初创公司尽调
尽调报告 Robotics / hardware — general-purpose humanoid robots Series B private 2026-05-15

Sanctuary AI

深厚 IP 与灵巧操作领先,但未披露客户、收入或估值,定论仍过早

Sanctuary AI 拥有一流人形机器人 IP 和可信战略支持方,但未披露客户、收入或估值;再叠加 2024 年 11 月联合创始人离任和融资报道互相矛盾,现在下判断还太早。

封面要素

成立时间 01
2018 [CO001]
总部 02
Vancouver, BC, Canada [CO002]
最近公告融资 03
US$100M Series B (June 2024, structure disputed) [CO009, CO011, CO022]
已披露估值 04
Not disclosed [CO023]
已披露收入 / ARR 05
Not disclosed [CO024]
建议 06
research-more

公司概况

Sanctuary AI(Sanctuary Cognitive Systems Corporation)是一家位于温哥华的人形机器人公司,由 Geordie Rose、Suzanne Gildert 和 Olivia Norton 于 2018 年创立。公司开发通用人形机器人 Phoenix,目前已到第八代,并搭配 Carbon AI 认知控制系统;公司在灵巧操作上投入很重,21 个自由度液压手、仿真到现实迁移、零样本手内操作等演示获得广泛报道。Morgan Stanley 研究把 Sanctuary AI 列入全球人形机器人 IP 第一梯队,公司也宣布了与 Magna 和 Microsoft 的合作。但公司从未披露估值、收入、ARR、客户名称或试点地点;联合创始人兼 Chief Science Officer Suzanne Gildert 在 2024 年 11 月 9 日领导层变动中离任;2024 年 6 月融资结构(US$100M Series B 与 CAD $30M EDC 贷款之间的关系)在公开资料中仍相互矛盾。

官网
www.sanctuary.ai
成立时间
2018-01-01
创始人
Geordie Rose, Olivia Norton, Suzanne Gildert
创立地点
Vancouver, British Columbia, Canada
总部
Vancouver, British Columbia, Canada
产品
Phoenix 是一款双足通用人形机器人,目前为 Generation 8(2024 年 12 月),搭配公司自研 AI 认知控制系统 Carbon。产品路线图强调灵巧双手操作,包括 21 个自由度液压手(2024 年 12 月)、触觉感知皮肤(2024 年 12 月)、操作策略的仿真到现实迁移(2025 年 3 月),以及零样本手内操作演示(2025 年 5 月)。
客户
面向工业自动化、物流和零售劳动力替代流程中的 B2B 工业客户;公司尚未公开披露具体客户名称、试点地点或已签约部署。
商业模式
面向 B2B 的人形机器人平台,目标场景是工业自动化、物流和零售;商业模式(机器人销售、租赁、机器人即服务或服务合同)与单位经济模型尚未公开披露。
阶段
Series B private
融资情况
Sanctuary AI 于 2023 年 3 月完成 CAD $58.5M Series A,并于 2024 年 6 月 12 日宣布 US$100M Series B,战略投资方包括 Microsoft、Accenture、Bell、Verizon Ventures、BDC Capital、Export Development Canada、Magna、Workday Ventures 和 Evok Innovations。公开报道对于 2024 年 6 月宣布的 CAD $30M EDC 贷款是否属于 US$100M Series B 的一部分,还是一项独立工具,存在矛盾。BDC Thrive 和 InBC 于 2024 年 7 月宣布一项金额未披露的战略投资。公司从未披露估值。
[CO001, CO002, CO005, CO009, CO011, CO017, CO018, CO023]

执行摘要

主要优势

  • 人形机器人 IP 处于第一梯队;Morgan Stanley 2025 年 2 月 11 日研究将其排全球第三。
  • 差异化灵巧操作栈——21 自由度液压手、触觉传感、sim-to-real 迁移,以及截至 2025 年 5 月的零样本手内操作演示。
  • 战略投资人与合作伙伴包括 Microsoft、Magna、Accenture、Bell、Verizon Ventures、BDC Capital、EDC 和 Workday Ventures,可带来分销、制造和 AI 平台触达。
  • 创始人履历扎实,CEO Geordie Rose 曾在 D-Wave Systems 和 Kindred 留下记录,Vancouver 深科技生态仍在支撑。
  • 产品节奏活跃,Phoenix Gen 7(2024 年 4 月)和 Gen 8(2024 年 12 月)已发布,2024–2025 年有多次能力演示。

主要风险

  • 未公开客户、试点场地、签约部署、收入、ARR 或单位经济模型;数据室外完全无法验证商业逻辑。
  • 联合创始人兼 Chief Science Officer Suzanne Gildert 于 2024 年 11 月 9 日离职,带来治理、研究方向和文化问题,公开报道尚未解答。
  • 2024 年 6 月融资结构报道互相矛盾(US$100M Series B 与 CAD $30M EDC 贷款),累计资本、工具组合和稀释都不清楚。
  • 资金更充足同行(Figure、1X、Agility、Apptronik、Tesla Optimus、Unitree、中国入局者)的人形机器人竞争升温,带来价格和资本强度风险。
  • 人形硬件资本强度高,现金跑道和烧钱速度又未披露;若下一轮按市场情绪而非商业牵引定价,融资压力会压顶。

未决问题

  • 2024 年 6 月 US$100M Series B 与 CAD $30M EDC 贷款究竟是同一工具还是两项独立工具,以及由此得到的累计资本和稀释图景。
  • 客户名称、试点场地、签约部署量,以及当前是否产生任何收入或 ARR。
  • 股权结构表、清算优先权、董事会构成,以及 2024 年 11 月之后的治理安排。
  • Phoenix 的单位经济模型(物料清单、毛利率、服务强度)以及计划采用的商业模式(销售、租赁、RaaS 或服务)。
  • 灵巧操作主张是否能在精心演示之外、按生产负载周期获得独立技术验证。

目录

Chapter 01

01公司概况

1.1 身份、产品与商业模式

Sanctuary Cognitive Systems Corporation(以 Sanctuary AI 名义运营)是一家总部位于加拿大 British Columbia 省温哥华的机器人与人工智能公司,由 Geordie Rose(CEO)、Suzanne Gildert(前 Chief Science Officer,2024 年 11 月离任)和 Olivia Norton(COO)于 2018 年创立。三位创始人来自量子计算先驱 D-Wave Systems,带来了物理、AI 研究和工业运营上的深厚技术背景。 Sanctuary 的公开使命是打造通用人形机器人,让机器人能在工业、物流、零售以及潜在消费场景中执行广泛人类任务。公司的产品架构建立在两个一体化平台上:Phoenix,即目前已到第八代的人形机器人形态(Gen 8 于 2024 年 12 月 16 日发布);以及 Carbon,即公司自研 AI 控制系统,公司称其模仿人脑子系统,包括记忆、视觉、听觉和触觉。 商业模式聚焦 B2B,部署试点面向物流仓库、制造设施以及潜在零售环境。不同于固定式工业机器人或单任务自动化,Sanctuary 把 Phoenix 定位为通用平台,靠遥操作、强化学习和仿真到现实迁移技术训练,让它适配不同工作流,而无需为每个任务定制编程。截至 2026 年 5 月运行日期,公司尚未公开披露具体客户、部署规模、定价结构或收入数字。 公司在加拿大 British Columbia 省温哥华运营,研发、工程和试点职能集中在总部。法律登记主体为 Sanctuary Cognitive Systems Corporation。公司尚未宣布任何国际扩张计划。 [CO001, CO002, CO003, CO004, CO005, CO020]

1.2 创始人、领导层与治理

Geordie Rose(CEO、联合创始人)是一名物理学家和连续创业者,此前创办 D-Wave Systems,这家量子计算公司实现了商业部署并获得了大量风险投资支持。Rose 在复杂软硬件系统上的技术背景,以及此前的融资履历,让他成为可信的深科技创始人。他的公开露出包括媒体采访、会议演讲和新闻稿引语,确立了他作为 Sanctuary 使命和产品愿景主要发言人的角色。关键人风险很实质:Rose 是公司的门面、主要融资人和战略决策者,公开资料中没有指向明确的继任者。 Suzanne Gildert(联合创始人、前 Chief Science Officer)曾任职 D-Wave,专长包括量子计算、AI 和认知架构。Gildert 对 Carbon AI 平台早期技术方向至关重要。2024 年 11 月 9 日,Sanctuary 宣布一次「重大领导层变动」,但未说明 Gildert 后续角色或离任;二级报道和归档新闻引用显示,她在该节点离开公司。这构成了实质性的创始人离任,也可能给技术路线图带来风险,不过公司将此次变动描述为有计划安排。 Olivia Norton(COO、联合创始人)负责运营、合作伙伴关系和规模化工作。相比 Rose 或 Gildert,她的履历公开信息较少,但其联合创始人身份和运营角色显示,她参与制造合作(例如与 Magna)、部署物流和投资者关系。 公司尚未公开披露董事会、独立董事、顾问或投资方董事席位。治理结构、股权所有权分布和创始人归属条款均未知。治理透明度不足提高了尽调风险,尤其考虑到 2024 年 11 月的领导层变动,以及公司缺少公开监管文件(Sanctuary 是一家加拿大私营实体,没有 SEC 或同等监管披露)。 [CO003, CO004, CO006, CO007, CO021]

领导层与创始人表
人员职务背景创始人-市场匹配关键人物风险
Geordie RoseCEO 兼联合创始人物理学家;创办 D-Wave Systems(量子计算);在深科技融资和商业化上有记录技术可信度深;融资能力已验证;具备软硬件系统经验重大:主要发言人、交易签署人、战略决策者;未披露继任者
Suzanne Gildert前首席科学官兼联合创始人D-Wave 前成员;量子计算和 AI 专家;November 2024 离开公司AI / 认知系统专长强;为 Carbon 平台提供技术愿景重大:创始人离职意味着技术联合创始人流失;对 Carbon 路线图的影响未知
Olivia NortonCOO 兼联合创始人运营和合作负责人;背景公开记录较少运营执行;合作伙伴管理(Magna、Microsoft)中等:运营连续性依赖 Norton;公开资料有限

Gildert 在 November 2024 离职,是重大治理事件。董事会构成和投资者席位未披露。

[CO003, CO004, CO006, CO007]

1.3 融资历史、估值与资本结构

Sanctuary AI 的融资历史同时包含已确认轮次和互相冲突的报道,需要谨慎引用。最早披露的轮次是 2023 年 3 月完成的 CAD $58.5 million Series A,依据包括 Pitchbook 和 Tracxn 引用在内的二级报道。Series A 的投资方名称或估值没有获得一手资料确认。 2024 年 6 月,Sanctuary 通过 2024 年 6 月 12 日 GlobeNewswire 新闻稿宣布 US$100 million Series B。Series B 的具名投资方包括 Microsoft、Accenture、Bell、Verizon Ventures、BDC Capital、Export Development Canada(EDC)、Magna、Workday Ventures 和 Evok Innovations。新闻稿称该轮融资将支持「持续产品开发、与 Magna 扩大制造能力,以及与 Microsoft 合作扩展 AI 模型训练」。公司未披露投前或投后估值。 关于 Export Development Canada 的出资,公开报道存在冲突。BetaKit 和 Globe and Mail 在 2024 年 6 月分别提到 EDC 提供 CAD $30 million 贷款(按 2024 年 6 月汇率约 US$22M),但没有说明这笔金额是 $100M Series B 的一部分,还是并行债务工具。最合理的解读是,Series B 总额 $100M 包含 EDC 债务出资以及其他具名投资方的股权承诺,但这一点仍未确认。精确资本结构、债务 / 股权拆分、利息条款和估值均存在证据缺口。 2024 年 7 月,BDC Capital 的 Thrive Venture Fund 和 InBC 宣布对 Sanctuary 进行战略投资,但未披露金额;这笔投资可能与 6 月 Series B 重叠,也可能是一笔较小的独立承诺。 累计披露融资约为 CAD $58.5M(Series A)加 US$100M(Series B),按汇率和时点假设约等于 US$160M。除 EDC 贷款外,公司未披露老股交易、down round、债务工具或 bridge round。公司从未公开披露估值。私募市场估值倍数未知,不过同业人形机器人公司 Figure AI(2024 年 2 月 Series B 后投后估值约 $2.6B)和 1X Technologies(媒体报道隐含约 $1-1.5B)提供粗略参照,显示如果 $100M Series B 以典型后期轮比例定价,Sanctuary 可能拿到 $500M-$1.5B 区间估值;但缺少硬数据时,这完全是推测。 [CO008, CO009, CO010, CO011, CO012, CO022]

融资轮次与资本事件
日期轮次金额币种已披露投资方估值来源可信度
March 2023Series A58.5CAD未披露中(仅二级报道)
June 2024Series B100USDMicrosoft、Accenture、Bell、Verizon Ventures、BDC Capital、EDC、Magna、Workday Ventures、Evok Innovations 等战略投资方高(新闻稿加媒体报道)
June 2024EDC 贷款(可能是 Series B 的组成部分)30CADExport Development Canada高(EDC 披露;是否计入 $100M Series B 不明确)
July 2024BDC Capital + InBC 投资BDC Capital Thrive Venture Fund、InBC 等投资方中(新闻稿;金额未披露;可能与 Series B 重叠)

关于 Series B 结构的报道相互冲突。EDC 贷款可能包含在 $100M 总额内,也可能单独计算。累计披露融资约 CAD 58.5M + USD 100M。

[CO008, CO009, CO010, CO011, CO012]
利益相关方 / 投资者图谱
利益相关方角色经济 / 控制重要性尽调问题
MicrosoftSeries B 投资者 + AI 合作伙伴战略投资者;Azure 集成;AI 模型合作;股权比例未披露核验 Microsoft 的战略意图:技术合作还是组合投资;评估 Azure 锁定风险
AccentureSeries B 投资者 + 合作伙伴战略投资者;咨询和实施合作;股权比例未披露如果 Accenture 为 Sanctuary 竞争对手提供建议,确认咨询合作范围并评估利益冲突
Magna InternationalSeries B 投资者 + 制造合作伙伴战略投资者;制造业试点伙伴;汽车级生产经验;股权比例未披露核验制造合同条款;评估排他性或产能分配
BDC CapitalSeries B 投资者 + July 2024 投资者通过 Thrive Venture Fund 投资的加拿大公共部门投资者;多轮参与;股权比例未披露确认累计投入资本和估值标记;评估投资逻辑是政策驱动还是回报驱动
Export Development Canada (EDC)贷款方 + 潜在投资者提供 CAD $30M 贷款(June 2024);可能包含认股权证或股权部分;贷款是否属于 Series B 总额不明确厘清 EDC 融资工具结构:贷款还是可转债;确认 CAD $30M 是否为 $100M USD Series B 之外的增量
Bell CanadaSeries B 投资者电信战略投资者;潜在部署伙伴;股权比例未披露评估电信用例匹配度;核验 Bell 是否有试点或部署承诺
Verizon VenturesSeries B 投资者电信战略投资者;潜在部署伙伴;股权比例未披露评估电信用例匹配度;核验 Verizon 是否有试点或部署承诺
Workday VenturesSeries B 投资者HR 软件战略投资者;潜在集成或部署伙伴;股权比例未披露评估软件-机器人集成策略;核验 Workday 客户试点或产品集成
Evok InnovationsSeries B 投资者加拿大风险基金;聚焦能源 / 工业;股权比例未披露确认投资逻辑和组合协同;评估能源行业部署潜力
Geordie RoseCEO 兼联合创始人创始人控制;主要决策者;未披露持股比例确认创始人持股和投票控制;评估归属期安排和锁定;评估继任计划

未披露董事会构成。利益相关方图谱反映截至 May 2026 已公开具名的投资者和创始人。所有各方的经济权益和治理权利均未披露。

[CO003, CO004, CO006, CO008, CO009, CO010]
FO001: 融资时间线与资本结构

Sanctuary AI 自 2023 年 3 月 Series A 到 2024 年 7 月 BDC+InBC 投资的融资里程碑。

Series B 结构不清晰;EDC 贷款可能重叠。Series A 和 BDC+InBC 日期按月份近似。

[CO008, CO009, CO010, CO011, CO012]

1.4 规模、里程碑与运营覆盖

Sanctuary AI 尚未披露收入、ARR、员工数、客户数或机器人产量。公司招聘页面列出了机器人工程、AI/ML 研究、仿真、机械设计、电气工程和运营等开放岗位;按岗位多样性和融资规模估算,技术团队可能在 100-300 人区间,但这只是估计。 2024 年至 2026 年 5 月的关键产品和技术里程碑包括:2024 年 4 月 25 日发布 Phoenix Gen 7 人形机器人;2024 年 12 月 16 日发布 Phoenix Gen 8,硬件和软件改进未具体说明;2024 年 12 月 12 日演示 21 个自由度液压手技术,突出灵巧操作能力;2024 年 12 月 19 日发布新的触觉传感器技术,以改善手内物体感知;2025 年 2 月 11 日,Morgan Stanley 人形机器人知识产权排名将 Sanctuary 列为全球第三(较 2024 年 11 月第四名上升),显示公司仍在积极申请专利并开发 IP;2025 年 3 月 18 日演示灵巧操作策略的仿真到现实迁移,显示公司可在仿真中训练行为并迁移到实体机器人;2025 年 4 月 1 日与 Microsoft 一起参加 Hannover Messe 展会,释放工业自动化 go-to-market 信号;2025 年 5 月 15 日(运行日期)发布零样本手内操作演示,展示机器人在没有显式任务训练的情况下自主调整带字母立方体朝向。 合作里程碑包括 2024 年 4 月 11 日宣布与 Tier 1 汽车供应商 Magna International 建立战略制造合作,目标是规模化试点机器人制造;以及 2024 年 5 月 1 日宣布与 Microsoft 合作开发通用机器人 AI 模型(可能使用 Azure OpenAI 和相关基础设施)。 公司尚未披露客户名称、试点地点、部署数量或商业牵引指标。没有具名客户或公开试点,是一个实质尽调缺口;相较 Agility Robotics(Amazon 合作)、Apptronik(Mercedes 合作)和 Figure AI(BMW 合作)等竞争对手,这一点尤其突出。 [CO013, CO014, CO015, CO016, CO017, CO018]

里程碑表
日期事件类型参与方含义
2018公司成立创立Geordie Rose、Suzanne Gildert、Olivia Norton(三位创始人)起步;D-Wave 分拆团队
March 2023Series A CAD $58.5M融资未披露投资者早期研发资本
April 11, 2024Magna 战略合作合作Magna International制造业试点;汽车级生产经验
April 25, 2024Phoenix Gen 7 发布产品Sanctuary AI第七代人形机器人平台
May 1, 2024Microsoft AI 合作合作MicrosoftAI 模型开发;Azure 集成
June 12, 2024Series B USD $100M融资Microsoft、Accenture、Magna、Bell、Verizon、BDC、EDC、Workday、Evok 等战略投资方大额资本注入;战略投资者协同
July 3, 2024BDC Capital + InBC 投资融资BDC Capital、InBC加拿大公共部门支持
November 9, 2024领导层调整(Gildert 离职)反向Suzanne Gildert创始人 / CSO 退出;技术连续性风险
November 27, 2024Morgan Stanley IP 排名 #4规模化Morgan Stanley Research专利组合获得认可
December 12, 202421-DoF 液压手演示产品Sanctuary AI灵巧操作里程碑
December 16, 2024Phoenix Gen 8 发布产品Sanctuary AI第八代人形机器人平台
December 19, 2024触觉传感技术发布产品Sanctuary AI增强触觉感知
February 11, 2025Morgan Stanley IP 排名 #3规模化Morgan Stanley ResearchIP 组合提升;排名上升一位
March 18, 2025Sim-to-Real 迁移演示产品Sanctuary AI训练效率;仿真验证
April 1, 2025与 Microsoft 参加 Hannover Messe规模化Microsoft、Sanctuary AI工业展会曝光
May 15, 2025零样本手内操作演示产品Sanctuary AI泛化能力

里程碑时间线从创立延伸至运行日。公司没有公开披露客户部署或收入里程碑。

[CO001, CO002, CO008, CO009, CO010, CO013]
KPI 快照表
指标数值日期可信度证据缺口
成立时间20182018
总部加拿大 BC 省 Vancouver2026-05-15
已披露融资约 ~USD $160M(CAD 58.5M + USD 100M)2026-05-15Series B 结构不明确
估值未披露;无监管文件或可信泄露
收入 / ARR未披露
员工数基于岗位多样性估计 100-300;未确认
客户数未披露具名客户或试点场地
已生产机器人数量未披露
Morgan Stanley IP 排名#3 全球(Feb 2025)2025-02-11
最新产品代际Phoenix Gen 82024-12-16

运营指标覆盖稀疏。融资数字的结构报道相互冲突。估值从未披露。没有公开监管文件(加拿大私营实体)。

[CO001, CO002, CO008, CO009, CO010, CO013]
FO002: 产品和 IP 里程碑(2024–2025)

2024 年 4 月至 2025 年 5 月的主要产品发布、技术演示和 IP 认可事件。

[CO013, CO014, CO015, CO016, CO017, CO018]
FO003: 关键 KPI 快照

截至 2026 年 5 月的关键公司指标,包含置信度和缺口标记。

融资金额根据 CAD/USD 汇率转换和互相冲突的 Series B 结构报道近似估算。

[CO001, CO008, CO009, CO018, CO015]

1.5 附录

Chapter 02

02市场分析

2.1 市场边界与定义

Sanctuary AI 瞄准的是人形机器人市场,属于更广义的工业自动化和服务机器人行业的一个子集。市场边界覆盖具有人类形态的实体机器人(双足行走、类人上半身、灵巧操作器),目标是在为人类工人设计的环境中运行,而无需专门改造基础设施。纳入支出类别包括:人形机器人本体资本开支、相关 AI 控制软件和仿真平台、部署服务(集成、训练、维护),以及经常性软件订阅或按任务收费许可。排除类别包括传统工业机械臂(SCARA、6 轴关节型机器人、没有腿或完整躯干的协作机器人)、不具备操作能力的自主移动机器人(AMR)、外骨骼和可穿戴辅助设备、远程呈现机器人,以及纯软件自动化(RPA、工作流 bot)。 现状替代品因用例而异。在仓储和物流中,人形机器人面对的是输送系统、固定龙门架上的传统工业机械臂、轮式底盘上的协作机械臂(cobot)、自动叉车,以及由外骨骼或拣选辅助系统增强的人类劳动力。在制造装配中,替代方案包括专用自动化单元、SCARA 机器人、高速 pick-and-place 用 delta 机器人,以及人工装配线。在零售和酒店场景中,替代品是专业服务机器人(清洁机器人、配送机器人、信息亭)、人类员工和自助服务亭。 相邻领域包括通用移动操作平台(非人形但具备灵巧性)、机器人 AI 基础模型(行为克隆、模仿学习、视觉-语言-动作模型)、仿真和数字孪生平台(Isaac Sim、MuJoCo、Gazebo),以及传感器技术(触觉传感器、深度相机、力矩传感器)。Sanctuary 的 Carbon AI 控制系统把公司放在机器人硬件和具身 AI 软件的交汇处,也模糊了机器人 OEM 与 AI 平台提供商之间的边界。 [CM001, CM002, CM003, CM004, CM005]

市场定义表
细分 / 类别纳入支出排除支出主要买方 / 付款方与 Sanctuary 关系
制造装配人形机器人 capex、部署服务、AI 软件订阅固定自动化单元、SCARA 机械臂、输送机工厂运营 / 工程团队、capex 预算Tier 1 汽车 / 电子试点;Magna 合作释放该垂直信号
物流与仓储机器人本体、集成、按任务授权、软件订阅无操作能力的 AMR、输送系统、叉车3PL、电商运营、物流 capex短期体量潜力最高;重复周期部署模型
零售与酒店餐饮机器人本体、店内部署服务、软件门店自助终端、自助结账、纯软件 POS门店运营、门店 opex 预算近期优先级较低;客户接受风险高
医疗与养老护理助力搬运机器人、护理辅助、感染控制医疗器械(FDA 批准)、药品、EHR / 软件医院管理、养老院运营、facility opex长期潜力高;监管门槛最高;Sanctuary 目前未瞄准
服务与现场作业危险 / 远程任务机器人(建筑、公用事业、巡检)固定工业自动化、专用巡检机器人设施运营或现场作业团队潜在未来垂直;不是 Sanctuary 当前落地路径
AI 软件平台(Carbon)认知平台授权、仿真、API 访问、模型训练第三方机器人硬件 OEM 销售技术 / 创新团队、平台买方跨垂直;凸显 Sanctuary 的软件 + 硬件公司定位

支出边界基于 Sanctuary 公开表述的产品范围和市场可比对象;所有排除类别均由二级市场定义(IFR、IBISWorld、Goldman Sachs)确认。鉴于 Sanctuary 截至 May 2026 的商业化路径不指向医疗,监管支出(FDA、医疗器械)排除在主要市场定义之外。

[CM001, CM002, CM003, CM004, CM005, CM013]
FM001: 市场规模判断框架

人形机器人 TAM/SAM/SOM 层级,展示从全球可寻址市场到 Sanctuary 自身可获取市场的各层市场。

SAM 的 $5-8B 估算由分析师推导,并非来自已发布一手来源。公开数据无法计算 SOM。Goldman Sachs 数值来自其 2024 年 2 月研究报告。

[CM006, CM007, CM011, CM012, CM037]

2.2 市场规模与增长预测

Goldman Sachs Research 在 2024 年 2 月发布估计,乐观采用情景下,全球人形机器人市场到 2035 年可能达到 $38 billion;基准情景为 $6 billion,保守情景为 $1 billion,反映采用时点和渗透率上的极大不确定性。该机构建模的潜在部署领域包括制造($14B)、物流和仓储($10B)、零售($5B)、医疗健康($4B)和其他服务($5B)。这些估计假设机器人单价从 2025 年每台 $150,000 降至 2035 年低于 $50,000。在乐观情景下,2025 至 2035 年隐含复合年增长率约为 40-50%,前提是技术成本下降和买方采用里程碑能够兑现,而这些尚未得到验证。 International Federation of Robotics(IFR)报告称,2023 年全球安装工业机器人 553,052 台,同比增长 6%,累计运营存量超过 3.9 million 台。截至 2026 年,人形机器人占比可以忽略不计。CB Insights 识别出截至 2025 年初 27 家融资充分的人形机器人初创公司,累计披露融资超过 $3 billion。World Economic Forum Future of Jobs 2025 报告预测,到 2030 年,自动化将在全球替代 92 million 个岗位并创造 170 million 个新岗位,为包括人形机器人在内的自动化技术提供了广义宏观需求信号。 换用不同市场规模测算框架,结果会大幅分化。一个自下而上的劳动力替代模型,使用 U.S. Bureau of Labor Statistics 关于物料搬运工和仓库工人的数据(2024 年美国约 4.3 million 个岗位),显示如果 10 年内渗透率达到 2-10%,潜在 SOM 为 100,000 至 500,000 台人形机器人;按每台 $50K 计算,隐含累计资本支出为 $5B 至 $25B。截至 2026 年中,尚不存在商业规模部署,因此所有规模估计都带有极大不确定性。 [CM006, CM007, CM008, CM009, CM010, CM011]

TAM/SAM/SOM 测算视角表
发布方年份地域数值CAGR方法可信度限制
Goldman Sachs Research2024 发布 / 2035 预测全球$38B 乐观 / $6B 基准 / $1B 保守~40-50% CAGR 乐观按用例自下而上;单价下降模型;劳动力替代逻辑情景区间宽,反映采用不确定性;2035 距今 10+ 年;方法细节未完全公开
Goldman Sachs Research2024 发布 / 2030 近期全球~$300M(近期估计)n/a近期爬坡假设大规模采用前需要 50-75% 成本下降近期估计对技术成熟度高度敏感;尚无客户验证
IFR World Robotics Report2023 实际值全球553,052 台新增工业机器人装机;总保有量 3.9M 台2023 年同比 +6%成员 OEM 工业机器人出货量年度普查未拆出人形机器人子集;总保有量包含非人形机械臂
CB Insights 人形机器人版图2025 年初全球(初创公司)27 家已融资初创公司;累计融资 >$3Bn/a风险融资数据库覆盖人形机器人类初创公司融资数据可能不完整;聚焦 VC 支持公司;不是总体市场规模
BLS 劳动力数据(自下而上代理)2024美国4.3M 物料搬运 / 仓储工人;平均总薪酬约 $50Kn/a职业就业与工资统计;美国劳工统计局劳动力替代只作代理指标;采用率、监管和 TCO 假设尚未验证
WEF《未来就业 2025》2025全球到 2030 年,自动化带来 170M 个新增岗位 / 92M 个被替代岗位;净新增 78Mn/a调研覆盖 22 个行业集群的 1,000+ 家全球雇主WEF 覆盖广义自动化(非人形机器人专项);估算来自雇主情绪预测,不是市场规模

所有数值均为估算或预测;截至 2026 年 5 月,任何人形机器人厂商都没有公开商业部署数据。Goldman Sachs 方法论只披露了一部分;IFR 和 BLS 是政府 / 行业一手来源。

[CM006, CM007, CM008, CM009, CM010, CM011]
FM002: 市场估算区间

跨多个来源和情景的人形机器人市场规模低、中、高估算。

所有数值均为十亿美元。Goldman Sachs 数值来自 2024 年 2 月研究。BLS 劳动力代理表示:若 10 年内以 $50K/台替代 2-10% 美国物料搬运岗位,需要的资本开支。CB Insights 数值为已披露 VC 融资,不是市场规模。

[CM006, CM007, CM009, CM011]

2.3 买方与细分市场格局

人形机器人的买方-用户-付款方结构因细分市场而显著不同。在制造业,买方通常是工厂运营或自动化工程团队,用户是生产线主管或技术员,付款方是用于提高生产率的企业资本开支预算。预算所有权在运营或工程负责人手中,而非 IT。在物流和仓储中,买方是电商运营商或 3PL 的履约运营或供应链负责人。用户是仓库现场经理。截至 2024 年,Amazon 运营超过 750,000 台移动机器人,但尚未大规模部署人形机器人。采用触发因素包括旺季劳动力短缺,以及 TCO 与临时工成本打平。 在零售和酒店业,采用驱动因素包括年员工流失率超过 100% 和最低工资上涨。约束因素包括客户接受度(恐怖谷担忧)和未经验证的 ROI。医疗和养老是高潜力但高度监管的细分市场。U.S. Bureau of Labor Statistics 预计,由于退休因素,到 2032 年每年会出现 203,000 个护理岗位空缺。约束因素包括严格的医疗器械监管、责任风险和患者同意问题。国防和政府细分市场在危险环境自动化上有长期潜力,政府采购由资本开支拨款和战备要求驱动,不过目前没有任何人形机器人厂商拥有重要国防合同。 [CM013, CM014, CM015, CM016, CM017, CM018]

细分市场与买方地图
细分市场买方用户付款方预算负责人采用触发因素
制造装配工厂自动化 / 工程团队(一级汽车供应商、电子)产线主管、技术员企业资本开支(生产率 / 用工短缺)运营 / 工程 VP 或工厂总经理用工短缺、工资年涨幅 >5-10%、绿地建设、安全要求
物流 / 仓储3PL 运营或电商履约运营仓库现场经理、班组长物流资本开支或运营开支供应链 VP、CFO(大额交易)旺季用工稀缺、新配送中心建设、3-5 年内 TCO 与临时工打平
零售 / 酒店餐饮门店运营或设施管理门店经理、一线员工门店运营开支或企业地产区域运营 VP>100% 年流失率、最低工资上调、客户体验要求
医疗 / 养老医院管理层、养老院运营方护理人员、照护员机构运营开支(非保险报销)CFO 或 CEO(机构层面)护理短缺(BLS 称每年 203K 个岗位空缺)、辅助搬运安全要求、感染控制
国防 / 政府DoD 或政府设施运营技术操作员、基地管理者政府采购(资本开支拨款)合同官 / 项目经理危险环境自动化、人员成本下降、战备要求

买方—用户—付款方结构来自行业类比(工业自动化、专用服务机器人)和二手劳动力市场数据;截至 2026 年 5 月,公开来源没有 Sanctuary 客户引用数据。

[CM013, CM014, CM015, CM016, CM017, CM018]
FM003: 买方与细分市场地图

按目标细分展示人形机器人部署中的买方、用户、付款方关系和采用触发因素。

买方-用户-付款方映射基于行业类比和二级劳动力 / 采购研究。没有 Sanctuary AI 的一手客户访谈数据。

[CM013, CM014, CM015, CM016, CM017, CM018]

2.4 增长驱动因素与采用约束

劳动力短缺和工资通胀是主要需求驱动因素。美国 JOLTS 数据显示,2026 年 4 月职位空缺为 8.1 million 个,失业工人为 7.4 million 人,结构性劳动力供需失衡仍在持续。仓储和物流岗位年流失率为 40-60%。物料搬运工工资在 2020-2025 年间年均增长 4.2%,跑赢生产率提升,改善了机器人替代的回本测算。OECD 国家的人口趋势(老龄化、日本、韩国和欧洲部分地区劳动年龄人口下降)创造了一个持续 10-15 年的需求信号,不依赖短期技术成熟度。 监管和政策驱动因素包括 OSHA 人体工学标准、碳足迹减排目标和产业政策激励(U.S. CHIPS Act、EU Horizon 机器人资助、中国 Made in China 2025 目标)。AI 基础模型成熟(视觉-语言-动作模型、行为克隆、仿真到现实迁移)降低了 Sanctuary 的 Carbon 平台训练新任务策略的时间和成本,是通用可行性的关键使能因素。 技术和经济约束仍然严峻。一台人形机器人 5 年总拥有成本包括采购价(截至 2026 年为 $100K-$150K)、集成服务($50K-$100K)和年度维护($10K-$20K/year),合计约 $200K-$250K。相比一名仓库工人约 $52K 的年度总薪酬,机器人必须达到 90%+ 利用率,才能在 3 年回本期内打平。截至 2026 年 5 月,没有厂商发布来自客户部署、经过验证的生产率基准。非结构化环境中人机协作安全标准(ISO TC 299)尚未覆盖通用人形机器人。机器人造成工伤的责任框架仍不清楚。资本密集度制造了鸡生蛋、蛋生鸡的采用难题;技术成熟度风险也让人形机器人区别于成熟自动化替代方案。Rodney Brooks 和 IEEE Spectrum 分析把这些商业化障碍称为当前人形机器人初创公司的结构性风险,而不只是短期风险。 [CM019, CM020, CM021, CM022, CM023, CM024]

增长驱动因素与约束表
驱动因素 / 约束方向时点对 Sanctuary 的含义尽调问题
美国结构性劳动力短缺(2026 年 4 月,8.1M 个岗位空缺 vs 7.4M 名失业者)驱动因素当前;多年仓储和制造业部署人形机器人的核心拉力信号;推高买方紧迫感核实 Sanctuary 试点客户是否把用工短缺列为 #1 采购触发因素
物料搬运工工资通胀(2020-2025 年 CAGR 4.2%)驱动因素当前;持续改善人形机器人 TCO 回本测算;抬高具备成本竞争力的替代门槛确认 Sanctuary TCO 模型采用现实利用率假设;追问客户回本假设
人口结构:OECD 人口老龄化;日本、韩国、欧洲部分地区劳动年龄人口下降驱动因素中期;结构性形成持续 10-15 年的需求信号;可能加快 Sanctuary 进入日本 / 亚洲市场评估 Sanctuary 是否有日本 / EU 市场进入策略;公开来源尚未看出
产业政策顺风(CHIPS Act、EU Horizon、中国 MIC 2025 机器人目标)驱动因素2024-2030政府补助和补贴降低客户集成成本;减轻 Sanctuary 试点融资负担查明 Sanctuary 或其客户是否获得过政府激励项目(未披露)
AI 基础模型成熟(VLA、行为克隆、模仿学习)驱动因素2025-2030 年加速降低 Carbon 平台训练新任务策略的时间 / 成本;对通用可行性至关重要索取 Carbon 平台当前策略泛化基准和 sim-to-real 迁移成功率
高前期 TCO(每台机器人 5 年全包 $200K-$250K)vs 人力(每名 FTE 全包 $52K)约束近期;随规模下降按 2026 年价格,多数买方 ROI 目前不划算;需要 3-4 年回本和 90%+ 利用率核实 Sanctuary 是否有任何试点已经实现或测算出正回本;公开无数据
人机协作安全 / 监管不确定性约束中期;2-5 年内厘清非结构化环境里,人形机器人缺少 ISO/OSHA 标准,增加部署摩擦和责任风险跟踪 ISO TC 299(机器人安全)进展;评估 Sanctuary 是否参与标准组织
技术成熟度与可靠性缺口约束商业可行性窗口 2026-2028 年当前一代人形机器人缺少生产环境验证过的正常运行时间和任务完成率向 Sanctuary 索取试点部署的任务完成率和正常运行时间基准(尚未公开)

所有驱动因素 / 约束判断均来自公开数据源和专家分析;目前没有 Sanctuary 专属性能数据可用于校准这家公司具体采用时点。

[CM019, CM020, CM021, CM022, CM023, CM024]
FM004: 人形机器人采用漏斗

示意性采用漏斗,展示从运营方意识到问题到商业平台部署的转化。

漏斗转化率是基于工业自动化买方旅程类比得出的示意性估计。没有可用的人形机器人专属买方旅程数据。数值表示相对指数,不代表绝对交易数量。

[CM013, CM014, CM031, CM032]

2.5 附录

Chapter 03

03竞争格局

3.1 竞争格局概览

截至 2026 年中,人形机器人竞争格局包括直接人形机器人同业、相邻移动操作平台、既有工业机器人厂商和替代自动化方案。在直接人形机器人竞争者中,Figure AI 在融资上领先($675M+),也是最早宣布具名 OEM 客户的公司(2024 年 1 月宣布 BMW 试点)。获得 OpenAI 支持、已融资约 $125M 的 1X Technologies,在工业 EVE 产品之外,将重心转向家庭辅助机器人(NEO)。Agility Robotics(融资 $250M,Amazon 战略投资)于 2023 年在 Amazon 仓库试点设施部署 Digit 机器人,是物流场景中最可信的真实世界人形机器人部署验证。Apptronik(融资 $160M,与 Google DeepMind 合作)开发了 Apollo,采用 NASA 传承设计,目标场景是制造和医疗健康。Boston Dynamics 于 2024 年推出 Atlas 商业版,从长期 R&D 姿态转向商业化,并有 Hyundai 约 $1.1B 收购作为后盾。Tesla Optimus 仍在 Tesla 内部部署,Musk 声称目标到 2025 年在 Tesla 工厂使用最多 1,000 台(时间表未经验证);如果对外商业化,将形成巨大的潜在规模优势。中国的 Unitree Robotics 以约 $16,000 提供 G1 人形机器人,在成本上挑战美国高端竞争者。 Sanctuary 处在中等融资梯队($160M),落后 Figure AI 约 $515M。既有工业机器人 OEM(ABB、KUKA、Fanuc、Yaskawa)尚未发布通用人形机器人平台,但它们拥有既有客户关系、服务基础设施和制造规模,可能成为市场 incumbents。Honda ASIMO(2022 年搁置)和 SoftBank Pepper(2021 年停产)是通用人形机器人商业化失败的历史尝试,为所有当前入局者提供了反向先例。 [CP001, CP002, CP003, CP004, CP005, CP006]

竞争对手画像表
公司类别累计融资(USD)目标细分市场核心产品差异点关键限制
Figure AI直接人形机器人竞争对手$675M+制造业(BMW 试点)Figure 01/02VC 支持强、BMW OEM 合作、Microsoft/OpenAI 投资方未披露商业收入;快速烧钱风险
1X Technologies直接人形机器人竞争对手~$125M家庭辅助 + 轻工业NEO(家庭)、EVE(工业)OpenAI 是战略投资方;已转向家庭市场规模较小;家庭市场未经验证;工业牵引有限
Agility Robotics直接人形机器人竞争对手~$250M物流仓储(Amazon 试点)DigitAmazon 投资 + 试点;双足且带机械臂Amazon 依赖风险;外部客户群有限;没有灵巧手
Apptronik直接人形机器人竞争对手~$160M制造业 + 医疗ApolloNASA 背景;Google DeepMind 合作;2024 年 9 月 $160M Series A仍处 Series A 阶段;融资少于 Figure;披露的商业规模有限
Tesla (Optimus)直接人形机器人竞争对手内部投入(估计 >$1B)内部使用(Tesla 工厂)Tesla Optimus Gen 2垂直整合(电机、电池、AI 芯片);若部署,规模巨大截至 2026 年不对外销售;若 Tesla 开放市场准入,将构成竞争威胁
Boston Dynamics (Atlas)人形机器人既有玩家 / 相邻玩家被 Hyundai 收购(~$1.1B)从工业研发走向商业化Atlas(电动版,2024)30+ 年机器人研发;品牌认知;Hyundai 母公司背书产品从研究转向商业;规模化部署未验证;成本高
Unitree Robotics中国 OEM / 低成本竞争对手未披露(VC 支持)研究 + 早期商业化G1 / H1 人形机器人低于 $20K 的价格点(G1 约 $16K);硬件迭代快;中国制造成本AI 复杂度弱于美国同业;出口管制风险;任务完成率未验证
Sanctuary AI标的公司~$160M制造业 + 物流Phoenix Gen 8 / Carbon AIMorgan Stanley IP 排名 #3;21-DoF 液压手;Microsoft/Accenture/Magna 生态未披露客户;联合创始人离职;相对 Figure AI 有 $515M 融资差距

融资数据来自公开新闻稿、CB Insights、PitchBook 和媒体报道。Tesla Optimus 资金估计计入 Tesla CAPEX,未单独披露。所有数据截至 2026 年 5 月,可能受未公告轮次影响。

[CP001, CP002, CP003, CP004, CP005, CP006]
FP001: 竞争定位图

基于证据支持的序数评分,在 AI 成熟度(x 轴)和硬件成熟度 / 部署规模(y 轴)上定位竞争对手。

坐标轴评分为序数(1-10),反映分析师基于截至 May 2026 公开披露的 AI 架构细节、部署公告和产品成熟度证据作出的判断。没有头对头基准数据可验证相对定位。

[CP001, CP002, CP003, CP004, CP005, CP006]

3.2 竞争者画像与差异化

Figure AI 于 2024 年 2 月以约 $2.6B 的 Series B 后投后估值融资 $675M+,具名投资方包括 Microsoft、OpenAI、NVIDIA 和 Jeff Bezos。2024 年 1 月宣布的 BMW 制造试点,是美国人形机器人初创公司中最具商业可信度的验证。Figure AI 用端到端神经网络方法控制机器人,这一点不同于 Sanctuary 的 Carbon AI 认知架构。 1X Technologies(前身为 Halodi Robotics)由 OpenAI 支持,融资约 $125M,其产品路线图转向家庭辅助机器人 NEO,同时保留工业 EVE 平台。家庭市场转向扩大了 1X 的目标可服务市场,但也反映近期工业部署经济性仍有不确定性。OpenAI 的战略投资为任务理解提供了前沿语言和视觉模型入口。 Agility Robotics 已融资约 $250M,包括 Amazon 领投的战略投资,并于 2023 年在 Amazon 仓库试点设施部署 Digit 机器人。Amazon 投资创造了战略锚定客户关系,但也可能限制 Agility 服务其他物流运营商。Digit 的形态(双足、有手臂、没有灵巧手)限制了任务多样性,弱于 Sanctuary 的 21-DoF 液压手。 Apptronik 在 2024 年 9 月完成 $160M Series A,Google 为战略投资方,并与 Google DeepMind 合作开展机器人学习。Apollo 瞄准制造(Houston 地区)和医疗健康应用,并承继了 Apptronik 过去用于宇航员服装机器人的 NASA 背景。 Boston Dynamics 于 2024 年推出 Atlas 商业版(电驱动,取代 2024 年 4 月停用的液压研究版)。Hyundai 的收购(约 $1.1B)提供制造支持和潜在汽车部署路径。Boston Dynamics 拥有 30+ 年 R&D 可信度,但尚未证明商业人形机器人能规模化部署。 Tesla Optimus Gen 2 已在 Tesla 工厂内部部署,Musk 称目标到 2025 年部署 1,000 台,并最终年产数百万台。Tesla 在电机、电池和 AI 芯片(Dojo 超级计算机)上的垂直整合,如果对外商业化,将形成独特成本结构。截至 2026 年 5 月,产品尚未对外销售。 中国 Unitree Robotics 提供约 $16,000 的 G1 人形机器人和约 $90,000 的 H1,相比美国同业具有 8 倍或更高成本优势。Unitree 的快速硬件迭代和中国制造成本基础在价格上挑战西方高端竞争者,不过 AI 复杂度和任务完成率仍落后美国领先者。 [CP009, CP010, CP011, CP012, CP013, CP014]

功能与能力矩阵
能力 / 标准Sanctuary AIFigure AI1X TechnologiesAgility RoboticsBoston DynamicsTesla OptimusUnitree
灵巧操作(手部 DoF)21-DoF 液压手(2024 年 12 月)未披露未披露有限(平行夹爪)先进(Atlas)未披露基础夹爪
双足运动已演示已演示已演示已演示(可上楼梯)同类最佳(长期研发积累)已演示已演示
自研 AI 认知平台Carbon AI(模块化认知架构)端到端神经网络集成 OpenAI未披露Boston Dynamics AI源自 Tesla FSD(推测)有限
商业客户部署未披露(2026 年 5 月)BMW 试点(2024 年 1 月)无规模化部署Amazon 试点(2023)商业版 Atlas(2024)仅内部使用研究用途销售
IP 强度(Morgan Stanley 排名)#3 全球(2025 年 2 月)未知 / 前 5(估计)UnknownUnknown强(历史专利)强(Tesla 专利)
企业分销伙伴Microsoft + Accenture + Magna未披露(Microsoft 投资方)OpenAIAmazon(战略)Hyundai(母公司)Tesla(内部)None
价格点(估计单价)~$100K-$150K未披露未披露~$150K+~$150K+(商业版)N/A(内部)~$16K-$90K
制造规模能力借助 Magna 合作未披露有限未披露Hyundai(母公司)Tesla(内部)自有工厂(中国)

矩阵单元格反映截至 2026 年 5 月的公开信息。标为“未披露”的单元格表示没有公开披露;标为“未知”的单元格表示已查找但未找到。所有价格估算来自分析师推导或媒体报道;没有厂商发布官方标价。

[CP020, CP021, CP022, CP023, CP026, CP027]
融资对比表
公司累计融资(USD)最近一轮领投方估值(报道)备注
Figure AI~$675MSeries B(~$675M,2024 年 2 月)Microsoft、OpenAI、NVIDIA、Bezos 等Series B 后约 $2.6B融资最快的人形机器人初创公司;未披露收入
Agility Robotics~$250MSeries B(2023)Amazon(战略领投)、DCVC未披露Amazon 仓库试点验证物流用例
Apptronik~$160MSeries A($160M,2024 年 9 月)Google(战略)、ARK Invest未披露Google DeepMind 合作;NASA 背景;总部在 Houston, TX
1X Technologies~$125MSeries B(2023,$100M)OpenAI、Sandwater、EQT Ventures约 $500M(估计)NEO 转向家庭机器人;商业规模小于 Figure/Agility
Sanctuary AI~$160MSeries B($100M,2024 年 6 月)Microsoft、Accenture、Magna、Bell、BDC、EDC、Workday、Evok 等战略投资方未披露Morgan Stanley IP 排名 #3;未披露收入;联合创始人 2024 年 11 月离职
Unitree Robotics未披露(中国 VC 支持)未披露未披露的中国 VC未披露$16K 的 G1 价格在成本上挑战美国既有玩家;有出口管制风险
Boston DynamicsHyundai 以 ~$1.1B 收购(2021)N/A(子公司)Hyundai 母公司Hyundai 子公司2024 年推出 Atlas 电动人形机器人;商业产品定价未知
Tesla (Optimus)内部投入(估计 >$1B)N/A(Tesla CAPEX)Tesla 公司N/A非独立融资;若 2026 年后对外销售,将构成竞争威胁

所有融资数据来自 CB Insights、PitchBook、公开新闻稿和媒体报道,截至 2026 年 5 月。Tesla Optimus 资金没有单独追踪。标为“未披露”的估值表示没有公开一级披露;估计估值基于媒体报道和分析师估算。

[CP009, CP010, CP011, CP012, CP013, CP014]
FP002: 功能广度与能力图谱

主要竞争对手在灵巧操作、移动、AI、客户、合作伙伴和制造规模上的能力覆盖。

标注为未披露的矩阵单元表示缺少公开披露,不等于确认没有该能力。比较基于截至 May 2026 的公开公告、产品演示和媒体报道。

[CP020, CP021, CP022, CP023, CP026, CP027]

3.3 Sanctuary AI 差异化与竞争位置

Sanctuary AI 的竞争差异化建立在三根支柱上:Carbon AI 认知架构、21-DoF 液压灵巧手,以及 Morgan Stanley IP 排名(截至 2025 年 2 月全球 #3)。Carbon AI 不同于端到端神经网络方法(Figure AI、Tesla FSD 衍生路线):它采用模块化认知架构,模仿人脑子系统(记忆、视觉、听觉、触觉),Sanctuary 称这能带来更可泛化的任务学习。但这种架构差异尚未通过头对头基准比较验证。 2024 年 12 月 12 日演示的 21-DoF 液压灵巧手,在当前人形机器人竞争者中具有技术辨识度——多数同业(Figure AI、Apptronik、Tesla)采用电驱动,且尚未发布同等级 DoF 规格。液压驱动可提供高力重比,适合需要精密握力的制造任务,但相比电驱替代方案,也增加了系统复杂度、维护要求和潜在可靠性风险。 合作生态——Microsoft(Azure AI 基础设施、Azure OpenAI 合作)、Accenture(企业部署服务)和 Magna International(汽车级制造)——提供了多数人形机器人同业缺少的分发和制造使能。如果合作能转化为客户引荐和部署交易,这个生态将成为有意义的 go-to-market 差异化。 关键竞争弱点:Sanctuary 未披露任何具名商业客户或试点部署,而 Figure AI(BMW)、Agility Robotics(Amazon)和 Apptronik(据报道为 Mercedes-Benz)都有具名制造或物流锚点。相对 Figure AI 的 $515M 融资缺口限制了 R&D 开支和制造规模。2024 年 11 月联合创始人离任(Suzanne Gildert,CSO)构成技术领导风险。公司集中在加拿大 / 北美,也限制了全球市场覆盖,弱于拥有日本、欧洲或韩国制造关系的同业。 [CP020, CP021, CP022, CP023, CP025, CP032]

护城河耐久性与竞争风险登记表
护城河主张威胁严重性缓解措施 / 尽调问题
Carbon AI 认知架构(具身 AI 差异化)API 时代:基础模型(GPT-5、Gemini)可能让 AI 控制层商品化;源自 Tesla FSD 的路线正在收敛到类似结果判断 Carbon 能否在无需大规模采集数据的情况下为新任务重新训练;索取与竞争路线对比的基准;评估对 Microsoft Azure AI 集成的依赖
21-DoF 液压灵巧手(技术 IP)电动执行器替代方案(Figure AI、Apptronik)可能以更低成本和重量做到类似 DoF;液压会增加维护复杂度在工业场景验证液压执行器优势;评估 21-DoF 设计的专利覆盖;对标竞品手部设计
Morgan Stanley #3 IP 排名(专利组合)IP 排名是滞后指标;Figure AI 和 Tesla 正在积极申请专利;排名可能在 12-18 个月内变化要求提供专利组合分析:按子系统(移动、操作、AI)划分的覆盖、关键待审专利,以及竞争对手申请速度
Microsoft + Accenture 分销生态分销依赖合作伙伴;Microsoft 和 Accenture 并非独家,也可能同时与 Figure AI 或其他竞争对手合作确认 Microsoft 和 Accenture 合作是否包含任何排他条款;评估来自这些伙伴的管线归因相对于 Sanctuary 直接销售的占比
Magna International 制造合作Magna 是合同制造商;如果 Sanctuary 商业化停滞,Magna 关系可能无法放量,也可能转向其他客户评估 Magna 合作条款:产量承诺、单位经济、排他条款,以及产能扩张的里程碑触发条件
Vancouver/Canada 人才池和研发成本优势加元优势可能削弱;美国竞争对手能以类似或更低成本吸引全球人才;相较美国同行的资金约束带来关键研究人员留存风险确认 Sanctuary 员工构成,以及在相对 Figure AI 存在 $515M 融资差距时留住关键机器人 / AI 研究人员的能力

威胁评估基于截至 2026 年 5 月可公开观察的竞争动态。严重程度评级是定性判断,参考融资差距、市场动态和先例分析。Sanctuary 未披露合作协议中存在任何合同排他性。

[CP024, CP035, CP036, CP037, CP038, CP039]
FP003: 竞争护城河与就绪度 KPI

Sanctuary AI 相对于人形机器人领域的关键竞争地位和护城河指标。

所有数值来自截至 May 2026 的公开来源。Sanctuary 单价估计为 ~$125K,是分析师推导的 $100K-$150K 区间中点;公司未披露官方定价。

[CP020, CP021, CP022, CP023, CP024, CP025]

3.4 护城河耐久性与失败先例

Sanctuary 的 IP 位置(截至 2025 年 2 月 Morgan Stanley 排名 #3)显示其相较同业有一定专利覆盖,但 IP 排名是滞后指标——资金充足的竞争者,包括 Figure AI 和 Tesla,很可能也在激进申请专利。如果认知系统中的架构选择形成路径依赖,Carbon AI 平台的模块化可能提供持久差异化;但 API 时代风险真实存在:随着前沿基础模型(GPT-5、Gemini、Claude)通过 VLA(vision-language-action)API 具备机器人控制能力,自研认知架构的价值可能被压缩。如果 Azure OpenAI 集成成为标准部署路径,Microsoft 合作可能部分隔离 Sanctuary 的这一风险。 Magna International 制造合作对生产规模具有战略意义,但其耐久性取决于 Sanctuary 能否达到触发批量承诺的商业里程碑。该合作看起来并非排他,Magna 也可能把类似安排扩展给其他人形机器人厂商。Microsoft 和 Accenture 分发合作同样不是排他的。 反向先例提供了冷静的历史背景。Honda ASIMO 是一个持续 22 年的 R&D 项目(2000-2022),耗费数十亿美元研发资金,却未能实现规模化商业部署;Honda 找不到可行商业路径后将其搁置。SoftBank Pepper 于 2014 年发布,单价约 $2,000,在 2021 年停产前销量约 27,000 台,原因是商业采用不佳。两次失败共享同一种失败模式:在真实世界价格和真实世界运行条件下,任务效用不足——这正是包括 Sanctuary 在内所有当前人形机器人初创公司面对的挑战。Unitree 的低于 $20K 的 G1 价位从低端带来商品化压力,而基础模型 API 则从高端带来潜在替代。 [CP024, CP035, CP036, CP037, CP038, CP039]

3.5 附录

Chapter 04

04财务情况

4.1 收入来源与变现框架

截至 2026 年 5 月,Sanctuary AI 没有公开披露收入。公司曾概述 Robotics-as-a-Service(RaaS)收入模型,这与行业同业(Agility Robotics、Figure AI)一致:硬件部署在客户设施中,按使用量或订阅收费,而不是一次性售出。相比硬件直接销售,这种模式会推迟收入确认,但能提高客户终身价值,也通过降低前期资本要求来支持企业采用。 Sanctuary 估计硬件 ASP 为每台 $100,000-$150,000,这一估计由分析师根据可比人形机器人价格推导,并未得到公司确认。按 $5-$8 每机器人小时的 RaaS 混合费率(与每小时 $25-$35 加福利的人类工人相比,符合劳动力套利经济性),在双班制、50% 利用率下,每台机器人年收入约为 $44,000-$70,000。这意味着,只有当 Sanctuary 在结构化制造或物流环境中实现高利用率部署,单位经济模型才有实质意义。 次级收入来源包括:(1)通过 Carbon AI 平台向第三方硬件应用授权 AI 软件(未公开确认);(2)通过 Accenture 合作,为企业部署和集成获取专业服务收入(结构上可能,但未披露);(3)由 Export Development Canada(EDC)关系和 BDC Capital 参与支持的潜在政府和国防合同。上述收入来源均未获得商业披露确认。 Sanctuary 当前收入阶段是商业化前或早期商业化。公司 Series B 投资方名单(Microsoft、Accenture、Magna、Workday、Bell、BDC、EDC、Evok)显示其与收入路径存在战略匹配,但并不确认公司已经产生收入。战略投资方常常为了期权价值和合作入口而投资,而不是因为商业收入即将到来。 [CI001, CI002, CI003, CI004, CI005, CI006]

收入流表
收入流模式类型状态(2026 年 5 月)关键依赖预计年收入潜力(每单位 / 客户)
硬件销售(Phoenix 机器人)一次性资本销售商业化前 / 未确认商业客户获取每台 ~$100K-$150K
Robotics-as-a-Service(RaaS)按使用量或订阅商业化前 / 已提出但未确认试点部署成功并签下客户合同每机器人年 ~$44K-$70K(50% 利用率,$5-8/hr)
Carbon AI 平台授权SaaS / API 授权未公开确认Carbon AI 产品化和第三方硬件伙伴未披露 / 推测
专业服务(Accenture 联合销售)集成和部署服务未公开确认Accenture 合作激活为企业部署未披露
政府和国防合同合同研发和部署可能(BDC/EDC 信号)政府采购流程;安全许可要求未披露
AI 模型训练数据授权数据授权未公开确认足够规模的专有遥操作数据资产未披露 / 推测

所有收入估算均由分析师根据 RaaS 行业基准和可比硬件定价建模得出。截至 2026 年 5 月,Sanctuary AI 未披露任何收入、定价或已签合同。

[CI001, CI002, CI003, CI004, CI005, CI006]
定价和变现表
定价要素Sanctuary 估计区间同行对比来源依据置信度
硬件 ASP(每台)$100K-$150KAgility Digit ~$150K+;Unitree G1 $16K;Apptronik 未披露基于可比产品和媒体的分析师估算
RaaS 小时费率$5-$8/robot-hourAgility 假设 $6-$10;Tesla N/A;Unitree N/A劳动力替代经济模型(对比 $25-$35/hr 人类工人)
RaaS 年度合同(50% 利用率)每台机器人 $44K-$70K无公开同行披露按小时费率 × 8,760 小时 × 50% 正常运行时间建模
毛利率(硬件)40-60%工业机器人硬件同行:40-55%硬件机器人行业基准
Carbon AI 授权(假设)未披露N/A — 未确认存在同行产品推测;Sanctuary 未确认 AI 平台授权产品很低
$125K ASP 对比 $45K/年工人的回收期~5-8 年(不含维护)企业采用机器人的经济门槛劳动力替代模型;不含集成和维护

所有 Sanctuary AI 价格估算均由分析师建模。Sanctuary AI 未披露官方定价,也没有任何公开来源确认。同行定价参考来自媒体和分析师估算。

[CI007, CI008, CI009, CI010]
FI001: 收入模型桥接

Sanctuary AI 在 2018 至 2027 年间的融资事件、产品里程碑和预计收入拐点时间线。

2022 年 Series A 与 2018 年创立时间来自媒体和 LinkedIn,均为近似值。预计现金跑道期限为分析师建模结果,Sanctuary AI 未确认。收入拐点是有条件的预测。

[CI016, CI017, CI018, CI019, CI020, CI021]

4.2 成本结构与单位经济模型

Sanctuary AI 的成本结构主要由 R&D 支出(AI 与机器人工程师人力)、硬件 COGS,以及企业销售和合作伙伴管理中的 SG&A 构成。作为一家估计拥有 200-300 人组织规模的 Series B 机器人公司,按可比阶段机器人同业和已披露 BC 薪资数据估算,年度运营费用约为 $60M-$90M。这个估计没有任何公开 Sanctuary 财务披露确认。 机器人制造 COGS 包括材料(执行器、传感器、处理器、结构件)、Magna 合同制造人工和质量保证。按估计硬件 ASP $125K、行业典型机器人毛利率(硬件 40-60%)计算,隐含每台 COGS 为 $50,000-$75,000。这些数字完全来自模型;Sanctuary 尚未披露单位经济模型。 R&D 强度预计较高(估计占总 OpEx 的 50-60%),与商业化前阶段可比人形机器人初创公司一致。Carbon AI 的 AI 训练管线——包括通过人类遥操作收集数据、GPU 计算基础设施和模型迭代周期——构成了显著经常性成本,其中一部分由 Microsoft Azure AI 合作补贴。 劳动力成本套利是单位经济模型的关键论点:如果一台 ASP $125K 的 Phoenix 人形机器人(或 RaaS 模式下 $5-$8/hour)能够替代一名 $45,000/year 的仓库或制造工人,并在 5-8 年回本,价值主张可能有吸引力。但这个分析忽略了维护成本、正常运行时间波动、集成成本和软件升级要求——这些因素都会把真实总拥有成本推高到朴素硬件成本比较之上。 [CI009, CI010, CI011, CI012, CI013, CI014]

单位经济表
成本项目低位估计(USD)高位估计(USD)依据置信度
每台硬件 COGS$50K$75K基于 $125K ASP 的 40-60% 毛利率假设;行业硬件机器人可比公司
年度研发 OpEx$35M$55M占 $60M-$90M 总 OpEx 的 50-60%;同阶段机器人创业公司
年度 SG&A OpEx$15M$25M占总 OpEx 的 25-30%;企业销售 + 合作伙伴管理
年度总 OpEx(烧钱)$60M$90MSeries B 机器人公司,估计 200-300 名员工,按 Vancouver 人工成本;可比同行
GPU/云计算(AI 训练)$3M$8M部分由 Microsoft Azure 合作补贴;可比 AI 公司计算成本
Series B 后估计 runway(2024 年 6 月)24 个月(高烧钱)36 个月(低烧钱)累计融资 $160M,扣除 B 轮前部署资金;$60M-$90M/年烧钱率

所有估算均为分析师建模,置信度低。Sanctuary AI 未披露任何财务数据。Runway 计算假设 Series B 交割时,在早前轮次支出后仍有约 $90M-$100M 未部署资本。

[CI011, CI012, CI013, CI014]

4.3 公开财务牵引信号

Sanctuary AI 的公开牵引信号仅限于融资历史和合作公告。2024 年 6 月完成的 $100M Series B,战略投资方包括 Microsoft、Accenture、Magna International、Workday、Bell、BDC Capital、EDC 和 Evok Innovations。Series A($30M)于 2022 年完成,在此之前还有一轮金额未披露的种子轮。已披露总融资约为 $160M,不过更早的 Pre-Series A 融资仍不清楚。 合作质量可作为商业牵引的代理指标。Microsoft 的 Azure AI 投资和共同开发关系,暗示商业试点管线可能经由 Azure marketplace 或 Accenture 咨询实践导入。Accenture 的技术伙伴投资,暗示 Sanctuary 机器人至少进入了 Accenture 的企业产品目录,尽管尚未公开披露由 Accenture 促成的客户部署。Magna International 的投资暗示制造产能承诺与预期生产规模相匹配,而不仅仅是供货协议。 员工数轨迹是间接牵引信号。LinkedIn 数据显示,Sanctuary AI 已从 2022 年 Series A 完成时的小团队增长到约 200-300 名员工(精确人数未经验证),符合资本部署模式。2024 年 11 月联合创始人离任(Suzanne Gildert,联合创始人兼 Chief Science Officer)是 Sanctuary 公司历史中披露的最重要负面事件,代表了技术 AI 路线图上的关键人风险。 外部技术验证信号:2024 年 12 月 21-DoF 液压手演示,以及 Morgan Stanley IP 排名(截至 2025 年 2 月全球 #3),是技术进展最强的独立验证代理。公司尚未公开披露收入、合同金额或试点客户指标。 [CI015, CI016, CI017, CI018, CI019, CI020]

资本充足性表
融资事件日期金额(USD)关键投资者备注
种子轮 / Pre-Series A2018-2019(估计)未披露未披露公司成立于 2018 年;早期 pre-seed 和种子轮未披露
Series A2022~$30M(估计)未披露领投方;BDC Capital、Evok 等首轮机构增长融资;支持 Phoenix Gen 1-5 开发
Series B2024 年 6 月$100MMicrosoft、Accenture、Magna、Workday、Bell、BDC、EDC、Evok 等战略投资方单轮最大融资;战略投资者阵容;未披露估值
已披露资本总额2026 年 5 月~$160M(估计)多家战略和机构投资者包括 Series A 估计和 Series B;不含未披露早期轮次
估计剩余 runway2026 年 5 月Series B 交割后 24-36 个月N/A分析师按 $75M 中性情景烧钱率估计;未确认
隐含 Series C 需求2026-2027(估计)$150M-$300M(估计)需要新的机构 / 战略投资者需要把 runway 延长到 2027 年之后,并为制造规模化提供资金

除 2024 年 6 月 Series B(GlobeNewswire 新闻稿称 $100M)外,所有融资金额均为分析师估计。Sanctuary AI 未披露公司估值。Series C 需求为模型推算,非公司披露。

[CI015, CI016, CI017, CI018, CI019, CI020]
FI004: 财务进展与里程碑状态矩阵

对 Sanctuary AI 财务与商业进展里程碑的结构化评估,展示状态、证据质量和投资者相关性。

矩阵仅基于公开可得的进展信号和分析师判断。所有状态评估均基于公开披露的有无。私下商业活动可能与本评估有重大差异。

[CI015, CI018, CI019, CI020, CI021, CI024]

4.4 资本充足性与财务判断

按已披露资金和同阶段公司的烧钱速度估算,Sanctuary AI 在 2024 年 6 月 Series B 轮交割后 还有 24-36 个月现金跑道,资本窗口大致延伸到 2026 年中至 2027 年中。该估算假设年烧钱 $60M-$90M(中位情景 $75M),且短期内没有显著收入贡献。若进入更高烧钱情景($90M+), 现金跑道将缩短至 Series B 轮交割后约 24 个月(2026 年中),公司必须加快收入落地,或 推进 Series C 融资。 对机构投资者而言,财务结论高度依赖后续验证。Sanctuary AI 面临几项财务风险:(1) 未公开披露收入, 商业模式没有任何财务验证;(2) 相比 Figure AI 存在 $515M 融资缺口,限制竞争规模;(3) 联合创始人离任, 带来关键人物和 IP 连续性风险;(4) 与 Microsoft、Accenture、Magna 的合作均非排他,且没有合同性收入承诺; (5) Unitree 从低端切入,压缩入门级人形机器人应用的潜在定价权。 可能改写结论的财务机会包括:(1) 公布具名商业客户,或披露带有生产率指标的试点,可大幅降低商业模式风险; (2) Series C 融资若超过 $200M,可延长现金跑道,并传递投资人信心仍在的信号;(3) 与非机器人企业达成 Carbon AI 平台授权,可证明软件收入能多元化;(4) 借助 BDC/EDC 关系拿下政府或国防合同,可带来非周期性收入。 截至 2026 年 5 月,上述事项均未公开披露。 相比同业,Sanctuary 的资本效率仍不确定。Sanctuary 资本基础为 $160M,低于 Agility 的 $250M 和 Figure AI 的 $675M,这可能说明资本效率更高,也可能意味着关键 R&D 和销售职能投入不足。Morgan Stanley 将其 IP 排名列为第 #3,以及 21-DoF 灵巧手演示,都显示 R&D 投入有产出,但商业转化尚未证明。 [CI023, CI024, CI025, CI026, CI027, CI028]

公开财务缺口表
数据点状态重要性尽调路径
收入(任何金额)未披露商业模式的基础商业验证数据室访问;客户访谈;试点部署披露
烧钱率 / 年度 OpEx未披露资本充足性和 runway 评估数据室 P&L;可比同行基准
单位经济(COGS、毛利率)未披露RaaS 定价可行性和硬件经济可持续性数据室 BOM + 制造成本表
客户名称和合同金额未披露商业牵引和客户集中风险客户访谈;数据室中的 LOI 或合同披露
公司估值未披露(Series B 估值未公开)股权投资回报分析数据室 cap table 和 waterfall;审阅 Series B term sheet
员工数和地点拆分约 200-300 人估计(媒体 / LinkedIn)校准烧钱率,以及研发与 SG&A 分配数据室 HR 数据和组织架构图
政府合同状态(BDC/EDC)未确认非周期性收入多元化潜力政府合同登记;Sanctuary 管理层通话

本表所有项目反映截至 2026 年 5 月公开披露缺失。Sanctuary AI 是加拿大私营公司,不受强制公开财务报告约束。所有数据都需要通过投资者数据室访问或管理层直接披露获得。

[CI025, CI026, CI027, CI028, CI029, CI030]
FI002: 财务估算区间

分析师对 Sanctuary AI 关键财务指标的估算区间;缺少一手财务披露,所有项置信度都低。

所有区间均由分析师推导,置信度低。Sanctuary AI 未公开披露任何财务数据。区间基于可比阶段机器人和 AI 硬件公司。

[CI007, CI009, CI011, CI013, CI023, CI002]
FI003: 资本密集度与现金流图谱

汇总 Sanctuary AI 财务评估所需的公开财务事实、结构性财务信号和关键尽调缺口。

员工数估计来自 LinkedIn,未经验证。$0 收入反映没有任何公开披露;Sanctuary 可能存在未披露的试点收入。Series C 需求为分析师建模结果。

[CI015, CI019, CI021, CI022, CI024, CI025]

4.5 证据展示

Chapter 05

05产品与技术

5.1 产品概览——Phoenix 机器人与市场定位

Sanctuary AI 的核心商业产品是 Phoenix,一款通用人形机器人,面向仓库、制造设施,以及未来可能进入的零售等由人类搭建的工作环境。 Phoenix 搭配 Carbon,这是 Sanctuary 自研的 AI 认知控制系统。公司称 Carbon 模仿人脑子系统,覆盖记忆、视觉、听觉和触觉。 Sanctuary 将这套组合包装成垂直整合系统:从定制电机设计、液压驱动,到自研感知、规划和生成式 AI 层,全部自成一体; 这也是它相对常用第三方执行器和现成 AI 框架的竞争对手所强调的差异点。 截至 2026 年 5 月,Sanctuary 已连续推出八代 Phoenix,其中 Gen 7 于 2024 年 4 月 25 日发布,Gen 8 于 2024 年 12 月 16 日发布。公司技术页面明确称:“from motor design to generative AI algorithms for superhuman task performance, our solutions are designed, built, owned, and patented by Sanctuary.” 这把 Sanctuary 定位为全栈机器人公司, 而不是系统集成商;这意味着 R&D 投入更高,但 IP 控制更强,若能规模化商业化,也可能拥有更宽利润率。 GTM 策略是 B2B,瞄准需要在物流、汽车制造、食品加工等劳动密集型场景中自动化结构化手工任务的工业运营商。截至 2026 年 5 月运行日, 除 2024 年 4 月宣布、后来未再确认状态的 Magna International 制造试点外,公司没有公开具名客户、试点场地、出货量披露或商业牵引指标。 [CE001, CE002, CE003, CE004, CE005, CE006]

产品模块和能力矩阵
模块 / 组件主要用户 / 买方成熟度 / 状态关键差异化尽调缺口
Phoenix Gen 8 机器人(硬件)工业运营方:仓储、制造、物流商业化前 — 仅演示;未披露量产单位或客户部署21-DoF 液压手;人形形态;垂直整合执行器未公开规格(身高、重量、有效载荷、续航);生产就绪度未披露
Carbon AI(认知软件)机器人运营方;通过 RaaS 触达企业客户商业化前 — 已在受控环境演示;无第三方基准仿生脑启发架构;多模态(记忆、视觉、听觉、触觉)架构未公开记录;无第三方验证或基准
21-DoF 液压手灵巧操作用例:装配、包装、零件搬运演示 — 2024 年 12 月首次公开演示;2025 年 5 月零样本立方体重定向公开演示的人形手中 DoF 最高;液压顺应性未披露可靠性数据、MTBF 或现场耐久性指标
触觉传感器系统需要滑移检测的抓取和手内操作任务演示 — 2024 年 12 月发布;2025 年 2 月集成更新自研触觉传感与 Carbon AI 集成,支持闭环操作未披露技术规格和传感器密度;无对比基准
Sim-to-Real 转移管线AI 工程师、机器人运营团队已演示 — 2025 年 3 月公开演示灵巧策略转移在仿真中训练的策略可直接部署到硬件,无需手工微调未发布仿真保真度、转移差距指标和任务成功率
Carbon AI 训练管线(遥操作 + RL)AI 训练团队;通过数据收集合作触达客户商业化前 — 仅内部使用;无 API 或第三方访问人类遥操作用于模仿 + RL 用于优化 + sim-to-real 用于部署未披露训练数据量、每项任务标注成本和样本效率

成熟度评级根据新闻稿、新闻报道和产品演示视频推断。截至 2026 年 5 月,尚无公开确认的商业部署。所有差异化主张均来自公司表述或分析师推断;不存在第三方基准。

[CE001, CE004, CE007, CE009, CE010, CE011]
工作流和用例表
用户工作 / 任务现有 / 传统工作流Sanctuary 方案声称 / 推断收益限制
仓库拣选和包装(非结构化 SKU)人类工人;高流失率;重复性劳损;$18-25/hr 人工成本Phoenix 搭载 Carbon AI,靠遥操作训练策略执行拣选、分拣、包装任务降低人工成本;连续运行;降低受伤风险无公开部署数据;未披露任务错误率和换线时间
制造装配(汽车 / 消费品)具备训练灵巧度的人类装配工;周期时间约 30-60 sec/cyclePhoenix 液压手执行通过遥操作和 RL 训练的装配任务可在多类任务间重新部署,无需固定工装未确认具名制造部署;Magna 试点状态不清楚
手内操作(灵巧抓取)刚性夹爪自动化(平行夹爪、真空);局限于结构化抓取点21-DoF 液压手结合触觉反馈,适配抓取和重定向已演示零样本重定向(2025 年 5 月);液压顺应性适合易碎物品反复循环下的现场可靠性未发布;在多样物体上的性能未经验证
Sim-to-real 训练部署数月机器人本体数据采集;机器人操作人工成本高在仿真中训练,无需手工微调即可部署到硬件降低新任务训练成本和时间;提高策略开发安全性未发布转移差距指标,以及相对纯仿真基线的任务成功率
工业 AI 演示 / 渠道营销实地参观、静态演示设备Hannover Messe 与 Microsoft 现场演示(2025 年 4 月);YouTube 演示频道建立客户管线;借 Microsoft 验证合作关系未披露销售管线、转化数据或演示后的部署承诺

收益根据产品公告和可比行业基准由分析师推断。Sanctuary 未公开披露客户案例、生产率数据或部署结果。限制反映证据缺失,而非已确认失败。

[CE001, CE006, CE009, CE010, CE013, CE014]
FE002: Phoenix 机器人部署工作流

端到端 Phoenix 机器人部署流程从任务识别、训练管线执行,推进到策略验证、现场部署和遥测反馈。Carbon AI 贯穿每个阶段,把客户任务定义转化为训练好的机器人策略,并在 Phoenix 硬件上自主执行。

工作流根据公开产品描述、训练管线表述和演示证据推断。Sanctuary 未公开披露确切步骤边界、时间线和数据量。

[CE001, CE006, CE013, CE014, CE015, CE019]

5.2 硬件架构——Gen 7、Gen 8、液压手与传感

Phoenix 采用拟人形态:双臂、双腿、躯干和头部,比例接近成年工人。公司发布过视频演示,但没有披露任何一代产品的详细规格, 包括身高、体重、载荷、电池续航或最高速度。相比 Unitree、Agility、Apptronik 等经常发布数据表的电驱竞争对手, 这种缺少公开硬件规格的做法并不常见。 Sanctuary 最具技术辨识度的硬件发布,是 2024 年 12 月 12 日推出的 21 自由度(DoF)液压灵巧手。21 DoF 让这只手达到或超过人手灵活度。 Sanctuary 选择液压驱动,而不是多数竞争对手采用的电伺服或腱驱设计,反映了明确的架构取舍:液压系统顺应性更高、力控更好, 且可提供没有齿槽效应的连续扭矩,从而支撑 2025 年 5 月零样本立方体重定向视频中的精细手内操作。 Sanctuary 还在 2024 年 12 月 19 日发布了新型触觉传感器技术,并在 2025 年 2 月 26 日发布集成更新。触觉感知是多数人形机器人平台的关键短板: 没有触觉,机器人就难以稳定抓取多样物体或检测打滑。2025 年 3 月 18 日的 sim-to-real 迁移演示显示,在仿真中训练的策略可以直接部署到实体硬件上, 这是训练管线效率的重要验证。触觉传感器如何与 Carbon AI 集成、sim-to-real 迁移精度如何量化,均未公开披露技术细节。 [CE007, CE008, CE009, CE010, CE011, CE012]

产品路线图与发布时间线
日期里程碑类型状态影响 / 尽调备注
April 25, 2024Phoenix Gen 7 亮相产品发布已发布第七代硬件平台;仅有公开视频演示;规格未披露
May 1, 2024宣布与 Microsoft 开展 AI 模型合作合作活跃(推断)AI 训练合作;可能依赖 Azure 基础设施;范围未披露
December 12, 2024演示 21 自由度液压手产品里程碑演示阶段首次公开演示灵巧液压手;尚无现场部署确认
December 16, 2024Phoenix Gen 8 亮相产品发布已发布第八代平台;公开资料未细述相较 Gen 7 的改进
December 19, 2024触觉传感器技术亮相产品里程碑演示阶段新的自研触觉感知系统;宣布与 Carbon AI 集成
February 26, 2025触觉传感器集成更新产品里程碑活跃公布集成进展;技术细节仍未公开
March 18, 2025仿真到现实迁移演示产品里程碑已演示在仿真中训练出的灵巧策略已成功部署到物理硬件
April 1, 2025与 Microsoft 共同亮相 Hannover Messe市场拓展已完成工业展会;Microsoft 同场亮相,释放企业渠道意图
May 15, 2025零样本手内操作演示产品里程碑已演示无需针对任务微调,自主重定向带字母方块
未披露Phoenix Gen 9 或后续硬件产品路线图未宣布未公开披露代际节奏、规格目标或发布时间线

表格条目来自新闻稿、归档新闻页面和第三方报道。每个里程碑公告后的跟进状态,是根据缺乏更新推断;任何里程碑均未发布量产、交付或取消公告。

[CE002, CE003, CE009, CE011, CE012, CE013]

5.3 软件平台——Carbon AI、训练管线与泛化

Carbon AI 是 Sanctuary 自研的认知控制平台。公司称它受大脑启发,整合记忆、视觉、听觉和触觉等子系统,类似人脑皮层区域。 与基于 transformer 的视觉运动策略不同(例如 Figure AI 与 OpenAI 的路线,或 1X Technologies 自研 LLM 的路线), Carbon AI 架构没有公开技术细节。仿生叙事意味着它可能是模块化系统,由专门模块处理不同感知和规划领域;但具体神经网络架构、 推理栈和训练数据规模都是专有信息。 训练管线结合三种机制:(1) 遥操作数据采集,人类操作员穿戴动作捕捉或触觉外骨骼系统演示任务,生成模仿学习数据集;(2) 强化学习, 智能体在仿真中根据奖励信号优化;(3) sim-to-real 迁移,将仿真中训练的策略部署到实体硬件。2025 年 3 月的 sim-to-real 演示验证了 这条管线能在无需实体机器人手工微调的情况下迁移灵巧操作策略,是能力成熟度的重要信号。 2025 年 5 月的零样本手内操作演示显示,Phoenix 液压手在没有任务专项训练的情况下,可自主重定向带字母的立方体。零样本泛化是操作系统最难的测试, 因为策略必须在没见过具体任务的情况下推理物体几何和接触关系。若该演示能在多样条件下复现,它就是 Sanctuary 公开技术组合中最强信号之一。 但这次演示使用的是受控单物体场景;在非结构化环境中面对多样真实物体的表现,尚未公开基准测试。 2024 年 5 月 1 日宣布的 Microsoft 合作,重点是为通用机器人开发 AI 模型。该合作很可能借助 Azure OpenAI 和 Azure ML 基础设施训练 Carbon AI 模型,但具体使用哪些 Azure 服务尚未确认。2025 年 4 月,Sanctuary 与 Microsoft 一同在 Hannover Messe 展示 Phoenix, 释放出通过 Microsoft 企业渠道争取工业自动化客户的信号。 [CE004, CE015, CE016, CE017, CE018, CE019]

技术和运营架构
层 / 组件角色关键依赖风险
液压执行器系统(手 + 关节)为灵巧操作和移动提供力与顺应性自定义液压组件;流体供应;密封件制造商(供应商未披露)比电动更重;维护复杂;敏感环境存在流体泄漏风险
触觉传感器阵列为抓取适配和滑移检测提供闭环反馈自研传感器设计(专有);与 Carbon AI 运动皮层类比模块集成技术规格未公开;传感器膜片现场耐久性未知
Carbon AI — 感知栈(视觉、听觉)多模态感知:视觉场景理解、音频线索摄像头和麦克风硬件;推理可能使用 Azure Vision 或类似云 API实时任务中的云推理延迟;依赖摄像头校准质量
Carbon AI — 操作 / 运动皮层类比模块策略执行:生成关节扭矩以达成操作目标GPU 计算(可能为 Azure GPU 实例或端侧);sim-to-real 转移准确性架构封闭;无公开基准;到真实环境的转移差距未量化
Carbon AI — 记忆 / 上下文系统任务上下文保留;多步骤指令跟随;情节记忆训练数据量;标注质量;可能使用 LLM backbone(未披露)未披露记忆容量上限;长期任务中的遗忘行为未刻画
Microsoft Azure(推断的云依赖)AI 模型训练基础设施;可能为 Azure ML + Azure OpenAI;Hannover Messe 演示Microsoft 合作(2024 年 5 月);Azure 定价和服务可用性伙伴集中风险;API 弃用或定价变化可能影响训练成本
机器人硬件制造(Magna 合作)Phoenix 机器人机架和总成的量产制造Magna International(2024 年 4 月宣布;截至 2026 年 5 月状态不清楚)制造就绪度未确认;未报告 Magna 正在生产 Phoenix 单位

架构根据产品描述、合作关系和技术演示推断。云服务商、硬件供应商和内部软件栈细节均未公开确认。依赖风险是合理推断,不是已确认失败。

[CE004, CE009, CE010, CE015, CE016, CE022]
FE001: Carbon AI 平台架构栈

Sanctuary AI 的产品架构是一套四层垂直栈:(1)物理硬件层,包括 Phoenix 机器人机身和自研液压 / 执行器系统;(2)感知层,纳入视觉、音频和触觉模态;(3)Carbon AI 认知软件层,包含记忆、感知、规划和控制等专用模块;(4)AI 训练基础设施层,承载 sim-to-real、遥操作和强化学习管线。

层级结构根据产品描述、合作公告和技术演示推断。内部模块名称和接口属于 Sanctuary 自有信息,未公开记录。Azure 依赖基于 Microsoft 合作关系推断;未确认其为独家云服务商。

[CE004, CE015, CE016, CE017, CE020, CE021]

5.4 技术依赖、风险与 IP 格局

Sanctuary 最大的技术依赖是 Microsoft:AI 模型开发合作、潜在 Azure 基础设施依赖,以及 Hannover Messe 联合露面,都指向它与 Microsoft AI 生态的深度整合。如果 Sanctuary 的 Carbon AI 训练管线与 Azure ML 或 Azure OpenAI 紧密绑定,定价变化、API 废弃, 或合作关系恶化,都可能实质影响 AI 开发路线图。没有公开来源确认 Sanctuary 是否具备多云冗余。 相比电驱竞争对手,21-DoF 灵巧手以及更广泛机器人设计中可能采用的液压驱动,会引入运营风险。液压系统需要定期更换流体、维护密封件, 在现场环境中也容易泄漏。在食品加工、洁净室或敏感材料场景中,液压油污染是足以排除部署的安全与合规风险。相较无刷电机, 液压系统的重量和复杂度劣势也会影响载荷与续航经济性。Sanctuary 是否计划未来推出纯电版本或混合设计,尚未披露。 IP 方面,Morgan Stanley 在 2024 年 11 月将 Sanctuary 人形机器人专利排名列为全球第四,2025 年 2 月列为第三,反映其在电机设计、 操作算法和传感器集成上持续提交专利。公司在 patents.google.com 上以 Sanctuary Cognitive Systems Corporation 检索可见结果, 确认其仍在积极推进专利申请;但具体专利号、权利要求范围、申请司法辖区和授权状态都没有公开逐项列明。2024 年 4 月宣布、且进入 Series B 投资人名单的 Magna 制造合作,理论上可提供规模化制造能力;但截至 2026 年 5 月,没有公开更新说明 Magna 的生产角色, 也没有披露任何由 Magna 制造的 Phoenix 机组。 Sanctuary 的 GitHub 组织存在,但公开代码活动很少,说明公司没有推进开源开发者策略。除产品公告讨论外,未发现 Hacker News 技术深度讨论、 Stack Overflow 标签、npm 或 PyPI 包,或开发者社区信号。这限制了外部对 Carbon AI 技术主张的独立验证,也减少了可加速第三方集成的自然开发者采用。 [CE021, CE022, CE023, CE026, CE027, CE028]

信任、质量、合规和安全
控制 / 认证状态范围缺口 / 尽调要求
ISO 10218 / ISO/TS 15066(机器人安全)Phoenix 未公开确认人机协作环境中的协作机器人安全大多数司法辖区商业部署需要;要求提供认证或时间表
CE marking / UL 认证(产品安全)未披露面向美国 / 欧盟市场的电子、电气设备和机械指令合规没有 CE 或 UL,Phoenix 不能在欧盟或美国商业市场合法销售;需核实状态
OSHA / WorkSafeBC(职业安全)未披露;Sanctuary 在加拿大 BC 省运营,受 WorkSafe BC 管辖试点环境中操作员和同事的工作场所安全索取 Phoenix 试点环境的 WorkSafe BC 批准记录
专利组合(Sanctuary Cognitive Systems Corp.)活跃 — Morgan Stanley 全球排名 #3(2025 年 2 月);patents.google.com 确认有申请记录电机设计、操作算法、传感器集成、Carbon AI 架构具体专利号、授权状态、到期日和司法辖区分布未披露
数据安全 / 模型安全未披露;未发布信任中心、SOC 2 或安全审计AI 训练数据、客户部署数据、机器人遥测数据索取信息安全政策、数据处理协议及任何渗透测试报告
出口管制(ITAR / EAR / Canadian Export Controls)现阶段不适用于商业机器人,但军民两用技术可能需要审查若客户贴近国防,或向受管制国家开展国际销售,则可能触发确认液压系统、AI 和传感器技术的出口管制分类

所有状态均反映缺乏公开披露,并不代表已确认不合规。Sanctuary 是加拿大私营公司,没有第三方可获取的监管文件。认证可能已存在,但尚未公开宣布。

[CE026, CE027, CE028, CE031]
FE003: 关键依赖图谱

Sanctuary AI 的产品交付依赖硬件、软件、基础设施和合作伙伴关系构成的网络。关键上游节点包括未披露的液压组件供应商、Microsoft Azure AI 基础设施、Magna 制造(状态不确定)和 Carbon AI 训练管线。下游客户价值取决于所有上游节点保持健康。

依赖结构根据合作关系、产品描述和架构表述推断。硬件供应商和云服务商未公开识别。由于 2024 年 4 月后没有更新,Magna 制造路径显示为不确定。

[CE009, CE022, CE023, CE025, CE029]

5.5 竞争定位——液压灵巧性 vs. 电驱规模化

2023 年以来,人形机器人竞争格局明显升温。Sanctuary 面对多家资金充足的电驱玩家:Figure AI(已融资 $675M+,在 BMW 生产部署)、 1X Technologies(电驱人形机器人 EVE 和 NEO,OpenAI 合作)、Agility Robotics(Digit 已部署到 Amazon,商业验证最成熟)、 Apptronik(Apollo,NASA 背景,Mercedes-Benz 合作)、Boston Dynamics(Atlas,数十年 R&D,但没有商业人形机器人产品)、 Unitree(低成本 H1 和 G1,面向科研和较轻商业应用)。此外,Tesla Optimus 也是潜在进入者,资本充足,并具备消费级制造能力。 Sanctuary 相对这一赛道的差异化主要压在三项主张上:(1) 21-DoF 液压手,灵巧度超过多数竞争对手夹爪;(2) Carbon AI 的仿生认知架构, 公司认为它比规模化 transformer 模型更适合人类任务泛化;(3) 垂直 IP 整合,让 Sanctuary 控制技术栈每一层。不过,液压手优势可能被其引入的运营复杂度削弱, Carbon AI 架构也尚未通过公开基准验证。 IEEE Spectrum 分析师、Rodney Brooks 等独立机器人研究者质疑,人形机器人估值是否反映了近期商业就绪度。Brooks 记录在案的预测指出, 机器人部署一贯比技术演示暗示的时间更长。Sanctuary 未披露商业部署、客户名称和单位经济模型,很难与演示阶段竞争对手区分开来。 与已披露生产部署的同业相比(Agility 在 Amazon,Figure 在 BMW),按公开证据看,Sanctuary 的牵引力明显更弱。 [CE032, CE033, CE034, CE035, CE036, CE037]

竞争格局对比
公司机器人平台驱动类型关键差异化商业牵引(公开证据)融资 / 规模优势
Sanctuary AIPhoenix Gen 8液压(手部);腿部可能为电驱21 自由度液压手;Carbon AI 仿生平台;垂直 IP 集成无具名客户或部署;Magna 试点未确认已披露约 $160M;Morgan Stanley IP 排名 #3
Figure AIFigure 02电伺服与 OpenAI 合作支持语言落地任务;已部署 BMW 制造场景BMW 生产试点(2024 年确认);已披露商业部署已融资 ~$675M+;估值 ~$2.6B(2024 年 2 月)
1X TechnologiesNEO 人形;EVE 轮式电驱OpenAI 为投资方;轻量化设计;安全优先理念NEO 已交付员工做家庭测试(2025 年);商业披露有限已融资 ~$130M+
Agility RoboticsDigit电驱与 Amazon 合作;商业成熟度最高;建成 Agility Factory 用于生产Amazon 履约中心试点(2024 年确认);商业部署验证最充分已融资 ~$220M+;Hyundai 持股
ApptronikApollo电驱NASA 工程积累;Mercedes-Benz 合作;聚焦物流 / 制造Mercedes-Benz 试点(2024 年确认)已融资 ~$100M+
Boston DynamicsAtlas(电驱)电驱(2024 年重新设计)数十年运动控制研发;被 Hyundai 收购;硬件积累最深未销售商业人形机器人产品;仅为研究平台Hyundai 支持;并非纯机器人创业公司
Unitree RoboticsH1、G1电驱低成本设计;广泛进入研究者群体;全球销售全球可商业购买;用于研究和轻工业估值较低;部署量最高

竞争对手融资和牵引数据来自截至 2026 年 5 月的公开公告和分析师报道。数字为近似值,可能未反映最新轮次。Sanctuary 腿部系统的驱动类型根据产品视频推断;公司尚未发布完整传动系统规格。

[CE032, CE033, CE034, CE035, CE036, CE037]
FE004: 竞争对手能力成熟度矩阵

从六个能力维度评估 Sanctuary AI 与五家主要竞争对手。评分为 1–10 的序数估计,基于公开证据。绿色(分数 ≥7)= 强或已验证;黄色(4–6)= 部分或早期;红色(≤3)= 有限或未验证。Sanctuary 在操作灵巧度上最高,但在商业部署证据上最低。

评分为基于截至 May 2026 公开证据的 1–10 序数估计。未经独立方审计。商业部署评分仅基于已公开确认的部署;可能存在未披露部署。融资数字为近似值。

[CE032, CE033, CE034, CE038, CE039, CE040]

5.6 证据展示

Chapter 06

06客户情况

6.1 客户基础概览——披露缺失

截至 2026 年 5 月,Sanctuary AI 的客户基础几乎完全未披露。除 2024 年 4 月与 Magna International 的单一战略试点合作外, 公司没有公开客户数量、年经常性收入(ARR)、活跃部署或具名生产客户。本分析时点,Sanctuary AI 官网没有客户证言、参考账户、案例研究或能证明活跃使用的 logo。 公司瞄准 B2B 工业市场,主要是仓储、汽车制造、零售运营和餐饮服务。Sanctuary 对外定位是向面临结构性劳动力短缺的工业运营商提供 Phoenix 人形机器人, 作为劳动力补充。公司将 co-pilot 模式描述为客户上线路径:初始部署时由人类监督员看管机器人。尽管目标清晰,但没有合同、部署协议或试点后续延展被公开确认。 Microsoft、Accenture、Magna International、Verizon Ventures、Workday Ventures、BDC Capital 和 InBC 等战略投资人, 可能代表通往企业客户的间接渠道;这在早期 B2B 机器人公司中很常见。但在 Sanctuary AI 案例中,投资人关系是否转化为商业客户协议尚未确认。 公司 2025 年 4 月与 Microsoft 在 Hannover Messe 联合展示,显示其仍在努力触达制造垂直的工业买家,但公开记录中没有看到该活动带来交易。 已表述的市场野心与有记录的客户牵引之间的证据缺口,是本章最重要风险。 [CU001, CU002, CU003, CU004, CU005, CU022]

客户分层表
买方细分目标垂直买方 / 用户 / 付款方地理渠道已知客户状态 / 证据
汽车制造装配 / 质量检测一级汽车制造商(付款方)北美(为主)企业直销Magna International(试点,Apr 2024)仅试点;未确认量产
仓储 / 物流拣货 / 分拣 / 包装3PL 和电商运营商(付款方)北美企业直销 / 合作伙伴(Accenture)未披露已声明目标市场;无具名客户
零售运营货架补货 / 库存大型零售商(付款方)北美直销 / Microsoft 企业渠道未披露公司声明的目标;无具名客户
餐饮服务 / 酒店备餐 / 服务任务餐厅和酒店运营商北美企业直销未披露公司声明的目标;无具名客户
一般工业 / 制造结构化手工任务自动化工业运营商(付款方)全球(公司愿景)直销 / Accenture 系统集成未披露(仅具名 Magna)商业化前阶段

买方细分基于公司官网和新闻报道中声明的目标市场。仅 Magna International 获公开确认。其余细分截至 2026 年 5 月只代表公司战略意图,缺乏具名客户证据。

[CU001, CU003, CU004, CU022]
FU001: Sanctuary AI 客户旅程图

Sanctuary AI 工业买家的获客与采用旅程,展示从认知到扩展的五个阶段及每个关口的证据质量。截至 May 2026,灰色阶段(生产与扩展)没有确认依据。

旅程阶段基于公司表述的 GTM 路径,并从单一 Magna 试点合作推断。生产和扩展阶段是假设性的,截至运行日期没有确认走通。

[CU001, CU006, CU021, CU023]

6.2 Magna International 试点——唯一确认合作

Sanctuary AI 唯一公开确认的客户合作,是与 Magna International 的战略伙伴关系。Magna 是加拿大 Tier-1 汽车零部件制造商,全球拥有超过 350 座制造设施。 该合作于 2024 年 4 月 11 日通过 Sanctuary AI 新闻页面宣布,并获得 Betakit、The Robot Report 等加拿大和机器人媒体报道。公告称, Phoenix 机器人将在 Magna 制造设施场景中部署,目标是验证机器人在工业场景执行结构化手工任务的能力。 截至 2026 年 5 月,也就是初次公告后 13 个月,Sanctuary AI 或 Magna 均未发布后续新闻稿、结果报告或案例研究来确认 Magna 试点结果。 也没有 Magna 高管被引用背书 Phoenix 表现。按行业标准,这种沉默并不寻常;成功试点通常会在六到十二个月内产生联名宣传。 这种缺席强烈暗示:试点可能已经结束但未达成商业协议,可能被搁置,也可能仍受禁止公开披露的保密协议约束。 Magna 合作的性质和范围很关键。Magna International 也是 Sanctuary AI 2024 年 6 月宣布的 Series B 轮战略投资人, 这意味着双方关系同时包含财务投资和运营试点评估。投资人兼潜在客户的双重角色在早期机器人公司中很常见,但会让外界更难判断合作深度, 也削弱其作为客户证明的独立性。若试点最终没有转化为商业协议,将对 Sanctuary AI 的商业化叙事构成重大挫折。 [CU006, CU007, CU008, CU009, CU010, CU011]

客户采用轨迹表
日期事件客户 / 合作伙伴信号类型截至 2026 年 5 月的状态
2023-03Series A 轮 CAD $58.5M — 投资方包括 MagnaMagna International(投资方)融资 / 投资方信号Magna 既是投资方,后来也成为试点伙伴
2024-04-11宣布与 Magna International 达成战略合作Magna International具名试点客户公告唯一确认的客户侧公告;无后续进展
2024-06-12宣布 Series B 轮 US$100M;Magna 列入投资方Magna International投资方角色再次确认客户与投资方角色重叠;试点状态未更新
2025-04-01与 Microsoft 在 Hannover Messe 做工业演示Microsoft(合作伙伴,非客户)展会 / 需求生成活动未宣布商业交易;仅是行业曝光信号
2025-05-15发布零样本手内方块操作演示无具名客户技术里程碑(无客户信号)技术有进展,但未披露商业牵引
2026-05-15运行日期 — 2025–2026 年无新客户公告无具名客户缺乏披露距上一次客户侧公告(Magna)已有 24 个月空窗

时间线来自公司新闻归档和第三方媒体报道。所有客户侧条目均仅与 Magna 合作有关。2024 年 5 月至 2026 年 5 月没有公告,是已记录的证据缺口。

[CU006, CU007, CU008, CU011, CU033]
具名客户证据表
客户细分部署 / 用例量产 vs 试点已披露结果数据证据质量最后更新
Magna International汽车制造Phoenix 机器人进入制造设施(结构化任务)试点(未确认量产)未公开披露低 — 单次公告,无结果、无高管引述Apr 2024(此后无更新)
未披露零售 / 物流运营商零售 / 仓储货架补货或拣货试点(公司暗示)UnknownNone很低 — 仅从目标市场表述推断Unknown
未披露制造合作伙伴一般工业制造结构化装配或检测任务(公司暗示)UnknownNone很低 — 仅从目标市场表述推断Unknown

部分列举。Sanctuary AI 尚未公开披露任何客户名单。第 2、3 行代表暗示的目标细分,不是已确认客户。证据质量评级反映公开文档的丰富度:结果指标、高管引述、第三方佐证、时效性和量产清晰度。

[CU006, CU007, CU009, CU010]
FU002: Sanctuary AI 客户采用漏斗

从总可用市场到已确认生产部署的 Sanctuary AI 估算客户转化漏斗。从试点(1)到生产(0)的断崖式下滑,是核心商业风险。

市场规模是估计参考值;知晓以下所有转化阶段均未知或为推断。单一 Magna 试点是唯一确认的客户数据点。漏斗数值代表分析师估计和公开证据,不是公司披露。

[CU002, CU004, CU006, CU015]

6.3 竞争对手客户基准——Sanctuary AI 位置的参照

截至 2026 年 5 月,Sanctuary AI 的客户牵引力明显弱于最接近的竞争对手。Figure AI 在 2024 年 1 月宣布与 BMW Manufacturing Co. 达成生产部署合作,用于汽车制造线。Agility Robotics 自 2023–2024 年起将 Digit 机器人部署到 Amazon 履约中心和 GXO Logistics 设施, 已在至少两个具名企业账户中从试点转为持续商业运营。1X Technologies 已在斯堪的纳维亚部署人形机器人用于商业仓库运营。Boston Dynamics 的四足机器人 Spot 虽非人形机器人,但拥有包括 Ford 和 BMW 在内的多个已确认企业客户,可作为商业机器人部署时间线的参照点。 在这一背景下,无论相比直接人形机器人同业,还是更广泛工业机器人行业,Sanctuary AI 与 Magna 的单一未确认试点都是相当薄的客户证明。 Morgan Stanley 2025 年 2 月将 Sanctuary AI 人形 IP 列为全球 #3,反映的是知识产权强度,而非客户牵引;这是商业尽调中必须区分的一点。 IP 排名可能说明长期竞争位置,但无法证明近期客户愿意付费。 CB Insights 和 Tracxn 的人形市场分析在最新覆盖中,均未把 Sanctuary AI 列为拥有已确认生产部署的公司。更广泛的人形机器人行业本身也面临质疑: IEEE Spectrum 的“humanoid robot bubble”文章和 Rodney Brooks 对部署时间线的持续评论都指出,绝大多数已宣布的商业合作并未推进到规模化生产运营。 Sanctuary AI 的处境符合这一行业模式,但相对已走过试点阶段的 Figure AI 和 Agility Robotics,没有任何生产客户是一个突出的风险因素。 [CU012, CU013, CU014, CU015, CU016, CU017]

FU003: 客户证明质量矩阵——Sanctuary AI 与竞争对手

从五个客户证明维度比较 Sanctuary AI 唯一确认合作(Magna)与最接近的人形机器人同行 Figure AI、Agility Robotics、1X Technologies 的证据质量。评分为 0–10 的序数质量评级,分数越高越强。数据来自公开新闻稿和媒体报道。

分数为 0–10 序数估计,来自公开证据质量评估。它们反映结果数据丰富度、高管背书、独立佐证、证据新鲜度和生产部署清晰度。未经审计。竞争对手分数基于截至 May 2026 的公开公告。

[CU012, CU013, CU014, CU015]

6.4 获客策略——证据有限

Sanctuary AI 没有公开正式 GTM 文件或面向客户的材料,因此获客方式主要只能从公司表述和投资人沟通中推断。公司把自己定位为直销型 B2B 企业销售组织, 面向需要自动化结构化手工任务的工业运营商。co-pilot 模式需要人类操作员在初始任务学习阶段监督 Phoenix,公司将其描述为从首次部署走向自主运行的路径, 这意味着上线路径和销售周期都更长。 Accenture、Microsoft、Workday Ventures 等战略投资人可能提供渠道助力:Accenture 是工业客户的技术集成商,Microsoft 提供云基础设施和企业关系, Workday Ventures 可把 Sanctuary 连接到对劳动力增强感兴趣、与 HR 技术相邻的企业。但这些关系都没有被明确描述为带收入分成或客户转介绍条款的渠道销售协议。 销售团队规模、配额结构、线索生成策略或获客成本(CAC)数据均未披露。公司 LinkedIn 和截至 2026 年中的招聘信息显示技术岗位仍在招聘, 但没有明显的销售负责人或首席营收官职位,这可能说明公司尚未进入结构化商业规模化阶段。平均销售周期、交易规模参数和合同期限模板,在公开来源中完全未知。 GTM 证据缺失与技术优先开发阶段相一致,但也让商业轨迹预测存在显著不确定性。 [CU019, CU020, CU023, CU031, CU034, CU038]

留存、重复使用与满意度指标表
指标报告值置信度来源 / 依据尽调询问
净收入留存率(NRR)未披露N/A公开记录中无客户留存数据向管理层索取 NRR;对标工业 SaaS 常模(120%+)
总收入留存率(GRR)未披露N/A公开记录中无客户留存数据向管理层索取 GRR 和流失队列细节
合同续约率未披露N/A无合同披露;Magna 试点未确认续约确认 Magna 试点是否转为商业协议
客户满意度 / NPS未披露N/A没有公开可用的评价、评分或 NPS 调查索取客户满意度数据;若已上架,检查 G2/Capterra
平均合同期限未披露N/A公开记录中无交易参数向管理层索取交易结构和合同期限

截至 2026 年 5 月,所有数值均未披露。Sanctuary AI 未在公开沟通中发布任何客户留存、满意度或合同指标。本表记录这些维度上公共客户证据的完全缺失。

[CU023, CU024]

6.5 反向信号、集中度风险与证据缺口

Sanctuary AI 的客户位置暴露出几项会实质影响投资判断的反向信号。第一,公司单独依赖一个未确认试点(Magna),这是极端客户集中度风险。 若未来出现任何规模化客户公告,Sanctuary 都是在从零到一,初始部署将定义其总收入和运营学习曲线中不成比例的一大部分。 该单一关系中的任何挫折都会产生放大影响。 第二,首席科学官 Suzanne Gildert 在 2024 年 11 月领导层调整中离任,让支撑产品开发和客户试点的技术领导连续性出现不确定性。 Gildert 曾与 Geordie Rose 联合创办 Sanctuary AI;此前两人也共同创办 D-Wave Systems。她的离任带走了一位奠基人物, 而她在深科技领域潜在企业客户中的可信度原本是一项资产。 第三,Rodney Brooks 是机器人研究中最受尊重的声音之一。他在公开文章中持续主张,由于任务泛化和部署经济性挑战尚未解决, 通用人形机器人未来十年内不会实现商业规模化。他的预测记分卡和人形机器人评论,是来自权威独立来源的一致反向信号。 IEEE Spectrum 对“humanoid robot bubble”的分析也指出,2023–2025 年宣布的许多高调人形机器人部署,并未转化为批量商业协议。 第四,从 2024 年 5 月到 2026 年 5 月的 24 个月里,Magna 试点和 2024 年 6 月 $100M Series B 之后,没有披露任何新客户公告。 这段时间里 Sanctuary 的资本化和技术里程碑都在推进,但商业消息沉默;这可能是公司刻意压住商业新闻,也可能是企业销售周期推进缓慢, 或确实缺乏商业协议。三种解释都影响投资判断。本章记录的证据缺口——没有客户数量、没有 ARR、没有具名部署、没有客户证言——合在一起, 构成 Sanctuary AI 公开记录中最大的尽调缺口。 [CU024, CU025, CU026, CU027, CU028, CU029]

客户集中度与扩张风险表
风险因素描述可能性严重性可用缓释措施尽调路径
单一客户集中Magna 是唯一确认接触;任何挫折都会形成完全集中风险高(当前适用)关键极少 — 未披露可分散集中度的管线确认 Magna 状态;索取客户管线指标
投资方与客户重叠Magna 既是投资方,也是唯一试点客户;独立性不清楚高(当前适用)重大需要区分投资方与客户治理审查投资协议中的客户义务
试点转量产缺口未确认任何试点转为商业量产协议高(当前适用)重大Magna 试点若运营成功,可能打开转化获取 Magna 试点结果;评估转化时间线
领导层离职的负面影响CSO Suzanne Gildert 于 2024 年 11 月离职;可能伤及客户信心重大新技术领导层保持连续性确认 CTO/CSO 接任者及面向客户的技术团队深度
企业销售周期慢Co-pilot 导入模式需要较长试用期高(结构性)重大Hannover Messe 和 Microsoft 渠道可能缓解对标 Agility Robotics 和 Figure AI 的销售周期
未披露客户管线合格线索、在推进试点或后期交易均无公开数据Unknown重大公司可能有 NDA 下未披露管线通过 NDA 向管理层索取客户管线指标

风险评级是基于缺乏客户证据和对已披露合作的直接分析得出的定性估计。严重性反映风险兑现时的潜在投资影响。没有管理层访问,所有缓释措施都只是推测。

[CU025, CU029, CU030, CU034]
FU004: 已确认具名客户数量——人形机器人同行(May 2026)

条形图比较截至 May 2026 Sanctuary AI 与最接近的人形机器人竞争对手的公开确认具名客户合作数量(试点和生产)。Sanctuary AI 落后于所有列示同行。

计数仅代表公开宣布或确认的具名客户关系。计数是基于公开新闻稿和第三方新闻的保守估计。未具名或未披露客户关系不计入。Boston Dynamics 计数针对 Spot(四足机器人),这是最接近的商业化先例;Atlas(人形)没有商业客户。

[CU016, CU036, CU030, CU037]

6.6 证据展示

Chapter 07

07风险

7.1 技术与产品风险

Sanctuary AI 的 Phoenix 机器人依赖 21 自由度液压手;公司称其灵巧度行业领先。液压驱动的力密度高于可比电驱系统,但也带来电驱竞争对手无需面对的一组可靠性和运营风险。 液压流体回路容易泄漏、热膨胀、密封件退化和污染。在食品加工、制药、半导体制造或医疗器械装配等生产型工业环境中,液压油泄漏不只是维护问题, 而是监管和产品责任问题。Sanctuary AI 没有公开披露液压流体封装设计、密封件寿命规格、维护间隔或污染风险缓释措施。 Carbon AI 认知软件通过 Microsoft Azure 云基础设施运行推理。没有可靠互联网连接的环境、气隙设施(国防、核能、制药),以及 Azure 出现宕机或价格层级变化的场景, 都会把这一依赖变成机器人运行的单点故障。Sanctuary AI 网站确认了 Microsoft 合作,但未披露边缘推理能力、本地回退逻辑或延迟容忍规格。 sim-to-real 迁移缺口——机器人在仿真环境和真实物理部署中的表现差异——是 Sanctuary AI 最具技术后果的未解风险。公司在 YouTube 和 Hannover Messe 2025 等展会上展示了令人印象深刻的操作能力。但持续生产部署数据、错误率统计或正常运行时间指标都未披露。2025 年 3 月的 sim-to-real 灵巧操作演示证明了进展, 但没有验证商业可靠性。在持续部署数据发布前,迁移缺口仍是任何投资逻辑中的重大假设。 专利强度是一把双刃剑。Morgan Stanley 截至 2025 年 2 月将 Sanctuary AI 列为全球人形 IP 第 #3,这是正面信号。但组合中许多专利仍处于申请中, 竞争对手可以在授权前研究已公布申请,并绕开权利要求设计。若关键待审专利被驳回或缩窄——尤其是覆盖液压手或 Carbon AI 推理架构的专利—— Sanctuary AI 的 IP 护城河将被实质削弱。 [CR001, CR002, CR003, CR004, CR005, CR006]

运营与技术风险登记表
故障模式可能性严重性缓释成熟度剩余敞口未解决缺口
敏感制造环境(食品、制药、半导体)中的液压油泄漏或密封失效低 — 未披露液压油围堵规格液压油污染产品线,会带来监管召回风险和 Sanctuary AI 的客户责任液压围堵架构、密封寿命规格和维护间隔数据未公开披露
Microsoft Azure 云中断或延迟飙升,打断 Carbon AI 机器人运行低-中 — Microsoft Azure SLA 覆盖可用时间;Sanctuary AI 未确认边缘推理后备方案中断期间机器人停止或行为不可预测;未披露安全停机或本地后备规格未确认边缘推理能力、故障安全行为规格,以及 Azure SLA 对 Carbon AI 的适用性
仿真到现实迁移缺口导致生产部署中的操作错误部分 — 已确认多次仿真到现实演示;未披露生产部署可靠性工业场景中的操作错误(掉落产品、损坏设备、安全事件)会带来责任敞口没有公开的生产部署错误率、可用时间或任务成功率指标
长时间生产条件下触觉传感器校准漂移低 — 触觉传感器已在 2024 年 12 月展示;长期校准稳定性未形成文件漂移会造成操作错误并随时间累积;维护负担和现场服务成本未知校准漂移规格、MTBF 数据和重新校准流程未公开披露
商业化前液压手 IP 遭专利挑战或无效部分 — Morgan Stanley待授权专利可被竞争对手绕开设计;核心专利若被驳回,将削弱 IP 护城河核心液压操作和 Carbon AI 权利要求的专利授权状态未公开列举

各行先按严重性、再按可能性排序。缓释成熟度仅依据公开披露评估。公开记录中没有 Sanctuary AI 机器人相关安全事件记录,但部署规模不足,缺失记录不具备统计意义。

[CR001, CR002, CR003, CR004, CR005, CR006]
FR001: 风险热力图:影响程度、发生可能性与缓解成熟度

该矩阵按影响程度(关键 / 高 / 中)、发生可能性(高 / 中 / 低)和缓解成熟度(强 / 部分 / 弱)给 Sanctuary AI 的主要风险类别打分。没有客户和财务不透明在影响 × 发生可能性的综合评分上最高。技术风险和对 Microsoft 的依赖影响程度高,缓解成熟度介于部分和弱之间。

风险调性和严重程度评级来自截至 2026 年 5 月的公开证据综合,不基于内部风险登记册或公司私下披露。

[CR001, CR003, CR008, CR019, CR024, CR031]

7.2 市场与竞争风险

Sanctuary AI 最紧迫的近期风险,是截至 2026 年 5 月完全没有披露商业客户。唯一公开具名客户合作,是 2024 年 4 月 11 日宣布的 Magna International 战略制造试点。后续没有任何公告确认该关系延续、扩大到生产规模或商业条款。Magna 自身投资人沟通中也没有提到活跃的 Sanctuary AI 机器人部署。 这种沉默说明试点可能已经结束,或仍处于延长评估阶段;这对 Sanctuary AI 的商业叙事构成重大压力。 自 Sanctuary AI 2024 年 6 月 Series B 以来,竞争环境已实质恶化。Figure AI 拿下 BMW Group 的生产合作,用于制造部署——这是与 Sanctuary AI 目标市场直接可比的客户类型。Agility Robotics 拿下 Amazon 作为仓库部署客户,确认 Sanctuary AI 所瞄准的可服务市场已经被竞争对手服务。 1X Technologies 追加融资 $110M,并报告活跃商业部署。Boston Dynamics 已在多个工业垂直建立 Atlas 和 Spot 的企业销售。Neura Robotics 和 UBTech 也带着资金充足的项目加入竞争。 所有主要竞争对手——Figure、1X、Agility、Boston Dynamics、Unitree——的商业机器人都采用电驱。液压驱动带来生态碎片化劣势: 围绕电驱系统建立的客户、维护队伍和供应链,较难迁移到 Sanctuary AI 的液压平台。若市场标准化到电驱——考虑到竞争对手压倒性的偏好,这看起来很可能—— Sanctuary AI 的液压差异化可能从优势变成竞争负担。 Goldman Sachs 预计 2035 年人形机器人市场将达到 $38B,但 IEEE Spectrum 和独立分析师对近期采用时间线提出了可信担忧。Rodney Brooks 是该领域最受尊重的从业者之一, 他持续预测通用人形机器人到 2020 年代末仍会面对商业可行性的重大障碍。在 Sanctuary AI 将技术演示转化为有公开参考案例的付费、重复客户之前, 公司的商业可行性仍未得到证明。 [CR008, CR009, CR010, CR011, CR012, CR013]

市场与竞争风险登记表
风险竞争对手或驱动因素证据严重性缓释措施投资含义
竞争对手已在 Sanctuary AI 目标市场拿到具名生产客户Figure AI(BMW Group 客户)、Agility Robotics(Amazon 客户)Figure-BMW 合作和 Agility-Amazon 部署已在 2024-2025 年公开确认Sanctuary AI 的 Gen 8 和触觉传感器进展仍可能形成技术差异每过一个季度仍无具名客户,都会扩大与获融资竞争对手的客户背书缺口
电动驱动成为市场标准,不利于 Sanctuary AI 的液压平台Figure AI、1X Technologies、Agility Robotics、Boston Dynamics(均为电动)主要竞争对手均采用电动驱动;洁净室和食品级规格往往排除液压Morgan Stanley长期平台匹配风险;客户采购规格可能排除液压
低成本中国人形机器人让硬件层商品化Unitree(G1、H1 机型)、UBTech WalkerUnitree G1 定价约 USD $16,000;UBTech 和 UBTECH Walker 面向企业市场高端认知差异化(Carbon AI)可能维持相对纯硬件竞争对手的定价权若中国竞争压缩硬件利润率,Sanctuary AI 的软件护城河必须驱动回报
Rodney Brooks 和 IEEE Spectrum 对近期商业化时间表提出可信质疑学界 / 分析师怀疑者(多方)Brooks 2023 年预测记分卡和 IEEE Spectrum 的「humanoid robot bubble」文章指出多项时间风险公司 2024-2026 里程碑在灵巧性方面早于 Brooks 最初的怀疑时间表负面分析师共识可能影响融资情绪和企业客户风险偏好

竞争风险图景基于公开宣布的合作和部署。竞争对手融资轮次和员工数数据来自公开申报和公告;Sanctuary AI 缺少可比公开数据。

[CR008, CR010, CR011, CR012, CR013, CR014]
FR002: 风险传导图

展示 Sanctuary AI 的主要风险向量如何传导为收入延后、资本约束和估值影响。没有客户和财务不透明共同指向融资难度。Microsoft 依赖和技术风险都会流向客户采用不确定性。

因果链条根据已披露公司信息和可比公司风险分析推断;未量化传导概率。

[CR002, CR008, CR019, CR024, CR031, CR032]

7.3 执行与运营风险

Sanctuary AI 联合创始人、前首席科学官 Suzanne Gildert 在 2024 年 11 月领导层调整中离任。Gildert 与 Geordie Rose 共同创办公司, 她对 Phoenix 认知和机器人架构的技术贡献,是公司早期差异化的核心。公司尚未公开宣布接任的首席科学官或首席技术官。Sanctuary 历史上依赖两位创始人级技术专家; 在这种背景下缺少具名技术负责人,会让持续产品推进以及面向企业客户和投资人的深层技术可信度产生执行风险。 2024 年 4 月宣布的 Magna International 制造合作,是 Sanctuary AI 获取汽车级制造规模和具名工业伙伴的唯一已披露路径。没有后续更新,也没有活跃 Magna 部署的可见证据,因此该合作状态是首要执行不确定性。若 Magna 已退出合作,Sanctuary AI 没有披露可替代的制造规模化伙伴; 在公司需要从原型生产转向商业批量制造的阶段,这是关键缺口。 人才留存是叠加的执行风险。AI 和机器人强人才市场竞争激烈:超大规模云厂商(Google DeepMind、Microsoft、Amazon、Meta)、 自动驾驶公司、资金充足的人形机器人初创公司(Figure、1X、Physical Intelligence),都在争夺同一批机器人工程师、机器学习研究员和具身 AI 专家。 Sanctuary AI 位于 Vancouver AI 生态,该生态还包括 Cohere、D-Wave 等资金充足的公司。没有薪酬数据、员工数或留存指标披露时,外界无法评估 Sanctuary AI 守住人才基础的能力。 Sanctuary AI 要从 R&D 阶段原型演示转向商业批量部署,需要资本密集型制造能力;公司没有公开确认已经获得这些能力。机器人制造爬坡——模具、质量保证、 供应链搭建、现场服务网络——通常需要多年提前期,并且需要显著高于 R&D 支出的资本开支。 [CR017, CR018, CR019, CR020, CR021, CR022]

合作伙伴与依赖风险登记表
依赖项对手方角色集中度失败情景严重性缓释措施剩余敞口
Microsoft Azure + AI 合作Microsoft Corporation (NASDAQ: MSFT)Carbon AI 的云推理基础设施和 AI 模型开发;在 Hannover Messe 联合开拓市场关键 — 未确认替代云或边缘推理平台Microsoft 终止合作或重新定价;收购竞争人形 AI 平台;战略优先级下调极高战略投资者关系;多次公开合作公告;联合演示中使用 Microsoft 品牌未公开合同条款、排他性条款或定价结构;单一供应商云依赖尚未缓释
Magna International 制造试点Magna International Inc.(NYSE/TSX: MGA,制造合作伙伴)唯一具名客户 / 制造合作伙伴;汽车级验证和试点部署高 — 唯一披露的客户合作;没有具名替代合作伙伴试点结束或停滞;Magna 投资者沟通未提及仍在进行的 Sanctuary 机器人部署战略投资者关系;初始合作公告获得正面媒体报道没有后续公告;试点是继续、暂停还是停止未知
Export Development Canada 贷款(CAD $30M)Export Development Canada(联邦国有企业)政府债务融资;CAD $30M 贷款偿还义务中 — 约占总融资的 20-25%;贷款条款未公开披露因收入不及预期或财务状况恶化而违反贷款契约;EDC 要求提前还款政府贷款方,对出口导向公司条款更灵活;补助公告中可见政治支持贷款条款、契约和还款时间表未公开披露
BDC Capital / InBC 战略投资者BDC Capital(Thrive Venture Fund)/ InBC Strategic Investments(公共资本)提供资本和加拿大生态资源的战略投资者低-中 — 多名投资者之一;无独占依赖政府投资授权调整或 InBC 缩减,减少从这些来源获得后续资本的机会投资于 2024 年 7 月宣布;BDC Capital Thrive Fund 是多基金载体,具备活跃投资部署授权投资条款未公开披露;投票权和董事会观察员权利未知
Accenture 和 Workday Ventures 作为企业投资者Accenture PLC / Workday Inc.具备企业销售渠道潜力的战略投资者低 — 渠道触达是间接的,尚未确认为商业渠道企业投资者退出或降低人形机器人关系优先级;没有渠道交易落地多家企业投资者拥有企业销售网络,降低单一渠道依赖没有公开商业协议显示 Accenture 或 Workday 企业客户因投资而签约

各行按严重性排序。Microsoft 同时是基础设施提供方和战略投资者,造成双重依赖风险。截至 2026 年 5 月,Magna 合作结果仍是 Sanctuary AI 最重要的未解决商业指标。

[CR009, CR029, CR030, CR031, CR041]
人员与执行风险登记表
角色或职能依赖或缺口可能性严重性缓释措施尽调路径
CEO Geordie Rose唯一公开具名的创始高管;公司身份、投资者关系和愿景沟通均依赖其人极高联合创始人有长期履历;多元投资者董事会提供部分连续性后备要求提供继任计划、董事会审批权限和关键人保险政策
首席科学官 / 首席技术官(公开空缺)Suzanne Gildert 于 2024 年 11 月离职;截至 2026 年 5 月,未公开确认替任 CTO 或 CSO持续产品发布(Gen 8、触觉传感器、仿真到现实)显示技术厚度;领导层缺口未确认确认是否已任命替任技术负责人;要求提供组织架构图和产品领导结构
AI 与机器人工程人才基础未披露员工数;面临超大规模云厂商、自动驾驶和资金充足的人形机器人公司竞争温哥华 AI 生态提供人才管线;独特使命可能吸引使命驱动型工程师要求提供员工数变化、流失率,以及相对 Figure AI、1X、Physical Intelligence 的薪酬基准
制造与硬件规模化执行Magna 合作状态不明;未披露替代制造规模化合作伙伴或内部设施加拿大政府支持(EDC、BDC)提供部分制造机构背书信号确认制造合作伙伴状态;要求提供生产路线图和目标商业化产量下的单位经济性

各行按严重性排序。CEO 层面的关键人风险偏高,但对现阶段创始人主导公司并不罕见。Gildert 离职是公开记录中最重大的未披露缺口。

[CR017, CR018, CR019, CR020, CR021, CR022]

7.4 财务与融资风险

Sanctuary AI 是私营公司,未披露收入、ARR、毛利率、经营亏损或烧钱速度。公司在 Series B(2024 年 6 月)中融资约 US$100M, 此前还完成 CAD $58.5M Series A 融资(2023 年 3 月),并获得 Export Development Canada 的 CAD $30M 贷款(2024 年 6 月)。 对一家收入前硬件公司而言,总融资额构成了实质资本基础;但没有财务披露时,外部投资者完全看不清现金跑道。 硬件商业化资本密集。人形机器人制造商需要大量投资模具、制造设备、供应链、质量保证、现场服务和保修准备金。软件业务可以用毛利支持增长, 机器人硬件业务则通常要等制造规模带来可支撑利润率的单位经济模型后,才可能摆脱负自由现金流。在低于 100 台的产量下,Sanctuary AI 几乎肯定处于深度负单位毛利。 Export Development Canada 贷款带来固定还款义务,不随收入表现变化。看不见 Sanctuary AI 的收入轨迹,就无法评估 CAD $30M 义务是否造成流动性风险。 公司依赖持续风险投资融资,意味着若投资人对 AI 机器人情绪恶化——无论由宏观环境、竞争对手失败,还是 Sanctuary AI 自身商业挫折触发—— 都可能限制公司以可接受条款完成下一轮融资。 工业机器人销售周期长(从初始接触到商业部署通常 12-24+ 个月),会拉长收入变现时间并放大现金消耗。收入前状态、未知烧钱速度、硬件成本结构和 EDC 贷款还款义务叠加, 构成重大流动性风险;没有私人财务披露,无法更精确刻画。 [CR024, CR025, CR026, CR027, CR028, CR029]

财务风险登记表
风险可能性严重性量化缓释措施剩余敞口尽调路径
烧钱速度和现金跑道未知确定(外部未知)极高烧钱速度完全未披露;根据可比的尚无收入人形硬件公司估算,范围为 USD $5-20M/monthUS$100M B 轮加上此前 CAD $58.5M A 轮,提供了有意义的资本基础缺少财务披露,无法评估偿付能力;没有依据确认现金跑道是否超过 12 个月要求提供经审计财务报表、经营现金流和至少 18 个月现金预测
商业规模硬件制造资本强度高硬件制造需要大量前期工装、供应链和质量保证投入政府融资(EDC CAD $30M)部分抵消资本需求;战略投资者可能共同投资100 台以下产量的单机经济性几乎肯定深度为负;规模拐点未知要求提供单位成本拆分、BOM 分析,以及 100/500/1000 台产量下的降本曲线预测
在不确定 AI 投资环境中依赖持续 VC 融资VC 市场环境可能变化;若竞争对手失败,AI 机器人情绪可能恶化明星投资者(Microsoft、Accenture、BDC)提供一定稳定性,并向后续投资者释放质量信号若下一轮融资前未证明商业牵引,可能出现估值下调轮风险或过桥融资缺口评估当前投资者财团承诺;要求提供任何进行中股权或债务融资条款
EDC 贷款偿还义务面对未知收入基础CAD $30M 固定偿还义务;契约和时间表未披露EDC 是有耐心的贷款方,明确肩负支持加拿大出口商的授权;历史上可重组如果收入没有兑现,EDC 贷款形成固定现金义务,可能需要修订或再融资要求 Sanctuary AI 管理层提供贷款条款、契约和还款时间表

所有财务风险量化均根据可比的尚无收入人形硬件公司推断,应在私下尽调中用 Sanctuary AI 实际财务报表验证。公司没有公开财务披露。

[CR024, CR025, CR026, CR027, CR028, CR029]
FR003: 依赖图:关键外部伙伴与监管机构

梳理 Sanctuary AI 的关键外部依赖:Microsoft(云 + AI 模型 + 商业化)、Magna(制造 + 客户验证)、EDC/BDC/InBC(政府资本)和监管机构。所有能产生收入的依赖都要经过 Phoenix 机器人和 Carbon AI 平台。

依赖强度未量化。通过 Accenture 和 Workday 的企业渠道具备潜力,但商业层面尚未确认。

[CR029, CR030, CR031, CR032, CR034, CR036]

7.5 战略、监管与劳动力风险

Sanctuary AI 最集中的战略依赖是 Microsoft 合作。Microsoft 为 Carbon AI 推理提供 Azure 云基础设施,参与 AI 模型开发合作,并通过 Hannover Messe 2025 等联合亮相提供共同营销。若 Microsoft 因收购竞争性人形 AI 平台、战略转向或商业争议而缩减、重定价或终止合作,Sanctuary AI 的技术基础设施和企业 GTM 策略都会受扰动。 合作的合同条款、排他性约定或定价结构均未公开披露。 人形机器人与人类工人在工业场所并肩作业的监管框架仍处早期。在加拿大,WorkSafeBC 和各省职业健康与安全机构监管工作场所安全, 但尚未为人形机器人工友场景建立专门标准。在美国,OSHA 现有机器人安全标准(主要基于工业机器人 ISO 10218)并非为双足、 具备高级操作能力的通用人形机器人设计。缺少清晰监管框架带来两类风险:(1) 当人形机器人造成工伤时,客户面对责任法律不确定性,可能推迟采用; (2) Sanctuary AI 可能遭遇追溯性监管要求,被迫进行昂贵的机器人设计修改或运营限制。 在市场走向电驱的背景下,Sanctuary AI 的液压定位带来长期战略对齐风险。如果主流工业机器人安全标准、保险政策或客户采购要求开始指定仅限电驱 (部分洁净室规格已出现这种情况),Sanctuary AI 将面临被迫架构转向。公司目前提供竞争差异化的 IP 组合集中在液压灵巧手技术上, 因此转向电驱的技术成本会很高。 Honda ASIMO、SoftBank Pepper 等失败的通用人形机器人项目,是历史前车之鉴:即便资本充足、技术成熟的项目,也可能无法达成商业可行性。 这些前辈面对类似挑战:演示强、客户部署有限、最终商业停摆。虽然 Sanctuary AI 在 2024-2026 年的进展使其不同于早期失败项目, 但结构性相似仍是投资者应当权衡的可信反向信号。 [CR031, CR032, CR033, CR034, CR035, CR036]

监管 / 法律风险登记表
风险 / 规则 / 义务司法辖区状态可能性严重性缓释措施剩余敞口尽调路径
与人类并肩作业的人形机器人工作场所安全认证加拿大(WorkSafeBC、省级 OHS)/ 美国(OSHA)目前没有专门针对人形机器人工作场所安全的标准;ISO 10218 覆盖工业机器人,但不覆盖双足人形机器人Sanctuary AI 尚未公开认证或合规声明客户围绕人形机器人造成工伤的责任存在法律不确定性,可能拖慢企业采用确认 Phoenix 是否已提交任何安全认证;要求提供合规路线图
AI 机器人技术的出口管制 / 军民两用技术风险加拿大(Export and Import Permits Act 下的 ITAR 类似规则)/ 美国(EAR/ITAR)任何 AI 或机器人技术出口均负有现行义务;Sanctuary AI 未公开出口合规姿态Microsoft 合作可能施加商业使用限制;未公开出口许可证披露加拿大 AI 机器人技术若出口至非盟友市场,可能触发监管行动;美国销售需要遵守 EAR要求提供出口合规计划文件及既往监管问询记录
机器人采集的工作场所视频和传感器数据涉及个人数据 / 隐私合规加拿大(PIPEDA / Bill C-27 / BC PIPA)/ 欧盟(任何 EU 部署适用 GDPR)截至 2026 年 5 月,Bill C-27(加拿大拟议 AI 监管)仍处于议会程序;没有已生效、专门约束 AI 机器人数据的法律Sanctuary AI 隐私政策未公开细节;Carbon AI 数据保留政策未披露机器人执行任务时采集员工生物识别数据,客户运营方承担责任风险要求提供 Carbon AI 数据保留、匿名化和同意架构文件
Sanctuary Cognitive Systems Corp. 在 BC Business Corporations Act 下的义务与公司治理加拿大不列颠哥伦比亚省有效 — Sanctuary Cognitive Systems Corporation 已在 BC 注册;公司申报按 BC Act 管理标准公司治理;年度申报要求按常规处理无重大剩余敞口;BC 注册处确认公司状态有效查看 OpenCorporates / BC Registry 的最新高管披露及任何特别决议申报

各行按严重性排序。截至 2026 年 5 月,公开记录中未发现针对 Sanctuary AI 的现行诉讼。人形机器人监管框架仍处早期,未来可能追溯性施加设计要求。

[CR031, CR032, CR033, CR034, CR036]
缓释与止损标准表
风险可监测触发因素阈值或事件行动含义
B 轮后仍无披露客户Sanctuary AI 新闻稿、公司网站、合作伙伴公告到 2026 日历年末仍未宣布具名商业客户重新评估投资论点;大幅下调对商业可行性的信心
Magna 合作确认结束Magna 投资者沟通、Sanctuary AI 新闻稿、行业新闻Magna 移除任何 Sanctuary AI 相关提及,或 Sanctuary AI 宣布替代制造合作伙伴若 Magna 已结束且没有替代方:制造规模化论点破裂;需要替代路径证据
Microsoft 合作范围缩减Microsoft AI 合作公告、Sanctuary AI 网站更新、联合活动露出Microsoft 停止与 Sanctuary AI 联合营销;2026 年主要贸易活动无联合亮相Azure 依赖 + 市场拓展依赖同时断裂;评估战略替代方案
融资轮失败或条款严重恶化融资公告、VC 市场新闻、显示裁员的员工 LinkedIn 离职动态到 2026 年下半年仍未完成 C 轮,或以大幅稀释宣布过桥融资公司存续风险信号;现金跑道可能不足以支撑商业化里程碑
公开演示出现技术失败或安全事件YouTube、新闻报道、社交媒体、IEEE Spectrum演示或部署中发生液压油事故、造成财产损失的操作错误或安全事件立即重新评估技术风险和声誉损害;评估客户信心影响
CTO 或同等技术负责人离职且无替补LinkedIn、公司公告、媒体报道另一名高级技术负责人离职,且 90 天内没有公开具名替任者技术执行连续性面临风险;评估创始人层级以下工程领导梯队厚度

止损标准是投资论点层面的触发器,不是运营质量保证指标。阈值仅作示例;投资者应按自身回报画像和尽调权限校准。

[CR009, CR019, CR024, CR031, CR041]
风险评分摘要
风险类别可能性影响缓释成熟度剩余严重性主要证据缺口
无披露商业客户(市场风险)确定极高薄弱极高Magna 试点结果未知;没有具名替代客户
烧钱速度和现金跑道未知(财务风险)确定(未知)极高薄弱极高没有财务披露;无法评估现金跑道
Microsoft 合作集中(战略风险)极高部分合同条款和 Azure 定价未披露
电动主导市场中的液压驱动(技术 / 市场风险)部分未披露洁净室或食品级认证
Magna 合作状态不明(执行风险)薄弱2024 年 4 月公告后没有更新
CTO/CSO 领导缺口(执行风险)薄弱Gildert 离职已确认;未公开具名替任者
生产环境仿真到现实落差(技术风险)部分未披露生产部署可靠性指标
工作场所人形机器人缺少监管框架(监管风险)OSHA 或 WorkSafeBC 尚无人形机器人专门标准

评分基于截至 2026 年 5 月的公开证据综合。可能性和影响为定性判断,来自已披露里程碑、竞品基准和行业类比。 无法访问内部风险登记册。

[CR001, CR008, CR024, CR031, CR032, CR041]

7.6 证据展示

Chapter 08

08估值

8.1 估值概览与信息限制

Sanctuary AI 是加拿大私营公司,没有法律义务向 SEC、SEDAR 或任何其他公开监管机构提交财务报表。截至 May 2026,公司未披露任何融资轮估值、年经常性收入(ARR)、客户数或员工数。本章使用的所有财务指标,要么是新闻稿层面的事实(融资金额、投资方名称),要么是分析师根据可比人形机器人和服务机器人公司推算的估计值。信息不透明是估值分析最大的约束。 最近一次已确认融资是在 May–July 2024 完成的 Series C,BDC Capital(Thrive Venture Fund)领投,InBC Strategic Investments、Obvious Ventures、Verizon Ventures、Accenture Ventures、Microsoft、Magna International、Workday Ventures 和 Bell 跟投。媒体报道给出的融资规模约为 $99.8M–$100M。June 2024,Export Development Canada 另外提供 CAD $30M(约 USD $22M)的政府支持研发贷款。所有轮次累计披露资本约 $243.8M。截至 May 2026 运行日,公司尚未宣布 Series D。 在没有披露财务指标的情况下,可用的估值代理有三类:(1)对照同类私营机器人公司的融资倍数;(2)使用同行公司(Figure AI、Boston Dynamics)披露估值的可比交易分析;(3)基于 Morgan Stanley 专利排名数据和 Microsoft Azure AI 合作的 IP 与战略伙伴溢价。三种方法不确定性都很高,应视为方向性区间,而不是点估值。估计公允价值区间为 $400M–$1.5B,中点接近 $700M。 [CV001, CV002, CV003, CV004, CV005, CV006]

Sanctuary AI 历史融资轮次
轮次大致日期金额(USD)领投 / 主要投资方备注
种子轮 / Pre-Series A 轮2018–2022Unknown内部 / 天使机构轮前;金额未披露
Series A 轮Q1 2023~$43M (CAD $58.5M)Evok Innovations首个机构轮;CAD 计价
EDC 政府贷款2024 年 6 月~$22M (CAD $30M)Export Development Canada非稀释债务;资金用途指向研发
Series B / Series C 轮2024 年 6–7 月~$100MBDC Capital(领投)、InBC、Microsoft、Accenture、Magna、Workday、Bell、Evok、Verizon Ventures、Obvious VenturesGlobeNewswire 新闻稿称为「Series B」;PitchBook 称为「Series C」;各来源轮次口径不一致
已披露融资总额截至 2024 年 5 月~$243.8M多家战略和财务投资方包含 EDC 贷款;不含任何未披露 pre-seed;各轮均未披露估值

所有金额均为估算,来自新闻稿和第三方数据库(PitchBook、CBInsights);Sanctuary AI 未披露任何轮次估值。 总额包含 EDC 贷款。各来源对轮次命名不一致(Series B vs. Series C),正文已记录。

[CV001, CV002, CV003, CV004, CV006, CV007]
关键未披露指标——估值阻断项
指标是否披露?估算区间估算依据缺口严重度
公司估值$400M–$1.5B累计融资 2x–6x 代理估算重大
年收入 / ARR$0–$10M未确认商业客户;假设暂无收入阻断
员工人数200–300 人(估)LinkedIn 档案、新闻提及、可比阶段同业轻微
商业客户数量0–2 个已确认无公开部署公告;Magna 试点状态未验证阻断
烧钱速度 / 现金跑道$60M–$90M/年(估)可比阶段人形机器人同业分析重大
清算优先权堆栈1x 时约 $244M;若参与分配则更高累计融资;假设采用标准 VC 条款重大

所有估算均由分析师根据可比公司数据和公开基准推导。Sanctuary AI 未发布这些数字。「阻断」表示该缺口会阻碍可靠估值分析, 出资前必须通过数据室访问解决。

[CV005, CV008, CV029, CV032, CV033, CV034]
FV001: Sanctuary AI 融资时间线

按时间展示 Sanctuary AI 从创立到 2024 年 Series C 的已披露融资事件。

种子轮日期为估计值(公司创立于 2018 年,早期资本很可能发生在 2019-2021 年)。Series A 的准确日期基于研究缓存笔记估计为 2023 年 Q1。EDC 贷款时点可能与 Series B/C 交割重叠。

[CV001, CV002, CV006, CV004, CV021]

8.2 融资历史与隐含估值区间

Sanctuary AI 至少通过四个可识别批次募集资本。最早披露的机构资本是 early 2023 的 Series A,融资约 CAD $58.5M(按当时汇率约 USD $43M),由 Evok Innovations 和其他加拿大科技投资者支持。更早的种子轮或 pre-Series A 可能存在,但未披露。June 2024,公司宣布一笔 USD $100M 融资;GlobeNewswire 新闻稿称为 “Series B”,但 PitchBook 和 CBInsights 数据库交叉引用为 “Series C”,反映了可转债、政府贷款和独立战略交割被合并进一个报道总额时常见的术语不一致。BDC Capital 的 Thrive Venture Fund 被列为领投方,InBC Strategic Investments——British Columbia 政府支持的基金——以及多家企业战略投资者共同参与。同月,Export Development Canada 提供一笔非稀释性 CAD $30M 贷款,进一步抬高了媒体报道的资本注入总额。 采用标准的 2x–6x 已融资资本估值代理——这是可比阶段尚未产生收入的深科技公司的常见区间——$243.8M 的累计融资意味着估值底部约 $487M(2x),顶部约 $1.46B(6x)。该区间中点约 $730M。关键是,这一代理假设投资者以市场化入场价格进入,没有战略折扣或政府补贴影响有效资本成本。由于 BDC Capital 和 EDC 是政府关联实体,使命不止于财务回报,它们的定价可能不同于公平交易的商业 VC,可能把隐含倍数压到低于市场常态。 没有出现二级市场交易、要约回购或数据室披露来校准这个区间。到 May 2026 仍未完成 Series D——距 Series C 约 24 个月——使公司的融资轨迹存疑。按每年 $60M–$90M 的估计烧钱速度(对照同阶段人形机器人同行),Series C 资金到 mid-2026 会基本耗尽,意味着公司要么压低支出,要么有未披露收入,要么即将融资。 [CV001, CV002, CV003, CV004, CV006, CV007]

Series B / C 投资方概况
投资方投资方类型轮次战略逻辑独立性
BDC Capital(Thrive Venture Fund,领投方)政府 VCSeries C 轮(领投)支持加拿大深科技生态;使命包括打造国家级冠军独立
InBC Strategic Investments政府 VC(BC 省)Series C 轮BC 省技术投资使命;创造就业、留住 IP独立
Microsoft企业 VC / 战略合作伙伴Series B / C 轮Azure AI 云合作;企业机器人管线;AI 模型共同开发合作伙伴
Accenture Ventures企业 VC / 咨询Series B / C 轮企业部署、咨询项目管线、集成服务收入合作伙伴
Magna International战略 / 制造合作伙伴Series B / C 轮Phoenix 制造试点;生产产能选择权合作伙伴
Workday Ventures企业 VC(HR 科技)Series B / C 轮未来劳动力自动化;HR 软件与机器人部署集成合作伙伴
Obvious Ventures财务 VC(影响力)Series B / C 轮影响力投资使命;AI 和可持续自动化组合独立
Verizon Ventures企业 VC(电信)Series B / C 轮IoT 连接用例;5G 赋能机器人部署假设合作伙伴

投资方名单来自 GlobeNewswire Series B/C 新闻稿和二级数据库来源(PitchBook、BDC Capital 官方通信、InBC 官方通信)。未披露各投资方出资额。战略逻辑为分析师推断;投资方未公开详细确认这些理由。

[CV002, CV003, CV007, CV036, CV037, CV038]
人形机器人同业资本效率对比
公司累计融资(USD)隐含 / 已披露估值估值 / 融资本金倍数收入状态(截至 2026 年)
Figure AI$675M$2.6B(2024 年 2 月披露)3.85xBMW 试点;规模化前暂无收入
Agility Robotics~$150M未披露UnknownAmazon 试点;商业化有限
1X Technologies~$135M未披露Unknown有限试点部署
Apptronik~$350M未披露Unknown暂无收入;NASA 和 GE 试点
Sanctuary AI$243.8M未披露;估算 $400M–$1.5B未知;代理估算 1.6–6.2x暂无收入;无已确认客户

非 Sanctuary 公司融资总额来自二级数据库(Wikipedia、PitchBook、CBInsights);可能不含未披露轮次。 Sanctuary AI 融资总额来自新闻稿和任务简报研究。非 Figure AI 公司的估值倍数未知;Figure AI 的 3.85x 倍数是唯一可用同业数据点。

[CV011, CV014, CV015, CV017, CV018, CV040]
FV004: 投资 KPI

面向 IC 的关键绩效指标,覆盖截至 2026 年 5 月 Sanctuary AI 的资本、估值、商业化、IP 和投资者结构。

所有估值和现金跑道数字均为分析师估计;Sanctuary AI 没有公开可得财务数据。投资者结构百分比按投资者数量计算,而非按资本额计算。Morgan Stanley 排名来自 2025 年 2 月,可能已经变化。

[CV001, CV002, CV003, CV021, CV041]

8.3 可比公司分析

为 Sanctuary AI 搭建估值可比组,需要桥接三类公司:(1)披露估值的人形机器人初创公司;(2)披露估值或收购价格的服务和工业机器人公司;(3)拥有 SaaS 或授权收入的 AI 机器人平台公司。没有单一可比公司能同时覆盖所有维度。 Figure AI 是最接近的公开基准。February 2024,Figure AI 以 $2.6B 投后估值融资 $675M,由 Microsoft、OpenAI、Nvidia、Samsung、Jeff Bezos 等组成的财团领投。该轮同时披露 BMW 制造业试点,提供了 Sanctuary 缺乏的商业验证。按 $2.6B 估值和 $675M 融资额计算,Figure AI 的已融资资本倍数为 3.85x。把这一倍数外推到 Sanctuary 的 $243.8M 融资额,理论估值接近 $940M;但 Sanctuary 未披露客户,需要对该数字打折。 Agility Robotics 在 2023 年从 Amazon Industrial Innovation Fund 融资约 $150M,建立了隐含的企业客户验证,而 Sanctuary 尚未复制。Agility 未披露估值。Apptronik 在 2024 年获 GV(Google Ventures)和 Honda 支持融资 $350M,也未披露估值。Boston Dynamics 在 2021 年被 Hyundai Motor Group 以约 $1.1B 企业价值收购(包含此前 SoftBank 投资),提供了唯一完全披露的机器人 M&A 交易基准;收购时,Boston Dynamics 经常性收入有限,估值主要来自 IP、品牌和战略可选性。Tesla 的 Optimus 项目嵌在 Tesla 公司主体内,无法独立估值。 Sanctuary 在 Morgan Stanley 人形机器人 IP 排名中位列全球 #3(February 2025,由 November 2024 的 #4 上调),提供了部分 IP 质量信号。但专利数量和排名并不能完美代表可商业化价值;没有客户牵引时,单靠 IP 通常拿不到创收同行的倍数。相对于无客户基线的 IP 溢价估计为 10%–25%,主要来自阻挡竞争对手设计的防御性护城河价值,以及未来授权收入的可能性。 [CV011, CV012, CV013, CV014, CV015, CV016]

可比估值表
公司最近已知估值(USD)累计融资(USD)估值 / 融资本金倍数商业客户关键备注
Figure AI$2.6B(2024 年 2 月)$675M3.85xBMW(已确认试点)Microsoft、OpenAI、Nvidia、Jeff Bezos 领投
Boston Dynamics~$1.1B(Hyundai 收购,2021 年)N/A(已被收购)N/A多个(Atlas、Spot)Hyundai 收购;此前 SoftBank 投资总额约 $1.1B
Agility Robotics未披露~$150MUnknownAmazon(试点)Amazon Industrial Innovation Fund 为投资方
Apptronik未披露~$350MUnknownNASA, GEGV(Google Ventures)、Honda 支持
1X Technologies未披露~$135MUnknownAker Solutions(试点)挪威初创公司;NEO 平台
Tesla Optimus包含在 Tesla 估值中N/AN/A仅内部使用(截至 2026 年)未单独估值;Tesla 市值包含 Optimus 期权价值
Unitree Robotics未披露(私营)UnknownUnknown科研 / 教育中国制造商;低成本消费级人形机器人
Sanctuary AI未披露$243.8MUnknown截至 2026 年 5 月未确认按融资倍数代理估算为 $400M–$1.5B;无商业客户

未上市公司估值来自新闻稿和二级数据库来源(PitchBook、Wikipedia、CBInsights);均未经独立审计。 Sanctuary AI 估值仅由分析师按融资倍数推导。Boston Dynamics 数值反映 Hyundai 收购企业价值(约数)。Tesla Optimus 不纳入有意义的倍数分析。非 Sanctuary 公司的「累计融资」可能不含未披露轮次。

[CV011, CV012, CV014, CV015, CV016, CV017]
估值情景分析
情景概率隐含估值(USD)关键条件退出时间窗口倍数依据
乐观20%$800M–$1.5BSeries D 轮披露估值 >$1B;确认商业客户;Microsoft 合作仍在推进2027–2028累计融资 3–6x
基准55%$400M–$800M继续商业化前开发;IP 组合完整;无收入;新增战略共同投资方2028–2030累计融资 1.5–3x
悲观25%$100M–$350M下行轮;到 2026 年 Q4 仍无商业客户;联合创始人流失;竞争性资金转向 Figure AI2028 年以后或核销累计融资 0.4–1.5x

概率和区间为分析师估算;没有 Sanctuary AI 财务数据可校准。概率加权估值 = 0.20 × $1.15B + 0.55 × $600M + 0.25 × $225M ≈ $618M。若披露任何已确认商业收入,乐观情景概率会大幅上升。

[CV009, CV032, CV033, CV034, CV035, CV040]
FV002: 人形机器人公司估值 — 可比公司组

截至 2026 年 5 月,选定人形机器人公司的已知或隐含估值;Sanctuary AI 区间为分析师估计。

Sanctuary AI 柱形为分析师用融资倍数参照估算;实际估值未知。Apptronik 按同业倍数,以融资额约 1.4x 推断。Boston Dynamics 数字为 Hyundai 2021 年前后的收购企业价值。所有数值均以百万美元计。

[CV011, CV016, CV017, CV018, CV040]
FV003: Sanctuary AI 估值情景区间

乐观、基准和悲观情景估值区间,附概率权重和关键触发条件。

区间和概率为分析师估计;没有可用于校准的 Sanctuary AI 财务数据。乐观情景要求披露估值高于 $1B 的 Series D,和 / 或确认商业客户。悲观情景假设到 2026 年 Q4 仍没有商业牵引,却完成降估值融资。数值以百万美元计。

[CV009, CV033, CV040, CV041]

8.4 反向信号与下行风险评估

多重信号给 Sanctuary AI 的隐含估值区间带来实质下行。 最重要的反向信号,是完全没有已确认商业客户。截至 May 2026,公司没有发布企业部署公告、收入披露或客户证言。April 2024 随 Phoenix Gen 7 发布一同宣布的 Magna International 制造业试点,之后没有跟进商业合同公告。人形机器人从试点转商业的历史转化率低,原因包括集成复杂、安全认证周期长,以及买方对总拥有成本的担忧。Magna 试点可能已经推进为经常性商业关系且未公开宣布,但这一点未获验证。 联合创始人 Suzanne Gildert(首席科学官)在 November 2024 领导层调整中离职。Gildert 是两位创始人之一(另一位是 CEO Geordie Rose),共同建立了 Sanctuary 的通用智能 AI 方法论,此前还联合创办 D-Wave Systems。商业化前 AI 公司看重关键人物风险,这会显著压低估值;首席科学官在商业化前关键阶段离开,使研究方向稳定性受到质疑。 业内权威声音的怀疑进一步放大风险。IEEE Spectrum 文章 “The Humanoid Robot Bubble” 记录了行业营销口径与现实成本、运行时长基准之间的差距。Rodney Brooks 是全球引用最多的机器人研究者之一,也是连续创业者,他公开维护的预测记录持续指出人形机器人和自主机器人的时间表过于乐观。Figure AI 在已确认 BMW 合作的情况下拿到 $2.6B 估值,挤占了该领域高确定性端的资本和注意力,未披露客户的公司更难获得资源。 在悲观情景下,如果 Sanctuary AI 到 late 2026 仍未宣布商业客户,并需要降价轮延长现金跑道,隐含投后估值可能压缩至 $150M–$350M——低于优先股堆叠门槛——导致早期普通股股东被大幅稀释,并可能触发 Series C 优先股投资者的反稀释保护。 [CV023, CV024, CV025, CV026, CV027, CV028]

8.5 结论与投资建议

Sanctuary AI 在人形机器人格局中占据技术上可信的位置——专利 IP 全球排名 #3、与 Microsoft Azure AI 共同开发、Gen 8 Phoenix 平台配备 21-DoF 液压手,并与 Magna International 建立制造合作——但公司没有披露任何收入、估值、客户或员工数。在本阶段尽调中,无法有信心地配置资本。 按乐观(20%)、基准(55%)和悲观(25%)情景加权,估值得到的期望值约为 $600M–$750M。乐观情景($1.2B,概率 20%)假设公司以高于 $1B 的披露估值完成 Series D,并有商业客户公告和正向 Microsoft 合作指标支持。基准情景($650M,概率 55%)假设公司继续处于商业化前开发阶段,新增战略投资者维持 IP 和平台价值。悲观情景($200M,概率 25%)假设降价轮、到 Q4 2026 没有商业客户,并可能出现领导层流失。 概率加权结果 = 0.20 × $1.2B + 0.55 × $650M + 0.25 × $200B = $240M + $357.5M + $50M = $647.5M。与基于融资额代理的隐含中点($730M)相比,$648M 的概率加权估值有小幅折价,符合潜在投资者应向一家零公开财务披露公司要求的信息风险溢价。 建议为继续研究。若要上调至观察或买入,需要出现以下任一项:(1)披露 Series D 估值高于 $800M,且新的公平交易机构投资者领投;(2)确认商业客户部署并产生经常性收入;(3)大型工业或科技平台提出战略收购要约,并以披露价格创造流动性。缺少这些催化剂,现在投入资本为时过早。 [CV040, CV021, CV022, CV003, CV005, CV009]

8.6 附录

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;做出任何投资决定前,应直接向管理层和原始文件核实。

证据索引

结论
编号陈述可信度来源
CO001 Sanctuary Cognitive Systems Corporation was founded in 2018. SO001, SO002, SO008
CO002 Sanctuary AI is headquartered in Vancouver, British Columbia, Canada. SO001, SO002, SO021
CO003 Geordie Rose is the CEO and co-founder of Sanctuary AI. SO001, SO002, SO010
CO004 Olivia Norton is the COO and co-founder of Sanctuary AI. SO002
CO005 Sanctuary AI's core products are the Phoenix humanoid robot and the Carbon AI cognitive control system. SO001, SO013
CO006 Suzanne Gildert was a co-founder and Chief Science Officer of Sanctuary AI. SO010
CO007 Sanctuary AI announced a significant leadership transition on November 9, 2024, which involved Suzanne Gildert's departure. SO010
CO008 Sanctuary AI raised CAD $58.5 million in a Series A round in March 2023. SO007, SO008
CO009 Sanctuary AI announced a US$100 million Series B financing on June 12, 2024. SO004, SO005, SO006
CO010 The Series B investors included Microsoft, Accenture, Bell, Verizon Ventures, BDC Capital, Export Development Canada, Magna, Workday Ventures, and Evok Innovations. SO004
CO011 Export Development Canada provided a CAD $30 million loan to Sanctuary AI in June 2024, with conflicting reports on whether this is part of the $100M Series B or separate. SO004, SO009
CO012 BDC Capital's Thrive Venture Fund and InBC announced a strategic investment in Sanctuary AI in July 2024 without disclosing the amount. SO010
CO013 Sanctuary AI unveiled the seventh generation of the Phoenix humanoid robot on April 25, 2024. SO010
CO014 Sanctuary AI demonstrated a 21 degrees-of-freedom hydraulic hand technology on December 12, 2024. SO010
CO015 Sanctuary AI unveiled the eighth generation of the Phoenix humanoid robot on December 16, 2024. SO010, SO011
CO016 Sanctuary AI unveiled new tactile sensor technology on December 19, 2024. SO010, SO011
CO017 Morgan Stanley's research division ranked Sanctuary AI fourth globally for patent holdings in general-purpose robotics and dexterous manipulation as of November 27, 2024. SO010
CO018 Morgan Stanley's research division ranked Sanctuary AI third globally for humanoid robotics intellectual property as of February 11, 2025. SO011
CO019 Sanctuary AI demonstrated sim-to-real transfer of dexterous manipulation policies on March 18, 2025. SO011
CO020 Sanctuary AI's business model is B2B-focused targeting industrial automation, logistics, and retail environments. SO001, SO013
CO021 No board of directors, independent directors, or investor board seats have been publicly disclosed for Sanctuary AI as of May 2026. SO002, SO007
CO022 The precise structure of the Series B financing is unclear, with conflicting reports on whether the EDC loan is included in the $100M total. SO004, SO009
CO023 Sanctuary AI has never publicly disclosed its valuation. SO001, SO004, SO007
CO024 Sanctuary AI has not disclosed revenue, ARR, or any financial metrics as of May 2026. SO001, SO007
CO025 Sanctuary AI demonstrated zero-shot in-hand manipulation capabilities on May 15, 2025, showing autonomous reorientation of a lettered cube. SO012
CO026 Sanctuary AI announced a strategic partnership with Magna International on April 11, 2024, focused on manufacturing pilot and scaling production capabilities. SO010
CO027 Sanctuary AI announced a collaboration with Microsoft on May 1, 2024, for AI model development for general-purpose robots. SO010
CO028 Sanctuary AI has not disclosed any customer names or pilot deployment sites as of May 2026. SO001, SO012
CO029 Geordie Rose previously founded D-Wave Systems, a quantum computing company. SO016
CO030 Suzanne Gildert worked at D-Wave Systems before co-founding Sanctuary AI. SO016
CO031 The Carbon AI control system is described as mimicking human brain subsystems including memory, vision, audition, and touch. SO001, SO013
CO032 Sanctuary AI participated in Hannover Messe trade show with Microsoft in April 2025. SO011
CO033 Sanctuary AI's technology is described as vertically integrated, from motor design to generative AI algorithms. SO013
CO034 Sanctuary AI is a private Canadian company with no SEC filings or equivalent public regulatory disclosures. SO025
CO035 Rodney Brooks, a prominent robotics expert, has published skeptical views on humanoid robotics deployment timelines. SO017
CO036 IEEE Spectrum has published critical coverage questioning the humanoid robotics investment bubble. SO018
CO037 Sanctuary AI maintains open-source repositories on GitHub. SO024
CO038 Sanctuary AI's careers page lists positions across robotics engineering, AI/ML research, simulation, mechanical design, and electrical engineering, suggesting a technical team in the range of 100-300 people. SO003
CO039 Cumulative disclosed funding for Sanctuary AI is approximately CAD $58.5M plus US$100M, totaling roughly US$160M depending on exchange rates. SO004, SO008
CO040 Figure AI, a competitor humanoid robotics company, raised approximately $675M in February 2024 at a roughly $2.6B valuation. SO028
CM001 The humanoid robotics market boundary encompasses physical robots with human-like form factors (bipedal locomotion, anthropomorphic upper body, dexterous manipulators) designed to operate in human environments without specialized infrastructure modification. SM014, SM018
CM002 Included spend categories in the humanoid robotics market are capital expenditure on robot units, AI control software and simulation platforms, deployment services (integration, training, maintenance), and recurring software subscriptions. SM014, SM018
CM003 Status-quo substitutes for humanoid robots in warehousing and logistics include conveyor systems, industrial arms on fixed gantries, collaborative robot arms (cobots) on wheeled bases, autonomous forklifts, and human labor augmented by exoskeletons. SM007, SM010, SM018
CM004 Status-quo substitutes for humanoid robots in manufacturing assembly include dedicated automation cells, SCARA robots, delta robots for high-speed pick-and-place, and manual assembly lines. SM007, SM018
CM005 Sanctuary AI's Carbon AI control system positions the company at the intersection of robotics hardware and embodied AI software, differentiating it from pure hardware OEMs and pure software providers. SM014, SM015
CM006 Goldman Sachs Research published a February 2024 estimate projecting the global humanoid robotics market could reach $38 billion by 2035 under an optimistic adoption scenario. SM001, SM011
CM007 Goldman Sachs published a base case of $6 billion and a conservative case of $1 billion for the humanoid robotics market by 2035, reflecting extreme uncertainty in adoption timing and penetration rates. SM001, SM009
CM008 Goldman Sachs estimates humanoid robot unit prices declining from $150,000 per robot in 2025 to under $50,000 by 2035 due to manufacturing scale and component cost reduction. SM001
CM009 CB Insights identified 27 well-funded humanoid robotics startups as of early 2025, with cumulative disclosed funding exceeding $3 billion. SM006
CM010 The International Federation of Robotics reported 553,052 industrial robots installed globally in 2023, a 6% increase year-over-year, with cumulative operational stock exceeding 3.9 million units; humanoid robots represent a negligible share. SM002, SM012
CM011 The U.S. Bureau of Labor Statistics estimates approximately 4.3 million material movers and warehouse workers in the U.S. labor force as of 2024, representing a potential humanoid robot SOM proxy of $5B-$25B cumulative capital spend at $50K per unit if 2-10% of roles are substituted. SM003, SM004
CM012 The serviceable addressable market for humanoid robots in manufacturing and logistics is estimated at approximately $5-8 billion by 2030, contingent on cost reduction to below $75K per unit and achievement of 90%+ task completion rates. SM001, SM011
CM013 In the manufacturing segment, the humanoid robot buyer is typically the plant automation or engineering team, the payer is the corporate capex budget, and the budget owner is the operations VP or plant GM. SM007, SM009
CM014 In the logistics and warehousing segment, the buyer is 3PL ops or e-commerce fulfillment leadership; adoption triggers include peak-season labor scarcity and TCO parity with temporary labor over 3-5 years. SM008, SM009
CM015 Amazon operates over 750,000 mobile robots as of 2024 but has not deployed humanoid robots at commercial scale. SM011, SM009
CM016 In healthcare and elder care, humanoid robots would need to be funded through facility opex budgets (not insurance reimbursement), with buyers being hospital administrators and nursing home operators. SM008, SM018
CM017 The U.S. Bureau of Labor Statistics projects 203,000 annual nursing job openings through 2032 due to retirements, creating a durable healthcare automation demand signal. SM003, SM004
CM018 The retail and hospitality segment faces annual labor turnover exceeding 100% and rising minimum wages, creating structural automation motivation, but customer acceptance constraints (uncanny valley) are a material adoption risk. SM008, SM009
CM019 U.S. JOLTS data (April 2026) showed 8.1 million job openings and 7.4 million unemployed workers, sustaining a structural labor supply-demand imbalance that is a primary demand driver for humanoid robots. SM004, SM003
CM020 Warehouse and logistics roles face annual turnover rates of 40-60%, driving recurring hiring costs and creating ongoing automation incentive for logistics operators. SM009, SM011
CM021 Wage growth for material movers averaged 4.2% annually from 2020-2025, outpacing productivity gains and improving the payback calculation for robot substitution over the same period. SM003, SM004
CM022 OSHA ergonomic standards, EU Machinery Directive updates, and U.S. CHIPS Act manufacturing credits provide regulatory and policy tailwinds for humanoid robot adoption in manufacturing settings. SM005, SM009
CM023 Industrial policy programs including China Made in China 2025 robotics targets, EU Horizon robotics grants, and U.S. advanced manufacturing credits provide government support for humanoid robot development and deployment. SM005, SM009
CM024 AI foundational model maturation (vision-language-action models, behavior cloning, sim-to-real transfer) reduces Carbon platform training costs and time, and is a critical enabler for general-purpose robot viability. SM011, SM013
CM025 Aging OECD populations and declining working-age populations in Japan, South Korea, and parts of Europe create a durable 10-15 year structural labor scarcity demand signal for humanoid robots. SM005, SM009
CM026 Total cost of ownership for a humanoid robot over 5 years is estimated at $200K-$250K all-in, including purchase price ($100K-$150K), integration services ($50K-$100K), and annual maintenance ($10K-$20K/year). SM011, SM009
CM027 A warehouse worker earning approximately $40K annually with 30% benefits burden totals approximately $52K total compensation; a humanoid robot must achieve 90%+ utilization for a 3-year payback at 2026 prices. SM003, SM009
CM028 No humanoid robot vendor, including Sanctuary AI, Figure AI, Agility Robotics, or Apptronik, has published validated productivity and uptime benchmarks from customer deployments as of May 2026. SM011, SM013, SM029
CM029 ISO Technical Committee 299 (robotics safety) standards do not yet cover general-purpose humanoid robots operating in unstructured human environments as of 2026; applicable standards are limited to industrial robot arms in fixed cells. SM016, SM011
CM030 Liability frameworks for robot-caused workplace injuries remain unclear for general-purpose humanoid robots; no established insurance product for humanoid robots at commercial scale has been announced as of May 2026. SM016, SM009
CM031 Capital intensity creates a chicken-and-egg adoption problem for humanoid robots — customers need fleet scale to justify integration overhead, but scale requires committed customers, creating an adoption bottleneck. SM011, SM016
CM032 Technology maturity risk, unproven reliability, and unknown failure modes distinguish current humanoid robots from mature automation alternatives and are cited by IEEE Spectrum as structural barriers to commercial viability. SM016, SM011
CM033 The World Economic Forum Future of Jobs 2025 report projects automation including AI and robots will displace 92 million jobs and create 170 million new jobs globally by 2030, providing a broad macro-demand signal for automation. SM005
CM034 Humanoid robot market sizing estimates range from $1B (conservative, 2035) to $38B (optimistic, 2035), a 38x spread reflecting deep uncertainty in adoption timing, cost reduction, and technology maturity validation. SM001, SM009
CM035 Defense and government segments represent a potential future market for humanoid robots in hazardous environment automation, with government procurement driven by capex appropriations and readiness mandates, though no current Sanctuary go-to-market targets this segment. SM018, SM013
CM036 Customer acceptance constraints including uncanny valley concerns and preference for human interaction in retail and hospitality settings are material adoption barriers not captured in standard TCO models. SM009, SM016
CM037 Sanctuary AI does not publicly disclose any serviceable obtainable market estimate, pilot deployment count, or revenue figure; market sizing for Sanctuary specifically cannot be performed from public data.
CM038 Rodney Brooks, a renowned roboticist and former MIT professor, has publicly stated that turning robotic ideas into deployment at scale is much harder than commonly projected and questions the commercial viability timelines of humanoid robotics companies. SM017, SM016
CM039 Honda's ASIMO humanoid robot was shelved in 2022 after 22 years of R&D without achieving commercial deployment at scale, and SoftBank's Pepper was discontinued in 2021 after selling approximately 27,000 units, demonstrating the difficulty of humanoid robot commercialization. SM016, SM018
CM040 The Goldman Sachs optimistic humanoid robotics market scenario of $38B by 2035, starting from a near-term base of approximately $300M, implies a compound annual growth rate of approximately 40-50% over the 2025-2035 period. SM001
CP001 Sanctuary AI competes in the humanoid robotics market with a mid-tier funding position of approximately $160M, positioned between well-funded US leaders such as Figure AI ($675M+) and low-cost Chinese competitors such as Unitree ($16K price point). SP021, SP024
CP002 Figure AI has raised approximately $675M+ at an approximately $2.6B post-Series B valuation (February 2024), with named investors including Microsoft, OpenAI, NVIDIA, and Jeff Bezos, and announced a BMW manufacturing pilot in January 2024. SP001, SP016, SP024
CP003 1X Technologies has raised approximately $125M including a strategic investment from OpenAI, and pivoted its product roadmap toward the NEO home assistance robot alongside its industrial EVE platform. SP002, SP011
CP004 Agility Robotics has raised approximately $250M including an Amazon strategic lead investment, and deployed its Digit robot in an Amazon warehouse pilot facility in 2023 — the most credible logistics humanoid deployment validation to date. SP003, SP017, SP024
CP005 Boston Dynamics launched a commercial electric version of Atlas in 2024, backed by Hyundai parent acquisition (approximately $1.1B in 2021), transitioning from its 30-year R&D posture to commercial humanoid deployment. SP005, SP014, SP019
CP006 Tesla Optimus remains deployed internally at Tesla factory operations as of May 2026, with Elon Musk citing targets of up to 1,000 units for Tesla factories by 2025 (unverified), and no external commercial availability announced. SP006
CP007 Unitree Robotics offers the G1 humanoid robot at approximately $16,000 and the H1 at approximately $90,000, representing a sub-$20K price point that challenges premium US and European humanoid competitors on cost. SP007, SP015
CP008 Apptronik raised approximately $160M in a Series A round in September 2024 with Google as a strategic investor and Google DeepMind partnership for robot learning, targeting manufacturing and healthcare applications with the Apollo platform. SP004, SP013, SP024
CP009 Figure AI's approximate $675M total raised represents the highest disclosed funding in the humanoid robotics sector as of May 2026, creating a $515M capital advantage over Sanctuary AI's approximately $160M. SP001, SP024, SP025
CP010 Agility Robotics has raised approximately $250M, Apptronik approximately $160M, and 1X Technologies approximately $125M, establishing a peer group of humanoid robotics companies with $125M-$675M in total disclosed capital. SP003, SP004, SP002, SP024
CP011 Tesla's Optimus development program is funded internally as Tesla CAPEX, estimated to exceed $1B in total investment, representing an unfunded competitive threat if Tesla opens Optimus to external customers. SP006
CP012 Boston Dynamics was acquired by Hyundai Motor Group for approximately $1.1B in 2021, giving it a parent company with manufacturing capacity and an automotive deployment pathway for the commercial Atlas platform. SP005, SP019
CP013 Unitree Robotics' total funding is not publicly disclosed; it is VC-backed with a Chinese investor base and operates its own manufacturing facility in China enabling the sub-$20K G1 price point. SP007, SP015
CP014 The named investor roster for Agility Robotics includes Amazon as a strategic lead investor and DCVC, making Amazon simultaneously a potential customer and investor with a potential conflict of interest for other logistics operators. SP003, SP024
CP015 Apptronik's Google strategic investment and Google DeepMind partnership for robot learning provide access to frontier AI infrastructure that differentiates Apptronik from non-Google-affiliated humanoid peers. SP004, SP013
CP016 1X Technologies' OpenAI strategic investment provides access to frontier language and vision models for task understanding, creating a robot-AI integration that differentiates its product from non-OpenAI-affiliated competitors. SP002, SP011
CP017 The humanoid robotics competitive landscape as of mid-2026 comprises at least 7 major funded competitors including Figure AI, Agility Robotics, 1X Technologies, Apptronik, Boston Dynamics Atlas, Tesla Optimus, and Unitree, plus Sanctuary AI as the subject company. SP024, SP026
CP018 Incumbent industrial robot OEMs including ABB, KUKA, Fanuc, and Yaskawa have not released general-purpose humanoid platforms as of May 2026, but represent potential future market entrants with existing customer relationships and service infrastructure. SP024, SP029
CP019 No commercial-scale humanoid robot deployment has been publicly confirmed by any vendor as of May 2026; all announced pilots are small-scale and results have not been publicly validated with productivity benchmarks. SP026, SP028
CP020 Morgan Stanley's research division ranked Sanctuary AI third globally for humanoid robotics intellectual property as of February 2025, up from fourth globally in November 2024. SP022, SP021
CP021 Sanctuary AI's approximately $160M in disclosed funding represents a $515M deficit versus Figure AI's approximately $675M, constraining R&D investment, manufacturing scale, and enterprise sales capacity. SP023, SP025, SP001
CP022 As of May 2026, two of the seven major humanoid robot competitors have announced named commercial customers or pilots (Figure AI with BMW and Agility Robotics with Amazon); Sanctuary AI has disclosed no named customers. SP026, SP028, SP021
CP023 The Unitree G1 price point of approximately $16,000 represents approximately an 8x discount versus Sanctuary AI's estimated $125,000 ASP, creating significant cost-competition from the Chinese market. SP007, SP015, SP021
CP024 Honda ASIMO was a 22-year R&D program (2000-2022) that never achieved commercial deployment at scale and was shelved in 2022; SoftBank Pepper was discontinued in 2021 after approximately 27,000 unit sales at a loss, providing adverse precedents for current humanoid market entrants. SP008, SP009, SP020
CP025 Sanctuary AI has publicly disclosed no commercial customers, no named pilot deployment partners, and no revenue as of May 2026, representing a material commercial validation gap relative to Figure AI (BMW) and Agility Robotics (Amazon). SP021, SP022, SP023
CP026 Sanctuary AI demonstrated a 21 degrees-of-freedom hydraulic dexterous hand in December 2024, a technical specification not publicly matched by any direct competitor as of May 2026. SP022, SP021
CP027 Figure AI uses an end-to-end neural network approach for robot control, architecturally distinct from Sanctuary's modular Carbon AI cognitive system; neither approach has been validated in a head-to-head task performance benchmark. SP010, SP016, SP021
CP028 Agility Robotics' Digit robot uses a parallel gripper without dexterous hand capability, limiting its addressable task set compared to Sanctuary's 21-DoF hydraulic hand design. SP003, SP017
CP029 Apptronik's Apollo carries NASA heritage from the company's prior work on astronaut exosuits; this design provenance provides credibility for high-reliability industrial applications. SP004, SP013
CP030 Boston Dynamics transitioned Atlas from a hydraulic research robot (discontinued April 2024) to a commercial electric-actuated version, leveraging over 30 years of bipedal locomotion R&D and existing enterprise customer relationships from Spot deployments. SP005, SP019, SP014
CP031 Tesla's vertical integration in actuators (motors), energy storage (batteries), and AI silicon (Dojo supercomputer, FSD chip) provides a potential cost structure advantage for Optimus if externally commercialized that no pure-play humanoid startup can replicate. SP006
CP032 Sanctuary AI's enterprise partnership ecosystem — Microsoft (Azure AI), Accenture (deployment services), and Magna International (manufacturing) — provides distribution and manufacturing enablers that most humanoid competitors lack as of May 2026. SP023, SP021, SP022
CP033 Sanctuary AI's Carbon AI cognitive architecture is differentiated from end-to-end neural network approaches by its modular design mimicking human brain subsystems, though this differentiation has not been validated by independent technical benchmarks. SP021, SP022
CP034 Sanctuary AI's go-to-market approach leverages enterprise distribution partnerships (Accenture for services, Microsoft for cloud AI) rather than building a direct sales force, contrasting with Agility Robotics' anchor-customer direct deployment approach. SP023, SP021
CP035 The non-exclusive nature of Sanctuary's Microsoft and Accenture partnerships creates a distribution risk — both partners could simultaneously support Figure AI or other humanoid competitors without breaching any disclosed agreement. SP023, SP021
CP036 Foundation model APIs (GPT-5, Gemini, Claude) capable of robot control via vision-language-action interfaces represent a potential commoditization threat to Sanctuary's Carbon AI cognitive architecture moat over the 2025-2030 period. SP026, SP028
CP037 Rodney Brooks' public scorecard of robotics predictions indicates structural skepticism about the commercial viability timelines of humanoid startups, noting that scale deployment is orders of magnitude harder than lab demonstration. SP018, SP030
CP038 The Magna International manufacturing partnership is strategically significant for production scale but its durability is contingent on Sanctuary achieving commercial milestones; Magna is a contract manufacturer serving multiple OEMs and the partnership appears non-exclusive. SP023, SP021
CP039 The ASIMO failure demonstrates that even a well-resourced, technically credible humanoid program (Honda spent estimated billions over 22 years) cannot achieve commercial success without a compelling customer value proposition at achievable price points. SP008, SP020
CP040 Independent technical reviews from IEEE Spectrum and Rodney Brooks identify the reliability gap, absent task completion benchmarks, and unproven real-world uptime as the primary shared weakness across all current-generation humanoid robots including Sanctuary AI. SP026, SP018, SP030
CI001 Sanctuary AI operates a Robotics-as-a-Service (RaaS) revenue model wherein robots are deployed at customer facilities and billed on a usage or subscription basis, consistent with industry peers Agility Robotics and Figure AI. SI001, SI007, SI010
CI002 At RaaS blended rates of $5-$8 per robot-hour, annual revenue per robot at 50% utilization on a two-shift basis is approximately $44,000-$70,000, potentially creating compelling unit economics relative to human labor costs of $45,000-$60,000 per worker per year. SI010, SI011
CI003 Secondary revenue streams for Sanctuary AI potentially include Carbon AI platform licensing, Accenture professional services co-sell, and government and defense contracts enabled by EDC and BDC relationships, none of which have been publicly confirmed as generating revenue. SI003, SI007
CI004 Sanctuary AI's strategic investor roster (Microsoft, Accenture, Magna, Workday, Bell, BDC, EDC, Evok) provides structural alignment with commercial revenue pathways, as strategic investors typically invest for deployment optionality with their own enterprise operations. SI003, SI012
CI005 Revenue from AI software licensing via Carbon AI for third-party hardware applications has not been publicly confirmed and is speculative; Sanctuary has not disclosed any licensing agreements or SaaS revenue as of May 2026. SI001, SI002, SI007
CI006 The Accenture partnership creates a structural possibility for professional services revenues from enterprise deployment and integration projects, but no Accenture-brokered Sanctuary deployment or revenue has been publicly disclosed as of May 2026. SI003, SI012
CI007 Sanctuary AI's hardware ASP is analyst-estimated at $100,000-$150,000 per Phoenix unit, based on comparable humanoid robot pricing; this estimate has not been confirmed by any company disclosure or commercial transaction. SI010, SI011
CI008 At an estimated ASP of $125,000 versus a human worker cost of approximately $45,000 per year, the hardware payback period without RaaS pricing is approximately 5-8 years excluding maintenance, integration costs, and software upgrades. SI010, SI011
CI009 Estimated gross margin on Sanctuary AI hardware at an ASP of $125,000 and COGS of $50,000-$75,000 per unit implies a 40-60% gross margin, consistent with industrial robot hardware peers but unconfirmed by any Sanctuary financial disclosure. SI010, SI011
CI010 No humanoid robot vendor has published official list pricing as of May 2026; all price estimates in the market are analyst-derived, media-reported, or based on market comparables. SI010, SI011, SI016
CI011 Sanctuary AI's estimated hardware COGS per unit ranges from $50,000 to $75,000, assuming 40-60% gross margin on an estimated $125,000 ASP and a Magna contract manufacturing cost structure. SI010, SI003
CI012 Sanctuary AI's estimated annual R&D OpEx is $35M-$55M, representing 50-60% of an estimated $60M-$90M total operating expense at an approximately 200-300 person organization; none of these figures have been confirmed by Sanctuary financial disclosure. SI008, SI015
CI013 Total estimated annual burn rate for Sanctuary AI is $60M-$90M per year at the current stage, based on comparable Series B robotics and AI hardware companies with similar headcount; Sanctuary has not disclosed any financial data. SI008, SI010
CI014 The Microsoft Azure partnership partially subsidizes Sanctuary AI's AI training compute costs; the specific financial value of the subsidy has not been publicly disclosed. SI003, SI007
CI015 Sanctuary AI has raised approximately $160M in disclosed capital across multiple rounds, including an estimated $30M Series A in 2022 and a confirmed $100M Series B in June 2024. SI003, SI008, SI005
CI016 Sanctuary AI was founded in 2018 by Geordie Rose (co-founder, formerly of D-Wave Quantum) and Suzanne Gildert (co-founder and Chief Science Officer until November 2024) in Vancouver, British Columbia. SI005, SI006
CI017 Sanctuary AI raised an estimated $30M Series A in 2022 with BDC Capital and Evok Innovations among investors, enabling Phoenix Gen 1-5 humanoid robot development through 2023. SI005, SI008
CI018 Sanctuary AI demonstrated the Phoenix Gen 7 robot completing autonomous grocery shelving and folding tasks at a Loblaws retail environment, the most public commercial-context validation of the Carbon AI platform to date. SI001, SI007, SI005
CI019 The June 2024 Series B round raised $100M from Microsoft, Accenture, Magna International, Workday, Bell Canada, BDC Capital, Export Development Canada, and Evok Innovations; no corporate valuation was publicly disclosed. SI003, SI012
CI020 Sanctuary AI demonstrated a 21 degrees-of-freedom hydraulic dexterous hand in December 2024 and received a Morgan Stanley SI007, SI005, SI001
CI021 Suzanne Gildert, co-founder and Chief Science Officer of Sanctuary AI, departed the company in November 2024, representing the most significant disclosed adverse event in Sanctuary's corporate history and a key-person risk to the technical AI roadmap. SI005, SI006
CI022 LinkedIn and media sources indicate Sanctuary AI has grown to approximately 200-300 employees as of May 2026, consistent with Series B capital deployment patterns; the exact headcount has not been officially confirmed. SI009, SI015
CI023 At an estimated $75M mid-case annual burn and approximately $90M-$100M of undeployed capital at Series B close (after earlier round spending), Sanctuary AI's estimated runway extends through mid-2026 to mid-2027, requiring a Series C or revenue acceleration. SI008, SI010
CI024 Sanctuary AI will likely require a Series C of $150M-$300M by 2026-2027 to fund manufacturing scale, enterprise sales, and AI R&D at competitive levels, consistent with peer fundraising dynamics. SI008, SI010
CI025 Sanctuary AI has disclosed no revenue, no customer names, no pilot deployment results, and no financial performance data as of May 2026; all financial assessments rely entirely on analyst modeling and secondary source inference. SI001, SI002, SI016
CI026 All core financial data required for institutional investment analysis — P&L, burn rate, COGS, gross margin, customer names, contract values, and valuation — is absent from public sources and would need to be obtained through investor data room access. SI001, SI002, SI016
CI027 Sanctuary AI's approximately $160M in total disclosed capital versus Figure AI's approximately $675M represents a $515M competitive funding gap, constraining investment in manufacturing scale, AI R&D, and enterprise sales relative to the category leader. SI003, SI008
CI028 Non-exclusive partnership arrangements with Microsoft, Accenture, and Magna provide no contractual revenue commitment; these partners could simultaneously support Figure AI or other competitors without breaching any disclosed agreement. SI003, SI012
CI029 Sanctuary AI's partnership quality (Microsoft Azure, Accenture, Magna) provides structural commercial pipeline access that partially compensates for absent public revenue disclosure, as strategic investors at Series B typically invest for deployment optionality. SI003, SI012, SI007
CI030 Sanctuary AI has not disclosed any government contracts, defense contracts, or publicly announced revenues from EDC or BDC beyond investment; the potential for government revenue exists structurally but is unconfirmed. SI001, SI002, SI003
CI031 IEEE Spectrum and Rodney Brooks have independently identified the commercial viability and financial sustainability of humanoid robotics startups as facing structural barriers, citing the gap between lab demonstrations and real-world deployment economics. SI016, SI017
CI032 Sanctuary AI has not raised any disclosed capital beyond the June 2024 Series B as of May 2026; no Series C announcement has been made and no additional strategic investor disclosures have appeared in public sources. SI001, SI002, SI008
CI033 Export Development Canada (EDC) participation in Sanctuary's Series B round as a strategic investor provides non-dilutive export financing optionality and signals Canadian government support for Sanctuary's commercial export strategy. SI003, SI020, SI018
CI034 BDC Capital's Series B co-investment provides Sanctuary AI with a government-backed institutional investor relationship that may facilitate future non-dilutive government R&D funding or Small Business Financing Programs in Canada. SI003, SI019, SI018
CI035 Sanctuary AI's financial verdict is conditional pass — technical differentiation (IP rank, dexterous hand) and strategic investor quality are above-average for the stage, but zero revenue disclosure, co-founder departure, and a $515M funding gap vs. Figure AI are material risk factors requiring data room validation before investment. SI001, SI016, SI003
CE001 Phoenix is Sanctuary AI's flagship general-purpose humanoid robot designed to perform a wide range of human tasks in industrial and logistics environments. SE001, SE002
CE002 Sanctuary AI unveiled the seventh generation of the Phoenix humanoid robot on April 25, 2024. SE003
CE003 Sanctuary AI unveiled the eighth generation of the Phoenix humanoid robot on December 16, 2024. SE003, SE004, SE007
CE004 Carbon AI is described as a cognitive control system that mimics human brain subsystems including memory, vision, audition, and touch to enable general-purpose task performance. SE001, SE002
CE005 Sanctuary's technology page states the company is vertically integrated from motor design to generative AI algorithms, with all solutions designed, built, owned, and patented by Sanctuary. SE001, SE002
CE006 Phoenix targets industrial automation, logistics, manufacturing, and potentially retail as primary deployment use cases. SE001, SE002
CE007 Phoenix robots have an anthropomorphic human-scale form factor designed to work in human-built environments without infrastructure modification. SE001, SE011
CE008 No public technical specifications for Phoenix Gen 7 or Gen 8 have been disclosed, including height, weight, payload capacity, battery runtime, or DoF count for the full body. SE007, SE011
CE009 Sanctuary AI demonstrated a 21 degrees-of-freedom (DoF) hydraulic dexterous hand on December 12, 2024, claiming it provides human-like in-hand manipulation capability. SE003, SE006, SE011
CE010 The 21-DoF hydraulic hand uses hydraulic actuation to provide compliance, force regulation, and continuous torque without cogging, enabling fine in-hand manipulation. SE006, SE011
CE011 Sanctuary AI unveiled new tactile sensor technology on December 19, 2024, designed to improve in-hand object sensing and grasp feedback. SE003, SE009
CE012 Sanctuary AI announced an update to its tactile sensor integration on February 26, 2025, indicating active development of the sensing pipeline. SE004, SE009
CE013 Sanctuary AI demonstrated sim-to-real transfer of dexterous manipulation policies on March 18, 2025, showing policies trained in simulation can be deployed on physical hardware. SE004, SE008
CE014 Sanctuary AI demonstrated zero-shot in-hand manipulation on May 15, 2025, with its hydraulic hand autonomously reorienting a lettered cube without task-specific fine-tuning. SE004, SE005, SE008
CE015 Carbon AI's training pipeline combines three mechanisms — teleoperation data collection, reinforcement learning, and sim-to-real transfer — to develop robot manipulation policies. SE001, SE002
CE016 Carbon AI is described as a brain-inspired or biomimetic system with specialized cognitive modules analogous to human cortical regions rather than a single monolithic transformer model. SE001, SE002
CE017 Sanctuary announced a collaboration with Microsoft on May 1, 2024 to develop AI models for general-purpose robots, implying use of Azure infrastructure for Carbon AI training. SE003, SE010
CE018 Sanctuary AI demonstrated Phoenix with Microsoft at Hannover Messe on April 1, 2025, targeting industrial automation customers through Microsoft's enterprise channel. SE004, SE010
CE019 Zero-shot manipulation refers to task execution without task-specific fine-tuning; the May 2025 cube demo represents the most challenging public generalization test in Sanctuary's portfolio. SE005, SE008
CE020 Carbon AI is described as proprietary, company-owned, and patented technology spanning every layer from motor design to generative AI algorithms per Sanctuary's technology page. SE001, SE002
CE021 Sanctuary's vertically integrated approach encompasses custom motor design, hydraulic actuation, tactile sensors, and Carbon AI cognitive software, all developed and patented in-house. SE001, SE002
CE022 Sanctuary's Microsoft collaboration and Hannover Messe co-appearance imply significant reliance on Azure cloud infrastructure for AI model training and large-scale inference. SE003, SE010
CE023 No public source confirms that Sanctuary uses Azure exclusively; multi-cloud or on-premises training configurations cannot be ruled out from available disclosures. SE001
CE024 Hydraulic actuation offers superior compliance, continuous torque, and force regulation compared to electric servo actuators for fine manipulation tasks requiring gentle contact. SE011, SE013
CE025 Hydraulic actuation systems are heavier than equivalent electric designs, require pressurized fluid circuits and periodic seal maintenance, and pose contamination risks in food, pharmaceutical, and cleanroom environments. SE011, SE013
CE026 Morgan Stanley's humanoid robotics research ranked Sanctuary AI fourth globally for patent holdings in general-purpose robotics and dexterous manipulation as of November 27, 2024. SE003, SE011
CE027 Morgan Stanley's humanoid robotics research ranked Sanctuary AI third globally for humanoid robotics intellectual property as of February 11, 2025, reflecting active patent filing. SE004, SE011
CE028 Sanctuary's patent strategy spans motor design, manipulation algorithms, and sensor integration, consistent with the vertically integrated platform claim; Google Patents confirms active filings under Sanctuary Cognitive Systems Corporation. SE015, SE002
CE029 The Magna International manufacturing partnership announced April 11, 2024 has produced no public updates, production milestones, or Phoenix unit delivery announcements through May 2026, leaving its status uncertain — potentially stalled or quietly discontinued. SE003, SE019
CE030 Sanctuary AI's GitHub organization (github.com/sanctuary-ai) has minimal public repositories with sparse commit activity as of May 2026, indicating no meaningful open-source developer community strategy. SE016, SE017
CE031 No ISO certifications, CE marking, UL listing, safety approval records, or regulatory compliance documentation for Phoenix robots have been publicly disclosed by Sanctuary AI as of May 2026. SE001, SE005
CE032 Figure AI has raised over $675 million at approximately a $2.6 billion valuation (February 2024) and operates a production pilot at BMW manufacturing facilities. SE022, SE029
CE033 1X Technologies develops electric humanoid robots including NEO (home-testing phase 2025) with a lighter form factor and different actuation philosophy from Sanctuary's hydraulics. SE023, SE024
CE034 Agility Robotics deployed Digit bipedal robots at Amazon fulfillment centers and operates its own manufacturing facility (Agility Factory), representing the most commercially validated humanoid deployment as of May 2026. SE027, SE030
CE035 Boston Dynamics' Atlas robot represents decades of R&D investment and has no disclosed commercial humanoid product pathway as of May 2026; Atlas transitioned from hydraulic to electric actuation in 2024. SE026, SE030
CE036 Unitree Robotics offers H1 and G1 humanoid robots at price points substantially below Sanctuary's estimated ASP, targeting research institutions and lighter commercial applications globally. SE028, SE030
CE037 Apptronik's Apollo robot targets logistics and manufacturing with electric actuation and NASA engineering heritage; a Mercedes-Benz pilot was confirmed in 2024. SE025, SE030
CE038 Sanctuary's 21-DoF hydraulic hand provides superior in-hand dexterity compared to the parallel-jaw, gripper, or lower-DoF end-effectors used by most electric humanoid competitors including Figure, Agility, and Unitree. SE001, SE011
CE039 IEEE Spectrum has published critical analyses questioning whether humanoid robotics valuations and timelines reflect actual technological readiness for commercial deployment at scale. SE012, SE013
CE040 Rodney Brooks, a pioneer in mobile robotics, has documented predictions indicating that dexterous manipulation and humanoid deployment timelines consistently underestimate the difficulty of real-world deployment at commercial scale. SE014, SE012
CU001 Sanctuary AI targets B2B industrial customers in automotive manufacturing, warehousing, retail operations, and food service, as stated on its official homepage and news releases. SU001, SU002
CU002 Sanctuary AI has not disclosed any customer count in public communications as of May 2026. SU003, SU004
CU003 Sanctuary AI officially describes its go-to-market target as industrial operators seeking to automate structured manual tasks through a co-pilot model in which human supervisors oversee Phoenix during initial task learning. SU001, SU002
CU004 Sanctuary AI has not disclosed any annual recurring revenue (ARR), contract value, or revenue run rate in any public filing, press release, or investor communication as of May 2026. SU004, SU016
CU005 Sanctuary AI's official website contains no customer testimonials, reference accounts, case studies, or logo walls as of May 2026. SU001, SU002
CU006 Sanctuary AI and Magna International announced a strategic partnership on April 11, 2024 to deploy Phoenix humanoid robots in Magna's manufacturing facilities for structured task automation. SU003, SU019
CU007 The Magna International partnership announced in April 2024 is the only named customer relationship officially announced by Sanctuary AI as of May 2026. SU003, SU004
CU008 No follow-up press releases, outcome data, or case studies confirming the results of the Magna pilot have been published by Sanctuary AI or Magna International for a period of 13 months after the initial April 2024 announcement. SU003, SU004
CU009 The Magna engagement is classified as a pilot deployment rather than a full production agreement; no commercial contract details or deployment scale figures have been disclosed. SU003, SU019
CU010 No Magna International executive quote or endorsement of Phoenix's effectiveness appeared in any publicly accessible source reviewed for this chapter as of May 2026. SU019, SU020
CU011 The absence of any public reference to the Magna partnership in Sanctuary AI's 2025 news releases suggests the pilot may have concluded or been placed on hold without a commercial production agreement. SU004, SU022
CU012 Figure AI announced a production deployment partnership with BMW Manufacturing Co. in January 2024 for humanoid robots performing structured assembly tasks in automotive manufacturing. SU014, SU023
CU013 Agility Robotics has been commercially deploying its Digit humanoid robot in Amazon fulfillment centers and GXO Logistics warehouse operations since 2023–2024 in a confirmed production deployment. SU015, SU024
CU014 1X Technologies has deployed humanoid robots for commercial warehouse operations in Scandinavia in confirmed commercial customer engagements as of 2024–2025. SU013, SU025
CU015 Sanctuary AI's disclosed customer traction — one unconfirmed pilot with Magna — is materially weaker than humanoid robot peers Figure AI, Agility Robotics, and 1X Technologies, all of which have confirmed production customers as of May 2026. SU015, SU014
CU016 Boston Dynamics' Spot quadruped robot has multiple confirmed enterprise customers including Ford, BP, and several construction and utility companies across North America and Europe, serving as a commercial precedent for robotics deployment timelines. SU008, SU025
CU017 The gap between Sanctuary AI's customer evidence and that of key competitors suggests the company remains in a pre-commercial phase as of May 2026, consistent with broader industry patterns but a differentiating risk factor. SU010, SU011
CU018 Morgan Stanley's February 2025 humanoid robotics IP ranking placed Sanctuary AI at #3 globally, reflecting intellectual property strength rather than commercial customer traction. SU007, SU019
CU019 Sanctuary AI's go-to-market strategy is B2B direct enterprise sales, with Microsoft as a technology collaboration partner, and Accenture and Workday Ventures as strategic investors who could serve as channel relationships. SU001, SU003
CU020 Sanctuary AI has not publicly disclosed any information about its sales team size, quota structure, sales cycle length, or customer acquisition cost in any source reviewed. SU004, SU016
CU021 Magna International's dual role as both a strategic investor in Sanctuary AI's Series B and the only named pilot customer reduces the independence and strength of Magna as customer proof evidence. SU005, SU003
CU022 Sanctuary AI positions Phoenix as a workforce supplement rather than replacement, targeting industrial operators in sectors facing structural labor shortages including logistics, automotive manufacturing, and retail. SU001, SU002
CU023 No third-party review platform, procurement database, or customer satisfaction tool lists Sanctuary AI with customer reviews or satisfaction scores as of May 2026. SU001, SU002
CU024 Sanctuary AI has not disclosed any NRR, GRR, customer churn rate, or contract renewal metric in any publicly available communication as of May 2026. SU004, SU016
CU025 Sanctuary AI's reliance on a single unconfirmed pilot engagement creates an extreme customer concentration risk: any setback in the Magna relationship represents 100% of known customer traction. SU007, SU009
CU026 Independent industry analysts and researchers have expressed skepticism about the near-term commercial viability of general-purpose humanoid robots, noting unproven deployment economics, complex integration challenges, and labor substitution uncertainties. SU010, SU011
CU027 Rodney Brooks, co-founder of iRobot and a leading robotics researcher, has publicly written that general-purpose humanoid robots will not achieve meaningful commercial scale within the next decade due to unresolved task generalization and deployment economics challenges. SU011, SU021
CU028 IEEE Spectrum's "humanoid robot bubble" analysis noted that investor enthusiasm in the humanoid robotics sector materially outpaces actual commercial deployment rates, with many announced partnerships not progressing to scaled production. SU010, SU025
CU029 Suzanne Gildert, co-founder and Chief Science Officer of Sanctuary AI, departed in a leadership transition announced November 9, 2024, removing a foundational technical leader whose credibility was an asset in enterprise customer conversations. SU027, SU018
CU030 The 24-month period from May 2024 to May 2026 produced zero publicly disclosed new customer or commercial partnership announcements for Sanctuary AI following the Magna pilot and the $100M Series B. SU004, SU022
CU031 Sanctuary AI's stated pursuit of general-purpose robot applications across multiple verticals simultaneously — manufacturing, logistics, retail, food service — may spread limited commercial resources across too many adoption challenges for a pre-revenue stage. SU001, SU022
CU032 Strategic investors including Accenture, Microsoft, and Workday Ventures represent plausible channel relationships for enterprise customer introduction, but no revenue- sharing or customer referral agreements have been publicly disclosed. SU001, SU019
CU033 Sanctuary AI and Microsoft presented together at Hannover Messe in April 2025, demonstrating industrial automation applications to potential manufacturing customers, but no commercial deal emerged from that event in public disclosures. SU003, SU004
CU034 The co-pilot model — requiring a human supervisor during initial Phoenix deployments — likely extends the sales cycle compared to competitors offering more autonomous operation, since customers must budget for both robot hardware and supervisory staffing costs. SU001, SU017
CU035 Canadian tech media including Betakit and The Logic have not reported any new named commercial customer announcements for Sanctuary AI beyond Magna during their coverage through mid-2026. SU018, SU020
CU036 The humanoid robotics industry shows highly variable customer acquisition rates in 2024–2026, with leading companies achieving production deployments while most others remain in extended pilot or pre-commercial phases. SU008, SU025
CU037 Given Sanctuary AI's US$100M Series B raise in June 2024, the absence of publicly disclosed paying customers in the subsequent 24 months through May 2026 is consistent with either a deliberate technology-first strategy or slower-than-expected commercial progress. SU005, SU007
CU038 Sanctuary AI emphasizes its Carbon AI cognitive system and 21-DoF hydraulic dexterous hand as differentiators for complex manipulation tasks, but no customer has publicly confirmed these capabilities as decision factors in a deployment. SU001, SU002
CU039 Rest of World and international tech media have not reported any Sanctuary AI commercial deployments or customer relationships outside North America as of mid-2026, consistent with a domestic-only commercial footprint. SU026, SU022
CU040 Sanctuary AI's sustained focus on technology milestones (21-DoF hand Dec 2024, Gen 8 Phoenix Dec 2024, sim-to-real transfer Mar 2025, tactile sensors Feb 2025) rather than customer contract announcements in 2025 indicates a technology-first prioritization over near-term commercial scale. SU003, SU025
CR001 Sanctuary AI's Phoenix robot uses a 21-degree-of-freedom hydraulic hand that requires pressurized fluid circuits subject to seal degradation, leakage, and contamination risks not present in electric-actuated competitors. SR012, SR019
CR002 Sanctuary AI's Carbon AI cognitive software relies on Microsoft Azure cloud infrastructure for inference, creating a single point of failure risk in environments requiring air-gap, offline operation, or high-frequency low-latency response. SR012, SR018
CR003 Sanctuary AI has demonstrated sim-to-real transfer of dexterous manipulation policies as recently as March 2025, but has not disclosed production deployment reliability metrics, error rates, or uptime data that would validate the transfer at commercial scale. SR012, SR019
CR004 Hydraulic actuation creates specific contamination risks in food-grade, pharmaceutical, and semiconductor manufacturing environments where hydraulic fluid spills are a regulatory and product liability issue distinct from electric-actuated robot failure modes. SR002, SR006
CR005 Morgan Stanley ranked Sanctuary AI #3 globally in humanoid IP as of February 2025, but many of the company's patents remain pending, meaning competitors can study published applications and design around claims before grant. SR016, SR019
CR006 Sanctuary AI has not publicly disclosed the grant status of its core hydraulic hand or Carbon AI patents, seal life specifications, or patent claim scope, making it impossible to independently assess IP moat strength. SR012, SR016
CR007 Hydraulic actuation components including specialized seals, high-precision pumps, and hydraulic fluid management systems represent supply chain dependencies that electric-actuated competitors do not face. SR002, SR012
CR008 Sanctuary AI has disclosed no commercial customers, no revenue, and no confirmed production deployments as of May 2026; the company's commercial viability remains entirely undemonstrated in the public record. SR012, SR016, SR024
CR009 The Magna International strategic manufacturing pilot, announced April 11, 2024, has had no publicly disclosed follow-on announcement, continuation, or commercial outcome as of May 2026. SR017, SR019, SR020
CR010 Figure AI has secured a production partnership with BMW Group for automotive manufacturing deployment, representing direct competition in Sanctuary AI's target automotive market segment. SR015, SR023
CR011 Agility Robotics has deployed Digit robots in Amazon warehouse facilities commercially, demonstrating that a competitor has already captured the warehouse automation market segment that Sanctuary AI also targets. SR008, SR029, SR023
CR012 1X Technologies has raised over $110M and reports active commercial deployments, creating direct competition for the enterprise humanoid robot market Sanctuary AI is pursuing. SR010, SR028, SR023
CR013 Unitree and UBTech offer significantly lower-cost humanoid robots that could commoditize the hardware layer, with Unitree G1 priced at approximately USD $16,000, putting pressure on the premium hardware pricing that Sanctuary AI would need to sustain its business model. SR005, SR022
CR014 All principal humanoid robot competitors — Figure AI, 1X Technologies, Agility Robotics, Boston Dynamics, Unitree — use electric actuation, creating an ecosystem standardization dynamic that disadvantages Sanctuary AI's hydraulic platform. SR008, SR009, SR015, SR028, SR029
CR015 Boston Dynamics has established enterprise sales of Atlas and Spot robots across multiple industrial verticals with years of operational data and customer relationships that Sanctuary AI does not yet have. SR009, SR030
CR016 Goldman Sachs projected the humanoid robot market at USD $38B by 2035, but IEEE Spectrum and independent robotics experts have raised substantive concerns about near-term adoption timelines that could delay market development. SR014, SR021, SR025
CR017 Sanctuary AI has not disclosed a replacement for the Magna International manufacturing partnership as of May 2026, creating uncertainty about the company's path to commercial- volume Phoenix robot production. SR012, SR017, SR019
CR018 Transitioning from R&D prototype production to commercial-volume manufacturing requires capital-intensive investment in tooling, quality assurance, supply chain, and field service infrastructure that Sanctuary AI has not publicly confirmed acquiring. SR012, SR016
CR019 Co-founder Suzanne Gildert departed Sanctuary AI in a leadership transition announced November 9, 2024; no replacement Chief Science Officer or Chief Technology Officer has been publicly named as of May 2026. SR019, SR024, SR027
CR020 No publicly confirmed head of engineering, CTO, or CSO has been identified at Sanctuary AI in the 18 months following Suzanne Gildert's November 2024 departure, creating a material technical leadership continuity uncertainty. SR012, SR024
CR021 Talent competition for AI robotics engineers from hyperscalers, autonomous vehicle companies, and well-funded humanoid startups including Figure AI, 1X Technologies, and Physical Intelligence creates ongoing retention risk for Sanctuary AI's engineering team. SR013, SR023
CR022 Sanctuary AI's headcount is not publicly disclosed, making it impossible to assess organizational capacity for scaling from R&D to commercial deployment, or to evaluate attrition from the November 2024 leadership transition. SR012, SR016
CR023 Sanctuary AI's commercial deployment transition requires sustained engineering team stability across robotics hardware, machine learning, tactile sensing, and enterprise software disciplines simultaneously, which is a high-risk organizational requirement. SR012, SR021
CR024 Sanctuary AI has never disclosed revenue, ARR, burn rate, or operating cash flow; its cash runway cannot be independently assessed from public information as of May 2026. SR016, SR024
CR025 Sanctuary AI raised US$100M in a Series B round (June 2024) and has received a CAD $30M government loan from Export Development Canada; total financing disclosed to date exceeds approximately CAD $190M across all known rounds. SR020, SR016
CR026 Without disclosed financials, investors cannot assess whether Sanctuary AI's capital base is sufficient to reach a commercial revenue milestone, making financial risk assessment dependent entirely on private diligence. SR016, SR001
CR027 Hardware commercialization for humanoid robots requires significant capital expenditure on tooling, manufacturing equipment, supply chain, quality assurance, and field service infrastructure; comparable pre-revenue hardware companies typically require multiple funding rounds before reaching unit-margin positive operations. SR023, SR021
CR028 Industrial robotics sales cycles typically range from 12 to 24 or more months from initial customer engagement to signed deployment contract, extending Sanctuary AI's time to revenue and amplifying its cash consumption before commercial operations begin. SR023, SR026
CR029 Sanctuary AI's dependence on venture capital continuation means any deterioration in AI robotics investor sentiment — whether from macro conditions, competitor failures, or the company's own commercial setbacks — could constrain its next fundraising round. SR016, SR003
CR030 Export Development Canada's CAD $30M loan creates a fixed repayment obligation for Sanctuary AI regardless of revenue performance; loan terms and covenants are not publicly disclosed. SR019, SR020
CR031 Microsoft is Sanctuary AI's most strategically concentrated dependency, providing Azure cloud infrastructure for Carbon AI inference, AI model development collaboration, and joint go-to-market exposure; no alternative cloud or AI model partner has been disclosed. SR012, SR018, SR019
CR032 No OSHA standard, WorkSafeBC guideline, or Canadian federal workplace safety regulation specifically governs bipedal humanoid robots operating alongside human workers in industrial settings as of May 2026. SR001, SR002
CR033 Canada's proposed AI regulation (Bill C-27 / Artificial Intelligence and Data Act) remains in parliamentary process as of May 2026 and has not enacted humanoid robot-specific obligations, creating regulatory uncertainty about future compliance requirements. SR001, SR007
CR034 OSHA's existing robot safety standards based on ISO 10218 were designed for stationary industrial robots and do not address the specific hazards of bipedal humanoid robots with advanced dexterous manipulation operating in shared human workspaces. SR001, SR002
CR035 Hydraulic fluid in Sanctuary AI's Phoenix hand creates contamination risk in food, pharmaceutical, and clean-room manufacturing environments; industry procurement specifications for these sectors typically require hydraulic-free designs. SR002, SR006
CR036 Many of Sanctuary AI's patents remain pending as of May 2026, meaning competitors can review published claims and design around them before grant, particularly for the hydraulic hand and Carbon AI cognitive architecture. SR016, SR019
CR037 Sanctuary AI's "general-purpose" robot positioning may delay customer adoption compared to purpose-built solutions for specific tasks such as warehouse picking or welding, where narrow-AI competitors can offer better task-specific performance at lower cost. SR013, SR023
CR038 Sanctuary AI has no publicly disclosed customer case studies, reference accounts, or testimonials as of May 2026, creating a reference customer gap that extends enterprise sales cycles for subsequent customers. SR012, SR027
CR039 Rodney Brooks, former MIT CSAIL director and a leading robotics practitioner, has consistently predicted that general-purpose humanoid robots face material commercial viability barriers extending into the late 2020s. SR013, SR025
CR040 IEEE Spectrum's "humanoid robot bubble" analysis raises structural concerns about whether current investment levels in humanoid robotics are justified by near-term commercial deployment timelines, which is an adverse signal for the entire sector including Sanctuary AI. SR014, SR026
CR041 The Magna International partnership status remains unconfirmed as of May 2026; if discontinued, it removes Sanctuary AI's only disclosed manufacturing partner and automotive-sector customer proof point simultaneously. SR017, SR019
CR042 No public safety incident reports involving Sanctuary AI robots have been found in the research record; however, the absence of deployment scale (no confirmed production units operating commercially) means this absence of incidents is statistically uninformative. SR012, SR001
CR043 Hydraulic fluid, even food-grade variants, poses contamination risk in pharmaceutical and semiconductor manufacturing environments that have zero-contamination specifications; these sectors represent a potentially large humanoid robot market that Sanctuary AI's hydraulic design may be excluded from. SR002, SR006
CR044 Historical general-purpose humanoid robot programmes — Honda ASIMO (discontinued 2022) and SoftBank Pepper (production halted 2021) — both demonstrated strong technology but failed to achieve commercial viability, establishing an adverse historical precedent for general-purpose humanoid commercial deployment. SR014, SR025, SR026
CR045 Worker displacement concerns from humanoid robot deployment in manufacturing could trigger regulatory backlash, union opposition, and political risk that increases enterprise customer hesitation and lengthens procurement timelines. SR022, SR021
CV001 Sanctuary AI has raised approximately $243.8M in total disclosed capital across all funding rounds through May 2024. SV002, SV003
CV002 The most recent Sanctuary AI funding round (Series C per PitchBook, labeled Series B in the GlobeNewswire press release) raised approximately $99.8M–$100M and was led by BDC Capital's Thrive Venture Fund in May–June 2024. SV005, SV002, SV014
CV003 The Series B/C investor roster includes BDC Capital, InBC Strategic Investments, Microsoft, Accenture Ventures, Magna International, Workday Ventures, Bell, Verizon Ventures, Obvious Ventures, and Export Development Canada. SV005, SV006, SV014, SV015
CV004 Export Development Canada provided a CAD $30M (~USD $22M) non-dilutive government loan to Sanctuary AI in 2023 to support R&D into general-purpose robotics. SV010, SV011
CV005 Sanctuary AI has not publicly disclosed a post-money valuation for any funding round as of May 2026; the company has no SEC or SEDAR filing obligation as a private Canadian corporation. SV001, SV002
CV006 Sanctuary AI raised a Series A of approximately CAD $58.5M (~USD $43M) in early 2023, making it the first identified institutional round for the company. SV002, SV004
CV007 BDC Capital Thrive Venture Fund and InBC Strategic Investments separately announced their investment in Sanctuary AI on July 3, 2024, either as part of the June 2024 Series B/C round or as a follow-on close. SV006, SV013, SV014, SV015
CV008 No Series D funding round has been publicly announced or reported for Sanctuary AI as of the May 2026 run date, approximately 24 months after the Series C close. SV001, SV002
CV009 Using a standard 2x–6x capital-raised valuation proxy for pre-revenue deep-tech companies, Sanctuary AI's $243.8M total raised implies an estimated valuation range of $488M–$1.46B, with a midpoint near $730M. SV023, SV033
CV010 The earliest institutional capital raised by Sanctuary AI is a Series A in 2023; any seed or pre-Series A round amount remains undisclosed, making total lifetime capital figures potentially understated. SV002, SV031
CV011 Figure AI raised $675M in February 2024 at a post-money valuation of $2.6B, led by Microsoft, OpenAI, Nvidia, Samsung, and Jeff Bezos. SV016, SV017
CV012 Figure AI secured a commercial manufacturing pilot agreement with BMW following its $2.6B raise, making it the first humanoid robot company to announce a named Tier-1 automotive production partner. SV016, SV017
CV013 Figure AI's $2.6B valuation at $675M raised implies a 3.85x capital-raised multiple; extrapolating this multiple to Sanctuary AI's $243.8M raised suggests a theoretical comparable valuation of approximately $940M before a no-customer discount. SV016, SV023, SV033
CV014 1X Technologies (Norwegian startup) raised approximately $135M by January 2024 and operates the NEO humanoid platform; its valuation is not publicly disclosed. SV018
CV015 Agility Robotics raised approximately $150M in 2023 from Amazon's Industrial Innovation Fund to deploy the Digit humanoid in Amazon warehouses; valuation not disclosed. SV019
CV016 Boston Dynamics was acquired by Hyundai Motor Group in June 2021 for an enterprise value of approximately $1.1B including prior SoftBank investment; this is the only fully disclosed M&A valuation benchmark for a humanoid/advanced robotics company. SV020
CV017 Apptronik raised approximately $350M in 2024 backed by GV (Google Ventures) and Honda to develop the Apollo humanoid; valuation was not publicly disclosed. SV021, SV024
CV018 The range of known or implied valuations for humanoid robot companies spans from approximately $225M (bear-case inferred) to $2.6B (Figure AI disclosed), with most competitors at undisclosed valuations, making peer benchmarking unreliable. SV016, SV024
CV019 Goldman Sachs projects the global humanoid robotics market to reach $38B by 2035, providing a long-term TAM anchor that supports double-digit valuation multiples for leading platform companies that achieve commercial scale by 2028–2030. SV025
CV020 CB Insights identifies over 50 humanoid and service robotics startups competing for enterprise manufacturing, logistics, and healthcare deployments, with few having disclosed commercial revenue as of 2025. SV024
CV021 Morgan Stanley ranked Sanctuary AI #3 globally in humanoid robotics intellectual property in February 2025, upgraded from #4 in November 2024, reflecting a growing patent portfolio that provides defensive value and potential licensing upside. SV006, SV007
CV022 Sanctuary AI's #3 Morgan Stanley patent ranking places its IP above Agility Robotics and 1X Technologies in the humanoid IP hierarchy, though below Figure AI (#1 or #2 in the same ranking), and represents a partial floor to valuation independent of commercial traction. SV006, SV007
CV023 No confirmed commercial customer for Sanctuary AI has been publicly announced as of May 2026; the company has not disclosed any enterprise deployment contract, revenue figure, or customer testimonial. SV001, SV003
CV024 The Magna International manufacturing pilot announced April 2024 has not been followed by a commercial contract announcement as of May 2026; the pilot's outcome—success, ongoing, or discontinued—remains undisclosed. SV001, SV003, SV006
CV025 IEEE Spectrum published "The Humanoid Robot Bubble," arguing that humanoid robots face fundamental cost and operational-time barriers that current industry timelines cannot overcome, citing specific benchmarks where robots fall short of human performance. SV028
CV026 Rodney Brooks, former MIT AI Lab director and serial robotics entrepreneur, has publicly predicted on a dated scorecard that humanoid robots will not achieve general-purpose commercial scale by 2030, citing consistent overestimation of near-term AI and motor capability. SV029, SV030
CV027 Figure AI's $2.6B valuation with confirmed BMW manufacturing pilot creates a competitive funding differential: Sanctuary AI has raised $431M less ($675M vs $243.8M) and lacks any confirmed commercial customer, potentially reducing its attractiveness to institutional investors seeking humanoid exposure. SV016, SV024
CV028 Co-founder Suzanne Gildert (Chief Science Officer) departed Sanctuary AI in November 2024 per a leadership transition announcement, representing a key-person risk to the company's proprietary AI methodology and research continuity. SV006, SV031
CV029 No revenue figure, ARR, or financial metric has been publicly disclosed by Sanctuary AI as of May 2026; the company's revenue is assumed to be near-zero or zero pending confirmation of a commercial customer. SV001, SV003
CV030 As a private corporation incorporated in British Columbia, Canada, Sanctuary AI has no legal obligation to file financial statements with the SEC, SEDAR, or any other public market regulator, making independent financial verification impossible without data-room access. SV001, SV031
CV031 The 2023–2025 humanoid robotics hype cycle, characterized by large venture rounds at high multiples and limited commercial revenue across the sector, shares structural characteristics with prior technology bubbles (2021 SPAC/EV, 2000 dotcom) that produced significant valuation compression. SV027, SV028
CV032 Sanctuary AI's estimated annual operating expense of $60M–$90M (benchmarked against comparable Series C humanoid robotics companies) implies that Series C proceeds of ~$100M would be substantially exhausted within 12–20 months, raising the prospect of a 2025–2026 funding requirement. SV024, SV025
CV033 Cumulative venture capital preference stack of approximately $243.8M at 1x non-participating preferred means any Sanctuary AI exit below $244M would return nothing to common equity holders including employee option pools. SV032, SV033
CV034 If Series C investors hold 1x participating preferred, the preference stack captures the first ~$244M of any exit proceeds plus a proportionate share of the residual, significantly elevating the breakeven exit price for common shareholders. SV032, SV033
CV035 Humanoid robot companies historically require 5–10 years from first prototype to scalable commercial deployment; Sanctuary AI's Phoenix was demonstrated in 2022 and as of 2026 has not achieved confirmed commercial scale, placing it on track with but not ahead of historical robotics commercialization timelines. SV029, SV035
CV036 InBC Strategic Investments is a British Columbia provincial government-backed fund with a mandate to invest in BC-headquartered technology companies; its investment in Sanctuary AI reflects both financial and economic-development objectives. SV015
CV037 BDC Capital's Thrive Venture Fund focuses on Canadian deep-technology startups with breakthrough potential; its participation as Series C lead investor provides government-backed validation but may reflect national-champion policy objectives in addition to commercial return expectations. SV014
CV038 Obvious Ventures is a mission-aligned impact VC firm with a portfolio spanning AI and sustainable automation; its investment in Sanctuary AI is consistent with its published focus but provides limited commercial signal given its non-industrial orientation. SV022
CV039 Verizon Ventures invested in Sanctuary AI, suggesting a hypothesis that 5G-connected robotics deployments could enable remote teleoperation, fleet management, or edge-AI applications in Sanctuary's future commercial products. SV002, SV003
CV040 Sanctuary AI's implied valuation range of $400M–$1.5B is unconfirmed and constructed entirely from funding-multiple benchmarks; actual valuation could be higher (post-Series D at a disclosed price) or lower (down-round) depending on commercial progress through 2026. SV002, SV023, SV033
CV041 A probability-weighted expected valuation across the three scenarios yields approximately $618M: (0.20 × $1,150M) + (0.55 × $600M) + (0.25 × $225M) = $230M + $330M + $56M ≈ $616–$618M base estimate. This weighted midpoint is the most analytically defensible single-point proxy absent disclosed figures. SV023, SV033
来源
编号出版方标题引文
SO001 Sanctuary AI Sanctuary AI Homepage The Carbon™ AI control system mimics subsystems found in the human brain, such as memory, sight, sound, and touch. When applied with Phoenix™ general purpose robots, our systems will be able to take on just about any human task.
SO002 Sanctuary AI Sanctuary AI About Page
SO003 Sanctuary AI Sanctuary AI Careers
SO004 GlobeNewswire Sanctuary AI Raises US$100M in Series B Financing Sanctuary AI has raised US$100 million in Series B financing led by new and existing investors including Microsoft, Accenture, Bell, Verizon Ventures, BDC Capital, Export Development Canada, Magna, Workday Ventures, and Evok Innovations.
SO005 Artificial Intelligence News Sanctuary AI raises $100M Series B
SO006 Axios Sanctuary AI series B funding
SO007 PitchBook Sanctuary AI Company Profile
SO008 Tracxn Sanctuary Cognitive Systems Company Profile
SO009 Export Development Canada (Wikipedia) Export Development Canada
SO010 Sanctuary AI Sanctuary AI News Archive (Wayback Machine 2024) Today we are announcing a significant transition within our leadership team.
SO011 Sanctuary AI Sanctuary AI News Archive (Wayback Machine 2025)
SO012 Sanctuary AI Sanctuary AI News Page (Current) In their latest video, their proprietary hydraulic hand autonomously manipulates a lettered cube, continuously reorienting it.
SO013 Sanctuary AI Sanctuary AI Technology Page Sanctuary's approach is vertically integrated. From motor design to generative AI algorithms for superhuman task performance, our solutions are designed, built, owned, and patented by Sanctuary.
SO014 Wikipedia Microsoft
SO015 Wikipedia Magna International
SO016 Wikipedia D-Wave Systems
SO017 Rodney Brooks My Dated Predictions Having ideas is easy. Turning them into reality is hard. Turning them into being deployed at scale is even harder.
SO018 IEEE Spectrum Humanoid Robot Bubble
SO019 IEEE Spectrum Humanoid Robot Coverage
SO020 Wikipedia Accenture
SO021 Wikipedia Vancouver
SO022 LinkedIn Sanctuary AI Company LinkedIn Page
SO023 CB Insights Sanctuary AI Company Profile
SO024 GitHub Sanctuary AI Organization
SO025 US Securities and Exchange Commission EDGAR Search for Sanctuary
SO026 Sanctuary AI Sanctuary AI Blog
SO027 Sanctuary AI Sanctuary AI YouTube Channel
SO028 Wikipedia Humanoid Robot
SM001 Goldman Sachs Research The Global Market for Robots Could Reach $38 Billion by 2035 The global market for humanoid robots could reach $38 billion by 2035 in an optimistic scenario.
SM002 International Federation of Robotics IFR Press Releases and Statistics
SM003 U.S. Bureau of Labor Statistics Laborers and Material Movers, Hand: Occupational Outlook Handbook
SM004 U.S. Bureau of Labor Statistics Job Openings and Labor Turnover Survey (JOLTS) - Table 1
SM005 World Economic Forum The Future of Jobs Report 2025
SM006 CB Insights Humanoid Robots Startups and Funding Landscape
SM007 Wikipedia Industrial Robot
SM008 Wikipedia Service Robot
SM009 The Economist Humanoid Robots Business Coverage
SM010 Wikipedia Cobot (Collaborative Robot)
SM011 IEEE Spectrum Humanoid Robots 2024 Industry Review
SM012 Wikipedia International Federation of Robotics
SM013 IEEE Spectrum General-Purpose Humanoid Robots Coverage
SM014 Sanctuary AI Sanctuary AI Homepage
SM015 Sanctuary AI Sanctuary AI Technology Page
SM016 IEEE Spectrum Humanoid Robot Bubble Analysis
SM017 Rodney Brooks My Dated Predictions Having ideas is easy. Turning them into reality is hard. Turning them into being deployed at scale is even harder.
SM018 Wikipedia Humanoid Robot
SM019 Wikipedia Magna International
SM020 PitchBook Sanctuary AI Company Profile
SM021 Sanctuary AI Sanctuary AI News Archive (Wayback Machine 2024)
SM022 Sanctuary AI Sanctuary AI News Archive (Wayback Machine 2025)
SM023 Sanctuary AI Sanctuary AI News Page (Current)
SM024 Axios Sanctuary AI Series B Funding
SM025 Tracxn Sanctuary Cognitive Systems Company Profile
SM026 U.S. Securities and Exchange Commission EDGAR Company Search - Sanctuary
SM027 Wikipedia Accenture
SM028 Wikipedia Robot Density
SM029 The Robot Report Sanctuary AI Coverage
SM030 Artificial Intelligence News Sanctuary AI Raises $100M Series B
SP001 Wikipedia Figure AI
SP002 Wikipedia 1X Technologies
SP003 Wikipedia Agility Robotics
SP004 Wikipedia Apptronik
SP005 Wikipedia Boston Dynamics
SP006 Wikipedia Tesla Optimus
SP007 Wikipedia Unitree Robotics
SP008 Wikipedia ASIMO
SP009 Wikipedia Pepper (robot)
SP010 Figure AI Figure AI Official Website
SP011 1X Technologies 1X Technologies Official Website
SP012 Agility Robotics Agility Robotics Official Website
SP013 Apptronik Apptronik Official Website
SP014 Boston Dynamics Boston Dynamics Official Website
SP015 Unitree Robotics Unitree Robotics Official Website
SP016 IEEE Spectrum Figure AI Humanoid Robot Coverage
SP017 IEEE Spectrum Agility Robotics Coverage
SP018 Rodney Brooks Predictions Scorecard 2024 January 01
SP019 Wikipedia Atlas Robot
SP020 Wikipedia SoftBank Robotics
SP021 Sanctuary AI Sanctuary AI Homepage
SP022 Sanctuary AI Sanctuary AI News Archive (Wayback Machine 2024)
SP023 GlobeNewswire Sanctuary AI Raises US$100M in Series B Financing
SP024 CB Insights Humanoid Robots Startups and Funding Landscape
SP025 PitchBook Sanctuary AI Company Profile
SP026 IEEE Spectrum Humanoid Robot Bubble Analysis
SP027 IEEE Spectrum Humanoid Robot Coverage
SP028 IEEE Spectrum Humanoid Robots 2024 Industry Review
SP029 Wikipedia Humanoid Robot
SP030 Rodney Brooks My Dated Predictions Having ideas is easy. Turning them into reality is hard. Turning them into being deployed at scale is even harder.
SI001 BetaKit Sanctuary AI Phoenix Gen 7 Coverage
SI002 BNN Bloomberg Sanctuary AI Coverage
SI003 GlobeNewswire Sanctuary AI Raises US$100M in Series B Financing Sanctuary AI, a leader in the development of human-like intelligence for general purpose robots, today announced the close of its US$100M Series B financing.
SI004 Artificial Intelligence News Sanctuary AI Raises 100M Series B
SI005 Wikipedia Sanctuary AI
SI006 Wikipedia Geordie Rose
SI007 Sanctuary AI Sanctuary AI Series B Announcement
SI008 PitchBook Sanctuary AI Financial Profile
SI009 LinkedIn Sanctuary AI Company Profile
SI010 IEEE Spectrum General-Purpose Humanoid Robots Industry Coverage
SI011 IEEE Spectrum General-Purpose Humanoid Robots Industry Analysis
SI012 Axios Sanctuary AI Series B Funding Coverage
SI013 Government of Canada Government of Canada Investing in Sanctuary AI General Purpose Robotics
SI014 BC Business Sanctuary AI Vancouver Humanoid Robot Coverage
SI015 Built In Sanctuary AI Company Overview
SI016 IEEE Spectrum Humanoid Robot Bubble The commercial deployment of humanoid robots faces fundamental economic barriers that current funding levels cannot paper over.
SI017 Rodney Brooks My Dated Predictions (Humanoid Robots) Having ideas is easy. Turning them into reality is hard. Turning them into being deployed at scale is even harder.
SI018 Sanctuary AI Sanctuary AI Technology Page
SI019 Sanctuary AI Sanctuary AI About Page
SI020 Wikipedia Export Development Canada
SI021 CB Insights Sanctuary AI Company Profile
SI022 Crunchbase Sanctuary AI Organization Profile
SI023 Wikipedia Accenture
SI024 Wikipedia Microsoft
SI025 Wikipedia Magna International
SE001 Sanctuary AI Sanctuary AI Homepage The Carbon™ AI control system mimics subsystems found in the human brain, such as memory, sight, sound, and touch.
SE002 Sanctuary AI Sanctuary AI Technology Page From motor design to generative AI algorithms for superhuman task performance, our solutions are designed, built, owned, and patented by Sanctuary.
SE003 Sanctuary AI Sanctuary AI News Archive 2024 (Wayback Machine)
SE004 Sanctuary AI Sanctuary AI News Archive 2025 (Wayback Machine)
SE005 Sanctuary AI Sanctuary AI News Page (Current) In their latest video, their proprietary hydraulic hand autonomously manipulates a lettered cube, continuously reorienting it.
SE006 The Robot Report Sanctuary AI unveils 21-DoF hydraulic hand
SE007 The Robot Report Sanctuary AI unveils eighth generation of its Phoenix humanoid robots
SE008 The Robot Report Sanctuary AI demonstrates sim-to-real dexterous manipulation
SE009 The Robot Report Sanctuary AI unveils tactile sensor technology
SE010 The Robot Report Sanctuary AI at Microsoft Hannover Messe 2025
SE011 IEEE Spectrum Sanctuary AI Phoenix Humanoid Robot
SE012 IEEE Spectrum Is There a Humanoid Robot Bubble?
SE013 IEEE Spectrum The Humanoid Robots of 2024
SE014 Rodney Brooks My Dated Predictions Having ideas is easy. Turning them into reality is hard. Turning them into being deployed at scale is even harder.
SE015 Google Patents Sanctuary Cognitive Systems patent portfolio search
SE016 GitHub Sanctuary AI GitHub Organization
SE017 GitHub Sanctuary AI Organization Repositories
SE018 Hacker News Hacker News discussions referencing Sanctuary AI
SE019 TechCrunch Sanctuary AI coverage (TechCrunch tag page)
SE020 Ars Technica Sanctuary AI raises $100M for humanoid robots
SE021 MIT Technology Review Sanctuary AI Series B coverage
SE022 Figure AI Figure AI Homepage
SE023 1X Technologies 1X Technologies Homepage
SE024 1X Technologies NEO Humanoid Product Page
SE025 Apptronik Apollo Humanoid Robot Product Page
SE026 Boston Dynamics Atlas Robot Page
SE027 Agility Robotics Agility Robotics Homepage
SE028 Unitree Robotics Unitree H1 Humanoid Robot Page
SE029 Wikipedia Sanctuary AI
SE030 Wikipedia Humanoid Robot
SU001 Sanctuary AI Sanctuary AI Official Homepage
SU002 Sanctuary AI Sanctuary AI News Page
SU003 Sanctuary AI (via Wayback Machine) Sanctuary AI News Archive 2024 — Magna Partnership Announcement Apr 11 2024: Magna strategic partnership announced — Phoenix humanoid robot pilot in Magna manufacturing facility
SU004 Sanctuary AI (via Wayback Machine) Sanctuary AI News Archive 2025 — Post-Magna Period
SU005 GlobeNewswire Sanctuary AI Raises US$100M in Series B Financing Magna International is named as a strategic investor in Sanctuary AI's US$100M Series B financing.
SU006 Wikipedia Magna International
SU007 PitchBook Sanctuary AI Company Profile
SU008 CB Insights Humanoid Robots Startup Landscape
SU009 Tracxn Sanctuary Cognitive Systems Company Profile
SU010 IEEE Spectrum The Humanoid Robot Bubble The gap between humanoid robot announcements and actual commercial deployments remains wide across the sector.
SU011 Rodney Brooks My Dated Predictions — Robots and AI Rodney Brooks has consistently predicted that general-purpose humanoid robots will not achieve commercial scale deployment within the next decade.
SU012 Figure AI Figure AI Official Homepage
SU013 1X Technologies 1X Technologies Official Homepage
SU014 Wikipedia Figure AI
SU015 Wikipedia Agility Robotics
SU016 CB Insights Sanctuary AI Company Profile
SU017 TechCrunch Sanctuary AI Coverage — TechCrunch Tag
SU018 Law and Style Sanctuary AI — Vancouver Humanoid Robotics Coverage
SU019 Betakit Sanctuary AI Coverage — Betakit Canadian Tech Media
SU020 The Logic Sanctuary AI Coverage — The Logic Canadian Business Media
SU021 Rodney Brooks Rodney Brooks — Humanoid Robot Analysis and Commentary Brooks argues that the engineering and economic challenges of general-purpose humanoid robots are routinely underestimated, and that stated commercial timelines are typically five to ten times longer in practice than publicly announced.
SU022 The Verge Humanoid Robot Coverage — The Verge
SU023 Figure AI Figure AI News Page — Customer and Partnership Announcements
SU024 Agility Robotics Agility Robotics News — Commercial Deployment Announcements
SU025 IEEE Spectrum Humanoid Robotics Industry Coverage
SU026 Rest of World Humanoid Robot Global Coverage
SU027 Wikipedia Suzanne Gildert
SU028 The Canadian Press Sanctuary AI — Canadian Press Business Coverage
SR001 U.S. Securities and Exchange Commission EDGAR Full-Text Search — Sanctuary AI public filings
SR002 Wikipedia Robot safety — Wikipedia
SR003 EU Startups EU Startups — Sanctuary AI coverage
SR004 Neura Robotics Neura Robotics — Humanoid Robots Homepage
SR005 UBTech Robotics UBTech Robotics — Walker Series Humanoid Robots
SR006 Wikipedia Generative artificial intelligence — Wikipedia
SR007 OpenCorporates OpenCorporates — Sanctuary Cognitive Systems Corporation (BC Registry)
SR008 Agility Robotics Agility Robotics — About Page
SR009 Boston Dynamics Boston Dynamics — About Page
SR010 1X Technologies 1X Technologies — About Page
SR011 Sanctuary AI Sanctuary AI — Official X / Twitter Profile
SR012 Sanctuary AI Sanctuary AI — Official Website
SR013 Rodney Brooks My Dated Predictions — Rodney Brooks Brooks has consistently predicted that general-purpose humanoid robots face 10+ year timelines to commercial viability, citing the gap between demonstration performance and real-world deployment reliability.
SR014 IEEE Spectrum The Humanoid Robot Bubble — IEEE Spectrum
SR015 Figure AI Figure AI — Official Website
SR016 PitchBook Sanctuary AI — PitchBook Company Profile
SR017 Wikipedia Magna International — Wikipedia
SR018 Wikipedia Microsoft — Wikipedia
SR019 Sanctuary AI (archived) Sanctuary AI News Archive 2024 — web.archive.org Nov 9 2024: Leadership transition announced. Apr 11 2024: Magna strategic partnership announced. May 1 2024: Microsoft collaboration confirmed.
SR020 GlobeNewswire Sanctuary AI Raises US$100M in Series B Financing
SR021 IEEE Spectrum Humanoid Robots in 2024 — IEEE Spectrum
SR022 Rest of World Humanoid robot coverage — Rest of World
SR023 CB Insights Humanoid Robot Startups Landscape — CB Insights
SR024 Wikipedia Sanctuary AI — Wikipedia
SR025 Rodney Brooks Predictions Scorecard: 2024 January 01 — Rodney Brooks
SR026 IEEE Spectrum Humanoid Robot Coverage — IEEE Spectrum
SR027 The Logic The Logic — Sanctuary AI coverage
SR028 1X Technologies 1X Technologies — Official Website
SR029 Agility Robotics Agility Robotics — Official Website
SR030 Boston Dynamics Boston Dynamics — Official Website
SV001 Sanctuary AI Sanctuary AI — News Index
SV002 PitchBook Sanctuary AI Company Profile — Funding & Investors
SV003 CB Insights Sanctuary AI Company Profile
SV004 Tracxn Sanctuary Cognitive Systems — Company Profile
SV005 GlobeNewswire Sanctuary AI Raises US$100M in Series B Financing Sanctuary AI today announced US$100M in Series B financing to accelerate development and deployment of the Phoenix general purpose robot.
SV006 Wayback Machine / Sanctuary AI Sanctuary AI News Archive — 2024 Milestones
SV007 Wayback Machine / Sanctuary AI Sanctuary AI News Archive — 2025 Milestones
SV008 Axios Sanctuary AI raises $100M Series B
SV009 Artificial Intelligence News Sanctuary AI raises $100M Series B to deploy humanoid robots
SV010 Government of Canada / ISED Government of Canada Investing in Sanctuary AI — $30M Loan for R&D The Government of Canada is investing $30 million in Vancouver-based Sanctuary AI to support research and development into general purpose robotics.
SV011 BNN Bloomberg Sanctuary AI Coverage
SV012 BC Business Sanctuary AI Vancouver — Humanoid Robot Startup
SV013 Betakit Sanctuary AI — Betakit Coverage
SV014 BDC Capital BDC Capital Thrive Venture Fund — Program Overview
SV015 InBC Strategic Investments InBC Strategic Investments — Portfolio
SV016 Wikipedia Figure AI — Wikipedia
SV017 IEEE Spectrum Figure AI and the Humanoid Robot Race
SV018 Wikipedia 1X Technologies — Wikipedia
SV019 Wikipedia Agility Robotics — Wikipedia
SV020 Wikipedia Boston Dynamics — Wikipedia
SV021 Apptronik Apptronik — Apollo Robot
SV022 Obvious Ventures Obvious Ventures — Portfolio
SV023 Wikipedia Venture Capital — Wikipedia
SV024 CB Insights Humanoid Robots Startups Landscape
SV025 Goldman Sachs The Global Market for Robots Could Reach $38 Billion by 2035
SV026 World Economic Forum The Future of Jobs Report 2025
SV027 The Economist Humanoid Robots — The Economist
SV028 IEEE Spectrum The Humanoid Robot Bubble Humanoid robots face fundamental cost and performance barriers that current investment levels cannot overcome on the timelines promoters project.
SV029 Rodney Brooks My Dated Predictions — Rodney Brooks Commercial general-purpose robots in the home or at human-level dexterity in manufacturing will not be widespread by 2030.
SV030 Rodney Brooks Predictions Scorecard 2024
SV031 Wikipedia Sanctuary AI — Wikipedia
SV032 Wikipedia Preferred Stock — Wikipedia
SV033 Wikipedia Post-Money Valuation — Wikipedia
SV034 Tesla Tesla AI — Optimus Robot
SV035 Wikipedia Robotics — Wikipedia
SV036 SEDAR+ SEDAR+ — Sanctuary AI Filing Search (No Public Filings Found)