Skild AI
全形态机器人基础模型:任意机器人,任意任务
Skild AI 是硬件无关机器人 AI 的早期平台龙头,技术差异真实;但 $14B 估值约等于 467x ARR,几乎不给执行风险留余地——在审计经济性和公开基准出现前,应继续研究。
封面要素
公司概况
Skild AI 是一家总部位于 Pittsburgh 的机器人基础模型公司,2023 年由 CMU/FAIR 研究员 Deepak Pathak、Abhinav Gupta 和 Ashish Kumar 创立。旗舰产品 Skild Brain 是一个硬件无关的 AI 平台,在 100,000+ 种模拟机器人形态上训练,可让任意机器人在不了解自身硬件的情况下执行复杂任务。公司已获 $1.83B+ 融资,SoftBank 领投 Series C 后估值 $14B,是全球资本最充足的软件优先型机器人 AI 公司。
- 官网
- skild.ai
- 成立时间
- 2023-01-01
- 创始人
- Deepak Pathak, Abhinav Gupta, Ashish Kumar
- 创立地点
- Pittsburgh, Pennsylvania, USA
- 总部
- Pittsburgh, Pennsylvania, USA
- 产品
- Skild Brain 是通过企业 API 交付的硬件无关机器人基础模型,在 Multiverse 仿真环境中横跨 100,000+ 种模拟机器人形态训练。它把分层架构(高层规划器 + 底层运动控制器)和上下文内适应结合起来,对未见过的机器人硬件做零样本泛化。2026 年 4 月收购 Zebra Symmetry 后,公司补上企业级机器人机队编排和实时监控。
- 客户
- 面向在安防巡检、仓储物流、楼宇巡检、建筑、数据中心维护和制造等场景部署自主机器人的企业 OEM 和系统集成商。
- 商业模式
- 基于 API 的授权——客户按机器人实例或 API 调用付费,将 Skild Brain 嵌入现有机器人平台。次要收入来自机器人机队编排(Zebra Symmetry),以及借助 IQT 投资人关系刚起步的政府 / 双用途渠道。
- 阶段
- series-c
- 融资情况
- Series C 轮于 2026 年 1 月完成,投后估值 $14B,SoftBank Vision Fund 2 领投融资 $1.4B。此前轮次:$300M Series A(2024 年 7 月,约 $1.5B)、$500M Series B(2025 年 5 月,$4.7B)。累计:$1.83B+。
执行摘要
主要优势
- 基于 100K+ 形态训练的 omni-bodied 架构没有公开直接对标;CMU/FAIR 创始团队拥有 physical AI 领域最深的学术到商业 IP 转移能力。
- 已从战略协同投资人(NVIDIA、Samsung、LG、SoftBank)融资 $1.83B,这些投资人同时也是分销伙伴,压低机器人 AI 常见的硬件 go-to-market 门槛。
- 收购 Zebra Symmetry 带来企业级机队编排和继承客户群,把纯 API 业务变成切换成本更高的托管服务方案。
主要风险
- NVIDIA GR00T N1 开放模型和 Physical Intelligence pi0 在核心用例上直接竞争,且许可成本为零,对定价和利润率形成结构性商品化压力。
- $14B 估值约为 467x trailing ARR,是该领域已披露最高收入倍数;若倍数向软件优先可比公司($3-7B 区间)压缩,入场价格回报将为负。
- IQT(In-Q-Tel)投资人带来军民两用 / 出口管制敞口;Form D 和审计财务均未公开;Zebra Symmetry 的 IP 链条和整合深度未验证。
未决问题
- 审计收入、毛利率、烧钱速度和现金余额均未公开;$30M ARR 来自第三方推断,无法用公司披露数据核实。
- 尚无具名企业客户获得公开确认;公司声称的部署垂直缺少参考客户,若没有一手访问,商业牵引力无法验证。
- 与 IQT 投资绑定的军民两用限制、出口管制义务,以及任何政府数据共享要求均未披露。
目录
01公司概况
1.1 身份与业务概览
Skild AI 于 2023 年 5 月注册成立,2024 年 7 月宣布 $300M Series A 后走出隐身状态。公司总部位于 Pittsburgh, Pennsylvania,并在 San Francisco Bay Area 设办公室;2026 年 2 月之后又在 Bengaluru, India 设点。 Skild AI 的核心产品是「Skild Brain」,公司称其为行业首个统一机器人基础模型。过去的机器人软件往往为特定机型和任务定制;Skild Brain 则是「全形态」的:无论四足机器人、人形机器人、桌面机械臂还是移动机械臂,它都能在事先不了解机器人具体物理形态的情况下完成控制。机器人可由此处理从简单家务到高强度工业作业的一系列任务。 公司的商业模式是软件平台提供商:Skild Brain 充当通用 AI 层,硬件厂商、系统集成商和企业客户可用它激活不同机器人机队。Skild 的收入来自安防 / 设施巡检、最后一公里配送、仓库、制造、数据中心和建筑等行业的企业部署。2026 年 4 月收购 Zebra Technologies 的 Robotics Automation 业务(包括 Symmetry Fulfillment 平台)后,Skild 进一步扩展到端到端仓库自动化解决方案。 截至 2026 年 1 月,Skild AI 累计融资超过 $2B,估值超过 $14 billion,是全球估值最高的私营机器人公司之一。公司的长期目标是发展根植于物理世界的人工通用智能(AGI),挑战 AGI 只能从数字知识中产生的主流看法。 [CO001, CO002, CO003, CO004, CO005, CO020]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 / 备注 |
|---|---|---|---|---|
| 估值 | $14B+ | Jan 14, 2026 | 高 | Series C 交割后估值 |
| 累计融资 | >$2B(CEO 披露);Crunchbase 为 $1.83B | Jan 2026 | 高 | 种子轮具体金额未披露 |
| 收入(年化运行率) | ~$30M(从 $0 增长) | Late 2025 | 中 | 公司披露;未经第三方验证 |
| 收入增长 | 数月内从 $0 到 ~$30M | 2025 | 中 | 缺少上一年基线,无法计算 YoY |
| 员工数(LinkedIn) | ~85 名员工 | May 2026 | 低 | LinkedIn 统计;可能不含承包商 |
| 员工数(法律实体,Dec 2024) | 34 名员工 | Dec 31, 2024 | 中 | Tracxn / 法律实体来源 |
| 阶段 | Series C | Jan 14, 2026 | 高 | 公司公告确认 |
| 总部 | Pittsburgh, PA | 2023–present | 高 | |
| 办公室 | Pittsburgh、SF Bay Area、Bengaluru 三地 | Feb 2026 | 高 | |
| 成立 | May 2023 | May 2023 | 高 |
收入数据由公司披露,未经独立审计。LinkedIn(~85)与法律实体文件(截至 Dec 2024 为 34) 的员工数不一致;Series C 后快速招聘可能解释这一差异。估值为 Series C 交割后估值。
[CO001, CO003, CO004, CO013, CO017, CO018]| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 / 背景 | 含义 |
|---|---|---|---|---|---|
| May 2023 | Skild AI 由 Deepak Pathak 和 Abhinav Gupta 创立 | 创立 | 两位 CMU Robotics Institute 教授离开学界 | 显示深厚领域能力进入商业 AI 机器人赛道 | |
| 2023 (Q3–Q4) | Sequoia Capital 种子轮融资 | 融资 | 未披露 | Sequoia Capital(Stephanie Zhan 领投) | 最早期验证创始愿景;快速拿到机构背书 |
| 2023–2024 | Skild Brain 在隐身期启动研发 | 产品 | 内部;首次演示全形态具身控制 | 建立核心 IP 与技术差异化 | |
| 2024 (H1) | 机器人上下文学习突破;获最佳论文提名 | 产品 | 在顶级机器人会议发表研究 | 首次证明机器人无需重新训练即可上下文适应 | |
| Jul 9, 2024 | Series A 交割;走出隐身并公开亮相 | 融资 | 估值 $1.5B,融资 $300M | Lightspeed(领投)、Coatue、SoftBank、Bezos Expeditions、Felicis、Sequoia、Menlo、General Catalyst、CRV、 Amazon、SV Angel、CMU | 公告时为机器人 AI 迄今最大 Series A;公司正式公开 |
| Jul 10, 2024 | SKILD AI, INC. 法律实体注册成立 | 监管 | 美国注册;CIN 7456248 | 公告后取得正式法律主体地位 | |
| 2025 (H1) | 收入数月内从 $0 增至 ~$30M | 规模化 | ~$30M 收入 | 企业部署覆盖仓储、制造、安全、配送、建筑、数据中心 | 机器人软件最快收入爬坡;验证商业模式 |
| Jun 12, 2025 | Series B 交割 | 融资 | ~$135M,估值 $4.5B | SoftBank(领投 ~$100M)、Nvidia($25M)、Samsung($10M) | 11 个月估值翻三倍;新增战略硬件支持方 |
| Jan 14, 2026 | Series C 交割;估值 $14B+ | 融资 | $1.4B,估值 $14B+ | SoftBank(领投)、NVentures、Macquarie Capital、Jeff Bezos、Samsung、LG、Schneider Electric、CommonSpirit、 Salesforce Ventures、IQT 等 | 机器人软件最大单笔融资;按估值巩固 Skild 全球顶级机器人 AI 公司地位 |
| Feb 19–20, 2026 | Bengaluru, India 办公室开业 | 规模化 | 首个国际办公室;R&D 和工程中心 | 扩大全球人才容量;首次落点亚太 | |
| Apr 2026 | 收购 Zebra Technologies 的机器人自动化业务 | 产品 | 以股权换资产交易(Zebra 获得 Skild 股权) | 包括 Symmetry Fulfillment 编排平台 | 拼出首个端到端仓库自动化方案;立即具备企业级扩展能力 |
Series B 的 $135M 金额来自 Crunchbase 和第三方报道;Skild 尚未正式确认 Series B 的具体总额。 种子轮金额未披露。
[CO001, CO002, CO003, CO005, CO015, CO016]Skild AI 从 2023 年 5 月创立到 2026 年 4 月收购 Zebra 的关键里程碑,展现出异常压缩的融资与商业化节奏。
种子轮日期根据 Sequoia 博客估算为 2023 年末(该博客发表于 2024 年 7 月 9 日,称种子轮发生在「九个月」前),金额报道为未披露。Series B 日期据 Tracxn。
[CO001, CO002, CO015, CO016, CO017, CO018]截至报告日期(2026 年 5 月),捕捉 Skild AI 成熟度、资本位置、商业牵引力和团队规模的关键指标。
收入数字为公司口径,未经独立审计。估值为 Series C(2026 年 1 月)投后估值,可能已随后续 Zebra 收购中的股权调整而变化。
[CO001, CO003, CO004, CO005, CO018, CO019]1.2 创始人与领导团队
Skild AI 由 Deepak Pathak(CEO)和 Abhinav Gupta(总裁)共同创立。两人都是全球机器人与 AI 领域引用次数最高的研究者之一,合计 h 指数超过 150,学术引用超过 90,000 次,体现出他们在该领域的特殊影响力。 Deepak Pathak 在印度小城长大,拿到 IIT Kanpur 计算机科学金牌,随后在 UC Berkeley 师从 Alyosha Efros 和 Trevor Darrell 完成 AI 博士。他曾在 Facebook AI Research(FAIR)做基础研究,之后成为 CMU Robotics Institute 的 Raj Reddy 副教授。Pathak 开创了机器人好奇心驱动探索、自监督学习和快速运动适应等方向。他获得过 Sloan Research Fellowship、MIT TR35 Innovator Under 35,以及 ICRA、CVPR、RSS、CoRL 等顶级会议的多个最佳论文奖。 Abhinav Gupta 是 CMU Robotics Institute 终身教授,也是 FAIR Robotics(Facebook/Meta)的创始成员和研究负责人。他的研究聚焦自监督学习、视觉表示学习,以及机器人系统的大规模数据。他获得过 ONR Young Investigator Award、PAMI Young Researcher Award、Sloan Fellowship 和 Okawa Research Grant。Gupta 和 Pathak 在 2023 年初共同创立 Skild 之前,已经讨论创业超过十年。 更广泛的 Skild AI 团队来自 Meta、Tesla、Nvidia、Amazon、Google、CMU、Stanford 和 UC Berkeley。LinkedIn 显示,截至 2026 年初,平台上约有 85 名员工;Tracxn 基于法律实体备案的数据则显示,截至 2024 年 12 月 31 日有 34 名员工——差异很可能来自 Series B 和 Series C 融资后的快速招聘。 [CO006, CO007, CO008, CO009, CO010, CO011]
| 人物 | 职务 | 背景 | 创始人与市场匹配度 | 关键人风险 |
|---|---|---|---|---|
| Deepak Pathak | CEO 与联合创始人 | IIT Kanpur(计算机科学金牌),UC Berkeley PhD;FAIR 研究员;CMU Robotics Institute Raj Reddy 副教授 | 好奇心驱动探索、自监督学习、快速运动适应的先行者;Sloan Fellow;MIT TR35 Innovator Under 35 | 高——首要技术愿景提出者和对外代表 |
| Abhinav Gupta | 总裁与联合创始人 | CMU Robotics Institute 终身教授;FAIR Robotics(Meta)创始成员与研究负责人 | 面向具身智能体的大规模自监督学习先行者;ONR Young Investigator、PAMI Young Researcher、Sloan Fellow、Okawa Grant | 高——同等联合创始人,机构和学术网络深 |
| Stephanie Zhan | 董事 / 投资人(Sequoia) | Sequoia Capital 合伙人;领投种子轮和 Series A | 曾称「机器人领域的 GPT-3 时刻即将到来」;从种子轮起就高确信投入 | 中——投资人董事席位 |
| Raviraj Jain | 董事会观察员 / 投资人(Lightspeed) | Lightspeed 合伙人;共同领投 Series A | 称 Skild 是「独一无二的公司」 | 低——观察员角色 |
| Dennis Chang | 战略伙伴(SoftBank) | SoftBank Investment Advisers 管理合伙人;领投 Series B 和 C | SoftBank 多轮领投方 | 低——投资人角色 |
除投资人席位外,董事会构成未公开披露。仅 Pathak 和 Gupta 被确认担任执行官。 完整 C 级和 VP 级领导层未公开披露。
[CO006, CO007, CO008, CO009, CO010, CO011]1.3 融资历史与投资人基础
Skild AI 自创立以来融资节奏极快。公司 2023 年从 Sequoia Capital 融到一轮金额未披露的种子轮;随后在 2024 年 7 月 9 日完成标志性的 $300M Series A,投后估值 $1.5B——Lightspeed Venture Partners、Coatue、SoftBank Group 和 Jeff Bezos(通过 Bezos Expeditions)领投,Felicis Ventures、Sequoia、Menlo Ventures、General Catalyst、CRV、Amazon、SV Angel 和 Carnegie Mellon University 等跟投。 2025 年 6 月,Skild 完成约 $135M Series B,由 SoftBank 领投(据报道出资 $100M),Nvidia 出资 $25M,Samsung 出资 $10M,估值 $4.5B。七个月后,2026 年 1 月,公司完成 SoftBank Group 领投的 $1.4B Series C,估值超过 $14B。该轮引入 NVentures(NVIDIA 的 VC 部门)、Macquarie Capital、Jeff Bezos(Bezos Expeditions)、Samsung、LG、Schneider Electric、CommonSpirit Health、Salesforce Ventures、IQT(In-Q-Tel)等战略投资人。此前投资人 Lightspeed、Felicis、Coatue 和 Sequoia 也都在 Series C 继续加码。 CEO Deepak Pathak 于 2026 年 1 月表示,公司迄今融资超过 $2B。Crunchbase 截至 Series C 跟踪到四轮合计 $1.83B。Series C 之后,Skild 于 2026 年 4 月收购 Zebra Technologies 的 Robotics Automation 业务;交易中 Zebra 获得 Skild 股权,股权结构表因此进一步纳入战略工业伙伴。 投资人基础值得注意:既有 Sequoia、Lightspeed、Coatue、Felicis 等硅谷一线 VC,也有横跨硬件(NVIDIA、Samsung、LG)、工业自动化(Schneider Electric)、医疗(CommonSpirit)、企业软件(Salesforce Ventures)、美国国防 / 情报(IQT/In-Q-Tel)和全球基础设施资本(Macquarie Capital、SoftBank)的主要战略 / 企业投资人。 [CO015, CO016, CO017, CO018, CO019, CO034]
| 投资人 / 利益相关方 | 类型 | 轮次 | 战略重要性 | 尽调问题 |
|---|---|---|---|---|
| SoftBank Group | 财务 / 战略 VC | A、B(领投)、C(领投) | 三轮领投方;承诺 $1B+;全球机器人分销能力 | SoftBank 获得了哪些治理权和董事会席位? |
| Sequoia Capital | VC(早期领投) | 种子轮(领投)、A、C(加注) | 首个机构支持方;董事会成员(Stephanie Zhan);强信号价值 | 确认董事会构成和 Sequoia 治理权 |
| Lightspeed Venture Partners | VC | A(共同领投)、C(加注) | 共同领投 Series A;Raviraj Jain 观察员;企业网络强 | 领投地位是否附带反稀释条款? |
| Coatue | VC / 跨界基金 | A、C(加注) | 聚焦科技的跨界基金;验证机构级兴趣 | 持仓规模及是否有老股交易 |
| NVentures (NVIDIA) | 企业 VC | C | 带来 AI 算力生态对齐;有望接入 NVIDIA Omniverse / Isaac 仿真 | 确认投资是否附带商业协议 |
| Bezos Expeditions (Jeff Bezos) | 个人 / 家族办公室 | A、C | 高知名度背书;连接 Amazon 生态 | 是否有补充协议或商业承诺? |
| Samsung / LG | 企业战略投资 | B 轮(Samsung)、C 轮(Samsung、LG) | 亚洲硬件制造生态;机器人硬件供应链 | 投资是否附带 OEM 或授权协议? |
| In-Q-Tel (IQT) | 美国情报 VC | C | 释放美国国家安全 / 国防应用兴趣信号 | IQT 投资是否绑定合同义务、IP 限制或出口管制? |
| Carnegie Mellon University | 学术 / 机构 | A | 来自创始人母校的机构背书;人才管线 | CMU 除财务投资外的关系性质 |
| Zebra Technologies | 战略方(以股权换并购资产) | C 轮后 | 以机器人自动化业务换取 Skild 股权;现在成为绑定伙伴 | 并购条款、Zebra 持股比例及整合里程碑 |
轮次参与信息由多份新闻稿汇总而来。各投资人的单独投资金额未披露。
[CO015, CO016, CO017, CO034, CO035, CO036]1.4 关键里程碑与战略时间线
从隐身创业公司到 $14B+ 公司,Skild AI 走出了一条相对投入资本异常压缩的时间线。自 2023 年 5 月创立以来,公司以很高速度完成了多项技术、商业和战略里程碑。 技术上,Skild 发表了机器人上下文内学习研究;公司称这是该领域首个研究突破,并获得顶级机器人会议最佳论文提名。Skild Brain 展示出无需重新训练即可适应机器人形态剧烈变化的能力,例如失去肢体、车轮卡死,或全新的身体几何结构。公司还搭建了一个数据飞轮,把大规模仿真(数万亿条合成经验)、互联网视频(数十亿个人类动作视频)、遥操作和真实部署数据结合起来,训练数据量约为竞争性机器人基础模型的 1000x。 商业上,Skild 在 2025 年「短短几个月」内从零收入做到约 $30M,并落地于仓库、制造设施、数据中心、建筑工地和安防应用。2026 年 4 月收购 Zebra 并纳入 Symmetry Fulfillment 仓库编排平台后,Skild 成为首家能够为现有仓库提供端到端自动化方案的公司:从人形机器人抓取放置、机器狗巡检,到 AMR 物料搬运,都由同一 AI 层编排。 2026 年 2 月开设 Bengaluru, India 办公室,是公司首次国际扩张,也把工程能力延伸到印度深厚的 AI 和机器人技术人才池。 [CO023, CO024, CO025, CO026, CO027, CO028]
展示 Skild AI 的身份、产品、资本、客户部署和关键依赖如何相互连接,形成当前商业位置和竞争护城河。
[CO011, CO020, CO025, CO026, CO018, CO029]1.5 展示材料
02市场分析
2.1 市场定义与范围
Skild AI 位于机器人硬件和人工智能软件的交汇处,具体瞄准物理 AI 或具身智能层:让机器人在非结构化真实环境中感知、推理、规划并行动的 AI 软件。这让 Skild 区别于传统机器人硬件制造商(如 ABB、KUKA、Fanuc),也区别于按任务定制的自动化软件供应商。 本分析的市场边界包括:(1)面向商用机器人的 AI 基础模型软件和 API;(2)机器人智能中间件、仿真平台和训练基础设施;(3)端到端机器人自动化软件方案,例如 Skild 收购后的 Symmetry Fulfillment 平台。排除的支出包括机器人硬件(执行器、传感器、机架)、人工操作机械,以及经典基于规则的工业自动化软件(PLC、SCADA)。相邻市场——仓库管理系统、自动驾驶车辆和无人机软件——共享技术 DNA,但服务不同买家和采购路径。 物理 AI 软件的现状替代方案包括:(a)示教再现编程,由人类专家手工演示每一个机器人动作;(b)按应用定制的机器学习模型,每个任务和机器人类型都需要单独训练;(c)继续使用人工,这是多数非结构化工作流的既有做法。这些替代方案的失效点——成本高、缺乏弹性、无法泛化——正是 Skild 切入的楔子。从固定用途自动化转向通用机器人智能,是结构性市场迁移,而不是渐进式功能改良。 [CM001, CM002, CM003, CM004, CM005]
| 细分市场 / 类别 | 包含支出 | 不包含支出 | 买方 / 付款方 | 与 Skild 的相关性 |
|---|---|---|---|---|
| 物理 AI 基础模型软件 | 机器人 AI API、SDK 许可、模型权重、训练基础设施、仿真平台 | 机器人硬件、传感器、执行器、人工操作机械 | 机器人 OEM、企业运营方、系统集成商 / 软件预算 | 直接 TAM——Skild Brain 销售切入这一层 |
| 工业机器人硬件 | 机械臂、协作机器人、AMR、四足平台、末端执行器 | AI 软件、SaaS 编排、维护服务 | 制造业 OEM、系统集成商 / 资本开支预算 | Skild 可激活的装机基础;硬件 CAGR 决定可触达机器人保有量增长 |
| 仓库自动化系统 | 输送线、AS/RS、机器人拣选系统、WMS 软件、车队编排 | 传统叉车作业、人工拣选作业 | 物流运营商、3PL、大型零售商 / 资本开支 + 运营开支预算 | Zebra 交易后,Skild 直接竞争端到端仓库自动化软件 |
| 人形机器人平台 | 全尺寸人形硬件 + 嵌入式 AI、双足执行、长时程任务执行 | 外骨骼、假肢、医疗康复机器人 | 工业 OEM、消费电子公司 / R&D + 资本开支预算 | 长期期权:Skild 的全形态模型瞄准人形机器人 OEM,向其授权 SDK |
市场边界反映 Skild 截至 April 2026 的主要商业定位。包含 / 不包含支出的区分由作者基于已发布的 公司描述和分析师范围披露定义;部分厂商跨越边界(例如 Skild 在 Zebra 之后的仓库平台横跨软件 + 编排)。预算归属会随公司规模和细分市场变化。
[CM001, CM002, CM003, CM004, CM014, CM015]2.2 市场规模分析
物理 AI / 具身智能软件的市场规模测算很难,因为各研究机构的范围定义差异很大。Grand View Research 估算,具身 AI 市场 2025 年为 $4.67B,到 2033 年增至 $67.6B,CAGR 为 39.7%。MarketsandMarkets 给出的 2025 年锚点相近,为 $4.44B,但 2030 年目标更保守,为 $23.1B,CAGR 为 39.0%。这些估算代表了最贴近 Skild 的可服务软件口径。更宽的市场估算——工业机器人硬件(2024 年 $16.9–34.0B)和仓库自动化系统(2025 年 $19.2–30.0B)——包含 Skild 不直接捕获的硬件支出,但定义了它可激活的机器人存量基础。 按 TAM/SAM/SOM 看:Skild 的 TAM 是全球具身 AI 软件市场,2025 年为 $4.4–4.7B,并在 2030–2033 年走向 $23–68B。它的 SAM 是面向工业和物流机器人的企业机器人 AI 软件市场,是 TAM 中剔除消费、医疗和国防保密细分后的子集,2025 年估计约 $2–3B。SOM 覆盖仓库自动化、离散制造和设施巡检等垂直领域,Skild 已在这些领域有活跃合作或部署;当前估计为 $200–500M,并会随着 Zebra 收购带来企业分发而快速扩大。 人形机器人预测提供了另一条更长期的市场向量:Goldman Sachs 预计人形机器人 2025–2028 年可在工厂落地,2030–2035 年进入消费应用;Morgan Stanley 估算,到 2050 年,人形机器人生态总规模为 $5 trillion。Skild 的全形态模型架构专门面向人形机器人平台设计,因此无论哪家人形机器人 OEM 率先规模化,它都可能受益。此类长期预测存在很大的分析师不确定性,应被视作情景数据,而非基准情景。 2025 年全球机器人风险投资融资总额达到 $13.8B,高于 2024 年的 $7.8B,显示投资人对各细分市场近期机器人采用保持强烈信心。 [CM006, CM007, CM008, CM009, CM010, CM011]
| 发布方 | 发布时间 | 地域 | 数值(基准年) | 数值(预测) | CAGR | 方法备注 | 置信度 | 关键限制 |
|---|---|---|---|---|---|---|---|---|
| Grand View Research | 2024 | 全球 | $33.96B (2024) | $60.56B (2030) | 9.9% | 自下而上按机器人出货量 + 软件测算;包含硬件、软件、服务 | 中 | 范围宽于纯软件口径;硬件占比高 |
| MarketsandMarkets | 2024 | 全球 | $16.89B (2024) | $29.43B (2029) | 11.7% | 硬件中心口径;范围更窄,聚焦工业机器人本体 | 中 | 因排除软件显著低于 GVR;口径不匹配 |
| Grand View Research | 2024 | 全球 | $19.23B (2023) | $59.52B (2030) | 18.7% | 仓库自动化系统,包含硬件、软件、WMS、机器人 | 中 | 包含硬件;Skild 只捕捉软件 / 智能层 |
| Mordor Intelligence | 2025 | 全球 | $29.98B (2025) | $65.74B (2031) | 13.98% | 包含移动机器人(占市场 41.4%)、件拣选(15.27% CAGR)、软件 | 中 | 基准年高于 GVR;方法不同;完整方法未公开 |
| MarketsandMarkets(经 PRNewswire) | 2025 | 全球 | $4.44B (2025) | $23.06B (2030) | 39.0% | 软件中心口径;具身 AI 包含机器人 AI、自主系统、智能家电 | 高 | 定义较宽,纳入非机器人自主系统;抬高 TAM |
| Grand View Research | 2025 | 全球 | $4.67B (2025) | $67.63B (2033) | 39.7% | 具身 AI 硬件 + 软件;物流与供应链细分增速最快,CAGR 42.2% | 高 | 包含硬件部分(收入占比 51.2%);仅软件 SAM 更小 |
| Goldman Sachs | 2024 | 全球 | 工厂可行(2025–2028) | 消费者场景可行(2030–2035) | N/A | 采用路径分析;并非单一 TAM 数字;成本需每年下降 15–20% | 低 | 仅情景分析;没有单一市场规模数字;2035 年 $38B 数字来自 GS 研究笔记 |
| Morgan Stanley | 2025 | 全球 | 早期 | 2050 年达到 $5T | N/A | 覆盖整个生态(硬件 + 软件 + 供应链 + 维护);1B+ 台情景 | 低 | 50 年周期;不确定性极高;单价假设($200K→$50K)未经验证 |
所有数字均来自面向公众的报告摘要。完整方法论报告在分析师付费墙后;置信度评级反映范围清晰度和交叉验证程度。MarketsandMarkets 和 GVR 是该领域被引用最多的两家公司,也因范围差异呈现最大分歧。Goldman Sachs 和 Morgan Stanley 使用采用情景框架,而非传统市场规模测算,直接与其他行比较会误导。CAGR 为 'N/A' 的行不应被解读为增长较慢。
[CM006, CM007, CM008, CM009, CM010, CM011]Skild AI 所处物理 AI / 具身智能软件市场的 TAM、SAM 与 SOM。TAM 采用 MarketsandMarkets 的具身 AI 市场估算。SAM 为作者估算的企业工业与物流 AI 软件市场。SOM 反映 Skild 在收购 Zebra 后活跃的商业细分。
[CM011, CM012, CM017, CM018, CM019]使用多家分析机构来源,对五个机器人相邻市场细分给出低 / 基准 / 高估算。区间反映各机构真实口径差异,而非预测不确定性本身。所有数值均为十亿美元,已四舍五入。工业机器人和仓库自动化包含硬件;具身 AI 数字以软件为中心。各行不可相加——市场存在重叠。
[CM006, CM007, CM009, CM010, CM011, CM012]2.3 买家与细分市场地图
Skild AI 的买家版图可分为五个清晰细分,每个细分都有不同买家、用户、预算所有者和采用触发因素。在仓库与物流——Skild 当前最活跃的商业垂直——主要买家是物流运营商(3PL)或大型零售商。付费方是希望降低人工成本、提升吞吐量的 CFO 或运营副总裁;用户是机器人机队经理。预算周期通常是多年期资本开支项目,软件金额常在 $500K–$5M 区间,并经常与硬件打包。2026 年 4 月收购 Zebra 让 Skild 直接接触这些企业买家。 离散制造领域的买家是工业 OEM 或 Tier 1 供应商。采用触发因素不是单纯成本,而是质量一致性、可重复性和劳动力可得性。预算归制造工程部门所有,采购需要安全认证,因此销售周期更长(12–36 个月)。人形机器人 OEM 合作仍处早期,但战略意义关键:Skild 通过 SDK 或云 API 提供机器人“大脑”,OEM 再将其授权到自家硬件中,类似移动 OS 授权模式。国防和安防买家通过政府项目办公室以多年期合同采购;超大规模云厂商和 REITs 的设施巡检则提供推进更快、毛利更高的账户,因为 24/7 覆盖替代夜班人工巡检,ROI 清晰。 [CM020, CM021, CM022, CM023, CM024, CM025]
| 细分市场 | 主要买方 | 终端用户 | 付款方 | 自动化覆盖的核心流程 | 预算归属 | 采用触发因素 |
|---|---|---|---|---|---|---|
| 仓储 / 物流 | 3PL 运营商或大型零售商(如 Amazon、DHL、XPO) | 机器人机队运营团队 | 运营副总裁 / CFO | 拣选包装、分拣、码垛、库存盘点 | 供应链 / 运营资本开支预算 | 劳动力短缺 + 吞吐需求;Skild-Zebra 捆绑方案提供端到端解决方案 |
| 离散制造 | 工业 OEM 或一级汽车 / 电子供应商 | 工艺 / 质量工程团队 | 制造副总裁 / 工厂经理 | 装配、质量检测、机床上下料、焊接 | 制造资本开支计划 | 质量一致性、可重复性、劳动力可得性;销售周期较长,为 12–36 个月 |
| 人形机器人 OEM | 人形机器人制造商(Figure、Agility Robotics、1X、Tesla Optimus 团队) | 机器人“大脑” / AI API 工程团队 | CTO / 研发预算 | 通用操作、运动控制、长时程任务规划 | 研发 / 产品工程预算 | 需要通用 AI 层;避免自建基础模型 |
| 国防 / 安防 | 美国 DoD、主承包商(Lockheed、Raytheon)、国家安全机构 | 任务操作员或自主系统项目办公室 | 政府项目办公室 / FFRDC 预算 | 周界巡逻、ISR、争议环境物流、EOD | DoD 项目预算;In-Q-Tel 管线释放兴趣信号 | 自主系统任务要求、兵力缩减、高风险环境作业 |
| 设施巡检 | 超大规模云厂商(AWS、Google、Azure 数据中心)或商业 REIT 管理方 | 设施 / 数据中心运营团队 | 基础设施副总裁 / 设施经理 | 7×24 安保巡逻、HVAC 巡检、火灾探测、设备监控 | 设施运营开支 / 安保预算 | 夜班人力成本、7×24 可用性要求、多站点规模化 |
买方和付款方估计来自 Skild 公开披露的用例、Zebra 收购公告以及行业一般做法。预算区间由作者按行业基准估算;单个部署合同未公开披露。人形机器人 OEM 一行反映新兴渠道,尚未为 Skild 贡献实质收入。
[CM020, CM021, CM022, CM023, CM024, CM025]映射 Skild AI 五个主要商业细分中的买方、用户、付款方和采用触发因素。单元格基于公开披露和行业惯例,呈现已观察或估算特征。
[CM020, CM021, CM022, CM023, CM024, CM025]物理 AI 部署的价值链,展示从机器人硬件制造到企业部署的参与方。Skild 位于硬件 OEM 与企业运营方之间,充当 AI 智能层。关键是,运营部署会产生真实世界数据飞轮,并回流给 Skild 改进模型——这是结构性竞争优势。
[CM020, CM021, CM022, CM029, CM033]2.4 增长驱动与采用约束
最持久的增长驱动来自结构性劳动力短缺:US Chamber of Commerce 报告目前有 1.7M+ 个制造业岗位空缺,National Association of Manufacturers 预计到 2030 年空缺将达 2.1M 个。这不是周期性缺口——人口结构趋势和制造业回流要求会让自动化需求在可预见未来保持结构性高位。电商增长进一步放大这一点:仓库自动化软件细分到 2031 年的 CAGR 为 14.87%(Mordor Intelligence),快于硬件,验证了“稀缺价值正在从钢铁转向智能”的投资逻辑。 第二个主要驱动是 AI 基础模型突破。大型多模态模型第一次能够以极少重新训练跨机器人形态和任务泛化。Skild 最早期支持者 Sequoia Capital 明确把这笔投资类比为语言 AI 的 GPT-3 时刻——通用智能架构的到来。单件拣选机器人是商业价值最高、技术难度也最高的子任务之一,预计到 2031 年 CAGR 为 15.27%(Mordor Intelligence),验证了市场对 Skild 灵巧操作能力的需求。 主要约束包括训练数据稀缺(MarketsandMarkets 将其列为关键市场挑战)、高初始资本开支强度(仅工业机器人硬件就需 $15K–$75K,尚未计入集成),以及在非结构化环境中部署 AI 系统的复杂性(需要专门的机器人集成商和大量操作员培训)。医疗和国防的安全认证周期可延长至 24–48 个月,限制受监管细分的近期收入。一个结构性尽调问题仍然存在:开源基础模型(Google RT-2、OpenVLA)是否会在 Skild 夯实数据飞轮优势和客户切换成本之前,更快地把 AI 层商品化。 [CM027, CM028, CM029, CM030, CM031, CM032]
| 因素 | 方向 | 时间 | 对 Skild 的影响 | 尽调问题 |
|---|---|---|---|---|
| 劳动力短缺(结构性) | 驱动因素 | 现在 – 持续 | 美国制造业 1.7M 个岗位空缺,为 Skild 目标细分市场带来持续自动化需求 | Skild 赢单时,主要驱动因素是劳动力成本,还是能力驱动的替代? |
| AI 基础模型突破 | 驱动因素 | 现在 – 2027 | 通用机器人 AI 无需逐任务 ML 工程即可迁移任务,移除了大规模部署的关键障碍 | Skild 的模型性能相较 Google RT-2、OpenVLA、Physical Intelligence π0 表现如何? |
| 电商吞吐需求 | 驱动因素 | 现在 – 2030 | Amazon 定下的履约标准迫使 3PL 自动化;仓储自动化软件以 14.87% CAGR 增长 | Skild-Zebra 管线中,电商客户相较其他垂直行业贡献多大比例? |
| 自动化价值向软件转移 | 驱动因素 | 2025 – 2030 | 仓储自动化的软件部分增长快于硬件(14.87% vs. 整体 13.98% CAGR),验证智能层论点 | 在竞争对手的硬件捆绑压力下,Skild 能否拿到软件利润率? |
| 单件拣选机器人增长 | 驱动因素 | 2025 – 2031 | 单件拣选是增长最快的机器人子细分市场,CAGR 为 15.27%;Skild 的灵巧操作 AI 正面切入这一需求 | Skild 的单件拣选成功率相较行业基准如何(通常要求 >99%)? |
| 训练数据稀缺(约束) | 约束 | 现在 – 2028 | 真实世界机器人训练数据昂贵且专有;MarketsandMarkets 将其列为 #1 市场挑战 | Skild 已累积多少机器人小时训练数据?它是否构成可防御 IP? |
| 初始 CapEx 高(约束) | 约束 | 现在 – 2028 | 加上 AI 软件前,工业机器人系统成本已达 $15K–$75K+;若没有融资方案,中小企业市场基本难以触达 | Skild 是否提供 RaaS(Robotics-as-a-Service)或融资选项,以降低 CapEx 门槛? |
| 集成复杂度(约束) | 约束 | 现在 – 2029 | 部署需要机器人工程师、生产经理和 IT 协同;合格集成商稀缺 | Skild 的合作生态中有多少认证系统集成商? |
| 安全与监管壁垒(约束) | 约束 | 现在 – 2030 | 医疗和国防部署需要 24–48 个月完成安全认证;收入确认因此推迟 | Skild 的 AI 系统已通过哪些监管标准(ISO 10218、IEC 62443)? |
| 开源 AI 商品化风险(约束) | 约束 | 2026 – 2029 | Google RT-2、OpenVLA 和 Physical Intelligence π0 已公开可用;商品化可能压缩 Skild 的软件利润率 | 相较标准化基准上最好的开源模型,Skild 性能优势是什么? |
方向、时间和影响由作者根据公开市场数据与投资者评论评估。约束严重度评级为定性判断。尽调问题截至本文写作时仍未解决。
[CM027, CM028, CM029, CM030, CM031, CM032]03竞争格局
3.1 竞争格局概览
Skild AI 所处的物理 AI 软件市场正在快速成形,竞争可分为五个清晰层级。第一层也是最相关的一层,是直接的机器人基础模型同行——构建通用机器人 AI 模型的创业公司,其中最突出的是 Physical Intelligence(π.ai),到 2025 年末已融资 $1.07B、估值 $5.6B,并开源了 π₀ 模型架构。第二层是来自超大规模云厂商和半导体公司的平台威胁:NVIDIA 的 Isaac GR00T N1 于 2025 年 3 月开源发布,借助 NVIDIA 在 GPU 上的主导地位把机器人 OEM 拉入其生态;Google DeepMind 的 Gemini Robotics 套件(2025 年 3 月发布)则利用 Alphabet 的前沿 AI 研究基础设施,以及与 Apptronik、Boston Dynamics 的既有 OEM 合作。第三层是垂直一体化人形机器人厂商——Figure AI(融资 $2B+,2025 年 9 月估值 $39B)、Agility Robotics(Amazon 持多数股权)、1X Technologies 和 Tesla Optimus——它们自建 AI 智能层,可能封锁 OEM 渠道。第四层是传统机器人龙头(ABB、Fanuc、KUKA、Yaskawa),它们在 300,000+ 台存量机器人基础上叠加软件和 AI。第五层是开源和内部自研替代方案:OpenVLA、HuggingFace LeRobot、Amazon 对 Covariant IP 的部署,以及 Tesla 内部 Optimus 项目。Skild 的 $14B 估值和 $2B+ 融资反映出投资人相信,一个中立、跨形态的 AI 平台可以整合碎片化机器人 AI 市场——类似 Android 在移动生态中的角色——但竞争强度也在随市场机会同步上升。2026 年 2 月,Alphabet 将机器人软件子公司 Intrinsic 并回 Google;再加上 OpenAI 在 2025 年推出专门的机器人部门,说明最大型 AI 组织已把物理 AI 视为核心产品领域,而不只是投资主题。
| 竞争对手 | 类别 | 融资 / 规模 | 目标细分市场 | AI 路线 | 相较 Skild 的关键优势 | 相较 Skild 的关键限制 |
|---|---|---|---|---|---|---|
| Physical Intelligence(π.ai,竞争者) | 直接同业 | 已融资 $1.07B;估值 $5.6B(2025 年 11 月) | 跨形态操作;研究 + OEM 授权 | π₀ VLA(3B 参数 PaliGemma + 流匹配);通过 openpi 开源 | 开源生态有牵引力;CapitalG(Alphabet)支持;学术社区强 | 没有企业分销;没有系统集成打法;披露数据集小于 Skild 声称规模 |
| NVIDIA GR00T / Isaac | 平台威胁 | 市值约 $3T;机器人业务是战略增长部门 | 人形机器人 OEM、机器人实验室、任何搭载 NVIDIA GPU 的机器人 | GR00T N1 双系统(VLM + 扩散),开源;Isaac Lab / Sim / Cosmos 生态 | 计算基础设施占主导;机器人训练事实上的 GPU 标准;免费模型推动 OEM 采用 | 不是纯 AI 软件公司;GPU 收入模式限制其对 AI 模型本身收费 |
| Google DeepMind | 平台威胁 | Alphabet(市值约 $2T);Intrinsic 已整合(2026 年 2 月) | 人形机器人 OEM 伙伴(Apptronik、Boston Dynamics);企业制造 | Gemini Robotics VLA + ER;Intrinsic IVM + Flowstate;端侧版本(2025 年中) | 研究算力近乎无限;领先 VLM 基础模型(Gemini 2.0);叠加 Intrinsic 工业 SDK | 过去未能把机器人研究商业化;截至 2025 年商业部署仍有限 |
| Figure AI | 垂直整合(硬件 + 软件) | 已融资 >$2B;估值 $39B(2025 年 9 月) | 汽车制造(BMW);物流 | Helix 专有 AI 平台 + Figure 02 人形机器人;BotQ 制造设施 | 资本体量巨大(估值 $39B);BMW 商业部署;垂直整合形成全栈护城河 | AI 绑定特定硬件(不跨形态);高资本开支;OpenAI 合作破裂释放不稳定信号 |
| Covariant | 直接同业(仓储 AI) | 新融资 $100M(2025 年 2 月);Amazon 人才收购后重组 | 仓储 / 物流拣选与操作 | RFM-1 仓储基础模型;Amazon 非独家 IP 授权 | 最大仓储专用真实世界训练数据集;Amazon 授权验证模型质量 | 组织不稳定;创始团队转入 Amazon;人才密度下降 |
| Apptronik | 相邻玩家(人形机器人硬件 + Google AI) | 累计融资 $935M(2026 年 2 月);估值约 $5–5.5B | 企业制造(Mercedes-Benz、GXO、Jabil、John Deere) | Apollo 人形机器人集成 Gemini Robotics-ER;硬件 + Google AI 软件捆绑 | Google DeepMind AI 合作面向 OEM 拼出强大的硬件 + 基础模型组合 | 不是 AI 软件平台;依赖 Google DeepMind;范围窄于 Skild 的跨形态覆盖 |
| Unitree / AgiBot | 相邻玩家(低成本硬件 + AI) | Unitree:2025 年收入 $235M(同比 335%);AgiBot:估值 $2.1B(2025 年) | 制造、物流(Unitree);中国工业(AgiBot);全球扩张中 | Unitree G1 价格 $13.5K–$21.5K;AgiBot ViLLA/GO-1;快速放量 | 硬件价格大幅更低;2025 年 Unitree 出货 5,500+ 台、AgiBot 出货 5,100+ 台 | AI 平台能力验证较少;数据主权担忧;西方市场准入受限 |
| ABB / KUKA / Fanuc | 在位者(传统机器人 OEM) | ABB:2024 年机器人收入 $2.3B;全球市场份额并列第一 | 工业制造、汽车、电子(企业 OEM 渠道) | OmniCore AI-ready 控制器(ABB);FANUC Physical AI 倡议(2025 年 12 月);支持 ROS2/Python | 300,000+ 台机器人装机基数;工业市场份额 70%+;企业信任与合规基础 | 软件能力受限;AI 能力只是附加项,不是基础;适应基础模型范式较慢 |
融资数字来自最新披露轮次;估值为已报道轮次的投后估值。'目标细分市场' 反映主要商业焦点。NVIDIA 和 ABB 属于大型公司旗下部门;如有数据,融资 / 估值指机器人部门。
[CP002, CP003, CP004, CP005, CP006, CP012]Skild AI 与七个主要竞争对手在两个轴上的序数定位。横轴:跨机体泛化性(0–10,10 = 单一模型兼容最广机器人且无需重新训练)。纵轴:企业部署就绪度(0–10,10 = 完整企业 SLA、分销网络和规模化创收部署)。分数由作者基于截至 2026 年 Q1 的公开披露和产品发布评估。没有独立基准验证这些序数分数。
[CP001, CP002, CP003, CP004, CP005, CP006]3.2 竞争对手画像:直接同行与平台威胁
Physical Intelligence(π.ai)由前 Google Brain 和 Berkeley 机器人研究员创立,是 Skild 在理念和商业上最接近的同行。它的 π₀ 架构把 3B 参数 PaliGemma VLM 主干与 300M 参数、使用流匹配的动作专家结合起来,并用来自 7–8 个机器人平台、68+ 项任务的 10,000+ 小时示范数据训练。开源的 “openpi” 仓库带来强学术社区参与,但也带来商品化风险:免费释放模型权重,等于表明价值在训练数据和部署,而不在架构本身——这与 Skild 自身定位一致。PI 到 2025 年底累计融资 $1.07B(2024 年 11 月 $400M Series A,2025 年 11 月 CapitalG/Alphabet 领投 $600M Series B)。Alphabet 持有 CapitalG,意味着它在 Skild 最接近的直接竞争对手中有财务权益。 NVIDIA 的 GR00T N1 是性质完全不同的威胁:NVIDIA 不向 AI 模型收费,而是把 GR00T 作为免费开源产品,推动 GPU 硬件采用和 NVIDIA Cosmos 仿真订阅。双系统架构(System 2 VLM 负责推理,System 1 扩散 Transformer 负责实时动作)于 2025 年 3 月以宽松许可证发布。早期采用者包括 1X Technologies、Agility Robotics、Boston Dynamics、Mentee Robotics 和 NEURA Robotics——基本覆盖领先人形机器人 OEM。Google DeepMind 的 Gemini Robotics(2025 年 3 月)同时覆盖通用 VLA,以及具备深度 3D 空间理解的 Gemini Robotics-ER。合作伙伴包括 Apptronik(Apollo,截至 2026 年 2 月融资 $935M)、Boston Dynamics Atlas 和 Agility Robotics Digit。 在垂直一体化人形机器人厂商中,Figure AI 资本最充足:2024 年 3 月以 $2.6B 估值完成 $675M Series B,随后在 2025 年 9 月以 $39B 估值完成 $1B+ Series C。其 “Helix” AI 平台为自研且绑定硬件,2024 年 12 月首次向 BMW 商业交付。不过,Figure 在 2025 年终止了与 OpenAI 的 AI 合作,而 OpenAI 随后推出自有机器人部门——前合作伙伴由此变成直接竞争对手。Covariant 的 RFM-1 仓库 AI 基础模型曾是先行者,但 Amazon 在 2024 年对其创始团队做收购式招聘(并取得 Amazon 的非独占 IP 许可),从根本上削弱了公司;Covariant 于 2025 年 2 月融资 $100M 后正在重建。Covariant 的先例——大客户通过人才收购实质性掏空供应商——是 Skild 企业战略的重大风险。
| 能力标准 | Skild AI | Phys. Intel.(π.ai,竞品) | NVIDIA GR00T | DeepMind | Covariant | ABB / Fanuc |
|---|---|---|---|---|---|---|
| 跨形态通用性 | 高(声称) | 高(已发表) | 中(聚焦人形机器人) | 高(多平台) | 低(仅仓储) | 无(OEM 专用) |
| 专有数据集规模 | 高*(1,000x 声称,未经验证) | 中(10K+ 小时,8 个平台) | 高(合成 + 真实,Omniverse) | 高(多年研究数据) | 中(仓储数据) | 低(机器人专用日志) |
| 企业支持 / SLA | 高(Zebra 后) | 低(研究导向) | 高(NVIDIA Enterprise) | 低(少数测试方) | 中(重建中) | 高(在位者) |
| 开源可用性 | 无(专有) | 高(openpi 权重) | 高(GR00T 宽松许可) | 无(选择性开放) | 无(专有) | 无(OEM 专有) |
| 真实世界部署足迹 | 中(Zebra 后估计) | 低(研究试点) | 低(通过 OEM 伙伴) | 低(通过 OEM 伙伴) | 高(Amazon 仓库) | 高(装机基数) |
| 硬件无关性 | 高(声称) | 高(多机器人) | 低(需要 NVIDIA GPU) | 中(Apptronik + Atlas) | 中(仓储机器人) | 无(专有硬件) |
评级由作者根据截至 2026 年 Q1 的公开产品披露和研究论文评估。Skild 的数据集评级('高*')基于公司声称的 '1,000x' 数字,该数字尚未经过独立基准测试——以星号标注。标为 'U' 的单元格表示缺少公开披露,并非确认没有能力。
[CP010, CP011, CP012, CP013, CP014, CP023]Skild AI 与五个竞争平台在六项评估标准上的能力覆盖和强度。评级为高、中、低、无;Skild 的数据集分数标为高*(公司声称,未经独立基准验证)。基于截至 2026 年 Q1 的公开产品披露和研究发表。
[CP010, CP011, CP012, CP013, CP014, CP023]3.3 对比分析:能力、定价与分发
跨形态通用性——不针对单一机器人重新训练,也能在任意机器人形态上运行——是 Skild 声称的主要差异化。Sequoia 合作公告称,Skild 的数据集“比多数竞争对手大 1,000 倍”。这个说法很惊人,但没有独立基准测试验证:Physical Intelligence 的 π₀ 在 8 个机器人平台、68+ 项任务的 10,000+ 小时示范数据上训练;NVIDIA GR00T 则用真实轨迹加 Omniverse 大规模生成的合成数据训练。没有已发布的头对头基准测试能验证 Skild 的数据优势。企业就绪度上,收购 Zebra 之后,Skild 相比 PI 和 DeepMind 有实质优势:Fetch Robotics AMR 机队、企业 WMS 集成,以及 Zebra 销售团队,提供了纯 AI 研究型组织缺少的商业基础设施。 竞争格局中的定价大多不透明。Skild 的企业交易没有公开披露;可比 AI 机器人平台合同显示,机队部署的年合同额可能为 $200K–$5M+。NVIDIA GR00T 免费,变现靠 GPU 硬件和云计算。Covariant 未发布定价。ABB 等传统龙头把软件与硬件打包,以混合毛利变现。中国竞争对手(Unitree G1 每台 $13.5K–$21.5K;AgiBot GO-1)代表市场最低硬件单台成本。Unitree 在 2025 年出货 5,500+ 台人形机器人,实现 $235M 收入(同比 335%);AgiBot 出货 5,100+ 台,估值 $2.1B——但它们的企业 AI 平台能力仍未充分建立,并在西方市场面临数据主权顾虑。 分发能力是关键战场。ABB 全球 300,000+ 台存量机器人构成 AI 创业公司难以复制的分发护城河。Skild 收购 Zebra 是直接反制:通过收购企业机器人基础设施,Skild 获得客户入口,而不是只依赖 OEM 合作。Figure AI 的 BMW 试点、Agility 在 Amazon 和 GXO 的部署(2025 年达到 100,000 次周转箱搬运里程碑),以及 Apptronik 与 Mercedes-Benz、John Deere 的合作,都形成了分发锁定;Skild 必须用自己的企业销售能力去竞争。
| 公司 / 平台 | 定价 / 变现模式 | 入门价格(估计) | 企业合同(估计) | 竞争含义 |
|---|---|---|---|---|
| Skild AI | 企业软件许可 + 专业服务;按机队或按机器人签年度合同 | $50K–$200K 试点 / POC | 大型机队每年 $500K–$5M+ | 企业定价支撑资本效率,但必须相较免费 / 更便宜替代方案证明 ROI 清晰 |
| Physical Intelligence | 开放模型免费(openpi);商业企业许可未知 | 开放模型权重实际为 $0 | 未知;企业条款未披露 | 开放模型降低采用摩擦,但若没有专有企业层,变现受限 |
| NVIDIA GR00T | 模型免费(开源);收入来自 NVIDIA GPU 硬件、AI Cloud、Omniverse 仿真 | 模型 $0;Omniverse / Cloud 每年 $5K–$50K | H100 集群:资本开支 $100K+;Cloud:可变 | 免费模型带来定价压力;NVIDIA 变现基础设施,而非 AI 模型本身 |
| Covariant | 企业软件;Amazon 非独家授权;定价未公开 | Unknown | 未知;Amazon 获得 IP 访问后,可能无需向 Covariant 付费即可自助使用 | Amazon 的 IP 地位带来定价不对称风险,削弱 Covariant 谈判筹码 |
| ABB / KUKA | 硬件捆绑软件;OmniCore 控制器 + AI 附加模块;订阅式 AI 模块开始出现 | 软件通常随机器人硬件捆绑($10K–$100K/台) | AI 附加模块:每个工作单元每年 $10K–$100K | 捆绑定价为在位者制造切换成本,但限制 AI 能力溢价 |
| Unitree / AgiBot | 硬件单机以激进价格销售;软件 / AI 能力仍在发展 | 每台人形机器人 $13.5K–$21.5K(硬件) | 企业 AI 软件条款未披露 | 若中国机器人获得 AI 平台能力,价格压缩会威胁 Skild 的 OEM 伙伴经济性 |
Skild 定价完全由作者根据可比企业 AI 机器人平台交易估算。Physical Intelligence 和 Covariant 定价未公开披露。NVIDIA 通过硬件和云计算变现,而不是模型授权。中国厂商定价指机器人硬件单机,不是 AI 软件授权。null 或“未知”单元格表示缺少披露。
[CP005, CP006, CP011, CP014, CP017, CP018]截至 2026 年 Q1,Skild AI 的竞争耐久性简表。关键对比指标刻画 Skild 相对本章主要同行集合的竞争位置。
[CP002, CP004, CP009, CP030, CP034, CP040]3.4 护城河耐久性、锁定与替代风险
Skild 最可信的护城河是数据飞轮:真实机器人部署产生专有训练数据,且据称无需人工标注,从而持续放大数据集优势。Zebra 收购又加入了一支 AMR 机队,在企业规模上产生物流和拣选数据。不过,这条护城河面临三类结构性威胁。第一,资源雄厚的竞争对手开源——NVIDIA GR00T N1(宽松许可证)、Physical Intelligence openpi 权重——可能把模型架构层商品化,把价值完全推向数据。第二,垂直一体化玩家(Amazon 通过 Covariant IP + Agility Robotics,Figure AI 的 Helix + BotQ 制造,Tesla Optimus 自研)正在封闭生态内部打造自我强化的数据飞轮,可能永久把 Skild 排除在其硬件渠道之外。第三,中国竞争对手(Unitree 每台 $13.5K–$21.5K,AgiBot)正在以威胁 Skild OEM 合作伙伴经济性的规模出货。 企业切换成本是一条关键但滞后的护城河。一旦 Skild 模型进入机器人机队的控制栈——包含专有 API、Zebra WMS 集成和按任务微调——迁移到竞争对手估计需要 6–18 个月工程投入。这会形成真实锁定,但只发生在初始部署之后。因此,最早的一批企业合同胜利对长期护城河建设格外重要。竞争替代风险最高来自 NVIDIA:如果 GR00T N1.x 在跨形态泛化上追平 Skild(它正通过 N1.5、N1.6、N1.7 更新快速改进),模型层可能成为嵌入 GPU 硬件的商品,压垮定价权。Google DeepMind 通过研究转产品路径带来类似风险,不过 Alphabet 过去商业化机器人研究的困难提供了一些安慰。最不利的信号是:OpenAI 在 2025 年决定直接进入机器人领域,此前它曾投资 Figure AI 和 Physical Intelligence;这表明全球最强的 LLM 组织把物理 AI 视为核心产品。如果 OpenAI 成功把 LLM 分发转化为机器人智能平台,Skild 的定位可能面临生存级威胁。
| 护城河主张 | 主要威胁 | 严重度 | 影响时间(估计) | 缓解措施 / 尽调问题 |
|---|---|---|---|---|
| 机器人无需人工标注即可生成训练数据,数据集优势持续复利 | 开源模型商品化;NVIDIA Cosmos 大规模合成数据;Open X-Embodiment 数据集 | 高 | 2–4 年 | 独立验证实际数据量;确认“零人工介入”的标注质量 |
| Zebra 收购带来企业客户、WMS 集成和企业销售队伍 | 集成风险:Zebra AMR 机队架构不同于 Skild AI 模型 | 中 | 1–3 年 | 跟踪 Fetch/Zebra 机队向 Skild AI 模型迁移;确认交叉销售收入相较独立销售的贡献 |
| 深度 API + 微调集成让每个客户迁移成本预计达到 6–18 个月 | 竞争对手 API 拉平;NVIDIA GR00T N1.x 功能趋同;OpenAI 机器人平台入场 | 中 | 2–5 年 | 按参考客户梳理集成深度;验证迁移成本估算 |
| 声称无需重新训练即可跑在任何机器人上;数据集比竞争对手大 1,000x | NVIDIA GR00T N1.x 持续迭代;Google DeepMind Gemini Robotics 支持多平台 | 高 | 1–3 年 | 委托独立基准测试:在标准操作任务上比较 Skild、π₀、GR00T N1.5 |
| 深厚资本支撑持续 R&D、企业销售和收购策略 | Figure AI(估值 $39B)和 Apptronik(融资 $935M)资本体量相当或更大;中国竞争对手有国家资本支持 | 中 | 持续 | 跟踪烧钱速度与 ARR 爬坡;确认 Series D 资金续航相较竞争对手资本投放是否足够 |
| 中立 AI 平台定位吸引不想依赖硬件竞争对手的机器人 OEM | 垂直整合竞争对手(Figure、Tesla、Amazon/Agility)缩小 OEM 伙伴池;NVIDIA GR00T 创造另一个 OEM 标准 | 高 | 2–4 年 | 梳理现有 OEM 合作关系;核实排他条款;逐家 OEM 评估自研 AI 风险 |
严重度评级(高 / 中 / 低)由作者根据截至 2026 年 Q1 的竞争证据评估。'影响时间' 为估计值;前沿模型能力提升可能使实际动态加速。
[CP025, CP030, CP031, CP033, CP034, CP040]04财务情况
4.1 收入模型与定价
Skild AI 的主要收入机制,是通过云端 API 和 SDK 向企业客户授权 Skild Brain 基础模型。客户——机器人 OEM、系统集成商和大型企业运营方——付费获得机器人智能层,而不是自研专有控制系统。这把 Skild 放在机器人技术栈中的横向软件平台位置,类似硬件厂商接入的操作系统。 公司公开描述的收入来源包括:(1)面向机器人硬件厂商的基础模型授权;(2)面向安防巡检、仓库编排和制造的垂直软件模块;(3)用于推理和训练的云基础设施服务;(4)收购 Zebra 之后,端到端仓库自动化,把 Skild Brain 与 Zebra 的 Symmetry Fulfillment 编排平台结合起来。Skild 官网描述了一个 “Mobile Manipulation Platform”,把抓取、交接、导航等技能抽象为 API 调用,显示出一种可编程接口模式,企业开发者可在其上构建应用。 精确定价没有公开披露。基于 Sacra 分析和可比企业 AI 平台基准,定价很可能采用多年期企业订阅,按单台机器人或机队层级授权,并可能叠加基于用量的云计算费用。2025 年企业机器人 AI 平台标价通常为每台机器人每年 $10,000–$100,000+,取决于能力和批量折扣。 公司披露 2025 年收入约 $30M——“短短几个月”内从零增长而来。这个数字由公司自述,尚未被独立审计或第三方确认。$30M 意味着少数大型企业合同(几十到数百台机器人、高 ACV),或更多小型交易;两种情况公司都没有公开拆分。Series C 和 Zebra 收购后,管理层称公司正在“指数级增长”,释放出 2026 年收入大幅加速的预期。 [CI001, CI002, CI003, CI004, CI011, CI012]
| 收入来源 | 机制 | 计价单位 | 当前状态 / 价值 | 收入质量 | 尽调问题 |
|---|---|---|---|---|---|
| 基础模型授权(Skild Brain API) | 机器人 OEM 和企业运营方付费调用跨本体 Skild Brain 模型 API,用于机器人控制 | 按机器人机队或企业授权计费(年度订阅;具体价格未披露) | 已上线;主要收入来源;贡献 2025 年约 $30M 收入 | 公司声称;未经审计 | 核实合同结构、ACV 区间、活跃合同数量、续约率 |
| 垂直软件模块 | 基于 Skild Brain 构建的行业能力层(安防巡检、仓储、制造) | 增购订阅,或与基础授权打包 | 已上线;多个垂直场景已部署(仓储、安防、制造、配送、数据中心、建筑) | 公司声称;未经审计 | 核实各垂直场景收入占比;模块是增量 ACV,还是打包销售? |
| 云端推理与训练服务 | 为客户部署托管推理服务和微调;AI 工厂私有云打包方案 | 按用量计费的云计算 + 软件毛利 | 已上线(Sacra 报告描述);公开财务未单列披露 | 根据产品描述和分析师报告推断 | 获取各收入流毛利率;拆分云端 COGS 与纯软件经济性 |
| Zebra Symmetry 履约平台 | 收购后获得的仓库编排平台,已有企业客户基础;协调机器人机队 + 一线员工 | 企业 SaaS;假设现有 Zebra 客户合同可转移 | 2026 年 4 月收购;尚未纳入已披露财务 | 第三方报道;未披露收入数字 | 核实收购时 Symmetry ARR;所有权转移后的客户留存 |
| 移动操作 SDK / API(开发者) | 面向抓取、交接和导航的可编程 API;支持第三方基于 Skild Brain 开发应用 | 开发者层级定价(可能按用量或按席位);未公开披露 | 已在 skild.ai 上线;具体采用情况未披露 | 已观察到(产品页);未披露收入归因 | 核实开发者层级是否产生实质收入,还是仅作为销售管道 / 线索获取渠道 |
收入按来源拆分未公开披露。约 $30M 的 2025 年收入数字由公司给出且未经审计;各收入流归因属于估算。Zebra Symmetry 的收入未计入 $30M 数字(收购于 2026 年 4 月完成)。所有收入流定价均未披露;估算基于 Sacra 分析和企业 AI 平台基准。
[CI001, CI011, CI012, CI013, CI014, CI018]| 层级 / 产品 | 标价估算 | 实际成交价说明 | 折扣 / 未知项 | 来源 |
|---|---|---|---|---|
| 企业基础模型授权 | 估计每台机器人每年 $10K–$100K+(机队定价;无公开标价) | 未披露;早期战略客户成交价可能低于标价 | 预计有批量折扣;试点价格可能明显偏离全面部署 ACV | Sacra 分析;企业 AI 平台基准;根据收入和员工数推断 |
| 垂直模块增购 | 未知;估计与基础授权打包,或每台机器人每年增量 $5K–$30K | 未披露 | 早期合同可能打包销售,以推动采用 | 根据 Skild 产品描述推断;无公开定价页 |
| 云端推理 / AI 工厂 | 按用量计费;估计每机器人小时推理 $0.10–$5.00(粗略基准;无 Skild 专属数据) | 未披露;早期部署可能按成本价或低毛利 | 可能由算力合作伙伴(NVIDIA)或 SoftBank 基础设施补贴 | 根据 AI 算力基准(Kruze Consulting)推断;无 Skild 专项披露 |
| Zebra Symmetry 平台(继承) | 既有企业 SaaS;Zebra 过往定价未公开披露 | 假设过渡期沿用 Zebra 原合同费率 | 收购后存在合同重定价风险;存量客户忠诚度不确定 | roboticsandautomationnews.com;Skild 收购公告 |
Skild AI 没有公开定价页。所有定价估算均为分析师推导的基准。实际成交价(扣除折扣、试点和收入确认时点影响后)未知。Zebra Symmetry 定价反映 Zebra Technologies 的历史合同,Skild 现已继承这些合同。
[CI011, CI012, CI013, CI023, CI036, CI001]4.2 成本结构与单位经济
Skild AI 的成本结构由算力主导——大型机器人基础模型的训练和推理天然资本密集。2024–2025 年,前沿 AI 模型单次训练成本为 $30M–$200M+;按 Epoch.ai 数据,算力成本约每年 2.4x 增长(每 8–10 个月翻倍)。相较文本或视觉模型,机器人基础模型还需要额外数据基础设施:仿真环境、遥操作数据集、传感器融合管线和真实部署反馈回路,都需要持续工程和算力投入。 公司尚未披露毛利率、收入成本、运营费用或资本开支数据。结构上,云端 API 交付的软件平台模式在规模化后应有高毛利;企业 AI 平台 SaaS 的行业 benchmark 显示,60–80% 毛利率可以实现。但 $30M 的早期收入不足以摊销基础模型训练成本。推理服务成本、客户导入和现场部署支持,短期很可能把毛利率压到长期目标以下。 单位经济(CAC、LTV、流失率、回本周期)没有公开披露。企业机器人细分通常销售周期长(6–18 个月)、ACV 高;一旦机器人机队基于某一 AI 层完成训练和集成,切换成本也高。如果数据飞轮随着每次部署提升模型质量,既有客户获得复利式价值,客户终身价值(LTV)结构上会很高。不过,Skild AI 没有公开客户层留存数据、流失率或续约条款。 Zebra Technologies 收购增加了硬件集成复杂度:Symmetry Fulfillment 编排平台的毛利经济与纯软件授权不同。把机队级编排与 Skild Brain 集成,可能在企业仓库合同扩张时暂时推高收入成本,因为专业服务和部署支持会随之增长。 [CI020, CI021, CI022, CI023, CI029, CI030]
示例收入模型,展示企业机器人部署如何转化为 Skild AI 的确认收入和估算毛利,基于已披露和估算的财务输入。所有数字均为近似值;毛利率和成本估算由分析师推断,并非公司披露。
收入 $30M 为公司口径;40–60% 毛利率是基于早期规模的软件平台基准估算;所有其他项目均为分析师估算。未经审计。
[CI001, CI020, CI021, CI022, CI023, CI029]估算 Skild AI 在 Series C 阶段(2026 年)的年度经营现金流出构成,展示机器人基础模型公司的资本密集属性。所有数字均为分析师估算;公司未披露经审计拆分。
所有项目均基于 AI 初创公司基准、员工人数、办公布局和算力成本报告估算。公司未披露,也未经审计。
[CI001, CI007, CI018, CI019, CI020, CI021]4.3 资本充足性与融资结构
以公司年龄和阶段看,Skild AI 的融资速度极不寻常。融资时间线(见第 1 章 / 公司概况,并在此通过 claimRefs 引用)从 2023 年种子轮延伸到 2026 年 1 月 $1.4B Series C,共四轮。CEO Deepak Pathak 公开表示,公司累计融资超过 $2B;Crunchbase 跟踪到的数字是四轮合计 $1.83B。差异来自未披露的种子轮规模,以及尚未记录的任何分批交割或资本调用结构。 账上现金没有公开披露。假设 $1.4B Series C 已在 2026 年 1 月全额完成并进入公司资产负债表,募资总额提供了充足现金跑道。按每月 $10–50M 的估计烧钱速度(符合这一规模和抱负的 AI 研究与企业部署公司),仅 Series C 就意味着 28–140 个月的总现金跑道。更保守地按每月 $30–50M 烧钱估算,用于模型训练、员工(约 85+ 人)、办公室扩张和 Zebra 整合成本,则 2026 年 1 月交割后的现金跑道为 28–47 个月,可支撑公司到 2028–2030 年而无需追加资本。 公司披露的募资用途是:继续扩大模型训练,并推动技术的未来部署。没有披露销售 / 市场、R&D、G&A 或基础设施资本开支的具体分配。也没有公开披露债务融资、信贷额度或项目融资义务。Zebra 收购采用现金加股权结构(Zebra 获得 Skild AI 股份),意味着并非所有 Series C 资金都在收购交割时以现金流出。 投资人基础集中:SoftBank 领投或参与了种子轮之后的每一轮(A、B、C),估计在这些轮次中投入 $1B+。这带来财务集中风险——SoftBank 持续支持对后续融资很重要,SoftBank 投资姿态的任何变化都可能削弱 Skild 以有利估值进入资本市场的能力。NVIDIA、Samsung、LG 和 Zebra 等战略投资人也持有股权,带来一致性,也带来潜在治理复杂度。 [CI005, CI007, CI008, CI009, CI010, CI015]
| 项目 | 数值 / 状态 | 置信度 | 说明 |
|---|---|---|---|
| 账上现金(估计,Series C 后,2026 年 1 月) | 估计 $1.2–1.4B(Series C 募资扣除 Zebra 收购现金对价) | 低 | 精确数字未披露;Zebra 交易结构(现金部分)未披露;估计假设 $1.4B 中大部分进入资产负债表 |
| 月度总烧钱(估计) | 估计 $30–50M/月(模型训练、员工、办公室、基础设施、Zebra 整合) | 低 | 与同等规模 AI 研究公司一致;无公开披露 |
| 估计现金跑道(总额) | 估计自 2026 年 1 月起 28–47 个月(即覆盖至 2028 年中到 2029 年末) | 低 | 基于 $1.2–1.4B 现金和 $30–50M/月总烧钱;收入会部分拉长跑道 |
| 计划募资用途(公司表述) | 扩大模型训练;推动技术未来部署(Series C 新闻稿措辞) | 高 | 公司官方披露;未提供具体分配比例 |
| 债务 / 项目融资义务 | 无公开披露 | 中 | 私营公司;无公开债务文件;未披露信贷额度或项目融资安排 |
| 下一轮融资触发因素 | 未公开披露;可能是商业里程碑(ARR)、技术里程碑或时间驱动(2027–2028) | 低 | 未公开说明 Series D 时点或触发条件 |
| Zebra 收购现金流出 | 未披露;Zebra 获得现金 + 股权对价 | 低 | 现金部分未知;会压缩 Series C 募资形成的可用跑道 |
所有现金、烧钱和跑道数字均为估计;没有可用审计数据。Series C 于 2026 年 1 月 14 日完成;Zebra 收购于 2026 年 4 月完成。跑道估算假设公司在下一轮之前不再融资。历史轮次(种子轮、Series A、B、C)记录在本地论据 CI005–CI009;更多背景见第 1 章公司概况。
[CI005, CI007, CI008, CI009, CI010, CI030]4.4 公开指标与证据缺口
对于一家具有这种规模和投资人画像的公司,Skild AI 披露的财务指标异常有限。主要公开财务数据点是 2025 年约 $30M 收入,这一数字披露于 Series C 新闻稿和后续投资人沟通。公司没有提供 ARR、ACV、客户数、留存指标或队列数据。 $14B 估值的 $1.4B Series C,意味着相对于 2025 年尾随收入约 467x 的收入倍数。这个数字只有在几种假设下才说得通:(a)近期收入增长率极高(公司“指数级增长”的表述);(b)相信基础模型平台具备赢家拿走大部分的动态,并能支撑庞大最终装机基础;或(c)来自 Zebra 收购、IQT 国家安全信号和 NVIDIA 物理 AI 路线图的战略期权价值。上述假设都无法用公开数据验证。 仍为私密的关键指标包括:毛利率、运营亏损、月度烧钱速度、下一次融资触发条件、客户数、平均合同价值、流失率和净收入留存。公司是私营企业,不受公开报告要求约束,也没有提交注册声明。经审计财务报表没有公开。 Zebra Technologies 收购引入了来自 Symmetry Fulfillment 平台及其既有客户基础的新收入流,但双方都没有披露 Zebra 机器人部门的收入、利润率或 EBITDA 数据。Zebra Technologies 是上市公司,可能在 2026 年 4 月剥离后的 10-Q 或 8-K 文件中披露了更多细节。 [CI034, CI035, CI037, CI038, CI004, CI027]
| 缺失指标 | 对投资判断的影响 | 具体尽调路径 |
|---|---|---|
| 经审计收入和损益表 | 无法独立核实 $30M 收入数字、成本结构或盈利能力 | 向 CFO / 审计机构索取 GAAP 审计财务报表;获取四大或同等机构审计意见 |
| 按收入流拆分的毛利率 | 缺少收入流层级经济性,就无法评估盈利路径或资本效率 | 要求按产品线提供收入和 COGS(授权、云、专业服务、Zebra Symmetry) |
| 客户数量和集中度 | 收入集中度风险未知;$30M 可能来自 1 个客户,也可能来自 300 个 | 要求提供收入前 10 大客户、占总收入比例、合同期限和续约状态 |
| 合同结构和 ARR / ACV 明细 | 无法判断经常性收入与一次性收入、试点合同与生产合同的占比 | 要求按客户队列提供收入瀑布,列示新增 ARR、扩张 ARR、流失,以及已确认收入与递延收入 |
| 月度烧钱速度和现金余额 | 缺少当前数字,就无法评估现金跑道或资本充足性 | 要求提供截至尽调日期、经 CFO 认证的烧钱证明和银行对账单 |
| 收购时 Zebra Symmetry ARR | 缺少基线,就无法评估收入增厚或整合风险 | 要求提供收购时 Zebra Symmetry 客户清单、ARR 和合同到期时间表 |
| 单位经济模型(CAC、LTV、流失) | 无法评估 GTM 可扩展性或长期利润率轮廓 | 要求按渠道提供 CAC、平均销售周期、赢单率,并按队列提供 NRR 和 GRR |
| 资本配置明细(R&D vs S&M vs G&A) | 无法评估管理层烧钱效率或投资优先级 | 要求按职能拆分 opex;拆分算力 capex 与 opex;按部门提供员工数 |
这张表列出在当前 $14B 估值下,严肃投资决策所需的最低财务尽调议程。早期私营公司不公开这些维度的指标并不罕见,但每个缺口都是风险,必须在按当前估值做投资判断前解决。
[CI001, CI004, CI034, CI035, CI018, CI026]4.5 财务结论
Skild AI 的财务画像,是一笔对物理 AI 中软件平台经济学的高信念下注:早期收入牵引力很强,但尽调阻碍也很大。如果 $30M 收入真实且主要可经常性确认,它说明企业客户愿意为横向机器人 AI 层付费。对一家深科技基础模型公司来说,几个月内从零变现到 $30M 很罕见,也支撑了领投方提出的“机器人领域 GPT-3 时刻”投资逻辑。 近期资本充足性很强:即便按前沿模型训练和企业快速扩张所隐含的激进烧钱速度,$1.4B Series C 也提供了充足现金跑道。战略投资人(NVIDIA、Samsung、LG、Zebra、IQT)提供的不只是资本,还有分发、硬件入口和市场信号。 财务风险集中在三条线上:(1)收入质量——$30M 未经审计,可能来自非经常性里程碑确认,也可能集中在少数尚未转为经常性合同的试点;(2)利润率路径——算力成本呈指数级上升,维持模型质量所需的机器人数据基础设施,可能在公司规模化前显著压缩利润率;(3)估值锚点——$14B 估值相当于 2025 年收入 467x,不给商业执行风险或倍数压缩留下余地;若收入不能持续指数级增长,公司通向公开市场退出或二级市场流动性的路径会很难。 最重大的尽调缺口,是缺少经审计财务,以及对收入数字、客户数或合同结构的任何独立验证。任何严肃尽调流程在按当前估值出手前,都必须取得审计报表、按客户和队列拆分的详细收入瀑布,并看到至少 2–3 个季度持续经常性收入的证据。 [CI001, CI004, CI023, CI029, CI030, CI034]
| 指标 | 数值 / 估计 | 置信度 | 重要性 | 尽调问题 |
|---|---|---|---|---|
| 毛利率 | 未披露;估计规模化后(纯软件)为 60–80%;近期受早期算力和部署成本拖累,可能为 30–50% | 低 | 决定规模化后的盈利路径和资本效率 | 要求提供经审计损益表,并拆分收入成本 |
| 月度总烧钱速度 | 估计 $10–50M/月(根据员工数、办公室、算力规模和模型训练节奏推断) | 低 | 决定现金跑道和下一轮融资时点 | 获取 Series C 后月度现金流量表 |
| 月度净烧钱(扣除收入后) | 在 $30M ARR 和估算毛利率下,估计为 $7.5–47.5M/月 | 低 | 真实现金跑道指标;收入会部分抵消烧钱 | 要求 CFO 签署烧钱证明,并提供截至 2026 年 1 月 + 2026 年 5 月的现金余额 |
| 估计现金跑道(Series C 后) | 自 2026 年 1 月起 18–47 个月(融资 $1.4B;估计烧钱 $30–50M/月) | 低 | 核心资本充足性指标;决定对下一轮融资的依赖 | 核实尽调时点账上现金和烧钱速度 |
| 获客成本(CAC) | 未披露;企业深科技 CAC 通常为 $50K–$500K+,包含现场销售和 PoC 支持 | 低 | 决定 GTM 可扩展性和效率 | 要求按季度提供销售与营销 opex、平均销售周期、赢单率 |
| 平均合同价值(ACV) | 未披露;根据收入和可能客户数推断,企业机队合同为 $500K–$5M | 低 | 衡量单合同企业机会规模和收入可预测性 | 要求按客户队列提供收入和 ACV 分布 |
| LTV(客户终身价值) | 未披露;高切换成本和数据飞轮锁定效应使其结构性偏高 | 低 | 决定 LTV:CAC 比率和商业模式可行性 | 要求提供客户留存数据、合同期限、续约率和扩张收入 |
| 人均收入 | ~$353K($30M 收入 / ~85 名员工);对软件公司偏低,反映早期阶段 | 低 | 效率代理指标;早期比例偏低正常,但规模化后需要改善 | 获取按职能划分的员工数(R&D vs S&M vs G&A)和总薪酬支出 |
所有单位经济模型指标要么未披露,要么来自推断和基准估算。没有可用的审计数据。估算使用公司声称的 $30M 收入数字,以及基于 LinkedIn 的约 85 人员工数估计。“未披露”字段构成任何严肃投资人财务尽调的核心议程。
[CI001, CI023, CI029, CI030, CI031, CI035]低 / 基准 / 高三档估算,覆盖 Skild AI 核心财务指标;依据已披露数据、分析师基准和可比 AI 初创公司画像。除标注公司披露外,所有区间均为分析师估算。区间代表真实不确定性,不是预测情景。
2025 年收入($30M)为公司口径;其余区间均由分析师基于可比 AI 初创公司基准和 Skild 融资历史估算。没有可用的经审计数据。
[CI001, CI007, CI023, CI029, CI030, CI031]Skild AI 核心财务指标概览,标明哪些由公司披露、哪些为分析师估算、哪些完全未知。当前 $14B+ 估值下,公开财务数据存在重大证据缺口,投资测算因此受限。
收入($30M)为公司口径;其他数值均为分析师估算,或已确认公开渠道不可得。Skild AI 没有可用经审计财务数据。
[CI001, CI004, CI007, CI008, CI023, CI025]05产品与技术
5.1 产品定义与平台模块
Skild AI 提供三类产品资产,并把它们组织成一个垂直一体化平台。核心产品是 Skild Brain,公司称其为行业首个统一机器人基础模型。过去的机器人软件通常为特定机器人类型和任务定制;Skild Brain 则是全形态的:它能控制任意机器人——四足机器人、人形机器人、桌面机械臂和移动机械臂——无需事先了解机器人的具体物理形态。模型让机器人能够处理从简单家务(装洗碗机、煎蛋)到高强度工业作业(在湿滑地形中导航、工厂装配)的任务。 第二个产品是 Mobile Manipulation Platform,把抓取、交接、拾放和导航等物理技能封装在 API 抽象层之后。企业客户和系统集成商可因此构建机器人应用,而无需管理底层运动控制细节。Skild 称它让“用户能够构建应用,而不必担心非结构化、混乱真实世界中的细节”。 第三个产品来自 2026 年 4 月对 Zebra Technologies 的 Robotics Automation 业务的收购,即 Symmetry Fulfillment Platform——一个已经验证的机队编排层,可在物流环境中协调异构机器人机队和一线员工。三类模块合在一起,覆盖 AI 智能(Skild Brain)、技能执行(Mobile Manipulation Platform)和机队协同(Symmetry),使 Skild 能以单一供应商身份提供端到端仓库自动化。 商业部署覆盖安防和设施巡检、最后一公里配送、仓库拣选和分拣、工厂装配、数据中心运营,以及建筑工地监控。2025 年,公司在短短几个月内从零收入增长到约 $30M,客户跨越多个行业。 [CE001, CE002, CE003, CE004, CE005, CE006]
| 模块 / 资产 | 主要用户 | 状态 / 成熟度 | 核心差异化 | 尽调缺口 |
|---|---|---|---|---|
| Skild Brain(基础模型) | 机器人 OEM、系统集成商、企业客户 | 商业化;自 2025 年起在 6+ 个行业活跃部署 | 跨本体;新机器人本体无需重新训练;可在上下文中适应硬件故障 | 无同行评审基准;性能声明仅来自公司自述或 NVIDIA 报道 |
| 移动操作平台 | 企业应用开发者、集成商 | 商业化;API 受限开放;截至 2026 年 5 月无公开 SDK | 将抓取、交接、拣放、导航能力抽象为 API | 无公开文档;开发者生态有限;GitHub 为空 |
| Symmetry 履约平台 | 仓储和物流运营方 | 生产级(2026 年 4 月从 Zebra Technologies 收购) | 已验证可编排异构机器人机队 + 人类一线员工 | 收购后与 Skild Brain 的整合深度未验证;存在 Zebra 客户迁移风险 |
| 仿真训练基础设施 | 内部 R&D(Skild AI) | 已运行;NVIDIA Isaac Lab + Cosmos Transfer | 生成数万亿条合成经验;数天算力即可累积千年级经验 | GPU 集群规模、成本和算力单位经济性未披露 |
| 数据飞轮 / 后训练流水线 | 内部 R&D;由商业部署供数 | 已活跃;随每一次新增部署增长 | 自我强化的竞争护城河;竞争对手拿不到自有真实世界数据 | 数据来源、清洗质量和标注流程未经独立验证 |
成熟度评级为分析师估计,基于公开部署信号和合作伙伴公告;Skild AI 未披露官方产品阶段或 TRL 信息。
[CE001, CE002, CE003, CE004, CE005, CE008]端到端流程从机器人 OEM / 系统集成商开始,经过 API 技能调用、Skild Brain 推理、机器人动作,再回到数据飞轮反馈。Symmetry Fulfillment 在主推理路径旁编排多机器人机队。
[CE002, CE003, CE004, CE017, CE019, CE024]从七个产品能力维度评估成熟度和证据质量。Skild 最强的维度是跨本体泛化和机队编排(通过 Symmetry)。最弱的是开发者平台开放度和安全认证。
[CE001, CE002, CE003, CE019, CE031, CE033]5.2 架构与训练管线
Skild Brain 采用基于 transformer 的两层分层架构。高层组件以低频运行,负责语义层任务规划——决定采取什么动作、排列子任务顺序,并整合操作员给出的语言或视觉指令。低层组件以高频运行,把高层命令转成具体物理机器人各关节的电机扭矩和角度。这种拆分接近人类分解意图与运动执行的方式,也让同一个高层规划器无需修改就能驱动形态差异很大的机器人身体。 训练分两阶段。预训练阶段使用大规模物理仿真(NVIDIA Isaac Lab),在多种机器人形态和多样环境中并行运行数千个机器人实例,生成数万亿条合成动作经验。NVIDIA Cosmos Transfer 用环境变化扩充训练数据集——光照、纹理、天气条件——尽可能提高鲁棒性。第二类预训练来源是互联网级人类视频:数十亿段人类执行操控和移动任务的片段,模型把人当作生物机器人来学习物体可供性。Skild 称,这使其能在数天内获得相当于一千年的机器人经验。 后训练还使用两类来源:遥操作数据(图像和本体感知映射到关节扭矩,通过可扩展接口采集,被称为最丰富的训练信号),以及商业机器人机队持续生成的真实部署数据。数据飞轮是 Skild 竞争策略的核心:每一次新的商业部署都会贡献后训练数据,改进模型,再带来更多、更好的部署。 公司声称的关键突破是上下文学习:Skild Brain 遇到新情况——断腿、车轮卡住、陌生机器人身体或新环境——会在几次尝试中与环境交互并调整行为,不需要梯度更新或重新训练。NVIDIA 案例研究称,车轮卡住后 2–3 秒内恢复,断腿后数次尝试恢复,并能零样本泛化到踩高跷行走,腿长与身体比例超出训练分布。在 Pittsburgh 城市测试中,类人机器人在从未见过的环境中采集数据数小时后,任务表现达到 60–80%。 [CE009, CE010, CE011, CE012, CE013, CE014]
| 层级 / 组件 | 角色 | 关键依赖 | 风险 |
|---|---|---|---|
| 高层规划器(transformer) | 语义任务规划;整合语言和视觉指令;编排子任务 | 基于仿真 + 互联网视频预训练;以低控制频率运行(约 ~10-30 Hz) | 面对训练分布未覆盖的新语义场景,可能规划错误或幻觉 |
| 低层运动控制器(transformer) | 高频将高层指令转换为各关节扭矩和角度 | 本体感知 + 摄像头输入;以 ~1000 Hz 输出;学习参数不绑定具体本体 | 高速操作受延迟约束;未发布独立延迟基准 |
| NVIDIA Isaac Lab 仿真 | 面向合成训练数据的大规模物理强化学习环境 | NVIDIA GPU 算力;Isaac Lab 开源框架;并行实例化机器人 | 仿真到现实落差;依赖 NVIDIA 仿真生态 |
| NVIDIA Cosmos Transfer | 用环境变化(光照、纹理、天气)增强训练数据集 | NVIDIA Omniverse 基础设施;Cosmos 世界基础模型 | 增强质量上限未知;照片级真实感落差可能引入分布偏移 |
| 互联网视频摄取流水线 | 从数十亿个人类动作视频中提取可供性,用于预训练 | 可访问在线视频平台;先进可供性提取技术 | 训练视频版权和许可状态未解决;法律风险未量化 |
| 遥操作数据采集 | 最丰富的训练信号;图像 + 本体感知映射到关节扭矩 | 人类操作员;遥操作硬件接口;数据流水线基础设施 | 扩展速度慢于仿真;运营成本高;质量取决于操作员技能 |
| 真实世界部署反馈闭环 | 从商业机器人机队生成自有后训练数据 | 活跃客户部署;从已部署机器人到训练集群的数据流水线 | 数据质量取决于部署多样性;跨客户数据治理不清晰 |
| HPE + STN 私有 AI 即服务 | 为模型训练和推理工作负载提供安全、私有 GPU 算力 | HPE Cray XD670(NVIDIA HGX H200)用于训练;HPE ProLiant DL380a(NVIDIA L40S)用于推理 | 基础设施成本、GPU 供给约束;私有算力单一依赖 HPE 供应商 |
四层堆栈展示 Skild AI 平台如何组织:底层是硬件无关的训练基础设施,向上依次是 Skild Brain 基础模型、技能执行 API 和顶层的机队编排。每一层都在原始模型之上增加抽象和企业价值。
[CE001, CE002, CE003, CE009, CE010, CE011]Skild AI 关键技术依赖的有向无环图。NVIDIA 算力基础设施位于训练中心;HPE 提供私有算力层;互联网视频和机器人硬件伙伴分别提供数据和本体。
[CE010, CE011, CE012, CE016, CE033, CE037]5.3 部署、集成与路线图
Mobile Manipulation Platform 的 API 是企业客户和机器人 OEM 伙伴的主要集成界面。API 将离散技能——抓取、交接、导航——暴露为可调用函数,使应用开发者无需了解底层电机控制就能编排机器人行为。截至 2026 年 5 月,公司尚未发布公开 SDK 或开源开发者文档;GitHub 组织(github.com/skild-ai)没有公开仓库,说明访问权限仍面向企业伙伴把关。这限制了第三方开发者生态建设。 截至 2026 年初,部署领域包括安防巡逻、设施检查、最后一公里配送、仓库履约、工厂装配、数据中心运营和施工现场监控。2025 年 3 月 HPE 新闻稿称,Skild Brain 初期目标是建筑、制造和安防机器人。LG CNS 合作(2025 年 6 月)面向智能工厂、智能物流和城市服务部署,包括养老护理和设施巡逻,LG Technology Ventures 同时是投资方。收购 Symmetry Fulfillment(2026 年 4 月)带来生产级企业仓库编排能力,以及现有工业客户关系。 路线图围绕几件事:用 Series C 轮资金扩张四类数据来源,深化模型架构和训练算法研究,并扩大真实世界部署来产生收入。公司宣称的长期目标,是为所有机器人形态、所有任务、所有场景打造一个以动作为中心的单一大脑。基础设施正在从早期公有云扩到私有 AI-as-a-service 部署(训练用 HPE Cray XD670 服务器搭配 NVIDIA HGX H200;推理用 HPE ProLiant DL380a 搭配 NVIDIA L40S),由 HPE 伙伴 STN 管理,提供安全、可定制、可扩展的算力。 [CE019, CE020, CE021, CE022, CE023, CE024]
| 用户任务 / 行业 | 当前工作流 | Skild 方案 | 可衡量收益 | 限制 |
|---|---|---|---|---|
| 安防与设施巡检 | 人类保安巡逻;CCTV 监控;人工报告 | 搭载 Skild Brain 的四足 / 人形机器人自主巡逻;适应新环境 | 24/7 覆盖;无需重新建图即可处理消防疏散通道、户外地形、障碍物 | 边缘场景处理;自主巡逻的监管批准未公开说明 |
| 仓库拣选与履约 | 固定路径 AMR + 人类拣货员;劳动力成本和场地重构成本高 | 移动操作平台 + Symmetry Fulfillment 协调异构机队 | 混合机器人机队端到端自动化;无需固定路径基础设施 | Symmetry 与 Skild Brain 整合深度未验证;现有 Zebra 客户存在过渡风险 |
| 工厂装配与制造 | 按任务配置机械臂,并为每个产品变体定制编程 | Skild Brain 通过 API 控制台面和移动操作机械臂 | 省去逐任务重新编程;可在上下文中适应新零件和夹具 | 精密制造成功率未公开披露;工业重复性未验证 |
| 最后一公里与园区配送 | 人类配送员;需要建图的固定路线 AMR | 人形 / 四足机器人在非结构化室内外环境中导航 | 无需预先建图即可适应新楼宇布局;实时处理障碍物 | 公共空间自动配送的监管批准未公开说明 |
| 建筑工地监测 | 人工巡检;无人机测绘需要操作员 | 足式机器人在建筑工地巡航,用于隐患检查和报告 | 可进入不稳定地形;自主适应工地每日变化 | 建筑安全认证和责任框架未披露 |
| 数据中心运维 | 人类技术员负责线缆追踪、设备检查和环境监控 | 自主机器人执行例行巡检和环境监控任务 | 降低人员暴露于高压 / 高密度环境的风险;24/7 可用 | 面向任务关键场景的可靠性标准(uptime SLA)尚未在机器人系统上公开验证 |
60-80% 任务表现数字仅来自 NVIDIA 针对 Pittsburgh 城市部署的案例研究,可能无法推广到受控工业环境。各行业表现未单独披露。
[CE004, CE005, CE020, CE021, CE022, CE024]| 日期 / 阶段 | 功能 / 里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2017(历史) | 好奇心驱动探索论文(ICML 2017);奠定自主学习技术基础 | 已发表;引用 6,000+ 次;直接 IP 脉络 | 学术 IP 根基;创始人与纯工程团队拉开差异 | SE009 |
| 2021(历史) | RMA 快速运动适应论文(RSS 2021);仿真到现实适应突破 | 已发表;获 CoRL 2021-22 Best Robotic System Award | Skild Brain 上下文学习的直接技术前身;仿真到现实脉络已得到验证 | SE010 |
| July 2024 | Series A 融资 $300M;Skild Brain 和 Mobile Manipulation Platform 结束隐身并发布 | 已完成 | 建立市场存在感和商业产品;获得首个机构背书 | SE006 |
| March 2025 | HPE + NVIDIA + STN 私有 AI-as-a-service 基础设施部署完成 | 运行中 | 扩大安全训练规模;降低公共云依赖;加快迭代 | SE004 |
| June 2025 | 与 LG CNS 建立工业人形机器人战略合作;LG Technology Ventures 参投 | 进行中;部署目标已定义 | 打开亚洲工业市场;验证平台适配智能工厂和物流场景 | SE011 |
| January 2026 | Series C 融资 $1.4B,估值 $14B+;资金用于数据扩容和研发 | 已完成 | 支撑未来 2-3 年现金跑道;重点扩大数据规模和商业部署 | SE007 |
| April 2026 | 收购 Zebra Technologies 的 Robotics Automation(含 Symmetry Fulfillment) | 已完成 | 补上企业车队编排能力;切入全栈仓储自动化 | SE003 |
| 2026(公司表述) | 面向所有机器人形态、所有任务、所有场景的单一动作中心大脑 | 开发中;公司表述的战略路线图 | 如果跑通,可消除面向特定硬件的定制,最大化平台 TAM | SE002 |
前瞻性里程碑来自公司表述;未确认独立第三方验证或承诺交付日期。历史里程碑基于公开公告。
[CE006, CE007, CE008, CE021, CE022, CE023]5.4 差异化、知识产权与数据护城河
Skild AI 的技术差异化建立在四根互相强化的支柱上。第一,训练数据规模:公司称其以动作为中心的数据量比竞争机器人模型多 1,000 倍,来源包括大规模仿真、互联网视频摄取和不断扩大的真实部署机队。第二,全机体泛化:不同于按机器人类型或任务分别训练模型的竞争对手,Skild Brain 可借助上下文学习适应新身体和新场景,完全不需要再训练。第三,涌现能力:演示包括在抓取中途接住滑落物体、旋转物体校正朝向、从硬件故障中恢复、踩高跷行走——这些行为并非显式编程,而是从大规模训练中涌现。第四,复利式数据飞轮:商业部署持续生成专有后训练数据,没有在线部署机队的竞争对手无法复制。 IP 基础直接来自创始人的研究。Deepak Pathak 在 ICML 2017 发表的自监督预测驱动好奇心探索论文,奠定了用内在动机进行自主智能体探索的基础技术;这篇论文引用超过 6,000 次,直接承接到 Skild Brain 无需显式奖励塑形即可泛化的能力。Kumar、Pathak 等人在 RSS 2021 的 Rapid Motor Adaptation(RMA)论文,通过两阶段方法(特权训练加在线历史编码器)证明了腿式机器人电机控制可实时适应新地形、负载和损伤——这是 Skild Brain 上下文学习的直接技术前身。Abhinav Gupta 的「Supersizing Self-Supervision」工作确立了大规模数据范式(50,000+ 次机器人抓取尝试、700 个机器人小时),支撑 Skild 的训练策略。Pathak 和 Gupta 的 CMU 团队凭借 2021–2022 年仿真到现实迁移工作,在 Conference on Robot Learning 获得 Best Robotic System Award。 尚未发布将商业版 Skild Brain 与竞争模型比较的同行评审基准。性能指标来自公司自述或 NVIDIA 伙伴案例研究。相对于具体竞争对手的 1,000 倍数据说法,也没有经过独立验证。 [CE027, CE028, CE029, CE030, CE031, CE032]
5.5 信任、安全、合规与质量控制
Skild AI 的信任与安全姿态由其部署场景决定:安防巡逻、施工现场、仓库运营都带来物理后果,机器人失效可能伤人或损坏财产。公司把 Skild Brain 的上下文学习描述为通过实时环境适应提升安全:传感器发现异常(负载变化、地形变化、硬件退化)时,模型会调整行为,而不是沿着不安全轨迹继续执行。不过,Skild Brain 尚未公开披露第三方物理安全认证(如协作机器人的 ISO 10218、ISO/TS 15066)或正式 AI 安全审计。 在基础设施和数据安全上,Skild 已从公有云转向私有 AI-as-a-service 部署(HPE + STN),并将其定位为能为用于后训练的客户部署数据提供更高安全性和数据隐私。公开资料中看不到公司发布安全白皮书或数据处理政策文件。 从 Zebra Technologies 收购的 Symmetry Fulfillment 平台带来企业级机队管理软件和既有工业部署,为运营可靠性增加一层已验证基础。互联网视频训练数据的法律风险是重大未解决风险:公司摄取数十亿个第三方视频,却没有公开披露授权或合理使用法律分析;这一风险类似大语言模型提供商面临的诉讼。IQT(In-Q-Tel)是美国情报界的风险投资机构,也是 Series C 轮投资方,可能使 Skild Brain 技术平台面对出口管制审查和军民两用分类要求。截至 2026 年 5 月,公开资料没有报道召回、安全事故或监管执法行动。 [CE035, CE036, CE037, CE038, CE039, CE040]
| 控制项 / 认证 / 指标 | 状态 | 范围 | 缺口 / 尽调问题 |
|---|---|---|---|
| 物理安全(机器人 - 人类交互) | 未公开披露第三方认证 | 所有人机共处的部署行业 | 核实 ISO 10218 / ISO/TS 15066 合规性;企业采购前要求提供安全审计报告 |
| AI 模型安全与对齐 | 未公开披露;未发布 ASIMOV 或同等框架结果 | Skild Brain AI 决策与规划层 | 核实内部安全测试协议;索取红队或对抗场景测试结果 |
| 数据隐私(客户部署数据) | 私有 AI 即服务(HPE + STN);非公有云;无公开数据政策 | 客户机器人车队的后训练数据 | 核实与客户的数据处理协议;数据留存、删除和匿名化政策 |
| 版权 / 知识产权(视频训练数据) | 未披露;已确认存在互联网视频摄取管线,但法律审查未披露 | 预训练数据来源(数十亿条互联网视频) | 获取 Skild AI 律师法律意见;对照 LLM 版权诉讼判例 |
| 监管(美国国防 / 出口管制) | IQT (In-Q-Tel) 是 Series C 投资方;潜在 ITAR / EAR 影响公开层面尚未处理 | 广义技术平台;未来任何政府部署 | 确认出口管制审查;评估 Skild Brain 双用途分类及 CFIUS 风险敞口 |
| Zebra Symmetry 合规态势 | 生产级;既有 Zebra 企业部署带有现行合规义务 | 仓储 / 物流车队编排层 | 确认收购后继承的认证和监管义务;评估迁移风险 |
未公开认证不等于不合规;尽调应直接向 Skild AI 索取文件。严重性评级是分析师对企业采购风险的判断。
[CE035, CE036, CE037, CE038, CE039, CE040]5.6 附录
06客户
6.1 客户群分层
Skild AI 的客户群可分为四个截然不同的分层,买方特征、采购路径和战略价值各不相同。 第一类也是核心分层,是机器人 OEM 伙伴——将 Skild Brain 授权嵌入自有机器人产品线的硬件制造商。已确认 OEM 伙伴包括 ABB Robotics(全球最大工业机器人厂商之一)、Universal Robots(协作机器人全球龙头,Teradyne 旗下)和 Mobile Industrial Robots(MiR,同属 Teradyne)。这些合作是分销倍增器:Skild Brain 嵌入 ABB 和 UR 数万台已安装机器人组合后,Skild 无需承担直接销售开销,就能触达对方整个客户群。OEM 伙伴关系类似 AI 软件供应商向硬件 OEM 授权——Skild 按机器人或按使用收取许可收入,OEM 则把 AI 智能捆绑成差异化卖点。 第二类是企业直接部署方:大型企业采购 Skild Brain,在自有设施内自动化特定工作流。Foxconn(通过 NVIDIA 合作)是首个公开点名的直接部署方,正在 Houston 的 Blackwell GPU 服务器装配线上使用 Skild Brain。公开资料还确认仓储、建筑和巡检领域存在其他直接部署方,但身份未披露。 第三类是战略投资者客户:参与 Skild 融资并表明运营兴趣的实体,包括 Amazon(Bezos Expeditions)、Samsung、LG(包括 LG CNS 商业合作)、Schneider Electric、CommonSpirit Health 和 Salesforce Ventures。它们既是财务支持者,也是潜在部署客户,使收入归因变得模糊。 第四类是国防和政府,由 IQT(In-Q-Tel)投资释放信号。尚无公开宣布的国防部署,但 IQT 的使命是把技术桥接到美国国家安全机构。 [CU002, CU003, CU004, CU005, CU006, CU007]
| 细分市场 | 买方 / 付款方类型 | 主要用例 | 规模 / 市场规模 | 收入 / 战略价值 | 证据缺口 |
|---|---|---|---|---|---|
| 机器人 OEM 合作伙伴(ABB、Universal Robots、MiR) | 硬件制造商被授权方 | 将 Skild Brain 嵌入机器人产品组合,用于工业和协作自动化 | ABB: 每年 400k+ 台机器人;UR: 全球已安装 100k+ 台协作机器人 | 战略价值高;无需直营销售即可触达长尾分销;合同条款未披露 | 未披露生产结果指标;ABB 细节保密 |
| 企业直连部署方(Foxconn 经 NVIDIA) | 生产设施运营方 | 在自有设施自动化复杂装配、拣选和检测 | 全球数十家大型制造商和物流运营商 | 首个确认的生产收入来源;合同金额未披露 | 仅有一个具名部署;结果指标(吞吐量、错误率)未发布 |
| 战略投资方客户(Amazon、Samsung、LG、Schneider、CommonSpirit) | 明确表达部署意向的战略投资方 | 各不相同:物流(Amazon)、消费电子(Samsung/LG)、能源 / 基础设施(Schneider)、医疗(CommonSpirit) | 合计市值 >$3T;未来潜在部署基础庞大 | 模糊:投资可能早于或伴随部署;收入可能来自试点而非生产合同 | 除 LG CNS 合作外,任何投资方客户都没有确认运营部署 |
| 国防 / 政府(IQT 信号) | 美国国家安全机构(通过 IQT 间接) | 国防场景下的自主巡检、监视和物流 | 美国 DoD 机器人预算 >$1B/yr | 仅早期信号;未确认政府合同或收入 | 无公开合同;IQT 被投公司通常受保密限制 |
| 通过 Zebra Symmetry 触达的仓储运营商(CTL Global、Encore、Geneva10) | 第三方物流 / 履约运营商 | AMR 辅助拣选、批量 / 集群工作流、人机协同 | 全球数百家 3PL 和履约运营商 | 通过收购获得带有已记录生产率指标的客户基础(30-40%+ 提升);这些站点尚未确认集成 Skild Brain | 收购后尚未确认 Skild Brain 部署在这些设施 |
| 医疗(CommonSpirit Health 战略投资方) | 医院 / 医疗体系运营方 | 医院物流、临床任务辅助、设施巡查 | CommonSpirit: 140+ 家医院,收入 $20B+;更广义美国医疗市场 >$4T | 战略投资释放意向信号;未确认部署 | 无公开医院机器人部署;医疗机器人面临监管和责任约束 |
收入 / 战略价值评估为分析师估计,依据细分市场规模、交易复杂度和战略投资者行为。各细分市场实际收入未披露。 投资方客户(Amazon、Samsung、LG、Schneider、CommonSpirit)的部署状态仅反映公开公告信息;可能存在未披露试点。
[CU002, CU003, CU004, CU005, CU006, CU007]展示不同客户细分如何推进 Skild AI 采用生命周期,从初始认知,到试点、生产部署,再到平台扩展。OEM 伙伴走的是先集成路径,企业直采部署方走的是先用例路径。生产阶段的数据飞轮形成自我强化的扩张循环。
[CU007, CU008, CU009, CU015, CU018, CU021]6.2 采用轨迹与增长证据
Skild AI 的采用轨迹特点是从零起步后收入增长异常迅速,同时几乎所有能让分析师评估增长质量、集中度或耐久性的底层指标都不透明。 收入在 2025 年从零增长到约 $30M——这一点由公司官方 Series C 博文、BusinessWire Series C 新闻稿确认,并得到多家独立新闻来源佐证。公司称收入「指数级增长」,technical.ly 报道还提到公司在 2025 年实现盈利。对于一家 2024 年 7 月才走出隐身模式的公司,这些里程碑非常罕见。 公司声明确认的部署足迹覆盖六个垂直领域:安防与设施巡检、最后一公里配送、仓库、制造、数据中心和建筑。已确认 Skild Brain 运行在 30 多种不同机器人类型上。Foxconn/NVIDIA 生产部署(2026 年 3 月)是首个公开点名的大规模客户部署。Skild 的首席幕僚向 technical.ly 确认,仓储、建筑和巡检领域此前已有部署,但未披露客户名称。 2026 年 4 月收购 Zebra Technologies,为 Skild 装机基础增加了成熟仓库运营商客户(CTL Global Solutions、Encore Fulfillment、Geneva10 Fulfillment),立刻带来物流领域的生产级客户关系。这些 Zebra 客户有记录在案的生产率指标:在保持吞吐的同时减少 30% 机器人需求(CTL Global、Encore),以及 40%+ 生产率提升(Geneva10)。 同样重要的是缺失项:没有客户数量、没有单客户收入、没有 ARR 拆分、没有队列数据,也没有流失或续约历史。极快的收入增速说明两种可能:少数几个超大企业合同,或几十个较小部署带来的真实广泛采用;但区分两者所需的数据并未公开。 [CU001, CU002, CU009, CU010, CU011, CU020]
| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 年收入 | ~$30M | 2025 | 公司表述(Series C 新闻稿) | 高 | 从零快速商业化;对成立 2 年的公司而言异常突出 | 开始产生收入的日期;客户数;ARR 与一次性收入拆分 |
| 收入增长率 | $0 到 $30M,用了「几个月」 | 2025 | Skild AI 官方博客 + BusinessWire | 高 | 增速极快;更像大型企业合同落地,而不是逐步爬坡 | 首次收入的确切季度或月份;增长是在加速还是进入平台期 |
| 具名付费客户 | 未披露;公司称有「多个客户」 | 2026 | Technical.ly + Sacra 分析 | 低 | 集中度未知;可能是 3 个大客户,也可能是 50 个小客户 | 实际客户数;单客户收入;最大客户收入占比 |
| 已确认活动的部署垂直领域 | 6(安防、配送、仓储、制造、数据中心、建筑) | 2026 | Series C 新闻稿 + 官方网站 | 高 | 横向覆盖广,降低单一行业依赖风险 | 各垂直领域具名客户;各垂直领域活跃部署数量 |
| 具名生产部署(公开确认) | 1(Foxconn/NVIDIA,Houston TX) | 2026-03 | 多家独立媒体来源 | 高 | 首个生产级部署公开证明;投资者可参考样本仍有限 | 其他具名客户;Foxconn 部署的结果指标 |
| 支持的机器人类型 | 30+(基于公司广泛形态主张推断) | 2026 | 公司主张 + NVIDIA 案例研究 | 中 | 广泛 OEM 兼容性对授权模式至关重要;支撑 ABB/UR 合作价值 | 准确清单;生产测试与仿真机器人类型拆分;各类型性能指标 |
| Zebra Symmetry 收购带来的客户(具名) | 3 家(CTL Global Solutions、Encore Fulfillment、Geneva10 Fulfillment) | 2025-2026 | Zebra 投资者关系 + IWLA | 高 | 通过收购立即获得生产仓储运营商装机客户基础;收购后 Skild Brain 集成仍不清楚 | Symmetry 客户总数;收购客户收入;Skild Brain 集成时间表 |
多数指标来自公司主张或根据媒体报道推断。客户数量和单客户收入为分析师估计;Skild AI 未披露这些数字。 缺失分母列说明还需要哪些数据,才能让每项指标有上下文。
[CU001, CU002, CU009, CU010, CU020, CU023]估算截至 2026 年 5 月,可触达机器人 OEM 和企业潜在客户到已确认生产部署的漏斗。阶段数量基于公开可用部署证据估算。合作公告与已确认生产部署之间的落差,凸显商业牵引仍处早期。
阶段估算为分析师基于公开信息的近似值。Skild AI 不披露管线、阶段数量或转化指标。“伙伴 / 投资者”计数只包含公开点名实体。
[CU001, CU007, CU008, CU009, CU010, CU011]6.3 具名客户证明表
截至报告日,Skild Brain 本身的具名客户证明仅限于一个已确认生产部署。如果把 Zebra Symmetry 收购带来的客户基础和 OEM 伙伴背书纳入,具名证据基础会大幅扩大。 Foxconn/NVIDIA 合作(2026 年 3 月宣布)是 Skild 质量最高的具名客户证明。Foxconn 正使用 Skild Brain 自动化 Houston 设施中双臂机器人执行复杂 GPU 服务器机架装配——一个多步骤操控任务(母排放置、限位块、16 颗螺丝钻孔、移除限位块),目前由人工完成。公司官方博客、多家独立媒体报道,以及 NVIDIA 副总裁 Deepu Talla 的公开声明均确认了这一点。不过,双方都未发布量化结果指标(吞吐提升、错误率、劳动力节省、ROI)。 在 OEM 伙伴层面,ABB Robotics 总裁 Marc Segura 和 Universal Robots CEO Jean-Pierre Hathout 已公开为 Skild Brain 集成背书,构成战略一致性的伙伴证明。ABB 提到「细节仍属机密」,确认合作处于活跃状态,但也限制了外部对生产部署状态的验证。 三家具名 Zebra Symmetry 客户——CTL Global Solutions、Encore Fulfillment 和 Geneva10 Fulfillment——是 Symmetry 编排平台真实的生产级客户,并有记录在案的绩效结果(30–40%+ 生产率提升)。不过,这些部署早于 Skild 收购,不能证明 Skild Brain 本身已经部署到其设施。收购后 Skild Brain 与 Symmetry 的集成,尚未通过具体客户或时间线公开确认。 CommonSpirit Health 和 Schneider Electric 是战略投资方;尚未宣布 Skild Brain 在 CommonSpirit 医院或 Schneider Electric 设施中的部署。 [CU007, CU008, CU009, CU012, CU013, CU014]
| 实体 | 细分市场 | 部署 / 用例 | 生产 / 试点 | 结果 / 证据 | 限制 |
|---|---|---|---|---|---|
| Foxconn(经 NVIDIA) | 企业直连部署方 – 高端制造 | 双臂机器人在 Houston, TX 组装 NVIDIA Blackwell GPU 服务器机架;16 步操作序列,包括钻孔 | 生产(March 2026 确认) | 公司博客 + NVIDIA 副总裁引述 + 5+ 家独立媒体来源;展示了任务复杂度和上下文恢复能力 | Foxconn 和 Skild 均未披露量化结果指标(吞吐量、错误率、人工节省) |
| ABB Robotics | 机器人 OEM 合作伙伴 | 将 Skild Brain 集成进 ABB 工业机器人组合,用于制造、物流和检测 | 已宣布合作;生产集成推进中 | ABB 总裁 Marc Segura 公开背书;NVIDIA 投资者新闻稿确认合作 | ABB 称细节仍保密;尚未确认经 ABB 渠道落地的客户部署 |
| Universal Robots(Teradyne) | 机器人 OEM 合作伙伴 – 协作机器人 | 将 Skild Brain 集成进 UR 协作机器人和 MiR 自主移动机器人组合 | 已宣布合作;集成时间表未披露 | UR CEO Jean-Pierre Hathout 公开背书;Skild 官方博客确认合作 | 未确认具名 UR 客户部署;集成深度和时间表未公开 |
| CTL Global Solutions(客户) | 仓储运营商 – 通过 Zebra Symmetry 收购 | Symmetry 编排下的 AMR 辅助仓库拣选;在保持吞吐量的同时减少 30% 机器人 | Symmetry 平台生产部署(收购前) | Zebra 投资者关系和 BusinessWire 新闻稿具名,并给出量化结果指标 | 部署在 Zebra Symmetry 上,不在 Skild Brain 上;收购后 Skild Brain 集成未确认 |
| Geneva10(G10)Fulfillment | 仓储运营商 – 通过 Zebra Symmetry 收购 | 通过 Symmetry 协调 AMR 车队,实现批量和集群拣选自动化 | Symmetry 平台生产部署(收购前) | IWLA 行业协会具名,并给出 40%+ 生产率提升指标 | 部署在 Zebra Symmetry 上,不在 Skild Brain 上;Skild Brain 集成时间未知 |
| CommonSpirit Health | 医疗战略投资方 – 潜在未来客户 | 医院物流、临床任务辅助、设施巡查(投资方表述意向) | 仅投资方;未确认部署 | BusinessWire 和 Skild 博客确认其为 Series C 战略投资方;IQT 引述提到国家重要性 | 未宣布部署;医疗机器人面临监管(FDA)、责任和采购壁垒 |
这是部分列举:Skild AI 未披露多数客户名称。 本表覆盖所有已公开识别且关系已确认或可推断的实体。 「生产 / 试点」状态反映目前最佳证据;未披露部署可能属于任一类别。 Zebra Symmetry 客户(CTL Global、Encore、Geneva10)已确认使用 Symmetry 平台,但收购后尚未确认使用 Skild Brain。
[CU007, CU008, CU009, CU012, CU013, CU014]评估每个已点名客户实体的证据质量、生产成熟度、结果具体性和留存可见度。评级反映可获得公开证据的强度,不必然代表底层部署质量。高 = 强公开证据;中 = 有限或间接证据;低 = 无公开证据;N/A = 不适用。
[CU007, CU008, CU009, CU012, CU013, CU014]按每个已点名客户或伙伴层级给出证据质量分数(1–5),权重来自独立来源数量、佐证质量和生产部署确认。Foxconn 以 4.5 领先(多个独立来源加 NVIDIA 官方确认);Amazon 最低,为 1.0(仅战略投资人,无商业关系)。全部八个实体平均 2.8,反映 Skild AI 商业部署仍早期且大多不公开。
[CU007, CU008, CU009, CU010, CU011, CU013]6.4 留存、耐久性与切换成本
Skild AI 的留存动态无法用公开数据直接衡量:NRR、GRR、流失率、合同期限和客户满意度评分都未披露。公司直到 2025 年才开始产生显著收入,这意味着即使是内部 cohort 数据,截至报告日也只有不到 24 个月历史。 数据飞轮创造了结构性留存机制。每一次客户部署都会向 Skild 后训练管线贡献专有机器人运营数据。数据改进共享的 Skild Brain 模型,继而改善平台上所有客户的结果。贡献过部署数据的客户,会受益于已学习其具体环境和任务类型的模型。切换到竞争机器人 AI 平台会放弃这些累积模型改进,形成类似云机器学习平台的经济性切换成本。 考虑到把 Skild Brain 嵌入机器人硬件组合需要很深的工程投入,OEM 合作(ABB、Universal Robots、MiR)很可能采用多年集成协议。这些合作创造了不依赖任何单一终端客户部署的耐久关系。机器人 OEM 如果把 Skild Brain 集成进产品架构,再替换它会面临高切换成本,因为必须为整个机队重新工程化智能层。 Zebra Symmetry 平台增加了一层软件编排依赖:使用 Symmetry 协调机队的仓库运营商,如果替换平台会承担迁移成本。这是收购继承来的留存资产,但其耐久性取决于能否与 Skild Brain 能力成功集成。 目前未发现负面证据——投诉、部署失败、客户流失或公开争议——但这与其说一定代表真实客户满意,不如说同样可能来自早期阶段不透明。尽调应向公司索取 NRR、GRR 和客户队列数据。 [CU018, CU019, CU023, CU024, CU029, CU030]
| 指标 | 数值 / 状态 | 细分市场 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| 净收入留存率(NRR) | 未披露 | 全部 | 低 | 要求按客户队列(2025 队列与 2026 年初队列)和细分市场(OEM 与企业直连)提供 NRR |
| 总收入留存率(GRR)/ 流失 | 未披露 | 全部 | 低 | 要求提供 GRR 和年化流失率;询问首次部署以来是否有客户退出 |
| 合同期限 / 续约期 | 未披露;推断 OEM 集成为多年期 | OEM 合作伙伴、企业直连 | 低 | 要求提供典型合同期限、续约率,并确认现有合同是已续约还是已失效 |
| 客户满意度 / NPS | 未披露;没有第三方评价 | 全部 | 低 | 要求提供净推荐值(NPS)或 CSAT 数据;确认是否有客户可供尽调访谈背书 |
| 数据飞轮锁定(切换成本代理) | 通过专有部署数据贡献形成结构性锁定;未量化 | 全部 | 中 | 要求确认客户合同中的数据贡献条款;若客户迁移到竞品平台,量化切换成本 |
所有留存和满意度指标均为 null,因为 Skild AI 未公开披露,且公司商业化历史不足 24 个月。 尽调问题代表评估客户耐久性所需的具体数据请求。
[CU018, CU019, CU023, CU024, CU030]6.5 扩张、集中度风险与战略展望
Skild AI 的扩张潜力由三股复利力量推动:数据飞轮(每次新部署都会改进模型,让未来部署更好)、OEM 伙伴生态(ABB 和 Universal Robots 各自拥有全球数十万台可用 Skild Brain 升级的装机机器人),以及 Zebra Symmetry 收购来的客户基础(立即进入生产仓库运营商群体)。 公司宣称的扩张路线,是从半结构化工业环境(工厂、仓库)推进到结构化更弱的环境(医院、酒店),最终进入完全非结构化的消费者家庭。这提供了一条几十年的扩张弧线,但当前客户证据集中在最早、最结构化的一层。 集中度风险是最重要的客户尽调缺口。收入约 $30M、客户数量未知,因此单一客户(Foxconn 或某个未披露大型企业)可能贡献了大部分收入。战略投资者与客户重叠带来第二层集中度担忧:如果投资者客户(Amazon、Samsung、LG)主要通过探索性商业试点贡献收入,而不是通过公平市场交易的生产合同贡献收入,那么收入耐久性低于标题数字所暗示。Amazon 正在开发自有 DeepFleet 机器人 AI(1M+ 机器人),这提出一个问题:Bezos Expeditions 的这笔投资究竟是为 Skild 成功下注,还是对冲其成功。 OEM 合作通过成熟硬件销售渠道分销 Skild 技术,缓解了渠道风险。不过,Skild 没有公开证据显示其具备自助式数字销售动作、开发者市场或系统集成商认证计划——这些缺口限制了长尾客户获取速度。LG CNS 合作是亚太部署渠道 / 伙伴模式的早期信号。 [CU003, CU004, CU006, CU021, CU022, CU025]
| 驱动因素 / 风险 | 类型 | 影响 | 尽调路径 |
|---|---|---|---|
| OEM 合作伙伴生态(ABB、Universal Robots、MiR) | 扩张驱动因素 | 高 – 无需直营销售成本即可触达数十万台已安装机器人 | 确认收入分成模式和时间;要求提供 OEM 客户激活管线;了解是否存在排他条款 |
| 数据飞轮复利(更多部署 = 更好模型 = 更多部署) | 扩张驱动因素 | 高 – 规模若达到临界点,可形成结构性竞争护城河 | 确认合同中的数据贡献条款;评估数据飞轮是在加速还是仍处早期;将数据量与实体 AI 同行对比 |
| Zebra Symmetry 装机基础(CTL Global、Encore、Geneva10 + 其他未披露客户) | 扩张驱动因素 | 中 – 立即获得仓储领域生产客户关系 | 要求提供完整 Zebra Symmetry 客户名单和 ARR;确认 Skild Brain 在 Symmetry 平台上的集成路线图;评估客户迁移风险 |
| Amazon 投资方自建竞品 DeepFleet(1M+ 台机器人) | 集中度 + 竞争风险 | 高 – Amazon 可能不会成为 Skild 客户,还可能挤压 OEM 采用 | 厘清 Amazon-Bezos 投资关系性质;判断 Amazon 是否有任何 Skild Brain 部署,还是纯财务 / 对冲投资 |
| 客户收入集中度(未知,推断偏高) | 集中度风险 | 高 – 短期内 $30M 收入可能集中在 1-3 个大合同 | 要求提供前十大客户收入集中度;确认是否有单一客户收入占比超过 20% |
| 战略投资方收入与独立交易收入 | 集中度 + 质量风险 | 中 – 投资方客户可能代表试点,而非持久签约的生产收入 | 区分投资方客户收入和非投资方收入;确认投资方客户是多年期合同还是采购订单型合作 |
影响评级为分析师判断,依据收入规模($30M)、公司年龄和可比机器人 AI 公司。尽调路径说明 解决每项风险或确认每个机会所需的具体信息。
[CU003, CU004, CU021, CU025, CU031, CU034]07风险
7.1 技术与产品风险
Skild AI 最核心的技术风险,是 Skild Brain 的「全机体」泛化主张尚未获得独立验证。截至 2026 年 5 月,所有性能证据都来自公司陈述或伙伴报告(NVIDIA 案例研究);商业模型尚无同行评审基准发布。仿真到现实缺口是最基础的技术风险:主要在仿真中训练的模型,能否在非结构化物理环境中可靠运行,本就是公认难题。Skild 在 NVIDIA Isaac Lab 中训练数万亿条合成经验,并用互联网视频和遥操作数据补充,但迁移到多样、不可控真实环境,仍是机器人 AI 社区承认的开放研究问题。 泛化失败可能有几种表现:(1)离开训练分布环境后,任务成功率显著下降;(2)遇到边缘物理配置时,机器人行为不可预测;(3)传感器噪声、光照变化或新物体几何形态导致电机命令错误,并带来安全后果。公司的「上下文学习」能力——车轮卡住后 2-3 秒恢复、断腿后适应——在受控演示中令人印象深刻,但对生产级泛化的证明有限。 算力强度是第二个结构性风险。截至 2025 年,前沿 AI 模型单次训练成本为 $30M–$200M+,机器人数据基础设施还要叠加仿真、遥操作和部署管线成本。Skild 声称的「1,000 倍数据规模」训练意味着相应更高的算力支出,带来持续运营费用风险和资本依赖。部署在物理环境中的大型基础模型若出现涌现行为,会带来责任敞口:在人机共享工作空间里,意外电机命令可能造成人身伤害、财产损失或客户流失。 截至 2026 年 5 月,Skild Brain 没有公开披露的第三方物理安全认证(ISO 10218、ANSI/RIA R15.06 或同等标准);对于医疗、政府和公共基础设施等受监管行业客户,这通常是采购门槛。 [CR001, CR002, CR003, CR004, CR005, CR006]
| 失效模式 | 可能性 | 严重性 | 缓释成熟度 | 剩余风险敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| 仿真到现实泛化失败:Skild Brain 在训练分布之外的非受控真实环境中失效 | 高 — 仿真到现实差距是机器人 AI 的核心未解难题;尚无独立验证发布 | 阻断性 | 低 — 已使用域随机化;未披露鲁棒性测试方法或通过 / 失败标准 | 高 — 大范围部署失败会摧毁客户信任和投资逻辑 | 缺少独立基准;未披露失效模式测试框架;无公开回滚程序 |
| 涌现的不安全物理行为:大模型输出意外电机指令,造成人员受伤或财产损失 | 中 — 大模型输出具有非确定性;部署范围越广,边缘情形越多 | 阻断性 | 低 — 未披露模型层面的安全监控、推理时护栏或急停集成 | 高 — 单起高关注度机器人伤害事件就可能触发监管行动并造成客户流失 | 未披露行为安全层;未发布安全测试协议;未描述人在回路的防护机制 |
| 客户部署中的实体机器人伤害事件 | 中 — 概率会随部署机器人数量和部署小时数成比例上升 | 重大 | 低 — Skild 未披露事故响应协议、责任框架或保险覆盖 | 高 — 声誉和财务敞口高;可能引发监管执法 | 未披露事故历史;未披露保险覆盖;无公开安全认证 |
| 算力基础设施中断或供应冲击:NVIDIA GPU 可得性受限,扰乱训练管线 | 低至中 — NVIDIA GPU 需求仍高度紧张;半导体供应存在地缘政治风险 | 重大 | 中 — SoftBank 关系可能带来优先 GPU 使用权;NVIDIA 是战略投资方 | 中 — 训练延误会压缩模型改进节奏和竞争位置 | 未披露算力采购协议;无公开 GPU 短缺应急方案 |
| 生产中的模型漂移或退化:数据飞轮引入低质量部署数据,污染重训 | 中 — 真实部署数据可能包含对抗性、边缘情形或错误交互数据 | 重大 | 低 — 未披露训练后管线的数据质量过滤方法 | 中 — 模型质量退化可能同时影响所有已部署实例(全机群风险) | 未披露训练后管线的数据质量治理;未描述回滚机制 |
可能性和严重性由分析师基于已发布的机器人 AI 研究、监管背景和可比部署事故评估。Skild 在 2024–2025 年大部分时间仍处商业化前阶段,公开渠道没有 Skild 专属事故数据。
[CR001, CR002, CR003, CR006, CR007, CR038]| 风险 | 可监控触发项 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 仿真到现实泛化失败 | Skild Brain 真实任务成功率的独立第三方基准;客户部署失败率报告 | 非受控环境中的任务成功率 <70%,或 >2 起已报告生产部署因泛化失败造成客户流失 | 投资逻辑破裂 — 没有独立验证,技术差异化主张撑不住;当前估值下投资不可辩护 |
| 估值 / 收入脱节 | 季度收入增长轨迹;2026 年全年收入相对 2025 年实际值 | 2026 年收入 <$100M,且下一轮估值较 $14B 需下调 >20%;或下一轮在任何估值下都无法完成 | 投资逻辑破裂 — $14B 估值无法支撑;可能出现按市值计价损失;以不利条款完成 Series D 是压力信号 |
| SoftBank 投资方集中失效 | SoftBank Vision Fund 2 季度业绩报告;媒体对 SoftBank 组合优化的报道;Skild 下一轮融资公告 | SoftBank 拒绝参与 Series D,且 9 个月内没有新的机构领投方完成投资 | 重大关切 — 迫使公司接受过桥融资、摊薄性条款或战略出售;新领投方确认前降低敞口 |
| 欧盟监管不利裁决 | 欧盟委员会 AI Act 监管机构执法行动;欧盟法院就涉及 Skild 或可比系统的 AI 机器人责任作出判决 | 执法行动点名 Skild 或其硬件伙伴;法院裁定将 Product Liability Directive 具体延伸至 Skild Brain | 投资逻辑需修正 — 欧盟市场准入承压;调整 TAM;增持前评估合规补救成本 |
| 关键联合创始人离职 | 公开公告;LinkedIn 状态变化;董事会 8-K 等效披露(IPO 后)或媒体报道 | Pathak 或 Gupta 任一人在非计划交接中离开,且未同时任命可信的外部继任者 | 投资逻辑破裂 — 投资人和人才信心可能崩塌;观察 90 天窗口;退出,除非立即宣布同等可信度的替代 CEO 才考虑继续持有 |
终止标准针对五项最高严重性风险定义。触发项设计成可从公开来源或常规尽调沟通中观察到。「投资逻辑破裂」指实质性降低或退出投资仓位,而不只是尽调标记。
[CR001, CR003, CR009, CR020, CR026, CR028]7.2 业务与商业风险
最具结构性担忧的商业风险,是估值与收入倍数。Skild AI 估值 $14B+,2025 年收入约 $30M——收入倍数约 467 倍——跻身全球估值最高的未规模化 AI 公司。该倍数预设公司能在 24–36 个月内实现快速、持久增长,收入达到数亿美元。$30M 收入数字来自公司披露且未审计,没有第三方验证、没有披露客户名单、没有集中度数据,也没有按产品线或垂直领域拆分。缺少这些输入,就无法评估收入质量。 硬件 + 软件一体化机器人的商业化周期天然长于纯软件。企业采购机器人自动化通常包括:试点评估(3–6 个月)、现场安全评估、与现有仓库管理或制造执行系统集成(3–12 个月)、全面部署(6–18 个月)。这一结构性现实意味着,即便资金充足、销售能力更强的竞争对手,也需要 2–4 年才能达到有意义的企业收入规模。Skild 2023 年成立后到 2025 年才出现 $30M 收入,符合这一模式;但这也意味着通向 $200M+ ARR 的路需要多年持续执行。 客户集中度是未披露风险。考虑到 Skild 的企业销售属性和业务早期阶段,当前收入大部分来自少数战略投资方(Samsung、LG、Schneider Electric、CommonSpirit Health)并非不可能。客户离开或战略伙伴降温,可能造成与客户数量不成比例的重大收入下滑。没有披露任何客户名称或合同结构,使这一风险无法量化。 Zebra Robotics 收购(2026 年 4 月)带来第二条收入流,但也带来集成风险、平台转换引发的潜在客户流失,以及关键商业化阶段额外的运营复杂度。 [CR008, CR009, CR010, CR011, CR012, CR013]
有向图展示主要风险类别如何级联为下游投资假设风险。技术泛化失败和竞争替代都会通过收入停滞传导到估值受损。监管失败有一条直接通往市场准入丧失的路径。SoftBank 集中风险直接汇入资本约束,并放大所有其他风险。
[CR001, CR009, CR013, CR019, CR026, CR028]7.3 竞争与市场风险
Skild 所处的是技术领域资本最拥挤的竞争环境之一。Physical Intelligence (pi) 截至 2025 年 11 月累计融资约 $1.07B,估值 $5.6B,并已在多种机器人类型上部署其基础模型,提供每台机器人每月 $300 的订阅模式。Figure AI 在 2025 年 9 月 Series C 中融资超过 $1B,估值 $39B,凭借自有「Helix」VLA 系统同时瞄准工业和消费类人形机器人。两者都是硬件无关竞争对手,基础模型路径、融资规模和目标市场都与 Skild 相近。 Google DeepMind(RT-2、Gemini Robotics)、Amazon(家用机器人、履约 AI)、Tesla(Optimus)和 Meta(家用机器人研究)都拥有平台级算力、专有硬件生态和自有数据管线,资源量级远超任何初创公司。一旦它们投入资源对外授权机器人 AI 层,每一家都可能形成潜在替代威胁。 开源威胁在 2026 年 2 月变成现实,当时 Alibaba 开源了一个机器人 AI 基础模型,证明资源充足的参与者愿意将基础模型层商品化,以推动生态采用。如果一个能力足够的开源机器人基础模型获得广泛硬件伙伴采用——类似 LLaMA 对商业 NLP 模型定价的冲击——Skild 的专有模型授权定价将面对严重下行压力或被替代。Partnership on AI 和 Georgetown CSET 都记录了开源扩散给专有 AI 在位者带来的商业风险。 硬件 OEM 依赖风险是对称的:Skild 当前战略投资方(Samsung、LG、Schneider Electric)同时也是硬件渠道。如果任何伙伴发展自有 AI 能力、收购竞争对手或减少机器人生产,Skild 会同时失去收入来源和部署向量。 [CR013, CR014, CR015, CR016, CR017, CR018]
7.4 监管、安全与法律风险
Skild 的监管敞口重大且跨多个司法辖区,EU 带来的近期合规负担最重。已经生效的 EU AI Act 建立了分层风险框架,预计会把共享人类工作空间中的 AI 控制机器人归为高风险系统。高风险分类要求合格评定、人类监督协议、全面数据治理和审计轨迹。EU Machinery Regulation 2023/1230 将于 2027 年 1 月生效,对 AI 具身机器人提出三项关键新要求:(1)自治阈值,要求对表现出「通过经验自我演化行为」的系统提供安全证明(这正是 Skild 上下文学习所代表的能力);(2)所有联网机器人的全生命周期网络安全义务;(3)人机共享工作空间的协作风险映射。 2024 年 12 月生效的 EU Product Liability Directive 显著扩大了 Skild 的责任敞口。根据该指令,AI 软件系统即使没有硬件故障,也可能单独因缺陷承担责任。受伤方只需证明 AI 系统造成损害;缺陷推定会把举证责任转移给制造商,要求其证明安全。关键在于,该指令把供应链责任延伸开来:数据标注方和算法训练方——也就是 Skild——可能对已部署机器人的缺陷承担共同责任。 截至 2026 年 5 月,美国没有针对通用工作场所机器人的统一联邦监管框架。各州 AI 工作场所法律拼图(Colorado、Illinois、New York City)不直接监管机器人物理安全,但可能影响 Skild 客户使用 AI 驱动招聘或劳动力管理功能。OSHA 现有机器防护法规适用于已部署机器人,但早于基础模型 AI;面向自主 AI 机器人的执法姿态仍未明确。 Skild 获得 In-Q-Tel 投资,引入军民两用出口管制考量。In-Q-Tel 是 CIA 的风险投资机构;若技术具有情报或国防应用,其被投公司可能在 ITAR(International Traffic in Arms Regulations)和 EAR(Export Administration Regulations)下接受更严格审查。Skild 未公开披露这项投资带来的任何限制。 用互联网级人类视频剪辑训练 Skild Brain,会产生类似其他基础模型公司(Stability AI、OpenAI)面临的潜在版权敞口。公司未公开披露授权条款或合理使用法律分析。 [CR018, CR019, CR020, CR021, CR022, CR023]
| 规则 / 案件 | 司法辖区 | 状态 | 影响可能性 | 严重程度 | 缓释措施 | 残余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| EU 产品责任指令(Dec 2024) | EU | Dec 2024 生效;预计 2026–2027 年全面适用 | 高 — Skild 供应部署在 EU 实体机器人中的 AI 软件 | 阻断性 | 建立成文安全论证;维护可反驳缺陷推定的证据 | 高 — 任何涉及 Skild Brain 的 EU 机器人伤害事件,都可能让 Skild 作为算法训练方承担责任 | 获取关于 Skild 在 Art. 10 供应链条款下风险敞口的法律意见;确认产品责任保险 |
| EU AI Act — 高风险分类 | EU | 已生效;高风险义务在 2025–2026 年逐步实施 | 高 — 共享工作空间中的 AI 控制机器人很可能被归为高风险 | 阻断性 | 准备合格评定;搭建数据治理框架;制定人类监督协议 | 高 — 不合规会阻断 EU 市场准入;估计 Skild 目标市场 40%+ 在 EU | 确认 Skild Brain 是否被归为高风险;获取法律意见;审查合规路线图 |
| EU 机械法规 2023/1230 | EU | 已发布;January 2027 生效 | 高 — Skild 的上下文学习触发「自我演化行为」门槛 | 重大 | 立即启动合格评定流程;建立沙盒测试和生命周期文档 | 重大 — Jan 2027 后不合规会阻断 EU 销售;需要多年合规准备期 | 要求提供 Skild 的机械法规合规计划和时间表;确认 EU OEM 合作伙伴是否已将其纳入合作条款 |
| 美国州级 AI 职场法律(CO、IL、NYC) | 美国(州) | 已生效 — Colorado(May 2024)、Illinois(Sept 2024)、NYC Local Law 144 | 中 — Skild 客户面临合规要求;Skild 通过客户责任承担间接风险 | 较小 | 客户合同应纳入就业决策中 AI 使用的合规义务 | 轻微至中等 — 碎片化监管格局会让企业客户采购的合规复杂度上升 | 确认 Skild 合同是否将 AI 工作场所合规责任分配给客户 |
| In-Q-Tel 军民两用 / 出口管制(ITAR / EAR) | 美国(联邦) | 未知 — In-Q-Tel 投资已完成;具体限制未披露 | 中 — In-Q-Tel 被投公司面临更强的军民两用审查 | 重大 | 对投资条款做法律审查;如有需要,制定技术管制计划 | 重大 — 出口限制可能压缩某些地区客户群,或约束跨境 IP 共享 | 获取 In-Q-Tel 投资协议及任何相关技术管制计划副本;确认国际客户准入资格 |
| 互联网视频训练数据 — 版权风险 | 美国及国际 | 未披露授权;无公开法律意见;截至 2026 年 5 月未有诉讼提交 | 中 — 基础模型版权诉讼趋势(Stability AI、OpenAI 先例) | 重大 | 训练数据授权计划;客户合同中的赔偿条款 | 重大 — 不利判决可能要求只用授权数据重训,显著推高成本并拉长时间线 | 要求 Skild 提供训练数据版权风险法律分析;确认企业合同是否包含 AI 来源赔偿 |
严重性按阻断性到轻微排序。鉴于 Skild 已表述国际部署雄心,欧盟监管风险短期内属重大风险。美国联邦层面对工作场所机器人立场仍不明确(截至 2026 年 5 月没有专门机构或规则)。In-Q-Tel 限制需等待文件审查,目前未知。版权风险是所有基础模型公司共同面对的开放问题。
[CR018, CR019, CR020, CR021, CR022, CR023]5×3 风险热力图,把 Skild AI 主要风险类别映射到影响和可能性两个维度。右上象限(高可能性、高影响)包含技术泛化失败和估值 / 收入脱节,二者都可能打破投资假设。SoftBank 投资人集中和 EU 监管暴露属于高影响但中等可能性的近期触发因素。人员和运营风险很重要,但部分可缓释。
[CR001, CR009, CR019, CR026, CR033, CR038]Skild AI 关键依赖关系图,展示公司与主要资本、算力、硬件和监管对手方的关系。SoftBank 是最关键单一节点,且没有替代者。NVIDIA 既是投资人,也是算力依赖。硬件 OEM 伙伴同时是战略投资人和收入来源;若其自研 AI,将形成结构性利益冲突。
[CR013, CR022, CR026, CR027, CR036]7.5 财务与资本结构风险
Skild 的资本结构特点是投资者极度集中,烧钱与收入轨迹高度不确定。SoftBank 领投了 Series A($300M,2024 年底)、Series B($135M,估值 $4.5B,2025 年 6 月)和 Series C($1.4B,估值 $14B,2026 年 1 月)。这意味着单一投资者控制了全部三轮机构融资的主要资本生命线。最可能持有 Skild 敞口的 SoftBank Vision Fund 2,在截至 2025 年 3 月的财年因组合减记报告 $3.6B 亏损。更广义的 Vision Fund 到 2023 年两年累计亏损约 $48B。SoftBank 资本受限、组合再平衡或内部授权转向,都可能在 Skild 的关键节点削弱后续投资能力。 $14B 估值是在短短七个月内从 $4.5B Series B 估值翻了三倍后达到的。这种私募市场估值膨胀速度,更多由叙事动能和战略投资者热情推动,而不是收入轨迹推动,造成结构性风险:单个负面数据点(错过收入目标、安全事故、竞争对手突破)就可能触发显著估值重置,损害未来融资条款和员工股权留存。 月度总烧钱率未公开披露。按公司规模(Zebra 收购后估计 200–400 名员工)、模型训练基础设施(GPU 算力)、办公室扩张和收购整合估算,总烧钱率为每月 $30–60M。在这一速度下,$1.4B Series C 资金(扣除 Zebra 收购现金对价,金额未披露)意味着自 2026 年 1 月起约 18–36 个月总资金续航——Series D 触发点落在 2027 年底或 2028 年初。收入会部分抵消烧钱,但 $30M ARR 折算到月度约为每月 $2.5M,相对于估计 $30–60M 月度烧钱,抵消很小。 Zebra Robotics 收购增加了收购整合成本、潜在业绩对赌义务和 Fetch Robotics 遗留运营成本。收购价未披露;任何重大现金组成都会使净 Series C 资金续航低于标题估算。 [CR026, CR027, CR028, CR029, CR030, CR031]
| 依赖项 | 交易对手 | 角色 | 集中度 | 失效情景 | 严重性 | 缓释措施 | 剩余风险敞口 |
|---|---|---|---|---|---|---|---|
| 资本 — 主要投资方 | SoftBank(Vision Fund) | 领投 Series A、B、C;主要机构资本提供方 | 极高 — 三轮均由单一投资方主导;没有其他投资方领投过任一轮 | SoftBank Vision Fund 2 流动性约束、组合再平衡或授权范围变化,迫使 Skild 接受过桥条款或失去后续投资 | 阻断性 | 在 Series D 分散投资者基础;培养 NVIDIA、Samsung 等战略投资方成为联合领投 | 高 — SoftBank Vision Fund 2 在 2025 财年录得 $3.6B 亏损;后续投资能力和意愿不确定 |
| 算力与仿真基础设施 | NVIDIA | Isaac Lab 仿真;Cosmos Transfer;NVentures 共同投资;可能拥有优先 GPU 使用权 | 高 — 训练管线建在 NVIDIA 仿真栈上;切换成本很高 | NVIDIA 降低 Skild 优先级或减少算力分配;GPU 出口限制影响供应 | 重大 | 多云算力策略;合同锁定算力配额;NVIDIA 的战略共同投资提供一定利益一致性 | 中 — NVIDIA 同时也是直接竞争对手的基础设施伙伴;竞争动态可能变化 |
| 硬件部署渠道 | Samsung、LG、Schneider Electric、CommonSpirit 等合作方 | 战略投资方兼客户;Skild Brain 的主要硬件部署渠道 | 高 — 四家战略投资方可能贡献当前收入大头 | 任一伙伴自研 AI、收购竞争对手或减少机器人产量 | 重大 | 分散硬件 OEM 合作;维持技术优势,让伙伴选择授权而非自研有理由 | 高 — 硬件 OEM 自研 AI 是机器人行业已有记录的风险 |
| 机群编排与企业客户基础 | Zebra Technologies / Fetch Robotics(已收购) | 提供 Symmetry 平台、企业客户关系和运营型机器人经验 | 中 — 收购后 Zebra 不再是独立交易对手,但整合执行成为风险 | 整合失败、关键人才流失,或从 Zebra 向 Skild 迁移平台期间客户流失 | 重大 | 设立专项整合管理;为关键人员提供留任包;用合同约定客户迁移 SLA | 高 — 收购于 2026 年 4 月完成;整合仍在进行,未披露整合里程碑或成功指标 |
| 军民两用国家安全投资方 | In-Q-Tel(CIA 风投部门) | 战略投资方;引出军民两用考量 | 低至中 — 少数股投资方;角色偏战略而非运营 | 出口管制限制损害国际销售或跨境数据共享;伙伴国家施加限制 | 重大 | 对投资条款做法律审查;维持技术管制计划;如有需要,拆分出口受控开发线 | 中 — 限制未公开披露;没有文件审查就无法量化风险 |
各行按严重性排序。SoftBank 集中度是最尖锐的结构性依赖。由于缺少已披露客户 / 收入数据,硬件 OEM 集中风险目前无法量化。Zebra 整合风险集中在收购后 12–18 个月窗口内。
[CR026, CR027, CR028, CR029, CR032, CR036]7.6 运营与人员风险
关键人物风险很高。Deepak Pathak(CEO)和 Abhinav Gupta(President)都是 Carnegie Mellon Robotics Institute 教授,2023 年创立 Skild AI。他们的声誉、学术网络,以及对研究和架构的直接技术贡献,支撑公司吸引顶尖人才、获得战略投资者并维持企业客户技术可信度的能力。两人都没有从早期阶段一路搭建并规模化一家商业科技公司至数亿美元收入的经历。他们合计 25+ 年 AI/机器人学术经验是深厚优势,但不能直接转化为规模化商业运营经验。 人才竞争激烈。Skild 必须从很小的全球机器人 AI 研究员和工程师人才池中招聘,而 Google DeepMind、Meta AI、Tesla Optimus、Amazon、Figure AI 和 Physical Intelligence 也在争夺这些人——这些公司都能提供有竞争力的薪酬、算力资源和各自有吸引力的使命。资深研究人员流失,或未能按所需速度招聘,都会拖慢模型开发、降低部署质量并削弱数据飞轮。 Zebra Robotics 收购(2026 年 4 月,包括 Fetch Robotics 传承团队)增加组织整合风险。Skild 是研究密集型 AI 初创公司;Zebra Technologies 是成熟企业硬件公司。文化对齐、系统集成和被收购团队留任,是企业机器人收购中有记录的难题。如果整合处理不当,构建 Symmetry 平台的关键 Fetch Robotics 技术人员可能离开,使 Skild 失去服务现有 Zebra 企业客户所需的制度知识。 公司没有披露联合创始人继任计划,没有披露具备企业规模化经验的 COO 或首席商务官,也没有证据显示其拥有通常需要的成熟企业 SaaS 领导团队(VP Sales、VP Customer Success),以支撑在企业自动化市场从 $30M ARR 扩到 $200M ARR。学术转商业的领导力转换风险,是深科技高校分拆公司常见失败模式。 [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 |
|---|---|---|---|---|---|
| Deepak Pathak — CEO 兼联合创始人 | 首要技术愿景制定者、主要对外代表和核心募资人;未披露接班人 | 低(离职)/ 高(集中风险) | 阻断性 | 带 cliff 的长期股权归属;董事会治理要求重大决策取得联合创始人同意 | 确认归属安排、离职后锁定期和董事会构成;评估创始人以外技术领导层的厚度 |
| Abhinav Gupta — 总裁兼联合创始人 | 共同核心技术架构师;CMU 网络锚点;主要研究可信度来源 | 低(离职)/ 高(集中风险) | 阻断性 | 同上;组织对两人的共同依赖使连续离职风险更尖锐 | 同上;评估联合创始人以下研究领导团队的厚度 |
| 从学术到商业化的领导力转换 | 两位创始人均未商业化放大过企业软件公司;销售 VP、客户成功 VP、CFO 角色未披露 | 中 — 缺口是结构性的,不是个人问题;公司 ARR 超过 $50M 后压力会加剧 | 重大 | 招聘有经验的企业 SaaS 和机器人商业化高管;搭建与研究领导层并行的商业领导团队 | 确认商业职能的 C-level 和 VP 级招聘;评估销售人数和销售管线管理流程成熟度 |
| 机器人 AI 人才获取与留存 | 研究人才争夺来自 Google DeepMind、Meta AI、Tesla Optimus、Amazon、Figure AI、Physical Intelligence | 高 — 人才短缺是结构性的;薪酬和股权竞争激烈 | 重大 | CMU 毗邻优势带来人才管线;$14B 估值下股权上行仍有意义,但已部分摊薄 | 按职能确认员工数;评估自愿流失率;对照同业审查薪酬基准 |
| Zebra / Fetch Robotics 收购整合 | 关键 Symmetry 平台工程师可能在收购后离开;AI 初创公司与企业硬件公司的文化需要整合 | 中 — 收购后人才流失很常见;文化错配是已有记录的风险 | 重大 | 留任包;清晰汇报线;共享 OKR;面向现有 Zebra 客户的客户连续性 SLA | 要求提供收购公告以来的员工留存数据;评估收购前后 Fetch Robotics 工程团队人数变化 |
严重性按阻断性到重大排序。创始人关键人风险是短期最尖锐、也最难缓释的人才风险。公司试图把收入扩大到 $100M 以上后,商业执行风险将从 2027 年起占主导。
[CR033, CR034, CR035, CR036, CR037]7.7 附录
08估值
8.1 当前估值与收入倍数
Skild AI 于 2026 年 1 月 14 日宣布的 $1.4B Series C,由 SoftBank 领投,确立了超过 $14B 的投后估值。这相当于在七个月内从 $4.5–4.7B Series B 估值(2025 年 6 月)翻了三倍,相比 $1.5B Series A 估值(2024 年 7 月)约提升 9.3 倍。Crunchbase 显示各轮已披露融资总额为 $1.83B,CEO Deepak Pathak 称总额超过 $2B。 Series C 新闻稿披露的 2025 年收入约为 $30M,描述为 2025 年「短短几个月」内从零增长而来。这意味着历史收入倍数约 467 倍($14B / $30M)。即使用保守的 2026 年 50% 收入增速(至 $45M),2026 年前瞻倍数仍约 311 倍。即使用乐观的 200% 增速(至 $90M),前瞻倍数也约 156 倍。 作为参照,公开市场中领先企业 AI SaaS 公司在高速增长阶段的前瞻收入倍数为 15–50 倍。前沿 AI 基础设施公司(如 IPO 时的 CoreWeave)前瞻收入倍数为 20–30 倍。机器人硬件公司(如高峰期 iRobot)收入倍数为 2–8 倍。467 倍历史倍数只有在「赢家拿走大部分」平台论成立时才说得通:Skild 需要在十年内拿下数万亿美元物理 AI 市场中不成比例的一大块份额。 这一估值由成熟机构投资者给出(SoftBank、NVIDIA、Samsung、LG、Schneider Electric、Macquarie、Bezos Expeditions、Salesforce Ventures、IQT),为牛市论提供了一定验证。不过,2024–2026 年 AI 机器人私募市场估值被战略投资者系统性推高,这些投资者追求生态站位,不只追求纯财务回报——这一因素会把估值推到基本面支撑之外。 [CV001, CV002, CV003, CV004, CV005, CV006]
Skild AI 从估算种子轮到 Series C 的估值演进。每一次跃升代表该轮融资新增的估值。种子轮到 Series C 的总估值增长约 $13.9B,历时 24 个月——属于私营 AI 公司历史上最快的估值爬坡轨迹之一。
种子轮估值为分析师估算。Series A、B、C 估值来自公开新闻稿。跃升金额根据披露估值计算。
[CV001, CV002, CV003, CV005]8.2 可比公司估值
机器人基础模型和物理 AI 赛道在 2024–2026 年经历了私募市场估值快速升级,背后是机构资本追逐「通用机器人大脑」论。关键可比公司: Physical Intelligence (pi):2025 年底以 $5.6B 估值融资 $600M。Physical Intelligence 据称收入很少,隐含 ARR 倍数基本无法定义。其主要资产是 pi0 VLA 模型和强研究团队(来自 Google、Berkeley、CMU)。Physical Intelligence 于 2025 年 2 月开源 pi0,形成一个直接竞争 Skild 产品的免费模型。 Figure AI:2026 年初以 $39B 估值融资,并以 BMW 制造部署作为具名客户证据。Figure AI 打造自有人形机器人硬件(Figure 01/02),结合硬件和软件经济性。$39B 估值是物理 AI 赛道最高估值,并由有形部署证据锚定。 1X Technologies:挪威人形机器人公司,据报道正在洽谈最高 $10B 估值融资。1X 打造 NEO 人形机器人硬件,并开发自有 AI 层,类似 Figure AI 的硬件 + 软件一体化模式。 Boston Dynamics(Hyundai):不是直接可比公司(已被收购,非 VC 融资),但 Spot 和 Atlas 部署为企业机器人采用时间线,以及从试点扩到量产的难度,提供了基准。 Covariant(RFM-1):2024 年被 Amazon 收购,此前融资 $222M;Covariant 被 Amazon 以未披露溢价收购,释放出物流巨头对 AI 机器人软件层战略价值的认可。 放在这组同业中,Skild 的 $14B 估值仅次于 Figure AI,也是该赛道最高的估值 / 收入比。Skild 的溢价来自其全机体、硬件无关基础模型论,以及 NVIDIA、Samsung 和 IQT 释放的投资者信号。 [CV008, CV009, CV010, CV011, CV012, CV013]
| 公司 | 最新估值 | 累计融资 | 收入 / ARR | 收入倍数 | 商业模式 | 主要投资方 |
|---|---|---|---|---|---|---|
| Skild AI | $14B(Jan 2026) | $1.83B+ | ~$30M(2025,公司披露) | ~467x 追踪收入 | 纯软件基础模型;OEM/SI 渠道 | SoftBank、NVIDIA、Samsung、LG、Bezos、IQT 等投资人 |
| Figure AI | $39B(2026) | $1.5B+ | 未披露(BMW 部署已确认) | N/A(已披露商业化收入前阶段) | 硬件 + 软件人形机器人;直销企业客户 | Microsoft、NVIDIA、OpenAI(前投资方)、BMW、Intel Capital |
| Physical Intelligence(pi) | $5.6B(late 2025) | ~$600M | 未披露(商业化收入前) | N/A | 纯软件基础模型;已开源 pi0 | Bezos、Sequoia、OpenAI Fund |
| 1X Technologies | ~$10B(估计,2025–2026) | ~$500M+ | 未披露 | N/A | 硬件 + 软件人形机器人(NEO) | EQT、Samsung、NordicNinja |
| Covariant(已收购) | N/A(2024 年被 Amazon 收购) | 累计融资 $222M | 未披露 | N/A | 机器人 AI 软件(RFM-1);聚焦仓储 | Amazon(收购方)、Index Ventures |
| Boston Dynamics(Hyundai) | ~$1B(Hyundai 2021 年收购) | N/A(已收购) | 未披露 | N/A | 硬件机器人(Spot、Atlas);软件层 | Hyundai(100% 所有者) |
估值数据来自截至 2026 年 5 月的公开新闻稿、投资方公告和 Crunchbase。收入倍数仅使用已披露 ARR 数据计算。Figure AI、Physical Intelligence 和 1X 收入未公开披露。Covariant 被 Amazon 收购,估值未披露。
[CV001, CV002, CV007, CV008, CV009, CV010]| 情景 | 假设 2026 年 ARR 增长 | 2026 年 ARR 估计 | $14B 估值隐含前瞻倍数 | 公开市场可比倍数 | 结论 |
|---|---|---|---|---|---|
| 悲观情景 | 50% YoY | $45M | 311x | 15–20x(企业 AI SaaS) | 严重高估;需要倍数压缩或增长显著加速 |
| 基准情景 | 100% YoY | $60M | 233x | 20–30x(超高增长 AI 平台) | 按财务指标看估值偏高;战略期权价值可部分抵消 |
| 乐观情景 | 200% YoY | $90M | 156x | 30–50x(前沿 AI 平台) | 估值拉伸,但在赢者拿大部分平台假设下可能站得住 |
| 超乐观情景 | 400% YoY | $150M | 93x | 50–100x(AI 平台峰值倍数) | 理论上说得通;高度依赖市场占领和护城河耐久性 |
| 退出可验证情景(2028) | 到 2028 年 CAGR 150% | $750M–$1B ARR | 14–19x(成熟平台倍数) | 10–20x(规模化企业 SaaS) | 可支撑 $14B 入场价;需要 3 年内执行到位,且没有重大竞争扰动 |
所有 ARR 预测均为分析师估计,基于已披露的 $30M 2025 年收入和假设增长率。Skild AI 未提供收入指引。公开市场可比倍数基于截至 2026 年 Q1 的 SaaS/AI 平台可比公司。
对比物理 AI 与机器人基础模型赛道的收入倍数。Skild AI 的 467x 尾随 ARR 倍数, 在已披露收入的公司中最高,说明其商业化相对估值仍处在最早期。
收入倍数按公开披露的估值和 ARR 计算。未披露 ARR 的公司标为「pre-revenue」或「N/A」。除另有注明外,所有倍数均为尾随口径。
[CV001, CV004, CV008, CV009, CV010, CV011]8.3 牛市与熊市估值情景
$14B 估值在乐观情景下能撑住,在悲观情景下也很容易被挑战: 乐观情景:Skild AI 在 2025 至 2030 年实现 200%+ 的收入 CAGR,到 2029–2030 年 ARR 超过 $1B。Goldman Sachs 预计,到 2035 年人形机器人市场 规模将达 $38B;MarketsandMarkets 预计,到 2030 年 AI 机器人市场规模将达 $33.4B(40.4% CAGR)。如果 Skild 拿下 $33B 市场的 5–10%,收入运行率将达到 $1.65B–$3.3B。按 10x 远期收入倍数(规模化高溢价 SaaS 倍数)计算,隐含估值为 $16.5B–$33B——足以解释当前 $14B 的入场价。这个情景要求 Skild 顶住开源替代品, 继续守住软件优先领先位置,并把 OEM / SI 渠道扩到全球。 基准情景:Skild 以 80–120% CAGR 增长,到 2028 年 ARR 达到 $200–300M。按 15–20x 远期收入倍数(这一规模下更现实的企业 AI 倍数)计算,隐含估值为 $3B–$6B——相对当前 $14B 有明显下行。这个情景对应的是:公司部署跑通,但算力成本 和开源竞争挤压利润率。 悲观情景:收入增速降至 50% CAGR 以下,到 2027 年 ARR 仅为 $60–100M。 开源 GR00T 或 pi0 被 Skild 的 OEM 渠道伙伴采用。按 10x 远期收入计算,隐含估值为 $600M–$1B——较 Series C 入场价回撤 90%+。考虑到 IQT 和 NVIDIA 的战略投入, 这一情景概率较低;但如果基础模型层商品化快于预期,并非不可能。 467x ARR 倍数不给商业执行风险和倍数压缩留下缓冲,使 Skild 成为 2026 年私人市场 对估值最敏感的 AI 投资之一。 [CV015, CV016, CV017, CV018, CV019, CV020]
| 轮次 | 日期 | 金额 | 投后估值 | 较上一轮跃升 | 领投方 | 当轮关键里程碑 |
|---|---|---|---|---|---|---|
| 种子轮 | 2023 | 未披露(估计 ~$10M) | 未披露(估计 ~$50–100M) | N/A(首轮外部融资) | 未披露 | 公司成立;初版 Skild Brain 原型;CMU 孵化 |
| Series A | July 9, 2024 | $300M | ~$1.5B | ~15–30x(种子轮到 A 轮) | Lightspeed Venture Partners、Coatue、SoftBank 等投资人 | 走出隐身期;6 个垂直场景启动;在 100K+ 种仿真形态上训练 |
| Series B | June 2025 | ~$135M($100M SoftBank + $25M NVIDIA + $10M Samsung,战略资本) | ~$4.5–4.7B | ~3x(A 轮到 B 轮) | SoftBank(领投) | LG CNS 合作;HPE 基础设施合作;收入向 $30M 爬升 |
| Series C | January 14, 2026 | $1.4B | >$14B | ~3x(B 轮到 C 轮) | SoftBank(领投) | 披露 2025 年 ARR 约 $30M;9 家具名投资方;IQT/国防信号;全球扩张 |
| Series C 后(Zebra) | April 2026 | N/A(现金 + 股权对价) | N/A(战略交易) | N/A | Zebra Technologies(股权接收方) | 收购 Symmetry Fulfillment;仓储垂直场景扩大 |
种子轮金额和估值为分析师估计,未公开披露。Series A、B、C 数据来自公开新闻稿(BusinessWire、TechCrunch、Crunchbase)。跃升倍数基于已披露估值近似计算。CEO 称累计融资超过 $2B;Crunchbase 追踪为 $1.83B。
8.4 市场规模与退出路径
Skild AI 的估值押注的是物理 AI 市场预测增长,市场规模远高于当前披露收入。 Goldman Sachs 将 2035 年人形机器人市场预测上调至 $38B,高于此前 $6B 的预测, 理由是 AI 突破和机器人制造成本下降 40%。MarketsandMarkets 预计,AI 机器人市场 将从 $6.1B(2024)增至 $33.4B(2030),CAGR 为 40.4%。Morgan Stanley 预计, 到 2050 年全球人形机器人机会将达 $5T,依据是大规模部署情景。 按这些预测,只要 Skild 拿下不高的份额,TAM 就足以支撑 $14B 私募估值。 但结构性障碍也很清楚:(1)人形机器人硬件市场由大型工业 OEM(FANUC、ABB、 KUKA、Yaskawa)主导,天然限制纯软件平台利润率;(2)训练和运行前沿机器人模型 的算力成本会随部署规模上升,压缩长期毛利率;(3)2026–2028 年窗口里,AI 模型层 的开源商品化可能比预期更快夺走市场份额。 退出路径包括:(1)IPO——若估值 $20B+、ARR $200M+ 且增长持续,则有可能; 最早也要 2027–2028 年;(2)战略收购——NVIDIA、Samsung、LG 或 SoftBank 投资组合公司收购具备战略合理性;(3)SPAC 或直接上市——有可能,但在当前市场环境下 概率较低。IQT 投资打开潜在国防 / 情报采购渠道,在部分地区可能成为相对直接竞争对手的 战略护城河。 [CV022, CV023, CV024, CV025, CV026, CV027]
| 风险因素 | 类别 | 严重性 | 对估值影响 | 缓释因素 | 证据基础 |
|---|---|---|---|---|---|
| 未经审计收入($30M) | 财务质量 | 高 | 若收入虚高或非经常性,467x 倍数相对真实 ARR 仍被低估 | 获取经审计财务;确认经常性合同结构 | 仅有公司新闻稿;无第三方验证 |
| 开源商品化(GR00T、pi0) | 竞争 | 高 | 若 OEM 伙伴转向免费替代方案,Skild 的软件定价权会坍塌 | 专有数据飞轮;企业级支持护城河;1000x 训练数据优势 | GR00T 于 2025 年 3 月开源;pi0 于 2025 年 2 月开源 |
| 三轮融资均集中依赖 SoftBank | 财务 / 治理 | 高 | SoftBank Vision Fund 2 在 FY2025 录得超过 $3.6B 亏损;后续跟投资金能力不确定 | NVIDIA、Samsung、LG 带来更分散的战略资本基础 | 多家财经媒体报道;SoftBank 业绩披露 |
| 缺少独立基准测试 | 技术 | 高 | 没有同行评审的性能数据,Skild 所称「1000x 训练数据优势」无法核验 | 上下文适应演示;NVIDIA 案例研究;HPE 合作 | 截至 2026 年 5 月,尚未发布学术论文或第三方基准测试 |
| 467x ARR 倍数——估值倍数极端压缩风险 | 估值 | 高 | 增长一旦降速或倍数收缩,Series C 价格可能回撤 50–90%+ | 乐观情形需要 2028–2030 年 ARR 达到 $1B+,才撑得住入场价格 | 按 $14B/$30M 计算;企业 AI 中没有可比公司达到这一倍数 |
| Zebra 整合风险 | 运营 | 中 | 整合 Symmetry Fulfillment 会增加硬件编排复杂度;过渡期存在客户流失风险 | Zebra 股权绑定;分阶段整合路径 | 收购于 2026 年 4 月宣布;未披露整合指标 |
| IQT / 双用途监管暴露 | 监管 | 中 | IQT 投资释放出国防 / 情报用途信号;带来出口管制义务(ITAR、EAR),并限制投资者基础 | 亲美合规框架;IQT 提供监管路径支持 | IQT 已确认为 Series C 投资方;监管暴露为推断 |
| 视频训练数据版权 | 法律 | 中 | Skild 使用互联网来源的视频数据训练;潜在版权侵权责任类似 AI 图像生成器案件 | 行业共性问题;法律结果不确定;可能需要授权协议 | 根据训练方法推断;截至 2026 年 5 月,未见针对 Skild 的公开诉讼 |
风险严重度由分析师根据公开证据评估。「对估值的影响」反映风险落地后的定性影响。「高」严重度风险指可能让估值受损 50%+,或造成不可恢复结果的风险。
关键物理 AI 与机器人细分市场的规模预测区间,展示支撑 Skild AI $14B 估值的总可用市场(TAM)扩张。 所有预测均来自 2024–2026 年的第三方研究机构或投资银行。
所有市场预测均来自第三方来源(Goldman Sachs、MarketsandMarkets、Morgan Stanley),后续可能修订。Skild AI 尚未公开自身 TAM 估算。
[CV022, CV023, CV024, CV025]8.5 估值结论
相对已披露财务指标,Skild AI 的 $14B 估值偏高。467x 的历史 ARR 倍数是物理 AI 赛道最高水平,退出时要解释这个估值,需要一组很苛刻的结果同时兑现:到 2028 年收入 CAGR 持续维持 150–200%,面对开源竞争仍守住软件层定价权,且市场环境在公开市场流动性 下仍支持高溢价 ARR 倍数。 这个估值并非完全站不住脚——投资人名单(SoftBank、NVIDIA、Samsung、LG、 Bezos Expeditions、IQT)覆盖赛道内最老练的一批机构,IQT 信号也带来竞争对手没有的 国防护城河。如果平台模型真的占据「机器人领域 Android OS」的位置,极高估值可以成立。 但缺少独立基准、收入未经审计、没有具名终端客户证据,意味着 $14B 主要由战略意图和 投资人逻辑支撑,而不是可验证的财务基本面。 任何接近 Series C 价格的投资,都必须押在三项前提上:(a)经审计财务验证的乐观情景 收入轨迹;(b)相对 GR00T / pi0 已确认的竞争护城河;(c)在 $14B+ 入场价上能产生 有意义回报的可行退出估值。若 $14B 投入只要求 1.5x(温和回报),投资期内退出估值也必须 超过 $21B——只有乐观情景能做到。 [CV001, CV004, CV006, CV017, CV020, CV028]
| 维度 | 评估 | 证据基础 | 投资含义 |
|---|---|---|---|
| 建议 | 继续研究 / 跟踪 | 收入未经审计;缺少独立基准测试;开源风险未解除 | 若没有经审计收入、客户分群数据和独立技术基准,不应以 $14B+ 估值投资 |
| 置信度 | 低 | 467x ARR 倍数;公司口径 $30M 收入;未公开客户名单 | 基础收入质量和竞争护城河耐久性高度不确定 |
| 风险评级 | 高 | 开源商品化;SoftBank 集中度;缺少独立基准;极端估值倍数 | 四项高严重度风险并发,Series C 价格出现 >50% 回撤的概率不低 |
| 估值立场 | 偏贵 | $14B 对应 467x 历史 ARR,而高增长 SaaS 可比倍数为 15–50x | 只有乐观收入轨迹兑现,估值才有条件说得通 |
| 入场纪律 | 当前价格放弃;跟踪里程碑 | 即便采用乐观增长假设,467x 历史 ARR 倍数也没有安全边际 | 若经审计 ARR 超过 $100M 且竞争护城河得到验证,可在 Series D 重新评估 |
| 决策含义 | 等待经审计财务 + 连续 2+ 个季度 ARR | 现有证据不足以区分战略价值和基本面价值 | 触发条件:经审计 $100M ARR、具名 Tier-1 客户、独立基准测试 |
建议仅由分析师基于公开证据评估,不构成投资建议。关键输入包括未经审计的 $30M 收入、$14B Series C 估值,以及 GR00T 和 pi0 带来的开源商品化风险。
[CV003, CV004, CV028, CV029, CV030]| 主题 | 缺失证据 | 为什么重要 | 责任方或尽调路径 |
|---|---|---|---|
| 经审计收入 | 经审计财务报表,确认 2025 年 $30M ARR 为经常性且已确认收入 | 没有审计确认,467x 倍数只能靠信心承接;收入质量(经常性 vs 一次性)未知 | 向公司索取;至少包括前 2 个财年的审计,以及本年度管理账 |
| 客户收入瀑布 | 具名客户清单,包含收入归因、分群年份、合同期限和流失率 | 没有这些数据,就无法评估增长率和留存;OEM 与终端用户收入拆分未知 | 公司数据室;至少通过 3 通客户推荐访谈验证 |
| 分收入流毛利率 | 按计算成本调整后的毛利率,拆分 Skild Brain 软件与 Symmetry 硬件服务 | 软件层估值要靠 60–80%+ 毛利率支撑;未披露的计算成本可能显著压缩毛利 | 要求按收入流披露毛利率;与 SaaS 基准对比 |
| 独立技术基准测试 | 第三方或同行评审评估,在标准机器人任务上比较 Skild Brain、GR00T 与 pi0 | 「1000x 训练数据优势」未获验证;竞争护城河是 $14B 平台溢价的核心依据 | 委托独立基准测试或确认学术合作;索取所有内部基准数据 |
| SoftBank 承诺 | 确认 SoftBank Vision Fund 2 对 Skild AI 的后续跟投资金能力和储备政策 | SoftBank 领投全部 3 轮机构融资;Vision Fund 2 在 FY2025 录得亏损;融资集中风险实质存在 | SoftBank 投资者关系;投资条款清单中的按比例跟投权披露 |
| IQT / 出口管制框架 | 正式 ITAR/EAR 合规文件,以及对美国以外部署的任何合同限制 | IQT 投资带来出口管制义务;限制可能压缩 Skild 可触达的国际 TAM | 公司法律顾问;索取合规备忘录和司法辖区限制 |
| Zebra / Symmetry ARR | 收购时 Symmetry Fulfillment ARR、客户数量和整合时间表 | Zebra 收购改变收入基础和成本结构;指标未披露,无法更新倍数计算 | 公司 M&A 披露;索取收购日 Symmetry 财务报表 |
这些尽调问题是在当前 $14B+ 估值下投入资本前的门槛要求。前四项中任一项结果不合格,单独就足以否决以这一价格投资。
[CV030, CV029, CV003, CV019, CV006, CV026]从市场规模、商业验证、竞争护城河、财务、风险和估值,一路推到在当前 $14B Series C 入场价下 「继续研究 / 观察」的建议。每个节点是一项关键尽调维度;每条边是逻辑依赖。
该流程图呈现分析师基于公开证据判断的决策逻辑。建议取决于证据质量;后续若有更多披露,可能调整。
[CV003, CV004, CV017, CV018, CV021, CV028]8.6 展示材料
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;任何投资决策前,都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 | Skild AI was founded in May 2023 by Deepak Pathak and Abhinav Gupta. | 高 | SO001, SO009, SO021 |
| CO002 | Skild AI is headquartered in Pittsburgh, Pennsylvania. | 高 | SO001, SO002, SO007 |
| CO003 | Skild AI also maintains an office in the San Francisco Bay Area. | 高 | SO001, SO002 |
| CO004 | Skild AI opened a Bengaluru, India office in February 2026, its first international expansion. | 高 | SO002, SO013, SO014 |
| CO005 | The Bengaluru office was announced on February 19, 2026 and opened as Skild's first office outside the US. | 高 | SO013, SO016 |
| CO006 | Deepak Pathak is CEO and co-founder of Skild AI. | 高 | SO001, SO002, SO004 |
| CO007 | Abhinav Gupta is President and co-founder of Skild AI. | 高 | SO001, SO002, SO004 |
| CO008 | Deepak Pathak earned a gold medal in Computer Science from IIT Kanpur and completed a PhD at UC Berkeley. | 高 | SO004, SO022 |
| CO009 | Deepak Pathak conducted foundational research at Facebook AI Research (FAIR) before joining CMU's Robotics Institute as the Raj Reddy Associate Professor. | 高 | SO004, SO022 |
| CO010 | Abhinav Gupta is a tenured professor at CMU's Robotics Institute and was a founding member and research leader at FAIR Robotics (Facebook/Meta). | 高 | SO004, SO015 |
| CO011 | The founders have a combined h-index of over 150 and more than 90,000 academic citations. | 高 | SO001, SO021 |
| CO012 | Skild AI's team includes robotics and AI experts recruited from Meta, Tesla, Nvidia, Amazon, Google, CMU, Stanford, and UC Berkeley. | 中 | SO001, SO021 |
| CO013 | LinkedIn shows approximately 85 employees at Skild AI as of early 2026. | 低 | SO011 |
| CO014 | Tracxn's data (based on legal-entity filings) recorded 34 employees at Skild AI as of December 31, 2024. | 中 | SO012 |
| CO015 | Skild AI raised an undisclosed seed round from Sequoia Capital in 2023, led by partner Stephanie Zhan. | 高 | SO004, SO022 |
| CO016 | Skild AI raised a $300M Series A on July 9, 2024 at a $1.5B valuation, led by Lightspeed Venture Partners, Coatue, SoftBank Group, and Jeff Bezos (Bezos Expeditions); additional investors included Felicis Ventures, Sequoia, Menlo Ventures, General Catalyst, CRV, Amazon, SV Angel, and Carnegie Mellon University. | 高 | SO001, SO004, SO008, SO021 |
| CO017 | Skild AI raised approximately $135M in a Series B in June 2025 at a $4.5B valuation, led by SoftBank (approximately $100M), with Nvidia ($25M) and Samsung ($10M). | 中 | SO007, SO010, SO023, SO024 |
| CO018 | Skild AI raised approximately $1.4B in a Series C on January 14, 2026 at a valuation exceeding $14B, led by SoftBank Group; investors included NVentures (NVIDIA), Macquarie Capital, Jeff Bezos, Samsung, LG, Schneider Electric, CommonSpirit, Salesforce Ventures, IQT, and others; Lightspeed, Felicis, Coatue, and Sequoia doubled down. | 高 | SO002, SO005, SO006, SO007, SO017 |
| CO019 | CEO Deepak Pathak stated in January 2026 that Skild AI had raised more than $2 billion in total; Crunchbase tracked $1.83B raised across four rounds. | 高 | SO005, SO007 |
| CO020 | Skild AI is building the 'Skild Brain,' which the company describes as the industry's first unified robotics foundation model. | 高 | SO002, SO006, SO019 |
| CO021 | The Skild Brain is omni-bodied and can control any robot—including quadrupeds, humanoids, tabletop arms, and mobile manipulators—without prior knowledge of the robot's exact physical form. | 高 | SO002, SO006 |
| CO022 | The Skild Brain demonstrates the ability to adapt to unpredictable scenarios such as loss of limbs, jammed wheels, increased payload, or an entirely new robot body, without retraining or fine-tuning. | 中 | SO002, SO006 |
| CO023 | Skild AI achieved an in-context learning breakthrough for robotics—which the company describes as a first in the field—earning Best Paper Nominations at top robotics conferences. | 中 | SO002, SO006 |
| CO024 | Skild AI trains its model on approximately 1000x more data than competing robotics models. | 中 | SO001, SO021 |
| CO025 | The Skild Brain's training data comes from four sources: large-scale simulation (trillions of synthetic experiences), internet videos (billions of human action videos), teleoperation, and real-world deployments. | 高 | SO006, SO020 |
| CO026 | Skild AI's data flywheel means each real-world deployment generates additional training data, continuously improving the Skild Brain's generalization capabilities. | 高 | SO006, SO020 |
| CO027 | The Skild Brain demonstrates emergent capabilities—behaviors not explicitly present in its training data—such as catching a slipping object or rotating an object to the correct orientation. | 中 | SO009 |
| CO028 | Skild AI's long-term goal is to develop AGI rooted in the physical world, challenging the prevailing notion that AGI can arise solely from digital knowledge. | 高 | SO001, SO004 |
| CO029 | Skild AI's revenue grew from zero to approximately $30M in just a few months in 2025 and is described by the company as growing exponentially. | 中 | SO002, SO006, SO007 |
| CO030 | Skild AI is deployed in security/facility inspection, last-mile and point-to-point delivery, warehouses, manufacturing, data centers, and construction tasks. | 高 | SO002, SO006 |
| CO031 | Skild AI plans to ultimately deploy its robotics technology in consumer homes, with enterprise applications as the first use case. | 中 | SO002, SO006 |
| CO032 | Skild AI announced the acquisition of Zebra Technologies' Robotics Automation business, including the Symmetry Fulfillment orchestration platform, in April 2026. | 高 | SO003, SO026 |
| CO033 | In the Zebra acquisition, Zebra Technologies received an equity stake in Skild AI as consideration for the transaction. | 中 | SO003 |
| CO034 | The Zebra acquisition combines Skild's omni-bodied AI with Zebra's battle-tested Symmetry orchestration platform to create end-to-end warehouse automation with humanoids, robotic dogs, arms, and AMRs. | 高 | SO003, SO026 |
| CO035 | Lightspeed Venture Partners co-led Skild AI's Series A and participated again in the Series C. | 高 | SO001, SO002 |
| CO036 | NVentures (NVIDIA's venture capital arm) participated in Skild AI's Series C round. | 高 | SO002, SO007 |
| CO037 | Strategic investors in Skild's Series C included Samsung, LG, Schneider Electric, CommonSpirit Health, and Salesforce Ventures. | 高 | SO002, SO007 |
| CO038 | In-Q-Tel (IQT), the US intelligence community's venture arm, participated in Skild AI's Series C round, signaling US national security interest. | 高 | SO002, SO012 |
| CO039 | Skild AI's product description as of May 2026 emphasizes a Mobile Manipulation Platform enabling skills like grasping, handover, and navigation via an API abstraction layer. | 中 | SO019 |
| CO040 | Deepak Pathak received the Sloan Research Fellowship, was named MIT TR35 Innovator Under 35, and received multiple Best Paper awards at top AI/robotics conferences including ICRA, CVPR, RSS, and CoRL. | 高 | SO004, SO022 |
| CO041 | Abhinav Gupta received the ONR Young Investigator Award, PAMI Young Researcher Award, Sloan Fellowship, and Okawa Research Grant. | 高 | SO015, SO004 |
| CO042 | Skild AI's key-person risk is concentrated in Deepak Pathak and Abhinav Gupta; the company has not publicly disclosed a succession plan or deep leadership bench beyond the two co-founders. | 中 | SO001, SO002 |
| CO043 | SoftBank Group has led or participated in Skild AI's Series A, Series B, and Series C—committing the majority of capital across the company's growth rounds and making it Skild's most important single institutional backer. | 高 | SO001, SO010, SO007, SO002 |
| CM001 | Skild AI operates in the physical AI / embodied intelligence software market — AI that enables robots to perceive, reason, plan, and act in unstructured real-world environments. | 高 | SM021, SM022, SM005, SM006 |
| CM002 | The global robotics market spans three main IFR-defined categories: industrial robots, professional service robots, and medical robots; the global 2025 consensus estimate is $50–55B across all segments. | 中 | SM007, SM001 |
| CM003 | Warehouse / logistics is the fastest-growing application area in warehouse automation; retail and e-commerce dominate segment share and are also growing fastest per Grand View Research. | 中 | SM003, SM004 |
| CM004 | Skild's specific market boundary is AI software (foundation models and APIs) for commercial robots, not the robot hardware itself; Skild does not manufacture physical robot platforms. | 高 | SM021, SM015, SM022 |
| CM005 | Status-quo substitutes for physical AI software include: teach-and-repeat programming, custom per-application ML models, and continued human labor — all of which fail to generalize across tasks or robot types. | 中 | SM015, SM018, SM016 |
| CM006 | MarketsandMarkets estimates the industrial robots market at $16.89B in 2024, growing to $29.43B by 2029 at a CAGR of 11.7%; this scope focuses on hardware units. | 高 | SM002, SM007 |
| CM007 | Grand View Research estimates the industrial robotics market at $33.96B in 2024, growing to $60.56B by 2030 at a CAGR of 9.9%; this scope includes hardware, software, and services. | 高 | SM001, SM007 |
| CM008 | The $16.9B vs. $34B industrial robotics market size discrepancy in 2024 reflects scope differences — MarketsandMarkets counts hardware units only; Grand View Research includes software and services — not measurement error. | 中 | SM001, SM002 |
| CM009 | Grand View Research estimates the warehouse automation market at $19.23B in 2023, growing to $59.52B by 2030 at a CAGR of 18.7%; North America led with 36.7% revenue share in 2023. | 高 | SM003, SM004 |
| CM010 | Mordor Intelligence estimates the warehouse automation market at $29.98B in 2025, growing to $65.74B by 2031 at a CAGR of 13.98%; mobile robots captured 41.4% of 2025 market share. | 中 | SM004, SM003 |
| CM011 | MarketsandMarkets projects the global embodied AI market at $4.44B in 2025, growing to $23.06B by 2030 at a CAGR of 39.0%; logistics and supply chain is the fastest-growing vertical. | 高 | SM005, SM006 |
| CM012 | Grand View Research estimates the embodied AI market at $4.67B in 2025, growing to $67.63B by 2033 at a CAGR of 39.7%; hardware leads at 51.2% revenue share; logistics is fastest-growing at 42.2% CAGR. | 高 | SM006, SM005 |
| CM013 | The global robotics market (all segments: industrial + service + medical) totals approximately $50–55B in 2025 by consensus across ABI Research, GM Insights, SkyQuest, and Future Market Insights, with 2030 forecasts ranging from $111B to $300B. | 中 | SM007, SM002 |
| CM014 | Goldman Sachs projects humanoid robots could be economically viable in factory settings between 2025–2028 and in consumer applications between 2030–2035, contingent on 15–20% annual cost reduction and battery life improvements. | 中 | SM008, SM007 |
| CM015 | Morgan Stanley projects the total humanoid robot ecosystem (hardware + software + supply chain) could reach $5 trillion by 2050, with ~1 billion units deployed and per-unit prices declining from $200K (2024) to $50K (2050). | 低 | SM009, SM010 |
| CM016 | Global robotics startup funding reached $13.8B in 2025, up 77% from $7.8B in 2024, representing the largest annual funding total in robotics investment history per available data. | 中 | SM025, SM024 |
| CM017 | Skild AI's TAM is the global embodied AI / physical AI software market, estimated at $4.4–4.7B in 2025 and growing toward $23–68B by 2030–2033 depending on scope. | 中 | SM005, SM006, SM011 |
| CM018 | Skild AI's SAM is estimated at approximately $2–3B in 2025, representing the enterprise robotic AI software market for industrial and logistics robots, excluding consumer, medical, and defense-classified segments. | 低 | SM005, SM006 |
| CM019 | Skild AI's near-term SOM is estimated at $200–500M in 2025, encompassing warehouse automation, discrete manufacturing, and facility inspection verticals where Skild has active partnerships or pilot deployments. | 低 | SM014, SM013 |
| CM020 | In the warehouse and logistics segment, primary buyers are 3PL operators or large retailers; Skild's April 2026 acquisition of Zebra's Robotics Automation business provides direct enterprise customer access to this segment. | 高 | SM014, SM012, SM015 |
| CM021 | Zebra Technologies' partnership (and ultimately acquisition by Skild) signals that hardware incumbents will license AI foundation models rather than build proprietary robot brains, validating an OS-style software licensing model for robotics AI. | 中 | SM014, SM015 |
| CM022 | In the warehouse segment, the payer is typically the VP Operations or CFO, the user is a robot fleet manager, and the budget ranges from $500K–$5M annually for AI software, often bundled in a larger automation capex program. | 中 | SM014, SM013 |
| CM023 | In discrete manufacturing, buyers are industrial OEMs or Tier 1 suppliers; adoption triggers are quality consistency and labor availability; sales cycles are typically 12–36 months due to safety certification requirements. | 中 | SM002, SM016 |
| CM024 | Humanoid OEM partners represent a nascent but strategically important buyer segment; Skild's omni-bodied model is specifically architected to serve humanoid platforms via SDK or cloud API licensing. | 中 | SM011, SM008, SM022 |
| CM025 | Defense and security is an emerging segment for physical AI; In-Q-Tel's participation in Skild's Series C signals active US intelligence and defense community interest in the company's autonomous capabilities. | 中 | SM013, SM015 |
| CM026 | Facility inspection at hyperscalers and data centers is a fast-growing segment for Skild; 24/7 robot patrols and equipment monitoring replace expensive night-shift human security and inspection operations. | 中 | SM013, SM021 |
| CM027 | The US Chamber of Commerce estimates 1.7M+ unfilled manufacturing jobs in the US, providing a structural demand driver for robotic automation; this figure is cited in Skild's Series A press release. | 高 | SM012, SM016, SM015 |
| CM028 | The National Association of Manufacturers (NAM) projects 2.1M unfilled manufacturing jobs by 2030, reinforcing labor scarcity as a durable, structural tailwind for industrial automation demand. | 高 | SM012, SM018 |
| CM029 | AI foundation model advances — specifically large multimodal models enabling generalist robot behavior — represent a technology breakthrough that removes the key historical barrier to scalable robot deployment; Sequoia Capital explicitly calls this the 'GPT-3 moment for robotics.' | 高 | SM015, SM006, SM011 |
| CM030 | The software segment of warehouse automation is projected to grow at 14.87% CAGR through 2031 — faster than the overall market (13.98%) — indicating the intelligence and orchestration layer is capturing an increasing share of market value. | 中 | SM004, SM003 |
| CM031 | Piece-picking robots are forecast to grow at 15.27% CAGR to 2031 — the fastest sub-segment of warehouse automation — validating high market demand for Skild's dexterous manipulation capabilities. | 中 | SM004, SM007 |
| CM032 | E-commerce growth has established an automation standard (exemplified by Amazon) that forces 3PL operators and retailers to invest in robotic fulfillment; this creates recurring enterprise demand for AI-powered picking and sorting. | 中 | SM003, SM004 |
| CM033 | Skild's data flywheel — real-world training data generated with zero human annotation as robots operate — is described as a structural competitive advantage that compounds with each new enterprise deployment. | 中 | SM013, SM021, SM015 |
| CM034 | Real-world robot training data scarcity is identified by MarketsandMarkets as the key market challenge in embodied AI; this directly validates Skild's data flywheel strategy as addressing a market-wide bottleneck. | 中 | SM005, SM012 |
| CM035 | Industrial robot system costs range from $15K–$75K for hardware alone, with total integration projects costing significantly more; this capital intensity creates adoption barriers for SMEs and constrains Skild's near-term market. | 中 | SM002, SM016 |
| CM036 | Deployment of robot AI systems requires coordination among robotics engineers, production engineers, and floor operators; qualified system integrators are scarce and represent a deployment bottleneck. | 中 | SM002, SM007 |
| CM037 | Cobots and robot AI systems must be frequently reprogrammed as product lines and consumer demands change; task-specific robot programming faces diminishing ROI in high-mix manufacturing, creating a pull for general-purpose AI. | 中 | SM002, SM018 |
| CM038 | Safety certification and regulatory compliance requirements (ISO 10218, ISO/TS 15066 for cobots) create deployment timelines of 24–48 months in healthcare and defense, limiting near-term revenue from regulated segments. | 中 | SM007, SM016 |
| CM039 | Industrial robotics market estimates for 2024 range from $16.9B (MarketsandMarkets, hardware-only) to $34.0B (Grand View Research, hardware+software+services) — a 2x range reflecting scope definitions, not measurement quality differences. | 高 | SM001, SM002, SM007 |
| CM040 | Humanoid robot market forecasts differ by more than two orders of magnitude in absolute terms — Goldman Sachs projects factory viability by 2025–2028 while Morgan Stanley sizes the 2050 ecosystem at $5T — indicating fundamental analyst disagreement on adoption pace and ecosystem definition. | 中 | SM008, SM009, SM010 |
| CM041 | Open-source robotics foundation models (Google RT-2, OpenVLA from UC Berkeley, Physical Intelligence π0) are publicly available and represent a commoditization risk for Skild's software layer if performance gaps are closed. | 中 | SM011, SM015 |
| CM042 | Well-resourced incumbents — NVIDIA (Isaac/GR00T platform), Google DeepMind, Amazon (warehouse robot data), ABB, and KUKA — represent competitive threats to Skild's AI model positioning; none have Skild's cross-robot generalization architecture but all have more resources. | 中 | SM011, SM024 |
| CM043 | Global robotics venture capital investment reached $13.8B in 2025, up from $7.8B in 2024, signaling strong investor conviction in near-term commercial robot adoption across warehousing, manufacturing, and humanoid segments. | 中 | SM025, SM024 |
| CP001 | Skild AI competes across five tiers: direct foundation model peers (Physical Intelligence), platform threats (NVIDIA, Google DeepMind), vertically integrated humanoid makers (Figure AI, Amazon/Agility, Tesla), legacy incumbents (ABB, KUKA, Fanuc), and open-source or internal-build substitutes. | 高 | SP003, SP004, SP007, SP018, SP023 |
| CP002 | Physical Intelligence (π.ai) raised $1.07B total by end 2025 ($400M Series A in November 2024 led by Bond/Thrive at $2.4B valuation; $600M Series B led by CapitalG/Alphabet in November 2025 at $5.6B valuation) — Skild's closest direct foundation model peer. | 高 | SP001, SP012, SP002 |
| CP003 | NVIDIA's Isaac GR00T N1 open-source humanoid robot foundation model, released March 2025, represents Skild's most dangerous platform threat due to NVIDIA's dominant GPU infrastructure position, free model strategy, and OEM partner base including 1X Technologies, Agility Robotics, and Boston Dynamics. | 高 | SP003, SP013 |
| CP004 | Figure AI raised over $2B total by late 2025 ($675M Series B in March 2024 at $2.6B valuation; $1B+ Series C in September 2025 at $39B valuation), representing the highest capitalization among humanoid robot companies and the largest valuation in the vertically integrated tier. | 高 | SP005, SP022 |
| CP005 | Covariant's 2024 Amazon acqui-hire — in which Amazon hired Covariant's founding team and received a non-exclusive IP license to Covariant's RFM-1 foundation model — fundamentally altered Covariant's competitive position, talent density, and organizational independence without constituting a full acquisition. | 高 | SP007, SP008 |
| CP006 | Legacy robotics incumbents ABB, KUKA, Fanuc, and Yaskawa collectively hold over 70% of the global industrial robot installed base; ABB Robotics achieved $2.3B in revenue in 2024 with more than 80% of offerings incorporating AI or software-enabled capabilities, establishing a distribution moat that pure-play AI startups cannot easily replicate. | 高 | SP018, SP019 |
| CP007 | Open-source alternatives to Skild's proprietary foundation model include OpenVLA (Berkeley), Octo (Berkeley + Stanford), HuggingFace LeRobot, Physical Intelligence's openpi, and NVIDIA GR00T N1 — all available without licensing fees, collectively lowering the adoption barrier for robot AI but also constraining pricing power for commercial model vendors. | 中 | SP002, SP003, SP012 |
| CP008 | Competitive intensity in the robot AI software market is rising rapidly; $13.8B in global robotics VC funding in 2025 represents a 77% increase from 2024, with significant allocations to foundation model startups and humanoid hardware companies that are building competing AI platforms. | 中 | SP001, SP005, SP011 |
| CP009 | Internal-build substitutes represent a significant competitive risk: Amazon is deploying Covariant IP in its own fulfillment centers, Tesla Optimus is developed in-house, and Figure AI's Helix platform is proprietary — large enterprises may choose to develop or co-develop robot AI capabilities rather than license a third-party platform like Skild. | 中 | SP007, SP005, SP022 |
| CP010 | Physical Intelligence's pi-zero (π₀) architecture uses a 3B-parameter PaliGemma VLM backbone combined with a 300M-parameter action expert using flow matching; it was trained on 10,000+ demonstration hours across 7–8 robot platforms and 68+ tasks, and outperforms OpenVLA and Octo on standard manipulation benchmarks. | 高 | SP002, SP012 |
| CP011 | NVIDIA GR00T N1 uses a dual-system architecture: System 2 (a vision-language model for slow, deliberate reasoning and planning) and System 1 (a diffusion transformer for fast, real-time motor action generation); released as open-source in March 2025 under a permissive license, with subsequent versions N1.5 and N1.6 released in 2025–2026. | 高 | SP003, SP013 |
| CP012 | Google DeepMind launched Gemini Robotics in March 2025 — comprising a generalist VLA model and Gemini Robotics-ER (Embodied Reasoning) with deep 3D spatial understanding — with hardware partners including Apptronik Apollo, Boston Dynamics Atlas, and Agility Robotics Digit; an on-device variant was released mid-2025. | 高 | SP004, SP011 |
| CP013 | Figure AI's 'Helix' AI platform is proprietary and hardware-specific to the Figure 02 form factor; commercial deliveries began December 2024 with BMW manufacturing facilities, making Figure the first among the vertically integrated humanoid makers to achieve revenue-generating commercial deployment. | 高 | SP005, SP022 |
| CP014 | Covariant's RFM-1 (Robotics Foundation Model) is trained on the world's largest warehouse manipulation dataset and enables robots to adapt to new SKUs in minutes; the 2024 Amazon deal granted Amazon a non-exclusive license to RFM-1, potentially allowing Amazon to deploy Covariant's AI capabilities without paying ongoing licensing fees. | 高 | SP007, SP017 |
| CP015 | Apptronik raised $935M in total funding by February 2026 (Series A) at an estimated $5–5.5B valuation; investors include Google, Mercedes-Benz, B Capital, John Deere, and Qatar Investment Authority; the Apollo humanoid is integrated with Google DeepMind's Gemini Robotics-ER model, creating a hardware + AI software bundle that competes directly with Skild's OEM channel. | 高 | SP011, SP004 |
| CP016 | 1X Technologies raised $100M in its January 2024 Series B led by EQT Ventures with participation from Samsung NEXT, OpenAI Startup Fund, and Tiger Global; its NEO bipedal humanoid was available for pre-order at $20K in October 2025, but NEO tasks remained partially teleoperated at launch. | 中 | SP006 |
| CP017 | Unitree Robotics shipped 5,500+ humanoid robot units in 2025 at a $13.5K–$21.5K price point, achieving $235M in revenue (335% YoY growth); Unitree is targeting an IPO at a $7B valuation and a volume of 20,000 units in 2026, establishing Chinese manufacturers as the volume leaders in the global humanoid market. | 中 | SP015, SP016 |
| CP018 | AgiBot reached a $2.1B valuation by 2025 with 5,100+ units shipped and investors including Tencent, BYD, CATL, LG, and JD.com; its GO-1 robot uses the ViLLA embodied foundation model framework, positioning AgiBot as a direct competitor to Skild's physical AI platform in the Chinese market. | 中 | SP016 |
| CP019 | Sanctuary AI raised approximately US$100M (CA$140M) in total funding by 2025; its Phoenix Gen-7 humanoid is designed for general-purpose labor and is backed by Accenture, Magna International, and Workday; Sanctuary is a relatively smaller player but represents the Canadian ecosystem's entry into physical AI. | 中 | SP014 |
| CP020 | Toyota Research Institute's Large Behavior Model (LBM) represents one of the world's largest robot manipulation datasets and enables natural language-based robot control; TRI has partnered with Boston Dynamics for Atlas AI development, making it a research-tier threat to Skild's foundation model positioning. | 中 | SP021 |
| CP021 | OpenAI established a dedicated robotics division in 2025 under Caitlin Kalinowski, with investments in Physical Intelligence and Figure AI (both now partially competitive with OpenAI's own efforts) and plans to adapt GPT-4/5 models for robot instruction and control. | 高 | SP020, SP026 |
| CP022 | Figure AI broke its AI platform partnership with OpenAI in 2025; OpenAI subsequently launched its own robotics division, converting a previously complementary relationship into direct competition at the AI platform layer. | 高 | SP026, SP020 |
| CP023 | Skild claims its training dataset is '1,000 times larger than most competitors'; this claim is sourced from Sequoia Capital's investment announcement and is not independently benchmarked — Physical Intelligence's pi-zero is trained on 10,000+ hours across 8 robot platforms and 68+ tasks, representing substantial but unquantified coverage relative to Skild's undisclosed corpus. | 中 | SP023, SP002, SP012 |
| CP024 | NVIDIA's strategy of releasing GR00T N1 as open-source under a permissive license creates hardware lock-in by ensuring robot OEMs adopt NVIDIA GPUs (H100, GH200, Thor) for inference and training, enabling NVIDIA to capture the economic value of the AI layer indirectly through hardware and cloud compute sales rather than model licensing. | 高 | SP003, SP013 |
| CP025 | Cross-embodiment generality — the ability to run on any robot form factor without per-robot retraining — is Skild's primary claimed differentiation from competitors; no independent public benchmark has validated that Skild outperforms Physical Intelligence pi-zero, NVIDIA GR00T N1, or Google DeepMind Gemini Robotics on this dimension. | 中 | SP023, SP002, SP012 |
| CP026 | Skild's enterprise pricing is not publicly disclosed; author-estimated enterprise software license + professional services contracts are likely in the $200K–$5M+ annual range for large fleet deployments, based on analogous AI platform deals in adjacent markets. | 低 | SP023, SP024 |
| CP027 | NVIDIA GR00T model weights are available free under a permissive open-source license; NVIDIA's monetization is via GPU hardware (H100 cluster: $100K+ capital cost), NVIDIA AI Cloud compute (estimated $5K–$50K/year per use case), and NVIDIA Omniverse simulation licenses. | 中 | SP003, SP013 |
| CP028 | Physical Intelligence has open-sourced its pi-zero model weights and code via the 'openpi' repository on GitHub; this lowers adoption friction for research and OEM integration but creates a tension with PI's commercial model — if the architecture is freely available, PI must differentiate on training data, deployment support, and enterprise services. | 高 | SP002, SP012 |
| CP029 | ABB's OmniCore controller platform, with AI-ready software capabilities and 300,000+ robot installed base, represents a distribution advantage that pure-play AI startups cannot easily replicate; ABB's enterprise trust, compliance certifications, and existing customer relationships create switching cost that operates in the incumbent's favor. | 高 | SP018, SP019 |
| CP030 | Skild's primary claimed competitive moat is a data flywheel in which real-world robot deployments generate proprietary training data with claimed zero human annotation, compounding the dataset advantage over time; the Zebra acquisition amplifies this by adding an AMR fleet that generates logistics and picking data at enterprise scale. | 中 | SP023, SP025 |
| CP031 | Enterprise switching cost for Skild is estimated at 6–18 months of re-integration engineering effort per customer, created by deep API dependencies, task-specific fine-tuning of Skild's model for the customer's robot fleet, and Zebra WMS integrations — this switching cost represents a structural lock-in mechanism but only materializes after initial deployment. | 中 | SP023, SP025 |
| CP032 | Agility Robotics (Amazon-majority-owned) reached a milestone of 100,000 tote moves with GXO Logistics in 2025 using its Digit robot; Amazon's vertical integration of Agility hardware and Covariant IP creates a closed-ecosystem competitive moat that may permanently exclude third-party AI platform vendors like Skild from Amazon logistics. | 高 | SP010, SP007 |
| CP033 | Tesla's Optimus program experienced production delays in 2025 — missing its earlier stated trajectory toward 100,000+ units per month by 2030 — and multiple leadership turnovers; Optimus represents a large-scale internal-build risk to Skild's humanoid OEM channel if it succeeds, but its execution difficulties reduce near-term probability of displacing Skild's OEM partnerships. | 中 | SP016, SP001 |
| CP034 | The open-sourcing of robot foundation models by NVIDIA (GR00T N1 permissive license) and Physical Intelligence (openpi weights on Hugging Face) creates structural commoditization pressure on the AI model layer; Skild's moat must increasingly rely on proprietary data scale, deployment depth, and vertical integration rather than the model architecture itself. | 高 | SP002, SP003, SP012, SP013 |
| CP035 | Skild claims its training dataset is 1,000x larger than 'most competitors'; Physical Intelligence disputes this implicitly by citing broad cross-embodiment training coverage (10,000+ hours across 8 robot platforms) and claims its π₀ architecture achieves superior cross-embodiment generalization — neither company has disclosed absolute dataset sizes or agreed on a common benchmark. | 中 | SP023, SP002 |
| CP036 | NVIDIA frames GR00T as a pro-ecosystem, non-threatening offering that helps all robot AI companies by providing a common infrastructure layer; Skild's investors and industry analysts argue the free model strategy is designed to commoditize the AI software layer and capture value through hardware and cloud compute sales, ultimately threatening independent robot AI platform companies. | 中 | SP003, SP013 |
| CP037 | Chinese humanoid manufacturers (Unitree G1 at $13.5K–$21.5K; AgiBot GO-1) are pricing robot hardware at dramatically lower levels than Western competitors — estimated 40–70% below comparable Western humanoid robots — threatening Skild's OEM partner economics and potentially compressing hardware margins, which would reduce OEM appetite for third-party AI software licensing. | 中 | SP015, SP016 |
| CP038 | Covariant's 2024 organizational disruption — in which Amazon acquired the founding team via an acqui-hire while leaving Covariant as an independent entity — resulted in diminished talent density, leadership rebuilding, and reduced organizational confidence, providing a cautionary precedent for Skild's hiring pipeline and key-employee retention risk. | 高 | SP007, SP008 |
| CP039 | Google DeepMind's long history of failed robotics commercialization — including the 2016 Google Robotics shutdown, the Replicant project, and Intrinsic's multi-year period without clear commercial traction before being folded back into Google in February 2026 — raises legitimate questions about whether frontier AI labs can consistently productize robot AI at enterprise scale. | 高 | SP004, SP009 |
| CP040 | Skild's acquisition of Zebra Technologies' robotics automation business (Fetch Robotics AMR fleet) creates integration risk: Zebra's AMR fleet uses architectures and control systems developed independently of Skild's AI foundation model, requiring either platform migration (estimated 12–36 months per deployment) or dual-track maintenance during transition. | 中 | SP025, SP023 |
| CP041 | The robot AI software market is likely to show winner-take-most dynamics: cross-embodiment data advantages compound over time, suggesting the first company to reach a critical proprietary dataset scale threshold may sustain an insurmountable lead — Skild currently claims this position but it is not independently verified, and NVIDIA's synthetic data generation at scale via Omniverse/Cosmos represents an alternative path to dataset sufficiency. | 中 | SP023, SP003 |
| CP042 | OpenAI's 2025 entry into robotics — following its prior investments in Physical Intelligence and Figure AI — creates a potential long-term existential risk for Skild: if OpenAI successfully converts its dominant LLM distribution into a robot intelligence platform, it would compete directly with Skild's model positioning and could leverage existing enterprise LLM relationships to displace Skild in enterprise accounts. | 中 | SP020, SP026 |
| CP043 | Intrinsic (formerly Alphabet's robot software subsidiary) was officially folded back into Google in February 2026; its IVM (Intrinsic Vision Model), Flowstate platform, and Foxconn factory deployments are now part of Google's enterprise robotics portfolio, significantly strengthening Google DeepMind's industrial robotics go-to-market relative to a standalone startup. | 高 | SP009, SP004 |
| CI001 | Skild AI reported approximately $30 million in revenue for 2025, growing from zero in 'just a few months.' | 中 | SI001, SI003, SI009, SI019 |
| CI002 | Skild AI's 2025 revenue grew from zero during the course of 2025, suggesting the $30M is not a full-year figure but an annualized run rate reached in 2025. | 中 | SI001, SI003 |
| CI003 | Skild AI described its revenue growth as 'exponential' as of the January 2026 Series C press release. | 高 | SI001, SI003 |
| CI004 | Skild AI's $30M revenue figure is company-stated and has not been independently audited or verified by a third party. | 高 | SI001, SI002, SI003 |
| CI005 | Skild AI raised a $300M Series A on July 9, 2024 at a $1.5B post-money valuation, led by Lightspeed Venture Partners, Coatue, and SoftBank. | 高 | SI005, SI024 |
| CI006 | Skild AI raised approximately $135M in a Series B in June 2025 at a $4.5B post-money valuation, led by SoftBank with Nvidia ($25M) and Samsung ($10M). | 高 | SI006, SI007, SI008 |
| CI007 | Skild AI raised approximately $1.4B in a Series C on January 14, 2026 at a valuation exceeding $14B, led by SoftBank. | 高 | SI001, SI002, SI003, SI004 |
| CI008 | CEO Deepak Pathak stated in January 2026 that Skild AI had raised more than $2 billion in total across all rounds. | 高 | SI002, SI003 |
| CI009 | Crunchbase tracked $1.83 billion in total funding raised by Skild AI across four rounds as of the Series C close. | 高 | SI003, SI002 |
| CI010 | The stated use of Series C proceeds is to continue scaling model training and growing future deployment of technology. | 高 | SI001, SI009 |
| CI011 | Skild AI operates a B2B enterprise software platform model, licensing the Skild Brain foundation model to robot OEMs, system integrators, and enterprise operators. | 高 | SI013, SI020, SI026 |
| CI012 | Skild AI's revenue streams include foundation model licensing, vertical-specific software modules (security, warehouse, manufacturing), and cloud infrastructure services (inference and AI-factory offerings). | 中 | SI013, SI020 |
| CI013 | Skild AI's official website describes a 'Mobile Manipulation Platform' where skills (grasping, handover, navigation) are abstracted via API calls, indicating a programmable interface pricing model. | 高 | SI020, SI013 |
| CI014 | Skild AI is actively deployed in six enterprise verticals: security and facility inspection, last-mile and point-to-point delivery, warehouses, manufacturing, data centers, and construction. | 高 | SI001, SI009 |
| CI015 | SoftBank led Skild AI's Series B with approximately $100M and also led the Series C, making it the largest single investor by committed capital. | 高 | SI006, SI001, SI003 |
| CI016 | Nvidia contributed $25M to Skild AI's Series B and its venture arm (NVentures) participated in the Series C. | 高 | SI006, SI007, SI001 |
| CI017 | Samsung contributed $10M to Skild AI's Series B and also participated in the Series C as a strategic investor. | 高 | SI006, SI008, SI001 |
| CI018 | Skild AI acquired Zebra Technologies' Robotics Automation business in April 2026 in a transaction structured as cash plus Skild equity issued to Zebra Technologies. | 中 | SI009, SI021, SI029, SI030 |
| CI019 | The Zebra acquisition included the Symmetry Fulfillment orchestration platform, formerly part of Fetch Robotics, which coordinates robot fleets with frontline workers in warehouse environments. | 中 | SI009, SI023, SI029 |
| CI020 | Training and inference for large robotics foundation models is compute-intensive, with frontier model training runs costing $30M–$200M+ per run as of 2024–2025. | 高 | SI015, SI016, SI017 |
| CI021 | AI foundation model training costs for frontier-scale models have been: GPT-4 at ~$79M, Gemini Ultra ~$192M, Llama 3.1-405B ~$170M, Grok-2 ~$107M—indicating the cost scale Skild AI may face. | 高 | SI016, SI015 |
| CI022 | Training compute costs for large-scale AI models are doubling approximately every 8 months, implying rapidly escalating capex for Skild's model training roadmap. | 高 | SI015, SI018 |
| CI023 | Software-first robotics AI platforms are expected to achieve gross margins of 60–80% at scale, consistent with top-tier enterprise SaaS; near-term margins at $30M revenue are likely lower due to compute and deployment costs. | 低 | SI013, SI026, SI017 |
| CI024 | Crunchbase confirmed the Series B amount as approximately $135M, corroborating Bloomberg reporting; Skild AI did not officially confirm a specific total for the Series B. | 中 | SI003, SI006 |
| CI025 | CEO Deepak Pathak told Bloomberg at the time of the Series C that Skild AI had raised more than $2B to date, exceeding the Crunchbase-tracked figure of $1.83B (possibly reflecting undisclosed seed and unreported tranches). | 高 | SI002, SI003 |
| CI026 | The Zebra Technologies acquisition was structured with Zebra receiving both cash and an equity stake in Skild AI as consideration, per multiple third-party sources. | 中 | SI009, SI021, SI022, SI023, SI029, SI030 |
| CI027 | In-Q-Tel (IQT), the US intelligence community's venture arm, participated in Skild AI's Series C, signaling potential US government/defense customer revenue opportunity. | 中 | SI001 |
| CI028 | Skild AI's stated long-term plan is to deploy robotics in consumer homes after establishing enterprise as the first application, representing a multi-year TAM expansion pathway. | 高 | SI001, SI003 |
| CI029 | Skild AI's gross margin is not publicly disclosed; estimates of 60–80% at scale are analyst-derived and not audited or confirmed by the company. | 低 | SI013, SI026 |
| CI030 | Skild AI's monthly burn rate is not publicly disclosed; analyst estimates of $10–50M/month are inferred from headcount, compute scale, office footprint, and comparable AI startup profiles. | 低 | SI014, SI017, SI018 |
| CI031 | Post-Series-C runway is estimated at 28–47 months from January 2026 (assuming $1.2–1.4B cash and $30–50M/month gross burn), extending to mid-2028 to late-2029 without additional fundraising. | 低 | SI001, SI003, SI014, SI017 |
| CI032 | Foundation model training and AI research infrastructure represent the primary capital expenditure for Skild AI, consistent with the stated use of Series C proceeds ('scale model training'). | 中 | SI001, SI015, SI016 |
| CI033 | Strategic investors (NVIDIA, Samsung, LG, Zebra, Schneider Electric, IQT) provide non-financial value including hardware access, distribution channels, and market signal, beyond pure capital contribution. | 中 | SI001, SI006, SI009 |
| CI034 | Skild AI's revenue quality is uncertain: the $30M figure is company-stated, unaudited, may include one-time pilot payments, and may be concentrated in a small number of large enterprise contracts. | 中 | SI001, SI004 |
| CI035 | Skild AI has not publicly disclosed unit economics including CAC, LTV, churn rate, net revenue retention, or payback period for any enterprise segment. | 高 | SI001, SI013, SI019 |
| CI036 | At $30M in revenue and an estimated small number of large enterprise contracts, Skild AI's implied ACV is likely in the range of $500K–$5M per enterprise deployment, though no per-customer data is publicly available. | 低 | SI001, SI013 |
| CI037 | At a $14B valuation and $30M in 2025 revenue, Skild AI's revenue multiple is approximately 467x trailing revenue — a premium only justifiable by very high near-term growth, winner-take-most platform dynamics, or strategic option value. | 中 | SI001, SI003, SI007 |
| CI038 | The Zebra acquisition adds potential revenue from the Symmetry Fulfillment platform's existing enterprise customer base, but no revenue, ARR, or margin figures for the Zebra robotics division have been disclosed by either party. | 低 | SI009, SI023, SI028 |
| CI039 | Skild AI is deploying at enterprise scale across multiple robotics verticals simultaneously, which drives both revenue and the data flywheel that continuously improves model quality. | 中 | SI001, SI009 |
| CI040 | SoftBank is estimated to have committed over $1 billion to Skild AI across Series A, B (leading $100M), and Series C (lead investor), representing significant LP concentration risk. | 中 | SI005, SI006, SI001 |
| CE001 | Skild Brain is described as the industry's first unified robotics foundation model capable of controlling any robot—quadrupeds, humanoids, tabletop arms, and mobile manipulators—without prior knowledge of the robot's exact body form. | 高 | SE001, SE002, SE003, SE005 |
| CE002 | The Mobile Manipulation Platform abstracts robot skills (grasping, handover, navigation) behind an API call, allowing application developers to build robot applications without managing low-level motor control. | 高 | SE001, SE002 |
| CE003 | In April 2026, Skild AI acquired Zebra Technologies' Robotics Automation business, including the Symmetry Fulfillment platform, which orchestrates heterogeneous robot fleets alongside human frontline workers in logistics environments. | 高 | SE014, SE015, SE024, SE027 |
| CE004 | Skild AI's commercial deployment sectors include security and facility inspection, last-mile delivery, warehouse fulfillment, factory assembly, data center operations, and construction site monitoring. | 高 | SE002, SE019, SE021 |
| CE005 | Skild AI's live revenue grew from zero to approximately $30M in just a few months in 2025, with multiple customers across deployment sectors. | 中 | SE002, SE019 |
| CE006 | Skild AI launched from stealth in July 2024 with a $300M Series A, simultaneously commercially releasing the Skild Brain and Mobile Manipulation Platform. | 高 | SE006, SE008, SE005 |
| CE007 | Skild claims its in-context learning capability represents a first-ever research breakthrough in robotics, earning Best Paper Nominations at top robotics conferences. | 中 | SE007, SE005 |
| CE008 | Skild AI's stated roadmap is to build a single action-centric brain for all robot embodiments, all tasks, and all scenarios; Series C capital is allocated to scaling all four data sources and expanding commercial deployments. | 高 | SE002, SE003 |
| CE009 | The Skild Brain uses a hierarchical two-tier transformer-based architecture with a high-level semantic planner (low-frequency) and a low-level motor controller (high-frequency) that outputs per-joint torques and angles. | 高 | SE018, SE003, SE005 |
| CE010 | Skild AI uses NVIDIA Isaac Lab for large-scale physics-based reinforcement learning simulation, running thousands of parallel robot instances across multiple embodiments and thousands of simulated environments. | 高 | SE003, SE004 |
| CE011 | Skild AI uses NVIDIA Cosmos Transfer to augment training datasets with environmental variations—lighting, texture, weather—to expand training data robustness beyond physics simulation alone. | 高 | SE003, SE004 |
| CE012 | The simulation training pipeline can generate a millennium of robot experience within days by running massive parallelized simulations across GPU clusters, making large-scale robotic training feasible at unprecedented speed. | 中 | SE003 |
| CE013 | Internet-scale human video (billions of clips) is used as a pretraining data source, with the model learning object affordances by treating humans as biological robots. | 高 | SE002, SE003 |
| CE014 | Teleoperation data—images and proprioception mapped to joint torques—is collected through scalable interfaces and is described as the richest form of post-training data signal. | 高 | SE002, SE004 |
| CE015 | Skild AI claims its model is trained on approximately 1,000 times more action-centric data than competing robotics models. | 低 | SE002, SE005 |
| CE016 | HPE Cray XD670 servers (equipped with NVIDIA HGX H200) are used for large-scale model training; HPE ProLiant DL380a Gen12 servers (NVIDIA L40S) are used for visualization and inference. | 高 | SE003, SE004 |
| CE017 | Commercial robot deployments from customer fleets continuously generate real-world data that feeds the post-training pipeline, creating a self-reinforcing data flywheel that compounds competitive advantage over time. | 高 | SE002, SE017 |
| CE018 | NVIDIA's case study reports that the Skild Brain recovered from jammed wheels within 2–3 seconds, recovered from broken legs after several attempts, and generalized zero-shot to walking on stilts with leg-to-body ratios beyond training parameters. | 中 | SE003 |
| CE019 | As of May 2026, the Skild AI GitHub organization (github.com/skild-ai) has no public repositories; no public SDK or open API documentation has been released; access is enterprise-gated only. | 高 | SE013, SE001 |
| CE020 | In Pittsburgh urban testing, Skild AI's humanoid robots achieved 60–80% task performance within hours of data collection in never-before-seen environments including city parks, streets, fire escapes, and obstacle courses. | 中 | SE003 |
| CE021 | The HPE partnership announcement (March 2025) identifies construction, manufacturing, and security robots as the initial deployment targets for the Skild Brain platform. | 高 | SE004, SE019 |
| CE022 | LG CNS signed a strategic partnership with Skild AI in June 2025 to jointly develop industrial AI humanoid robots targeting smart factory, smart logistics, and urban services; LG Technology Ventures invested in Skild. | 高 | SE011, SE012, SE025 |
| CE023 | The LG CNS–Skild AI partnership combines Skild's Robot Foundation Model with LG CNS's logistics and city business solutions, targeting elder care, facility patrol, factory automation, and urban service applications. | 高 | SE011, SE012 |
| CE024 | The Symmetry Fulfillment platform acquired from Zebra Technologies provides enterprise-grade fleet orchestration with existing production-grade industrial deployments coordinating heterogeneous robots with human workers. | 高 | SE014, SE015, SE027 |
| CE025 | Skild AI intends to use the Symmetry platform to offer end-to-end warehouse automation integrating humanoids, quadrupeds, robotic arms, and AMRs under a single orchestration and AI intelligence layer. | 中 | SE015, SE024 |
| CE026 | Deepak Pathak stated (HPE press release, March 2025) that the Skild Brain performs AI modeling and inferencing simultaneously in real time, dynamically collecting data unlike static train-then-infer AI paradigms. | 中 | SE004 |
| CE027 | Deepak Pathak's ICML 2017 curiosity-driven exploration paper (arXiv 1705.05363) has over 6,000 citations and is a direct research lineage of the Skild Brain's autonomous generalization capability. | 高 | SE009, SE005 |
| CE028 | The RMA: Rapid Motor Adaptation paper (Kumar, Pathak et al., RSS 2021) demonstrated real-time adaptation of legged robot motor control to novel terrain, payloads, and damage using privileged training plus an online history encoder—the direct technical precursor to Skild Brain in-context learning. | 高 | SE010, SE005 |
| CE029 | Pathak and Gupta's CMU group won the Best Robotic System Award at the Conference on Robot Learning (CoRL) for large-scale adaptive sim-to-real transfer work in 2021–2022. | 中 | SE005 |
| CE030 | Abhinav Gupta's Supersizing Self-Supervision work (50,000+ robot grasping tries, 700 robot hours) established the data-at-scale paradigm for robotics that directly underpins Skild AI's training data strategy. | 高 | SE026, SE005 |
| CE031 | Skild AI claims emergent capabilities arising from training at scale, including robots spontaneously catching slipping objects mid-grasp and rotating objects to correct orientation without explicit programming. | 低 | SE003, SE018 |
| CE032 | The data flywheel creates a compounding competitive moat: competitors without a live commercial deployment fleet cannot replicate the proprietary real-world post-training data that improves the Skild Brain with each deployment. | 中 | SE002, SE017 |
| CE033 | Skild Brain's omni-bodied design targets any machine that can move; no specific robot hardware OEM exclusivity agreements have been publicly disclosed, and the platform is positioned as hardware-agnostic. | 中 | SE002, SE005 |
| CE034 | No peer-reviewed benchmarks comparing the commercial Skild Brain to competing robotics foundation models have been publicly published; performance claims are company-asserted or from NVIDIA partner case study only. | 高 | SE007, SE016 |
| CE035 | No third-party physical safety certifications (ISO 10218, ISO/TS 15066, or equivalent) for the Skild Brain AI system or robot deployments have been publicly disclosed as of May 2026. | 高 | SE001, SE002 |
| CE036 | Skild AI transitioned from public cloud to a private AI-as-a-service infrastructure (HPE + STN), which the company positions as providing security and data privacy for customer deployment training data. | 中 | SE004 |
| CE037 | The copyright and licensing status of internet video training data used in Skild Brain pretraining has not been publicly disclosed, representing unquantified legal exposure analogous to LLM training data litigation. | 中 | SE023, SE013 |
| CE038 | IQT (In-Q-Tel), the US intelligence community's venture arm, is a Skild AI Series C investor, which may subject the Skild Brain platform to export control review and dual-use technology classification requirements. | 中 | SE007, SE016 |
| CE039 | No recalls, safety incidents, product liability claims, or regulatory enforcement actions related to Skild AI's commercial robot deployments have been publicly reported as of May 2026. | 中 | SE001, SE022 |
| CE040 | The Symmetry Fulfillment platform comes with existing enterprise production deployments and compliance obligations from its Zebra Technologies heritage; the depth of technical integration with the Skild Brain remains unverified post-acquisition. | 中 | SE014, SE015 |
| CU001 | Skild AI grew from zero to approximately $30M in annual revenue in just a few months in 2025, confirmed by the company's official Series C blog post and a BusinessWire press release. The company describes revenue as "growing exponentially." This marks the company's transition from research/pre-revenue stage to active commercial deployment. | 高 | SU001, SU003, SU009 |
| CU002 | Skild AI deploys the Skild Brain across six publicly confirmed verticals: security and facility inspection, last-mile and point-to-point delivery, warehouses, manufacturing, data centers, and construction. These are the company's stated deployment sectors as of the January 2026 Series C announcement. | 高 | SU001, SU003, SU020 |
| CU003 | Amazon participated in Skild AI's Series A funding round (2024) through its Amazon Industrial Innovation Fund, and Jeff Bezos invested personally through Bezos Expeditions in the Series C (2026). Amazon is therefore both a financial investor and a potential strategic customer, though no Amazon operational deployment of the Skild Brain has been publicly confirmed. | 高 | SU003, SU011 |
| CU004 | Samsung participated as a strategic investor in the Series B (2025) and Series C (2026) rounds. LG invested through LG Technology Ventures and via LG CNS, which signed a commercial partnership with Skild AI in June 2025 to deploy Robot Foundation Models for smart factory, smart logistics, and urban service applications including elder care. LG CNS is publicly described as a deployment customer, not only an investor. | 高 | SU001, SU003 |
| CU005 | Schneider Electric participated as a strategic investor in the Series C round. CommonSpirit Health, one of the largest nonprofit Catholic health systems in the United States with over 140 hospitals, is also a Series C strategic investor. Salesforce Ventures joined as a Series C investor. These entities are strategic investors with potential future deployment interest, but no confirmed operational deployments have been publicly announced. | 高 | SU001, SU003 |
| CU006 | In-Q-Tel (IQT), the non-profit strategic investor that serves as the technology bridge for U.S. intelligence and defense agencies including the CIA, DoD, and NSA, participated in the Series C round. IQT Partner Rita Waite stated: "Solving intelligence for the physical world unlocks enormous commercial value and long-term strategic national importance. Skild AI is uniquely positioned to do both." This signals potential defense and government market interest. | 高 | SU001, SU003 |
| CU007 | Skild AI announced a partnership with ABB Robotics to integrate the Skild Brain into ABB's industrial robot portfolio. ABB Robotics President Marc Segura stated: "Integrating Skild AI's generalized robot intelligence into our portfolio will help customers scale industrial-grade automation more quickly and address increasingly complex application scenarios across a broad range of industries." As of May 2026, ABB has confirmed the partnership but noted that details remain confidential. | 高 | SU002, SU006, SU004 |
| CU008 | Skild AI announced a partnership with Teradyne Robotics' Universal Robots (UR) and Mobile Industrial Robots (MiR) to integrate the Skild Brain into their cobot and autonomous mobile robot portfolios. Universal Robots CEO Jean-Pierre Hathout stated: "Working with Skild AI and NVIDIA allows us to bring advanced AI capabilities to our cobots — enabling them to handle more dynamic, variable tasks across industries." This OEM partnership gives Skild AI access to UR's installed base of hundreds of thousands of deployed robots globally. | 高 | SU002, SU006 |
| CU009 | Skild AI and NVIDIA are deploying the Skild Brain in a production manufacturing environment at a Foxconn facility in Houston, Texas. The deployment automates the assembly of NVIDIA's Blackwell GPU server racks using dual robotic arms performing complex tasks: picking and placing a busbar, placing a limit block, drilling 16 screws in succession, and removing the limit block. This is described as the first mass-scale public deployment of the Skild Brain after years of internal testing. | 高 | SU002, SU004, SU006, SU012, SU024 |
| CU010 | Technical.ly confirmed via Skild AI's chief of staff Aditya Raghunathan that the Skild Brain has already been deployed to customers in warehousing, construction, and inspections prior to the Foxconn announcement, but Skild does not share the names of these other companies. The Foxconn partnership represents the first public mass-deployment, confirming that undisclosed prior deployments exist. | 高 | SU004, SU001 |
| CU011 | Skild AI acquired Zebra Technologies' Robotics Automation business in April 2026, including the Symmetry Fulfillment orchestration platform. This acquisition brings an established set of enterprise warehouse operator customers previously using Zebra's Symmetry platform, effectively expanding Skild's active customer base with proven, revenue-generating warehouse deployments. | 高 | SU005, SU013, SU014, SU027 |
| CU012 | CTL Global Solutions, a third-party logistics provider, is a named customer of Zebra Technologies' Symmetry Fulfillment platform. The January 2025 Zebra press release identified CTL Global Solutions as a customer that adopted the expanded Symmetry Fulfillment solution, benefiting from AMR-assisted picking and 30% fewer robots needed. Following the April 2026 Skild acquisition, CTL Global Solutions becomes a Skild customer by inheritance. | 中 | SU008, SU016 |
| CU013 | Encore Fulfillment is a named customer of Zebra Technologies' Symmetry Fulfillment platform, identified in Zebra investor relations as adopting the expanded Symmetry solution with AMR coordination and 30% robot reduction. As a legacy Zebra Symmetry customer, Encore Fulfillment becomes a Skild customer following the April 2026 acquisition. | 中 | SU008, SU016 |
| CU014 | Geneva10 (G10) Fulfillment selected Zebra's Symmetry automation platform for batch and cluster picking operations. The IWLA (International Warehouse Logistics Association) noted that Geneva10 expected productivity gains exceeding 40% and improved management of surge order volumes. Geneva10 Fulfillment becomes a Skild customer following the Zebra acquisition. | 中 | SU015, SU008 |
| CU015 | Skild AI's business model is primarily B2B: the Skild Brain is licensed as a model-as-a- service via cloud API to robot OEM partners, system integrators, and enterprise customers. Hardware OEMs embed the Skild Brain into their robot portfolios; enterprise deployers access it to animate specific robot fleets. The Zebra acquisition adds a fleet orchestration software layer enabling end-to-end warehouse automation as a service. | 高 | SU001, SU022, SU003 |
| CU016 | HPE (Hewlett Packard Enterprise) announced a partnership with Skild AI in March 2025 to provide private AI-as-a-service infrastructure for Skild's robot brain training and inference. HPE provides HPE Cray XD670 servers with NVIDIA HGX H200 for training and HPE ProLiant DL380a with NVIDIA L40S for real-time inferencing. This positions HPE as a key infrastructure partner enabling customer deployments at scale. | 高 | SU007, SU025 |
| CU017 | STN provides production-grade cloud infrastructure for Skild AI, including custom- tailored GPU environments optimized for high-speed learning and task generalization across multiple robot embodiments. STN is a deployment infrastructure partner rather than a customer, but its involvement enables Skild's customer-facing deployment pipeline. | 中 | SU017 |
| CU018 | The data flywheel mechanism creates a natural form of customer lock-in: each deployment generates proprietary real-world robot training data that feeds back into the Skild Brain, making it smarter and more capable across all deployments. Customers who have contributed deployment data have an embedded incentive to remain on the platform as their proprietary context improves the shared model. | 中 | SU001, SU002, SU010 |
| CU019 | Net Revenue Retention (NRR), Gross Revenue Retention (GRR), churn rates, contract lengths, customer satisfaction scores, and renewal rates are not publicly disclosed by Skild AI. The company is early-stage in commercial operations, having reached $30M revenue only in 2025, making historical retention metrics unavailable in any case. | 低 | SU001, SU004 |
| CU020 | The exact number of paying customers is not disclosed by Skild AI. The Series C press release mentions "multiple customers" and deployment data that feeds the flywheel, but provides no count. Analyst sources infer "dozens of commercial users" based on revenue of approximately $30M and the nature of enterprise robotics contracts, but this is an estimate and not company-confirmed. | 低 | SU004, SU022, SU026 |
| CU021 | Skild AI's phased go-to-market is to first deploy in semi-structured settings (factories, warehouses), gather data for deploying in less structured environments (hospitals, hotels), and ultimately achieve general-purpose robots for unstructured environments including homes. This means enterprise customers in manufacturing and logistics are the early adopter base that enables future consumer and healthcare expansion. | 高 | SU002, SU003 |
| CU022 | NVIDIA is both an investor (through NVentures) and a technical partner providing Isaac Lab, Isaac Sim, Cosmos Transfer, and Jetson compute for Skild AI's training and inference pipeline. NVIDIA's involvement as a catalyst of the Foxconn deployment and the ABB/Universal Robots integrations makes NVIDIA a critical partner ecosystem node rather than a direct customer. | 高 | SU002, SU006, SU019 |
| CU023 | The Zebra Symmetry Fulfillment platform had proven deployments in some of the world's most demanding logistics and supply chain environments before the Skild acquisition. Named customers (CTL Global Solutions, Encore Fulfillment, Geneva10 Fulfillment) demonstrate the platform's production readiness and commercial track record in warehousing, providing Skild with an immediately monetizable installed base in the logistics vertical. | 高 | SU005, SU008, SU015, SU016 |
| CU024 | No independently verified third-party case studies, G2/Capterra reviews, Gartner Peer Insights reviews, or named customer testimonials for the Skild Brain itself exist as of May 2026. The Foxconn deployment is publicly confirmed by the company and press coverage, but no operational outcome metrics (throughput improvement, error rate, labor savings) from that deployment have been disclosed by either Foxconn or Skild. | 高 | SU004, SU018 |
| CU025 | Amazon has developed its own internal robot foundation model, DeepFleet, which was announced in July 2025 and deployed across 1 million+ robots in over 300 Amazon fulfillment centers globally. This demonstrates that Amazon is building its own competing infrastructure for robot AI—raising the question of whether Amazon's Bezos Expeditions investment in Skild represents a hedge, or a genuine deployment intent. | 高 | SU003, SU009 |
| CU026 | The Skild Brain enables robots costing $4,000–$15,000 to perform tasks that previously required $250,000+ specialized robotic systems, according to the NVIDIA case study. This cost reduction, if validated at customer sites, represents a compelling value proposition that could accelerate enterprise adoption and expand the addressable customer base to small and medium-sized businesses (SMBs), which Skild explicitly names as a target segment. | 中 | SU010, SU002 |
| CU027 | Skild AI's GitHub organization (github.com/skild-ai) contains no public repositories as of May 2026. This means there is no open-source developer community building on the Skild Brain platform. API access is gated to enterprise partners and customers, limiting third-party ecosystem development and independent developer adoption signals. | 中 | SU018, SU022 |
| CU028 | The Skild Brain has been tested and deployed across more than 30 distinct robot types, including quadrupeds, humanoids, tabletop arms, and mobile manipulators. This breadth of embodiment support is core to the OEM partner value proposition, as it means a single AI model can serve the entire portfolio of a multi-product OEM like ABB without separate models per robot family. | 中 | SU001, SU002, SU010 |
| CU029 | Skild AI's Zebra Symmetry platform—prior to the April 2026 acquisition—was expanded in January 2025 to reduce the total number of AMRs required in a warehouse by up to 30% while maintaining or improving picking throughput. Named beneficiaries include CTL Global Solutions and Encore Fulfillment. This operational metric represents independent, production-grade customer proof for the Symmetry layer of Skild's offering. | 高 | SU008, SU016 |
| CU030 | No public reports of failed Skild AI deployments, customer complaints, safety incidents, or customer churn have been identified as of May 2026. The absence of adverse evidence is partly explained by the early stage of commercial deployments (beginning meaningfully in 2025) and by the lack of public customer disclosure, which makes verification of success or failure outcomes equally impossible through third-party sources. | 低 | SU004, SU024 |
| CU031 | Skild AI's current procurement model for enterprise customers is inferred to be direct enterprise sales with negotiated licensing agreements. There is no public evidence of a self-serve developer portal, marketplace listing, or trial/freemium path as of May 2026. Enterprise procurement cycles for robotics AI software are typically long (6–18 months) and require integration with existing robot hardware and factory management systems. | 中 | SU018, SU022 |
| CU032 | The Skild Brain deployment in Foxconn's Houston facility involves a complex, long-horizon task: picking a busbar, placing a limit block, drilling 16 screws in sequence, and removing the limit block. The robot uses an end-to-end neural network fine-tuned with a small amount of task data, achieving recovery behaviors from disturbances that are difficult to program by hand. This provides a production-verified reference for the Skild Brain's manipulation capabilities in an electronics manufacturing setting. | 高 | SU002, SU012 |
| CU033 | The LG CNS partnership (announced June 2025) targets smart factory, smart logistics, and urban service robots including elder care and facility patrol. LG CNS is a system integrator subsidiary of LG Group and one of Korea's largest IT services firms. This partnership makes LG CNS both a channel partner and a deployment customer for the Skild Brain in the Asia-Pacific region. | 高 | SU001, SU004 |
| CU034 | Skild AI's customer concentration risk is unclear. Revenue of ~$30M concentrated across an unknown number of customers—inferred as potentially a small number of large enterprise contracts—creates execution risk if any major customer pauses or exits. The strategic investor-customer overlap (Amazon, Samsung, LG) means that investment relationships and commercial relationships are intertwined, which could make it difficult to distinguish genuine arms-length customer revenue from relationship-driven pilots. | 中 | SU001, SU022 |
| CU035 | Pittsburgh urban testing demonstrated that Skild humanoid robots achieved 60–80% task performance within hours of data collection in never-before-seen environments (parks, streets, fire escapes). While this is an internal performance benchmark rather than a customer outcome metric, it establishes a reference point for real-world deployment capability and suggests that production readiness for at least locomotion-heavy tasks is achievable with limited local training data. | 中 | SU010 |
| CR001 | Sim-to-real gap is a well-documented failure mode for robotics foundation models; models trained primarily in simulation frequently fail in unstructured real-world environments due to physics, sensor, and visual mismatches. | 高 | SR012, SR004 |
| CR002 | Domain randomization — Skild's primary sim-to-real mitigation — cannot fully replicate real-world environmental complexity including imperfect sensors, material variability, lighting changes, and novel object geometries. | 高 | SR012, SR020 |
| CR003 | Skild AI's 'omni-bodied' generalization claims for the Skild Brain are company-stated and partner-reported (NVIDIA case study); no independent third-party benchmarks on the commercial model have been published as of May 2026. | 高 | SR002, SR022 |
| CR004 | Skild AI uses NVIDIA Isaac Lab for simulation training (trillions of synthetic experiences) and Cosmos Transfer for data augmentation; the training pipeline is built predominantly on NVIDIA infrastructure. | 高 | SR014, SR002 |
| CR005 | Training frontier AI models costs $30M–$200M+ per run as of 2025; Skild's claimed 1,000x data scale training implies commensurately high compute expenditures, creating ongoing operating expense risk. | 中 | SR012, SR007 |
| CR006 | Emergent behavior in large foundation models deployed in physical environments creates liability exposure: unexpected motor commands in human-robot shared workspaces can cause injury, property damage, or customer loss. | 中 | SR012, SR004 |
| CR007 | Real-world deployment data fed back into foundation model post-training can introduce adversarial, edge-case, or erroneous interaction data, creating model drift or degradation risk across the entire deployed fleet. | 中 | SR012, SR020 |
| CR008 | Skild AI reported approximately $30M in revenue for 2025, growing from zero in 'just a few months'; this figure is company-stated and unaudited. | 中 | SR001, SR013 |
| CR009 | $14B valuation at approximately $30M in revenue implies a ~467x revenue multiple, placing Skild among the most richly valued pre-scale AI companies globally. | 高 | SR001, SR013 |
| CR010 | Skild AI has not disclosed a customer list, customer concentration data, contract terms, or segment-level revenue breakdown; revenue quality is not independently verifiable. | 高 | SR001, SR022 |
| CR011 | Commercialization of hardware-software integrated robotics typically takes 2–4 years from initial customer trials to meaningful enterprise run rates, due to procurement cycle length, safety assessment, and systems integration complexity. | 高 | SR008, SR030 |
| CR012 | Enterprise robotics procurement cycles are typically 6–18 months, including pilot evaluation, site safety assessment, integration with WMS/MES systems, and full fleet deployment. | 中 | SR025, SR008 |
| CR013 | Physical Intelligence raised approximately $1.07B total and was valued at $5.6B as of November 2025; its hardware-agnostic robot foundation model directly competes with Skild Brain at a significantly lower valuation multiple. | 高 | SR008, SR025, SR030 |
| CR014 | Figure AI raised over $1B in a September 2025 Series C at a $39B valuation, targeting both industrial and consumer humanoid robots with its proprietary Helix VLA system. | 高 | SR009, SR010 |
| CR015 | Alibaba open-sourced a robotics AI foundation model in February 2026, demonstrating that major platform players are willing to commoditize the foundation model layer to drive ecosystem adoption rather than proprietary licensing. | 高 | SR011, SR021 |
| CR016 | Google DeepMind, Meta AI, Tesla Optimus, and Amazon are actively competing for robotics AI researchers from the same small talent pool as Skild AI, driving compensation and equity expectations significantly higher. | 中 | SR014, SR023 |
| CR017 | Strategic investor-customers (Samsung, LG, Schneider Electric) that currently license Skild Brain could develop in-house AI capabilities or acquire a competitor, simultaneously removing a revenue source and a deployment channel. | 中 | SR012, SR021 |
| CR018 | As of May 2026, there is no unified federal regulatory framework specifically governing general-purpose AI-controlled robots in US workplaces; OSHA's existing machine-guarding rules predate foundation model AI. | 高 | SR005, SR006 |
| CR019 | The EU AI Act, now in force, is expected to classify AI-controlled robots operating in shared human workspaces as high-risk systems, requiring conformity assessments, human oversight protocols, and comprehensive data governance. | 高 | SR004, SR019, SR028 |
| CR020 | EU Machinery Regulation 2023/1230, effective January 2027, introduces three requirements for AI-embodied robots: autonomy thresholds for self-evolving systems, lifetime cybersecurity obligations, and collaborative risk mapping for human-robot shared spaces. | 高 | SR004, SR018, SR019 |
| CR021 | EU Product Liability Directive, in force December 2024, allows standalone liability claims against AI software systems for defectiveness without requiring a physical hardware fault; supply chain members including algorithm trainers can bear shared liability. | 高 | SR004, SR019 |
| CR022 | In-Q-Tel, the CIA's venture capital arm, is a disclosed investor in Skild AI; portfolio companies can face enhanced ITAR and EAR export control scrutiny if their technology has intelligence or defense applications. | 中 | SR014, SR022 |
| CR023 | Skild AI has not publicly disclosed licensing terms or legal analysis for its internet-scale video training dataset, creating potential copyright exposure similar to challenges facing other foundation model companies. | 中 | SR012, SR021 |
| CR024 | US state-level AI workplace laws (Colorado May 2024, Illinois September 2024, NYC Local Law 144) create a patchwork compliance environment that affects Skild's enterprise customers and indirectly complicates procurement. | 高 | SR005, SR006 |
| CR025 | EU Machinery Regulation requires enhanced conformity assessment and documented safety proofs for future operational states for machines demonstrating 'self-evolving behaviour through experience' — the exact capability Skild's in-context learning represents. | 高 | SR004, SR018 |
| CR026 | SoftBank led Skild AI's Series A (2024), Series B (June 2025), and Series C (January 2026), making SoftBank the sole lead investor across all three institutional funding rounds. | 高 | SR001, SR013 |
| CR027 | Skild AI has raised approximately $1.83B+ in total across all rounds per Crunchbase data as of January 2026. | 高 | SR001, SR013 |
| CR028 | SoftBank Vision Fund 2 posted a $3.6B loss in fiscal year ended March 2025, due to portfolio markdowns in difficult funding conditions. | 高 | SR007, SR017 |
| CR029 | SoftBank Vision Fund has reported approximately $48B in cumulative losses over two years through 2023, reflecting overexposure in concentrated high-risk bets. | 中 | SR026, SR027 |
| CR030 | Skild AI's monthly gross burn rate is not publicly disclosed; based on estimated headcount (200-400 post-Zebra), compute infrastructure, and prior-round burn benchmarks, gross burn is estimated at $30–60M per month. | 低 | SR001, SR022 |
| CR031 | Skild AI's valuation tripled from $4.5B (Series B, June 2025) to $14B (Series C, January 2026) in just seven months, reflecting private market narrative momentum rather than revenue trajectory. | 高 | SR001, SR013 |
| CR032 | The purchase price for the Zebra Technologies Robotics Automation Business (April 2026) is not publicly disclosed; any cash component reduces the net Series C runway below the headline estimate. | 中 | SR003, SR016 |
| CR033 | Deepak Pathak (CEO) and Abhinav Gupta (President) co-founded Skild AI in 2023; both are Carnegie Mellon University Robotics Institute professors with combined 25+ years of academic AI and robotics research expertise. | 高 | SR014, SR015, SR023 |
| CR034 | Neither Deepak Pathak nor Abhinav Gupta has previously built and scaled a commercial technology company from early-stage through to hundreds of millions in annual revenue. | 中 | SR022, SR023 |
| CR035 | Skild AI competes for robotics AI researchers and engineers against Google DeepMind, Meta AI, Tesla Optimus, Amazon Robotics, Figure AI, and Physical Intelligence — all of which offer significant compensation and compute access. | 高 | SR014, SR015 |
| CR036 | The April 2026 Zebra Technologies Robotics Automation Business acquisition adds organizational integration complexity, including legacy system migration, cultural integration between a research-intensive startup and an enterprise hardware company, and existing customer continuity obligations. | 中 | SR003, SR016, SR024 |
| CR037 | Talent retention from the Zebra Robotics / Fetch Robotics engineering team is a material risk; key personnel who built the Symmetry fleet orchestration platform may exit if integration is mishandled, depriving Skild of institutional knowledge for serving existing enterprise customers. | 中 | SR003, SR016 |
| CR038 | Physical robots in warehouse and factory settings can cause injury; industry-documented incidents (Amazon warehouse, logistics operations) demonstrate that robotic system failures in shared human-robot spaces produce real safety consequences. | 高 | SR004, SR018 |
| CR039 | No publicly disclosed third-party physical or AI safety certifications (ISO 10218, ANSI/RIA R15.06, or equivalent) have been published for the Skild Brain as of May 2026. | 高 | SR022, SR002 |
| CR040 | No public incident history exists for Skild AI robot deployments; the company was pre-commercial for most of 2024–2025, limiting available safety track record. | 中 | SR001, SR022 |
| CR041 | Open-source foundation models create sustained downward pressure on proprietary model pricing by lowering technical barriers for competitors and enabling hardware OEMs to build in-house AI capabilities without licensing fees. | 中 | SR011, SR020, SR021 |
| CR042 | EU Product Liability Directive supply-chain provisions place shared liability on data annotators and algorithm trainers — including foundation model providers — for defects in AI systems that harm end users. | 高 | SR004, SR019 |
| CR043 | No litigation, SEC filing, regulatory inquiry, or disclosed legal proceeding involving Skild AI has been identified in publicly available sources as of May 2026. | 中 | SR005, SR022 |
| CR044 | Partnership on AI identifies open-source foundation model proliferation as creating both safety risks (reduced control over dangerous capabilities) and commercial risks (margin compression for proprietary providers) that will intensify as open models improve. | 中 | SR020, SR021 |
| CR045 | Foundation models in robotics face documented out-of-distribution generalization failures in real-world environments; key challenges include visual and physical domain mismatch, sensor noise, and adversarial edge cases not represented in simulation training data. | 高 | SR012, SR020 |
| CV001 | Skild AI's Series C (January 2026) established a post-money valuation exceeding $14B, making it one of the most highly valued private robotics AI companies globally. | 高 | SV001, SV002, SV003, SV004 |
| CV002 | The Series C raised $1.4B at a $14B valuation, tripling the Series B valuation of $4.5–4.7B in just seven months — one of the fastest valuation ramp trajectories in private AI history. | 高 | SV002, SV003, SV024 |
| CV003 | At $14B valuation and $30M 2025 ARR, Skild AI's trailing ARR multiple is approximately 467x — the highest disclosed revenue multiple in the physical AI and robotics foundation model sector. | 高 | SV001, SV002, SV003 |
| CV004 | The 467x ARR multiple leaves no margin for commercial execution risk; any meaningful growth deceleration or multiple compression would result in a 50–90%+ drawdown from the Series C entry price. | 高 | SV001, SV008, SV021 |
| CV005 | The Series C investor list includes SoftBank (lead), NVIDIA (NVentures), Samsung, LG, Schneider Electric, Macquarie, Bezos Expeditions, Salesforce Ventures, and IQT/In-Q-Tel. | 高 | SV001, SV005, SV002 |
| CV006 | IQT's (In-Q-Tel) participation as a Series C investor signals defense/intelligence use-case positioning and creates potential access to US government procurement channels not available to competitors. | 中 | SV001, SV005 |
| CV007 | CEO Deepak Pathak stated that Skild AI has raised more than $2B in total across all rounds; Crunchbase tracks $1.83B, with the difference attributable to an undisclosed seed amount. | 高 | SV002, SV003, SV029 |
| CV008 | Physical Intelligence (pi) was valued at $5.6B in late 2025 after raising $600M total; it has not disclosed revenue, making the multiple effectively undefined, but the company open-sourced pi0 in February 2025. | 高 | SV011, SV012 |
| CV009 | Figure AI was valued at approximately $39B in 2026, supported by a BMW manufacturing deployment as named customer evidence; Figure builds its own humanoid hardware plus AI layer. | 高 | SV009, SV010 |
| CV010 | 1X Technologies (Norwegian humanoid robotics) was reportedly in discussions at a valuation of up to $10B in 2025–2026; 1X builds NEO humanoid hardware and develops its own AI layer. | 低 | SV008 |
| CV011 | Covariant (RFM-1) was acquired by Amazon in 2024 after raising $222M, with Amazon's co-founders joining. The acquisition signals that robotics AI software layers are valuable enough for Fortune 10 companies to acquire. | 中 | SV008 |
| CV012 | Among physical AI peers, Skild AI has the highest disclosed revenue-to-valuation multiple at 467x; Figure AI, Physical Intelligence, and 1X have not disclosed ARR, making direct multiple comparison limited. | 中 | SV003, SV008, SV009, SV011 |
| CV013 | NVIDIA open-sourced GR00T N1, a foundation model for generalist humanoid robots, in March 2025 — creating a free, high-quality alternative to proprietary models like the Skild Brain. | 高 | SV027, SV028 |
| CV014 | Physical Intelligence open-sourced pi0 in February 2025 under Apache 2.0 license, providing a freely available vision-language-action (VLA) model for robotics that competes with Skild Brain's foundation model offering. | 中 | SV012 |
| CV015 | Under a bull case (200% CAGR 2025–2026), Skild AI's 2026 ARR would reach approximately $90M, implying a forward revenue multiple of approximately 156x at the $14B Series C valuation. | 中 | SV001, SV008 |
| CV016 | Comparable public AI infrastructure companies (e.g., CoreWeave at IPO) traded at approximately 20–30x forward revenue; enterprise SaaS at hyper-growth trades at 15–50x forward revenue — far below Skild's 467x trailing multiple. | 中 | SV008, SV030 |
| CV017 | The $14B valuation is defensible only under a winner-take-most platform thesis, where Skild captures a disproportionate share of a projected $33–38B physical AI market by 2030–2035. | 中 | SV013, SV015, SV021 |
| CV018 | The $14B valuation is at risk if open-source models (GR00T, pi0) are adopted by Skild's OEM partners; the software pricing power that justifies the multiple depends on Skild maintaining proprietary differentiation. | 中 | SV027, SV012, SV008 |
| CV019 | SoftBank Vision Fund 2 posted losses in FY2025, raising questions about follow-on capital capacity for Skild AI in future rounds; SoftBank led all three institutional rounds (A, B, C). | 中 | SV008, SV032 |
| CV020 | To generate a 1.5x return on a $14B Series C entry, an exit valuation exceeding $21B is required; achieving this requires $1B+ ARR by 2028–2030 at a public-market revenue multiple of 20–30x. | 中 | SV001, SV021 |
| CV021 | Goldman Sachs revised its humanoid robot market forecast to $38B by 2035, citing AI breakthroughs and 40% reductions in manufacturing costs — the primary macro tailwind underpinning Skild's valuation. | 高 | SV013, SV014 |
| CV022 | MarketsandMarkets projects the AI robotics market to grow from $6.1B (2024) to $33.4B (2030) at a 40.4% CAGR — the primary market growth metric cited by Skild AI investors. | 高 | SV015, SV016 |
| CV023 | Morgan Stanley projects a $5T global humanoid robot opportunity by 2050, suggesting that physical AI could become one of the most valuable technology sectors in history. | 中 | SV017, SV023 |
| CV024 | If Skild AI captures 5–10% of a $33B AI robotics market by 2030, its ARR would be $1.65B–$3.3B; at a 10x forward revenue multiple, the implied valuation would be $16.5B–$33B. | 低 | SV013, SV015, SV021 |
| CV025 | Skild AI's Bengaluru office (opened February 2026) and LG CNS Korean partnership signal active international expansion — consistent with a platform company pursuing global market capture. | 高 | SV005, SV022 |
| CV026 | The Zebra Technologies acquisition (April 2026) expands Skild's total addressable market into established warehouse automation enterprise accounts, potentially accelerating revenue growth and diversifying the customer base. | 中 | SV005, SV022 |
| CV027 | The strategic investor base (NVIDIA, Samsung, LG, Schneider Electric, IQT) provides distribution, hardware access, and market signal that goes beyond pure capital — supporting the platform valuation through ecosystem lock-in. | 中 | SV001, SV005, SV025 |
| CV028 | The $14B valuation is characterized as stretched relative to disclosed financial metrics; it is conditionally justifiable if Skild achieves 150%+ CAGR through 2028 without major competitive disruption. | 中 | SV001, SV008, SV021 |
| CV029 | The absence of independent benchmarks for the Skild Brain means that the technical superiority claim ('1000x more training data, better generalization') is unverified — a risk that could impair the valuation if benchmarks reveal unexpected limitations. | 中 | SV008, SV021 |
| CV030 | Any serious diligence process must obtain audited financials, a detailed revenue waterfall by customer and cohort, and evidence of at least 2–3 quarters of sustained recurring revenue before investing at the current $14B+ valuation. | 高 | SV001, SV021 |
| CV031 | Skild AI's Series A ($300M at $1.5B, July 2024), Series B (~$135M at ~$4.7B, June 2025), and Series C ($1.4B at $14B, January 2026) represent a compressed funding timeline consistent with companies on an accelerated commercialization path. | 高 | SV018, SV019, SV001 |
| CV032 | The Series B led by SoftBank followed just 11 months after the Series A; the Series C followed just 7 months after the Series B — an unusually compressed funding cycle reflecting both market enthusiasm and SoftBank's commitment to the thesis. | 高 | SV002, SV003, SV024 |
| CV033 | Skild AI is the second most highly valued private robotics company globally as of May 2026, behind Figure AI ($39B) and ahead of Physical Intelligence ($5.6B) and 1X Technologies (~$10B). | 中 | SV009, SV011, SV008, SV001 |
| CV034 | The Skild Brain's 'omni-bodied' hardware-agnostic positioning means Skild's TAM is theoretically the entire enterprise robotics market rather than a single hardware category — a structural advantage that justifies a premium multiple relative to hardware-specific competitors. | 中 | SV005, SV025, SV030 |
| CV035 | The private market AI robotics sector has experienced systematic valuation inflation in 2024–2026 driven by strategic investors pursuing ecosystem positioning; Skild's $14B may partially reflect strategic premium above fundamental justification. | 中 | SV008, SV021, SV030, SV031 |
| CV036 | Under the base case (100% CAGR 2025–2028), Skild AI would reach approximately $240M ARR by 2028. At a 15x forward multiple, the implied valuation would be $3.6B — a 75% drawdown from the $14B Series C entry price, representing a materially negative return for Series C investors. | 中 | SV001, SV008, SV030 |
| CV037 | No public-market pure-play robotics software company provides a direct revenue multiple comparable to Skild AI. The closest analogues are pre-revenue AI platforms, which trade at 20–100x next-twelve-month revenue — still far below 467x trailing multiple. | 中 | SV013, SV016, SV030 |
| CV038 | Exit readiness for Skild AI is assessed as low-to-medium for 2026–2027: audited revenue is absent, ARR is sub-scale for a public listing, and competitive moat is unverified. An IPO is not feasible before 2028 at the earliest under base-case assumptions. | 中 | SV001, SV021, SV030 |
| CV039 | The three thesis-break triggers for Skild AI's investment thesis are: (1) any major OEM partner publicly adopting GR00T or pi0 over the Skild Brain; (2) revenue growth decelerating below 50% CAGR for two consecutive quarters; and (3) SoftBank withdrawing or significantly reducing follow-on capital commitment. | 中 | SV027, SV012, SV019 |
| CV040 | Bull-case assumptions for justifying the $14B valuation require simultaneously: 150%+ CAGR through 2028, software gross margins above 70%, at least one Tier-1 manufacturer as a named public customer, and no major adoption of open-source alternatives by Skild's OEM channel — all four must hold concurrently. | 中 | SV001, SV013, SV008 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Business Wire | Skild AI Raises $300M Series A To Build A Scalable AI Foundation Model For Robotics | The round was led by Lightspeed Venture Partners, Coatue, SoftBank Group, and Jeff Bezos (through Bezos Expeditions)... The funding brings the company to a valuation of $1.5B. |
| SO002 | Business Wire | Skild AI Raises $1.4B, Now Valued Over $14B | The latest funding brings the company's valuation to over $14 billion. |
| SO003 | Business Wire | Skild AI Acquires Zebra Technologies' Robotics Automation Business | Skild AI today announced the acquisition of Zebra Technologies' Robotics Automation business, including its Symmetry Fulfillment orchestration platform. |
| SO004 | Sequoia Capital | Partnering with Skild: The Future of Embodied Intelligence | I first met Deepak and Abhinav on a Thursday afternoon, as they were raising their initial seed round, and we were partners by Tuesday. |
| SO005 | TechCrunch | Robotics software maker Skild AI hits $14B valuation | Skild AI CEO Deepak Pathak told Bloomberg that the company has now raised more than $2 billion to date. |
| SO006 | Skild AI | Announcing Series C | Live revenue grew from zero to about $30M in just a few months in 2025, and is growing rapidly with multiple customers. |
| SO007 | Crunchbase News | Robotics Startup Skild AI Lands $1.4B, Tripling Valuation To $14B In Just 7 Months | The fundraise comes just over seven months after Skild raised a $135 million Series B at a $4.5 billion valuation. |
| SO008 | The Robot Report | Skild AI grabs $300M to build foundation model for robotics | Yesterday, Skild AI emerged from stealth mode and announced that it has closed a $300 million Series A round. |
| SO009 | Inc. Magazine | This Startup Raised $300 Million to Build a Better AI Brain for Robots | Skild was founded in May 2023 by Abhinav Gupta and Deepak Pathak, two ex-professors from Carnegie Mellon University. |
| SO010 | Yahoo Finance (GuruFocus) | Nvidia, Samsung Back Skild AI's $4.5 Billion Valuation | Skild still needs to prove its model by landing big enterprise deals and scaling up those adaptive systems beyond pilot labs. |
| SO011 | Skild AI | LinkedIn Company Profile | Building general purpose robotic intelligence. Company size 11-50 employees | |
| SO012 | Tracxn | Skild – 2026 Company Profile & Team | 34 (As on Dec 31, 2024) |
| SO013 | Skild AI | Skild AI Expands Global Footprint To Bengaluru | We are proud to announce that we've officially opened our newest office in Bengaluru, India. |
| SO014 | Analytics India Magazine | US-Based Robotics Startup Skild AI Opens Office in Bengaluru | |
| SO015 | CMU School of Computer Science | Abhinav Gupta Homepage | |
| SO016 | The South First | After Anthropic and Open AI, Skild AI to set up office in Bengaluru | |
| SO017 | Morningstar / Dow Jones | Robotics Startup Skild AI Raises $1.4 Billion at Valuation Over $14 Billion | |
| SO018 | PitchBook | Skild AI 2026 Company Profile: Valuation, Funding & Investors | |
| SO019 | Skild AI | Skild AI Homepage | Mobile Manipulation Platform. Our AI can execute low-level skills like grasping, handover, and navigation on mobile platforms. |
| SO020 | Skild AI (Blog – Series C Detail) | Skild AI Data Flywheel and Deployment Strategy | Live revenue grew from zero to about $30M in just a few months in 2025, and is growing rapidly with multiple customers. |
| SO021 | Business Wire | Skild AI Raises $300M Series A – Full Release Text | Gupta and Pathak have been Carnegie Mellon University professors with a combined 25 years of experience between them in robotics and AI... they have a 150+ h-index, over 90k citations |
| SO022 | Sequoia Capital | Partnering with Skild – Detailed Founder Story | He went on to pursue a Ph.D. in AI at Berkeley while joining Facebook AI Research (FAIR), co-founded a startup that was later acqui-hired, and then became an assistant professor at the Robotics Institute at CMU. |
| SO023 | Yahoo Finance (GuruFocus) | Nvidia, Samsung Back Skild AI's $4.5 Billion Valuation (Series B detail) | Nvidia quietly dropped $25 million into Skild AI's Series B, and Samsung chipped in another $10 million, joining SoftBank's $100 million lead. |
| SO024 | Crunchbase | Skild AI – Crunchbase Series B Detail (Jun 12, 2025) | |
| SO025 | Analytics India Magazine | Skild AI opens Bengaluru office (Feb 2026) | |
| SO026 | Briefglance | Skild AI Hits $14B Valuation on Bet to Build One Brain for All Robots | Skild AI's strategy stands in contrast. By focusing exclusively on the 'brain' and remaining hardware-agnostic, it aims to become the 'Android' or 'Windows' for the entire robotics industry. |
| SM001 | Grand View Research | Industrial Robotics Market Size, Share | Industry Report, 2030 | The global industrial robotics market size was estimated at USD 33,956.1 million in 2024 and is projected to reach USD 60,562.0 million by 2030, growing at a CAGR of 9.9% from 2025 to 2030. |
| SM002 | MarketsandMarkets | Industrial Robots Market Report 2024-2029 (Global Forecast) | The industrial robots market is projected to grow from USD 16.89 billion in 2024 to USD 29.43 billion by 2029, registering a CAGR of 11.7% during the forecast period. |
| SM003 | Grand View Research | Warehouse Automation Market Size And Share Report, 2030 | The global warehouse automation market size was estimated at USD 19.23 billion in 2023 and is projected to reach USD 59.52 billion by 2030, growing at a CAGR of 18.7% from 2024 to 2030. |
| SM004 | Mordor Intelligence | Warehouse Automation Market – Industry Size & Growth 2025–2031 | The Warehouse Automation Market size is expected to increase from USD 29.98 billion in 2025 to USD 34.17 billion in 2026 and reach USD 65.74 billion by 2031, growing at a CAGR of 13.98% over 2026–2031. |
| SM005 | MarketsandMarkets (via PRNewswire) | Embodied AI Market worth $23.06 billion by 2030 – Exclusive Report by MarketsandMarkets | The global embodied AI market is projected to grow from USD 4.44 billion in 2025 to USD 23.06 billion by 2030, at a CAGR of 39.0%. |
| SM006 | Grand View Research | Embodied AI Market Size & Share | Industry Report, 2033 | The global embodied AI market size was estimated at USD 4.67 billion in 2025 and is projected to reach USD 67.63 billion by 2033, growing at a CAGR of 39.7% from 2026 to 2033. |
| SM007 | RobotToday | Global Robotics Industry: Comprehensive Sector Overview 2025 | the global robotics market in 2025 totals approximately $50–55 billion USD. This report anchors to ABI Research's $50 billion figure. |
| SM008 | Goldman Sachs | Humanoid Robots: Sooner Than You Might Think | Goldman Sachs suggests humanoid robots could be economically viable in factory settings between 2025 to 2028, and in consumer applications between 2030 and 2035. |
| SM009 | CNBC | Morgan Stanley says humanoid robots will be a $5 trillion market by 2050. How to play it | Morgan Stanley projects that the global humanoid robot market could reach $5 trillion by 2050, with over 1 billion units in use. |
| SM010 | Investing.com | Humanoid robots seen as $5 trillion global opportunity at Morgan Stanley | Humanoid robots seen as $5 trillion global opportunity at Morgan Stanley, with prices declining from $200,000 per robot in 2024 to $50,000 by 2050. |
| SM011 | CB Insights | The physical AI models market map: Behind the arms race to control robot intelligence | Physical AI models market map covering the arms race to control robot intelligence across foundation model providers, hardware OEMs, and enterprise deployers. |
| SM012 | Business Wire | Skild AI Raises $300M Series A To Build A Scalable AI Foundation Model For Robotics | The U.S. Chamber of Commerce estimates that there are currently 1.7 million open manufacturing jobs in the U.S. and the National Association of Manufacturers estimates that 2.1 million jobs will go unfilled by 2030. |
| SM013 | Business Wire | Skild AI Raises $1.4B, Now Valued Over $14B | Skild AI's robots generate training data with zero human in the loop, creating a data flywheel that compounds the model's advantage with every deployment. |
| SM014 | Business Wire | Skild AI Acquires Zebra Technologies' Robotics Automation Business | Skild AI today announced the acquisition of Zebra Technologies' Robotics Automation business, including the Symmetry Fulfillment platform, to accelerate end-to-end warehouse automation deployments. |
| SM015 | Sequoia Capital | Partnering with Skild: The Future of Embodied Intelligence | We believe we are at the GPT-3 moment for robotics — the inflection point at which a generalist AI architecture enables breakthrough capability across the entire domain. |
| SM016 | Inc. Magazine | This Startup Raised $300 Million to Build a Better AI Brain for Robots | The startup's robot AI can control a wide range of different robot 'bodies' without needing to be retrained for each — unlike competing systems that must be purpose-built for specific robot types. |
| SM017 | Yahoo Finance (GuruFocus) | Nvidia, Samsung Back Skild AI's $4.5 Billion Valuation | Nvidia and Samsung have backed Skild AI's $4.5 billion valuation in a new funding round that underscores growing corporate interest in generalist robotics AI platforms. |
| SM018 | The Robot Report | Skild AI grabs $300M to build foundation AI model for robotics | Skild AI's foundation model approach represents a fundamental departure from classical robot programming — instead of writing code for each task, the Skild Brain learns generalizable skills. |
| SM019 | Analytics India Magazine | US-Based Robotics Startup Skild AI Opens Office in Bengaluru | Skild AI's expansion to Bengaluru signals its ambition to scale globally and tap India's deep AI and robotics talent pool. |
| SM020 | Tracxn | Skild – 2026 Company Profile & Team | Skild AI competes in the physical AI and robotics foundation model space, targeting enterprise automation across warehousing, manufacturing, and inspection. |
| SM021 | Skild AI | Announcing Series C | Skild's robots generate real-world training data at scale with zero human annotation overhead, creating a compounding data advantage that widens with every deployment. |
| SM022 | Skild AI | Skild AI Homepage | The Skild Brain is the world's first unified foundation model for robotics — one brain for any robot, any task, any environment. |
| SM023 | The South First | After Anthropic and Open AI, Skild AI to set up office in Bengaluru | Skild AI's Bengaluru office will focus on AI and robotics research, expanding global headcount as the company scales commercial deployments. |
| SM024 | Morningstar / Dow Jones | Robotics Startup Skild AI Raises $1.4 Billion at Valuation Over $14 Billion | Skild AI's $14B+ valuation makes it the most highly valued private robotics software company globally as of January 2026. |
| SM025 | PitchBook | Skild AI 2026 Company Profile: Valuation, Funding & Investors | Skild AI has raised $2B+ across four rounds from 2023 to 2026, representing one of the fastest capital formation trajectories in robotics AI. |
| SM026 | Briefglance | Skild AI Hits $14B Valuation on Bet to Build One Brain for All Robots | Skild AI's unique market position is its omni-bodied model that generalizes across robot types and tasks — a technical moat that competitors have struggled to replicate. |
| SP001 | PR Newswire / Physical Intelligence | Physical Intelligence Raises $400 Million to Create Foundation Models for Robots | Physical Intelligence raises $400 million to build general-purpose foundation models for robots. |
| SP002 | Physical Intelligence | pi0: A Vision-Language-Action Flow Model for General Robot Control | We introduce pi-zero, our first generalist robot policy, trained on a diverse set of tasks across multiple robot types. |
| SP003 | NVIDIA Developer Blog | NVIDIA GR00T N1: An Open Physical AI Model for Generalist Humanoid Robots | NVIDIA GR00T N1 is the world's first open, general-purpose foundation model designed for humanoid robots. |
| SP004 | Google DeepMind | Gemini Robotics: Bringing AI into the Physical World | Gemini Robotics is a family of AI models that brings together the intelligence of Gemini with physical dexterity. |
| SP005 | PR Newswire / Figure AI | Figure Raises $675M at $2.6B Valuation and Signs Collaboration Agreement with OpenAI | Figure has raised $675 million in a Series B funding round at a post-money valuation of $2.6 billion. |
| SP006 | Business Wire | 1X Technologies Raises $100M to Accelerate Development of General-Purpose Humanoid Robots | 1X Technologies has raised $100M to accelerate development of general-purpose humanoid robots. |
| SP007 | TechCrunch | Amazon Snaps Up Covariant Co-founders in Major AI Robotics Deal | Amazon has poached the co-founders of warehouse robotics AI startup Covariant in a deal that transfers the founders without buying the company outright. |
| SP008 | TechCrunch | Covariant Raises $100 Million in Fresh Funding | Covariant has raised $100 million in fresh funding as it rebuilds following the departure of its co-founders to Amazon. |
| SP009 | TechCrunch | Intrinsic Joins Google After Years as an Alphabet Moonshot | Intrinsic, Alphabet's robotics software startup, has officially joined Google after years operating as a semi-independent moonshot. |
| SP010 | Agility Robotics | Agility Robotics and GXO Reach 100,000 Tote-Move Milestone with Digit | Agility Robotics' Digit robot has completed more than 100,000 tote moves at GXO Logistics facilities. |
| SP011 | Apptronik | Apptronik Closes Series A Funding Round | Apptronik has closed its Series A funding round with participation from Google, Mercedes-Benz, John Deere, B Capital, and Qatar Investment Authority. |
| SP012 | The Robot Report | Physical Intelligence pi-zero Foundation Model for Robotics | The pi-zero model demonstrates cross-embodiment generalization across seven robot platforms, outperforming OpenVLA and Octo on standard benchmarks. |
| SP013 | The Robot Report | NVIDIA GR00T N1 Open Foundation Model for Humanoid Robots | NVIDIA's dual-system GR00T N1 reflects a platform strategy designed to lock humanoid OEMs into NVIDIA's compute infrastructure. |
| SP014 | Sanctuary AI | Phoenix Gen-7: Sanctuary AI's Most Advanced Humanoid Robot | Phoenix Gen-7 is Sanctuary AI's most advanced humanoid robot, designed for general-purpose labor in manufacturing and logistics. |
| SP015 | Unitree Robotics | Unitree G1 Humanoid Robot | The Unitree G1 is available starting at $13,500 for the base configuration. |
| SP016 | The Robot Report | Chinese Humanoid Robots Dominate Volume Shipping in 2025 | Unitree and AgiBot collectively shipped more than 10,000 humanoid robot units in 2025, establishing Chinese manufacturers as volume leaders globally. |
| SP017 | Covariant | RFM-1: Robotics Foundation Model for Warehouse Automation | RFM-1 is a universal robotics foundation model trained on the world's largest warehouse manipulation dataset. |
| SP018 | ABB Ltd. | ABB Robotics Annual Results 2024 | ABB Robotics achieved $2.3B in revenue in 2024, with more than 80% of offerings incorporating AI or software-enabled capabilities. |
| SP019 | FANUC Europe | FANUC Physical AI Initiative | FANUC's Physical AI initiative integrates ROS2, Python, and generative AI capabilities into its industrial robot controller platform. |
| SP020 | OpenAI | OpenAI Robotics | OpenAI has established a dedicated robotics division to develop AI systems that enable robots to operate in complex real-world environments. |
| SP021 | Toyota Research Institute | Large Behavior Models — TRI Robotics | TRI's Large Behavior Models represent one of the world's largest robot manipulation datasets, enabling robots to follow natural language instructions. |
| SP022 | Figure AI | Figure Achieves First Commercial Deployment Milestone with Figure 02 | Figure has achieved its first commercial deployment milestone, with Figure 02 robots operating at BMW manufacturing facilities. |
| SP023 | Sequoia Capital | Partnering with Skild AI | Skild's dataset is 1,000 times larger than most competitors and growing, creating a compounding data moat. |
| SP024 | Business Wire | Skild AI Raises $1.4B, Now Valued Over $14B | Skild AI has raised $1.4 billion in its Series C, bringing total capital raised to over $2 billion at a $14 billion valuation. |
| SP025 | Business Wire | Skild AI Acquires Zebra Technologies Robotics Automation Business | Skild AI has acquired Zebra Technologies' Robotics Automation Business, including the Fetch Robotics AMR fleet and enterprise software platform. |
| SP026 | Bloomberg | Figure AI Breaks AI Partnership With OpenAI as Robotics Ambitions Diverge | Figure AI and OpenAI have ended their robotics AI collaboration, with sources citing diverging strategic ambitions. |
| SI001 | Business Wire | 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. The new capital will be used to continue scaling the company's model training and growing the future deployment of its technology. |
| SI002 | TechCrunch | Robotics software maker Skild AI hits $14B valuation | The startup has raised a $1.4 billion Series C round that values it at more than $14 billion. Skild AI CEO Deepak Pathak told Bloomberg that the company has now raised more than $2 billion to date. |
| SI003 | Crunchbase News | Robotics Startup Skild AI Lands $1.4B, Tripling Valuation To $14B In Just 7 Months | The raise brings Pittsburgh-based Skild AI's total raised to over $1.83 billion, according to Crunchbase. The company says it grew from zero to about $30 million revenue 'in just a few months' in 2025. |
| SI004 | Bloomberg | Robotics Startup Skild Valued Above $14 Billion After SoftBank-Led Funding Round | |
| SI005 | Forbes | Skild AI Is Building A 'General Purpose Brain' For Robots | The company announced Tuesday it has raised $300 million at a $1.5 billion valuation in a Series A funding round led by Lightspeed Ventures, Softbank, Coatue and Amazon founder Jeff Bezos. |
| SI006 | Yahoo Finance | Nvidia And Samsung Back $4.5B Robotics Startup Skild AI With $35M As SoftBank And Jeff Bezos Drive Push Into Consumer Robots | The Series B funding round, which values Skild AI at approximately $4.5 billion, is led by a $100 million investment from Japan's SoftBank Group. Samsung has committed $10 million to the round. According to Bloomberg, Nvidia is contributing $25 million. |
| SI007 | CNBC TV18 | Nvidia, Samsung plan investments in robotics startup Skild AI | |
| SI008 | TrendForce | NVIDIA, Samsung Reportedly Back Startup Skild AI in Consumer Robotics Push | |
| SI009 | Robotics and Automation News | Skild AI acquires Zebra Technologies' robotics automation business | Skild AI grew from zero to approximately $30 million in revenue in just a few months in 2025 and is now positioned to scale enterprise deployments at a pace that was not previously possible. |
| SI010 | The There's a Robot for That | Skild AI Secures $1.4B | Universal Robot Brain | |
| SI011 | Tech Funding News | Robotics unicorn Skild AI grabs $1.4B to build a universal brain for every robot | |
| SI012 | The Outpost AI | Skild AI Triples Valuation to $14B in Seven Months as SoftBank Leads $1.4B Robotics Funding | |
| SI013 | Sacra | Skild AI funding, news & analysis | Compute economics: Training and running large robotic foundation models requires massive computational resources, creating ongoing infrastructure costs that scale with customer usage and potentially constraining unit economics compared to traditional software businesses. |
| SI014 | Kruze Consulting | Understanding AI Compute Costs for Startups | AI compute/hosting costs growing at a 300% CAGR for startups, compared to ~53% for SaaS peers. |
| SI015 | Epoch AI | Training compute costs are doubling every eight months for large-scale AI | Training compute costs are doubling every eight months for large-scale AI models. |
| SI016 | Visual Capitalist | Charted: The Surging Cost of Training AI Models | Frontier models (GPT-4, Gemini Ultra, Llama 3.1) require training runs costing $79M to $192M. |
| SI017 | Anelya.net | Your AI Startup Burns Differently Than SaaS. Here's the Math. | AI foundation model startups have significantly higher burn rates than traditional SaaS companies, with compute and infrastructure now dominating the cost structure. |
| SI018 | Gaurav Singh Ventures (GSV) | AI Startup Funding & Cost Challenges in 2025 | Only well-funded organizations can afford frontier-scale models; smaller startups typically train smaller models or use pre-trained open weights to control burn rate. |
| SI019 | Geo.sig.ai | Skild AI Revenue & Market Share 2026 | Skild AI reported approximately $30 million in revenue for 2025, growing from zero; B2B SaaS subscription model targeting enterprise robot fleet operators. |
| SI020 | Skild AI | Skild AI Official Website — Mobile Manipulation Platform | Our AI can execute low-level skills like grasping, handover, and navigation on mobile platforms. These skills are abstracted away using an API call, allowing users to build applications without worrying about details of the unstructured, messy real world. |
| SI021 | Evertiq | Skild AI acquires Zebra Technologies' robotics automation business | |
| SI022 | Saudi Tech Post | Skild AI acquires Zebra Technologies' robotics business | |
| SI023 | Photonics Media | Zebra Technologies Divests Robotics Automation Business | |
| SI024 | Maginative | Skild.ai Raises $300M Series A with $1.5B Valuation | |
| SI025 | Medium (creed_1732) | Skild AI Robotics Manufacturing Foundation Model Raised $1.4B | |
| SI026 | AI2Work | Skild AI's $1.4B Bet on Robot Foundation Models | Skild AI's universal brain slashes robot deployment costs for enterprises from ~$250,000 to as low as $4,000–$15,000 per unit, leading to much greater SaaS attach and upsell rates over time. |
| SI027 | Particle News | Skild AI Lands $1.4 Billion Series C at $14 Billion Valuation | |
| SI028 | SEC EDGAR — Zebra Technologies Corporation | Form 8-K Current Report — Zebra Technologies Corporation | Date of report (Date of earliest event reported): February 12, 2026 — Zebra Technologies Corporation Form 8-K filed with the SEC. |
| SI029 | Market Screener | Skild AI, Inc. acquired Robotics Automation Business of Zebra Technologies Corporation | |
| SI030 | Humanoids Daily | Skild AI Acquires Zebra's Robotics Division to Build the 'Orchestrated Warehouse' | |
| SE001 | Skild AI | Skild AI Official Website — Mobile Manipulation Platform | Our AI can execute low-level skills like grasping, handover, and navigation on mobile platforms. These skills are abstracted away using an API call, allowing users to build applications without worrying about details of the unstructured, messy real world. |
| SE002 | Skild AI | Skild AI Series C Announcement — The Skild Data Flywheel | Each deployment contributes to a growing data flywheel, helping improve performance across the fleet and adaptability to new scenarios. Live revenue grew from zero to about $30M in just a few months in 2025. |
| SE003 | NVIDIA | Skild AI Builds Omni-Bodied Robot Brain With NVIDIA | The model demonstrates remarkable adaptability to mechanical changes, recovering from jammed wheels within 2–3 seconds and broken legs after several attempts rather than experiencing failure. |
| SE004 | Hewlett Packard Enterprise | Skild AI Accelerates Development of Human-Like Robot Brain with AI Solutions from HPE | Skild AI worked with STN, an HPE Partner Ready Service Provider, to leverage the GPU One service based on HPE AI infrastructure and NVIDIA accelerated computing. |
| SE005 | Sequoia Capital | Partnering with Skild — The Future of Embodied Intelligence | Deepak and Abhinav leveraged large-scale data to build a foundation model using their adaptive architecture, based on transformers. What they got by doing this was totally unique: a way to unlock intelligence in the embodied, physical world. |
| SE006 | BusinessWire | Skild AI Raises $300M Series A To Build A Scalable AI Foundation Model For Robotics | |
| SE007 | BusinessWire | Skild AI Raises $1.4B, Now Valued Over $14B | |
| SE008 | The Robot Report | Skild AI Grabs $300M to Build Foundation AI Model for Robotics | |
| SE009 | arXiv / ICML 2017 | Curiosity-Driven Exploration by Self-Supervised Prediction (Pathak et al.) | We propose curiosity as an intrinsic reward signal, which is computed as the prediction error of an agent's knowledge about its own actions and their consequences. |
| SE010 | arXiv / RSS 2021 | RMA: Rapid Motor Adaptation for Legged Robots (Kumar, Pathak et al.) | |
| SE011 | LG CNS | LG CNS Partners with Skild AI to Develop Industrial AI Humanoid Robot Solution | LG CNS and Skild AI have signed a strategic partnership to jointly develop AI humanoid robot solutions for smart factory, smart logistics, and urban service environments. |
| SE012 | Analytics India Magazine | LG CNS Signs Deal with Skild AI to Build Industrial Humanoid Robots | |
| SE013 | GitHub | Skild AI GitHub Organization — No Public Repositories | |
| SE014 | BusinessWire | Skild AI Acquires Zebra Technologies' Robotics Automation Business | |
| SE015 | Robotics & Automation News | Skild AI Acquires Zebra Technologies' Robotics Automation Business | |
| SE016 | TechCrunch | Robotic Software Maker Skild AI Hits $14B Valuation | |
| SE017 | Crunchbase News | Robotics Startup Skild AI Triples Valuation with $1.4B Series C | |
| SE018 | xMaquina | How Skild AI Is Building a General-Purpose Humanoid Mind | |
| SE019 | Pulse 2.0 | Skild AI $1.4 Billion At $14 Billion Valuation For AI Robotics | |
| SE020 | Innovation Library | Skild AI — One Brain for Every Robot | |
| SE021 | The AI Insider | Skild AI Says It Has Created AI Capable of Controlling Any Type of Robot | |
| SE022 | Grokipedia | Skild AI — Grokipedia | |
| SE023 | arXiv | Foundation Models in Robotics: Applications, Challenges, and the Future | |
| SE024 | Yahoo Finance | Skild AI Acquires Zebra Technologies Robotics Automation Business | |
| SE025 | Business Korea | LG CNS Taps US Robotics Firm to Develop Industrial AI Humanoids | |
| SE026 | arXiv / ICRA 2016 | Supersizing Self-Supervision: Learning to Grasp from 50K Tries and 700 Robot Hours (Gupta et al.) | |
| SE027 | The AI Insider | Skild AI Acquires Zebra Technologies' Robotics Automation Business | |
| SU001 | Skild AI | Announcing Series C - Skild AI | |
| SU002 | Skild AI | The Reindustrial Revolution: Partnering with ABB Robotics, Universal Robots, and NVIDIA | |
| SU003 | Business Wire | Skild AI Raises $1.4B, Now Valued Over $14B | |
| SU004 | Technical.ly (via Wayback Machine) | Skild faces real-world test of its robot brain in Nvidia, Foxconn factory deal | |
| SU005 | Robotics and Automation News | Skild AI acquires Zebra Technologies' robotics automation business | |
| SU006 | AiThority | Skild AI Expands Generalized Robot Intelligence Across Industries With ABB Robotics, Universal Robots, and NVIDIA | |
| SU007 | Hewlett Packard Enterprise | Skild AI Accelerates Development of Human-like Robot Brain with AI Solutions from Hewlett Packard Enterprise | |
| SU008 | Zebra Technologies | Zebra Technologies Expands Symmetry Fulfillment Solution to Increase Productivity with 30% Fewer Robots | |
| SU009 | Crunchbase | Robotics Startup Skild AI Lands $1.4B, Tripling Valuation To $14B In Seven Months | |
| SU010 | NVIDIA | Skild AI Builds Omni-Bodied Robot Brain With NVIDIA | |
| SU011 | Business Wire | Skild AI Provides First Look at Its General-Purpose Robotic Brain | |
| SU012 | U.S. News & World Report | Skild AI, Nvidia Deploy Robot Brain on Blackwell Assembly Lines | |
| SU013 | National Law Review | Skild AI Acquires Zebra Technologies' Robotics Automation Business | |
| SU014 | The AI Insider | Skild AI Acquires Zebra Technologies' Robotics Automation Business | |
| SU015 | International Warehouse Logistics Association | Geneva10 Fulfillment Selects Zebra's Automation | |
| SU016 | Business Wire | Zebra Technologies Expands Symmetry Fulfillment Solution to Increase Productivity with 30% Fewer Robots - Business Wire | |
| SU017 | STN Inc. | Skild Partnership Case Study - STN | |
| SU018 | GitHub | Skild AI GitHub Organization | |
| SU019 | NVIDIA Investor Relations | NVIDIA and US Manufacturing and Robotics Leaders Drive America's Reindustrialization With Physical AI | |
| SU020 | Skild AI | Skild AI - Official Website | |
| SU021 | The Outpost | Skild AI Triples Valuation to $14B in Seven Months as SoftBank Leads $1.4B Robotics Funding | |
| SU022 | Sacra | Skild AI - Sacra Research Report | |
| SU023 | AIM Media House | How Is Skild AI Transforming Warehouse Automation? | |
| SU024 | Economic Times - Enterprise AI | Skild AI and Nvidia Unveil Advanced Robot Brain for Automated Manufacturing | |
| SU025 | Robotics and Automation News | Skild AI builds robot brain with HPE and Nvidia to merge physical and digital worlds | |
| SU026 | Sig.ai | Skild AI Revenue and Market Share 2026 | |
| SU027 | Rocking Robots | Skild AI Acquires Zebra Technologies Warehouse Robotics Unit | |
| SR001 | Crunchbase News | Robotics Startup Skild AI Lands $1.4B, Tripling Valuation To $14B In Just 7 Months | The raise brings Pittsburgh-based Skild AI's total raised to over $1.83 billion. The company says it grew from zero to about $30 million revenue 'in just a few months' in 2025, and 'is growing exponentially.' |
| SR002 | The Robot Report | Skild AI grabs $300M to build foundation model for robotics | Skild AI claims to be building the industry's 'first unified robotics foundation model' called the Skild Brain. |
| SR003 | The Robot Report | Skild acquires Fetch Robotics assets from Zebra | Skild AI acquires Zebra Technologies' Robotics Automation Business to transform warehouse automation. |
| SR004 | Osborne Clarke | Robotics at a global regulatory crossroads: compliance challenges for autonomous systems | For 'software as a product', under the revised directive, machine-learning models and other AI systems can face standalone liability claims for defectiveness, without the need for a fault in physical hardware. |
| SR005 | Fisher Phillips | Comprehensive Review of AI Workplace Law and Litigation as We Enter 2025 | There remains no federal law specifically regulating the use of AI in the workplace. We don't expect the Republican-controlled Congress to enact any workplace-related AI laws in 2025 or 2026. |
| SR006 | NCSL | Artificial Intelligence 2025 Legislation | Over 30 states have formed AI committees or taskforces that have begun issuing reports and recommendations, many of which will turn into proposed legislation. |
| SR007 | PitchBook | Vision Fund loss drags on SoftBank quarterly profit | Vision Fund 2 posted a $3.6 billion loss due to portfolio markdowns and difficult funding conditions. |
| SR008 | Bloomberg | Robotics Startup Physical Intelligence Valued at $5.6 Billion in New Funding | Physical Intelligence valued at $5.6 billion in new funding round, closing approximately $600M Series B. |
| SR009 | TechCrunch | Figure reaches $39B valuation in latest funding round | Figure reaches $39B valuation in its latest funding round, becoming one of the most highly valued robotics startups. |
| SR010 | Figure AI | Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation | Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation, with investors including NVIDIA, Intel Capital, Salesforce, LG Technology Ventures. |
| SR011 | StartupNews.fyi | Alibaba open-sources robotics AI model as competition in embodied AI intensifies | Alibaba open-sources robotics AI model as competition in embodied AI intensifies, signaling that major players view ecosystem adoption over exclusive ownership. |
| SR012 | arXiv | Foundation Models in Robotics: Applications, Challenges, and the Future | Foundation models in robotics face documented challenges around generalization to unseen environments; out-of-distribution inputs remain a central unsolved problem for deployment at scale. |
| SR013 | Humanoids Daily | Skild AI Secures $1.4 Billion Series C, Tripling Valuation to Over $14 Billion | Skild AI secures $1.4 billion Series C, tripling its valuation to over $14 billion in just seven months from a $4.5 billion Series B valuation. |
| SR014 | Lightspeed Venture Partners | Skild is bringing Generative AI to the real world | Lightspeed is proud to partner with Deepak Pathak and Abhinav Gupta, two of the world's most accomplished roboticists from CMU. |
| SR015 | Sequoia Capital | Partnering with Skild: The Future of Embodied Intelligence | Partnering with Skild because of Deepak Pathak and Abhinav Gupta's unique ability to combine frontier AI research with practical robotics deployment. |
| SR016 | Humanoids Daily | Skild AI Acquires Zebra's Robotics Division to Build the 'Orchestrated Warehouse' | Skild AI acquires Zebra's Robotics Division, including Fetch Robotics assets, to build the orchestrated warehouse combining Skild Brain with Symmetry fleet orchestration. |
| SR017 | CNBC | SoftBank Vision Fund swings to annual loss as investment gains slow | SoftBank's Vision Fund swings to annual loss as investment gains slow, with Vision Fund 2 posting significant markdowns on portfolio companies. |
| SR018 | Intertek | Changes to Robots — How the New Framework Addresses Autonomous Systems | Regulation 2023/1230 introduces three pivotal requirements for robotics manufacturing: autonomy thresholds, lifetime cybersecurity responsibilities and collaborative risk mapping, effective January 2027. |
| SR019 | Bird & Bird | Smart Robots, Dual Regulations — Navigating the AI Act and Machinery Compliance | Companies developing AI-embodied robots must navigate overlapping compliance obligations under the EU AI Act and EU Machinery Regulation 2023/1230, both effective by 2027. |
| SR020 | Partnership on AI | Risk Mitigation Strategies for the Open Foundation Model Value Chain | Open foundation model proliferation creates safety and commercial risks for proprietary AI incumbents; startups must pivot toward differentiated deployment capabilities. |
| SR021 | Georgetown CSET | Open Foundation Models: Implications of Contemporary Artificial Intelligence | Open-source foundation models substantially lower barriers to entry in AI, reducing the moat value of proprietary models and increasing competitive pressure on commercial incumbents. |
| SR022 | Grokipedia | Skild AI | Skild AI was founded in 2023 by Deepak Pathak and Abhinav Gupta, both Carnegie Mellon University professors in the Robotics Institute. |
| SR023 | Analytics India Magazine | IIT Graduates Founded Robotics Company Skild AI Raises $300M | Skild AI was co-founded by Deepak Pathak and Abhinav Gupta, both IIT graduates and CMU Robotics Institute professors without prior commercial enterprise CEO experience. |
| SR024 | Hoodline | Pittsburgh Robot Unicorn Gobbles Up Zebra Warehouse Unit | Pittsburgh-based Skild AI acquires Zebra Technologies' warehouse robotics unit, taking on both the Symmetry platform and the Fetch Robotics engineering team. |
| SR025 | Sacra | Physical Intelligence — Valuation, Funding, and News | Physical Intelligence raised approximately $1.07B total, valued at $5.6B as of November 2025; subscription model at $300/month per connected robot. |
| SR026 | Bloomberg Línea | SoftBank's Vision Fund Losses at $48 Billion, Yet Profit May Be Within Reach | SoftBank's Vision Fund has reported approximately $48 billion in cumulative losses over two years through 2023, reflecting overexposure in concentrated high-risk bets. |
| SR027 | Bitget Academy | SoftBank Vision Fund Investment Risks and Opportunities Analysis 2026 | SoftBank Vision Fund's concentrated, leveraged exposure means it experiences bigger swings in net asset value, increasing short-term losses and market perception risk for portfolio companies. |
| SR028 | Interoperable Europe (European Commission) | Robotics and Autonomous Systems Rolling Plan 2024 | The EU regulatory framework for robotics and autonomous systems requires compliance with the AI Act risk tiers and the new Machinery Regulation, creating substantial conformity obligations for AI-embodied systems. |
| SR029 | Tech in Asia | SoftBank faces $184.4M Q1 loss due to declining portfolio values | SoftBank faces $184.4M Q1 loss due to declining portfolio values, with listed Vision Fund companies facing collective losses of approximately $900M in Q1 2025. |
| SR030 | The Robot Report | Physical Intelligence raises $600M to advance robot foundation models | Physical Intelligence raises $600M to advance robot foundation models, with total funding now exceeding $1B at a $5.6B valuation. |
| SV001 | BusinessWire | Skild AI Raises $1.4B, Now Valued Over $14B | Skild AI has raised a $1.4 billion Series C round, now valued at over $14 billion. The company grew from zero to about $30M revenue in just a few months in 2025. |
| SV002 | TechCrunch | Robotics software maker Skild AI hits $14B valuation | The startup has raised a $1.4 billion Series C round that values it at more than $14 billion. |
| SV003 | Crunchbase News | Robotics Startup Skild AI Lands $1.4B, Tripling Valuation To $14B In Just 7 Months | The raise brings Pittsburgh-based Skild AI's total raised to over $1.83 billion, according to Crunchbase. |
| SV004 | Bloomberg | Robotics Startup Skild Valued Above $14 Billion After SoftBank-Led Funding Round | |
| SV005 | Skild AI Blog | Announcing Series C | |
| SV006 | Tech Funding News | Nvidia, SoftBank chase robotics brain Skild AI with $1B bet at $14B valuation | |
| SV007 | Morningstar / Dow Jones | Robotics Startup Skild AI Raises $1.4 Billion at Valuation Over $14 Billion | |
| SV008 | AI2Work | Skild AI's $1.4B Raise: Why Robotics Foundation Models Are 2026's Mega Bet | |
| SV009 | Figure AI | Figure achieves commercial deployment milestone | |
| SV010 | PRNewswire | Figure Raises $675M at $2.6B Valuation and Signs Collaboration Agreement with OpenAI | |
| SV011 | Physical Intelligence | Physical Intelligence Raises $400 Million | |
| SV012 | The Robot Report | Physical Intelligence open-sources Pi0 robotics foundation model | |
| SV013 | Goldman Sachs | The global market for humanoid robots could reach $38 billion by 2035 | Goldman Sachs projects the global humanoid robot market could reach $38 billion by 2035, up from an earlier estimate of $6 billion. |
| SV014 | Goldman Sachs Insights | Humanoid Robots — Climbing the Uncanny Valley | |
| SV015 | MarketsandMarkets | Embodied AI Market Worth $23.06 Billion by 2030 | |
| SV016 | MarketsandMarkets | Industrial Robotics Market | |
| SV017 | CNBC | How to play a $5 trillion market for humanoid robots by 2050 | |
| SV018 | Forbes | Skild AI Is Building A General Purpose Brain For Robots | The company announced Tuesday it has raised $300 million at a $1.5 billion valuation in a Series A funding round led by Lightspeed Ventures, Softbank, Coatue and Amazon founder Jeff Bezos. |
| SV019 | Yahoo Finance | Nvidia And Samsung Back $4.5B Robotics Startup Skild AI With $35M | The Series B funding round, which values Skild AI at approximately $4.5 billion, is led by a $100 million investment from Japan's SoftBank Group. |
| SV020 | Briefglance | Skild AI Hits $14B Valuation on Bet to Build One Brain for All Robots | |
| SV021 | Sacra | Skild AI — Company and Valuation Analysis | |
| SV022 | Technical.ly | Skild AI raises $1.4 billion at a $14 billion valuation | |
| SV023 | CNBC | Humanoid robots seen as $5T global opportunity at Morgan Stanley | |
| SV024 | The Outpost AI | Skild AI triples valuation to $14B in seven months as SoftBank leads $1.4B raise | |
| SV025 | Sequoia Capital | Partnering with Skild | |
| SV026 | Pulse2 | Skild AI $1.4 Billion Funding | |
| SV027 | NVIDIA | NVIDIA GR00T N1: An Open Physical AI Model for Generalist Humanoid Robots | |
| SV028 | The Robot Report | NVIDIA GR00T N1 open humanoid robot foundation model | |
| SV029 | Crunchbase | Crunchbase Company Profile — Skild AI | |
| SV030 | Market Analysis | Skild AI Funding Round Signals a Shift Toward Platform Economics in Robotics | |
| SV031 | The Information | The Robotics Startup Bubble: Valuations Outpace Reality | |
| SV032 | SoftBank Group | SoftBank Group Corp. Q3 FY2025 Results — Vision Fund Investment Disclosures |