当前标记一致。关键风险在于,这一中位数适用于财务已验证的公司,而 FieldAI 收入完全没有外部验证。 [CV017, CV018, CV019, CV020, CV021, CV022]
| 公司 | 阶段 / 上一轮 | 估值 | ARR / 收入信号 | 营收倍数(估计) | 对 FieldAI 的参考价值 | 局限 |
|---|---|---|---|---|---|---|
| Physical Intelligence | Series B 轮($600M,2025 年 11 月) | $5.6B(Bloomberg 2026 年 3 月称正洽谈 $11B+) | 披露收入接近零;未说明商业化时间表 | N/A(收入前阶段);期权性估值倍数 | 最接近的具身 AI 软件可比对象;走基础模型路线;不绑定硬件 | 没有收入 → 估值完全由研究期权性和品牌撑起;做基本面比较的参考价值较低 |
| Skild AI | Series C 轮($1.4B,2026 年 1 月) | $14B | $30M ARR(BusinessWire 2026 年 1 月);快速增长 | 按 $30M ARR 约 467x | 直接竞争者;同一代公司(2023 年成立);全形态具身基础模型;定位相近 | 收入更低却估值更高,显示投资人极度乐观;Skild 的收入验证也有限 |
| Apptronik | Series A 延伸轮(累计 $935M,2026 年 2 月) | $5.3B | 未披露;硬件优先的人形机器人 | 未披露;按硬件阶段看可能偏低 | 人形机器人集成伙伴;软件 + 硬件模式不同于 FieldAI 的纯软件模式 | 硬件经济性会拖累利润率,相比 FieldAI 纯软件模式,可比性较弱 |
| Gecko Robotics | Series D 轮($125M,2025 年 6 月) | $1.25B | 未单独披露;关键基础设施检测收入在增长 | 未披露;工业检测赛道 | 工业机器人 AI 平台;软件 + 硬件;团队有 DARPA 背书 | 聚焦检测而非通用自主;硬件成分限制软件倍数适用性 |
| Figure AI | Series C 轮($1B+,2025 年 9 月) | $39B | 未公开披露;人形硬件仍处商业化前阶段 | 未披露;期权性溢价极高 | 人形机器人最高估值;衡量赛道情绪上限的标尺 | 聚焦人形硬件,并与汽车 OEM(BMW)合作;结构上不同于 FieldAI |
| FieldAI(标的) | Series A1 轮($314M,2025 年 8 月) | $2.0B | >$100M 预订收入(OCBJ 2026 年 4 月匿名来源);$140M ARR 估计(GetLatka);$5M 估计(CompWorth) | 14–20x(若为 $100–140M);约 400x(若为 $5M) | 标的公司 | 收入未验证;估计区间相差 28x,带来基本面不确定性 |
所有估值均为最近一次披露私募轮标记,不是公开市场价格。可比公司的收入数字来自自报或第三方估计,均未经审计。「营收倍数」仅作示意。Physical Intelligence 和 Skild AI 的估值显示,市场给具身 AI 软件定价时,几乎不看当期收入,而是在为极高的期权性付费;如果收入为真,FieldAI 的 $2B 在这组可比公司中反而保守。
[CV017, CV018, CV019, CV020, CV021, CV022]在不同收入情景下,$2B 估值隐含的 EV/ARR 倍数,并以 Houlihan Lokey AI 原生软件中位数(21.2x)作为参照。
所有收入数值都是估算或第三方报道,均未审计。$2B 估值分母来自 Reuters/CNBC 援引知情人士。Houlihan Lokey 21.2x 中位数适用于已验证财务的 AI 原生软件公司;FieldAI 收入未验证,因此对比只是示意,不是精确匹配。
[CV022, CV023, CV024, CV028]8.4 牛市、基准与熊市情景
牛市情景要求 FieldAI 收入位于估算高端($140M ARR,年同比增长 80–100%),软件许可流毛利率接近 70–80%,由车队扩张驱动的 NRR 超过 120%,并在 2027 年以 $350–500M ARR 为基数、按 15–20x 远期 ARR 完成 Series B——对应 $5.25–10B 估值,相比 $2B 标记回报 2.6–5x。该情景的关键风险是收入数字未经验证,增长率未获独立确认,且市场环境必须继续支持 AI 软件高倍数。 基准情景假设 $100M ARR、40–60% 增长、混合毛利率 55–70%、NRR 110–120%。若 2027–2028 年 Series B 以 $200–300M ARR 为基数、按 10–12x 远期 ARR 定价,估值将为 $2.0–3.6B——相对 $2B 标记回报 0–1.8x。这意味着大体持平到温和升值,和风险水平相匹配。 熊市情景假设收入显著更低($20–50M ARR),行业逆风拖慢增长(企业采纳摩擦、宏观 capex 回撤),AI 软件倍数压缩把下一轮隐含估值拉到 $0.4–1.0B——相比 $2B 标记亏损 50–80%。该情景下降价轮风险最高。负面触发因素包括:若 Physical Intelligence 或 Skild AI 以低于预期的倍数融资,整个机器人赛道估值重置;FieldAI 客户削减部署或不续约;或出现证据显示「已预订收入」显著高估已确认 ARR。 [CV025, CV026, CV027, CV028, CV029, CV030]
| 情景 | 关键假设 | 退出 / 下一轮隐含估值 | $2B 标记估值下的回报逻辑 | 概率信号 |
|---|---|---|---|---|
| 牛市 | ARR:$140M(已验证),同比增长 80–100%;GM:70–80%;NRR:120%+;2027 年 B 轮按 $350–500M ARR 的 15–20x 定价 | B 轮估值 $5.25B–$10B;2030 年 IPO 估值 $15–30B+ | 从 $2B 入场为 2.6–5x;到 IPO 潜在 7–15x | 低-中:需要收入验证和持续增长同时成立;AI 倍数必须守住 |
| 基准 | ARR:$100M(未验证),同比增长 40–60%;GM:55–70%;NRR:110–120%;2028 年 B 轮按 $200–300M ARR 的 10–12x 定价 | B 轮估值 $2.0B–$3.6B;2031 年 IPO 估值 $6–9B | 从 $2B 入场为 0–1.8x;到 IPO 约 3–4.5x;尚可但不突出 | 中:与“好但不突出”的执行一致;需要避开行业压缩 |
| 熊市 | ARR:$20–50M(实际),增长放缓;行业倍数压缩至 8–10x;2027 年估值下调轮 | 下一轮估值 $0.4B–$1.0B;可能重组或人才收购 | 从 $2B 入场亏损 50–80%;以困境价格救援式收购 | 低-中:由收入意外叠加行业修正触发;若投资人做过真实尽调,则不太可能 |
所有情景均为基于公开数据和可比公司基准的模型估计。没有审计财务就无法验证。情景中的收入数字是假设区间,不是已确认数值。回报估计假设现有股权结构之外不再发生额外稀释。
[CV025, CV026, CV027, CV028, CV029]下一轮融资(2027–2028)在牛市、基准和熊市情景下的隐含估值结果,并以当前 $2B 标记作为参照。
所有情景区间均为建模估算,依据可比公司倍数和既定收入假设。没有审计财务,任何区间都无法验证。回报计算假设入场价为 $2B,且没有额外稀释;实际稀释会压低回报。数值以百万美元计。
[CV025, CV026, CV027, CV030]8.5 建议、否决触发点与最终尽调要求
建议为 TRACK(中等信心、高风险、估值偏紧)。投资论证在结构上成立:具身 AI 软件采用物理优先架构,部署已经在野外跑通;创始人与市场高度匹配;投资人组合也几乎是最强配置。但外部投资人如果拿不到经审计财务、已验证 ARR 和披露毛利率,就无法承销 $2B 价格——主导风险是信息不对称,不是技术或团队质量。按当前 $2B 标记,入场价格已经按收入区间高端的执行兑现定价;任何收入下修或全行业倍数压缩都会造成实质性稀释。 若满足三个条件,建议可上调至 BUY:(1)披露经审计或独立验证的 ARR ≥$100M,且年同比增长 ≥40%;(2)确认混合毛利率 ≥60%;(3)通过队列数据证明净收入留存 ≥110%。若不满足,TRACK 仍合适:继续观察公司披露的财务里程碑、Series B 公告和隐含估值抬升、战略合作收入披露,以及客户集中度风险信号。若出现以下情况,建议下调至 AVOID:(1)Series B 公告估值低于 $2B(降价轮);(2)重大客户流失或部署失败公开;或(3)已报告的 >$100M 已预订收入被大幅下修。 治理是明显缺口:董事会构成未披露,清算优先权和期权池规模未知,也没有独立财务报告。任何潜在投资人都必须在承诺前完成这些尽调。公司把自己定位为工业机器人的「软件大脑」层,确实差异化且短期内难以复制;但治理不透明叠加估值偏高,要求投资人保持纪律性的 TRACK 姿态,直到商业验证点可以被公开核验。 [CV031, CV032, CV033, CV034, CV035, CV036]
| 触发器 | 阈值 / 可衡量事件 | 对投资论点的传导 | 行动含义 |
|---|---|---|---|
| Series B 轮下调估值 | Series B 轮以低于 $2.0B 的估值完成 | 直接违背私募投资人论点;意味着收入不及预期或赛道重估 | 下调至 AVOID;立即退出或降低敞口 |
| 大客户流失或部署失败 | 公开宣布客户退出,或安全事故导致部署暂停 | 削弱「数百个站点、三大洲」的商业牵引叙事 | 复核仓位;行动前判断事件是个案还是系统性问题 |
| 收入修正显著低于 $100M 预订收入 | 公司披露或调查报道显示预订收入 < $50M,或预订额 ≠ ARR | 按真实 ARR 计算,估值倍数跳到 40x+;超出当前阶段的合理区间 | 下调至 AVOID;若倍数压缩到 10–15x,估值将落在 $500M–750M |
| 被超大云厂商替代 | NVIDIA Cosmos、Google Intrinsic 或 Amazon AWS Robotics 宣布可比 FFM 服务 | 授权费承压;平台风险兑现;护城河受到质疑 | 重估防御性;按影响范围下调至 TRACK 或 AVOID |
| 治理或法律风险事件 | IP 纠纷、监管行动、关键人物离职(CEO 或 CSO),或 SEC 执法 | 动摇团队和投资人信心;抬高执行风险 | 重大风险事件 → 下调至 TRACK,待解决;严重事件 → AVOID |
这些触发器都设在可观察阈值上,便于在无法获取私有财务数据时持续监控。五项均会推翻投资论点;任何一项发生都会实质改变建议。下调估值轮是最可操作的公开信号。
[CV031, CV032, CV033, CV034]| 主题 | 缺失证据 | 为什么重要 | 负责人或尽调路径 |
|---|---|---|---|
| 经审计 ARR 与收入确认政策 | 没有经审计财务;估计区间为 $5M–$140M;「预订收入」≠ 已确认 ARR | 没有验证过的收入锚点,无法承销 $2B 估值;它决定适用 14x 还是 400x 倍数 | 在 NDA 下索取经审计或审阅的 P&L 和 ARR 明细;可接受滚动 12 个月 ARR,但需附收入确认脚注 |
| 按收入流拆分的毛利率 | 未披露;软件授权与硬件集成占比未知;综合毛利率无法验证 | 软件级毛利率(70–80%)能支撑溢价;混合毛利率(55–65%)撑不起同一倍数 | 索取按授权与服务拆分的贡献毛利瀑布图;对照公开 AI SaaS 基准 |
| 净收入留存与 cohort 数据 | 未披露;公司称有「扩张合同」,但 NRR 未验证 | 要支撑先落地再扩张的 SaaS 倍数,NRR 需 >120%;若 <100%,将触发熊案 | 索取按产品线划分的季度 cohort 留存;确认 2024–2026 cohort 的 ACV 轨迹 |
| 股权结构、清算优先权和期权池 | 未披露;优先权负担和反稀释保护未知 | Series A/A1 的重清算优先权可能在基准情形下把普通股价值压到接近零 | 向 Series A 领投方律师索取股权结构摘要和瀑布分析;确认优先权结构 |
| 董事会构成与治理文件 | 未公开披露;独立董事成员未知 | 治理质量是机构投资人的关键尽调维度;治理弱 = 代理风险更高 | 索取董事会材料、观察员权利协议和投资人权利协议;确认审计委员会状态 |
| 客户集中度与合同条款 | 客户名称未披露;前 5 大客户收入集中度未知;合同期限 / 续约条款未知 | 若集中度高(前 3 大客户 >50% 收入),流失风险画像会实质改变 | 在 NDA 下索取脱敏客户明细;确认单一客户收入上限;审阅合同续约条款 |
六项要求都是投资前标准尽调问题。对这个阶段的私营公司而言都不异常;若要用任何基本面锚点承销 $2B 估值,这些证据都必须拿到。前三项(ARR、GM、NRR)是从 TRACK 上调至 BUY 的最低门槛;后三项(股权结构、治理、客户集中度)则是任何投委会批准前的必备材料。
[CV033, CV034, CV035, CV036]面向 IC 的 FieldAI 七项投资维度评分;0–10 分,10 分代表同类最佳。
评分是分析师基于截至 2026 年 5 月公开证据作出的定性判断。财务透明度得分 2/10,反映一家 $2B 公司不透明程度异常高;若公司通过尽调资料室补充披露,该分数会明显上调。综合得分(未加权平均)约为 5.9/10,因此更支持跟踪,而不是买入。
[CV009, CV010, CV011, CV014, CV015, CV033]8.6 附录
免责声明
本报告是基于公开证据的尽调快照,不构成投资建议。关键财务、法律、技术和合同事实仍未公开;作出任何投资决定前,应直接向管理层和一手文件核验。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | FieldAI was founded in 2023 by Ali Agha in Southern California. | 高 | SO003, SO005, SO008 |
| CO002 | FieldAI is headquartered in Irvine, California as of September 2025, having moved from approximately 13,000 sqft of space in Mission Viejo, CA. | 高 | SO003, SO004, SO005, SO015 |
| CO003 | FieldAI is a pure-play AI software company; it does not design, manufacture, or sell robot hardware. | 高 | SO001, SO003, SO014 |
| CO004 | FieldAI's core product is the Field Foundation Model (FFM), a physics-first, risk-aware embodied AI platform that acts as a universal robot brain. | 高 | SO003, SO005, SO009 |
| CO005 | FFMs are hardware-agnostic and have been proven across quadrupeds, humanoids, wheeled robots, and passenger-scale vehicles in production deployments. | 高 | SO003, SO004, SO009 |
| CO006 | FieldAI robots operate fully autonomously on-edge without cloud connectivity, GPS, pre-mapped environments, or predefined paths. | 高 | SO003, SO004, SO020 |
| CO007 | FieldAI operates a B2B model combining software licensing and hardware-integration services; subscription pricing reportedly ranges from tens of thousands to $500,000 per year. | 中 | SO009, SO014 |
| CO008 | FieldAI's current flagship product is called EDGE, described as a "General-Purpose Robot Brain." | 中 | SO001 |
| CO009 | FieldAI has production deployments across three continents—North America, Europe, and Asia-Pacific (including Japan)—spanning thousands of missions at hundreds of sites. | 中 | SO003, SO014 |
| CO010 | FieldAI serves customers in construction, energy, mining, manufacturing, logistics, urban delivery, inspection, security, and defense verticals. | 高 | SO003, SO009, SO012 |
| CO011 | Ali Agha is the Founder and CEO of FieldAI with nearly two decades of expertise in robotics AI and autonomous systems. | 高 | SO002, SO003, SO021 |
| CO012 | Before founding FieldAI, Ali Agha spent approximately seven years at NASA's Jet Propulsion Laboratory (JPL) as Principal Investigator for the DARPA Subterranean Challenge, DARPA RACER, NASA Autonomous Mars Cave Exploration, and Coordinated Autonomy for Mars Helicopter-Rover programs. | 高 | SO002, SO021 |
| CO013 | Ali Agha led the CoSTAR team (JPL-MIT-Caltech-KAIST-LTU) that won the Urban Circuit of the 2020 DARPA Subterranean Challenge, demonstrating multi-robot GPS-denied autonomy in complex urban environments. | 高 | SO002, SO011, SO020 |
| CO014 | Dr. Shayegan Omidshafiei is co-founder and Chief Science Officer of FieldAI; he was previously a leading Research Scientist at DeepMind and Google for over five years, and earned his PhD and SM from MIT where he first collaborated with Agha. | 高 | SO002, SO020 |
| CO015 | Dr. David Fan is co-founder and CTO of FieldAI; he served as chief technologist for DARPA SubT and DARPA RACER programs, and was a NASA JPL research fellow who worked closely with Agha prior to founding the company. | 高 | SO002, SO020 |
| CO016 | Dr. Eric Krotkov leads FieldAI Federal (the company's government/defense division); he is a former DARPA Program Manager, former Chief Science Officer of Toyota Research Institute, CMU professor, and the creator of the PackBot and Talon military robots. | 高 | SO002, SO014 |
| CO017 | Sebastian Scherer is FieldAI's Director of Fieldable Embodied AI and also serves as Associate Research Professor at Carnegie Mellon University's Robotics Institute. | 中 | SO002 |
| CO018 | Ali Agha earned his PhD in Computer Science and Engineering from Texas A&M University and held a postdoctoral position at MIT before working at Qualcomm Research and then NASA JPL. | 中 | SO020 |
| CO019 | Board composition and formal governance structure for FieldAI have not been publicly disclosed as of May 2026. | 中 | SO002 |
| CO020 | CEO Ali Agha is FieldAI's primary public spokesperson and fundraiser, representing a key-person dependency; no succession plan has been publicly disclosed. | 中 | SO002, SO005, SO008 |
| CO021 | FieldAI exited stealth and announced $405 million raised across two consecutive rounds on August 20, 2025. | 高 | SO003, SO004, SO005 |
| CO022 | The August 2025 funding announcement valued FieldAI at $2 billion, confirmed by Reuters, CNBC, TechCrunch, and FieldAI's own press release. | 高 | SO005, SO006, SO008 |
| CO023 | Prior to the August 2025 announcement, FieldAI had a $500 million valuation established in an earlier round that Reuters described as "a round last year" (i.e., approximately 2024). | 中 | SO008, SO009 |
| CO024 | The most recent of FieldAI's two disclosed rounds raised $314 million, making the first round approximately $91 million by subtraction from the $405 million total. | 中 | SO008, SO009 |
| CO025 | Gates Frontier (Bill Gates) and Samsung are described as "previous investors" in FieldAI, indicating they invested in the earlier round at the approximately $500 million valuation. | 高 | SO003, SO007, SO015 |
| CO026 | Bezos Expeditions, Prysm Capital, and Temasek co-led the $314 million Series A1 round announced August 2025. | 高 | SO003, SO007, SO009 |
| CO027 | Additional investors across FieldAI's disclosed rounds include Khosla Ventures, Intel Capital, BHP Ventures, Canaan Partners, Emerson Collective, and NVentures (NVIDIA's venture arm). | 高 | SO003, SO004, SO010 |
| CO028 | Vinod Khosla, founder of Khosla Ventures, stated: "FieldAI is at the forefront of the general-purpose robotics revolution, and its ability to rapidly deploy will unlock long-term economic and societal value." | 高 | SO003, SO010 |
| CO029 | The $314 million Series A1 round was oversubscribed, with most capital coming from investors reaching out to FieldAI rather than the company soliciting them, per CEO Ali Agha's statements. | 中 | SO003, SO018 |
| CO030 | The August 2025 funding will be used to accelerate global expansion, support product development in locomotion and manipulation, and double headcount by end of 2025. | 高 | SO003, SO004 |
| CO031 | FieldAI had approximately 30 employees at the end of 2024 and grew to approximately 130 by August 2025, planning to double headcount to approximately 260 by end of 2025. | 高 | SO008, SO005 |
| CO032 | LinkedIn listed FieldAI as having 201–500 employees as of early 2026, consistent with the company's doubling plan. | 中 | SO014 |
| CO033 | Sources familiar with the matter told the Orange County Business Journal in April 2026 that FieldAI has more than $100 million in booked revenue and partnerships with large industrial customers. | 中 | SO014 |
| CO034 | In late 2025, FieldAI completed a lease for the entirety of 3 Morgan, a 41,000-square-foot R&D building in Irvine Spectrum, moving from approximately 13,000 sqft in Mission Viejo. | 中 | SO015 |
| CO035 | FieldAI has established FieldAI Federal, a subsidiary division focused on government and federal defense applications. | 高 | SO002, SO014 |
| CO036 | FieldAI's production deployments span thousands of missions across hundreds of sites on three continents, with dozens of large enterprise customers as of April 2026. | 中 | SO014 |
| CO037 | In February 2026, Certis Group (Singapore's leading integrated security provider) and FieldAI formed a strategic partnership to deploy autonomous robots in real-world security operations globally. | 高 | SO012, SO001 |
| CO038 | On March 12, 2026, Boston Dynamics and FieldAI announced a partnership combining Spot's mobility with FieldAI's FFMs for construction and other dynamic environments; Marc Raibert personally endorsed the company. | 高 | SO011, SO017 |
| CO039 | FieldAI does not name any specific enterprise customers in its official announcements, declining to disclose customer identities as of its August 2025 funding release. | 高 | SO005, SO019 |
| CO040 | Active robotics deployments among surveyed construction contractors fell from 65% in 2024 to 46% in 2025, per BuiltWorlds data cited by Construction Dive, despite 95% positive evaluations—indicating a significant gap between willingness and actual deployment. | 中 | SO013 |
| CO041 | FieldAI had multi-million-dollar contracts in the U.S., Europe, and Asia as of August 2025, indicating early-stage commercial revenue at time of funding announcement. | 中 | SO008 |
| CO042 | Pre-diligence references to a "San Francisco" headquarters for FieldAI cannot be confirmed from any primary or credible secondary source; all reliable evidence consistently identifies Irvine, CA (and previously Mission Viejo, CA) as the company's location. | 高 | SO003, SO004, SO005, SO008, SO014, SO015 |
| CO043 | FieldAI's prior round breakdown described in some references as $150 million seed (August 2024) plus $255 million Series A (August 2025) cannot be confirmed from primary sources; Reuters' reporting implies a first round of approximately $91 million followed by a $314 million second round. | 中 | SO007, SO008, SO009 |
| CO044 | FieldAI has been recognized by Fast Company as one of the top innovative AI companies, per reporting by the Orange County Business Journal. | 低 | SO014 |
| CM001 | FieldAI's addressable market is the AI software intelligence layer for industrial robots deployed in unstructured environments, distinct from the traditional structured-factory automation market dominated by FANUC, ABB, and Yaskawa. | 高 | SM014, SM017 |
| CM002 | Unstructured industrial environments targeted by FieldAI include active construction sites, underground and open-pit mines, oil and gas topside and inspection environments, and flexible or high-mix manufacturing facilities where static robot programming fails. | 高 | SM004, SM005, SM008, SM018 |
| CM003 | FieldAI explicitly excludes structured factory automation (fixed assembly lines), consumer or logistics robots in controlled environments, robot hardware manufacturing, and subsea ROVs from its product scope. | 中 | SM014, SM017 |
| CM004 | Status-quo substitutes for FieldAI's platform include manual human inspection and monitoring crews, traditional fixed-path or single-task robots requiring extensive pre-programming, drone/UAV aerial surveys, and BIM/CAD modeling with manual data capture. | 中 | SM005, SM006 |
| CM005 | FieldAI's Boston Dynamics case study describes construction sites as among the most difficult environments for robotics due to constant evolution of terrain, workers and machines in motion, and materials arriving without notice. | 中 | SM005 |
| CM006 | Global Market Insights estimates the global AI-powered industrial robot market at $17.9 billion in 2026, growing at 7.1% CAGR to $33.3 billion by 2035, with North America as the largest market and Asia Pacific as the fastest-growing region. | 中 | SM002 |
| CM007 | The Business Research Company estimates the industrial artificial intelligence market (all software, platforms, services, hardware) at $13.69 billion in 2026 growing to $73.54 billion by 2030 at a 51.1% CAGR, driven by automation adoption in manufacturing, automotive, energy, and logistics. | 中 | SM010 |
| CM008 | The International Federation of Robotics reports 542,000 industrial robots installed globally in 2024—the second-highest annual total on record—with 4.66 million units in operational use, a 9% year-on-year increase. | 高 | SM001, SM025 |
| CM009 | IFR forecasts global industrial robot installations to grow 6% to 575,000 units in 2025 and to surpass 700,000 units per year by 2028, driven by continued demand across Asia, Americas, and Europe. | 高 | SM001, SM025 |
| CM010 | The Business Research Company and Research and Markets estimate the construction robotics market (broad scope) at approximately $7.79 billion in 2026, growing to $15.39 billion by 2030 at a CAGR of approximately 18%. | 低 | SM010 |
| CM011 | Grand View Research estimates the global construction robots market (narrow robot-only definition) at $1.4 billion in 2024, projected to reach $3.66 billion by 2030 at an 18% CAGR, driven by labor shortages and safety mandates. | 中 | SM003 |
| CM012 | Persistence Market Research values the global mining robotics market at $1.7 billion in 2026, expected to grow to $3.3 billion by 2033 at a CAGR of 9.8%, with open-pit mining accounting for approximately 63% of the market share. | 中 | SM009 |
| CM013 | MarkWide Research estimates the oil and gas robotics market at $3.8 billion in 2026, growing to $12.96 billion by 2035 at a CAGR of 14.6%, with growth driven by deepwater expansion, safety regulations, and AI-enabled inspection. | 低 | SM007 |
| CM014 | Fortune Business Insights estimates the inspection robotics in oil and gas market at $0.93 billion in 2026, growing to $1.51 billion by 2034 at a CAGR of 6.23%, with North America holding 32.2% market share in 2025. | 中 | SM019 |
| CM015 | Mordor Intelligence values the oil and gas automation market at $46.16 billion in 2026, growing to $63.19 billion by 2031 at 6.48% CAGR, with upstream operations accounting for 58.55% of revenue and software holding 66.12% of share. | 中 | SM020 |
| CM016 | Research and Markets projects the embodied AI market to grow at a CAGR of 17.5%, reaching $7.24 billion by 2030, covering automation and manufacturing, automotive, healthcare, and logistics verticals. | 低 | SM015 |
| CM017 | No published analyst estimate covers FieldAI's specific serviceable addressable market—AI software for general-purpose robots in unstructured industrial environments—requiring an author-derived SAM estimate using a 15–25% software-share assumption applied to combined vertical market data. | 低 | SM002, SM009, SM010, SM019 |
| CM018 | FieldAI raised $405 million across two consecutive funding rounds (seed August 2024, Series A August 2025), reaching a $2 billion valuation, with the latest round oversubscribed following rapid customer adoption. | 高 | SM008, SM016, SM017 |
| CM019 | FieldAI's platform is deployed in daily operations at numerous customer sites globally, spanning construction, energy, manufacturing, urban delivery, and inspection verticals across Japan, Europe, and the US. | 中 | SM014, SM018, SM024 |
| CM020 | FieldAI's NVIDIA collaboration demonstrated that what previously took customers 3.5 months of manual effort can now be completed in 12 hours with FieldAI's technology—over 200 times faster than manual methods. | 中 | SM004 |
| CM021 | Boston Dynamics and FieldAI announced a formal partnership in March 2026 to deploy autonomous robots in construction and other complex dynamic environments, combining Spot's mobility with FieldAI's Field Foundation Models. | 高 | SM005, SM006, SM016 |
| CM022 | FieldAI's Boston Dynamics case study documents that customers reduced inspection and documentation time by more than 90% compared to manual processes and avoided millions of dollars in potential cost overruns through earlier issue detection. | 中 | SM005 |
| CM023 | FieldAI has deployed Spot robots across construction sites spanning Asia, Europe, and North America over the past two years, with customers expanding initial pilots into fleet-wide deployments standardizing robotic autonomy across global operations. | 中 | SM005, SM006 |
| CM024 | Approximately 55% of the $3.55 billion invested in construction technology in Q1 2025 went toward funding next-generation robotics and AI-enabled technology, according to Nymbl Ventures data cited by Construction Dive. | 中 | SM016 |
| CM025 | FieldAI's enterprise buyer structure has payers at C-suite or VP Operations/Finance level, technical buyers at HSE or operations leadership, and day-to-day users at site superintendent or project manager level. | 中 | SM005, SM006, SM014 |
| CM026 | Construction segment buying power concentrates in large ENR-ranked general contractors, who bear cost-overrun risk and have internal digital transformation budgets for automation pilots; FieldAI's customers include some of the top 10 ENR-ranked construction firms. | 中 | SM005 |
| CM027 | Mining and energy buyers are concentrated around major operators whose HSE and operations leaders are incentivized by regulatory compliance, incident reduction, and fleet productivity metrics; industrial AI robotics spend falls within HSE or operational capex budgets. | 中 | SM009, SM019 |
| CM028 | Key construction adoption trigger is the inability to generate accurate as-built BIM data quickly enough at scale with available labor; robots that cut documentation time by 90%+ address a direct operational pain point. | 中 | SM005, SM016 |
| CM029 | Mining and energy adoption is primarily triggered by safety regulations (MSHA fatality reporting, OHS mandates) and persistent labor shortages, both of which are structural and growing rather than cyclical. | 中 | SM009, SM011 |
| CM030 | BHP Ventures, the venture capital arm of one of the world's largest mining companies, is listed as an investor in FieldAI's funding rounds, providing direct market validation of the mining vertical opportunity. | 中 | SM009, SM018 |
| CM031 | Industry estimates indicate that over 50% of the Western US mining workforce is expected to retire by 2029, creating structural labor shortages that drive automation adoption in mines. | 中 | SM009, SM011 |
| CM032 | Studies on mining automation indicate that autonomous and semi-autonomous systems can increase productivity by 20% in mines that have implemented them, while Rio Tinto's autonomous truck fleets demonstrate safer and more predictable cycle times. | 中 | SM009 |
| CM033 | The US Mine Safety and Health Administration (MSHA) recorded 31 mining fatalities in fiscal year 2024 at a fatal-injury rate of 0.0110 per 200,000 hours worked, sustaining regulatory pressure for autonomous alternatives in hazardous mining tasks. | 中 | SM009 |
| CM034 | Grand View Research and other analysts project construction robotics to grow at an 18% CAGR through 2030, driven by labor shortages, urbanization, and tightening safety mandates for hazardous construction environments. | 中 | SM003 |
| CM035 | IFR's World Robotics 2025 report confirms that global robot installations are expected to grow 6% to 575,000 units in 2025 and surpass 700,000 units by 2028, indicating sustained growth in the overall robotics installed base that FieldAI's software layer can serve. | 高 | SM001, SM025 |
| CM036 | McKinsey's Global Industrial Robotics Survey found that 71% of industrial companies cite the capital cost of robots and 61% cite lack of internal experience as the primary barriers to robotics and automation adoption. | 高 | SM012, SM013 |
| CM037 | McKinsey concludes that activities requiring high levels of human input—such as assembly, stamping, surface treatment, and welding—are less likely to be automated in the short to medium term, limiting the portion of industrial activity addressable by current AI robotics platforms. | 高 | SM012, SM013 |
| CM038 | Deloitte's 2026 TMT Predictions reports that annual industrial robot sales have remained flat at approximately 500,000 units since 2021 and warns that unless data quality, integration complexity, and cybersecurity bottlenecks are resolved, the market will stay at relatively modest annual growth. | 高 | SM013, SM012 |
| CM039 | BuiltWorlds' 2025 Equipment and Robotics Benchmarking report found that the share of construction firms reporting active robotics use fell from 65% in 2024 to 46% in 2025, even as positive evaluations jumped from 74% to over 95%—evidence of widespread pilot purgatory. | 中 | SM016 |
| CM040 | Fortune Business Insights identifies extensive capital and maintenance requirements as the primary market restraint for oil and gas inspection robotics, with significant upfront investment required and a shortage of skilled professionals to deploy and analyze robot data. | 中 | SM019 |
| CM041 | Deloitte estimates annual humanoid robot shipments for industrial use in the range of 5,000–7,000 units in 2025, growing to approximately 15,000 units in 2026—indicating the humanoid robot market is in its very early commercial stage. | 中 | SM013 |
| CM042 | Analytics Insight notes that the core value in industrial robotics is shifting from physical hardware to AI-driven software and intelligence, with factories beginning to select robots for compatibility with an intelligence ecosystem rather than standalone hardware performance. | 低 | SM022 |
| CP001 | As of May 2026, buyers can solve the industrial robot autonomy job through at least five distinct paths: FieldAI software, rival embodied-AI platforms, inspection-specialist verticals, robot-OEM bundled intelligence, or status-quo manual processes and in-house navigation. | 中 | SP001, SP014 |
| CP002 | Multi-homing is common among industrial enterprise buyers—most large operators use more than one robot platform or autonomy approach simultaneously, making switching costs softer than platform vendors imply. | 中 | SP002, SP006 |
| CP003 | FieldAI's Field Foundation Models operate fully on-edge with latency under 100 milliseconds, require no cloud connectivity, no GPS, and no pre-mapped environments, enabling deployment in air-gapped and GPS-denied locations. | 中 | SP014, SP020 |
| CP004 | No competitor reviewed in this chapter—Intrinsic, Physical Intelligence, Covariant, Viam—has confirmed on-edge, air-gappable embodied AI autonomy for unstructured outdoor GPS-denied industrial environments. | 中 | SP022, SP015, SP010, SP024 |
| CP005 | FieldAI's Multiagent Foundation Model (MFM) enables fleet-scale real-time coordination of heterogeneous robots; this multi-agent AI reasoning layer is distinct from Boston Dynamics Orbit's centralized fleet orchestration. | 中 | SP020, SP002 |
| CP006 | Intrinsic, originally an Alphabet X spin-out founded 2021, was folded into Google in February 2026, gaining access to Google DeepMind, Gemini AI models, and Google Cloud infrastructure. | 高 | SP003, SP004 |
| CP007 | In May 2026, Fanuc—the world's largest industrial robot manufacturer with 1.1 million deployed robots and roughly 16–18% of global robot shipments—announced it would integrate Google Cloud's Gemini and Intrinsic's Flowstate into its robot systems, giving Google/Intrinsic access to the largest installed base in industrial robotics. | 高 | SP005, SP004 |
| CP008 | Intrinsic laid off approximately 20 percent of its workforce in January 2023 after five years of development as an Alphabet Other Bet, before announcing its first product, Flowstate, and subsequently being absorbed into Google. | 高 | SP003, SP004 |
| CP009 | Physical Intelligence (Pi), founded in 2024 by former Google DeepMind researcher Sergey Levine and colleagues, has raised approximately $70M seed + $400M Series A + $600M Series B and is reportedly seeking $1 billion more at an $11 billion valuation as of early 2026—5.5× FieldAI's $2 billion valuation. | 中 | SP011, SP012 |
| CP010 | Physical Intelligence has not released a commercial product or disclosed a commercialization roadmap as of March 2026; all capital is directed at research, data collection, and model scaling. | 中 | SP011 |
| CP011 | Physical Intelligence's π0.7 model (April 2026) demonstrates emergent generalization across manipulation tasks it was not explicitly trained on, using a reinforcement-learning token approach enabling sub-millimeter precision learning in 15 minutes of online training. | 中 | SP015, SP021 |
| CP012 | Covariant's founders (Pieter Abbeel, Peter Chen, Rocky Duan) and approximately 25% of staff were hired by Amazon in August 2024 under a reverse acqui-hire arrangement that gave Amazon a non-exclusive license to RFM-1 and related robotic foundation model IP. | 高 | SP010, SP027 |
| CP013 | Boston Dynamics and FieldAI announced a formal partnership in March 2026 to extend Spot's autonomy into uncharted, dynamic environments including construction sites; Boston Dynamics simultaneously runs the Orbit software platform backed by Google Gemini AI. | 高 | SP002, SP006, SP016 |
| CP014 | Gecko Robotics reached unicorn status ($1.25B valuation) in June 2025 after raising $125M Series D led by Cox Enterprises; total raised is $347M including Series C, seed, and earlier rounds. | 高 | SP007, SP019 |
| CP015 | Gecko Robotics' Cantilever platform combines TOKA magnetic-crawling robots with AI-powered digital twin analytics, serving energy, defense, and manufacturing sectors; its focus is proprietary-robot inspection, not general robot autonomy. | 高 | SP007, SP008 |
| CP016 | Viam, founded in 2020 by former MongoDB CTO Eliot Horowitz, is an open-source robotics SDK platform that has raised $117M total as of Series C (March 2025) and positions itself as hardware abstraction middleware, not a trained autonomy system. | 中 | SP013, SP024 |
| CP017 | Machina Labs raised a $124M Series C in February 2026 from Lockheed Martin Ventures, Woven Capital (Toyota), and Balerion Space Ventures to launch a 200,000-sq-ft AI-driven metal forming factory; its RoboCraftsman platform is application-specific (sheet metal forming), not a general autonomy brain. | 高 | SP009, SP025 |
| CP018 | Boston Dynamics Orbit integrates Google's Gemini AI via its AIVI-Learning feature for industrial-environment reasoning, and manages multi-site fleets of Spot, Stretch, and eventually Atlas robots through centralized dashboards. | 高 | SP006, SP002 |
| CP019 | FieldAI's Field Foundation Models support multiple embodiment types including quadrupeds, humanoids, wheeled robots, tracked vehicles, and autonomous vehicles; no major embodied-AI competitor has confirmed validated deployment across a comparably wide range of robot types. | 中 | SP001, SP017 |
| CP020 | Intrinsic's Flowstate is hardware-agnostic across KUKA, Fanuc, Universal Robots, and other industrial arm brands, and offers no/low-code development for manufacturers without deep robotics experience; it relies on Google Cloud for compute. | 高 | SP022, SP003 |
| CP021 | FieldAI's software subscriptions reportedly range from tens of thousands to approximately $500,000 per year per deployment, with company-reported booked revenue exceeding $100M and a third-party estimate of $140M ARR for 2025. | 低 | SP014, SP018 |
| CP022 | Google and Intrinsic have not disclosed pricing for Flowstate; the platform is expected to follow a tiered enterprise SaaS + Google Cloud compute model based on the Fanuc partnership framing and developer challenge structure. | 低 | SP004, SP005 |
| CP023 | Gecko Robotics prices through multi-year service contracts that bundle robot deployment, Cantilever platform access, and engineering consulting; the March 2026 U.S. Navy IDIQ has an initial $54M award and a $71M ceiling over five years for 18 vessels. | 高 | SP023, SP008, SP019 |
| CP024 | Boston Dynamics Orbit is sold as a software layer on top of Boston Dynamics hardware; annual pricing has not been publicly disclosed but is referenced in enterprise case studies as ongoing operational software integrated with CMMS and WMS systems. | 中 | SP006 |
| CP025 | Viam offers a developer-friendly free-tier open-source core and enterprise pricing upon request; the open-source model drives developer adoption before upselling managed cloud infrastructure and enterprise support. | 中 | SP013, SP024 |
| CP026 | The U.S. Navy spends an estimated $13–$20 billion annually on ship maintenance; a shift to autonomous robotic inspection can reduce average vessel downtime from up to 90 days to under 30 days, representing a primary status-quo displacement opportunity for both Gecko and FieldAI. | 中 | SP023 |
| CP027 | FieldAI's claimed competitive moats are: (1) physics-first, risk-aware FFM data flywheel; (2) on-edge GPS-denied deployment; (3) multi-embodiment hardware agnosticism; (4) multi-agent fleet coordination; and (5) partner ecosystem anchored by Boston Dynamics, Certis Group, and CMU's Robotics Innovation Center. | 中 | SP001, SP002, SP020, SP028 |
| CP028 | Physical Intelligence's $11B valuation is 5.5× FieldAI's $2B valuation; Physical Intelligence has raised approximately $2.07B total, 5.1× FieldAI's $405M raised, despite having no commercial products or revenue. | 高 | SP011, SP018 |
| CP029 | If Physical Intelligence commercializes before FieldAI achieves a dominant data advantage, Pi's superior capital base and emergent VLA model capabilities could compress FieldAI's differentiation window in general-purpose robot intelligence. | 中 | SP011, SP021 |
| CP030 | UpsideList's March 2026 analysis of FieldAI assigns a 50% probability to a bear case in which NVIDIA Isaac or Boston Dynamics internalizes the autonomy layer, triggering a down-round that wipes common equity through the $506M preference stack on a $2B valuation (25% preference overhang). | 中 | SP026 |
| CP031 | Boston Dynamics is simultaneously FieldAI's most prominent hardware partner and the operator of the Orbit software platform, which integrates Google Gemini AI and could develop competing autonomy capabilities; no public exclusivity or IP-separation provisions have been disclosed. | 中 | SP002, SP006 |
| CP032 | Foundation-model commoditization risk is real: Physical Intelligence has open-sourced early π models, and VLA architectures are proliferating through academic releases and hyperscaler APIs, which could compress FieldAI's physics-first differentiation into a trade secret rather than a structural moat over time. | 中 | SP015, SP021, SP027 |
| CP033 | An independent robotics intelligence analysis rated Covariant's data moat as "NARROW, not wide," citing competing data accumulation by Physical Intelligence, Nimble, Nomagic, and Amazon; the same moat dynamics apply to any embodied AI platform, including FieldAI. | 中 | SP027 |
| CP034 | FieldAI's Boston Dynamics partnership documented a 90%-plus reduction in inspection and documentation time compared to manual processes on construction sites, which is the primary quantified ROI evidence for status-quo displacement. | 中 | SP002 |
| CP035 | FieldAI's $506M in preference capital on a $2B valuation represents 25.3% preference overhang; in a down-round or internalization scenario, common equity holders (including employees and early investors) face near-total dilution before preference holders are made whole. | 中 | SP026 |
| CP036 | Gecko Robotics won a five-year, up-to-$71M IDIQ contract with the U.S. Navy's Pacific Fleet in March 2026 to deploy robotic inspection and digital-twin technology on at least 18 vessels, beginning with a $54M initial award. | 高 | SP023, SP019 |
| CP037 | FieldAI became the inaugural corporate tenant at Carnegie Mellon University's Robotics Innovation Center in February 2026, occupying a 2,500-sq-ft lab and office suite at the Hazelwood Green facility alongside CMU's AirLab and LeCAR research groups. | 高 | SP028, SP001 |
| CI001 | FieldAI has raised $405M across two consecutive funding rounds announced simultaneously in August 2025. | 高 | SI001, SI012, SI005 |
| CI002 | FieldAI's most recent round raised $314M in August 2025, co-led by Bezos Expeditions, Prysm Capital, and Temasek, with participation from Khosla Ventures, NVentures, Intel Capital, Canaan Partners, BHP Ventures, and Emerson Collective. | 高 | SI003, SI004, SI019 |
| CI003 | FieldAI's first funding round raised approximately $91M (inferred from $405M total minus $314M most recent round) at a valuation of approximately $500M in 2024. | 中 | SI003, SI004, SI005 |
| CI004 | FieldAI's post-money valuation is $2B as of the August 2025 funding close, up from approximately $500M in the prior round. | 高 | SI004, SI005, SI006 |
| CI005 | Anonymous sources familiar with the matter told the Orange County Business Journal in April 2026 that FieldAI has more than $100M in booked revenue. | 中 | SI008 |
| CI006 | GetLatka's analyst platform estimates FieldAI's annual revenue at $140M as of November 2025, with 150 employees. | 低 | SI010 |
| CI007 | FieldAI's revenue model combines a one-time hardware integration fee when payloads are deployed on existing robots and a recurring annual software licensing fee for Field Foundation Model access and fleet analytics. | 中 | SI009, SI001 |
| CI008 | Third-party industry sources report FieldAI's subscription pricing ranges from tens of thousands to $500,000 per year depending on deployment scale; CEO Agha cited multi-million-dollar contracts in the US, Europe, and Asia. | 低 | SI009, SI004, SI013 |
| CI009 | FieldAI had approximately 30 employees at the end of 2024, grew to approximately 130 by the August 2025 funding announcement, and planned to double to approximately 260 by the end of 2025. | 中 | SI004, SI007 |
| CI010 | FieldAI's 2025 funding announcement stated capital will be used for global expansion, product development (locomotion, manipulation), and aggressive hiring to double headcount. | 中 | SI001, SI007 |
| CI011 | FieldAI's most recent funding round was oversubscribed, indicating investor demand exceeded the amount raised. | 中 | SI001, SI003 |
| CI012 | FieldAI became the inaugural corporate tenant at CMU's Robotics Innovation Center at Hazelwood Green in early 2026, providing access to high-bay labs and outdoor robot testing terrain. | 中 | SI024 |
| CI013 | FieldAI's capital efficiency ratio (post-money valuation / total capital raised) is approximately 3.95x–5x ($2B / $405M), which is premium relative to capital-intensive robotics hardware peers but aggressive relative to verified ARR. | 中 | SI011, SI005 |
| CI014 | Based on a headcount trajectory of 200–260 employees at fully-loaded cost of $250K– $350K/year, FieldAI's annualized labor expense is estimated at $50–90M/year, implying a monthly burn rate of approximately $5–10M inclusive of infrastructure and operations. | 低 | SI004, SI007, SI009 |
| CI015 | FieldAI's legal entity "Field Ai, Inc." was incorporated in California (formed in Delaware) on October 24, 2024, with document number 6436098; CEO/CFO/Secretary/Director is Aliakbar Aghamohammadi; status is active as of March 2026. | 中 | SI002, SI023 |
| CI016 | FieldAI's gross margin on software licensing is estimated at 65–80% based on enterprise AI software industry benchmarks; hardware integration services are estimated at 20–45% gross margin; blended margin is estimated at 55–75%. | 低 | SI009, SI013 |
| CI017 | CAC for enterprise industrial robotics software is estimated at $200K–$1M per customer given documented 12–24 month sales cycles, field demonstrations, and integration trials. | 低 | SI009, SI013 |
| CI018 | FieldAI's on-edge architecture (inference at <100ms on-device, no cloud dependency) reduces marginal variable cost per deployment compared to cloud-inference AI platforms, which is structurally favorable for gross margin at scale. | 中 | SI009 |
| CI019 | Revenue per employee at FieldAI is estimated at approximately $900K–$1M if GetLatka's $140M ARR estimate and 150-employee figure are both accurate, which would be strong productivity for an enterprise software company at this stage. | 低 | SI010, SI004 |
| CI020 | FieldAI's GTM motion includes direct enterprise sales, hardware-OEM channel partnerships (Boston Dynamics), and regional expansion partnerships (Certis Group), without disclosed channel revenue-share arrangements. | 中 | SI001, SI009 |
| CI021 | Enterprise industrial robotics software sales cycles are typically 12–24 months due to field demonstrations, integration testing, and multi-stakeholder procurement processes, resulting in structurally high CAC for FieldAI's primary market segments. | 中 | SI013, SI009 |
| CI022 | The $2B valuation implies an ARR multiple of approximately 14x against the $140M ARR estimate and approximately 20x against the $100M booked revenue signal; both multiples are aggressive for a pre-profitability AI company with undisclosed financials. | 中 | SI010, SI008, SI005 |
| CI023 | UpsideList's secondary-market analytical model estimates -2% upside for FieldAI at current implied valuation, flagging the company as pre-revenue from a secondary market perspective and noting 50–70% potential cumulative dilution over 3–5 more rounds. | 低 | SI015 |
| CI024 | An EDGAR full-text search for FieldAI or Field AI as of May 2026 finds no Form D filings directly attributable to FieldAI the robotics startup; the California entity "Field Ai, Inc." (document #6436098) is active but has no SEC Form D on file. | 中 | SI002, SI023 |
| CI025 | Revenue estimates for FieldAI range from approximately $5M (low analyst platforms) to $140M ARR (GetLatka), a 28x spread that reflects the fundamental challenge of estimating private-company financial metrics without audited data. | 中 | SI010, SI015, SI016 |
| CI026 | FieldAI has not publicly disclosed customer count, customer names, NRR, ARR breakdown by segment, gross margin, or monthly cash burn as of May 2026. | 高 | SI001, SI016 |
| CI027 | No public debt facility, revenue-based financing arrangement, or project finance obligation has been disclosed by FieldAI; the capital structure appears to be pure equity. | 中 | SI001, SI012 |
| CI028 | FieldAI's first patent application (US 2025/0252306) was filed February 5, 2025, covering uncertainty-aware traversability estimation; it is currently pending as of August 2025 publication. | 高 | SI017, SI025 |
| CI029 | FieldAI does not manufacture robots and carries no hardware inventory or manufacturing capex; this pure-software-and-services model eliminates working capital and supply-chain financing needs that capital-intensive robotics peers face. | 中 | SI009, SI001 |
| CI030 | FieldAI's $405M capital base combined with a pure-software operating model and an estimated $100M+ in cumulative revenue bookings implies at minimum 2–3 years of operating runway before the next fundraising event. | 低 | SI001, SI008, SI014 |
| CI031 | FieldAI's key diligence risk is information asymmetry, not near-term solvency; the gap between publicly verifiable financial metrics and what is required to underwrite the $2B valuation is unusually wide for a company at this funding level. | 中 | SI015, SI016, SI008 |
| CI032 | FieldAI's partnership with NVIDIA (Isaac platform) and CMU Robotics Innovation Center represents capital deployment into strategic R&D infrastructure rather than purely commercial spending. | 中 | SI018, SI024 |
| CI033 | FieldAI's patent application US 2025/0252306 reflects an IP strategy prioritizing risk-aware robotics navigation, consistent with the company's core FFM differentiation and signaling early IP investment as an R&D capital use. | 中 | SI017, SI025 |
| CI034 | Robotics Press reports FieldAI with 56% data completeness (11 sources, updated March 2026), indicating limited independent coverage and public information relative to comparable companies at comparable funding levels. | 中 | SI016 |
| CI035 | FieldAI's Series A round came at a 4x valuation markup from $500M to $2B in approximately 12 months, consistent with aggressive AI investment multiples in 2024–2025 but implying high investor return expectations embedded in the $2B price. | 中 | SI004, SI005, SI011 |
| CE001 | FieldAI's core product is the Field Foundation Model (FFM) platform — physics-first, risk-aware embodied AI software purpose-built for industrial robot autonomy. | 高 | SE001, SE007 |
| CE002 | FFMs natively accept multimodal inputs including vision, LiDAR, text, and audio to build a unified environmental belief-state. | 中 | SE001 |
| CE003 | FFMs are deployed entirely on-edge on the robot hardware with no cloud connectivity required for autonomous operation. | 高 | SE001, SE007, SE002 |
| CE004 | The Dynamics Foundation Model (DFM) integrates the robot's intrinsic kinematic and dynamic models into the cognitive layer, enabling the same FFM brain to operate across mechanically distinct robot platforms. | 中 | SE001 |
| CE005 | The Multiagent Foundation Model (MFM) enables multiple robots in a fleet to share environmental representations and reason collectively, acting as a distributed autonomous system at site scale. | 中 | SE001, SE002 |
| CE006 | FieldAI's "Context over Training" principle allows FFMs to make inferences about unforeseen scenarios without exhaustive pre-training for every possible situation. | 高 | SE001, SE003 |
| CE007 | FFMs enable robots to navigate and operate without pre-mapped layouts, GPS, fixed paths, or additional robot infrastructure. | 高 | SE001, SE007, SE011 |
| CE008 | The FFM architecture has been deployed across quadrupeds, humanoids, wheeled robots, and passenger-scale autonomous vehicles per FieldAI's official claims. | 中 | SE007, SE001 |
| CE009 | FieldAI's official Solutions page lists eight application categories: site mapping, facility inspection, condition anomaly detection, large-scale data capture, unmanned material transport, security and threat detection, telepresence/teleoperation, and search & ISR. | 高 | SE010, SE019 |
| CE010 | Boston Dynamics and FieldAI announced a partnership to advance robotics in construction and other complex environments on March 12, 2026. | 高 | SE002, SE006 |
| CE011 | The FieldAI-Boston Dynamics Spot integration reduces inspection and documentation time by more than 90% compared to manual processes, per the joint partnership announcement. | 中 | SE002 |
| CE012 | Certis Group and FieldAI announced a strategic partnership on February 23, 2026 to deploy autonomous robotics across large-scale, multi-site security operations globally. | 中 | SE011, SE018 |
| CE013 | Certis integrates FieldAI's autonomy technology with its proprietary Mozart orchestration platform, which coordinates robots, human teams, workflows, and command systems. | 中 | SE011 |
| CE014 | FieldAI deepened its collaboration with NVIDIA Omniverse in March 2026 to accelerate its operational data flywheel and improve digital twin generation for industrial customers. | 高 | SE004, SE015 |
| CE015 | FieldAI uses NVIDIA Omniverse NuRec to transform raw sensor data captured during routine robot operations into high-fidelity 3D digital twins of customer sites. | 中 | SE004, SE016 |
| CE016 | FieldAI integrates NVIDIA Isaac Sim and Isaac Lab to train and validate robot policies in simulation-ready environments reconstructed from real deployment data. | 中 | SE004, SE016 |
| CE017 | FieldAI is adopting the NVIDIA Physical AI Data Factory Blueprint via Microsoft Azure, using NVIDIA Cosmos open-world foundation models and NVIDIA OSMO for synthetic data pipeline automation. | 中 | SE004 |
| CE018 | An unnamed senior executive at a global industrial manufacturing company reported that a process taking 3.5 months now takes 12 hours with FieldAI's technology — over 200x faster — per CEO Ali Agha's cited claim. | 低 | SE004, SE016 |
| CE019 | FieldAI became the inaugural corporate tenant at CMU's Robotics Innovation Center at Hazelwood Green in February 2026, occupying 2,500 sq ft of lab and office space. | 中 | SE008 |
| CE020 | FieldAI's GitHub organization (field-ai) maintains active forks of robotics libraries including rosbridge_suite (updated Apr 2, 2026), spconv (Mar 5, 2026), Spot SDK, and unitree_sdk2. | 中 | SE012 |
| CE021 | FieldAI's Field AI Research Institute (FAIRI) publishes peer-reviewed research, including the NeBula paper documenting the DARPA SubT CoSTAR team's autonomy approach. | 中 | SE013 |
| CE022 | Ali Agha's publication record includes RSS 2024 and ICRA 2024 papers on risk-aware AI decision-making and low-frequency sampling in Model Predictive Path Integral control. | 中 | SE014 |
| CE023 | FieldAI occupies 41,000+ sq ft of offices, labs, and warehouse in Irvine and expects to outgrow this space before the end of 2026. | 中 | SE009 |
| CE024 | FieldAI has documented deployments in North America, Europe, Japan, and Asia-Pacific through its Certis partnership, with robots operating at hundreds of industrial sites globally. | 中 | SE004, SE007 |
| CE025 | FieldAI raised $405M in 2025 with capital explicitly earmarked for product development across locomotion and manipulation capabilities, plus plans to double headcount by end of 2025. | 高 | SE007, SE003 |
| CE026 | When FFMs encounter unfamiliar conditions, the model adapts by slowing down or choosing more conservative paths — an acknowledged throughput cost that is a direct consequence of the safety-first design. | 中 | SE016 |
| CE027 | FieldAI's Squarespace release notes page returns HTTP 401 Unauthorized to external visitors and is effectively access-blocked from public review. | 中 | SE022 |
| CE028 | No public API documentation, proprietary SDK, or developer portal has been identified for FieldAI's FFM platform as of May 2026; GitHub activity is limited to forks of third-party robotics libraries. | 中 | |
| CE029 | FieldAI's three-phase roadmap (per analyst coverage) positions inspection/monitoring for now through 2026, manipulation/intervention for 2026–2028, and general-purpose autonomy for 2028 and beyond. | 低 | SE005 |
| CE030 | The FFM architecture decouples the world model from the robot's physical dynamics model (DFM), enabling the same cognitive intelligence to operate across multiple robot morphologies. | 中 | SE001, SE007 |
| CE031 | No publicly disclosed safety certifications (ISO 10218, IEC 62443, ANSI/RIA), FedRAMP status, or independent compliance standards have been published for FieldAI's platform as of May 2026. | 中 | |
| CE032 | FieldAI plans to expand its Spot fleet to become "one of the largest third-party quadruped fleets in the world" through the Boston Dynamics partnership, with fleet expansion underway. | 中 | SE002 |
| CE033 | FieldAI differentiates its FFM from competitors by designing physics constraints and uncertainty management into the architecture from inception, rather than retrofitting large language or vision models for robotics. | 中 | SE003, SE007 |
| CE034 | CEO Ali Agha stated that customers can define a risk threshold for robot behavior, making the safety conservatism level configurable per deployment context. | 中 | SE003 |
| CE035 | FieldAI has offices in Tokyo, Singapore, San Francisco, Boston, and Pittsburgh in addition to its Irvine headquarters. | 中 | SE009, SE008 |
| CE036 | FieldAI sponsors research at CMU AirLab and LeCAR lab, and supports VectorRobotics — a student capstone team working on humanoid loco-manipulation for the MSRSD program. | 中 | SE008 |
| CE037 | CEO Ali Agha stated in the March 2026 Boston Dynamics announcement that the company is "at the inflection point where large fleet-scale deployment becomes possible." | 中 | SE002 |
| CE038 | Big-D Construction has deployed FieldAI robots for over two years and expanded to multiple projects in April 2026, with one catch reportedly saving $1.2 million through early defect detection. | 中 | SE017 |
| CU001 | FieldAI's stated commercial segments are construction, energy/oil & gas, mining, manufacturing, urban delivery and inspection, and federal/defense. | 高 | SU015, SU022, SU023 |
| CU002 | The typical FieldAI buyer persona is an enterprise technology or innovation leader — VP of Operations, Chief Strategy Officer, or VDC Director — while the payer is the enterprise organization itself. | 中 | SU001, SU021 |
| CU003 | The daily user of FieldAI systems is the field superintendent or safety manager who deploys and monitors the robot on-site; Big-D superintendent Bronson Dupaix and DPR superintendent Justin Schreiner are the named examples. | 高 | SU001, SU009 |
| CU004 | FieldAI subscription pricing is reported to range from tens of thousands of dollars to $500,000 per year depending on deployment scale, with hardware integration services added for first-time deployments. | 中 | SU008 |
| CU005 | FieldAI does not manufacture robots; it sells software and sensor-compute payload retrofits to enterprise buyers who own or separately procure robot hardware from OEM partners. | 高 | SU015, SU008 |
| CU006 | Construction is FieldAI's earliest and most commercially mature segment, with the only publicly named production customer case studies and deployments running continuously since 2024. | 高 | SU012, SU001, SU009 |
| CU007 | FieldAI's energy/oil & gas and industrial manufacturing solution pages describe autonomous equipment inspection, gauge and thermal monitoring, and digital-twin creation, but no named enterprise customer in these segments has been publicly confirmed. | 高 | SU022, SU023 |
| CU008 | Security operations entered FieldAI's addressable market via the Certis Group strategic partnership announced February 2026, targeting multi-site security deployments globally. | 高 | SU007, SU006 |
| CU009 | FieldAI Federal operates as a separately branded segment for federal and defense applications, with a Detroit Defense partnership referenced in prior chapters; no government contract award has been publicly confirmed. | 中 | SU025, SU015 |
| CU010 | FieldAI claimed "successful testing and deployments across hundreds of complex real-world industrial environments" as of the August 2025 funding announcement; this is a company assertion and has not been independently verified. | 低 | SU015, SU014 |
| CU011 | FieldAI customer deployments span Japan, Europe, and the United States as of August 2025 per company announcement; as of March 2026 the NVIDIA collaboration announcement cited customers across North America, Europe, and Asia. | 中 | SU015, SU005 |
| CU012 | The August 2025 funding round was described as oversubscribed following "rapid customer adoption and multiple expansion contracts," providing financial-signal corroboration that existing customers renewed or expanded; specific accounts are unnamed. | 中 | SU015, SU014 |
| CU013 | Sacra's independent business profile estimates FieldAI's B2B model as hardware integration services plus software licensing, with subscriptions ranging from tens of thousands to $500,000 per year depending on deployment scale. | 中 | SU008 |
| CU014 | Independent analyst robotsinconstruction.com tracked exactly four confirmed FieldAI construction deployments as of April 2026: two named (Big-D, DPR) and two unnamed top-10 ENR firms; the platform has been tracking since the product began shipping in 2024. | 高 | SU012, SU021 |
| CU015 | The Big-D Construction deployment has been continuously active for more than two years as of April 2026, with the company explicitly expanding to multiple projects across its portfolio. | 高 | SU001, SU002 |
| CU016 | The DPR Construction deployment at the Santa Clara data center was running approximately 1.5 years as of November 2025 per independent media reports. | 中 | SU009, SU010 |
| CU017 | As of March 2026, FieldAI described its fleet as "rapidly growing" with customers expanding across North America, Europe, and Asia in the NVIDIA collaboration announcement. | 低 | SU004, SU005 |
| CU018 | Boston Dynamics announced plans to expand the FieldAI-Spot deployment to create "one of the largest third-party quadruped fleets in the world" as part of the March 2026 partnership. | 中 | SU003, SU016 |
| CU019 | Big-D Construction is a named, confirmed production customer with over two years of active deployment across multiple construction jobsites in the United States. | 高 | SU001, SU002, SU012 |
| CU020 | Shaun Orr, a C-level executive and 25-year veteran of Big-D Construction, stated in April 2026 that "every project will have some representation of FieldAI tools," indicating commitment to fleet-wide adoption. | 高 | SU001, SU002 |
| CU021 | Big-D superintendent Bronson Dupaix described eliminating five-hour daily job-walk sessions using handheld cameras and scanners, replaced by autonomous robot missions. | 高 | SU001, SU002 |
| CU022 | Big-D VDC director Chantelle Menlove reported that a superintendent who had used the robot on one site immediately requested it for his next project, demonstrating organic bottom-up demand within the Big-D organization. | 高 | SU001, SU002 |
| CU023 | DPR Construction, one of the ten largest US contractors, is a named production customer with approximately 1.5 years of active deployment confirmed at a data center site in Santa Clara, California. | 中 | SU009, SU010, SU012 |
| CU024 | DPR superintendent Justin Schreiner stated publicly that the FieldAI system "makes us better at what we do" by improving efficiency, documentation, and freeing teams for critical tasks. | 中 | SU009, SU010 |
| CU025 | The DPR Santa Clara deployment captured 45,000+ photos, the robot walked 100+ miles, scanned 500,000 square feet of interiors, and documented 125,000 square feet of roofing. | 中 | SU009, SU011 |
| CU026 | Certis Group, Singapore's largest security and operations company employing services across a 30M+ worker global security industry, announced a strategic partnership with FieldAI in February 2026 for autonomous security robot deployment. | 高 | SU007, SU006 |
| CU027 | Certis' Mozart orchestration platform will integrate with FieldAI's FFMs for autonomous patrol, real-time incident detection, remote supervision, and coordinated human-robot response across public infrastructure, transport hubs, and commercial facilities. | 中 | SU007 |
| CU028 | Boston Dynamics cited an inspection and documentation time reduction of more than 90% compared to manual processes as a measured benefit from FieldAI-powered Spot deployments in construction. | 中 | SU003, SU016 |
| CU029 | Two unnamed top-10 ENR general contractor firms appear in the robotsinconstruction.com independent deployment tracker as confirmed production users of FieldAI technology. | 中 | SU012 |
| CU030 | An unnamed global industrial manufacturing executive was quoted in FieldAI's March 2026 NVIDIA press release saying a process that "once took three and a half months now takes just twelve hours" — a 200× improvement attributed to FieldAI technology. | 低 | SU004, SU005 |
| CU031 | FieldAI has not disclosed NRR, GRR, churn rate, cohort retention data, contract renewal rate, or any customer satisfaction score in any public filing, press release, or investor communication as of May 2026. | 中 | SU013, SU008 |
| CU032 | Big-D Construction's expansion from pilot to multi-project production and its stated plans for FieldAI tools on every future project are the strongest qualitative signal of customer retention and expansion, though no formal renewal or contract-value metric has been disclosed. | 中 | SU001, SU002 |
| CU033 | Big-D Construction plans to deploy FieldAI tools on every future project, indicating land-and-expand behavior driven by bottom-up demand from project teams. | 高 | SU001, SU002 |
| CU034 | FieldAI's data flywheel — where each deployment enriches foundation models through real-world sensor data — creates compounding switching costs by making each customer's models more accurate and site-specific over time. | 中 | SU015, SU005 |
| CU035 | The BuiltWorlds 2025 Equipment & Robotics Benchmarking report found that active construction robotics usage among contractors dropped from 65% in 2024 to 46% in 2025, a 19-percentage- point decline despite a rise in positive sentiment to 95%. | 高 | SU018, SU019, SU020 |
| CU036 | BuiltWorlds attributed the 2025 construction robotics adoption decline to a shift away from widespread experimental pilots toward more selective, proven implementations on smaller scales. | 高 | SU018, SU021 |
| CU037 | Robotics.press noted in April 2026 that FieldAI has no publicly disclosed revenue, ARR, contract values, or retention metrics, creating a meaningful gap between narrative and verifiable traction at a $2B valuation. | 中 | SU013 |
| CU038 | CNBC reported in March 2026 that autonomous AI systems deployed in real-world operations face "silent failure at scale" risks — where errors compound undetected over weeks — a systemic concern applicable to FieldAI's industrial deployments. | 中 | SU017 |
| CU039 | Robotics.press found that very few named, quantified FieldAI customer case studies exist in the public domain, and most deployment claims originate from company materials without third-party verification of intervention rates, uptime, or ROI. | 中 | SU013 |
| CU040 | Based on publicly confirmed data, FieldAI's construction customer base as of May 2026 consists of four accounts — two named (Big-D, DPR) and two unnamed ENR firms — meaning the loss of either named anchor customer would remove over half the visible evidence base. | 中 | SU012, SU013 |
| CU041 | The Certis Group partnership provides potential geographic diversification into APAC and European security markets via a channel-operator model rather than direct enterprise sales. | 中 | SU007, SU006 |
| CU042 | FieldAI's primary deployment platform at confirmed customer sites is the Boston Dynamics Spot quadruped robot; humanoid and wheeled platforms are referenced in company materials but no humanoid-specific customer deployment has been independently confirmed. | 高 | SU012, SU003 |
| CR001 | Colorado's AI Act (SB 24-205), the first comprehensive US state statute targeting high-risk AI systems, takes effect June 30, 2026, requiring impact assessments, consumer disclosures, and reasonable care to prevent algorithmic discrimination. | 高 | SR002, SR004 |
| CR002 | The EU AI Act's obligations for high-risk AI systems — including risk management, technical documentation, CE marking, and human oversight — were originally due by August 2, 2026, subject to a Digital Omnibus proposal that could extend the deadline to December 2027 if adopted before August 2026. | 高 | SR002, SR004 |
| CR003 | California SB 53 (Transparency in Frontier AI Act), effective January 2026, requires developers of large AI models (training compute >10²⁶ FLOPs) to publish risk frameworks, report critical safety incidents within 15 days, and implement whistleblower protections, with penalties up to $1 million per violation. | 高 | SR002, SR004 |
| CR004 | FieldAI's industrial AI autonomy systems deployed in EU markets likely qualify as high-risk AI under the EU AI Act's Annex III classification (safety components of products in critical infrastructure and industrial settings), triggering conformity assessment, CE marking, and post-market monitoring obligations. | 中 | SR004, SR002 |
| CR005 | No public regulatory enforcement actions, government investigations, or regulatory fines against FieldAI have been identified as of May 2026. | 中 | SR012, SR029 |
| CR006 | AI product liability litigation is consolidating around product-liability doctrine in US courts, treating deployed AI systems as products subject to design-defect, failure-to-warn, and foreseeable-misuse theories. | 高 | SR001, SR010 |
| CR007 | AI-related lawsuits increased 137% in 2025, with autonomous AI agents presenting the highest litigation risk; median claim sizes around $5M, with the largest claims (top 5%) accounting for most exposure. | 中 | SR001, SR010 |
| CR008 | The EU revised Product Liability Directive (PLD) treats software — including AI systems — as 'products' and extends strict liability concepts across the distribution chain, capturing parties that substantially modify AI systems; member states must transpose by December 2026. | 高 | SR001, SR002 |
| CR009 | No public lawsuits, IP disputes, or legal proceedings involving FieldAI as either plaintiff or defendant have been identified as of May 2026. | 中 | SR012, SR029 |
| CR010 | Product liability doctrine can extend upstream to AI component providers: Garcia v. Character Technologies allowed theories against 'upstream technology provider and the manufacturer' to proceed at the pleading stage, a precedent applicable to FieldAI as an AI software layer provider. | 高 | SR001, SR010 |
| CR011 | FieldAI Federal, the company's defense-oriented subsidiary, is operational and would be subject to ITAR and EAR export controls if its autonomous robotics systems cross into dual-use military technology classifications; no public ITAR authorisation documentation has been found. | 中 | SR026, SR029 |
| CR012 | USPTO patent search finds one published application (US 2025/0252306) associated with FieldAI as of May 2026 but no granted patents; IP protection relies primarily on trade secrets and publication velocity rather than granted patent moats. | 中 | SR030, SR026 |
| CR013 | OSHA and NIOSH document a persistent baseline of serious robot-related workplace injuries, including crushing, striking, and pinning incidents, often occurring during unexpected startup or AI misjudgment of human proximity. | 高 | SR006, SR007 |
| CR014 | Industrial safety standards for robots near humans (ISO 10218, RIA R15.06, IEC 61508) identify unexpected robot movements, software failures, poorly defined autonomous operation boundaries, and incomplete human-robot interaction protocols as primary risk factors. | 高 | SR007, SR006 |
| CR015 | FieldAI's robots operate entirely on-edge without GPS, pre-mapped environments, or cloud connectivity; on-device inference is the sole decision-making authority, meaning AI model failures manifest immediately as physical robot actions with no human-in-the-loop override by default. | 中 | SR026, SR013 |
| CR016 | No public safety incidents involving FieldAI-powered robots, no OSHA complaints citing FieldAI, and no customer safety-related disclosures have been identified as of May 2026. | 中 | SR029, SR012 |
| CR017 | FieldAI's construction, mining, and energy deployments place robots in the environments most likely to involve unexpected obstacles, human co-presence, and high-consequence failure modes (falls into excavations, equipment collisions, explosive atmospheres). | 中 | SR013, SR011 |
| CR018 | A 2025 Unitree H1 humanoid robot malfunction — flailing violently near a worker after software failure — and CES 2026 safety expert observations confirm that close-quarters human-robot interaction in autonomous mode is an unresolved safety frontier for the industry. | 中 | SR005, SR007 |
| CR019 | No ISO 10218, IEC 62443, SOC 2, CE marking, or equivalent third-party safety or security certification for FieldAI's FFM platform or edge compute payload has been found in any public source as of May 2026. | 中 | SR029, SR026 |
| CR020 | FieldAI has not published performance benchmarks, safety validation test reports, or failure-rate data for its FFMs under real-world industrial conditions; the absence prevents independent verification of the platform's safety profile. | 中 | SR028, SR026 |
| CR021 | The 2026 International AI Safety Report identifies rapid advancements in robot autonomy and 'corresponding emergent risks — including loss-of-control scenarios and inadequate real-world reliability testing' as areas requiring stronger regulatory frameworks and open incident reporting. | 高 | SR003, SR005 |
| CR022 | No public DPA, security whitepaper, penetration test disclosure, or IEC 62443 certification for FieldAI's cloud fleet analytics platform or NVIDIA pipeline integration has been identified as of May 2026. | 中 | SR029, SR018 |
| CR023 | FieldAI's Boston Dynamics partnership (announced March 12, 2026) combines FieldAI's FFMs with Spot quadruped hardware for construction and other complex environments, creating a deep bidirectional integration dependency. | 高 | SR013, SR014 |
| CR024 | The terms of the Boston Dynamics–FieldAI partnership, including exclusivity provisions, revenue sharing, minimum commitments, and termination rights, are not publicly disclosed. | 中 | SR013, SR020 |
| CR025 | Hyundai Motor Group, which owns Boston Dynamics, made a strategic investment in FieldAI in February 2026, creating a dual relationship (hardware partner + investor parent) that could produce conflicts of interest if Hyundai pursues a proprietary autonomy strategy for Boston Dynamics hardware. | 中 | SR008, SR019 |
| CR026 | FieldAI uses NVIDIA Omniverse NuRec for 3D reconstruction, Isaac Sim/Isaac Lab for simulation training, NVIDIA Cosmos for synthetic data generation, and OSMO for pipeline automation — creating a multi-layer technical dependency across NVIDIA's product stack. | 高 | SR018, SR026 |
| CR027 | GPU supply constraints in 2026 include HBM memory bottlenecks and 9–12 month lead times for data-centre GPUs; cloud hyperscalers capture most new GPU supply, squeezing availability for smaller AI firms without multi-year cloud commitments. | 中 | SR009, SR004 |
| CR028 | NVentures (NVIDIA's corporate venture arm) is a financial investor in FieldAI, reducing the probability of abrupt compute supply cutoff, but not eliminating pricing, API deprecation, or competitive-priority risks if NVIDIA pivots its strategy. | 中 | SR023, SR018 |
| CR029 | Certis Group's partnership with FieldAI (February 2026) introduces dependency on the Mozart fleet orchestration platform for security vertical deployments; platform incompatibility or Certis strategy change could limit security-vertical revenue. | 中 | SR025, SR029 |
| CR030 | FieldAI's FFM platform is marketed as hardware-agnostic, supporting quadrupeds, humanoids, wheeled robots, and passenger-scale vehicles, reducing single-OEM dependency at the platform layer, though named commercial partnerships are concentrated in Boston Dynamics (construction) and Certis (security). | 中 | SR026, SR013 |
| CR031 | FieldAI's SDK integration with Boston Dynamics Spot uses a custom Spot SDK fork (publicly visible on GitHub), making the integration a high-fidelity but tightly coupled dependency where Spot SDK version changes require FieldAI engineering response. | 中 | SR014, SR020 |
| CR032 | No evidence of diversified cloud provider contracts, alternative GPU sourcing agreements, or NVIDIA-independent training infrastructure at FieldAI has been identified; NVIDIA/cloud compute dependency appears concentrated. | 中 | SR018, SR009 |
| CR033 | FieldAI was founded in 2023 by Ali Agha, a former NASA JPL Principal Investigator and DARPA Subterranean Challenge Urban-circuit winner, and had grown to 130+ employees by August 2025. | 高 | SR022, SR023 |
| CR034 | FieldAI's disclosed executive team includes David Fan (CTO), Shayegan Omidshafiei (President & CSO), Duncan McIntyre (CFO), and Patrick Purwin (VP Sales), providing depth below the CEO level, though none has equivalent public brand recognition to Ali Agha. | 中 | SR021, SR024 |
| CR035 | No evidence of executive departures, organisational restructuring, or named leadership controversy at FieldAI has been identified as of May 2026. | 中 | SR029, SR012 |
| CR036 | FieldAI was in stealth until August 2025; as of May 2026, the company has less than one year of demonstrated enterprise sales capacity at scale, which is insufficient to confirm that its commercial execution muscle matches its technical capabilities. | 中 | SR022, SR017 |
| CR037 | Headcount growth from zero to 130+ employees in approximately two years (2023–2025) introduces cultural, coordination, and product-execution risk typical of rapidly scaling deep-tech startups. | 中 | SR022, SR023 |
| CR038 | FieldAI competes for senior robotics-AI engineers with Google Intrinsic (now integrated into Google AI), Physical Intelligence, Amazon Robotics, and Tesla Optimus — all of whom are hiring from the same narrow pool of PhDs and practitioners who can work on embodied AI at scale. | 中 | SR022, SR027 |
| CR039 | No evidence of post-funding key technical departures from FieldAI has been identified as of May 2026, though the absence of public evidence is not confirmation of stability. | 低 | SR029, SR021 |
| CR040 | Ali Agha is FieldAI's most prominent public figure, primary technical spokesperson (Robot Report interview, RoboBusiness keynote), and the originator of the DARPA SubT / NASA JPL pedigree that underpins the company's investor and customer credibility; no co-equal public-facing leader exists. | 高 | SR028, SR024 |
| CR041 | FieldAI raised $405M at a $2B post-money valuation across two consecutive rounds (Series A and A1) in 2025, making it the highest single-year valuation achieved by a robotics AI software company at Series A stage by public record. | 高 | SR015, SR016 |
| CR042 | The $2B valuation implies an approximate 20× multiple on the '$100M+ booked revenue' signal from anonymous OCBJ April 2026 sources, or approximately 14× against GetLatka's $140M ARR estimate (November 2025); both multiples are at the top of the enterprise AI software range and require sustained high-velocity growth to be defensible. | 低 | SR017, SR027 |
| CR043 | FieldAI has not publicly disclosed gross margin, NRR, CAC, burn rate, cash on hand, or any other financial metric; all valuation underwriting must rely on unaudited third-party revenue estimates and analogy to comparable deeptech companies. | 中 | SR027, SR019 |
| CR044 | FieldAI's capital adequacy is estimated at 2–4 years from the $405M raise at inferred burn rates consistent with 130+ employees at deep-tech compensation levels ($10–20M/month), with no public confirmation of cash deployment pace. | 低 | SR022, SR019 |
| CR045 | A sustained AI sector valuation correction, macroeconomic downturn, or failure to meet growth expectations could prevent future fundraising at the $2B entry valuation or higher, forcing a down round that activates investor preference protections at the expense of earlier holders. | 中 | SR016, SR027 |
| CR046 | FieldAI Federal's defence channel introduces government procurement risk: federal contracts have long procurement cycles (12–36 months), require security clearances and ITAR compliance, and can be cancelled or reduced by budget reallocation; no awarded federal contracts have been publicly disclosed. | 中 | SR026, SR029 |
| CR047 | GetLatka estimates FieldAI's ARR at $140M as of November 2025; the estimate is unconfirmed by the company, unaudited, and derived from proprietary third-party modelling that cannot be independently verified. | 低 | SR027, SR017 |
| CR048 | Revenue concentration among a small number of early customers is probable given FieldAI's two-year age and nascent commercial stage; the company has not disclosed customer count, top-10 customer revenue concentration, or ARR distribution. | 中 | SR027, SR012 |
| CR049 | FieldAI has one published US patent application (US 2025/0252306) and no granted US patents as of May 2026; IP defence relies primarily on trade secrets, model weights as proprietary assets, and speed of deployment data accumulation. | 中 | SR030, SR026 |
| CR050 | The 'hundreds of sites on three continents' customer-deployment claim is company-sourced and unverified by any independent third party; the absence of named customers or third-party deployment confirmation means the claim cannot be used to corroborate commercial traction. | 低 | |
| CV001 | FieldAI's post-money valuation was $2.0 billion following the August 2025 funding announcement, up from approximately $500 million in the preceding round. | 高 | SV001, SV004, SV006 |
| CV002 | FieldAI raised $405 million across two consecutive rounds: approximately $91 million in an earlier round (at ~$500M valuation) and $314 million in the August 2025 oversubscribed round. | 高 | SV002, SV003, SV005 |
| CV003 | GeekWire and Axios described the two FieldAI rounds as 'Series A and A1'; the company itself describes them as two consecutive rounds without publishing the formal round designation. | 中 | SV003, SV002 |
| CV004 | PremierAlts reported FieldAI total funding as $506.1 million across five rounds as of February 22, 2026, with the most recent round of $315 million closed in February 2026—suggesting possible additional capital beyond the August 2025 announcement. | 中 | SV009 |
| CV005 | SEC EDGAR full-text search for 'Field AI' Form D filings returned three results: HII Field AI Series I, II, and III—all structured as pooled investment fund series of HII Field AI, LLC—filed November 2025 and April 2026; no Form D filed directly by Field Ai, Inc. | 高 | SV020, SV019 |
| CV006 | FieldAI shares are actively listed on secondary markets including Nasdaq Private Market and Notice.co, with a secondary share price of approximately $34.32 as of May 2026. | 中 | SV021, SV022 |
| CV007 | FieldAI's LinkedIn employee count was reported at 201–500 as of early 2026, consistent with the company's stated plan to double headcount from 130 (August 2025) to approximately 260 by end of 2025. | 中 | SV007, SV002 |
| CV008 | FieldAI's capital efficiency ratio (valuation / total raised) is approximately 4–5x ($2B / $405M), consistent with high-growth AI software premium multiples for this stage. | 中 | SV009, SV001 |
| CV009 | FieldAI's founding team has exceptional field robotics credentials: CEO Ali Agha led the JPL/MIT CoSTAR team that won the DARPA Subterranean Challenge Urban Circuit, and CSO Shayegan Omidshafiei was a research scientist at DeepMind/Google. | 高 | SV002, SV005, SV029 |
| CV010 | The FieldAI Series A1 round was oversubscribed, with lead investors including Bezos Expeditions (Jeff Bezos' family office), Prysm Capital, and Temasek, and participation from Khosla Ventures, NVentures (NVIDIA), Intel Capital, Canaan Partners, BHP Ventures, and Emerson Collective. | 高 | SV001, SV005, SV032 |
| CV011 | Vinod Khosla publicly stated: 'FieldAI is at the forefront of the general-purpose robotics revolution, and its ability to rapidly deploy will unlock long-term economic and societal value.' | 高 | SV005, SV006 |
| CV012 | FieldAI CEO Ali Agha stated the company has multi-million-dollar contracts in the US, Europe, and Asia, and the company claims deployments across hundreds of sites on three continents. | 中 | SV004, SV005 |
| CV013 | A third-party analytics platform (GetLatka, November 2025) estimated FieldAI ARR at $140 million; a secondary market data source (CompWorth) estimated revenue at approximately $5 million—a 28x discrepancy reflecting the challenge of estimating private-company metrics externally. | 低 | SV008, SV023 |
| CV014 | Anonymous sources familiar with FieldAI's finances told the Orange County Business Journal (April 2026) that the company has more than $100 million in booked revenue. | 低 | SV007 |
| CV015 | No audited or company-confirmed revenue, gross margin, NRR, or cash position has been publicly disclosed by FieldAI; all financial estimates are third-party, anonymous, or analyst-estimated. | 中 | SV007, SV008, SV013 |
| CV016 | First Round Capital and B Capital chairman Howard Morgan publicly stated AI startup valuations are overheated, warning that 'buy high, sell higher works only inside a bubble.' | 中 | SV024, SV025 |
| CV017 | Physical Intelligence was in talks for a new round valuing the company above $11 billion as of March 2026, doubling its $5.6 billion valuation from November 2025, despite having no disclosed revenue or commercialization timeline. | 高 | SV015, SV016, SV030 |
| CV018 | Skild AI raised a $1.4 billion Series C round in January 2026 at a $14 billion+ valuation; the company reported approximately $30 million in ARR grown from zero in just a few months of 2025. | 中 | SV012, SV026 |
| CV019 | Apptronik raised $935 million total across its Series A rounds by February 2026, at a post-money valuation of approximately $5.3 billion, with investors including Google, Mercedes-Benz, and B Capital. | 高 | SV013, SV014 |
| CV020 | Gecko Robotics raised $125 million in a June 2025 Series D, reaching a $1.25 billion valuation—roughly double its $633 million Series C valuation from December 2023. | 高 | SV017, SV018 |
| CV021 | Figure AI was reported to have raised a Series C round exceeding $1 billion at approximately $39 billion valuation in September 2025, making it the highest-valued humanoid robotics startup. | 中 | SV023, SV016 |
| CV022 | Houlihan Lokey's Q1 2026 AI in Vertical Software report found that AI-native software companies commanded a median EV/Revenue multiple of 21.2x in VC rounds, versus 8.5x for AI-enabled software and 5.5x for legacy SaaS. | 高 | SV011, SV010 |
| CV023 | At FieldAI's $2B valuation and $100–140M ARR estimates, the implied revenue multiple is approximately 14–20x—within the AI-native software median but requiring the higher end of the revenue range to be real. | 中 | SV011, SV007, SV008 |
| CV024 | At FieldAI's $2B valuation and the CompWorth low-end revenue estimate of approximately $5 million, the implied revenue multiple would be approximately 400x—comparable only to pre-revenue speculative investments in the sector. | 中 | SV023, SV011 |
| CV025 | In the bull case, FieldAI achieves verified ARR of $140M growing at 80–100% annually, reaches $350–500M ARR by a 2027 Series B, and raises at 15–20x ARR, implying a $5.25–10B valuation. | 低 | SV007, SV011 |
| CV026 | In the base case, FieldAI achieves $100M ARR growing 40–60% annually, reaches $200–300M ARR by a 2028 Series B, and raises at 10–12x ARR, implying a $2–3.6B valuation. | 低 | SV011, SV008 |
| CV027 | In the bear case, FieldAI's actual revenue is $20–50M ARR, sector multiples compress to 8–10x, and the next funding round is a down-round at $400M–$1B valuation. | 低 | SV024, SV025 |
| CV028 | Seattle-area venture capitalists stated in December 2025 that early-stage private AI valuations have run 'well ahead of fundamentals,' with capital chasing startups at valuations that price in outcomes requiring years of execution to justify. | 中 | SV025 |
| CV029 | If FieldAI's actual ARR is at the low end of estimates ($5–20M) rather than the $100–140M booked-revenue level, the implied multiple at $2B (100–400x) would be unjustifiable even by the most optimistic AI-native software benchmarks. | 中 | SV011, SV024 |
| CV030 | Physical Intelligence co-founder Lachy Groom stated the company has 'no timeline for commercialization,' illustrating that extreme optionality premiums in embodied-AI software are disconnected from current revenue for the entire peer group. | 中 | SV016 |
| CV031 | A Series B down-round at FieldAI (valuation below $2B) would be a thesis-break trigger, contradicting the private-investor conviction that supports the current $2B mark. | 中 | SV025, SV024 |
| CV032 | FieldAI's board composition, liquidation preferences, option pool size, and investor rights agreements have not been publicly disclosed, representing a significant governance opacity for a company at $2B valuation. | 中 | SV010, SV008 |
| CV033 | The three minimum threshold conditions to upgrade the recommendation from TRACK to BUY are: verified ARR ≥$100M, confirmed gross margin ≥60%, and net revenue retention ≥110%—none of which are publicly available. | 中 | SV011, SV007 |
| CV034 | Key commercial proof points that would provide positive evidence for the TRACK-to-BUY upgrade include: a publicly confirmed Series B step-up above $2B, a named large enterprise customer, a disclosed contract renewal or NRR figure, or an independent financial audit. | 中 | SV007, SV008 |
| CV035 | Customer concentration risk is unverifiable without access to FieldAI's customer schedule; if top-3 customers account for more than 50% of revenue, churn risk would materially change the base and bear scenario assumptions. | 低 | |
| CV036 | The external-investor recommendation of TRACK reflects the combination of high team quality, high investor-quality signal, and a valuation that is stretched but within the comp set—offset by complete lack of verified financial metrics available to non-insiders. | 中 | SV011, SV001, SV024 |
| CV037 | FieldAI's hardware-agnostic software model—deploying AI without manufacturing robots—means the company avoids capital-intensive hardware supply chain risks that affect Apptronik, Figure AI, and Agility Robotics, implying a higher sustainable gross margin floor. | 中 | SV002, SV013 |
| CV038 | FieldAI's two major strategic partnerships announced in early 2026—Boston Dynamics (March) and Certis Group (February)—provide corroborating evidence of commercial traction beyond the anonymous revenue disclosures. | 中 | SV008, SV027 |
| CV039 | Skild AI's $14B valuation on $30M ARR implies a ~467x revenue multiple; Physical Intelligence's $11B+ valuation implies effectively infinite multiple on zero revenue; these peer benchmarks contextualize FieldAI's 14–20x (if $100–140M ARR) as moderate within the embodied-AI sector. | 中 | SV012, SV015, SV026 |
| CV040 | The SEC EDGAR Form D filing for HII Field AI Series III (CIK 0002126354, filed April 3, 2026) confirms ongoing structured investment activity around FieldAI equity as recently as April 2026, consistent with active secondary market demand. | 高 | SV020, SV019 |
| CV041 | Intel Capital's public endorsement of the FieldAI round included a formal post formalizing a platform integration and strategic relationship with Intel's hardware stack, indicating at least one strategic investor is treating this as a partnership investment rather than pure financial. | 中 | SV032, SV001 |
| CV042 | The most plausible exit paths for FieldAI are: (1) strategic acquisition by a large industrial conglomerate or tech platform (Honeywell, Emerson, Amazon, NVIDIA); (2) IPO in 2028–2030 if revenue scales to $500M+; or (3) strategic recap/continuation fund. | 低 | SV001, SV009 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | FieldAI | Redefining Industrial AI — FieldAI Homepage | Leading the frontier of Physical AI with deployments across three continents. Built by veterans from DeepMind, NASA JPL, Tesla, NVIDIA, Amazon, and beyond. |
| SO002 | FieldAI | Team | FieldAI | With nearly two decades of expertise in pioneering AI and autonomy algorithms across diverse robotic platforms, Dr. Ali Agha is leading FieldAI's strategic vision and product development. |
| SO003 | FieldAI | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | Headquartered in Irvine, CA, FieldAI is a pioneer in developing embodied AI software that is redefining autonomous robot operations in real-world environments. |
| SO004 | PR Newswire | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | Irvine, CA – August 20, 2025 – FieldAI, a leader in physical AI and robotic autonomy, today announced that it has raised $405 million in two consecutive rounds. |
| SO005 | TechCrunch | FieldAI raises $405M to build universal robot brains | FieldAI, an Irvine, California-based robotics startup, has raised $405 million across multiple previously undisclosed rounds to develop what it calls 'foundational embodied AI models.' |
| SO006 | CNBC | Nvidia, Bill Gates-backed robotics startup Field AI hits $2 billion valuation after recent raise | Gates, Nvidia-backed robotics firm Field AI hits $2 billion valuation. |
| SO007 | GeekWire | Robotics startup FieldAI, backed by Gates and Bezos, hits $2B valuation after latest funding | The investment, which came in consecutive Series A and A1 rounds, values the 2-year-old startup at $2 billion according to reports by Axios and CNBC on Wednesday. |
| SO008 | Reuters (via Yahoo Finance) | Robotics startup FieldAI raises $314 million in new funding, sources say | The Irvine, California-based startup is now valued at $2 billion, up from $500 million in a round last year... It had about 30 people at the end of 2024 and grew to 130 this year, and plans to double headcount by the end of 2025. |
| SO009 | Sacra | FieldAI valuation, funding & news | FieldAI raised $405 million across multiple funding rounds, including a $314 million round in August 2025 co-led by Bezos Expeditions, Prysm Capital, and Temasek. |
| SO010 | Intel Capital | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | |
| SO011 | Boston Dynamics | Boston Dynamics & FieldAI Partner to Bring Robots Into Uncharted, Dynamic Environments | We've known Ali and his team since they won the Urban-circuit of the DARPA Subterranean Challenge using Spot robots, while tackling similar challenges for NASA. |
| SO012 | Certis Group | Certis and FieldAI Form Strategic Partnership to Deploy Autonomous Robotics in Real-World Security Operations | |
| SO013 | Construction Dive | Robotic software startup FieldAI lands $405M in fresh funding | The share of firms reporting active robotics use fell from 65% last year to 46% now [2025]. |
| SO014 | Orange County Business Journal | FieldAI Robotics Growing, Will Need More Space | People familiar with the matter tell the Business Journal that the company has more than $100 million in booked revenue and partnerships with large industrial customers. |
| SO015 | Orange County Business Journal | $2B-Valued FieldAI Moves to Irvine | Robotics startup FieldAI, Orange County's newest unicorn, has completed a lease for a new, larger headquarters in Irvine. |
| SO016 | Robotics & Automation News | FieldAI raises more than $400 million to advance embodied AI at scale | |
| SO017 | Robotics & Automation News | Boston Dynamics and FieldAI partner to bring robots into construction and other complex, dynamic environments | |
| SO018 | TechStartups | FieldAI raises $405M in funding at $2B valuation, backed by Bill Gates, Jeff Bezos, and Nvidia | Agha himself spent nearly a decade at NASA's Jet Propulsion Laboratory, working on robotics autonomy before launching FieldAI in Irvine, California. |
| SO019 | TechFundingNews | FieldAI raises $405M at $2B valuation to build smarter robot brains | |
| SO020 | xmaquina.io | How FieldAI Is Building Real-World Robot Intelligence | After earning his Ph.D. in Computer Science and Engineering from Texas A&M University, Agha joined MIT as a postdoctoral researcher. There, he began working closely with future FieldAI co-founder Shayegan Omidshafiei. |
| SO021 | Emerson Collective | Dr. Ali Agha | Ali has nearly two decades of experience leading cutting-edge autonomy projects. Prior to founding FieldAI, he was a Principal Investigator at NASA's Jet Propulsion Laboratory (JPL). |
| SO022 | Sacra | FieldAI valuation, funding & news — product detail | |
| SO023 | FieldAI via PR Newswire | FieldAI Accelerates Industrial Customers' Adoption of AI in Collaboration with NVIDIA | |
| SO024 | FieldAI (via Intel Capital) | FieldAI Announces Over $400M — Intel Capital repost with investor statement | |
| SO025 | GeekWire | FieldAI backing and Series A/A1 details | The investment, which came in consecutive Series A and A1 rounds, values the 2-year-old startup at $2 billion. |
| SM001 | International Federation of Robotics | World Robotics 2025 Report — Global Robot Demand in Factories Doubles Over 10 Years | 542,000 robots installed in 2024 – more than double the number 10 years ago. Annual installations topped 500,000 units for the fourth straight year. |
| SM002 | Global Market Insights | AI-Powered Industrial Robot Market Trends, 2026–2035 | The global AI-powered industrial robot market was estimated at USD 16.8 billion in 2025. The market is expected to grow from USD 17.9 billion in 2026 to USD 33.3 billion in 2035, at a CAGR of 7.1%. |
| SM003 | Grand View Research | Construction Robots Market Size, Industry Report, 2030 | The Global Construction Robots Market size was estimated at USD 1.4 billion in 2024 and is projected to reach USD 3.66 billion by 2030, growing at a CAGR of 18% from 2025 to 2030. |
| SM004 | FieldAI (via PR Newswire) | FieldAI Accelerates Industrial Customers' Adoption of AI in Collaboration with NVIDIA | A senior executive at a global industrial manufacturing company noted that what once took three and a half months for them now takes just twelve hours with FieldAI's technology, over 200 times faster than manual methods. |
| SM005 | Boston Dynamics | Case Study: Practical Autonomy on Construction Sites | Customers have reduced inspection and documentation time by more than 90 percent compared to manual processes and avoided millions of dollars in potential cost overruns by identifying issues earlier. |
| SM006 | Robotics and Automation News | Boston Dynamics and FieldAI partner to bring robots into construction and other complex, dynamic environments | |
| SM007 | MarkWide Research | Robotics in Oil and Gas Market Size, Share, and Industry Trends Forecast 2026–2036 | Market Size in 2026: $3.8 Billion. Market Size in 2035: $12.96 Billion. CAGR (2026–2036): 14.6%. |
| SM008 | CNBC | Gates, Nvidia-backed robotics firm Field AI hits $2 billion valuation after recent raise | |
| SM009 | Persistence Market Research | Mining Robotics Market Size, Share, and Growth Forecast 2026–2033 | The global mining robotics market size is likely to be valued at US$ 1.7 Billion in 2026 and is expected to reach US$ 3.3 Billion by 2033, growing at a CAGR of 9.8%. |
| SM010 | The Business Research Company | Industrial Artificial Intelligence Global Market Report 2026 | The industrial artificial intelligence market size is expected to grow from $9.06 billion in 2025 to $13.69 billion in 2026 at a CAGR of 51.1%. |
| SM011 | MineListings | Autonomous Equipment Revolutionizes Mining Operations in 2026 | Autonomous trucks reaching 3,832 units by mid-2025…integration of digital maintenance programs has resulted in a substantial 50% reduction in unplanned downtime. |
| SM012 | McKinsey & Company | Unlocking the Industrial Potential of Robotics and Automation | Participants report that the primary challenges to adoption include the capital cost of robots and a company's general lack of experience with automation, cited by 71 percent and 61 percent of respondents, respectively. |
| SM013 | Deloitte | AI for Industrial Robotics, Humanoid Robots, and Drones — TMT Predictions 2026 | Annual sales of new industrial robots have remained flat at roughly 500,000 units since 2021…unless the broader technology, AI, and robotics ecosystem address bottlenecks related to data quality, integration, and cyber security, the market for industrial robots is likely to stay at its current level of relatively modest annual growth. |
| SM014 | FieldAI | FieldAI — Redefining Industrial AI (Official Website) | Leading the frontier of Physical AI with deployments across three continents. |
| SM015 | Research and Markets | Embodied AI Market Report 2026 | |
| SM016 | Construction Dive | Robotic Software Startup FieldAI Lands $405M in Fresh Funding | The share of firms reporting active robotics use fell from 65% last year to 46% now…positive evaluations of the innovative equipment jumped from 74% in 2024 to over 95% this year. |
| SM017 | TechCrunch | FieldAI Raises $405M to Build Universal Robot Brains | Since launching the company in 2023, FieldAI has secured contracts across industries including construction, energy, and urban delivery. |
| SM018 | Robotics and Automation News | FieldAI Raises More Than $400 Million to Advance Embodied AI at Scale | Deployments span a variety of robot types in high-complexity environments from Japan, to Europe, to the US, with some of the world's largest companies in industries including construction, energy, manufacturing, urban delivery and inspection. |
| SM019 | Fortune Business Insights | Inspection Robotics in Oil & Gas Market Size, Share 2026–2034 | The global Inspection Robotics in Oil & Gas Market size was valued at USD 0.84 billion in 2025. The market is projected to grow from USD 0.93 billion in 2026 to USD 1.51 billion by 2034, exhibiting a CAGR of 6.23%. |
| SM020 | Mordor Intelligence | Oil & Gas Automation Market — Size, Share & Industry Analysis 2026–2031 | The oil & gas automation market size was valued at USD 43.35 billion in 2025 and estimated to grow from USD 46.16 billion in 2026 to reach USD 63.19 billion by 2031, at a CAGR of 6.48%. |
| SM021 | MDPI (Processes Journal) | Recent Advances and Challenges in Industrial Robotics: A Systematic Review of Technological Trends and Emerging Applications | |
| SM022 | Analytics Insight | Industrial Robotics in 2026: Is the Brain More Important Than the Machine? | Factories are beginning to select robots the way companies select cloud infrastructure: for compatibility with an intelligence ecosystem rather than for standalone performance. |
| SM023 | Robotnik Automation | Industrial Robotics in 2025: Trends, Figures, and Global Outlook | According to IFR figures, factories worldwide installed 542,076 industrial robots in 2024, a historic level that confirms the strength of global automation growth. |
| SM024 | The AI Insider | FieldAI Raises $405 Million to Scale Advanced Robotics Foundation Models | FieldAI develops a single 'software brain' that can power many different kinds of robots. Its systems are already in use at customer facilities in Japan, Europe, and the U.S. |
| SM025 | Automation.com | Global Robot Demand in Factories Doubles Over 10 Years | Globally, robot installations are expected to grow by 6% to 575,000 units in 2025. By 2028, the 700,000-unit mark will be surpassed. |
| SP001 | FieldAI | FieldAI — One Autonomy for All Robots (Homepage) | "One brain across robots, tasks, and environments. Proven global deployment. ONE BRAIN FOR ANY MACHINE." |
| SP002 | Boston Dynamics | Boston Dynamics & FieldAI Partner to Bring Robots Into Uncharted, Dynamic Environments | "Inspection and documentation time reduced by more than 90% compared to manual processes." |
| SP003 | TechCrunch | Alphabet-owned robotics software company Intrinsic joins Google | "Intrinsic laid off 20% of its workforce in January 2023." |
| SP004 | CNBC | Former Alphabet 'moonshot' robotics company Intrinsic is folding into Google | |
| SP005 | The Next Web | Google is doing to factory robots what Android did to phones. Fanuc just became the Samsung of the equation. | "Fanuc commands roughly 16 to 18 per cent of global robot shipments, holds an estimated 50 to 60 per cent of the global CNC market, and has surpassed 1.1 million robots installed in factories." |
| SP006 | Boston Dynamics | Orbit Robot Fleet Management Software | Boston Dynamics | "AIVI-Learning Powered by Google Gemini Robotics — Industrial environments are incredibly complex… We integrated Gemini into AIVI-Learning because modern facilities require industrial AI that can better reason and adapt for your sites." |
| SP007 | Gecko Robotics | Gecko Reaches Unicorn Status | "Gecko Robotics… doubled its valuation from its previous funding round, with a Series D valuation of $1.25 Billion. The round was led by new investors Cox Enterprises." |
| SP008 | Sacra | Gecko Robotics valuation, funding & news | |
| SP009 | Business Wire | Machina Labs Raises $124 Million to Scale Manufacturing Infrastructure for Defense and Advanced Mobility | |
| SP010 | TechCrunch | Amazon hires the founders of AI robotics startup Covariant | "Amazon announced Friday evening that it has hired Covariant's founders — Pieter Abbeel, Peter Chen, and Rocky Duan — along with 'about a quarter' of the startup's employees. It's also signed a non-exclusive license to use Covariant's robotic foundation models." |
| SP011 | The Elec | Physical Intelligence Valued at $11 Billion as Robotics AI Investment Surges | "Notably, the company has yet to release a commercial product or industrial prototype and has not disclosed a commercialization roadmap." |
| SP012 | The Robot Report | Physical Intelligence raises $600M to advance robot foundation models | |
| SP013 | Viam | Viam Announces $45M Series B Funding | |
| SP014 | Sacra | FieldAI valuation, funding & news | |
| SP015 | Physical Intelligence | Physical Intelligence (π) — Company Website | |
| SP016 | Robotics and Automation News | Boston Dynamics and FieldAI partner to bring robots into construction and other complex, dynamic environments | |
| SP017 | TechCrunch | FieldAI raises $405M to build universal robot brains | "The mission is to build a single robot brain that can generalize across different robot types and a diverse set of environments." |
| SP018 | CNBC | Gates, Nvidia-backed robotics firm Field AI hits $2 billion valuation after recent raise | |
| SP019 | CNBC | Gecko Robotics raises $125 million in deal valuing critical infrastructure startup over $1 billion | |
| SP020 | FieldAI | Technology | Field AI | "Deployed entirely on the edge and designed to generalize responses to uncertain scenarios, FieldAI-enabled robots can make strong inferences about unforeseen scenarios to learn and adapt far more quickly and efficiently." |
| SP021 | arXiv | π0.7: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities | |
| SP022 | Intrinsic | Intrinsic — Intelligent Robotics (Homepage) | |
| SP023 | UBOS | Gecko Robotics Secures Multi-Year $71 Million Navy Robotics Deal | "In March 2026 the U.S. Navy awarded Gecko Robotics a five-year IDIQ contract beginning with an initial award of $54 million and carrying a ceiling of $71 million." |
| SP024 | Viam | Viam | Robotics Software Platform for Software Engineers | |
| SP025 | Machina Labs | Home | Machina Labs | |
| SP026 | UpsideList | Fieldai — Company Analysis | "Pre-revenue $2B valuation. $506M preferences mean bear case (50% probability) wipes common equity. Bear (50%)—NVIDIA Isaac/BD internalize; down-round; common wiped by $506M preferences." |
| SP027 | Robotics Press | Covariant: Competitive Response | "Our moat rating is NARROW, not wide. The data advantage is real but not unassailable—Physical Intelligence, Nimble, and Nomagic are all accumulating deployment data, and hyperscalers including Amazon are internalizing manipulation AI." |
| SP028 | CMU | CMU's Robotics Innovation Center Secures FieldAI as Inaugural Corporate Tenant | "When a company valued at $2 billion with $405 million in backing decides to expand its footprint into Pittsburgh and anchor at Carnegie Mellon University's Robotics Innovation Center, it is not a courtesy — it is a calculation." |
| SI001 | FieldAI | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | FieldAI announces that it has raised $405 million in two consecutive rounds. Investors include Bezos Expeditions, BHP Ventures, Canaan Partners, Emerson Collective, Intel Capital, Khosla Ventures, NVentures (NVIDIA's venture capital arm), Prysm, Temasek, and others. The latest round was oversubscribed, following rapid customer adoption and multiple expansion contracts. |
| SI002 | BizProfile.net (sourced from California Secretary of State) | Field Ai, Inc. Mission Viejo, CA — Filing Information | Field Ai, Inc., a Stock Corporation, is a business entity located in Mission Viejo, CA. Officially filed on October 24, 2024, this corporation is recognized under document number 6436098. Formed in Delaware. The data was extracted from the California Secretary of State's Registry (https://bizfileonline.sos.ca.gov/search/business/) as of 3/25/2026. |
| SI003 | TechCrunch | FieldAI raises $405M to build universal robot brains | The most recent round raised $314 million in August and was co-led by Bezos Expeditions, Prysm, and Temasek. FieldAI's other backers include Khosla Ventures, Intel Capital, and Canaan Partners, among others. |
| SI004 | Economic Times (Reuters) | Robotics startup FieldAI raises $314 million in new funding, sources say | The Irvine, California-based startup is now valued at $2 billion, up from $500 million in a round last year. CEO Ali Agha said the company had about 30 people at the end of 2024 and grew to 130 this year, and plans to double headcount by the end of 2025 to support multi-million-dollar contracts in the US, Europe and Asia. |
| SI005 | CNBC | Gates, Nvidia-backed robotics firm Field AI hits $2 billion valuation | The latest round values the two-year-old startup at $2 billion, according to a person familiar with the matter. Along with NVentures and Bezos Expeditions, the rounds included investments from Khosla Ventures, Temasek, Canaan Partners and Intel Capital. |
| SI006 | GeekWire | Robotics startup FieldAI, backed by Gates and Bezos, hits $2B valuation after latest funding | |
| SI007 | Built In Los Angeles | FieldAI Secures $405M to Scale General-Purpose Robotics Intelligence | With the new funding, FieldAI plans to accelerate global expansion, invest in product development for locomotion and manipulation and scale hiring efforts, with a goal of doubling headcount by year's end. |
| SI008 | Orange County Business Journal | FieldAI Robotics Growing, Will Need More Space | Sources familiar with the matter indicate the company has more than $100 million in booked revenue and partnerships with large industrial customers. |
| SI009 | Sacra | FieldAI valuation, funding & news | FieldAI operates a B2B model, offering hardware integration services and software licensing for autonomous robot control. All processing occurs on-device with latency under 100 milliseconds, enabling offline functionality. Data is uploaded to FieldAI's cloud platform for analytics, and the foundation models are continuously refined through federated learning. |
| SI010 | GetLatka | FieldAI Revenue 2025: $140M ARR, $2B Valuation | FieldAI generates $140M in revenue. FieldAI has 150 employees. Company data last updated Nov 24, 2025. |
| SI011 | Premier Alternatives | FieldAI Valuation 2026 — Private Company Worth | With a capital efficiency ratio of 3.95x, FieldAI has achieved a valuation that is 3.95 times the total capital raised. |
| SI012 | PR Newswire / FieldAI | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | |
| SI013 | Construction Dive | Robotic software startup FieldAI lands $405M in fresh funding | |
| SI014 | TechFundingNews | FieldAI raises $405M at $2B valuation to build smarter robot brains | |
| SI015 | UpsideList | Fieldai — Company Analysis | Our model estimates -2% upside. Pre-revenue; no secondary. Dilution Risk: 3-5 more rounds; 50-70% cumulative dilution. Questions to Ask: "Monthly burn and revenue timeline?" "Liquidation preference terms?" |
| SI016 | Robotics Press | FieldAI — Company Profile | Last updated: 2026-03-12. Sources: 11. Completeness: 56%. |
| SI017 | Parola Analytics | Patent Snapshot: FieldAI's vision for adaptable, real-world robotics | The patent application (US 2025/0252306), titled "System and method for uncertainty-aware traversability estimation with optimum-fidelity scan data," was filed on February 5, 2025, following a provisional filing in 2024. The application lists Samuel Triest, David Fan, and Ali Agha as inventors. |
| SI018 | FieldAI via PR Newswire | FieldAI Accelerates Industrial Customers' Adoption of AI in Collaboration with NVIDIA | |
| SI019 | Yahoo Finance (Reuters) | Robotics startup FieldAI raises $314 million in new funding, sources say | |
| SI020 | Robotics & Automation News | FieldAI raises more than $400 million to advance embodied AI at scale | |
| SI021 | Orange County Business Journal | FieldAI Raises $405M for $2B Valuation | |
| SI022 | TechStartups | FieldAI raises $405M in funding at $2B valuation, backed by Bill Gates, Jeff Bezos and NVIDIA | |
| SI023 | CA Company Registry | FIELD AI, INC. — California Companies Directory | Statement of Information BA20241893043. Initial Filing 6436098. Annual Report Due Date changed to 10/31/2025. |
| SI024 | Carnegie Mellon University | CMU's Robotics Innovation Center Secures FieldAI as Inaugural Corporate Tenant | FieldAI named as inaugural corporate tenant at CMU's Robotics Innovation Center at Hazelwood Green. The facility features specialized high-bay labs and a 1.5-acre outdoor running room designed for testing robots in varied terrain. |
| SI025 | Parola Analytics / USPTO | US Patent Application 2025/0252306 — System and method for uncertainty-aware traversability estimation | Patent application US 2025/0252306 filed February 5, 2025, covering uncertainty-aware traversability estimation. Inventors: Samuel Triest, David Fan, Ali Agha. |
| SE001 | FieldAI | Technology | Field AI | "Field Foundation Models™ are unlike any embodied robot AI ever developed, incorporating context and safety awareness into unknown conditions in a way that mimics how humans learn." |
| SE002 | Boston Dynamics | Boston Dynamics & FieldAI Partner to Bring Robots Into Uncharted, Dynamic Environments | "Inspection and documentation time reduced by more than 90% compared to manual processes." |
| SE003 | TechCrunch | FieldAI raises $405M to build universal robot brains | "Once [the] network starts getting access to that, it starts making much safer decisions… it tells you how confident it is, and you as a customer can define this risk threshold." |
| SE004 | FieldAI / PR Newswire | FieldAI Accelerates Industrial Customers' Adoption of AI in Collaboration with NVIDIA | "What once took three and a half months for them now takes just twelve hours with FieldAI's technology, over 200 times faster than manual methods." |
| SE005 | xmaquina.io | How FieldAI Is Building Real-World Robot Intelligence | "Near-term, right now through 2026, they're focused on inspection, monitoring, and data collection. These applications are already deployed and generating revenue." |
| SE006 | Robotics and Automation News | Boston Dynamics and FieldAI partner to deploy autonomous robots on construction sites | |
| SE007 | Intel Capital | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | |
| SE008 | Carnegie Mellon University | CMU's Robotics Innovation Center Secures FieldAI as Inaugural Corporate Tenant | "The RIC facility features specialized high-bay labs and a 1.5-acre outdoor 'running room' designed for testing robots in varied terrain." |
| SE009 | Orange County Business Journal | FieldAI Robotics Growing, Will Need More Space | |
| SE010 | FieldAI | Solutions | Field AI | |
| SE011 | Certis Group | Certis and FieldAI Form Strategic Partnership to Deploy Autonomous Robotics in Real-World Security Operations | "At the core of FieldAI's technology are its Field Foundation Models™, general-purpose autonomy software that acts as a brain for any robot type to operate safely in complex, dynamic and real-time environments without reliance on prior maps, prior routes, or supporting infrastructure." |
| SE012 | GitHub (FieldAI organization) | Field AI — GitHub organization | "field-ai/rosbridge_suite — updated Apr 2, 2026; field-ai/spconv — updated Mar 5, 2026; field-ai/spot-sdk — updated Oct 22, 2025." |
| SE013 | FieldAI | Field AI Research Institute (FAIRI) | Field AI | |
| SE014 | Ali Agha personal academic site | Ali Agha — Publications | |
| SE015 | Pure AI | Field AI Expands NVIDIA Collaboration to Accelerate Industrial AI Adoption | |
| SE016 | FieldAI | FieldAI and NVIDIA Omniverse: Building the Next Generation of Industrial AI | "When the model encounters something unfamiliar, it adapts in real time by slowing down or choosing a more conservative path. That conservatism has a cost in throughput and efficiency." |
| SE017 | FieldAI | Bringing General-Purpose Robots to Every Construction Site: Inside Big-D Construction's Expansion with FieldAI | "For over two years of field deployments, FieldAI has worked closely with the Big-D team to incorporate their feedback into the product." |
| SE018 | Morningstar / PR Newswire | Certis and FieldAI Form Strategic Partnership to Deploy Autonomous Robotics in Real-World Security Operations | |
| SE019 | FieldAI | FieldAI | One Autonomy for All Robots (Homepage) | |
| SE020 | FieldAI / PR Newswire | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | |
| SE021 | The Robot Report | FieldAI founder and CEO to discuss building risk-aware AI models at RoboBusiness | |
| SE022 | FieldAI | Release Notes — Field AI (private) | |
| SE023 | Robotics and Automation News | FieldAI raises more than $400 million to advance embodied AI at scale | |
| SE024 | Emerson Collective | Dr. Ali Agha — Emerson Collective Portfolio | |
| SE025 | AI Wiki | Field AI — AI Wiki | |
| SU001 | FieldAI | Bringing General-Purpose Robots to Every Construction Site: Inside Big-D Construction's Expansion with FieldAI | "I think the opportunity is that every project will have some representation of FieldAI tools." — Shaun Orr, C-level executive, Big-D Construction |
| SU002 | Robotics and Automation News | Big-D expands construction operations with FieldAI as robotics adoption accelerates | |
| SU003 | Boston Dynamics | Boston Dynamics & FieldAI Partner to Bring Robots Into Uncharted, Dynamic Environments | "Efficiency: Inspection and documentation time reduced by more than 90% compared to manual processes." |
| SU004 | PR Newswire | FieldAI Accelerates Industrial Customers' Adoption of AI in Collaboration with NVIDIA | "A senior executive at a global industrial manufacturing company noted that what once took three and a half months for them now takes just twelve hours with FieldAI's technology." |
| SU005 | FieldAI | FieldAI Accelerates Industrial Customers' Adoption of AI in Collaboration with NVIDIA | |
| SU006 | The Straits Times | Certis and FieldAI Partner to Advance Autonomous Robotics in Real-World Security Operations | |
| SU007 | International Security Journal | Field AI and Certis form strategic partnership | "For robotics to be viable at scale, they must integrate seamlessly with human teams, operational workflows and command systems." — Ng Tian Beng, President and Group CEO, Certis |
| SU008 | Sacra | FieldAI valuation, funding & news | |
| SU009 | Interesting Engineering | US robot dog patrols massive construction sites for faster progress | "The FieldAI system makes us better at what we do. Giving us greater efficiency, helping us document items more effectively, and taking some of the more mundane tasks off our plate." — Justin Schreiner, DPR superintendent |
| SU010 | ThomasNet | Robot Dog Takes on 'Mundane' Tasks in Construction (Video) | |
| SU011 | InspEnet | FieldAI Brain transforms construction with robotics | |
| SU012 | Robots in Construction | FieldAI | Robots in Construction | "4 Confirmed Deployments. 2 Named Customers (DPR Construction and Big-D Construction) and 2 unnamed top-10 ENR firms." |
| SU013 | robotics.press | FieldAI | robotics.press | "Very few named, quantified customer case studies in the public domain; most deployment claims originate from company materials without third-party verification of intervention rates, uptime, or ROI." |
| SU014 | Intel Capital | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | |
| SU015 | FieldAI | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | "The latest round was oversubscribed, following rapid customer adoption and multiple expansion contracts for FieldAI's general-purpose robotics intelligence, with successful testing and deployments across hundreds of complex real-world industrial environments." |
| SU016 | Robotics and Automation News | Boston Dynamics and FieldAI partner to bring robots into construction and other complex, dynamic environments | |
| SU017 | CNBC | 'Silent failure at scale': The AI risk that can tip the business world into disorder | "Autonomous systems don't always fail loudly. It's often silent failure at scale." — Noe Ramos, VP AI Operations, Agiloft |
| SU018 | Construction Dive | Contractor optimism for robots grows, but usage drops | "The share of firms reporting active robotics use fell from 65% last year to 46%." |
| SU019 | The Robot Report | Construction robotics finds interest, but adoption wavers, reports BuiltWorlds | |
| SU020 | Construction Owners Association | Construction Contractors Show Strong Support for Robotics but Adoption Slips in 2025 | |
| SU021 | Construction Dive | Robotic software startup FieldAI lands $405M in fresh funding | |
| SU022 | FieldAI | Construction | Solutions | FieldAI | |
| SU023 | FieldAI | Industrial & Energy | Solutions | FieldAI | |
| SU024 | FieldAI | Urban Operations | Solutions | FieldAI | |
| SU025 | FieldAI | Journey | Company | FieldAI | |
| SR001 | K&L Gates LLP | AI Product Liability: The Next Wave of Litigation | Product liability is different: It is built to evaluate mass-distributed technologies through the lenses of defect, warnings, and foreseeability, with liability that can extend across a chain of entities involved in making a product available. |
| SR002 | Baker Botts LLP | AI Legal Watch: January 2026 | Colorado's AI Act (SB 24-205)—the first comprehensive U.S. statute targeting 'high-risk' AI systems—takes effect June 30, 2026, requiring impact assessments, consumer disclosures, and reasonable care to prevent algorithmic discrimination. |
| SR003 | International AI Safety Report 2026 (Expert Advisory Panel, 100+ independent experts) | International AI Safety Report 2026 — Extended Summary for Policymakers | Rapid advancements in robot autonomy/capability and corresponding emergent risks—including loss of control scenarios and inadequate real-world reliability testing. |
| SR004 | Gunder LLP | 2026 AI Laws Update: Key Regulations and Practical Guidance | The EU has adopted binding legal frameworks that extend to non-EU based organizations, most notably the EU AI Act, which imposes significant obligations on high-risk and general-purpose AI models. |
| SR005 | SRES (Safety & Reliability Engineering Solutions) | CES Wrap-Up 2026: The Humanoid Robot Safety Question | Existing industrial safety protocols often fall short when robots share open workspaces with humans. |
| SR006 | CDC / National Institute for Occupational Safety and Health (NIOSH) | Center for Occupational Robotics Research | NIOSH established the Center for Occupational Robotics Research in September 2017 to address the safety of today's workers who use, wear, or work near robots. |
| SR007 | Control Engineering | Industrial robot safety considerations, standards and best practices to consider | |
| SR008 | Chosun Biz (Korea JoongAng Daily) | Hyundai Motor Group boosts US robot AI push with Field AI investment | |
| SR009 | VerticalData.io | AI Supply Chain Constraints: GPU Lead Times, Allocation & Hardware Procurement in 2026 | The AI industry generates headlines every week. But behind product launches and model announcements, there is a quieter dynamic shaping outcomes: which organizations can actually build at scale and which ones are left waiting. |
| SR010 | Global Law Lists | The 10 Most Consequential Legal Rulings on AI in 2025–2026 | |
| SR011 | Maedcore | Industrial Safety with Robotics: A Practical Guide 2026 | |
| SR012 | Tracxn | Field AI — 2026 Company Profile | |
| SR013 | FieldAI | Boston Dynamics and FieldAI Partner to Bring Robots Into Uncharted and Dynamic Environments | |
| SR014 | Boston Dynamics | Boston Dynamics & FieldAI Partner to Bring Robots Into Uncharted, Dynamic Environments | Combining Field AI's expertise in risk-aware autonomy with Spot's remarkable mobility allows us to tackle uncharted and highly dynamic environments together—spaces that were previously off-limits to robots. |
| SR015 | CNBC | Nvidia, Bill Gates-backed robotics startup Field AI hits $2 billion valuation after recent raise | |
| SR016 | GeekWire | Robotics startup FieldAI, backed by Gates and Bezos, hits $2B valuation after latest funding | |
| SR017 | Orange County Business Journal | FieldAI Raises $405M for $2B Valuation | The robotics company's valuation jumped to $2 billion from $500 million, placing it in the 'unicorn' category of startups worth more than $1 billion. |
| SR018 | FieldAI | FieldAI and NVIDIA Omniverse: Building the Next Generation of Industrial AI | NVIDIA Omniverse libraries provide the digital substrate that accelerates the transition from field data to intelligent digital worlds. |
| SR019 | Premier Alternatives | FieldAI Valuation: $2.0B (2026) | |
| SR020 | Robotics and Automation News | Boston Dynamics and FieldAI partner to bring robots into construction and other complex, dynamic environments | |
| SR021 | Premier Alternatives | FieldAI Team | Leadership & Board | |
| SR022 | TechCrunch | FieldAI raises $405M to build universal robot brains | |
| SR023 | PR Newswire | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | |
| SR024 | FieldAI | FieldAI Team | |
| SR025 | Certis Group | Certis and FieldAI Form Strategic Partnership to Deploy Autonomous Robotics in Real-World Security Operations | |
| SR026 | FieldAI | FieldAI Technology | |
| SR027 | Sacra | FieldAI — Private Company Research | |
| SR028 | The Robot Report | FieldAI founder and CEO discuss building risk-aware AI models at RoboBusiness | |
| SR029 | FieldAI | Latest News | FieldAI | |
| SR030 | Paróla Analytics | FieldAI Robotics Patents | |
| SV001 | CNBC | Gates, Nvidia-backed robotics firm Field AI hits $2 billion valuation | Field AI's robots are deployed in daily operations at numerous customer sites worldwide. |
| SV002 | TechCrunch | FieldAI raises $405M to build universal robot brains | |
| SV003 | GeekWire | Robotics startup FieldAI, backed by Gates and Bezos, hits $2B valuation after latest funding | The investment, which came in consecutive Series A and A1 rounds, values the 2-year-old startup at $2 billion. |
| SV004 | Economic Times | Robotics startup FieldAI raises $314 million in new funding, sources say | FieldAI, which develops systems for robots to operate safely in industrial environments, has raised $314 million in a new funding round, quadrupling its valuation. |
| SV005 | FieldAI | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale | FieldAI has raised $405 million in two consecutive rounds. |
| SV006 | Orange County Business Journal | FieldAI Raises $405M for $2B Valuation | |
| SV007 | Orange County Business Journal | FieldAI Robotics Growing, Will Need More Space | People familiar with the matter tell the Business Journal that the company has more than $100 million in booked revenue and partnerships with large industrial customers. |
| SV008 | Sacra | FieldAI valuation, funding & news | |
| SV009 | PremierAlts | FieldAI Valuation: $2.0B (2026) | Total Funding Raised: $506.1M across 5 rounds; Last Round: Early Stage VC $315.0M; Last Funding: Feb 2026. |
| SV010 | PitchBook | Field AI 2025 Company Profile: Valuation, Funding & Investors | |
| SV011 | Houlihan Lokey | AI in Vertical Software | Q1 2026 Market Update | AI-native software companies command the highest multiples, with a median of 21.2x EV/Revenue for venture capital rounds. |
| SV012 | TechCrunch | Robotics software maker Skild AI hits $14B valuation | Skild AI has raised a $1.4 billion Series C round that values it at more than $14 billion. |
| SV013 | TechCrunch | Humanoid robot startup Apptronik has now raised $935M at a $5B+ valuation | TechCrunch separately learned that its post-money valuation is now about $5.3 billion. |
| SV014 | CNBC | Apptronik raises $520 million at $5 billion valuation for Apollo robot | |
| SV015 | Bloomberg | AI Robotics Lab in Talks to Raise $1 Billion at $11 Billion Valuation | Physical Intelligence, the two-year-old San Francisco robotics startup, is in discussions to raise about $1 billion in new funding at a valuation exceeding $11 billion. |
| SV016 | TechCrunch | Physical Intelligence is reportedly in talks to raise $1B, again | Co-founder Lachy Groom told TechCrunch the company has no timeline for commercialization, an unusual posture that its investors don't seem to mind. |
| SV017 | CNBC | Gecko Robotics raises $125 million in deal valuing critical infrastructure startup over $1 billion | Gecko Robotics announced on Thursday that it raised $125 million in a Series D funding round, raising the AI and robotics company's valuation to $1.25 billion. |
| SV018 | Gecko Robotics | Gecko Reaches Unicorn Status | |
| SV019 | U.S. Securities and Exchange Commission | HII Field AI Series II — Form D Filing (CIK 0002095305) | HII Field AI Series II, a Series of HII Field AI, LLC — Pooled Investment Fund — Other Investment Fund — filed 2025-11-14. |
| SV020 | U.S. Securities and Exchange Commission (EDGAR) | SEC EDGAR Full-Text Search — Form D for 'Field AI' | Hits: HII Field AI Series I (filed 2025-11-14), HII Field AI Series II (filed 2025-11-14), HII Field AI Series III (filed 2026-04-03). No direct Form D found for Field Ai, Inc. |
| SV021 | Nasdaq Private Market | Sell or Invest in Field AI Stock Pre-IPO | |
| SV022 | Notice.co | FieldAI Stock — How to Buy, Valuation, Stock Price, IPO | FieldAI Stock $34.32 |
| SV023 | TechFundingNews | FieldAI raises $405M at $2B valuation to build smarter robot brains | |
| SV024 | Economic Times | AI startup valuations are overheated: First Round Capital cofounder Howard Morgan | 'In a bubble, everybody thinks buy high, sell higher works forever, but that only works inside the bubble,' said Howard Morgan. |
| SV025 | GeekWire | Is there an AI bubble? Investors sound off on risks and opportunities for tech startups in 2026 | There's clear froth in parts of the AI market, especially in early-stage private valuations where companies are priced well ahead of fundamentals. |
| SV026 | BusinessWire | Skild AI Raises $1.4B, Now Valued Over $14B | The company grew from zero to about $30M revenue in just a few months in 2025, and is growing exponentially. |
| SV027 | Orange County Business Journal | $2B-Valued FieldAI Moves to Irvine | |
| SV028 | TechStartups | FieldAI raises $405M in funding at $2B valuation, backed by Bill Gates, Jeff Bezos and Nvidia | |
| SV029 | The Robot Report | FieldAI founder, CEO discuss building risk-aware AI models at RoboBusiness | |
| SV030 | The Robot Report | Physical Intelligence raises $600M to advance robot foundation models | |
| SV031 | Robotics and Automation News | FieldAI raises more than $400 million to advance embodied AI at scale | |
| SV032 | Intel Capital | FieldAI Announces Over $400M in Funds Raised to Advance Embodied AI at Scale |