初创公司尽调
尽调报告 Industrial / Robotics / Embodied AI Series B / Unicorn 2026-07-23

Psibot

有资历的团队和一线战略投资人支撑了 US$1.48B 独角兽估值,但收入未披露、基准未独立验证、且存在中国特有监管敞口;这个估值押的是执行兑现,不是用业绩倍数定价。

Psibot 同时拿到了顶级具身 AI 团队、平台授权模式和头部战略投资人,估值也站上 $1.48B 独角兽档位——但收入未披露、性能基准来自自报,还暴露在中国特定监管风险下。按当前估值投入前,尽调必须先验证这张执行期权能否兑现。

封面要素

估值 01
1480 USD M [CO015]
成立时间 02
2024 [CO001]
总部 03
Beijing and Shanghai, China [CO001]
投资建议 04
research-more [CV009]

公司概况

Psibot (Lingchu Intelligence / 灵初智能) 是一家中国具身 AI 初创公司,2024 年成立,总部位于北京和上海。公司开发面向机器人灵巧操作的视觉-语言-动作(VLA)模型平台,旗舰系统 Psi R1 已展示 L3 级自主操作能力(30 分钟自主打麻将)。Psibot 不造自有机器人,而是把 AI “大脑”——智能层——授权给机器人制造商,并希望打造中国最大的灵巧手数据集,形成可复利的数据飞轮。公司已在物流仓库分拣场景落地。

官网
psibot.ai
成立时间
2024-01-01
创始人
Viktor Wang, Xiaojie Chai, Yaodong Yang, Yuanpei Chen
创立地点
Beijing and Shanghai, China
总部
Beijing and Shanghai, China
产品
Psibot 的产品是面向灵巧操作的 VLA 模型平台(Psi R1),配套数据生成和仿真流水线,并以硬件智能层的形式授权给机器人 OEM。公司希望借助多个硬件伙伴,拿到更像软件的经济性,同时不背硬件资本开支。
客户
采用 Psibot 模型和数据工具的机器人制造商与 OEM,早期落地场景包括物流仓库分拣和制造自动化。
商业模式
平台 / 智能层授权——Psibot 授权其 VLA “大脑”,并向硬件厂商销售数据和仿真服务,而不是直接销售机器人。
阶段
Series B / Unicorn
融资情况
累计融资超过 US$300M,包括 2026 年 7 月由 Chery Automobile 和 Lens Technology 领投的约 US$100M 轮次,投后估值 US$1.48B;此前还完成约 RMB 2B(约 US$280M)融资。公开来源未披露收入、ARR、利润率、烧钱速度、现金跑道、员工数或清算优先权栈。
[CO001, CO006]

执行摘要

主要优势

  • 平台授权模式切入具身 AI 技术栈里最容易规模化、也最能守住壁垒的一层,有机会在多家 OEM 身上跑出软件式经济性。
  • 创始团队履历罕见:覆盖 Peking University、Stanford / Fei-Fei Li 学术脉络,以及 Alibaba / Tencent 机器人经验。
  • Chery Automobile、Lens Technology 等头部战略投资人同时给资金和产业牵引,降低融资与制造落地风险。
  • 公司想搭建中国最大的灵巧手数据集,把数据飞轮越滚越厚;已展示 L3 操作能力,包括 30 分钟自主麻将。
  • 市场大且扩张快:WAIC 2026 具身 AI 公司合计估值超过 US$14.7B,2026 H1 中国新增 15 家独角兽。

主要风险

  • 收入、ARR、毛利率、烧钱速度、现金跑道和员工数均未披露,US$1.48B 估值更多押注团队、数据与投资人,而不是基本面。
  • 核心能力基准由公司自报,尚无独立验证;分析师估计,超过 85% 的具身 AI 部署仍停留在非生产试点。
  • 中国特定监管与数据访问风险(National Intelligence Law、修订后的 Cybersecurity Law、Data Security Law / PIPL)会压住海外需求和融资可选项。
  • 对 Nvidia 级算力和精密部件的依赖暴露在美国出口管制下;Unitree / DEEP / Leju 即将上市,也可能把估值锚拉回可审计指标,带来 down-round 风险。
  • 18-24 个月现金跑道耗尽后,行业预计在 2027-2028 年进入「淘汰年」;早期牵引也可能来自战略投资人拉动,而非独立第三方需求。

未决问题

  • 经审计财务数据——收入、ARR、毛利率、烧钱速度和现金跑道——以及最新一轮承诺资本与已到账资本的拆分。
  • 完整股权结构表和清算优先权堆叠,用于量化稀释与下行保护。
  • 对 L3 操作能力和 30 分钟自主麻将说法进行独立第三方验证。
  • 可具名、可联系的客户背调名单,并区分独立第三方收入与关联方(Chery / Lens)收入。
  • 中国 AI、数据和网络安全法律下的诉讼、执法与许可记录;鉴于出口管制风险,还需要部件 BOM 与算力供应计划。

目录

Chapter 01

01公司概览

1.1 身份、总部与商业模式

Psibot 在中文来源中以 Lingchu Intelligence(灵初智能)运营,是一家具身 AI 公司,2024 年 9 月 1 日成立,在北京、上海办公。公司自称是中国具身 AI 赛道的领导者,聚焦通用具身智能、大规模视觉-语言-动作(VLA)模型和灵巧操作算法;其使命是“用 AI 和机器人创造无限生产力”,愿景是“成为智能机器人全球领导者”。VLA 模型把机器人周围的摄像头图像和自然语言指令作为输入,直接输出底层电机指令;传统工业自动化里的感知、规划、控制三套链路被压成一条,机器人因此能在新物体、新工况下泛化,不必按任务重新编程。 商业假设不是硬件竞赛,而是平台玩法。Psibot 把 Psi 系列模型——智能“大脑”——授权给第三方机器人开发商,销售自研人手数据采集系统(Psi-SynEngine 外骨骼手套),并提供仿真和训练数据服务。管理层称其为“小全栈”:Psibot 掌控高价值设计要素(系统架构、自由度、运动范围和模型),组件生产与制造交给专业供应商。公司押注具身 AI 最可防守的位置不在机器人本体,而在软件大脑和背后的私有数据飞轮;无论最终哪种硬件形态胜出,Psibot 的智能层和灵巧操作数据集都能随整个行业增值。[CO001, CO002, CO003, CO004, CO005, CO006]

KPI 快照表
指标数值 / 状态日期 / 期间置信度缺口或注意事项
成立2024 年 9 月(2024 年 9 月 1 日)创立公司和独立来源的成立月份一致
总部中国北京和上海当前来源有的称其总部在北京,有的称其成立于上海
又名Lingchu Intelligence(灵初智能)当前国内报道均使用该中文名
阶段早期私营公司;据报已成独角兽截至 2026-07-23追踪平台列为 Series A 前;据报 2026 年 7 月轮接近完成
最新轮次(据报)近 US$100M,由 Chery Automobile 领投2026-07-23Bloomberg 报道;表述为接近完成,尚未交割
据报估值投后约 US$1.48B2026-07-23来自 Bloomberg 报道;公司未确认
累计融资(据报)成立以来约 US$300M截至 2026-07-23包括 ¥2B(约 US$280M)天使轮 + Pre-A,以及据报 7 月轮
旗舰模型Psi R1(VLA + 强化学习)2025-05演示 30+ 分钟自主麻将(“L3”)操作
最新模型Psi-R2 和 Psi-W0(人类数据预训练)2026-04-10开源 1,000 小时手部数据;声称储备 100,000 小时
商业模式平台 / “小全栈”;授权 AI 大脑 + 销售数据系统当前授权 Psi 模型;销售 Psi-SynEngine 数据采集手套
收入 / 年化收入未披露截至 2026-07-23留存来源均未给出任何收入数字
员工数未披露截至 2026-07-23留存来源均未给出员工数量
客户数未披露(试点有名称,无数量)截至 2026-07-23试点包括一家物流供应商和一家光纤电缆制造商

本表合并公司披露和独立报道。2026 年 7 月轮规模及约 US$1.48B 估值来自 Bloomberg 报道,且报道称尚未交割;收入、员工数和客户数均未披露,因此列为缺口而非估算。

[CO001, CO014, CO015, CO016, CO017, CO021]
FO002: Psibot 公司快照逻辑

Psibot 的身份、模型 / 数据平台、客户、资本和依赖关系如何串联。

1.2 创始人、领导层与关键人物依赖

Psibot 由创始人兼 CEO Viktor Wang 博士(中文名 Wang Qibin / 王启斌)领导。他拥有 George Washington University 博士学位,具备约二十年硬件和商业化经验,曾任 JD.com 机器人业务总裁、Yunji Technology 产品副总裁,更早曾在 BlackBerry 任职。技术班底是公司的标志性资产:联合创始人、工程负责人 Xiaojie Chai 博士在 Alibaba 和 Tencent 的机器人及自动驾驶项目中积累十五年以上经验,并主导过 L4 自动驾驶落地;首席科学家 Yaodong Yang 教授是 Peking University Institute for Artificial Intelligence 助理院长、Boya Scholar,也是在 UCL 受训的强化学习研究者,曾赢得 NeurIPS 2022 具身灵巧操作挑战赛,并牵头 PKU-PsiBot Joint Lab;联合创始人 Yuanpei Chen 是一名 Z 世代研究员,曾在 Stanford 跟随 Karen Liu 和 Fei-Fei Li 做访问学者,并称自己率先用强化学习在真实世界控制双臂多技能操作。 科学人才高度集中——Psibot 将自己包装为“科学家密度最高的具身 AI 公司”——也是最尖锐的治理风险。投资逻辑很大程度压在少数具名个人身上:CEO 的商业化履历、首席科学家的学术背书和 PKU 实验室,以及 Chen 的控制模型研究。公开来源未披露董事会构成、股权分配和正式治理安排,也看不到独立董事会或创始科学家之外的职业管理层厚度;关键人物依赖因此集中,学术创始人的留任成为实质尽调问题。[CO007, CO008, CO009, CO010, CO011, CO012]

管理层与创始人表
人物职务背景重要性关键人物 / 披露备注
Dr. Viktor Wang(Wang Qibin / 王启斌)创始人兼 CEOGeorge Washington University 博士;约 20 年硬件 / 商业化经验;曾任 JD.com Robotics 总裁、Yunji Technology 产品 VP;BlackBerry撑起商业化和市场拓展可信度对单一创始人 CEO 的关键人物依赖极高
Dr. Xiaojie Chai(柴晓杰)联合创始人兼工程负责人在 Alibaba 和 Tencent 做机器人与自动驾驶 15+ 年;主导 L4 自动驾驶落地补足全栈工程和规模化生产经验创始技术负责人;留任至关重要
Prof. Yaodong Yang(杨耀东)首席科学家PKU Institute for AI 副院长 / 博雅学者;UCL 博士;赢得 NeurIPS 2022 具身操作挑战赛;负责 PKU-PsiBot Joint Lab提供学术可信度和高校研究管线学界 / 产业双重角色;时间分配和 IP 归属是尽调项
Yuanpei Chen(陈源培)联合创始人兼 RL 负责人Gen-Z 研究者;在 Stanford 跟随 Karen Liu 和 Fei-Fei Li 做访问学者;开发 Psi-C0 控制模型;“Sequential Dexterity”支撑灵巧操作的核心控制模型研究年轻且价值高的研究者;留任和股权安排是关键人物风险
Prof. Ying Wen(温颖)RL 负责人上海交通大学 AI 学院;构建 DB1 多模态决策模型(据报优于 DeepMind 的 Gato)补强多智能体 / 决策模型研究深度仍有高校任职;是否兼职、投入多少未披露
董事会 / 治理未披露公开资料未列出董事名单、股权拆分或独立董事治理不透明,限制控制权评估对据报独角兽而言,这是重大披露缺口

创始人和科学家姓名可由公司官网、独立中文与英文报道交叉印证。董事会构成、股权和正式治理安排均未披露,因此标记为缺口,不做推断。

[CO007, CO008, CO009, CO010, CO011, CO012]

1.3 融资历史、投资人与估值

Psibot 在不到两年里走完了一整段早期融资曲线。公司在 2024 年 11 月完成天使轮,由 GL Ventures(Hillhouse)和 Lanchi Ventures 领投,China Development Bank Capital、Guozhong Capital、CCTV Media Convergence Industrial Investment Fund 等国家队资本参与。约 2026 年 3 月 10 日,公司在一份联合公告中披露,天使轮和 Pre-A 轮合计达到 ¥2B(约 US$280M);公告刻意卡在中国“两会”政治窗口期。Pre-A 由上海国资 Xuhui Capital 领投,Liangxi Sci-Tech Innovation Phase II Mother Fund(Bohua Capital 管理)、Xi Venture Capital、Pufeng Capital 和 Timing Capital 参与,多家老股东追加投资。 2026 年 7 月 23 日,Bloomberg 报道称 Psibot 接近完成新一轮近 US$100M 融资,投后估值约 US$1.48B,由车企 Chery Automobile 领投,Apple 和 Tesla 供应商、精密玻璃及传感器厂商 Lens Technology 参与。消息人士称该轮尚未交割;若完成,公司自成立以来累计融资约 US$300M,并跻身中国最新一批具身 AI 独角兽。投资人组合战略浓度很高:早期是国资产业基金,随后引入汽车和制造供应链资本,呼应中国把 AI 推向“实体经济”的大方向。2024 年至 2026 年初的轮次没有披露定价估值,US$1.48B 数字来自 Bloomberg 报道而非公司确认;因此轮次状态和估值都应按“已报道、未交割”事实处理。[CO014, CO015, CO016, CO017, CO018, CO019]

利益相关方 / 投资者图谱
投资者 / 利益相关方轮次 / 背景类型公开记录显示尽调追问
Chery Automobile据报 2026 年 7 月轮领投(约 US$100M)战略方(汽车)Bloomberg 点名为领投方;总部在芜湖、入选 Fortune Global 500 的车企确认是否交割、持股、董事会权利,以及任何商业 / 生产协议
Lens Technology据报参与 2026 年 7 月轮战略方(制造供应链)Apple 和 Tesla 的精密玻璃 / 传感器供应商;报道点名其参投确认获配份额,以及是否存在组件供应关系
GL Ventures(Hillhouse)天使轮共同领投(2024 年 11 月)创投公司公告点名为天使轮领投方确认持股比例、按比例跟投权和信息权
Lanchi Ventures天使轮共同领投(2024 年 11 月)创投公司公告将其与 GL Ventures 一并列为天使轮领投方确认持股及后续跟投参与
Xuhui CapitalPre-A 领投(2026 年 3 月披露)国资(上海)公开记录列为 Pre-A 领投方;上海国资投资人确认金额、估值和任何战略条件
China Development Bank Capital天使轮“国家队”投资人国资背景公开记录列为天使轮投资人;部分追踪平台列为领投核对其与 GL/Lanchi 的领投归因;确认持股
Guozhong Capital / CCTV 媒体融合基金天使轮投资人国资背景公开记录列为国资关联天使投资人确认获配份额及任何国资控制影响

以上行覆盖公司公告、中文报道和融资追踪平台中最影响决策的投资人。天使轮领投归因在不同来源之间不一致(GL Ventures/Lanchi 与 China Development Bank Capital);这里保留差异,不自行判定。此表不是完整股东名册。

[CO014, CO016, CO018, CO019, CO020, CO025]

1.4 封面指标、规模与披露画像

Psibot 可核验的封面指标几乎全是资本和团队信号,经营指标没有披露。截至报告日能够确认的是:公司成立于 2024 年 9 月;据报道估值 US$1.48B;累计融资约 US$300M;2026 年 7 月由 Chery 领投近 US$100M 轮次;产品节奏从 Psi R0(2024 年 12 月)推进到 Psi R1(2025 年 5 月),再到 Psi-R2 和 Psi-W0(2026 年 4 月)。相反,收入、收入运行率、毛利率、烧钱速度、员工数和硬客户数量,在任何留存公开来源中都未披露;公司自己也没有给早期轮次挂估值。 这使 Psibot 成为一家私有且披露不足的公司,目前只能用代理指标而非财务数据推断规模:投资人阵容的规模和战略质量,2026 年采集 100 万小时操作数据、打造中国最大灵巧手数据集的目标,从 100,000 小时储备中开源 1,000 小时,以及大型物流供应商和主要光纤光缆制造商的具名试点。每个缺乏支撑的封面指标都作为明确证据缺口保留,并配上具体尽调路径(经审计财务、HR 认证员工数、已签约客户名单);因为仅 2026 年中国就诞生了二十多家具身 AI 独角兽,要区分真实牵引力和资本驱动的热度,必须拿到公司尚未释放的经营数据。[CO001, CO017, CO021, CO022, CO023, CO024]

FO003: Psibot 快照 KPI

截至 2026-07-23 报告日的关键成熟度、资本和警示指标。

1.5 里程碑与发展轨迹

Psibot 有记录的时间线不到两年,但在成立、产品、融资和商业化四条线上密度异常高。公司于 2024 年 9 月 1 日成立;2024 年 12 月 30 日发布首个端到端强化学习具身模型 Psi R0;2025 年 1 月开始与关键客户推进概念验证和签约;2025 年 3 月发布 Psi R0.5;2025 年 4 月推出 Psibot V1 和 Psibot H1 硬件平台;2025 年 5 月发布旗舰 Psi R1,并用机器人连续自主打麻将超过 30 分钟,展示其定义的 L3(自主、长时程)灵巧操作能力。 融资和规模里程碑随后提速:2024 年 11 月天使轮;2026 年 3 月披露天使轮 / Pre-A 合计 ¥2B;2026 年 4 月 10 日发布用人类数据预训练的 Psi-R2 和 Psi-W0 模型(并开源 1,000 小时多模态手部操作数据);2026 年 7 月 23 日报道 Chery 领投近 US$100M 轮次,估值 US$1.48B。部分产品节点只在公司材料中精确到月份,2026 年 7 月轮次也仍是接近交割而非已完成;时间线保留这些限定,而不是抹平。留存公开来源中没有负面或治理事件(诉讼、高管流失、召回)——这符合一家极年轻私有公司的特征,但也更像披露有限,而非经审计记录完全干净。[CO001, CO014, CO016, CO027, CO028, CO029]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2024-09-01Psibot(Lingchu Intelligence)成立创立公司成立创始团队(Wang、Chai、Yang、Chen)为 VLA 平台论点打下法律和技术底座
2024-11天使轮融资金额未披露GL Ventures、Lanchi Ventures、China Development Bank Capital 等首笔机构资本;验证 RL 优先路径
2024-12-30Psi R0 发布产品首个端到端 RL 具身模型Psibot建立强化学习模型线
2025-01重点客户 POC 推进并签约规模化试点合作未披露客户首批商业验证信号
2025-03Psi R0.5 发布产品迭代模型Psibot延续快速模型节奏
2025-04Psibot V1 和 Psibot H1 发布产品硬件平台Psibot为技术栈加入自有硬件参考平台
2025-05Psi R1 发布产品VLA + RL “L3” 模型;30+ 分钟自主麻将演示Psibot旗舰能力展示;获得行业关注
2026-03-10披露 ¥2B 天使轮 + Pre-A 融资融资累计 ¥2B(约 US$280M)Xuhui Capital(Pre-A 领投)及国资 / 市场化基金释放国资和产业信心;时点贴近两会
2026-04-10Psi-R2 和 Psi-W0 发布产品人类数据预训练模型;开源 1,000 小时Psibot推进数据飞轮和开源定位
2026-04宣布自有整机量产(据中文报道)规模化宣布量产;未披露数量Psibot从演示走向商业部署
2026-07-23据报 Chery 领投轮融资约 US$100M,估值约 US$1.48B(接近完成)Chery Automobile(领投)、Lens Technology独角兽里程碑;汽车 / 供应链战略方背书
2026数据规模目标规模化目标:1M 小时数据;中国最大的灵巧手数据集Psibot支撑数据护城河复利的核心
截至 2026-07-23公开记录未见诉讼 / 召回 / 高管流失不利事项未发现(披露有限)记录干净,与公司年轻、披露薄相符;不等于审计结论

单一口径的时间线。产品日期来自公司官方里程碑页面;融资日期和金额由公司公告、中文报道及融资追踪平台交叉印证。2026 年 7 月轮和量产主张均来自报道,尚未独立确认已交割。

[CO001, CO014, CO016, CO027, CO028, CO029]
FO001: Psibot 公司里程碑时间线

Psibot 从 2024 年 9 月到 2026 年 7 月的成立、模型发布、融资步骤和商业化信号按时间展开。

1.6 要点

Chapter 02

02市场分析

2.1 市场边界、替代方案与相邻赛道

Psibot 的市场是具身 AI 的“大脑”层:公司授权 Psi 系列 VLA 加强化学习模型,销售自研人手数据采集系统,并提供仿真和训练数据服务,而不是制造机器人本体。纳入范围的支出因此是模型授权、数据和仿真服务、数据采集硬件;排除范围是 Psibot 外包、或授权客户自建的机器人底盘、执行器和整机装配。这个边界很关键,因为它把 Psibot 的可服务市场从自身出货量中拆出来:作为平台,公司机会随下游采用其智能层的机器人总装机基数放大,而不取决于 Psibot 自己卖出多少台机器。 Psibot 要替代的现状方案,是固定可编程自动化和 AGV/AMR 车队——它们速度快,但遇到新物体就不能无重新编程泛化——以及承担分拣、拣选和灵巧搬运的人力。相邻市场包括人形机器人(通用大脑的主要硬件载体)、仓储机器人(分拣是最近的商业应用)、物流机器人和更广义的工业机器人。由于不同机构对这些池子的定义不同——人形机器人、仓储机器人、物流机器人、“具身智能”——其头部总量彼此重叠,不能直接相加;下一节所有规模测算都受这个前提约束。[CM001, CM002, CM003, CM027, CM032, CM038]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方Psibot 为什么在意
具身 AI “机器人脑”(VLA + RL)模型授权、训练数据和仿真服务、数据采集硬件机器人本体制造、执行器、底盘机器人 OEM;企业 AI / R&D 预算Psibot 销售的核心市场
人形机器人完整人形机器人系统(硬件 + 嵌入式软件)纯组件供应企业、国企、集成商Psibot 大脑的主要硬件宿主
仓储机器人拣选、分拣、包装机器人与机械臂、AMR货架、仓储管理软件、输送线物流运营商、电商最近商业应用(分拣)
物流机器人物流场景中的移动 + 操作机器人干线卡车和末端配送车辆3PL、零售商相邻需求池
固定式工业自动化(替代项)固定 PLC 产线、AGV制造商Psibot 要替代的现状方案
人工劳动(替代项)人工分拣和操作劳动力雇主自动化要替代的基线

边界按 Psibot 对自身平台范围的描述,以及独立市场研究机构对机器人支出的切分来划;纳入 / 排除项反映 Psibot 的平台授权模型,而不是硬件制造商模型。

2.2 多视角市场测算

没有单一数字能概括 Psibot 的市场,因此用五个独立视角框定。第一,人形机器人:Global Market Insights 估算 2025 年全球市场约 US$7.9B,并预测从 2026 年 US$10.9B 增至 2031 年 US$54.2B、2035 年 US$192.7B,CAGR 37.6%;Research and Markets 同样将其定义为早期、指数级增长市场。第二,仓储机器人:Fortune Business Insights 估算 2025 年为 US$6.51B,并从 2026 年 US$7.35B 增至 2034 年 US$25.41B(CAGR 16.8%),Asia-Pacific 占 51.7%。第三,物流机器人:Global Market Insights 给出 2025 年 US$17.8B,并从 2026 年 US$20.7B 增至 2035 年 US$91.4B。第四,广义仓库自动化:Grand View Research 估算 2023 年 US$19.23B,2030 年达到 US$59.52B。 第五,也是地理上最贴近 Psibot 的视角,是中国:36Kr Research Institute 估算中国具身智能产业从 2018 年 ¥213.3B 增至 2025 年 ¥915B,并在 2026 年超过 ¥1T;Morgan Stanley 预测 2026 年中国人形机器人出货约 50,000 台(较 28,000 台增长约 79%),2030 年达 446,000 台,中国市场从 2026 年约 US$2B 增至 2030 年约 US$15B。销量印证很强:IDC 报告称 2025 年全球人形机器人出货超过 18,000 台,中国厂商贡献约 90% 单位量,并预计 2030 年出货超过 510,000 台,CAGR 约 95%。TrendForce 还预计中国人形机器人产量在 2026 年增长约 94%。这些视角方向一致、量级分歧明显,因此 2026 年人形机器人 TAM 在全球范围内更现实地落在 US$6B–11B 区间,长期数字仍偏投机。[CM004, CM005, CM006, CM007, CM008, CM009]

TAM/SAM/SOM 或规模测算口径表
口径 / 细分发布方年份地域规模CAGR置信度限制
人形机器人(TAM)Global Market Insights 报告2026全球US$10.9B37.6%,至 2035 年各发布方预测分歧大
人形机器人(长期)Global Market Insights 报告2035全球US$192.7B37.6%远期估算,投机性很高
中国人形机器人市场Morgan Stanley2026中国~US$2B106%,至 2030 年投行预测;由出货量推导
中国人形机器人市场Morgan Stanley2030中国~US$15B106%长期预测
仓储机器人Fortune Business Insights2026全球US$7.35B16.8%,至 2034 年口径较宽,包含 AMR 和 AGV
物流机器人Global Market Insights 报告2026全球US$20.7B到 2035 年 17.9%与仓储机器人定义重叠
中国具身智能36Kr 研究院2026中国>¥1 万亿(~US$140B)n/a“行业”口径过宽,不能与机器人 TAM 相加
仓储自动化Grand View Research2030全球US$59.52B到 2030 年 18.7%包含非机器人自动化

数值逐字引用各发布方;由于细分口径相互重叠(人形机器人、仓储、物流、具身智能),这些行只是不同观察口径,不能相加。

[CM004, CM006, CM007, CM008, CM009, CM010]
FM001: 市场规模拆解视角

从可承载机器人「大脑」的全球宽口径 TAM,到中国特定可服务市场,再到 Psibot 未披露的可获得份额;各层衡量的市场池不同,不是严格算术级联。

这是视角堆栈,不是严格 TAM-SAM-SOM 级联:各层来自不同机构,衡量不同地区里有重叠的市场池;底层 SOM 未披露,而不是估算值。

[CM004, CM007, CM008, CM010, CM031]
FM002: 市场估算区间

围绕 2030 年全球人形机器人市场的独立预测相差约四倍,分析师把这个跨度视为投机不确定性的标记,而不是已有共识。

所有数值均指接近 2030 年的全球人形机器人市场,单位为十亿美元。低位和中位边界来自独立怀疑者分析;高位边界使用 Global Market Insights 的 2031 年数字作为多头代理,因为没有发布清晰的 2030 年全球美元口径数字——这里明示该转换,而不是藏起来。

[CM023, CM013, CM004]

2.3 买方、用户、付费方与采用路径

Psibot 服务两条买方渠道,付费方不同。授权渠道里,买方是嵌入 Psi 大脑的机器人 OEM 或开发商,“用户”是机器人自己的控制栈,付费方是 OEM 的研发预算。直连企业渠道里,买方是物流运营商或制造商,用户是仓库和产线操作员,付费方是企业资本开支;在中国,越来越多还来自国企和政府采购。物流和电商是领先买方垂直——Fortune Business Insights 预计电商在 2026 年约占仓储机器人市场 47%——这与 Psibot 披露的分拣试点吻合。在中国,指令性产业政策把预算权推向承担政策试点的国企和政府实体。 采用路径从实验室演示走向小规模试点,再到场景验证,最后才是多站点规模化部署。到 2026 年,行业仍集中在试点和验证阶段:IDC 报告称,2025 年超过 85% 的人形机器人部署用于表演、教育、数据采集和导览,而非生产工作,制造和物流只有早期试点。Psibot 自身商业化信号——一家中国大型物流客户的仓库分拣项目,以及一家领先光纤光缆制造商的测试——也处在试点阶段,且二者都不是已确认的大规模付费客户;因此买方地图更多描述意图和结构,而不是已验证收入。[CM014, CM015, CM016, CM028, CM033, CM037]

细分市场 / 买方地图
细分市场买方用户付款方工作流预算负责人采用触发因素
物流 / 电商仓库物流运营商 / 3PL仓库运营人员企业资本开支分拣、拣选、打包运营 / 自动化部门人工成本与旺季吞吐量
机器人 OEM(授权)机器人制造商机器人自有控制栈OEM 研发预算将 Psi 大脑接入机器人OEM 产品团队需要有竞争力的智能层
精密制造工厂 / 车间产线工人企业资本开支灵巧装配与搬运工厂工程团队柔性自动化缺口
国有企业国企 / 政府机构国企运营部门政府采购政策要求的试点国企 + 国资委“十五五”部署要求
数据 / 仿真客户AI 和机器人开发者研究人员研发预算购买训练数据与仿真研发负责人VLA 训练数据稀缺

买方、用户、付款方随渠道不同而变化;OEM 授权模式下,付款方是机器人制造商;企业直接试点时,付款方是终端运营商的资本开支预算或国企采购。

FM003: 买家 / 细分市场图

Psibot 五个目标细分市场的买家、用户、付费方关系和采用成熟度;2026 年大多数细分仍停在试点或集成阶段。

采用成熟度是基于 IDC 部署结构数据和 Psibot 已披露试点信号作出的定性尽调判断,不是已披露客户名单。

[CM014, CM015, CM016, CM037]
FM004: 采用漏斗或价值链图

这是一条示意性采用漏斗,从广泛市场兴趣走到规模化生产部署;IDC 数据显示,2025 年大多数人形机器人活动还停在生产部署之前很远的位置。

只有顶部数字(18,000 台,IDC)是已披露数据;下方阶段套用了 IDC “>85% 位于非生产场景”的拆分,属于示意性估算。底部阶段为零,因为公开资料没有确认任何 Psibot 付费部署。

[CM013, CM016, CM033]

2.4 增长驱动与采用约束

最强需求驱动来自中国产业政策。第十五个五年规划(2026–2030)把具身智能提升为与量子、生物制造并列的国家战略支柱,配套国家投资、采购要求和 MIIT/SASAC 部署目标;国家标准自 2026 年初开始发布,MIIT 也把 2025 年定义为中国“人形机器人量产元年”。人口老龄化和结构性劳动力短缺带来持久替代需求,供应链本土化则让中国整机成本接近可比海外产品的 50%,改善采用经济性。资本本身也是驱动:2025 年前十一个月,中国具身智能融资达到 ¥33.5B(约为上一年同期四倍),全年超过 ¥38B、305 笔交易;这股动能延续到 2026 年上半年的密集融资潮。 约束同样具体。MERICS 认为中国人形机器人仍缺精度和灵巧度,现场专用试验成本高,成本至少还要砍半,并且 VLA 突破仍依赖美国研究。精密部件——高精度滚珠丝杠、齿轮和先进传感器——仍有部分进口依赖,这是供应和成本风险,也正是政策明确要攻克的方向。高单机成本、可靠性上限,以及灵巧操作周围的安全和保险壁垒,使大多数部署仍停在受控试点。独立分析更直白:真正用于生产的工业收入只占人形机器人销售约 3–5%,其余来自研究、展厅和尝鲜买家——这是收入质量警告,直接给行业头部增长打了折扣。[CM017, CM018, CM019, CM020, CM021, CM022]

增长驱动因素与约束表
驱动因素 / 约束方向时间影响尽调问题
“十五五”规划与工信部 / 国资委要求驱动(+)2026–2030国家需求与定向采购核实国资订单管线敞口
人口老龄化与用工短缺驱动(+)结构性持久的劳动力替代需求量化可替代的人工任务
供应链本地化(约 50% 成本)驱动(+)2026 年起采用经济性改善核实 BOM 成本相对海外水平
行业资金流入(2025 年 ¥38B)驱动(+)近期资金支撑扩张,也抬高泡沫风险跟踪烧钱速度与收入
精密部件进口依赖约束(–)近期供应与成本风险梳理滚珠丝杠、齿轮和传感器采购来源
精度 / 灵巧性与可靠性差距约束(–)近期限制规模化部署委托独立基准测试
单机成本高、ROI 证据弱约束(–)近期试点在规模化前卡住获取回本周期与 ROI 数据

驱动因素和约束来自中国政策追踪、36Kr 行业数据,以及 MERICS 和独立分析师的反向评估;时间相对于 2026-07-23 运行日表述。

2.5 规模测算缺口与相互冲突的预测

规模测算留下了尽调必须补齐的重大缺口。第一,没有公开来源拆出 Psibot 专属的可服务可得市场(SAM)或可获得市场(SOM);其服务市场只能被推断为人形机器人、仓储机器人和物流机器人 TAM 中的一块授权与数据切片,而 Psibot 没有披露收入、抽成率或授权单位数来锚定这块切片。第二,头部 TAM 不能相加:人形机器人、仓储机器人、物流机器人和“具身智能”在不同机构口径下定义不同且彼此重叠,相加会高估机会。 第三,预测离散度极高。独立分析指出,2030 年人形机器人市场预测相差约四倍——从约 US$4B 到超过 US$15B——并把这个差距解读为投机热度,而非稳定共识;同一批评也认为,行业里被报道的独角兽估值可能跑在真实部署和收入前面。第四,全球仓库自动化渗透率仍低,意味着巨大空间,也意味着短期转化尚未验证。这些矛盾被保留下来,而不是被调和:市场确实很大、增长很快,但真正能测算 Psibot 钱包份额的数字——SAM、SOM、抽成率和已确认客户数——仍是后续估值最敏感的开放问题。[CM022, CM023, CM031, CM032, CM034, CM036]

2.6 要点

Chapter 03

03竞争格局

3.1 竞争版图与替代选择

Psibot 位于具身 AI 智能层,所以竞争集合比“其他机器人制造商”更宽。五类玩家值得关注。直接同业是销售或开源通用“大脑”的机器人基础模型提供商:Physical Intelligence(π0/π0.7 模型族)和 Nvidia 的 GR00T 参考栈。既有玩家和硬件规模玩家是中国“三巨头”——AgiBot(Zhiyuan)、Unitree 和 UBTech——再加上纵向一体化的西方公司 Figure AI 和 Tesla,它们都自研大脑并规模出货。相邻玩家包括 Galaxy General 的 Galbot、Apptronik、Boston Dynamics、1X 和 Agility。替代方案和现状方案是固定工业自动化、AGV/AMR 和人力;从成本和可靠性看,它们仍是多数仓库默认选择。最后,自研也是现实选项:OEM 可以开发自己的控制模型,也可以采用 Nvidia 开放的 GR00T,而不是授权 Psibot。 竞争范围过宽,正是 Psibot 的核心问题。公司不卖本体,因此既要对抗免费 / 开放选项(Nvidia GR00T),也要对抗内部自研选项(Tesla、Figure、Unitree 和 AgiBot 都训练自有大脑);而纯大脑类别里已公开的对手 Physical Intelligence 和 Nvidia 资本实力远强于 Psibot。潜在进入者进一步抬高强度:OpenAI 牵头对 Figure 投资 US$6.7B,并运营一支 200 多名研究员的内部机器人基础模型团队(Project Atlas);Google 也推出 Gemini Robotics。因此版图更有利于规模和资本,Psibot 必须靠模型能力、灵巧度和私有数据打出差异,而不是靠分销取胜。[CP001, CP007, CP022, CP023, CP028, CP029]

FP001: 竞争定位图

按 AI 大脑 / 软件深度与硬件规模和制造能力,对 Psibot 及竞争对手做序位定位;Psibot 属于大脑重、硬件轻的一组,旁边是 Physical Intelligence 和 Nvidia GR00T,离中国硬件规模领先者很远。

轴向分数是有证据支撑的序位判断(0-10),不是原始来源公布的坐标。硬件规模分数跟随已披露出货量和制造版图;大脑深度分数跟随引用报道中的模型能力和数据策略。

[CP002, CP003, CP006, CP008, CP009, CP011]

3.2 竞争对手画像——规模、融资与方向

中国硬件领导者定义了规模标尺。AgiBot 成立于 2023 年,有 BYD 背书,2026 年 3 月 30 日生产出第 10,000 台人形机器人,从 5,000 台到 10,000 台约用三个月;2025 年出货约 5,168 台(Omdia 排名全球第一)。Unitree 在 2026 年 6 月 1 日用创纪录的 73 天通过 STAR Market 上市委审议,目标估值约 US$6.2B,同时融资约 ¥4.2B(约 US$583M);2025 年收入达到 ¥1.699B(约 US$240M),毛利率约 60%,人形机器人已贡献超过一半收入,H1 型号累计超过 11,000 台。UBTech 在 2026 年 6 月 30 日开放全尺寸 UWORLD U1 系列订单,首日累计订单突破 13,361 台,起价 ¥119,800;此前 2025 年全尺寸机器人出货 1,079 台,人形机器人收入 ¥821M。 纯大脑和西方全栈玩家定义了资本与能力标尺。Physical Intelligence 成立于 2024 年,构建通用机器人基础模型(π0、π0.7,后者带 RL Token 和 Multi-Scale Embodied Memory,可处理超过十分钟任务),据报道正以超过 US$11B 估值融资约 US$1B,四个月前估值还是 US$5.6B,背后投资人包括 Founders Fund、Lightspeed、Jeff Bezos 和 Nvidia。Figure AI 成立于 2022 年,2025 年 9 月以 US$39B 估值完成 Series C(累计融资约 US$1.9B),运行自研 Helix 大脑,正在放量 Figure 03,并有 OpenAI 牵头的 US$6.7B 投资支持。Tesla 到 2026 年 6 月已在 Giga Texas 部署约 1,000 台 Optimus Gen 3,目标年底内部部署 5,000 台。与这些公司相比,Psibot 约 US$1.48B 估值和未披露出货量,使其更像中腰部的能力与数据挑战者,而不是规模领导者。[CP002, CP003, CP004, CP005, CP006, CP008]

竞争对手画像表
竞争对手类别规模 / 融资目标细分市场差异化局限
Psibot (Lingchu)具身 AI 大脑(授权方)~US$1.48B 估值;已融资 ~US$300M机器人 OEM;物流 / 制造"小型全栈" 授权 + 灵巧手数据集未披露出货量;结构性锁定弱
AgiBot (Zhiyuan)中国硬件规模化厂商BYD 背书;2026 年 3 月第 10,000 台工业 / 通用制造爬坡最快;供应链标准化部署数为公司披露;生产性使用未经验证
Unitree中国硬件规模化厂商科创板 IPO 目标 ~US$6.2B;2025 年收入 ¥1.7B科研、商业、工业公开市场资本;11,000+ 台;~60% 毛利率工业收入仅 ~9%;2026 年 Q1 放缓
UBTech中国硬件规模化厂商U1 首日订单 13,361 台;2025 年人形机器人收入 ¥821M工厂、服务、护理全尺寸 U1 产线;88 DOF;品牌 + 规模已确认出货基数相对订单簿偏小
Physical Intelligence美国机器人基础模型实验室正以 >US$11B 估值融资 ~US$1B任意机器人 / 任意任务(大脑)π0.7 可控模型;RL Token;具身记忆无硬件;尚无收入的通用化押注
Figure AI美国垂直整合全栈US$39B 估值;已融资 ~US$1.9B先工厂,后家庭自研 Helix 大脑;Figure 03 爬坡;OpenAI 背书估值建立在近零收入上;安全诉讼
Tesla Optimus美国垂直整合全栈累计研发 ~US$2-3B;母公司上市先替代内部工厂劳动力垂直整合;FSD / Dojo 算力~1,000 台;外部销售要到 2027 年
Nvidia GR00T机器人基础模型 + 平台上市公司;Isaac / Jetson 生态所有人形 OEM(开放栈)开放参考平台;供应商 + 大脑供应商与竞争者身份冲突;不是机器人制造商
Galaxy General (Galbot)中国具身 AI 玩家独角兽级别;公开融资声量较低零售 / 物流操作仿真优先的数据路径公开能见度和披露规模较低

竞争对手规模和融资数字来自 IPO 申报报道、公司官网和独立追踪机构;多项部署数由公司披露,未经独立审计,风险登记表已标注。

[CP002, CP003, CP006, CP008, CP009, CP011]
定价 / 包装对比
竞争对手价格 / 合同模式包含能力折扣或未知项启示
Psibot模型授权 + 数据平台(未披露)Psi 模型授权、Psi-SynEngine 数据系统、仿真定价、抽成率、授权单元数均未知无法对标收入或单位经济性
UBTech (U1)单台起价 ¥119,800;三档全尺寸人形硬件 + 机载栈企业 / 批量折扣未知透明的硬件价格锚
Unitree (G1 / H1)~US$16,000(G1);可 RaaS 租赁人形硬件 + Unitree 软件企业报价不一低成本硬件压低西方同行价格
AgiBot从 ~US$5,000(入门)到更高档人形硬件 + 具身 AI 栈批量 / 工业定价未披露激进的价格驱动扩张
Figure AI面向工厂租赁(行业引用价 ~US$80k/年)全栈机器人 + Helix 大脑合同条款未公开服务模式,不是单机销售
Nvidia GR00T开放参考栈(平台经济性)基础模型 + Isaac / Jetson靠算力变现,不靠模型授权免费 / 开放选项压迫大脑授权方

价格来自 2026 年报道和厂商材料的引用;Psibot 自身定价是明确证据缺口,同行价格给出它必须被衡量的基准。

3.3 能力、定价与 GTM 对比

能力上,Psibot 拼的是模型质量和灵巧度,而不是单机产量。分层快 / 慢 VLA 加强化学习架构,以及灵巧操作基准(连续 30 多分钟自主麻将演示),是公司的招牌;但同样的快 / 慢、VLA 优先路径也出现在 Physical Intelligence(π0 到 Hi Robot)、Figure(Helix)、Google Gemini Robotics 和 Nvidia GR00T 身上,核心技术路线并不独特。原始规模上,Psibot 落后 AgiBot、Unitree 和 UBTech 数个数量级——对手是数千到一万多台,Psibot 仍是未披露试点量。 定价和 GTM 上,对比很尖锐。硬件同业公开单机价格和租赁条款——UBTech U1 起价 ¥119,800,Unitree G1 约 US$16,000,AgiBot 机型约 US$5,000 起,还提供 robot-as-a-service 租赁——并借助 BMW、BYD、NIO 的工厂试点、零售体验店和公开上市触达买方,拿到分销和资本。Psibot 的模型授权和数据平台定价没有披露;作为私有大脑授权商,公司缺少上市或捆绑式竞争对手的分销和资产负债表半径。抵消项是 Chery 和 Lens Technology 带来的战略投资人供应链及传感器入口,以及私有人手操作数据集。信任和监管姿态上,分析师提示中国人形机器人厂商有结构性法律风险,而文件并未披露;Morgan Stanley 买方调查显示,仅 23% 的潜在工业买方对现有产品满意——这是适用于整个赛道的需求侧警告。[CP014, CP015, CP016, CP017, CP018, CP019]

功能 / 能力矩阵
购买标准PsibotAgiBot / Unitree(中国硬件)Physical Intelligence(美国大脑)Figure AI(美国全栈)
通用 VLA 大脑强(Psi R1)中(自研)强(π0.7)强(Helix)
灵巧操作 / 手强(灵巧手数据集)
制造规模低(未披露)强(10,000+ 台)无(无硬件)中(Figure 03 爬坡)
专有数据飞轮强(声称中国最大数据集)强(RL Token)强(真实世界视频)
资本 / 估值中(~US$1.48B)强(IPO / BYD)强(>US$11B)强(US$39B)
分销 / 渠道低(非上市授权方)强(工厂、零售、IPO)低(早期)中(BMW 试点)
已确认工业收入未知(未披露)低(约占 Unitree 收入 9%)未知(收入前)低(百万级个位数)

单元格反映有证据支撑的序数判断;“未知”表示没有可靠公开数字,而不是猜测,符合章节质量标准。

FP002: 功能广度 / 能力图

这张图按具身 AI 大脑采购时真正关键的标准,比较各竞争者的能力覆盖和强弱;未知单元格明确标注,不做猜测。

强弱标签来自引用报道,是有证据支撑的序位判断;“无”表示玩家没有硬件制造,“自称”标记 Psibot 自报的数据领先,没有独立验证。

[CP014, CP015, CP016, CP024, CP025]

3.4 切换成本、锁定与分销权力

对大脑授权商来说,切换经济性是最核心的竞争问题。机器人 OEM 可以采用 Nvidia 开放的 GR00T 栈、授权 Physical Intelligence,或自建控制模型;除非 Psibot 的私有灵巧手数据集真正形成数据锁定,否则离开 Psibot 的切换成本结构性偏低。多栖接入进一步加压:OEM 可以同时整合多家基础模型提供商,并让它们互相压价,限制单一大脑供应商的定价权。内部自研威胁并非假设——Tesla、Figure、Unitree 和 AgiBot 都训练私有大脑,Nvidia 也把 GR00T 作为开放参考平台分发;因此 Psibot 的可授权大脑假设,同时面对免费 / 开放和内部自研替代。 分销权也偏向既有玩家。硬件同业掌握工厂关系(BMW、BYD、NIO)、零售渠道和公开市场入口;Nvidia 则同时是整个赛道许多公司的算力供应商和大脑竞争者:许多中国同业基于 Jetson 和 Isaac 构建,Nvidia 还点名 Unitree H2 Plus 机体作为 GR00T Reference Humanoid 的硬件基础。Psibot 的抗衡力量来自战略投资人的制造和传感器供应链,以及中国市场入口,但这些都依赖伙伴而非公司自有。净结果是:这是一家拥有可信数据飞轮护城河叙事的大脑供应商,但相较资本更足、纵向一体化或开源的对手,结构性锁定弱、分销有限。[CP020, CP021, CP022, CP032, CP037]

3.5 护城河耐久性与替代风险

Psibot 最能防守的护城河叙事,是其私有人手操作数据集和数据采集引擎,以及打造中国最大灵巧手数据集的目标。如果灵巧操作数据是可靠 VLA 模型的稀缺输入,早期且持续复利的数据领先可能具备耐久性。但耐久性尚未证明,时间窗口也有限。每个严肃对手都在放大数据:Figure 从内部真实世界视频中学习,Physical Intelligence 提取 RL Token 做快速在线强化学习,Nvidia 通过 Isaac 仿真大规模生成合成数据。如果开放模型或资金更足的模型补上灵巧度差距,大脑层会商品化,Psibot 的授权假设就会被侵蚀。 反向证据不少。行业层面,2025 年人形机器人部署绝大多数并非生产用途——研究、教育、演示和导览占主导——真正的工业收入只占低个位数;Unitree 自己的招股书显示,研究和教育占人形机器人收入 73.6%,真实产线收入在 2025 年前九个月仅约 ¥15.7M(US$2.2M)。Morgan Stanley 买方调查(满意度 23%)和 Unitree 2026 年 Q1 增速放缓都说明供给正在跑过已验证需求。在这个背景下,Psibot 的估值比 Figure(US$39B)和 Physical Intelligence(>US$11B)低一个数量级,也低于 Unitree 约 US$6B 的 IPO 目标;因此它既不是能力领导者,也不是规模领导者——存活逻辑压在数据飞轮锁定和中国市场入口上,而两者仍需证明。[CP024, CP025, CP026, CP027, CP030, CP035]

护城河耐久性 / 竞争风险登记表
护城河主张威胁严重性缓释措施 / 尽调问题
专有灵巧手数据集对手扩大数据规模(视频、RL、仿真)核实数据集规模、独特性与授权条款
更强 VLA + RL 模型(Psi R1)所有同行都走快 / 慢系统路径委托独立基准测试,对比 π0.7 / Helix / GR00T
授权 / 平台模式开放方案(GR00T)和自研栈可替代评估已签约 OEM 授权客户与切换成本
战略投资方供应准入(Chery、Lens)依赖伙伴,非自有确认有约束力的供应 / 分销协议
中国市场 / 政策准入资金充足的国内进入者 + 普惠政策梳理国企管线和任何优先采购
灵巧能力领先行业需求弱;工业收入约 9%获取付费部署和 ROI 证据

严重性反映威胁发生可能性及其对 Psibot 授权论点的影响;每一行都把护城河主张与验证它所需的具体尽调配对。

FP003: 护城河 / 就绪度 KPI

Psibot 相对行业的一张竞争耐久性快照;规模和收入缺口是硬约束。

KPI 数值混合了已披露数据(估值、收入结构、调研)和序位判断(架构独特性);Psibot 自身出货和收入 KPI 没有披露,因此按未披露处理。

[CP005, CP015, CP016, CP030, CP035]

3.6 要点

Chapter 04

04财务

4.1 收入模式、收入流与定价

Psibot 变现的是智能,不是硬件。公司材料描述的是“小全栈”模式:掌控机器人设计参数(结构、运动范围、自由度),但把组件生产和制造外包;收入来自向第三方机器人开发商授权 Psi 系列 VLA 模型,销售私有人手数据采集系统(Psi-SynEngine,使用 16-DOF 外骨骼手套,可实现亚毫米 3D 轨迹精度和指尖触觉感知),并提供仿真及训练数据平台。公司明确的商业目标是打造中国最大灵巧手数据集,并在 2026 年采集 100 万小时操作数据,这把数据本身也改写成可变现资产。这是 Android 式平台假设:把授权“大脑”铺进许多 OEM 本体,而不是销售单一机器人。 关键财务警告在于,这些收入流没有任何一个披露价格、抽成率、授权单位数或已确认收入。报道显示 Psibot 到 2026 年初已通过中国物流和制造商业试点产生部分收入,但金额、合同结构和收入确认时点全部不公开。硬件同业会公开透明单机价格——UBTech U1 起价 ¥119,800,Unitree G1 起价 ¥85,000(约 US$12,000),R1 Air 起价 ¥29,900(约 US$4,300)——而 Psibot 的授权和数据平台定价未披露,因此连间接三角测算收入都做不到。本章每条变现线索都标为公司主张(收入流存在)或不可获得(经济性),并配上具体尽调请求,而不是估算值。[CI001, CI002, CI003, CI004, CI005, CI021]

收入流表
收入流机制单位当前数值 / 状态质量尽调问题
模型授权向机器人 OEM 授权 Psi 系列 VLA 模型按授权 / 按单元 / 版税(未披露)试点活跃;未披露价格或授权单元公司口径;经济性不可得含定价、抽成率、授权单元数的已签协议
数据采集系统(Psi-SynEngine)销售外骨骼手套人手数据硬件 / 软件按系统 / 按席位(未披露)产品已存在;未披露销售公司口径;经济性不可得系统价目表与出货量
数据 / 数据集授权将专有灵巧手数据集变现按小时 / 数据集计价(未披露)100,000 小时储备;1,000 小时开源公司主张;经济性未披露数据集授权条款和买方
模拟与训练数据平台模拟和模型训练服务订阅 / 用量(未披露)已推出;未披露收入公司主张;经济性未披露平台定价和活跃客户
联合实验室 / R&D 服务PKU-PsiBot 联合实验室和伙伴 R&D资助 / 合同(未披露)实验室运营中;资金结构未知已确认;经济性未披露实验室资金和任何服务收入确认

每条收入流都有公司和第三方来源确认,但没有任何一条披露价格、单位数量或已确认收入;“当前价值”列反映的是状态,不是已入账收入。

定价 / 变现表
项目价格 / 单位 / 合同标价与实际折扣 / 未知项来源依据
Psi 模型授权未披露标价和实际成交价均未披露整体定价结构未知公司材料;未公布价格
Psi-SynEngine 数据系统未披露未披露数量 / 捆绑条款未知公司材料
同业锚点 — UBTech U1(硬件)每台 ¥119,800 起标价企业折扣未知同业披露(仅作基准)
同业锚点 — Unitree G1(硬件)¥85,000 起(约 US$12,000)标价可提供 RaaS 租赁Unitree 招股书(仅作基准)
同业锚点 — Unitree R1 Air(硬件)¥29,900 起(约 US$4,300)标价海外可比产品 US$30k-130k+Unitree 招股书(仅作基准)

Psibot 没有公布定价;同业硬件价格只用于给定价缺口定标,不是 Psibot 收入代理。标价不等于实际收入,也不等于利润率。

FI001: 收入模型桥

对一个大脑与数据授权方,客户活动如何转成收入和毛利;下游每个资金节点目前都未披露,所以这座桥是定性的。

定性桥:节点关系是结构性的,没有量化,因为 Psibot 没有披露收入、价格或成本数据。警示色标记完全依赖未披露私有数据的节点。

[CI001, CI004, CI010, CI011]

4.2 GTM 动作与销售效率代理指标

Psibot 的 GTM 动作不是大规模销售漏斗,而是少数深度战略合作,这与早期平台授权商相符。公开部署证据很薄,但具体:在一家中国大型物流客户处做小规模仓库验证(报告分拣效率提升),并在全球最大光纤光缆制造商之一处测试。2026 年 7 月轮次领投方本身就是渠道——Chery Automobile(Fortune Global 500 车企,拥有 AiMOGA 机器人子公司,并与 Nvidia 有战略合作)和 Lens Technology(Apple 与 Tesla 精密组件供应商,正在建设具身智能中心,目标年产 3,000 台人形机器人和 10,000 台机器狗)——因此,投资人牵引的需求和制造拉力构成 GTM 的核心部分。 常规销售效率指标完全不可得。公开来源没有销售周期、获客成本、回本周期、管线覆盖或渠道经济性,也无法可靠代理,因为授权价格和单客户收入未知。战略投资人结构是双刃代理:它可能缩短销售周期、降低制造入口风险,但也意味着早期收入可能是关联方或试点阶段收入,而非公平交易的商业需求——这个区别会实质影响收入质量。诚实判断是,Psibot 的 GTM 由资本和关系驱动;承销人想看的每个效率指标,目前都是尽调请求,不是数据点。[CI006, CI007, CI008, CI009, CI022, CI023]

4.3 成本结构、毛利率与资本强度

理论上,大脑和数据授权商应当比机器人制造商拥有更接近软件的成本结构:VLA 模型训练完成后,增量授权毛利率较高,最重的成本是研发人才、算力和数据采集,而不是物料清单和装配。Psibot 通过外包制造强化这个叙事,并声称外骨骼手套采集数据的成本约为传统遥操作的十分之一——这个说法未经验证,但结构上很关键,因为数据采集成本是数据飞轮模型里的主导可变成本。若属实,它会降低支撑整个投资逻辑所需数据集的资本强度。 Psibot 没有披露这些数据,因此无法用公开数据建模利润率路径,只能拿最接近的公开可比公司做参照。Unitree 招股书显示,一家纵向一体化硬件厂商靠自制电机和执行器,在 2025 年 ¥1.708B 收入上实现约 60% 毛利率(2022 年为 44%)。这个数字给深度一体化硬件同业设了一个偏上限的参照;纯软件授权商原则上可以在增量单位上超过它,但必须先用尚未验证的收入基数,吸收沉重的固定研发、算力和数据采集成本。因此资本强度问题反转了通常的机器人担忧:Psibot 的风险不在工厂资本开支,而在当前资本耗尽前,授权收入能否覆盖庞大且前置的模型与数据成本。Lens Technology 自身扩张(2025 年前九个月 R&D 投入 RMB 2.44B)说明周边具身 AI 硬件生态建设有多昂贵。[CI010, CI011, CI012, CI013, CI024, CI025]

单位经济模型表
指标数值 / null置信度为何重要尽调要求
授权价格 / 抽成率none决定收入上限,也决定能否与同业对比获取已签署授权的定价和版税条款
毛利率(按收入流)null(同业参考约 60%)决定授权模式能否规模化索取各收入流毛利率和成本拆分
CAC / 回本周期none检验销售效率和获客成本提供销售周期、CAC 和回本数据
数据采集成本优势约为遥操作的 1/10(公司主张)数据飞轮的主导变量成本独立验证每小时成本
ARR / 经常性收入none任何前瞻收入模型的核心披露经常性收入与一次性收入拆分
客户集中度 / 关联方占比none战略投资人需求可能不是公平交易关联方交易明细

Psibot 没有披露财务数据,因此所有量化单位经济字段都是 null;唯一非 null 的单元格是公司主张的成本比率,且明确未经验证。每个 null 都带着具体尽调要求。

FI002: 单位经济模型桥

单个授权部署的定性单位经济链,以最接近公开可比公司的约 60% 毛利率为锚,因为 Psibot 自身数据为空。

Psibot 输入项不可得;这座桥用 Unitree IPO 招股书中的同行毛利率锚来框定可能区间。所有 Psibot 专属节点都是定性占位,等待披露。

[CI010, CI012, CI025]

4.4 公开牵引力与私有指标缺口

Psibot 展示出来的东西与披露出来的东西之间存在落差,这是其财务画像的核心特征。公开侧有具体但非财务的信号:新近约 US$100M 轮次,估值 US$1.48B;不到两年累计融资约 US$300M;一线战略投资人;物流和光纤制造的具名商业试点;快速扩大的技术足迹(约十八个月内从 Psi R0 推进到 Psi-R2/W0,并从所称 100,000 小时数据储备中开源首批 1,000 小时);以及 PKU-PsiBot 联合实验室。私有侧,几乎所有硬财务指标都缺失:无收入、无 ARR、无毛利率、无月度烧钱速度、无现金跑道、无员工数、无授权单位数、无客户集中度披露。 这很重要,因为行业公开数据提醒投资人:不能把演示当需求。Unitree 招股书——目前最清晰的具身 AI 经济性窗口——显示,即使是销量领导者,也只有约 9% 的人形机器人收入来自真正的工业部署,74% 来自研究和教育,17% 来自商业“展示”用途;2025 年前九个月,真实产线收入仅约 ¥15.7M(US$2.2M)。Morgan Stanley 买方调查发现,只有 23% 的潜在工业买方对现有产品满意。在这种背景下,Psibot 的具名试点值得鼓励,但没有量化;投资人需要更高强度证据,证明它已经拿到真实、公平交易、经常性的收入。本节每个私有缺口都在缺口表中配有具体尽调路径。[CI014, CI015, CI016, CI017, CI026, CI027]

公开财务缺口表
缺失的私营公司指标影响精确尽调路径
按收入流划分的实际收入 / ARR收入质量和增长测算无法落地经审计 / 管理口径的分收入流收入 + 收入确认政策
毛利率和成本结构无法建模利润率路径按期间拆分成本(R&D、算力、数据、COGS)
现金、烧钱、跑道资本充足性只是假设,不是事实融资后资产负债表和月度烧钱
授权合同和单位数无法验证试点是否转化为收入已签署授权合同(含条款)和授权单位数
客户集中度 / 关联方占比掩盖公平交易需求与战略投资人需求的区别关联方交易和头部客户明细
员工数和 R&D 能力成本基数和扩张能力不清按职能划分员工数和薪酬年化支出

每个缺口都配有可消除疑点的精确证据;合在一起,就是财务尽调方按基本面给 Psibot 估值前必须完成的清单。

FI003: 财务估算区间

对少数可框定的 Psibot 指标(资本和估值)给出来源支撑区间,并加入以同行为锚的毛利率区间;Psibot 收入 / 现金消耗无法框定,故不列入。

资本和估值区间反映了不同来源之间的小幅报道差异;毛利率区间是 Unitree 已披露的 2022-2025 年区间,只用作基准。Psibot 收入、现金消耗和现金跑道不可得,因此刻意不绘制。

[CI018, CI019, CI012]

4.5 资本充足性与融资依赖

前瞻资本充足性是 Psibot 最强的财务维度。历史融资时间线已在公司概览覆盖;简言之,Psibot 先后叠加了天使轮(2024 年 11 月,由 GL Ventures 和 Lanchi Ventures 领投,国资背景共同投资人参与)、2026 年 3 月 10 日公告的天使轮加 Pre-A 合计约 ¥2B(约 US$280M),以及 2026 年 7 月由 Chery Automobile 领投、Lens Technology 参与、接近完成的约 US$100M 轮次,估值 US$1.48B——不到两年总资本约 US$300M。若九位数新轮次刚刚落袋,且没有制造资本开支负担,一家年轻的软件和数据公司理论上可以在下一次融资前支撑两到三年的研发、算力和数据采集扩张;但具体现金余额、烧钱速度和现金跑道仍未披露,必须确认。 融资依赖仍然真实存在。Psibot 没有披露收入基数来支撑自我造血,因此仍依赖资本市场持续开放,以及战略投资人的善意;这些投资人的利益部分来自产业而非纯财务。公开来源没有债务或项目融资义务,这符合轻资产授权商定位;但下一轮融资触发点实际上是能力与商业化里程碑:Psibot 必须在当前热度驱动的估值环境降温前(2026 年诞生 22 家以上具身 AI 独角兽;2026 年上半年中国具身 AI 融资 US$13.8B),把数据领先和试点转化为可展示的授权收入。战略投资人影响力集中,以及后续资本可能以技术里程碑为条件,是需要测试的关键融资风险。[CI018, CI019, CI020, CI028, CI029, CI030]

资本充足性表
维度状态依据尽调要求
累计融资额2024 年以来约 US$300M(天使轮 + Pre-A 约 US$280M;2026 年 7 月约 US$100M)第三方报道;多来源一致确认股权结构和各轮精确规模
最新估值2026 年 7 月投后估值 US$1.48BStraits Times / Tech Times / Tech in Asia 报道确认投后估值和股权类别条款
账面现金未披露无公开资产负债表获取本轮融资后的现金余额
月度烧钱 / 跑道未披露(仅推断 2-3 年,轻资产)仅为推断获取烧钱速度和跑道模型
计划资金用途R&D、算力、数据采集扩张(定性)公司 / 报道叙事董事会批准的资金用途计划
债务 / 项目融资义务未见迹象(轻资产授权方)缺乏公开证据确认不存在表外或债务义务

历史融资时间线已在公司概览中详述;本表聚焦前瞻资本充足性。融资额和估值有充分交叉验证;现金、烧钱和跑道未披露,只能推断。

FI004: 资本强度 / 现金流图

资产轻授权方与硬件制造商的资金流向对比;Psibot 避开工厂资本开支,但要前置投入模型、算力和数据采集,而收入线还没有跑通。

结构性现金流图,不是量化陈述;Psibot 没有披露成本或现金流数据。“无工厂资本开支”节点对应其制造外包模式,这是它区别于硬件同行的关键资本强度差异。

[CI013, CI020, CI024, CI028]

4.6 财务判断——收入质量、利润率路径与阻碍

看收入质量,结论是「尚未证明,但不是不存在」。Psibot 的变现模式可信且有差异化 (授权加可变现数据资产),也有具名试点;但公司没有披露任何已实现收入,没有独立第三方经常性 合同证据,早期收入停留在试点或关联方阶段的风险也真实存在。看利润率路径,软件授权模式最终 理应比硬件同行更高毛利,但没有成本或收入披露,路径无法建模;最诚实的参照点是 Unitree 约 ~60% 的硬件毛利率,以及其在 Q1 2026 同时因研发负担而利润腰斩,这说明即便是已经盈利的 龙头,具身 AI 经济性也会很快被压缩。看资本密集度,Psibot 有优势——轻资产、没有工厂资本开支 ——但未验证的收入线要承接前置的模型和数据成本,风险正好反过来。 尽调阻断项具体且严重。最大的一项是已实现收入和合同披露完全缺席:没有这些,收入质量、ARR 轨迹和单位经济模型都无法支撑投资判断。第二项是现金、现金消耗、现金续航三件套未披露,把资本充足性从 「大概率舒服」变成一个假设。第三项是客户集中度和关联方暴露,因为需求与战略投资方相关。 在 Psibot 打开已实现财务数据之前,这仍是一笔由团队、数据和投资方驱动的赌注;眼下的财务 报表不是数字,而是一组结构清楚的问题。本章图表和缺口表逐项列出这些阻断项,以及能够解除 阻断的确切证据。[CI031, CI032, CI033, CI034, CI035]

4.7 图表

Chapter 05

05产品与技术

5.1 用客户工作流定义产品

Psibot 卖的是智能,不是铁。核心产品是 Psi 系列视觉-语言-动作(VLA)模型——一个能感知场景、 理解自然语言指令并生成灵巧机器人动作的「大脑」——交付方式不是 Psibot 品牌整机,而是授权给 第三方机器人厂商的智能层。放到客户工作流里,软件接收一个任务(「分拣这些包裹」「给这排货架 补货」)以及摄像头和传感器输入,输出一串抓取、放置和修正动作,由机械臂或人形机器人执行, 面向仓储分拣、包装、补货和制造搬运。 Psibot 采用「小全栈」模式:公司定义机器人设计参数——结构、运动范围、自由度——但把零部件 生产和制造外包,自己保持轻资产并聚焦模型。产品能力最常被引用的证据点是 Psi R1 麻将演示: 机器人在开放环境中与人类自主打麻将超过 30 分钟,长时序任务把灵巧摸牌操作和策略推理揉在 一起,Psibot 用它定义 L3 级自主。由此看,产品定位是物流和制造现场的「操作系统」。[CE001, CE002, CE003, CE004]

工作流 / 用例表
用户任务当前工作流Psibot 方案可衡量收益限制
仓库包裹分拣人工分拣或固定功能输送线Psi 模型驱动 OEM 机器人完成灵巧抓取-放置报道称分拣效率提升(未量化)单次小规模验证;无吞吐数据
制造零件搬运人工操作员 / 刚性自动化VLA 引导多样零件操作无需重编程即可跨 SKU 保持灵活只在一家光纤制造商测试;范围窄
补货 / 包装人工或任务专用机器语言指令驱动多步骤操作跨任务泛化(公司主张)无公开生产部署
长周期灵巧任务(演示)L1 / L2 系统无法自动化Psi R1 自主麻将(30+ 分钟)展示 L3 推理 + 灵巧性演示,不是商业工作负载
用于模型训练的数据采集昂贵的遥操作Psi-SynEngine 外骨骼手套采集公司称成本约为遥操作的 1/10成本主张未经验证

用例来自公司材料和报道;收益大多是定性描述,因为 Psibot 没有披露已部署工作流的吞吐量、准确率或成本数字。

FE002: 客户工作流 / 运行流程

Psibot 驱动的机器人如何执行客户任务——从自然语言指令和摄像头输入,到规划、token 化并执行灵巧动作,再到纠错。

定性运行流程,抽象自 Psibot 描述的架构和麻将演示;内部消息传递和时序细节没有公开记录。

[CE002, CE004, CE012]

5.2 产品线与模块图谱

Psibot 的产品组合是一叠模型加一个数据采集工具,迭代速度很快。模型线包括 Psi R0(2024 年 12 月 30 日,被描述为行业首个端到端强化学习具身模型)、Psi R0.5(2025 年 3 月)、Psibot V1 和 H1 硬件参考平台(2025 年 4 月)、旗舰 Psi R1(2025 年 5 月),以及 Psi-R2 世界-动作模型与 Psi-W0 动作条件世界模型(2026 年 4 月 10 日)。头部模型之下还有两个功能 子模型——负责规划的 Psi-P0,以及负责控制的 Psi-C0;后者由联合创始人、RL 负责人 Yuanpei Chen 开发。 数据层是 Psi-SynEngine:一套自有采集系统,基于 16 自由度外骨骼手套,具备亚毫米级 3D 轨迹精度,以及 3D 指尖触觉和力反馈感知,用来规模化捕捉人手操作。报道称 Psibot 的灵巧机器 人手有 21 个自由度。Psi-R2 训练使用了 95,472 小时人类数据,覆盖 294 个场景和 4,821 个 任务;Psibot 已开源其所称 100,000 小时多模态数据储备中的前 1,000 小时。因此,产品线图谱 是一代代模型叠在一台定制数据采集引擎之上——整个投资论点靠这项资产复利。[CE005, CE006, CE007, CE008, CE009, CE010]

产品模块 / 资产矩阵
模块 / 资产用户状态 / 成熟度差异化尽调缺口
Psi R1(分层 VLA + RL 模型)OEM 机器人开发者(授权)2025 年 5 月发布;旗舰CoAT 支撑 L3 自主;快 / 慢脑除 DexGraspVLA 外无独立基准
Psi-R2 世界-动作模型OEM 开发者;内部训练2026 年 4 月 10 日发布基于 95,472 小时人类数据训练;延迟 <100ms训练集构成和评测未经审计
Psi-W0 动作条件世界模型内部 RL 飞轮2026 年 4 月 10 日发布用于规划的反事实推理无公开准确率或消融数据
Psi-P0 / Psi-C0 子模型内部使用(规划 / 控制)已使用;版本节奏不清规划-控制模块化拆分无公开规格表或接口文档
Psi-SynEngine(16-DOF 外骨骼手套)数据采集操作员;作为产品销售量产中;核心数据资产亚毫米 3D 精度 + 指尖触觉单价和出货量未披露
灵巧手(21-DOF)具身 OEM 硬件参考设计;外包制造高自由度,面向操作任务缺少可靠性 / 耐久性数据
Psibot V1 / H1 硬件平台参考 / 演示硬件2025 年 4 月发布Psi 模型的参考机身不是商业化重点;规格很薄

每个模块都有公司和第三方来源确认存在;成熟度反映发布状态,不代表经过验证的现场表现。每行都把资产和具体尽调缺口配对。

5.3 技术架构与运行模式

Psi R1 的架构是分层、端到端设计,把视觉-语言-动作模型与强化学习融合,在一个训练好的系统里 连接高层规划和低层灵巧控制。Psibot 称为「Chain of Action Thought」(CoAT)的推理机制, 让系统在开放、动态环境中拆解并执行长时序任务——公司把这种能力定义为 L3(自主推理和操作), 高于 L1(基础抓放)和 L2(没有认知链的人类式抓握)。为了打通思考和运动,设计使用了一个 连接规划与控制的「Action Tokenizer」,以及一个快/慢「S1 fast / S2 slow」双系统大脑,把 反应式控制和审慎规划分开。 2026 年的演进是 Psi-R2 / Psi-W0 双模型架构。Psi-R2 是世界-动作模型;Psi-W0 是动作条件 世界模型,用来做反事实推理并带动强化学习「飞轮」。关于这一双模型设计的报道称,它把推理 延迟从 2.2 秒降到 100 毫秒以内——这对实时操作是一个阶跃变化。学术实现 DexGraspVLA 采用 预训练视觉语言模型做高层规划器,用扩散式低层控制器,反复把多样输入映射为跨域不变表征, 让模仿学习能够泛化。公开设计目标是同时拿到高泛化、高灵巧、高成功率的「不可能三角」 (95% 验证、99.9% 规模化部署目标),路径是先用大量人类数据预训练,再用少量真实机器人 数据微调——这正面回应了具身 AI 的数据稀缺问题。[CE011, CE012, CE013, CE014, CE015, CE016]

技术 / 运行架构表
层 / 组件角色依赖风险
感知(视觉-语言输入)理解场景 + 自然语言任务预训练 VLM 基础模型依赖外部基础模型质量
规划(Psi-P0 / CoAT)拆解长周期任务;L3 推理模型训练数据和算力推理能力主张未经过独立基准验证
动作标记器 / 快慢脑连接规划与控制;S1/S2 拆分专有设计未文档化;无公开规格或消融
控制(Psi-C0 / 扩散策略)生成灵巧动作轨迹训练 / 推理依赖 Nvidia 算力算力暴露于美国出口管制
世界模型(Psi-W0)反事实推理;RL 飞轮大规模人类数据储备数据安全制度;准确率未经验证
数据引擎(Psi-SynEngine)采集人手操作数据外骨骼手套硬件;操作员成本 / 质量主张未经验证
具身硬件(灵巧手 / 机身)物理执行动作Schaeffler/THK/NSK 精密部件约 90% 滚珠丝杠来自海外

架构是在世界模型和数据引擎之上搭出的分层感知-规划-动作标记化-控制栈;两个最难的依赖是受出口管制的算力和海外精密部件。

FE001: 产品架构图

Psi 操作栈从感知到规划、动作 token 化、控制和世界模型,底层坐着 Psi-SynEngine 数据引擎,并最终落在 OEM 硬件上。

这是结构栈,不是运行时图;层级边界遵循 Psibot 描述的模块。算力(Nvidia Jetson/Isaac)和具身硬件位于数据引擎之下,并在依赖图中展示。

[CE011, CE013, CE009, CE007]

5.4 部署、集成、可靠性与路线图

Psibot 的部署模式直接来自其授权论点:公司不交付成品机器人,而是把 Psi 模型授权给机器人 OEM,同时销售 Psi-SynEngine 数据系统以及仿真和训练数据服务,把自己的大脑接入合作方硬件。 公开部署证据具体但单薄——一家大型中国物流客户的小规模仓储分拣验证(报道有效率提升),以及 在全球最大光纤光缆制造商之一处测试。研发和验证由 PKU-PsiBot Joint Lab for Embodied Dexterous Manipulation 锚定,该实验室产出 Psibot 公开论文。 可靠性和支持层面,画面是一处缺口:大约 18 个月内迭代五代模型加硬件,节奏异常快;但没有 独立的现场可靠性、在线率、平均故障间隔时间(MTBF)或支持 SLA 数据公开,外部无法 核验产品化成熟度。公司披露的 2026 年运营目标,是建设中国最大的灵巧手数据集,并在年内采集 100 万小时操作数据。从 Psi-R2/W0 发布看,路线图指向数据储备和 RL 飞轮扩张——继续加深模型 及其训练数据,而不是转向自有硬件——这与轻资产、平台授权方策略一致。[CE019, CE020, CE021, CE022, CE023, CE024]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
30 Dec 2024Psi R0 — 首个端到端 RL 具身模型已发布确立 RL 优先路径humanoid.guide / 公司
Mar 2025Psi R0.5 迭代已发布快速迭代节奏humanoid.guide
Apr 2025Psibot V1 / H1 硬件平台已发布模型参考机身humanoid.guide
May 2025Psi R1 — 分层 VLA+RL,L3,麻将演示已发布旗舰能力展示公司 / Rocking Robots
10 Apr 2026Psi-R2 / Psi-W0 双模型;1,000 小时开源已发布延迟 <100ms;数据储备策略公司 / Embodied Global
2026(目标)1,000,000 小时数据;最大的灵巧手数据集进行中数据飞轮扩张公司

路线图显示,Psibot 以异常快的发布节奏转向数据扩张策略;未来里程碑都是公司设定的目标,没有独立可靠性或交付验证。

5.5 差异化、IP 与关键依赖

Psibot 的差异化靠数据飞轮,不靠某个唯一算法。论点是「机器人版 Android」:通过外骨骼手套 Psi-SynEngine 采集自有、低成本的人手操作数据——公司称成本约为传统遥操作的十分之一——再把 数据喂给授权给多家 OEM 机身的 VLA 模型,让每一次部署都丰富数据集、反过来改进模型。公开 IP 足迹真实且外部可见:DexGraspVLA 框架(开源代码和数据集)、VLA 综述和论文清单 (「Awesome-VLA-Papers」),以及通过 Psi-Robot GitHub 组织发布的 Psi SynHand 模型;同时 还有创始人的学术履历(Yaodong Yang、Yuanpei Chen)。 但这种差异化必须放到强可比对象和硬依赖里看。Physical Intelligence 等全球实验室也在走类似 分层 VLA 路线,OpenVLA 这类开放基线(一个 7B 模型,表现超过 55B 闭源 RT-2-X)说明范式并非 专有——Psibot 的优势在数据和灵巧操作重点,不在独一无二的架构。关键依赖都在外部:Nvidia 算力(Jetson 边缘模块和 Isaac 仿真)被中国同行共同使用,也暴露在美国出口管制之下;精密 部件(特种滚珠丝杠约 90% 由德国 Schaeffler 和日本 THK/NSK 供应)是任何具身硬件都绕不开的 需求;PKU 联合实验室和战略投资方则牵引研发和制造需求。这些依赖被明确画出,因为约束 Psibot 操作空间的,与其说是算法,不如说是它们。[CE025, CE026, CE027, CE028, CE029, CE030]

FE003: 关键依赖图

Psibot 模型与数据假设依赖的外部供应商、平台、数据权利、监管和合作伙伴,以及它们如何连接到授权产品。

依赖图是方向性、定性的;边的存在表示重大依赖,不是量化暴露。警示色标记 Psibot 控制之外的依赖(算力、海外零部件、监管)。

[CE029, CE030, CE036, CE021]

5.6 信任、安全、合规与质量控制

信任图景被一条清晰分界线切开:研究经过验证,产品化尚未验证。能力层面,Psibot 的头部指标 ——30 分钟以上自主麻将、100% 翻牌准确率,以及 95%/99.9%「不可能三角」成功率——均由公司披露, 没有独立基准测试;唯一可被外部核验的能力证据,是同行评议的 DexGraspVLA 结果(AAAI 2026 Oral,90%+ 零样本抓取)。独立分析师给叙事降温:MERICS 认为中国人形机器人整体缺乏精度和 灵巧度,依赖特定场地试验,成本仍然太高,并且 VLA 突破仍看向美国研究——这直接检验了成熟度 主张。 安全、安保和质量控制层面存在有记录的缺口:没有公开证据显示 Psibot 拥有安全认证、功能安全 标准,或管控其数据采集和模型授权业务的数据隐私与安全控制。这个问题尤其尖锐,因为 Psibot 正在中国趋严制度下积累 100,000 小时以上人类操作数据储备——2026 年 1 月 1 日生效的《网络 安全法》修正案,罚款最高可达 RMB 10 million,且具备域外适用;数据安全框架也正好覆盖大型 行为数据集。因此,质量结论并不均衡:研究成熟度高(同行评议工作、快速迭代、延迟阶跃变化), 但产品化成熟度——现场可靠性、认证、披露的基准测试和算力供应韧性——薄弱、未验证,或被依赖 关系约束。每一处弱点都对应一项具体尽调请求,而不是一个公司声称的数字。[CE032, CE033, CE034, CE035, CE036, CE037]

信任 / 质量 / 合规表
控制 / 指标状态范围缺口
同行评审能力证据已有(DexGraspVLA,AAAI 2026 Oral)灵巧抓取,90%+ 零样本仅限抓取,不是完整 L3 工作流
独立基准(旗舰主张)缺失麻将、翻牌、95%/99.9% 成功率仅公司披露;未验证
安全 / 功能安全认证无公开证据已部署操作系统未披露标准或证书
数据安全 / 隐私控制无公开证据100,000+ 小时人类数据储备2026 年《网络安全法》下未披露
独立分析师评估反向(MERICS)中国人形机器人灵巧性 / 精度行业在 VLA 突破上落后于美国
现场部署生产率弱(IDC:2025 年超过 85% 非生产性)行业人形机器人部署无 Psibot 特定生产率数据

信任画像分裂很明显:研究能力已验证,但产品化、安全和数据安全控制未经验证;每个未验证项都带有尽调要求。

FE004: 产品成熟度 / 能力图

Psibot 核心能力维度的成熟度:研究 / 演示成熟度强,但产品化和合规信号弱且未验证。

成熟度评级是作者对引用证据的定性综合,不是量化分数;它映射的是已验证研究能力和未验证产品化之间的缺口。

[CE015, CE032, CE035, CE037]

5.7 图表

Chapter 06

06客户

6.1 客户群分层

Psibot 的客户模型直接来自平台授权策略:公司不把成品机器人卖给终端企业,而是把 Psi 视觉-语言-动作模型授权给机器人 OEM 和开发者,由后者把模型装进自己的硬件,再服务终端客户。 按买方/用户/付款方拆分,OEM 通常是买方和付款方,终端企业(仓库或工厂运营方)才是操作能力 的最终用户。垂直场景很具体:物流和仓储——条码分拣、高 SKU 服装配送拣选,这些场景里物体 多样、任务频繁变化,需要 VLA 模型提供泛化——以及制造业,证据是在全球最大光纤光缆制造商 之一处测试。 地理上,客户基础集中在中国,这与 Psibot 的北京/上海布局和中国国家支持的具身 AI 推进一致; 目标细分市场是物流、零售和智能制造仓储。因此,可触达客户集是中国快速扩张的人形机器人市场 里的众多机器人 OEM;通过它们,大型物流买家(State Grid、SF Express、China Post 一类企业) 才变得可达。Psibot 把条码分拣物流选为第一个真实部署场景,正是因为高 SKU 多样性和任务变化 会让刚性自动化失效,泛化大脑才有用武之地——这是有意选择的楔入场景,而不是一次广泛发布。[CU001, CU002, CU003, CU005, CU024, CU035]

客户分层表
客群买方 / 用户 / 付款方使用场景规模收入 / 战略价值缺口
机器人 OEM / 开发者(主要)OEM 购买并付款;终端企业使用将 Psi VLA 模型授权嵌入其机器人被授权方数量未披露核心变现渠道未披露被授权方名称或数量
物流 / 仓储运营商仓库运营商使用;OEM/Psibot 供给条码分拣、高 SKU 拣选一个大客户级试点 + 约 100 台数据采集单元切入点所在场景客户未具名;吞吐量未披露
制造业运营商工厂运营商使用灵巧零部件搬运在一家光纤厂商测试验证跨垂直场景可用仍在测试;范围窄
战略投资方作为渠道Chery / Lens 作为买方渠道OEM / 制造业牵引、生态入口2 家标志性战略支持方降低制造和销售风险关联方;需求并非独立第三方
研究 / 联合实验室PKU-PsiBot 联合实验室模型验证和人才1 个运行中的联合实验室带来可信度和 IP,不是收入不是商业客户

分层反映 Psibot 的平台授权模式:OEM 是直接客户,终端企业是用户;规模数字只是状态指标,不是已披露账户数。

FU001: 客户旅程图

Psibot 平台授权模型的客户旅程:从发现到试点和数据采集部署,再到 OEM 授权和数据飞轮扩张循环。

定性旅程抽象自公司材料和报道;阶段边界是结构性的,留存阶段明确标为未验证,因为没有续约数据。

[CU025, CU009, CU033]

6.2 采用与部署轨迹

采用轨迹仍早,且以试点为主。最实在的数字是到 2026 年初,北京大约部署了 100 台数据采集/ 操作单元,在物流场景采集分布式真实操作数据。关键在于,这一足迹既是部署,也是数据集建设: 仓库工人在真实作业中佩戴 Psi-SynEngine 数据手套,因此所谓「部署」既是一批付费企业账户, 也是一群操作员在拼 2026 年 100 万小时数据目标。这种边界模糊意味着,100 台设备本身不能 证明收入。 除数据采集足迹外,采用证据是某大型中国物流客户的小规模仓储分拣验证(报道有效率提升),以及 在一家主要光纤光缆制造商处测试——这是试点和测试,不是多站点生产账户。Psibot 没有披露活跃 账户数、部署量、利用率或复购数据,因此任何采用主张背后的分母都缺失。行业背景有两面性: Morgan Stanley 把 2026 年中国人形机器人出货预测上调一倍至 50,000 台,说明从试点走向生产 的动量真实存在,并利好 Psibot 的 OEM 被授权方;但市场本身仍早,Psibot 自己从试点转为稳定 生产账户的能力尚未证明。[CU004, CU006, CU008, CU009, CU023, CU026]

客户增长 / 采用轨迹表
指标日期来源置信度含义 / 缺失分母
北京数据采集单元约 100 台2026 年初报道(news18a / Houdao)操作者群体,不是付费账户;没有总分母
公开提及的物流试点1 个(未具名客户)2026Tech Times / Gasgoo无吞吐量、账户数或站点数
制造业测试1 个(一家光纤厂商)2026Tech Times / 公司测试阶段;结果未量化
活跃账户 / 被授权方未披露2026公司(未披露)None核心采用分母缺失
使用率 / 复购未披露2026公司(未披露)None无法区分使用和部署
2026 年数据采集目标1,000,000 小时2026(目标)公司表明操作者覆盖面扩大,不代表收入

唯一具体采用数字是约 100 台数据采集单元;所有商业采用分母(账户、使用率、复购)均未披露。

FU002: 采用 / 部署漏斗

从发现到扩张的路径,标出 Psibot 已知合作所处位置;生产和续约阶段没有披露,并相应标记。

定性漏斗以流程呈现,因为没有披露各阶段数量;下游两个阶段标为警示,以反映缺少生产和续约证据。

[CU006, CU011, CU036]

6.3 具名客户证据与参考质量

在尽调尺度上,这里的证据最弱。Psibot 没有公开披露任何具名生产客户;客户要么被泛称为「一家 大型物流客户」「全球最大光纤光缆制造商之一」,要么是战略投资方,而非独立第三方买家。光纤 测试和物流分拣试点是两个最具体的合作,但二者都停留在试点/测试阶段,结果未量化,也没有参考 客户背书。若按重视具名、生产级、结果可衡量参考客户的尽调标尺,尽管投资人验证很强,Psibot 的客户证据质量仍低。 生态背景必须谨慎解读,避免给过高信用。Chery Automobile 旗下 AiMOGA Robotics 发布了自己的 人形机器人(Mornine),并在 2026 年 4 月签下 1,000 台智能警用机器人订单;Chery/AiMOGA 生态覆盖 30 多个国家——但没有确认 AiMOGA 的机器人运行 Psibot 的 Psi 模型,因此这笔订单是 生态背景,不是 Psibot 的直接客户证据。诚实的说法是,Psibot 的具名证据表主要列出匿名试点 和战略投资方关系;短期最有价值的尽调动作,是拿到具名、生产级、结果披露的参考客户。[CU007, CU010, CU011, CU027, CU028, CU031]

具名客户证明表
客户客群部署 / 使用场景生产或试点结果限制
未具名中国大型物流客户物流 / 仓储仓库分拣验证试点据称分拣效率提升客户未具名;结果未量化
未具名光纤光缆厂商制造业灵巧零部件搬运测试测试跨垂直场景适用性信号测试阶段;未披露指标
机器人 OEM 被授权方OEM 渠道将 Psi 模型授权嵌入机器人试点 / 早期商业化核心变现路径未具名也未披露被授权方数量
Chery / AiMOGA 生态战略支持方 / 渠道生态合作;AiMOGA 人形机器人生态(未确认由 Psi 驱动)支持方 + 潜在 OEM 渠道未确认 AiMOGA 机器人运行 Psi 模型
Lens Technology战略支持方 / 制造渠道具身智能中心建设战略投资制造 + 需求牵引未披露为独立第三方客户

枚举不完整——Psibot 没有披露完整客户名单,因此本表覆盖所有公开可识别的合作(具名或泛称),而非完整名录;证据缺口指向缺失的具名生产参考。

[CU006, CU007, CU010, CU012, CU031]
FU003: 客户证明矩阵

从证据强度、结果具体性、生产成熟度和留存可见度四个维度评估客户证明质量。

评级是作者对引用证据的定性综合,不是量化分数;它显示所有合作的生产成熟度普遍偏低,且留存可见度缺失。

[CU010, CU028, CU031]

6.4 留存、复购与耐久性

留存没有可衡量数据,这种缺席本身就是结论。Psibot 没有披露净收入留存、毛留存、流失率、续约 率或合同期限,也没有任何队列或复用数据。两个结构性事实进一步放大缺口:公司成立于 2024 年,还太年轻,尚未跑完有意义的续约周期;采用仍在试点阶段,几乎没有可评估耐久性的 合同。客户满意度分数、NPS 或参考质量指标也没有披露。 关于耐久性,只能做定性且前瞻的判断。若 OEM 被授权方基于 Psi 模型构建产品并积累部署历史, 切换成本会上升,Psibot 的自有数据集和 OEM 关系会变成锁住客户的耐久资产。但这是一条论点, 不是已经展示出来的留存曲线。行业背景也提示谨慎:IDC 报告称,2025 年超过 85% 的人形机器人 部署属于非生产性,因此即便机器人已经部署,持续生产性使用——续约的前提——也远未得到保证。 本章每一个留存字段因此都是空值,并附带具体尽调请求。[CU014, CU015, CU019, CU032, CU033]

留存 / 重复使用 / 满意度表
指标值 / null客群置信度尽调问题
净收入留存(NRR)全部none按部署队列披露 NRR
总留存(GRR)全部none提供 GRR 和客户数留存
流失 / 续约率全部none按账户披露续约和流失
合同期限OEM 被授权方none典型授权合同期限
客户满意度 / NPS全部none满意度 / NPS 或客户背书质量
复购 / 扩张全部none增购和多站点扩张历史

所有留存和满意度字段都是 null;公司处于试点阶段且成立于 2024 年,因此还没有完整续约周期。每个 null 都对应一项具体尽调问题。

留存队列 / 重复使用数据可得性
队列 / 期间留存数据状态不可得原因
第 1 年(2024 年批次)不可得2024 年成立;没有完成续约周期
第 2 年(2025 年批次)不可得试点阶段;未披露合同续约
2026 年试点过早部署仍在进行;队列尚未成熟
数据采集操作者不适用操作者群体,不是付费客户队列

无法构建时间序列留存队列,因为 Psibot 未披露续约数据,且公司太年轻,尚无成熟队列;本表用可得性缺口替代队列图。

6.5 扩张、集中度与渠道依赖

Psibot 设想的扩张动作,是通过 OEM 授权先落地后扩张,再由数据飞轮放大——每一次部署丰富 数据集,数据集改进模型,模型再赢下下一个 OEM。但公司尚未披露账户扩张、追加销售或多站点 铺开证据,采购动态(销售周期、合同期限、定价)也未披露,因此无法评估买方摩擦。扩张 逻辑在结构上说得通,在经验证据上尚未证明。 集中度和渠道依赖是最尖锐的风险。早期需求与战略投资方相关——Chery、Lens Technology,以及 国资相关基金(GL Ventures、Lanchi Ventures、SDIC Advanced Manufacturing Fund、Jingxi Ruiling)——而不是与可证明的独立第三方商业买家相关。这类背书确实有价值:它缩短销售周期、 降低制造进入风险,即便没有具名生产客户,也构成高质量验证。但它是一把双刃剑,会把需求集中 到少数关联方,并提出一个问题:早期收入到底是商业性的,还是战略性的。中国地理集中度构成 第二条集中度轴线,把采用绑定到国内国企采用要求,同时限制国际扩张。尽调优先事项是关联方 交易清单和前几大客户集中度披露。[CU012, CU013, CU016, CU017, CU018, CU025]

扩张和集中度风险表
扩张驱动集中度风险影响尽调路径
OEM 切入后扩张 + 数据飞轮披露的被授权方很少扩张逻辑成立但未验证被授权方名单和扩张历史
战略投资方渠道(Chery/Lens)需求集中在关联方收入可能来自战略合作而非独立第三方关联方交易清单
国资关联资金牵引依赖政策驱动需求采用绑定国企指令国资关联买方需求占比
数据采集覆盖面扩张操作者,不是付费账户混淆部署和收入付费单元与数据采集单元拆分
中国地域集中单一国家风险敞口限制国际扩张海外管线和出口计划

OEM 授权让扩张在结构上说得通,但实证上尚未验证;主要风险是早期需求集中在战略投资方和国资关联买方,而非独立第三方客户。

6.6 图表

Chapter 07

07风险

7.1 按严重程度排序的风险概览

Psibot 的风险可以排出清晰的严重程度。严重度最高、最难缓释的一组,是中国特有的监管和法律 暴露——《国家情报法》、修订后的《网络安全法》,以及《数据安全法》/PIPL 跨境制度——因为它 是结构性的(附着在 Psibot 的国籍和数据中心模式上),直接限制可触达的国际客户和投资者基础, 且公司自身再怎么行动也无法完全中和。第二高的一组是商业证明和炒作风险:未经验证的基准测试、 试点阶段可靠性、全行业过度承诺,以及没有收入锚的估值。依赖风险(Nvidia 算力、精密部件) 和人才/执行风险(关键人物集中、公司 2024 年才成立)处在重要但可部分缓释的层级。 传导机制决定投资论点怎么受伤。监管风险会传导为国际需求和融资可选性受限;商业证明风险会 传导为试点转生产变慢,进而压到收入,最终压到估值;依赖风险会传导到成本、毛利率和供应连续性; 人才风险会传导到执行和路线图交付。因为 Psibot 没有披露收入,每一项下游风险最终都最重地 砸在估值上;当前估值靠团队质量、数据飞轮和战略投资方支撑,而不是基本面。下文热力图和传导 图会按可能性和残余严重度给这些暴露排序,并追踪它们如何流入财务模型。[CR040, CR031, CR037, CR009]

FR001: 风险热力图

按发生可能性、影响、缓释成熟度和残余严重度给风险排序;监管和商业证明风险主导高残余象限。

评级是作者对引用证据的定性综合,不是量化评分;缓释措施最难覆盖的是监管准入、商业验证和估值风险, 因此这些残余严重度最高。

[CR040, CR001, CR031]
FR002: 风险传导图

Psibot 的主要风险会传导到收入、客户、利润率、融资,最后落到估值;缺少收入锚时, 估值吸收的敞口最大。

定性传导模型;连线方向表示主导因果流。Psibot 未披露收入,因此每条下游路径都落到估值。

[CR031, CR009, CR040]

7.2 监管与法律风险

监管和法律暴露是 Psibot 的定义性风险。作为一家中国公司,它受《国家情报法》(2017 年)约束; 该法第 7 条写明:「任何组织和公民都应当依法支持、协助和配合国家情报工作。」独立法律分析 (China Law Translate)给恐慌式解读降温:这项义务位于总则,类似其他中国法律中的配合义务, 也没有专门执行机制——但同一分析也认为,中国公民或公司无法有意义地抵抗直接的国家安全请求, 也不能指望法院提供救济。对一家全部价值都在自有操作数据集上的公司来说,这种残余的强制访问 风险,对任何海外客户或投资者都很重要;美国国土安全部商业咨询通告也正式化了西方监管视角: 中国公司处理的数据带有这类暴露。 2026 年活跃的监管议程进一步放大暴露。修订后的《网络安全法》于 2026 年 1 月 1 日生效,把 一般行政罚款上限从 RMB 1 million 提高 10 倍至 RMB 10 million,明确纳入 AI 伦理、风险监测 和安全监管义务,并把域外适用扩大到任何「危害中国网络安全」的海外活动。其上还有《数据安全法》 和 PIPL,要求数据分类并设置跨境传输制度(CAC 安全评估、标准合同条款或认证)——Psibot 的 模型依赖大量操作数据的积累和流动,因此直接相关。外部地缘政治暴露又叠在上面:美国出口管制 审查,以及海外上市/采购限制(困扰同行 Unitree 的同类「间谍法」担忧)会约束 Psibot 出海销售 或海外融资。Psibot 没有披露诉讼、执法行动、IP 纠纷,或必须持有的具体牌照,留下了下方登记表 跟踪的尽调缺口。[CR001, CR002, CR003, CR004, CR005, CR006]

监管 / 法律风险登记表
规则 / 许可 / 案件司法辖区状态概率严重度缓释剩余风险敞口尽调路径
《国家情报法》(2017)第 7 条强制协助中国生效中数据本地化;可披露保障有限高 — 压低海外客户 / 投资人信任上限国家数据访问请求记录;律师意见
修订版《网络安全法》(2026 年 1 月 1 日生效)中国(域外适用)生效中已记录的合规项目;快速整改中高 — RMB 10M 罚款、AI 治理义务合规项目审查;事件响应准备度
《数据安全法》/ PIPL 跨境机制中国生效中CAC 评估 / SCCs / 认证中 — 约束数据流动和审计跨境传输批准;审计记录
美国出口管制 / 海外上市审查美国 / 盟友执行中国产算力替代;中国优先销售高 — 限制国际触达和融资出口管制律师意见;海外收入敞口图
诉讼 / IP / 执法历史中国 / 全球未披露Unknown未披露未知 — 披露前无法量化诉讼和执法清单;IP 自由实施分析

各行按剩余严重度排序。中国 AI / 数据监管在 2026 年仍快速演进,且 Psibot 未披露诉讼或执法历史,因此覆盖并不完整;证据缺口指向缺失的执法和许可记录。

[CR001, CR003, CR006, CR009, CR010]

7.3 运营、质量与安全风险

运营上,Psibot 的核心风险是灵巧操作尚未在生产级可靠性上得到证明。IDC 报告称,2025 年超过 85% 的人形机器人部署属于非生产性;MERICS 判断中国具身 AI 部署依赖特定场地、需要手工调校、 成本过高,广泛采用前成本至少要下降一半。Psibot 授权的是智能层而非自造机器人,因此真实世界 质量还取决于 OEM 合作方把 Psi 模型接入其硬件的效果——这是一片协同界面,Psibot 只部分控制, 因为它设计结构和运动范围,但外包零部件生产和制造。 叠加其上的是验证缺口:Psibot 的招牌能力主张——30 多分钟自主打麻将,以及低至同行十分之一的 训练数据成本——均由公司披露,没有独立基准测试;更广泛的行业还被批评为编排式演示,无法转化 为可靠、长时间、广变化的部署。模型依赖大规模人类手套数据采集,数据质量和数据集完整性风险 真实存在;Psibot 也没有披露产品安全记录、事故历史或责任保险覆盖。下方运营登记表按严重度 排列这些失效模式,并标出尚未解决的安全与可靠性披露。[CR011, CR012, CR013, CR014, CR015, CR016]

运营 / 质量 / 安全风险登记表
失效模式概率严重度缓释成熟度剩余风险敞口未解决缺口
生产规模下操作不可靠低 — 仅处于试点阶段高 — 2025 年超过 85% 部署未进入生产未披露生产可靠性指标
基准未获第三方验证低 — 仅公司自报中高 — Mahjong / 数据成本说法未经审计没有独立基准验证
依赖 OEM 集成质量低 — Psibot 控制设计,不控制制造中 — 质量随被授权方而变未披露集成 QA
数据质量 / 数据集完整性风险中 — 自有采集引擎中 — 手套数据有噪声 / 偏差风险未披露数据质量验证
产品安全 / 事故敞口低中未知 — 无披露未知 — 无事故或保险记录无安全认证、事故日志或责任险

各行按剩余严重度排序。核心运营风险是灵巧操作在规模化场景下还没有证明可靠,同时基准未经验证,安全记录也未披露。

7.4 合作方与依赖风险

Psibot 的依赖图谱有三个关键节点。第一是算力:中国具身 AI 开发者普遍依赖 Nvidia Jetson 边缘模块和 Isaac 仿真/训练工具,二者暴露于美国出口管制政策——一旦获取收紧,可能抬高成本或 削弱训练能力,是供应和能力尾部风险。第二是精密机械部件:行业供应链分析显示,海外供应商 (德国 Schaeffler、日本 THK 和 NSK)主导灵巧系统所需的特种滚珠丝杠和高端谐波减速器,这是 中国正在缩小但尚未消除的瓶颈。第三是资本和渠道对手方:Psibot 早期需求和融资集中在战略投资方 (Chery、Lens Technology)和国资相关基金,收入路径则经过 OEM 被授权方,而不是多元化的独立 第三方客户。 这些依赖会彼此部分抵消。中国控制着估计 63–70% 的人形机器人供应链,并在电机、执行器、电池 以及越来越多的减速器上快速国产化,这缓冲了机械部件风险,也给 Psibot 的被授权方带来成本和 物流优势。但前沿算力依赖更难本地化,早期需求集中在关联方也提出一个问题:Psibot 的进展 是商业性的还是战略性的。中国单一国家集中度把 Psibot 绑定到国内政策周期和国企采用,同时限制 国际多元化。下方依赖登记表和地图按集中度与残余暴露给每个对手方排序。[CR019, CR020, CR021, CR022, CR023, CR024]

合作伙伴 / 依赖风险登记表
依赖项交易对手角色集中度失效情景严重度缓释剩余风险敞口
前沿 / 边缘算力Nvidia(Jetson、Isaac)AI 算力和训练工具高(全行业)美国出口管制收紧准入国产 AI 加速器替代高 — 前沿差距难以本地化填补
精密部件Schaeffler、THK、NSK滚珠丝杠、高端减速器高(某些部件约 90%)进口限制或价格冲击中高中国本土产能上升中 — 正在收敛但未消除
战略资本 / 渠道Chery、Lens、国资关联基金资本、制造、需求高(关联方)支持方撤资或政策转向中高逐步分散头部支持方中高 — 关联方需求
收入渠道OEM 授权客户模型落到硬件高(未分散)OEM 采用慢或流失拓宽授权客户基础中 — 披露授权客户少
地理市场中国国内市场主要需求盘高(单一国家)国内政策或需求周期借 Chery / AiMOGA 输出生态中 — 国际分散度有限

各行按剩余严重程度排序。算力依赖最难本地化;机械部件依赖有中国 63–70% 供应链控制率缓冲;资本 / 渠道集中则对应关联方风险。

FR003: 依赖关系图

梳理 Psibot 的关键合作伙伴、供应商、资本提供方和监管者,标出生态中集中度和单一来源敞口所在的位置。

定性依赖关系图;节点颜色反映集中度和单一来源敞口,其中算力和监管节点标为最严重。

[CR019, CR021, CR026]

7.5 财务、模型与人才/执行风险

财务上,主导风险是不透明:Psibot 没有披露收入、ARR、毛利率、现金消耗率或现金续航,因此约 $1.48B 估值靠团队质量、数据飞轮和明星投资方支撑,而不是基本面。具身 AI 研发高度资本密集 ——同行每年投入数亿美元到十多亿美元——所以 Psibot 迄今约 $300 million 融资意味着现金续航 有限,未来很可能反复融资;一旦基准测试不及预期或资本市场收紧,公司会暴露于降价融资或 稀释风险。行业泡沫放大了估值风险:Morgan Stanley 调研发现,尽管预测被上调一倍,潜在工业 买家中只有 23% 对当前人形机器人产品满意;2026 年约 22 家具身 AI 独角兽诞生——这是典型泡沫 信号,会提高估值倍数压缩的概率。 人才和执行风险高度集中。Psibot 的可信度依赖一个小团队——CEO Viktor Wang,以及一群学术带头人 (Peking University 的 Yaodong Yang、来自 Stanford/Fei-Fei Li 圈层的 Yuanpei Chen、SJTU 的 Ying Wen)——任何关键人物离开,都会伤到能力和叙事;具身 AI 人才市场竞争也极为激烈。作为 一家 2024 年才成立的公司,Psibot 还要同时扩张硬件、数据运营和 OEM 关系,承担非常年轻组织 常见的执行风险。下方人才/执行登记表按严重度排序,并给出尽调路径。[CR029, CR030, CR031, CR032, CR033, CR034]

人员 / 执行风险登记表
角色 / 职能依赖或缺口可能性严重性缓释措施尽调路径
创始人 / CEO(Viktor Wang)愿景、商业化、融资低-中联合创始人梯队深创始人留任条款;归属期;继任安排
首席科学家 / 学界领军人(Yang、Chen、Wen)核心模型 IP 与可信度PKU 联合实验室;多名负责人留任、竞业限制和 IP 归属审查
技术人才留存稀缺具身 AI 工程师股权激励;学术人才管线流失数据;薪酬对标
年轻组织的执行2024 年成立;同步扩张引入有经验的运营管理者组织架构、招聘计划、运营成熟度审查

各行按严重性排序。公司依赖研究可信度,创始人和学术负责人身上的关键人集中度是最主要的人才风险。

7.6 缓释、监测与否决标准

Psibot 最强的缓释因素是投资方和数据资产。明星战略投资者(Chery 及其 AiMOGA 机器人子公司、 Lens Technology)和国资相关基金,降低了资本获取以及制造/商业化渠道风险;自有操作数据集 ——2026 年目标为 100 万小时——如果能证明会带来模型质量,就是一条真实且会复利的护城河。这些 缓释因素真实但不完整:它们更能解决融资和渠道风险,远不如解决监管准入、商业证明或估值风险; 后几项仍是支撑论点的关键不确定性。 监测上,价值最高的触发指标包括:第三方独立验证(或推翻)Psibot 的基准测试主张;付费、独立 第三方生产部署而非试点和数据采集单元的证据;美国出口政策下 Nvidia/算力获取的任何变化; 《国家情报法》或《网络安全法》下已披露的执法或强制数据访问事件;以及关键创始人或首席科学家 离职。论点失效标准也由此直接推出:核心基准测试可信验证失败、前沿算力获取长期丧失、 国家驱动的数据访问事件让国际需求失效,或降价融资重置估值论点。下方缓释与否决标准表把 每项主要风险配上可监测触发因素、阈值和行动含义;本章尽调事项——审计财务、具名生产参考客户、 关联方清单、部件 BOM,以及诉讼/执法历史——是解决残余不确定性的最快路径。[CR037, CR038, CR039, CR040]

缓释与叫停标准表
风险可监控触发点阈值 / 事件行动含义
未验证基准独立第三方评测核心主张被证伪或无法复现投资论点破裂 — 重新评估技术护城河
监管 / 数据访问暴露披露强制数据访问或执法事件任何国家数据访问事件拖累需求投资论点破裂 — 国际需求受损
算力依赖Nvidia / 美国出口政策变化持续失去前沿算力访问升级关注 — 模型训练和成本承压
缺乏基本面的估值下一轮融资条款降价轮或融资失败重置估值论点;稀释风险
商业验证付费、独立交易的生产账户12–18 个月内试点没有转化下调判断 — 牵引力来自战略而非商业
关键人依赖创始人 / 首席科学家离职任何关键人物离开升级关注 — 能力与叙事受损

每项主要风险都配一个可监控触发点、一个阈值事件,以及它对投资论点意味着什么;基准、数据访问和估值触发点最值得优先跟踪。

7.7 图表

Chapter 08

08估值

8.1 投资论点与反论点

Psibot 的投资故事建立在定位、团队和数据上,而不是任何已披露财务表现。牛市论点是,把一个 视觉-语言-动作(VLA)「大脑」授权给机器人 OEM,切中了具身 AI 技术栈里最可防守、最可扩张 的层:如果 Psi R1 平台成为中国灵巧操作的事实智能标准,Psibot 就能在许多硬件厂商之上获取 软件式经济性,而不背负它们的资本密集度。这个论点由一个履历异常强的创始团队强化 (Peking University 的 Yaodong Yang、经由 Yuanpei Chen 连接的 Stanford / Fei-Fei Li 传承,以及 Alibaba / Tencent 机器人老兵 Xiaojie Chai),也由建设中国最大灵巧手数据集作为 复利数据飞轮的目标、以及提供资本和产业牵引的战略投资方(Chery Automobile 和 Lens Technology) 强化。市场背景很大:WAIC 2026 参展公司合计估值超过 US$14.7 billion,仅 2026 年上半年就有 15 家中国具身 AI 初创公司跨入独角兽行列。 反论点同样具体。Psibot 没有披露收入、ARR、毛利率或员工人数;头部能力主张(L3、30 分钟自主 麻将演示)由公司自报,没有独立基准测试;行业分析师估计,超过 85% 的具身 AI 部署仍是非生产性 试点;《国家情报法》和数据安全制度限制了全球可比本应暗示的国际需求。早期进展也可能 来自关联投资方的战略牵引,而不是独立第三方商业需求,从而削弱价格背后的收入质量信号。简言之, US$1.48 billion 估值是对执行的期权,不是结果的倍数;举证责任完全在公司一边。[CV001, CV003, CV004, CV005, CV006, CV007]

投资论点 / 反论点表
论点支柱多头论据反论点反驳
商业模式授权「大脑」可拿到类似软件的可扩张经济性未披露收入、ARR、利润率或员工数
团队与数据精英团队叠加最大灵巧手数据集飞轮数据护城河未证实;能力主张缺少独立基准
市场TAM 大;WAIC 企业超过 US$14.7B;2026H1 新增 15 家独角兽超过 85% 部署仍是非生产性试点
支持方Chery 和 Lens 提供资本与产业拉力牵引力可能来自战略关系,而非独立交易需求
地域在中国快速扩张赛道处于领先位置NIL 和数据制度压住国际需求上限

多头论点支柱与对应空头反驳并列。

[CV001, CV003, CV005, CV006, CV007]

8.2 建议、信心、风险评级与估值立场

我们的建议是有条件观望:在拿到具名生产参考客户、审计财务报表和关联方收入清单之前,不要按当前 US$1.48 billion 投后估值领投或跟价。可选性真实存在,但按入场价看下行不对称,因此条款纪律 比对故事的信念更重要。我们对这一判断的信心为低到中:公开证据基础单薄,主要来自公司表述和一次 融资事件的媒体报道,几乎没有独立、可审计的运营数据。整体风险评级为高,驱动因素包括商业证明缺席、 能力基准未经验证、监管和数据访问暴露,以及分析师预计 2027-2028 年 18-24 个月现金续航 到期后,行业会进入全面「淘汰年」。 我们的估值立场是,US$1.48 billion 相对已披露基本面偏高——只有拿团队、数据飞轮、战略投资方信号, 以及中国和全球具身 AI 可比公司的高估值集合来解释才站得住;但这些都不能替代收入。建议逻辑经过 四道门:平台经济性是否可扩张?能力和商业牵引是否有独立证明?监管和执行风险是否受控? 入场价格是否有支撑?Psibot 通过了第一道门,今天没有通过第二道门,第三道门也只部分通过,第四 道门则是有条件通过——所以结论是守价格纪律的观望,而不是直接放弃或买入。下方尽调计分卡显示, 机会在市场和护城河上得分高,但在商业证明、单位经济模型、风险控制和证据质量上得分低。[CV009, CV010, CV011, CV012, CV014]

建议摘要表
维度评估依据
建议有条件观察 — 当前价格不领投可选性真实;US$1.48B 估值下下行不对称
置信度低至中公开证据薄;仅一个融资事件
风险评级缺少商业验证、基准未验证、存在监管暴露
估值立场相对基本面偏高只能靠团队、数据飞轮、支持方和可比公司支撑
决策含义严守价格纪律进场,留出降价轮安全边际要求具名参考客户、审计财务、关联方明细表

围绕核心决策维度给出一句话判断;每行都有建议部分的主张支撑。

[CV009, CV010, CV011, CV012]
FV001: 建议逻辑

决策闸门最终导向有条件观察建议;Psibot 通过了可扩展性关,但当前还没过验证关,风险约束只过了一半, 价格支撑也只是有条件成立。

关卡评估是作者对引用证据的定性综合;节点颜色标出 Psibot 在每一关是通过(正向)、未通过(负向) 还是部分通过(警示)。

[CV009, CV014]
FV004: 投资 KPI

尽调记分卡(0-10;分数越高越有利);机会的市场和护城河得分高,但验证、单位经济、风险约束和证据质量偏弱。

评分是作者在 0-10 分制下综合所引主张得出的定性尽调判断,不是量化模型输出。

[CV004, CV005, CV011, CV012]

8.3 融资背景与入场纪律

Psibot 累计融资超过 US$300 million;2026 年 7 月由 Chery Automobile 和 Lens Technology 领投 的一轮约 US$100 million 融资,将投后估值定在 US$1.48 billion;此前还有一轮约 RMB 2 billion(约 US$280 million)的融资。由于公司处在收入前阶段,这一估值来自投资人信号、 顶尖具身 AI 团队稀缺性,以及产业投资方的战略牵引,而不是单位经济模型——公开证据可以证实这轮 融资事实,却无法独立验证其公允价值。 因此,入场纪律压倒一切。既是投资方、也可能消费 Psibot 技术的战略投资方设定了出清价格,所以 独立第三方财务投资人应预期支付溢价,并应按一次或多次未来稀释性融资的真实概率控制仓位:行业 18-24 个月现金续航意味着流动性事件前会反复要钱,形成优先权堆栈压力。降价融资风险已经 偏高且还在上升,因为 Unitree、DEEP Robotics 和 Leju 即将公开上市,正在把中国机器人估值方法 从自上而下 TAM 外推,重置为可审计交付、收入和盈利能力——这种重置对一家靠叙事定价的收入前平台 结构性不利。在接近当前估值投入资本之前,我们会要求审计财务、具名客户参考清单、关联方收入清单 和优先权堆栈披露。[CV002, CV015, CV016, CV017, CV018, CV019]

8.4 牛市、基准与熊市情景

牛市情景假设 Psibot 的能力主张得到独立验证,OEM 授权在多个硬件合作方之间放大,灵巧手数据飞轮 复利成耐久护城河;在这些条件下,平台可重估至 US$4-6 billion 或更高,与软件中心型全球同行的 溢价一致。基准情景假设试点转化缓慢,能力在规模化上仍可信但未证明,Psibot 继续作为一家私有 独角兽,估值落在约 US$1.5-2.5 billion 区间,同时继续消耗资本来延长现金续航。熊市情景假设 基准测试不及预期、监管或算力获取冲击,或在市场收紧时融资失败,迫使降价融资或出现困境结果, 估值落在 US$0.3-0.7 billion 区间甚至更低。 主导下行触发因素包括头部基准测试被证伪、失去 Nvidia 级算力访问、数据安全或《国家情报法》执法 事件封死国际需求,以及现金续航到期前无法完成下一轮融资。基于今天可获得的公开证据,基准情景 是最可辩护的中枢估计;牛市情景需要尚不存在的独立证明,行业预期中的 2027-2028 年整合也提高了 熊市尾部权重。隐含价值对两个假设高度敏感:商业里程碑达成概率,以及一旦出现可审计指标,市场 愿意套用的收入倍数。[CV022, CV023, CV024, CV025, CV026, CV027]

多头 / 基准 / 空头情景表
情景关键驱动隐含估值
多头基准获验证、OEM 授权放量、数据护城河复利约 US$4-6B 或更高
基准试点转化慢,能力可信但未证实约 US$1.5-2.5B(维持私有独角兽)
空头基准落空、监管或算力冲击、融资失败约 US$0.3-0.7B 或陷入困境

用驱动因素和隐含估值区间框定情景。

[CV022, CV023, CV024]
FV002: 估值敏感性

不同情景和参考点对应的隐含企业价值(US$B);围绕当前估值,上下区间横跨一个数量级。

数值是由可比集和情景判断推导出的指示性情景中点,不是贴现现金流结果;Psibot 没有收入, 无法锚定精确倍数。

[CV028, CV022, CV023, CV024]
FV003: 估值 / 回报区间

低—中—高情景下的隐含估值(US$B);悲观到乐观的跨度凸显,估值有多大一部分压在尚未验证的里程碑上。

区间是锚定可比估值的情景判断;未按概率加权,也不包含所列区间之外的尾部结果。

[CV022, CV023, CV024]

8.5 可比估值

公开可比公司能框定争论,但不能终结争论。中国侧,Unitree 获批科创板上市意味着市值接近 RMB 42 billion(约 US$5.83 billion);CCB International 预计若计入品牌溢价,最高可达 RMB 109 billion(约 US$15.1 billion),相当于一家 2025 年出货超过 5,500 台机器人且已经盈利 的公司,对应 32x 市销率。DEEP Robotics 的收入约为 Unitree 的五分之一,但按约 RMB 13.9 billion (约 US$1.93 billion)发行估值计算,隐含市销率更高,达 41x。更广泛的群体中,至少 25 家国内 具身智能公司估值已经超过 RMB 10 billion(约 US$1.39 billion),其中 15 家在 2026 年上半年 成为独角兽。 全球侧,Figure AI 在 2025 年达到约 US$39 billion,Physical Intelligence 据报洽谈估值超过 US$11 billion,Skild AI 讨论过约 US$8-12 billion 区间,Apptronik 接近 US$5.5 billion 完成 融资。关键限制是可比性:这些同行大多处于收入前或按远期倍数定价,公开市场公司则是全栈硬件 集成商,而 Psibot 是软件 / 智能层授权方。这种错配意味着,可比公司能支撑巨大可寻址可选性, 但不能给 Psibot 具体商业模式一个精确公允价值;IPO 群体带来的估值方法重置转向可审计指标, 也可能压缩这一群体当前享有的账面溢价。[CV029, CV030, CV031, CV032, CV033, CV034]

可比估值表
公司估值 / 隐含价值倍数或备注
Unitree(科创板 IPO)隐含约 US$5.83B;计入品牌溢价最高 US$15.1B32x 市销率;已盈利;2025 年出货超 5,500 台机器人
DEEP Robotics发行估值约 US$1.93B41x 市销率;收入约为 Unitree 的五分之一
Figure AI约 US$39B(2025)全球领先人形机器人平台;按远期倍数定价
Physical Intelligence超过 US$11B(据报道谈判中)偏软件的美国基础模型同业
Skild AI / Apptronik约 US$8-12B / 约 US$5.5B全球具身 AI 可比公司的上沿
中国独角兽群25+ 家公司估值超过 US$1.39B;2026H1 新增 15 家多数尚未有收入;按叙事定价

覆盖并不完整:多数同业仍未有收入或保持私有,也没有纯软件授权模式的干净公开可比公司;证据缺口同样受这一完整性限制影响。

[CV029, CV030, CV031, CV033, CV034]

8.6 退出路径、论点失效触发因素与尽调事项

Psibot 可能的退出路径有两条:沿着 Unitree 和 GigaAI 正在打出的路线,走境内科创板或香港上市;或被 Chery 这类已经依赖其技术的产业支持方战略收购。短期退出准备度很低:公司尚未产生收入,仍是私营状态;任何公开市场窗口打开前,都需要拿出可验证、可审计的商业牵引力,尤其投资人很快会有三个更细颗粒度的上市参照,市场会奖励交付和盈利能力。 投资论点会在几种情况下失效:核心基准被独立证伪;授权业务无法越过绑定战略伙伴放大;监管或出口管制事件切断算力或数据访问;或公司在现金跑道耗尽前融不到钱。任何资本承诺前,最终尽调清单包括:经审计财务报表和收入确认政策;可联系的具名客户推荐名单,并区分独立交易收入与关联方收入;覆盖 Chery 和 Lens 的关联方交易表;考虑出口管制敞口后的组件 BOM 和算力供应计划;完整股权结构表及清算优先权层级;以及对 L3 操作和麻将自主性主张的独立第三方验证。总体看,机会高风险、高期权性;我们建议有条件通过,只有在上述证据交付、且价格按明确的下轮估值下调安全边际谈妥后,才具备可投性。[CV036, CV037, CV038, CV039, CV040]

论点破裂与叫停触发点表
触发点观察信号后果
基准被证伪独立测试无法复现 L3 / 麻将主张核心能力前提坍塌;空头情景
算力访问丧失出口管制切断 Nvidia 级别供应路线图停滞;竞争劣势
监管 / 数据事件NIL 或数据安全执法封死需求国际 TAM 消失;估值下调
授权无法放量除绑定支持方外,没有独立交易的 OEM 采用收入质量论点破裂
融资失败下一轮完成前现金跑道耗尽2027–2028 洗牌中降价轮或陷入困境

会推翻投资论点的事件,按与核心能力的关系由近到远排序。

[CV025, CV039, CV027]
最终尽调清单
尽调事项重要性状态
审计财务和收入确认政策用基本面校验估值未披露
具名客户参考清单区分独立交易收入与关联方收入未披露
关联方交易明细表评估 Chery / Lens 收入质量未披露
股权结构表和优先权堆栈量化稀释和下行保护未披露
独立基准验证确认核心能力前提未披露

任何投钱前都需要上述证据;目前每一项都未披露。

[CV038, CV013, CV019]

8.7 展示材料

免责声明

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

证据索引

结论
编号陈述可信度来源
CO001 Psibot (also known as Lingchu Intelligence / 灵初智能) was founded on 1 September 2024 and operates in Beijing and Shanghai, China. SO001, SO004, SO011
CO002 Psibot describes itself as a leading Chinese embodied-AI company focused on general-purpose embodied intelligence, VLA models, and dexterous-manipulation algorithms. SO001, SO021
CO003 A VLA model maps camera images and natural-language instructions directly to low-level motor commands, collapsing separate perception, planning, and control pipelines. SO003, SO014
CO004 Psibot's stated mission is to "create infinite productivity with AI and robots" and its vision is to "become a global leader in intelligent robotics." SO001, SO025
CO005 Psibot operates a "small full-stack" model, controlling system architecture, degrees of freedom, and motion range while outsourcing component production and manufacturing. SO006, SO007
CO006 Psibot licenses its Psi-series model to third-party robot developers and sells a proprietary human-hand data-collection system (Psi-SynEngine). SO003, SO008
CO007 Psibot's founder and CEO is Dr. Viktor Wang (Wang Qibin), who previously led JD.com's robotics business and was VP of products at Yunji Technology. SO003, SO004, SO006
CO008 Co-founder and engineering leader Dr. Xiaojie Chai has more than fifteen years of robotics and autonomous-driving experience at Alibaba and Tencent, including L4 self-driving deployment. SO002, SO008
CO009 Prof. Yaodong Yang, an assistant dean at Peking University's Institute for AI, is Psibot's chief scientist and heads the PKU-PsiBot Joint Lab. SO004, SO002
CO010 Co-founder and RL leader Yuanpei Chen was a visiting scholar at Stanford under Karen Liu and Fei-Fei Li and developed Psibot's Psi-C0 control model. SO004, SO002
CO011 Psibot markets itself as the embodied-AI company with the "highest density of scientists," anchored by a Peking University joint lab. SO002, SO006
CO012 Psibot's board composition, equity split, and formal governance structure are not disclosed in retained public sources. SO010, SO013
CO013 The investment thesis concentrates key-person dependence on a small group of named founders and academic scientists. SO006, SO002
CO014 On 23 July 2026 Bloomberg reported Psibot was close to finalizing a round of nearly US$100 million led by Chery Automobile with Lens Technology participating. SO003, SO004, SO011
CO015 The reported July 2026 round values Psibot at roughly US$1.48 billion post-money, a figure attributed to Bloomberg reporting rather than a company confirmation. SO004, SO005, SO011
CO016 Psibot closed an angel round in November 2024 led by GL Ventures and Lanchi Ventures. SO002, SO010
CO017 Psibot has raised about US$300 million in total since its 2024 inception. SO004, SO005
CO018 Psibot's combined angel and Pre-A rounds totalled 2 billion yuan (about US$280 million), disclosed around 10 March 2026. SO006, SO008, SO010
CO019 The Pre-A round was led by Shanghai state-owned Xuhui Capital with participation from Liangxi Sci-Tech, Xi Venture Capital, Pufeng Capital, and Timing Capital. SO008, SO010
CO020 State-backed "national team" investors in the angel round include China Development Bank Capital, Guozhong Capital, and the CCTV Media Convergence Industrial Investment Fund. SO007, SO008
CO021 Psibot is running pilots at a large Chinese logistics vendor and one of the world's largest fibre-optic cable makers. SO004, SO009
CO022 Psibot does not disclose revenue, run-rate, or gross margin in any retained public source as of the run date. SO013, SO010
CO023 Psibot aims to collect one million hours of manipulation data in 2026 and build China's largest dexterous-hand dataset. SO004, SO009
CO024 Psibot's own machine mass production and its data-collection cost claims are reported without disclosed shipment volumes, contracts, or customer counts. SO013
CO025 Lead backer Chery Automobile is a Wuhu-based automaker that made its Fortune Global 500 debut in 2024, and Lens Technology is a precision-glass and sensor supplier to Apple and Tesla. SO003, SO004
CO026 Psibot's thesis is that the defensible position in embodied AI is the software "brain" and proprietary data flywheel rather than the robot body. SO003, SO005
CO027 Psibot released its first end-to-end reinforcement-learning embodied model, Psi R0, on 30 December 2024, followed by Psi R0.5 in March 2025. SO001
CO028 Psibot began proof-of-concept engagements and signings with key clients in January 2025. SO001
CO029 Psibot launched its Psibot V1 and Psibot H1 hardware platforms in April 2025. SO001
CO030 Psibot released its flagship Psi R1 model in May 2025, demonstrating over 30 minutes of autonomous Mahjong play as an example of "L3" long-horizon dexterous manipulation. SO015, SO016, SO014
CO031 Psibot released the human-data-pretrained Psi-R2 and Psi-W0 models on 10 April 2026 and open-sourced 1,000 hours of multimodal hand-manipulation data. SO009
CO032 Chinese-language coverage reports Psibot announced own-machine mass production around April 2026 without disclosing shipment volumes. SO013
CO033 Psibot is best characterised as a private-undisclosed company whose scale must be inferred from capital and pilots rather than published operating metrics. SO010, SO013
CO034 Independent verification of Psibot's headline benchmark claims (30-minute Mahjong autonomy, roughly one-tenth data-collection cost) does not exist in retained public sources. SO003, SO013
CO035 Psibot's angel-round lead is attributed to GL Ventures and Lanchi Ventures by the company but to China Development Bank Capital by some trackers, an unresolved discrepancy. SO002, SO010
CO036 Psibot's platform-first, logistics-focused positioning contrasts with hardware-first humanoid rivals that emphasise dancing, kung-fu, and trade-show demos. SO006, SO007
CO037 Psibot's founders and scientist affiliations are corroborated across the company site and independent English- and Chinese-language coverage, but the "highest density of scientists" label is a company marketing claim. SO002, SO004, SO011
CO038 Psibot operates primarily out of Beijing and Shanghai, but a precise office footprint and headcount by location are not disclosed. SO004, SO010
CO039 Psibot's most recent disclosed milestones as of the run date are the April 2026 Psi-R2/Psi-W0 release and the July 2026 Chery-led funding report. SO009, SO003
CO040 No litigation, product recalls, or executive-churn events involving Psibot appear in retained public sources as of the run date. SO010, SO013
CM001 Psibot competes in the embodied-AI "robot brain" layer — VLA plus reinforcement-learning models and dexterous-manipulation software — which sits atop the robotics hardware stack rather than inside it. SM023, SM015, SM013
CM002 The status-quo substitutes Psibot's technology displaces are fixed industrial automation, AGV/AMR fleets, and manual labor, none of which generalize across novel objects without reprogramming. SM002, SM003, SM007
CM003 Adjacent markets bordering Psibot's core include humanoid-robot hardware, warehouse robotics, logistics robots, and industrial robotics, and its licensed model can span multiple hardware form factors. SM001, SM004, SM002
CM004 The global humanoid-robot market was valued at about US$7.9 billion in 2025 and is forecast to grow from US$10.9 billion in 2026 to US$54.2 billion by 2031 and US$192.7 billion by 2035, a 37.6% CAGR (Global Market Insights). SM001
CM005 Research and Markets and Global Market Insights both frame humanoid robotics as an early, exponentially growing market with 2035 estimates approaching US$190 billion. SM006, SM001
CM006 The global warehouse-automation market was about US$19.23 billion in 2023 and is projected to reach US$59.52 billion by 2030 at an 18.7% CAGR (Grand View Research). SM002
CM007 The global warehouse-robotics market was US$6.51 billion in 2025 and is projected to grow from US$7.35 billion in 2026 to US$25.41 billion by 2034 at a 16.8% CAGR, with Asia-Pacific holding 51.7% share in 2025 (Fortune Business Insights). SM003
CM008 The global logistics-robots market was US$17.8 billion in 2025 and is forecast to grow from US$20.7 billion in 2026 to US$91.4 billion by 2035 at a 17.9% CAGR (Global Market Insights). SM004
CM009 China's embodied-intelligence market rose from ¥213.3 billion in 2018 to ¥915 billion in 2025 and is expected to exceed ¥1 trillion in 2026 (36Kr Research Institute). SM007, SM025
CM010 Morgan Stanley forecasts China humanoid shipments of about 50,000 units in 2026 (up ~79% from 28,000) and 446,000 units by 2030, with the China market at about US$2 billion in 2026 rising to about US$15 billion by 2030. SM010, SM011
CM011 Global humanoid-robot shipments exceeded 18,000 units in 2025, a breakout year led by Chinese vendors. SM005, SM008, SM012
CM012 Chinese vendors accounted for roughly 90% of global humanoid-robot units shipped in 2025. SM005, SM008, SM012
CM013 IDC projects global humanoid shipments to exceed 510,000 units by 2030 at a ~95% CAGR, and reports that more than 85% of 2025 deployments were in performances, education, data collection, and guided-tour scenarios rather than production work. SM005
CM014 Psibot's buyers are robot OEMs and developers licensing the Psi brain plus end-user enterprises in logistics and manufacturing; the users are warehouse and line operators and the payers are enterprise capex and SOE procurement budgets. SM015, SM023, SM008, SM020
CM015 In China, embodied-AI budget ownership increasingly sits with state-owned enterprises and government procurement under directive industrial policy. SM008, SM013
CM016 The adoption path runs from lab demonstration to small-scale pilot to scenario validation to scaled multi-site deployment, with commercialization still concentrated at the pilot stage in 2026. SM005, SM013, SM015
CM017 Chinese industrial policy is a primary demand driver — the 15th Five-Year Plan (2026-2030) elevates embodied intelligence to a national strategic pillar with state investment, procurement mandates, and MIIT/SASAC deployment targets. SM008, SM007
CM018 Demographic pressure — population aging and structural labor shortages — is a structural growth driver for embodied AI in China. SM007, SM008
CM019 Supply-chain localization lets Chinese whole-machine costs sit near 50% of comparable overseas products, improving adoption economics. SM007, SM008
CM020 China embodied-intelligence financing reached ¥33.5 billion in the first eleven months of 2025 (about four times the year-earlier level), exceeding ¥38 billion across more than 305 deals for the full year. SM007
CM021 MERICS assesses that Chinese humanoids still lack precision and dexterity, run costly site-specific trials, must cut costs by at least half, and still depend on US research for VLA breakthroughs. SM013
CM022 Independent analysis argues the humanoid market is in a speculative phase, with truly productive industrial revenue only about 3-5% of humanoid sales and the rest from research, showrooms, and novelty buyers. SM009
CM023 2030 humanoid-market forecasts diverge roughly four-fold (about US$4 billion to over US$15 billion), a divergence independent analysis reads as a marker of speculative heat. SM009
CM024 High unit costs, reliability limits, and safety and insurance barriers around dexterous manipulation keep most humanoid deployments in controlled pilots rather than at transformative scale. SM009, SM005
CM025 TrendForce projects China humanoid output up about 94% in 2026, with Unitree and AgiBot together about 80% of shipments and China about 84.7% of global humanoid shipments. SM012
CM026 National standards for embodied intelligence began releasing in early 2026, with MIIT designating 2025 China's "first year of humanoid mass production" and 140+ manufacturers releasing 330+ models. SM008, SM007
CM027 Excluded from Psibot's serviceable market is pure hardware manufacturing, AGV/AMR fleets, and fixed programmable automation, which Psibot outsources or displaces rather than sells. SM023, SM002
CM028 The e-commerce segment is projected to hold about 47% of the warehouse-robotics market in 2026, indicating logistics and e-commerce as the leading buyer vertical (Fortune Business Insights). SM003
CM029 Wheeled humanoid form factors carry the highest near-term growth (IDC projected CAGR ~120%), favoring the semi-structured indoor logistics settings Psibot targets. SM005
CM030 Precision components — high-precision ball screws, gears, and advanced sensors — remain partially import-dependent, a supply constraint Chinese policy explicitly targets. SM008
CM031 No public source isolates a Psibot-specific SAM or SOM; its serviceable market can only be inferred as a licensing-and-data slice of the humanoid, warehouse-robotics, and logistics-robots TAMs. SM001, SM004
CM032 Market definitions vary by publisher (humanoid robots vs warehouse robotics vs logistics robots vs embodied intelligence), so headline TAMs overlap and are not directly additive. SM001, SM002, SM003, SM004
CM033 Psibot's own commercialization signals name a large Chinese logistics client (warehouse sorting) and a leading fibre-optic cable maker as pilot users, but neither is a confirmed paying customer at scale. SM015, SM023
CM034 China's H1-2026 embodied-AI funding surge and wave of new unicorns evidence strong capital-driven demand momentum entering the run-date window. SM007, SM014, SM018, SM019, SM022, SM024
CM035 Adverse commentary questions whether reported unicorn valuations across the sector, including newly minted ones, outrun grounded deployment and revenue. SM021, SM009
CM036 Warehouse-automation penetration remains low globally (only a minority of warehouses are highly automated), implying large headroom but also unproven near-term conversion. SM002, SM003
CM037 The buyer-user-payer split differs by channel — for licensed OEMs the payer is the robot maker, while for direct enterprise pilots the payer is the end operator's capex or SOE procurement. SM015, SM008
CM038 Psibot's platform and licensing position means its addressable market scales with the whole downstream robot install base rather than with its own unit shipments. SM023, SM016, SM017
CM039 China is positioned as one of the fastest-growing embodied-intelligence markets globally, supported by vertically integrated supply chains and independent model iteration (36Kr). SM007
CM040 Global humanoid TAM estimates for 2026 cluster in the US$6-11 billion range across independent researchers, materially smaller than long-run 2035 projections. SM001, SM006
CP001 Psibot competes in the embodied-AI brain layer against Chinese hardware-scale humanoid makers, US robot-foundation-model labs, and vertically integrated full-stack players, plus status-quo automation and internal build. SP016, SP005, SP019
CP002 AgiBot (Zhiyuan), founded in 2023 and BYD-backed, produced its 10,000th humanoid on 30 March 2026, accelerating from 5,000 to 10,000 units in about three months, and shipped roughly 5,168 units in 2025 (Omdia number one). SP001, SP003
CP003 Unitree cleared its STAR Market IPO listing-committee review on 1 June 2026 in 73 days, targeting an approximately US$6.2 billion valuation and raising about ¥4.2 billion (~US$583 million), with its H1 humanoid surpassing 11,000 cumulative units. SP001, SP002, SP011
CP004 Unitree's 2025 revenue reached ¥1.699 billion (~US$240 million, up about 335% year-over-year) at roughly 60% gross margin, with humanoids more than half of revenue, though Q1 2026 growth decelerated and profit nearly halved. SP001, SP002
CP005 Unitree's prospectus shows humanoid revenue was 73.6% research and education, 17.4% commercial, and only about 9% industrial from 2023 through Q3 2025, with genuine production-line revenue only ~¥15.7 million (US$2.2 million) in the first nine months of 2025. SP002, SP011
CP006 UBTech opened orders for its full-size UWORLD U1 line on 30 June 2026, passing 13,361 cumulative orders on day one, priced from ¥119,800, after shipping 1,079 full-size units for ¥821 million of humanoid revenue in 2025. SP006, SP003
CP007 Chinese vendors led by Unitree and AgiBot account for roughly 80-90% of global humanoid shipments, shipping more units than all Western competitors combined. SP024, SP012, SP001
CP008 US-based Physical Intelligence, founded in 2024, builds general-purpose robot foundation models (π0, π0.7) and is reportedly raising about US$1 billion at a valuation north of US$11 billion, up from US$5.6 billion four months earlier, backed by Founders Fund, Lightspeed, Jeff Bezos, and Nvidia. SP007, SP008
CP009 Figure AI, founded in 2022, raised a Series C in September 2025 at a US$39 billion valuation (about US$1.9 billion raised in total), runs its in-house Helix brain, is ramping Figure 03, and is backed by an OpenAI-led US$6.7 billion investment. SP004, SP005, SP009
CP010 Figure AI's US$39 billion valuation rests on essentially no revenue (low single-digit millions), and the company faces a whistleblower safety lawsuit and lost BMW's European expansion to another vendor. SP004
CP011 Tesla had deployed about 1,000 Optimus Gen 3 units (28 degrees of freedom) at Giga Texas by June 2026, targeting 5,000 internal units by year-end with external sales from Q1 2027, funded via corporate R&D of roughly US$2-3 billion cumulatively. SP005
CP012 Nvidia's GR00T is a robot foundation model and Isaac platform; on 1 June 2026 Nvidia named Unitree's H2 Plus body as the hardware foundation for its open GR00T Reference Humanoid, using Jetson Thor compute. SP001, SP005
CP013 Galaxy General's Galbot is part of the Chinese embodied-AI cohort with a simulation-first data approach but has drawn less headline funding than the hardware-scale leaders. SP018, SP003
CP014 Psibot's differentiation is a brain-and-data platform ("small full-stack") that licenses its Psi VLA-plus-RL model rather than manufacturing at scale, positioning it closer to Physical Intelligence and Nvidia GR00T than to hardware-scale Chinese peers. SP016, SP007, SP017
CP015 On raw manufacturing scale Psibot trails AgiBot, Unitree, and UBTech by orders of magnitude — thousands to more than ten thousand units versus Psibot's undisclosed pilot volumes — competing instead on model capability and data. SP001, SP006, SP016
CP016 Psibot's hierarchical fast/slow VLA architecture is echoed by Physical Intelligence (π0 to Hi Robot), Figure (Helix), Google Gemini Robotics, and Nvidia GR00T, so its core technical approach is not unique. SP007, SP019, SP017
CP017 Chinese hardware peers publish unit prices (UBTech U1 from ¥119,800; Unitree G1 ~US$16,000; AgiBot models from ~US$5,000) and robot-as-a-service leasing, whereas Psibot's model-licensing and data pricing is undisclosed. SP003, SP006
CP018 Hardware peers reach buyers via factory pilots (BMW, BYD, NIO), retail experience stores, and public listings, giving them distribution and capital advantages that Psibot lacks as a private brain-licensor. SP003, SP002
CP019 Analysts note Chinese humanoid makers carry structural legal and regulatory risks not disclosed in filings, and a Morgan Stanley survey found only 23% of prospective industrial buyers satisfied with current products. SP001, SP013
CP020 As a brain licensor Psibot faces low switching costs because OEMs can adopt Nvidia GR00T, license Physical Intelligence, or build in-house, unless its proprietary dexterous-hand dataset creates genuine data lock-in. SP007, SP012, SP005
CP021 Robot OEMs can multi-home across multiple foundation-model providers (GR00T, Physical Intelligence, in-house), pressuring any single brain vendor's pricing power. SP005, SP001
CP022 Vertically integrated players (Tesla, Figure, Unitree, AgiBot) build their own AI brains in-house and Nvidia offers an open GR00T stack, so the licensable-brain thesis competes against both free/open and internal alternatives. SP005, SP001, SP004
CP023 Psibot's strategic backers (Chery, Lens Technology) offer manufacturing and sensor supply access, a partner advantage relative to pure-software labs. SP014, SP015
CP024 Psibot's most defensible moat claim is its proprietary human-hand manipulation dataset and data-collection engine, aiming to build China's largest dexterous-hand dataset. SP016, SP017
CP025 Psibot's data moat is unproven and time-limited because rivals also scale data (Figure real-world video, Physical Intelligence RL tokens, Nvidia simulation), so commoditization risk is high if open models close the gap. SP004, SP007, SP009
CP026 Across the sector, 2025 humanoid deployments were overwhelmingly non-productive (research, education, demonstrations, guided tours), with genuine industrial revenue only a small single-digit share. SP024, SP002, SP011
CP027 China's 15th Five-Year Plan funds both "big brain" and "small brain" tiers, intensifying domestic competition and inviting well-funded entrants into the layer Psibot targets. SP019, SP018
CP028 Likely entrants threatening the brain layer include global labs — OpenAI's 200-plus-researcher Project Atlas and Google Gemini Robotics — and cash-rich incumbents, raising competitive intensity. SP005
CP029 Other well-funded players include Apptronik (~US$5.5 billion valuation, US$935 million Series A), Boston Dynamics, 1X, and Agility, broadening the humanoid field. SP005
CP030 Psibot's ~US$1.48 billion valuation is an order of magnitude below Figure (US$39B) and Physical Intelligence (>US$11B) and below Unitree's ~US$6B IPO target, positioning it as a mid-tier challenger. SP004, SP008, SP011, SP015, SP021
CP031 Chinese humanoid startups have collectively raced past ¥100 billion in valuation, with several unicorns each above roughly ¥10 billion (~US$1.4 billion). SP010, SP018, SP022
CP032 Many competitors build on Nvidia Jetson and Isaac, and Nvidia's own GR00T reference platform makes it simultaneously the compute supplier and a brain competitor to vendors like Psibot. SP001, SP005
CP033 Unlike hardware-scale peers, Psibot demonstrates capability through dexterity benchmarks (a 30-plus-minute autonomous Mahjong session) rather than shipment counts. SP017, SP016
CP034 The status-quo alternative — fixed automation, AGVs, and human labor — remains the default for most warehouses on cost and reliability, a competitive baseline Psibot must beat. SP013, SP024
CP035 A Morgan Stanley buyer survey (23% satisfaction) and Unitree's decelerating Q1 2026 growth signal that supply is outpacing validated demand across the humanoid sector. SP001, SP002, SP020
CP036 OpenAI's US$6.7 billion stake in Figure and its internal Project Atlas robot foundation model make the leading AI lab a direct competitor to independent brain vendors like Psibot. SP005
CP037 Psibot's differentiation durability hinges on data-flywheel lock-in and Chinese-market access via strategic backers, both unproven at scale. SP016, SP023, SP014
CP038 Physical Intelligence's π0.7 adds an RL Token and Multi-Scale Embodied Memory enabling tasks longer than ten minutes, a direct capability benchmark against Psibot's Psi R1. SP007
CP039 UBTech's U1 claims 88 degrees of freedom and sub-20-millisecond speech-to-lip synchronisation, illustrating hardware peers' spec-led competition. SP006
CP040 The competitive question is whether value accrues to the AI brain (Psibot, Physical Intelligence, Nvidia) or the integrated body (Tesla, Figure, Unitree, AgiBot); as of 2026 capital favors both but revenue favors hardware. SP004, SP011, SP005
CI001 Psibot monetizes an intelligence layer, licensing its Psi-series VLA model to third-party robot developers rather than manufacturing and selling robots itself. SI004, SI001, SI022
CI002 Psibot sells a proprietary human-hand data-collection system (Psi-SynEngine) using 16-DOF exoskeleton gloves with sub-millimetre 3D trajectory precision and fingertip tactile sensing. SI004, SI021
CI003 Psibot offers simulation and training-data platforms to robot developers as additional monetization lines alongside model licensing. SI022, SI024
CI004 Psibot discloses no licensing price, take-rate, licensed-unit count, or recognized-revenue figure for any of its monetization streams. SI004, SI022, SI017
CI005 Hardware peers publish transparent unit prices (UBTech U1 from ¥119,800; Unitree G1 from ¥85,000/~US$12,000 and R1 Air from ¥29,900/~US$4,300), against which Psibot's undisclosed licensing pricing cannot be triangulated. SI011, SI010
CI006 Psibot's public deployment evidence is a small-scale warehouse validation at a large Chinese logistics client (reporting sorting-efficiency gains) and testing at one of the world's largest fibre-optic cable makers. SI001, SI024
CI007 Psibot's July 2026 lead investors act as potential channels — Chery Automobile (a Fortune Global 500 automaker with an AiMOGA robotics subsidiary) and Lens Technology (an Apple/Tesla components supplier building an embodied-intelligence centre). SI001, SI008
CI008 Psibot discloses no sales cycle, customer-acquisition cost, payback period, pipeline coverage, or channel economics, and none can be reliably proxied without licensing-price data. SI004, SI022
CI009 Because early demand is strategic-investor-linked, Psibot's initial revenue may be related-party or pilot-stage rather than arm's-length commercial demand, a distinction material to revenue quality. SI007, SI001
CI010 As a manufacturing-outsourced licensor, Psibot should carry software-like incremental gross margins with heaviest costs in R&D talent, compute, and data acquisition rather than bill-of-materials. SI004, SI021
CI011 Psibot claims its exoskeleton-glove data-collection approach costs roughly one-tenth of traditional teleoperation, an unverified but structurally important assertion for the data-flywheel model. SI021, SI004
CI012 Unitree's IPO prospectus reports a ~60% gross margin (60.27% in 2025, up from 44.22% in 2022) achieved through in-house production of motors and actuators, a hardware-peer reference point for the margin discussion. SI010, SI013, SI011
CI013 Psibot's capital-intensity risk inverts the usual robotics concern — not factory capex but whether licensing revenue can cover a large, front-loaded model-and-data cost base before capital runs down. SI008, SI025
CI014 Psibot discloses no revenue, ARR, gross margin, monthly burn, runway, headcount, licensed-unit count, or customer-concentration figures. SI017, SI022, SI004
CI015 Despite financial gaps, Psibot shows concrete non-financial traction — a ~US$100M round at US$1.48B, ~US$300M raised since 2024, marquee backers, named pilots, a fast product cadence, and a PKU joint lab. SI002, SI001, SI024
CI016 Unitree's prospectus shows only ~9% of humanoid revenue from genuine industrial deployment, 74% from research and education and 17% from commercial display, warning against reading demos as demand. SI013, SI012
CI017 A Morgan Stanley survey found only 23% of prospective industrial buyers satisfied with current humanoid products, and more than 85% of 2025 deployments were non-productive per IDC. SI018, SI015
CI018 Psibot has raised roughly US$300 million since its September 2024 founding, culminating in a near-final ~US$100 million July 2026 round that set a US$1.48 billion valuation. SI002, SI001, SI016
CI019 Psibot's angel and Pre-A rounds totalled about ¥2 billion (~US$280 million), announced 10 March 2026 during China's Two Sessions. SI007, SI005, SI006
CI020 With a fresh nine-figure round and no manufacturing capex burden, Psibot can plausibly fund two to three years of R&D, compute, and data scaling, though exact cash, burn, and runway are undisclosed. SI002, SI004
CI021 Psibot reframes data as a monetizable asset, targeting China's largest dexterous-hand dataset and one million hours of manipulation data collected in 2026. SI004, SI021
CI022 Psibot's go-to-market is a small number of deep, strategic engagements rather than a volume sales funnel, consistent with an early-stage platform licensor. SI001, SI022
CI023 Lens Technology is building an embodied-intelligence centre targeting output of about 3,000 humanoids and over 10,000 robot-dogs a year, signalling manufacturing pull that could channel Psibot's models. SI009, SI008
CI024 Lens Technology committed more than RMB 20 billion to R&D since IPO, including RMB 2.44 billion in the first nine months of 2025, illustrating how capital-intensive the surrounding embodied-AI hardware ecosystem is. SI008, SI009
CI025 Unitree's 2025 revenue of ¥1.708 billion (up 335.36%) with net profit up 674% establishes that embodied-AI hardware can be profitable at scale, setting a demanding comparable for Psibot's unproven model. SI010, SI011
CI026 Psibot's technical footprint expanded from Psi R0 to Psi-R2/W0 within about eighteen months, and it open-sourced the first 1,000 hours of a stated 100,000-hour data reserve. SI021, SI024
CI027 Genuine production-line humanoid revenue for the comparable was only about ¥15.7 million (US$2.2 million) in the first nine months of 2025, underscoring that Psibot's unquantified pilots need realized-revenue proof. SI012, SI013
CI028 There is no public evidence that Psibot carries debt or project-finance obligations, which is appropriate for an asset-light intelligence-layer licensor. SI022, SI004
CI029 Psibot's next-round trigger is effectively a capability-and-commercialization milestone — converting its data lead and pilots into demonstrable licensing revenue before the funding environment cools. SI017, SI020
CI030 The 2026 environment is favourable for now — 22-plus embodied-AI unicorns minted and US$13.8 billion of Chinese embodied-AI funding in H1 2026 — but reliance on continued capital access is a financing risk if sentiment turns. SI017, SI025, SI020
CI031 The verdict on revenue quality is "unproven, not absent": a credible differentiated model and named pilots, but zero disclosed realized revenue and no evidence of arm's-length recurring contracts. SI017, SI001, SI022
CI032 The margin path is unmodellable on public data; the honest reference is Unitree's ~60% hardware gross margin alongside its Q1 2026 profit halving under R&D load. SI010, SI014
CI033 Unitree's adjusted Q1 2026 net profit fell about 52% year-on-year despite 68% revenue growth, showing how quickly embodied-AI economics can compress even for a profitable leader. SI014, SI015
CI034 Psibot's single largest diligence blocker is the complete absence of realized-revenue and contract disclosure, without which revenue quality, ARR, and unit economics cannot be underwritten. SI017, SI022
CI035 Secondary blockers are the undisclosed cash/burn/runway triad and unquantified customer concentration/related-party exposure given strategic-investor-linked demand. SI004, SI007
CE001 Psibot's core product is the Psi-series vision-language-action (VLA) model — an "intelligence layer" or robot "brain" — sold to third-party robot makers via licensing rather than as Psibot-branded hardware. SE013, SE001, SE017
CE002 In customer-workflow terms, Psibot's software converts a natural-language task plus camera and sensor input into a sequence of dexterous, multi-step robot actions for jobs such as warehouse sorting, packaging, and restocking. SE009, SE014
CE003 Psibot follows a "small full-stack" model — it defines robot design parameters (structure, range of motion, degrees of freedom) but outsources component production and manufacturing, keeping itself asset-light. SE013, SE011
CE004 Psibot's flagship Psi R1 demonstrated L3-level autonomous dexterous manipulation, playing Mahjong with humans autonomously for more than thirty minutes in an open environment. SE009, SE001, SE015
CE005 Psibot's product line advanced rapidly — Psi R0 (30 Dec 2024, first end-to-end RL embodied model), R0.5 (Mar 2025), Psibot V1/H1 hardware platforms (Apr 2025), Psi R1 (May 2025), and Psi-R2 plus Psi-W0 (10 Apr 2026). SE014, SE010, SE002
CE006 The Psi stack includes sub-models Psi-P0 for planning and Psi-C0 for control, the latter developed by co-founder and RL leader Yuanpei Chen. SE013, SE011
CE007 Psi-SynEngine is Psibot's proprietary data-acquisition system using 16-degree-of-freedom exoskeleton gloves with sub-millimetre 3D trajectory precision and 3D fingertip tactile and force-feedback sensing. SE013, SE010
CE008 Psibot's dexterous robot hand is reported at 21 degrees of freedom, among the more articulated hands in the sector. SE017
CE009 Psi-R2 is a world-action model trained on 95,472 hours of human data covering 294 scenarios and 4,821 tasks, and Psi-W0 is an action-conditioned world model enabling counterfactual reasoning and a reinforcement-learning flywheel. SE002, SE016
CE010 Psibot open-sourced the first 1,000 hours of a stated 100,000-hour multimodal human-hand manipulation data reserve. SE014, SE010
CE011 Psi R1 uses a hierarchical, end-to-end architecture combining a VLA model with reinforcement learning, linking high-level planning to low-level dexterous control in a single trained system. SE009, SE001
CE012 Psibot's architecture uses a "Chain of Action Thought" (CoAT) reasoning mechanism to decompose and execute long-horizon manipulation in open environments — the capability it defines as L3. SE001, SE009
CE013 Psibot's design uses an "Action Tokenizer" to bridge planning and control and a fast/slow ("S1 fast / S2 slow") dual-system brain separating reactive control from deliberative planning. SE009, SE002
CE014 DexGraspVLA, from the PKU-PsiBot Joint Lab, is a hierarchical framework using a pre-trained vision-language model as high-level planner and a diffusion-based low-level controller. SE005, SE006
CE015 DexGraspVLA reports a 90%+ zero-shot dexterous-grasping success rate across thousands of unseen cluttered scenes and was accepted as an AAAI 2026 Oral presentation. SE005, SE006, SE003
CE016 Reporting on the Psi-R2/W0 dual-model design credits it with cutting inference latency from 2.2 seconds to under 100 milliseconds. SE002
CE017 Psibot frames its technical goal as an "Impossible Triangle" of high generalization, high dexterity, and high success rate, citing 95% validation and 99.9% scaled-deployment success targets. SE012, SE009
CE018 The dual-model approach pre-trains on abundant human data before fine-tuning on minimal real-robot data, directly addressing embodied AI's core data-scarcity dilemma. SE002, SE013
CE019 Psibot's technology go-to-market is licensing the Psi model to robot OEMs plus selling the Psi-SynEngine data system and simulation and training-data services, integrating its brain into partner hardware. SE013, SE017
CE020 Psibot's public deployment evidence is a small-scale warehouse sorting validation at a large Chinese logistics client and testing at one of the world's largest fibre-optic cable makers. SE017, SE013
CE021 Psibot operates a PKU-PsiBot Joint Lab for Embodied Dexterous Manipulation that produces its published research and validates its models. SE006, SE013
CE022 Psibot's stated 2026 operating goals are to build China's largest dexterous-hand dataset and to collect one million hours of manipulation data during the year. SE013, SE010
CE023 Psibot's cadence of five model generations plus hardware in roughly eighteen months is unusually fast, but no independent field-reliability, uptime, or support-SLA data is public. SE014, SE019
CE024 Psi-R2 and Psi-W0 (Apr 2026) are Psibot's most recent releases, and the roadmap points toward scaling the data reserve and the RL flywheel rather than shipping proprietary hardware. SE002, SE016
CE025 Psibot's differentiation rests on a data flywheel — proprietary low-cost human-hand data collection feeding a VLA model licensed across many OEM bodies, an "Android for robots" thesis. SE013, SE017, SE001
CE026 Psibot claims its exoskeleton-glove data collection costs roughly one-tenth of traditional teleoperation, an unverified but structurally central cost claim. SE013
CE027 Psibot's published IP footprint includes the DexGraspVLA framework, an Awesome-VLA-Papers survey list, and Psi SynHand models released through its Psi-Robot GitHub organization. SE004, SE003
CE028 OpenVLA, an open-source 7B VLA model that outperformed the 55B closed RT-2-X, is a public technical benchmark showing Psibot competes against strong open baselines, not a unique paradigm. SE008
CE029 Psibot's stack depends on Nvidia compute (Jetson edge modules and Isaac simulation), shared across Chinese peers and exposed to US export controls. SE020, SE019
CE030 Precision components (special ball screws) are roughly 90% supplied by Schaeffler (Germany) and THK/NSK (Japan), a supply-chain dependency for any hardware embodying Psibot's model. SE019, SE023
CE031 Global labs such as Physical Intelligence (pi0) pursue similar hierarchical VLA approaches, so Psibot's architecture is differentiated by its data and dexterity focus rather than by a unique paradigm. SE024, SE008
CE032 Psibot's headline capability metrics — 30-minute autonomous Mahjong, 100% tile-flip accuracy, and 95%/99.9% success figures — are company-reported and not independently benchmarked. SE009, SE001
CE033 Independent analysts (MERICS) assess that Chinese humanoids broadly lack precision and dexterity, rely on site-specific trials, remain too expensive, and still look to US research for VLA breakthroughs. SE019
CE034 The only externally verifiable Psibot capability evidence is the peer-reviewed DexGraspVLA result (AAAI 2026 Oral, 90%+ grasping); other flagship claims rest on company demonstrations. SE005, SE006
CE035 No public evidence shows Psibot safety certifications, functional-safety standards, or data-privacy and security controls governing its data-collection and model-licensing operations. SE013, SE019
CE036 China's Cybersecurity Law amendments (effective 1 January 2026) and data-security regime impose penalties up to RMB 10 million with extraterritorial reach — directly relevant to a company amassing 100,000-plus hours of human-manipulation data. SE020, SE019
CE037 Psibot's technology maturity is uneven — research and demo capability is strong (peer-reviewed, fast cadence, latency step-change) while productization signals (field reliability, certifications, disclosed benchmarks) are weak or absent. SE019, SE014, SE021
CU001 Psibot's primary customer segment is robot OEMs and developers who license the Psi VLA model, with the OEM typically the buyer/payer and the end-enterprise the ultimate user. SU004, SU001
CU002 Psibot's end-use verticals are logistics/warehousing (barcode sorting, high-SKU clothing distribution) and manufacturing (fibre-optic cable handling). SU001, SU002, SU008
CU003 Psibot's initial real-world deployment is barcode-sorting in logistics warehouses, chosen because high SKU diversity and frequent task changes demand the generalization a VLA model provides. SU008, SU001
CU004 Psibot deployed roughly 100 data-collection/manipulation units in Beijing by early 2026 to gather distributed real-world manipulation data in logistics settings. SU008, SU015
CU005 Psibot's customer base is concentrated in China, with logistics, retail, and smart-manufacturing warehousing as the target segments. SU001, SU002
CU006 Psibot's adoption is pilot-stage — a small-scale warehouse sorting validation at a large Chinese logistics client reporting efficiency gains, with no disclosed throughput or account counts. SU001, SU003
CU007 Psibot is testing its models at one of the world's largest fibre-optic cable makers. SU001, SU005
CU008 Psibot discloses no active-account count, deployment count, utilization rate, or repeat-purchase data. SU004, SU001
CU009 Psibot's data-collection deployment doubles as adoption — warehouse workers wear Psi-SynEngine data gloves during real operations, blending customer deployment with dataset building. SU008, SU004
CU010 No named production customer is publicly disclosed; Psibot's clients are described generically ("a large logistics client", "a fibre-optic maker") or are strategic investors. SU001, SU008
CU011 Psibot's most concrete customer proof is pilot/testing-stage, not production deployment with disclosed outcomes. SU001, SU019
CU012 Chery Automobile is a strategic backer and potential OEM channel via its AiMOGA Robotics subsidiary, which unveiled the humanoid Mornine and signed a 1,000-unit intelligent-police-robot deal in April 2026. SU001, SU002, SU009
CU013 Lens Technology is a strategic backer building an embodied-intelligence centre, a potential manufacturing and demand channel for Psibot's model. SU001, SU002
CU014 Psibot discloses no retention metrics — no net revenue retention, gross retention, churn, renewal rate, or contract length. SU004, SU001
CU015 No cohort or repeat-usage data exists for Psibot; founded in 2024 and pilot-stage, its customer durability is unproven. SU002, SU019
CU016 Psibot's investor base includes state-linked funds (GL Ventures, Lanchi Ventures, SDIC Advanced Manufacturing Fund, Jingxi Ruiling), reflecting strategic and government-aligned demand pull. SU008, SU023
CU017 Psibot's intended expansion is land-and-expand via OEM licensing plus its data flywheel, but there is no disclosed evidence of account expansion or upsell yet. SU004, SU001
CU018 Customer concentration and related-party risk is elevated because early demand is linked to strategic investors (Chery, Lens) rather than demonstrably arm's-length buyers. SU001, SU018
CU019 No customer-satisfaction score, NPS, or reference-quality metric is disclosed for Psibot. SU004
CU020 IDC reports more than 85% of 2025 humanoid deployments were non-productive, warning that deployment does not equal sustained productive customer usage. SU019
CU021 A Morgan Stanley survey found only 23% of prospective industrial buyers satisfied with current humanoid products. SU020
CU022 MERICS assesses that Chinese embodied-AI deployments rely on site-specific trials and remain too expensive, limiting broad arm's-length customer adoption. SU018
CU023 Morgan Stanley doubled its 2026 China humanoid shipment forecast to 50,000 units, signalling pilot-to-production momentum that could benefit Psibot's OEM licensees. SU020, SU012
CU024 Psibot's platform-licensing model makes its addressable customers the many robot OEMs across China's scaling humanoid market, through which large logistics buyers become reachable. SU012, SU001
CU025 Psibot's channel dependence on strategic investors is double-edged — it shortens sales cycles and de-risks manufacturing but concentrates demand and raises related-party questions. SU001, SU010
CU026 Psibot's stated 2026 goal to collect one million hours of data implies substantially scaling its data-collection deployment footprint, its de-facto operator base. SU004, SU008
CU027 The Chery/AiMOGA ecosystem operates across 30-plus countries, offering Psibot a potential international OEM-adoption path if ecosystem partners adopt its model. SU010, SU009
CU028 Psibot's customer-proof quality is low on the diligence scale — unnamed clients, pilot-stage, unquantified outcomes — despite strong investor validation. SU001, SU019
CU029 Procurement friction for Psibot's licensing model is unclear; no disclosed sales cycle, contract length, or pricing exists to assess buyer procurement dynamics. SU004
CU030 Psibot's Beijing data-collection units represent an operator/worker base building the dataset, not paying enterprise customers, so they are not revenue proof. SU008
CU031 AiMOGA's 1,000-unit intelligent-police-robot signing is AiMOGA/Chery's own commercialization and is not confirmed to run Psibot's Psi model, so it is ecosystem context rather than direct Psibot customer proof. SU009, SU010
CU032 Psibot's repeat-purchase and renewal behaviour cannot be assessed because it discloses no contracts and is too young to show renewal cycles. SU002, SU004
CU033 The strongest durable customer asset is Psibot's proprietary dataset and OEM relationships, which create switching costs if a licensee builds its product on the Psi model. SU004, SU016
CU034 Psibot's named strategic investors (Chery, Lens, GL Ventures, Lanchi, SDIC, Jingxi Ruiling) constitute high-quality validation even though named production customers are absent. SU001, SU002, SU023
CU035 Customer geography is concentrated in China, exposing adoption to domestic state-owned-enterprise adoption mandates while constraining international expansion. SU002, SU018
CU036 Psibot's 2026 adoption trajectory is early — a handful of pilots and a data-collection footprint — with no disclosed conversion to multi-site production accounts. SU001, SU008
CU037 The customer verdict is investor-validated but commercially unproven — strong strategic backing and named pilots, but no named production customers, no retention data, and elevated related-party concentration risk. SU001, SU019, SU018
CR001 As a Chinese company, Psibot is subject to the National Intelligence Law (2017), whose Article 7 obligates organizations to support, assist, and cooperate with national intelligence work. SR003, SR004
CR002 Independent legal analysis finds Article 7 narrower and less novel than often portrayed and lacking a dedicated enforcement mechanism, while conceding a Chinese firm could not meaningfully resist a lawful state security request and courts offer no reliable remedy. SR003
CR003 China's amended Cybersecurity Law took effect on 1 January 2026, raising the general administrative-fine cap tenfold from RMB 1 million to RMB 10 million. SR001, SR006
CR004 The amended Cybersecurity Law expressly integrates AI ethics, risk-monitoring, and safety-oversight obligations, signalling AI governance will be treated as a core dimension of data and cyber compliance pending comprehensive AI legislation. SR001, SR002
CR005 The amended Cybersecurity Law broadens extraterritorial reach to any overseas activity that endangers China's cybersecurity, heightening exposure for cross-border operations. SR001
CR006 China's Data Security Law and PIPL impose data classification and a cross-border-transfer regime requiring a CAC security assessment, standard contractual clauses, or certification, with biennial audits for large-scale personal-data handlers. SR005, SR006
CR007 Psibot's model depends on accumulating and moving large volumes of manipulation data (targeting one million hours in 2026), which amplifies its exposure to China's data-classification, cross-border-transfer, and cybersecurity obligations. SR019, SR006
CR008 A US Department of Homeland Security business advisory formalizes the Western regulatory view that data handled by firms linked to the PRC carries compelled-access risk under Chinese law. SR004
CR009 US export-control scrutiny and foreign-listing/procurement restrictions — the same national-intelligence/spy-law concerns that dogged peer Unitree — constrain Psibot's ability to sell to or raise capital from foreign counterparties. SR017, SR008
CR010 Psibot discloses no litigation, enforcement actions, IP disputes, or the specific licenses and approvals it must hold, leaving its regulatory-compliance posture unverified. SR019, SR013
CR011 More than 85% of 2025 humanoid deployments were non-productive per IDC, indicating that deployment does not equal sustained productive usage and that Psibot's pilot reliability is unproven. SR015, SR014
CR012 MERICS judges Chinese embodied-AI deployments to be site-specific, dependent on hand-tuning, and too expensive, with costs needing to fall by half or more before broad adoption. SR014
CR013 The broader embodied-AI field is criticized for controlled, choreographed demonstrations that do not translate into reliable, long-duration, wide-variation real-world deployment. SR011, SR012
CR014 Psibot's marquee capability claims — 30-plus minutes of autonomous Mahjong play and training-data cost as low as one-tenth of peers — are company-reported and have not been independently benchmarked. SR020, SR011
CR015 Psibot discloses no product-safety record, incident history, or product-liability insurance coverage for its deployed systems. SR019
CR016 Psibot's model depends on human-glove data collection at scale, creating data-quality and dataset-integrity risk if collected data is noisy, biased, or non-representative of production tasks. SR019, SR007
CR017 Because Psibot licenses an intelligence layer rather than making robots, real-world reliability depends on how well OEM partners integrate the Psi model, a coordination surface Psibot only partly controls. SR019, SR028
CR018 Psibot designs structure, motion range, and degrees of freedom but outsources component production and manufacturing, so end-product quality is exposed to third-party hardware and integration. SR019, SR016
CR019 Chinese embodied-AI developers broadly rely on Nvidia's Jetson edge modules and Isaac simulation/training tooling, a compute dependency shared across the sector including Psibot's stack. SR016, SR008
CR020 Nvidia compute access is exposed to US export-control policy, a supply and capability tail-risk that could raise costs or degrade training for Chinese robotics firms if access tightens. SR008, SR009
CR021 Foreign suppliers — Schaeffler of Germany, THK and NSK of Japan — dominate the special ball screws and high-end harmonic reducers dexterous systems require, supplying roughly 90% of certain precision components. SR007, SR009
CR022 Precision reducers and harmonic drives remain a technology bottleneck where foreign suppliers set the standard, though Chinese local capacity is rising and closing the gap. SR007, SR009
CR023 China controls an estimated 63–70% of the global humanoid supply chain and is rapidly localizing motors, actuators, and batteries, which cushions mechanical-component dependency and gives Psibot's licensees cost and logistics advantages. SR009, SR007
CR024 Psibot's early demand and financing are concentrated in strategic investors (Chery, Lens Technology) and state-linked funds rather than diversified arm's-length customers, raising related-party concentration risk. SR016, SR013
CR025 Psibot's revenue path runs through OEM licensees rather than diversified end-customers, concentrating commercial dependence on a small set of robot developers. SR028, SR019
CR026 Psibot depends on strategic and state-linked capital providers (GL Ventures, Lanchi Ventures, and state funds), tying its financing to a concentrated backer set and to Chinese policy priorities. SR024, SR013
CR027 Single-country concentration in China ties Psibot's adoption to domestic policy cycles and state-owned-enterprise procurement while limiting international diversification. SR013, SR014
CR028 The concentration of Psibot's early demand in related parties raises the question of whether its traction is genuinely commercial or strategically sponsored. SR016, SR024
CR029 Psibot discloses no revenue, ARR, gross margin, burn rate, or runway, so its roughly $1.48B valuation is underwritten by team, data, and backers rather than by fundamentals. SR013, SR023
CR030 Embodied-AI R&D is highly capital-intensive, with peers spending hundreds of millions to over a billion dollars annually, so Psibot's roughly $300 million raised implies a finite runway and likely repeated future raises. SR027, SR010
CR031 With no revenue anchor, Psibot's valuation rests on team quality and the data flywheel and is exposed to down-round or multiple-compression risk if benchmarks disappoint or capital tightens. SR023, SR029
CR032 Roughly 22 embodied-AI unicorns were minted in 2026 amid record capital inflows and trillion-dollar market projections, a classic bubble signal raising the odds of valuation compression across the cohort. SR025, SR010
CR033 A Morgan Stanley survey found only 23% of prospective industrial buyers satisfied with current humanoid products even as shipment forecasts were doubled, an adverse demand signal for richly valued startups. SR026, SR021
CR034 Psibot's credibility is concentrated in a small group of founders and academics (CEO Viktor Wang; Yaodong Yang, Yuanpei Chen, Ying Wen), so departure of a key figure would damage both capability and narrative. SR016, SR013
CR035 The market for embodied-AI talent is fiercely competitive among Chinese and global players, raising retention risk for Psibot's key technical staff. SR022, SR009
CR036 Founded only in 2024, Psibot carries the execution risk of a very young organization scaling hardware design, data operations, and OEM relationships simultaneously. SR018, SR013
CR037 Psibot's strongest mitigants are marquee strategic investors (Chery/AiMOGA, Lens Technology) and state-linked funds that de-risk capital access and manufacturing/commercialization channels. SR013, SR016
CR038 Psibot's proprietary manipulation dataset (targeting one million hours in 2026) is a genuine, compounding moat that partly mitigates competitive risk if the model quality it feeds proves out. SR019, SR020
CR039 Thesis-break triggers for Psibot are a credible failure to verify core benchmarks, a durable loss of frontier compute access, a state-driven data-access event that poisons international demand, or a down-round that resets the valuation thesis. SR011, SR008
CR040 Psibot's net residual risk is elevated — regulatory-access, commercial-proof, and valuation risks remain load-bearing and only partly mitigated by strong backing — making these the primary uncertainties diligence must resolve. SR014, SR015, SR021
CV001 Psibot's bull thesis is that licensing a VLA "brain" to robotics OEMs targets the most defensible and scalable layer of the embodied-AI stack, capturing software-like economics without hardware capital intensity. SV011, SV028
CV002 Psibot's July 2026 round of roughly US$100 million, led by Chery Automobile and Lens Technology, set a post-money valuation of US$1.48 billion. SV010, SV011, SV013
CV003 The thesis is reinforced by an elite founding team, a Peking University research lineage, and a stated goal of building China's largest dexterous-hand dataset as a compounding data flywheel. SV011, SV028
CV004 Firms exhibiting at WAIC 2026 carried an aggregate valuation above US$14.7 billion and 15 Chinese embodied-AI startups crossed unicorn status in the first half of 2026. SV006, SV004
CV005 Psibot has disclosed no revenue, ARR, gross margin, or headcount, so the valuation is unsupported by any published financial fundamentals. SV010, SV012
CV006 Psibot's headline capability claims are self-reported and not independently benchmarked, and analysts estimate more than 85% of embodied-AI deployments remain non-productive pilots. SV026, SV024
CV007 The National Intelligence Law and data-security regime cap the international demand a global comparison would otherwise imply for a Chinese embodied-AI vendor. SV025
CV008 Early traction may reflect strategic pull from related backers rather than arm's-length commercial demand, weakening the revenue-quality signal behind the valuation. SV011, SV024
CV009 The recommendation is a conditional Watch — do not lead at the current US$1.48 billion post-money without named production references, audited financials, and a related-party revenue schedule. SV010, SV024, SV026
CV010 Confidence in the judgment is low-to-medium because the public evidence base is thin and dominated by company statements and coverage of a single financing event. SV010, SV012
CV011 The overall risk rating is high, driven by the absence of commercial proof, unverified benchmarks, regulatory exposure, and an expected 2027-2028 sector shakeout. SV002, SV025, SV026
CV012 The valuation stance is that US$1.48 billion is stretched relative to disclosed fundamentals and defensible only by reference to team, data flywheel, strategic-backer signaling, and sector comps. SV010, SV024
CV013 Entry discipline requires named production references, audited financial statements, and a related-party revenue schedule before any capital commitment near the current mark. SV024, SV026
CV014 The recommendation logic chains through four gates — platform scalability, independent proof of capability and traction, contained risk, and price support — which Psibot clears, fails, partly fails, and only conditionally clears respectively. SV024, SV028
CV015 Psibot has raised more than US$300 million cumulatively, including the roughly US$100 million July 2026 round. SV010, SV011, SV014
CV016 An earlier financing of approximately RMB 2 billion (about US$280 million) preceded the July 2026 unicorn round. SV012, SV014, SV015
CV017 Because Psibot is pre-revenue, its valuation is derived from investor signaling, team scarcity, and strategic-backer pull rather than from unit economics. SV024, SV010
CV018 Sector cash runways of 18-24 months imply repeated dilutive rounds before any liquidity event, creating preference-stack overhang for early investors. SV002, SV003
CV019 Public evidence corroborates the fact of the round but does not independently validate US$1.48 billion as fair value on fundamentals. SV010, SV024
CV020 Strategic backers who also consume the technology set the clearing price, so an arm's-length financial investor should expect to pay a premium. SV011, SV024
CV021 The imminent listings of Unitree, DEEP Robotics, and Leju are resetting China's robotics valuation methodology from top-down TAM extrapolation toward auditable deliveries, revenue, and profitability. SV003, SV023
CV022 The bull case (verified benchmarks, scaled OEM licensing, compounding data moat) could re-rate Psibot toward US$4-6 billion or more, in line with software-centric global peer premiums. SV016, SV017, SV005
CV023 The base case (slow pilot conversion, credible but unproven capability) keeps Psibot a private unicorn in roughly the US$1.5-2.5 billion band while consuming further capital. SV010, SV002
CV024 The bear case (benchmark disappointment, regulatory or compute shock, or failed raise) forces a down-round or distressed outcome in the US$0.3-0.7 billion range or worse. SV002, SV025
CV025 Dominant downside triggers are disproof of headline benchmarks, loss of Nvidia-class compute access, a data-security or National-Intelligence-Law enforcement event, and inability to close the next round before runway expires. SV026, SV025
CV026 On the public evidence available today the base case is the most defensible central estimate, and the bull case requires independent proof that does not yet exist. SV002, SV024
CV027 Analysts expect a sector "elimination year" in 2027-2028 as 18-24-month runways expire, raising the weight on the bear tail. SV002, SV003
CV028 Implied value is highly sensitive to the assumed probability of hitting commercial milestones and to the revenue multiple applied once auditable metrics exist. SV003, SV001
CV029 Unitree's approved STAR Market listing implies about US$5.83 billion, with CCB International projecting up to US$15.1 billion including brand premium — a 32x price-to-sales multiple on a profitable maker that shipped over 5,500 robots in 2025. SV009, SV003, SV022
CV030 DEEP Robotics, at roughly one-fifth of Unitree's revenue, carries an even richer implied 41x price-to-sales multiple at an approximately US$1.93 billion issuance valuation. SV003, SV020
CV031 Figure AI reached about US$39 billion in 2025, illustrating the paper premiums attached to leading global humanoid platforms. SV017, SV019
CV032 Physical Intelligence was reported in talks to raise above a US$11 billion valuation. SV016
CV033 Skild AI has been discussed in a roughly US$8-12 billion range and Apptronik closed near US$5.5 billion, framing the upper band of global embodied-AI comps. SV001, SV027
CV034 At least 25 domestic embodied-intelligence companies now carry valuations above RMB 10 billion (about US$1.39 billion), 15 of them minted in the first half of 2026. SV002, SV004, SV030
CV035 The comps' key limitation is comparability — most peers are pre-revenue or forward-multiple priced and the public names are full-stack hardware integrators, whereas Psibot is a software / intelligence-layer licensor. SV003, SV024
CV036 Psibot's plausible exit paths are a domestic STAR Market or Hong Kong listing along the Unitree and GigaAI trail, or a strategic acquisition by an industrial backer such as Chery. SV007, SV029
CV037 Near-term exit readiness is low because Psibot is pre-revenue and private and would need auditable commercial traction before any public window opens. SV003, SV021
CV038 Final diligence asks are audited financials, a named customer-reference list, a related-party transaction schedule, a component bill-of-materials and compute-supply plan, a full cap table with preferences, and independent benchmark verification. SV024, SV026
CV039 The thesis breaks if headline benchmarks are disproven, licensing fails to scale beyond captive partners, a regulatory or export-control event severs compute or data access, or the company cannot raise before runway expiry. SV025, SV026
CV040 On balance Psibot is a high-risk, high-optionality opportunity meriting a conditional pass that becomes investable only once commercial proofs are delivered and price is negotiated with a down-round margin of safety. SV010, SV024, SV026
来源
编号出版方标题引文
SO001 Psibot (Lingchu Intelligence) About Us - PsiBot PsiBot has launched the industry's first end-to-end reinforcement learning-based embodied models—Psi R0, R0.5, and R1.
SO002 Psibot (Lingchu Intelligence) GL Ventures and Lanchi Ventures Lead Investment in PsiBot PsiBot recently completed its angel round financing, led by GL Ventures and Lanchi Ventures.
SO003 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Psibot is finalizing a fundraise of close to $100 million at a $1.48 billion valuation, led by Chery Automobile.
SO004 The Straits Times (Bloomberg) PsiBot becomes latest AI startup to hit US$1 billion value PsiBot is raising close to US$100 million of funding at a US$1.48 billion valuation.
SO005 The Next Web China's Psibot hits a $1.48bn valuation betting on 'world models' Founded only in 2024, Psibot has now raised about $300m.
SO006 Bamboo Works PsiBot's $280 million fundraising bets on the brains behind embodied AI PsiBot was able to raise a hefty 2 billion yuan ($280 million) just two years after its founding, in its angel and Pre-A funding rounds.
SO007 Sahm Capital PsiBot's $280M Fundraising Signals China's Bet On Embodied AI The angel round drew state-backed and industrial capital, including Guokai Finance, Guozhong Capital, and a CCTV-affiliated industry fund.
SO008 Gasgoo Seeds | PsiBot Announces Completion of 2 Billion Yuan Financing PsiBot has completed its angel and Pre-A financing rounds, raising a total of 2 billion yuan.
SO009 Gasgoo Seeds | Embodied AI tech firm PsiBot closes new funding round On April 10, the company released Psi-R2 and Psi-W0 — large models pre-trained on human data — and open-sourced its first batch of 1,000 hours of multimodal human hand manipulation data.
SO010 InforCapital PsiBot - Robotics Startup, $280M Raised PsiBot has raised $280M across 2 funding rounds since 2024.
SO011 Tencent News (智能车参考 / AI4Auto) 车企砸钱北大师徒坐镇!具身机器人新晋独角兽来了 灵初智能正在完成新一轮融资,投后估值达到约14.8亿美元,约101亿人民币。
SO012 Sina Finance 中国企业灵初智能成为最新一家估值突破10亿美元的AI初创公司 灵初智能成为最新一家估值突破10亿美元的AI初创公司。
SO013 NeoDrop AI (AI科技评论) 奇瑞领投 PsiBot,机器人开始算量产账 PsiBot 已经宣布整机量产,但其官方公告没有披露出货量、合同金额或客户数量。
SO014 Humanoid.guide Psi R1 by PsiBot – VLA Embodied AI Model Overview Psi R1 is a Vision-Language-Action (VLA) model by PsiBot... Release date April 2025.
SO015 Rocking Robots PsiBot's R1 Robot Demonstrates Advanced Reasoning with Live Mahjong Gameplay R1 has maintained consistent reasoning and physical interaction for periods of up to 30 minutes while playing Mahjong.
SO016 Psibot (Lingchu Intelligence) The Real VLA is Coming: PsiBot's Psi R1 Ushers in a New Era of Embodied Intelligence PsiBot, with its hierarchical end-to-end VLA + reinforcement learning architecture Psi-R1, has delivered a compelling answer.
SO017 Psibot (Lingchu Intelligence) The Second Wave of Real VLA: Psi R1 Achieves Generalized Intelligence at the Brain Level PsiBot's dexterous hand is capable of executing complex grasping tasks—such as gripping the handle ring of a delivery bag and lifting it smoothly.
SO018 MERICS (Mercator Institute for China Studies) Embodied AI: China's ambitious path to transform its robotics industry China's humanoids still lack precision and dexterity and are mostly deployed in limited tasks and in site-specific trials.
SO019 China Biz Insider Morgan Stanley Raises China Humanoid Robot 2026 Forecast to 50,000 Units Morgan Stanley now projects China's humanoid robot market to reach US$2 billion in 2026 and US$15 billion by 2030.
SO020 TrendForce China's Humanoid Robot Output to Surge 94% in 2026; Unitree and AgiBot to Capture Nearly 80% Market Share Unitree Robotics and AgiBot... together... projected to account for nearly 80% of total shipments.
SO021 Psibot (Lingchu Intelligence) Home - PsiBot Toward a New Era of Embodied Intelligence.
SO022 Psibot (Lingchu Intelligence) Product_Psi R1 - PsiBot Psi R1 is a Vision-Language-Action (VLA) model that enables robots to perform complex, long-horizon tasks.
SO023 Taiwan News The Real VLA is Coming: Psi R1 Starts a New Era of Embodied AI The Real VLA is Coming - Psi R1 Starts a New Era of Embodied AI.
SO024 AI Weekly Morgan Stanley doubles China humanoid robot forecast to 50,000 Morgan Stanley doubled its China humanoid robot forecast to 50,000 units.
SO025 Psibot (Lingchu Intelligence) PsiBot official homepage (Chinese) 用AI和机器人创造无限生产力。
SM001 Global Market Insights Humanoid Robot Market Size, Forecasts Report 2026-2035 The global humanoid robot market was valued at USD 7.9 billion in 2025 and is expected to grow from USD 10.9 billion in 2026 to USD 54.2 billion in 2031 and USD 192.7 billion in 2035, at a CAGR of 37.6%.
SM002 Grand View Research Warehouse Automation Market Size And Share Report, 2030 The global warehouse automation market size was estimated at USD 19.23 billion in 2023 and is projected to reach USD 59.52 billion by 2030, growing at a CAGR of 18.7% from 2024 to 2030.
SM003 Fortune Business Insights Warehouse Robotics Market Size, Share Report | 2026-2034 The global warehouse robotics market size was valued at USD 6.51 billion in 2025 and is projected to grow from USD 7.35 billion in 2026 to USD 25.41 billion by 2034, exhibiting a CAGR of 16.80%. Asia-Pacific dominated with a share of 51.70% in 2025.
SM004 Global Market Insights Logistics Robots Market Size, Forecast Report 2026-2035 The global logistics robots market was estimated at USD 17.8 billion in 2025 and is expected to grow from USD 20.7 billion in 2026 to USD 91.4 billion in 2035, at a CAGR of 17.9%.
SM005 IDC Humanoid Robotics Commercialization Trends 2026: From Task Execution to Value Creation In 2025 the global humanoid robot market experienced a breakout year, led by Chinese vendors, with shipments exceeding 18,000 units; more than 85% of deployments were concentrated in performances, education, data collection and guided tour scenarios.
SM006 Research and Markets Humanoid Robot Market Report 2026 The humanoid robot market report tracks market size, growth rate and hotspots across 2020-2035 with segmentation and attractiveness analysis.
SM007 36Kr Research Institute 2026 Research Report on the Development of the Embodied Intelligence Industry The market scale of China's embodied intelligence has rapidly increased from 213.3 billion yuan in 2018 to 915 billion yuan in 2025 and is expected to exceed the trillion-yuan mark in 2026; 2025 financing reached 33.473 billion yuan in the first 11 months, four times the prior year.
SM008 RobotToday China's 15th Five-Year Plan (2026-2030): Embodied Intelligence as National Industrial Strategy 2025 has been formally designated China's 'first year of humanoid robot mass production' by MIIT; more than 140 domestic manufacturers released over 330 distinct models, and Chinese firms shipped approximately 90% of global humanoid robot units in 2025.
SM009 AI Robotic Daily Embodied AI Bubble: Humanoid Robot Market Valuation & Trends Truly productive revenue from industrial scenarios accounts for merely three to five percent of total sales; forecasts for 2030 range from around four billion dollars to over fifteen billion dollars, a fourfold difference within the exact same time frame.
SM010 AI Weekly Morgan Stanley Doubles China Humanoid Robot Forecast to 50,000 Morgan Stanley doubled its China humanoid robot shipment forecast to about 50,000 units in 2026, with the China market rising toward US$15 billion by 2030.
SM011 China Biz Insider Morgan Stanley Bets on China Humanoid Robots — 50K Units, $2B Market in 2026 Morgan Stanley projects roughly 50,000 China humanoid shipments in 2026 (up about 79% from 28,000) and 446,000 units by 2030, with the China market at about US$2 billion in 2026.
SM012 TrendForce China Humanoid Robot Output to Grow 94% in 2026 China's humanoid robot output is projected to grow about 94% in 2026, with Unitree and AgiBot together accounting for roughly 80% of shipments and China around 84.7% of global humanoid shipments.
SM013 MERICS (Mercator Institute for China Studies) Embodied AI: China's ambitious path to transform its robotics industry Chinese humanoid robots still lack the precision and dexterity for many tasks, run costly site-specific trials, must cut costs by at least half, and continue to look to US research for VLA breakthroughs.
SM014 The Straits Times China's Psibot becomes latest AI start-up to hit US$1 billion value Psibot has become the latest Chinese AI start-up to reach a US$1 billion valuation amid a wave of embodied-AI funding in 2026.
SM015 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Psibot's VLA platform has been validated in a large Chinese logistics client's warehouse for sorting and is being tested at one of the world's largest fibre-optic cable makers.
SM016 The Next Web China's Psibot hits US$1.48 billion valuation on world-models bet Psibot's licensing-first world-models strategy underpins a reported US$1.48 billion valuation.
SM017 Humanoid.guide Psi R1 Product Profile Psi R1 is a hierarchical VLA plus reinforcement-learning model demonstrated in extended autonomous dexterous manipulation.
SM018 Bamboo Works Psibot's $280 million fundraising bets on the brains behind embodied AI Psibot's roughly US$280 million angel-plus-pre-A haul is a bet on the software brain behind embodied AI amid a Chinese funding wave.
SM019 SAHM Capital Psibot's $280M Fundraising Signals China's Bet on Embodied AI The financing signals China's broader bet on embodied AI as a strategic industry.
SM020 Inforcapital Psibot Company Profile Psibot is a Beijing- and Shanghai-based embodied-AI company building VLA models for dexterous manipulation.
SM021 NeoDrop Psibot and the Embodied-AI Funding Frenzy Reported embodied-AI unicorn valuations, including newly minted ones, may be running ahead of grounded deployment and revenue.
SM022 Gasgoo Auto News Embodied AI tech firm Psibot closes new funding round Psibot has closed a new funding round backed by strategic investors as embodied-AI capital surges in China.
SM023 Psibot (Lingchu Intelligence) About Us - PsiBot PsiBot licenses its Psi-series embodied models and sells a proprietary human-hand data-collection system, controlling design while outsourcing component production.
SM024 Sina Finance Psibot completes new financing round (Chinese coverage) Chinese-language coverage reports Psibot's new financing round amid the domestic embodied-AI investment surge.
SM025 QQ News (Tencent) Psibot embodied-AI funding and mass-production coverage (Chinese) Chinese-language coverage situates Psibot within China's push toward humanoid mass production and embodied-intelligence deployment.
SP001 Tech Times Unitree IPO Cleared, AGIBOT Hits 10,000 Units: China Humanoid Robot Duopoly Takes Shape Unitree's 2025 revenue reached 1.699 billion yuan with gross margins reaching 60.13%; it targets a valuation of approximately $6.2 billion. AGIBOT produced its 10,000th humanoid on March 30, 2026, moving from 5,000 to 10,000 in three months.
SP002 China Biz Insider Unitree Races to Commercialize After Record-Speed IPO Approval Research and education accounted for 73.6% of humanoid revenue, commercial 17.4%, and industrial applications a combined 9%; genuine production-line revenue totaled only about RMB 15.7 million (US$2.2 million) in the first nine months of 2025.
SP003 Curionic BYD AgiBot vs Unitree vs UBTECH: China's Humanoid Robot Race in 2026 Compared BYD-backed AgiBot, Unitree, and UBTECH shipped more units than all Western competitors combined; AgiBot shipped 5,168 units in 2025 (Omdia
SP004 TechMarketBriefs Figure AI IPO 2026: $39B Valuation, Risks & Bull Case The September 2025 Series C tripled the cap table at a $39 billion mark on essentially no revenue; the bear case includes a whistleblower lawsuit alleging Figure cut safety and BMW choosing a different vendor for its European expansion.
SP005 Embodied Global US Embodied AI Companies Tracker 2026: OpenAI, Tesla, Figure As of June 2026 Tesla deployed approximately 1,000 Optimus units at Giga Texas, targeting 5,000 internally by year-end; OpenAI leads a $6.7 billion investment in Figure AI and runs an internal 200-plus-researcher robot foundation-model effort codenamed Project Atlas.
SP006 AI in Asia China's Humanoid Robot Industry Shifts Into Mass Production By the close of its 30 June launch day UBTECH's cumulative orders for the UWORLD U1 series passed 13,361 machines, priced from 119,800 yuan, with 88 degrees of freedom and speech-to-lip synchronisation inside 20 milliseconds.
SP007 Physical Intelligence Physical Intelligence — π0.7 Steerable Robotic Foundation Model We are developing learning algorithms to create a model that will control any robot to do any task; π0.7 extracts an RL Token from VLA models and uses Multi-Scale Embodied Memory to enable complex tasks longer than ten minutes.
SP008 The AI Insider Report: Physical Intelligence to Raise $1B with Valuation North of $11B Physical Intelligence is in advanced talks to raise about $1 billion at a valuation north of $11 billion, up from $5.6 billion four months earlier.
SP009 NextBigFuture Figure AI Humanoid Robots Valued at $39 Billion Figure AI closed a $1 billion Series C at a $39 billion valuation with backing from NVIDIA, Intel, Qualcomm, Salesforce, T-Mobile, and Brookfield.
SP010 NewsGlobeNow China Humanoid Robot Startups Race Past 100 Billion Yuan Chinese humanoid robot startups have collectively raced past 100 billion yuan in valuation, with several unicorns each valued above roughly 10 billion yuan (~US$1.4 billion).
SP011 TechMarketBriefs Unitree Stock & IPO 2026: Valuation, Risks & Bull Case Unitree's implied market capitalization stands at approximately RMB 42 billion (US$5.83 billion) even as first-quarter 2026 revenue growth decelerated sharply and profit nearly halved year-on-year.
SP012 TrendForce China Humanoid Robot Output to Grow 94% in 2026 Unitree and AgiBot together account for roughly 80% of China's humanoid shipments, with China around 84.7% of global humanoid shipments in 2026.
SP013 MERICS (Mercator Institute for China Studies) Embodied AI: China's ambitious path to transform its robotics industry Chinese humanoid robots still lack the precision and dexterity for many tasks and continue to look to US research for VLA breakthroughs, while status-quo automation remains cheaper and more reliable for most tasks.
SP014 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Chery Automobile and Lens Technology back Psibot, whose VLA platform is validated in a large Chinese logistics client's warehouse and tested at a leading fibre-optic cable maker.
SP015 The Straits Times China's Psibot becomes latest AI start-up to hit US$1 billion value Psibot reached a US$1.48 billion valuation, an order of magnitude below Figure AI and Physical Intelligence and below Unitree's IPO target.
SP016 Psibot (Lingchu Intelligence) About Us - PsiBot PsiBot licenses its Psi-series model and sells a proprietary human-hand data-collection system, aiming to build China's largest dexterous-hand dataset.
SP017 Humanoid.guide Psi R1 Product Profile Psi R1 uses a hierarchical fast/slow VLA plus reinforcement-learning architecture demonstrated in a 30-plus-minute autonomous Mahjong session.
SP018 36Kr Research Institute 2026 Research Report on the Development of the Embodied Intelligence Industry China's embodied-intelligence field hosts more than 140 humanoid manufacturers with vertically integrated supply chains and independent model iteration.
SP019 RobotToday China's 15th Five-Year Plan (2026-2030): Embodied Intelligence as National Industrial Strategy Chinese policy organises around a 'big brain' for high-level decision-making and a 'small brain' for real-time motor control, funding model development across both tiers in direct competition with Boston Dynamics, Figure AI, and others.
SP020 NeoDrop Psibot and the Embodied-AI Funding Frenzy Newly minted embodied-AI unicorn valuations may be running ahead of grounded deployment and revenue relative to better-funded rivals.
SP021 The Next Web China's Psibot hits US$1.48 billion valuation on world-models bet Psibot's US$1.48 billion valuation positions it as a mid-tier challenger among embodied-AI brain providers.
SP022 Gasgoo Auto News Embodied AI tech firm Psibot closes new funding round Psibot closed a new funding round backed by strategic automotive and components investors amid intense embodied-AI competition in China.
SP023 Inforcapital Psibot Company Profile Psibot builds VLA models for dexterous manipulation and licenses its intelligence layer to robot developers.
SP024 IDC Humanoid Robotics Commercialization Trends 2026 In 2025 Chinese vendors led a breakout humanoid market with shipments exceeding 18,000 units, and more than 85% of deployments were in performances, education, data collection, and guided-tour scenarios rather than production work.
SP025 Global Market Insights Humanoid Robot Market Size, Forecasts Report 2026-2035 The global humanoid robot market is forecast to grow from US$10.9 billion in 2026 toward US$54.2 billion by 2031, drawing dozens of well-funded entrants.
SI001 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Psibot closed a nearly US$100 million round led by Chery Automobile with Lens Technology, reaching a US$1.48 billion valuation, and licenses its VLA platform rather than manufacturing robots.
SI002 The Straits Times China's Psibot becomes latest AI start-up to hit US$1 billion value Psibot reached a US$1.48 billion valuation after a fundraise of nearly US$100 million, having raised about US$300 million since its 2024 founding.
SI003 Gasgoo Auto News Embodied AI tech firm Psibot closes new funding round Psibot closed a new funding round backed by strategic automotive and components investors, extending its capital base for embodied-AI development.
SI004 Psibot (Lingchu Intelligence) About Us - PsiBot PsiBot licenses its Psi-series model, sells a proprietary human-hand data-collection system, and aims to build China's largest dexterous-hand dataset.
SI005 The Bamboo Works PsiBot's $280 million fundraising bets on the brains behind embodied AI PsiBot raised about US$280 million across angel and Pre-A rounds by March 2026, betting on licensing the intelligence layer for embodied AI rather than selling hardware.
SI006 Sahm Capital PsiBot's $280M fundraising signals China's bet on embodied AI PsiBot's roughly US$280 million angel-plus-Pre-A raise was announced during China's Two Sessions in March 2026.
SI007 Gasgoo Auto News PsiBot announces completion of 2 billion yuan financing PsiBot announced completion of about 2 billion yuan (~US$280 million) in combined angel and Pre-A financing on 10 March 2026.
SI008 PR Newswire (Lens Technology) Lens Technology: Securing AI Edge Hardware Leadership Through Three Core Strengths and a Three-Year Roadmap Since its IPO Lens Technology has committed more than RMB 20 billion to R&D, including RMB 2.44 billion in the first nine months of 2025, extending into robot joints and dexterous hands for embodied intelligence.
SI009 The Bamboo Works Lens Technology approaches second 'iPhone moment' with AI and robotics Lens Technology's revenue rose 16.1% to 53.7 billion yuan in the first nine months of 2025 as it invests heavily in AI wearables and embodied intelligence as new growth engines.
SI010 Reportify (Unitree STAR Market IPO Prospectus, 招股说明书) 宇树科技 招股说明书 — Unitree Robotics STAR Market IPO Prospectus (filing) Unitree's prospectus reports 2025 revenue of 1,708.21 million yuan (up 335.36%) at a 60.27% gross margin, with humanoid shipments over 5,500 units, ranked first globally.
SI011 Yicai Global Unitree Robotics Files for USD608 Million IPO in Shanghai Unitree plans to raise about CNY4.2 billion (USD608 million); revenue rose 335% to CNY1.7 billion and net profit widened 674% to CNY600 million in 2025, with the G1 priced from CNY85,000 and R1 Air from CNY29,900.
SI012 China Biz Insider Unitree's STAR Market IPO Filing Puts China's Humanoid Robot Economics Under a Spotlight Unitree's revenue grew from RMB 123 million in 2022 to RMB 1.167 billion in the first three quarters of 2025, but the filing intensifies scrutiny on whether demand is driven by spectacle rather than practical industrial use.
SI013 HTX / Deep Tide TechFlow Decoding Unitree's IPO Prospectus — The True Picture of the Robot Market Unitree shows 335% revenue growth to ~US$252M in 2025 at ~60% gross margins on deep vertical integration, but 74% of humanoid sales are research, 17% display, and only 9% genuine industrial applications.
SI014 Humanoids Daily Inside Unitree's Prospectus: Revenue Climbs and Profits Dip as STAR Market IPO Hearing Approaches Unitree's Q1 2026 revenue jumped 68% year-on-year but adjusted net profit plunged 52% to 40.3 million yuan due to massive R&D and sales expenses.
SI015 Tech Times Unitree Robotics Nears Shanghai IPO: Profitable on Actuators, Exposed to China's Spy Law Unitree is profitable largely on actuator and component sales, and its filing exposes revenue-mix and legal-risk questions that apply across China's humanoid sector.
SI016 Tech in Asia Chery backs $100m funding for Chinese AI startup PsiBot Chery backs a roughly US$100 million funding round for Shanghai-based PsiBot.
SI017 The Next Web China's Psibot hits US$1.48 billion valuation on world-models bet Psibot's US$1.48 billion valuation reflects investor appetite for embodied-AI brains despite limited disclosed revenue.
SI018 IDC Humanoid Robotics Commercialization Trends 2026 In 2025 more than 85% of humanoid deployments were in performances, education, data collection, and guided-tour scenarios rather than production work.
SI019 MERICS Embodied AI: China's ambitious path to transform its robotics industry Chinese humanoids remain too expensive and lack the precision for many tasks, and costs must fall by at least half before broad commercial deployment becomes viable.
SI020 NeoDrop Psibot and the Embodied-AI Funding Frenzy Embodied-AI unicorn valuations may be running ahead of grounded revenue and deployment amid a 2026 funding frenzy.
SI021 Humanoid.guide Psi R1 Product Profile Psi R1 pairs a hierarchical VLA-plus-RL model with a proprietary exoskeleton-glove data-collection system for dexterous manipulation.
SI022 Inforcapital Psibot Company Profile Psibot builds and licenses VLA models for dexterous manipulation and offers simulation and training-data platforms to robot developers.
SI023 Sina Finance 灵初智能完成新一轮融资 (Lingchu Intelligence completes new funding round) 灵初智能 (Psibot) completed a new financing round in July 2026 led by strategic investors, valuing the company in the unicorn tier.
SI024 PsiBot (Lingchu Intelligence) PsiBot Products — Psi R1 PsiBot positions Psi R1 as a licensable intelligence layer with simulation and data services for robot developers.
SI025 Global Market Insights Humanoid Robot Market Size, Forecasts Report 2026-2035 The global humanoid robot market is forecast to grow from US$10.9 billion in 2026 toward US$54.2 billion by 2031, with heavy investment flowing into both hardware and intelligence layers.
SE001 PR Newswire (PsiBot) The Real VLA is Coming: Psi R1 Starts a New Era of Embodied AI PsiBot's robots play Mahjong with humans via L3 autonomous reasoning based on Chain of Action Thought (CoAT), performing long-horizon complex manipulation in open environments.
SE002 Embodied Global PsiBot Dual-Model Architecture for Embodied AI The Psi-R2 world action model integrates VLA capabilities trained on 95,472 hours of human data covering 294 scenarios and 4,821 tasks; Psi-W0 enables counterfactual reasoning, cutting inference from 2.2 seconds to under 100 milliseconds.
SE003 GitHub (Psi-Robot) Psi-Robot/DexGraspVLA: [AAAI'26 Oral] A Vision-Language-Action Framework Towards General Dexterous Grasping DexGraspVLA open-sources code and dataset for a hierarchical vision-language-action framework towards general dexterous grasping, accepted as an AAAI 2026 Oral.
SE004 GitHub (Psi-Robot) Psi-Robot organization repositories The Psi-Robot organization hosts DexGraspVLA, an Awesome-VLA-Papers survey list, and official models of Psi SynHand.
SE005 arXiv DexGraspVLA: A Vision-Language-Action Framework Towards General Dexterous Grasping (arXiv:2502.20900) DexGraspVLA is a hierarchical framework using a pre-trained vision-language model as high-level planner and a diffusion-based low-level controller, achieving 90+% dexterous grasping success under thousands of unseen cluttered scenes.
SE006 DexGraspVLA Project (PKU-PsiBot Joint Lab) DexGraspVLA project page Authored by the Institute for AI, Peking University and the PKU-PsiBot Joint Lab (AAAI 2026 Oral), DexGraspVLA demonstrates robust zero-shot generalization to unseen objects, backgrounds, and lighting.
SE007 Hugging Face Paper page — DexGraspVLA The DexGraspVLA paper page aggregates the framework's abstract, code, and community engagement for general dexterous grasping.
SE008 OpenVLA (Stanford/Google) OpenVLA — An Open-Source Vision-Language-Action Model OpenVLA, a 7B open VLA model, sets a new state of the art and outperforms the 55B-parameter closed RT-2-X across multiple robot platforms.
SE009 Psibot (Lingchu Intelligence) Psi R1 product page Psi R1 is a hierarchical end-to-end VLA-plus-RL model delivering L3 dexterous manipulation via Chain of Action Thought and a fast/slow dual-system brain.
SE010 Psibot (Lingchu Intelligence) PsiBot news — Psi-R2 / Psi-W0 and data reserve PsiBot open-sourced the first 1,000 hours of a 100,000-hour multimodal human-hand manipulation data reserve alongside the Psi-R2/W0 dual-model release.
SE011 Psibot (Lingchu Intelligence) PsiBot company news PsiBot describes a small full-stack model defining robot design parameters while outsourcing manufacturing, with sub-models for planning (Psi-P0) and control (Psi-C0).
SE012 Psibot (Lingchu Intelligence) PsiBot technology news PsiBot details its Psi-series model progression and its Impossible Triangle goal of high generalization, dexterity, and success rate.
SE013 Psibot (Lingchu Intelligence) About Us — PsiBot PsiBot licenses its Psi-series model, sells the proprietary Psi-SynEngine data-collection system, and aims to build China's largest dexterous-hand dataset with one million hours of data in 2026.
SE014 Humanoid.guide Psi R1 — product profile Psi R1 progressed from Psi R0 (Dec 2024) through R0.5, V1/H1 hardware, to R1 (May 2025), a hierarchical VLA-plus-RL system for dexterous manipulation.
SE015 Rocking Robots PsiBot's R1 robot demonstrates advanced reasoning with live Mahjong gameplay PsiBot's R1 played Mahjong with humans for more than thirty minutes autonomously, demonstrating long-horizon reasoning and dexterous tile manipulation.
SE016 The Next Web China's PsiBot hits $1.48 billion valuation on world-model bet PsiBot's bet centres on world models — Psi-W0 — that let robots reason about actions before executing them, underpinning its US$1.48 billion valuation.
SE017 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Psibot licenses its VLA platform rather than making robots, with a 21-DOF dexterous hand and deployments in logistics sorting and fibre-optic manufacturing.
SE018 The Straits Times China's Psibot becomes latest AI start-up to hit US$1 billion value Psibot, which licenses embodied-AI models for dexterous manipulation, reached a US$1.48 billion valuation after raising nearly US$100 million.
SE019 MERICS Embodied AI: China's ambitious path to transform its robotics industry Chinese humanoids still lack precision and dexterity, rely on site-specific trials, remain too expensive, and continue to look to US research for VLA breakthroughs.
SE020 Tech Times Unitree Robotics Nears Shanghai IPO; Profitable Actuators, Exposed to China's Spy Law Chinese embodied-AI firms depend on Nvidia compute and face data-security and national-security law exposure, including the Cybersecurity Law regime.
SE021 IDC Humanoid Robotics Commercialization in 2026 More than 85% of 2025 humanoid deployments were non-productive, underscoring the gap between demonstrated capability and reliable field performance.
SE022 Gasgoo Auto News Embodied AI tech firm Psibot closes new funding round Psibot, developer of the Psi-series VLA model for dexterous manipulation, closed a new strategic funding round to scale its embodied-AI platform.
SE023 Curionic BYD/AgiBot vs Unitree vs UBTech — China humanoid robot comparison 2026 Chinese humanoid makers share dependence on foreign precision components such as ball screws from Schaeffler, THK, and NSK.
SE024 Physical Intelligence Physical Intelligence — pi0 and generalist robot policies Physical Intelligence builds generalist vision-language-action foundation models (pi0, Hi Robot) for dexterous robot control, a global comparator to Psibot's approach.
SE025 Neodrop PsiBot embodied-AI unicorn analysis PsiBot positions its Psi model as an intelligence layer licensed to robot makers, compounding a proprietary dexterous-manipulation dataset.
SU001 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Psibot has run small-scale warehouse validations at a large Chinese logistics client and tested at one of the world's largest fibre-optic cable makers, licensing its VLA platform to robot makers.
SU002 The Straits Times China's Psibot becomes latest AI start-up to hit US$1 billion value Psibot, backed by Chery and Lens Technology, licenses embodied-AI models for dexterous manipulation and targets logistics automation in China.
SU003 Gasgoo Auto News Embodied AI tech firm Psibot closes new funding round Psibot, whose Psi model has been validated in Chinese logistics pilots, closed a new strategic funding round to scale deployment.
SU004 Psibot (Lingchu Intelligence) About Us — PsiBot PsiBot licenses its Psi-series model to robot developers and targets building China's largest dexterous-hand dataset with one million hours of data in 2026.
SU005 Psibot (Lingchu Intelligence) Psi R1 product page Psi R1 performs generalized dexterous manipulation for logistics and manufacturing tasks such as sorting and handling.
SU006 Psibot (Lingchu Intelligence) PsiBot news — Psi-R2 / Psi-W0 and data reserve PsiBot's data-collection strategy underpins its 100,000-hour reserve, feeding models deployed with logistics partners.
SU007 Psibot (Lingchu Intelligence) Home — PsiBot PsiBot demonstrates dexterous manipulation use cases — Mahjong, LEGO assembly, bolt tightening, piano — alongside logistics pick-and-place generalization.
SU008 News18A (Lingchu Intelligence coverage) Lingchu Intelligence Secures New Funding to Advance Embodied AI R&D and Deployment Lingchu Intelligence's financing included SDIC Advanced Manufacturing Fund and Jingxi Ruiling, state-linked investors backing its embodied-AI logistics deployment.
SU009 Robotics Business News AiMOGA Robotics Announces Humanoids and Secures 1,000-Unit Deal to Accelerate Global Commercialization Backed by Chery Group, AiMOGA unveiled the humanoid Mornine and signed 1,000 units of its Intelligent Police Robot, advancing large-scale embodied-AI deployment across 30-plus countries.
SU010 Digital Market Reports Chery Partners with AiMOGA Robotics to Expand Intelligent Ecosystem at Auto China 2026 Chery identified robotics as its third growth curve and partnered with AiMOGA Robotics to promote real-world embodied-intelligence applications under a scenario-driven, globally scalable approach.
SU011 AInvest Morgan Stanley Doubled China's 2026 Humanoid Robot Forecast — FOMO or Real Commercialization? Morgan Stanley nearly doubled China's 2026 humanoid shipment view from 28,000 to 50,000 units, but the question remains whether this reflects real commercialization or FOMO ahead of a catalyst window.
SU012 CoinLive China's Humanoid Robot Boom Gains Pace as Morgan Stanley Lifts Shipment Forecast Again Morgan Stanley raised its China humanoid shipment forecast again as robot makers accelerate production, pointing to a fast pilot-to-production shift.
SU013 TechBuzz Morgan Stanley Doubles China Humanoid Robot Forecast Morgan Stanley doubled its 2026 China humanoid robot shipment forecast, citing State Grid procurement and supply-chain capacity.
SU014 Tech in Asia Chery backs $100m funding for Chinese AI startup PsiBot Chery led a roughly US$100 million round in PsiBot, a Shanghai/Beijing embodied-AI startup licensing its model to robot makers.
SU015 Houdao AI Lingchu Intelligence: Collecting a Million Hours of Human Data with Data Gloves to Solve the Embodied AI Implementation Challenge Lingchu Intelligence collects a million hours of human manipulation data via data gloves to solve the embodied-AI implementation challenge in logistics.
SU016 Humanoid.guide Psi R1 — product profile Psi R1 targets dexterous manipulation for logistics and manufacturing, licensed as a model to third-party robot developers.
SU017 Rocking Robots PsiBot's R1 robot demonstrates advanced reasoning with live Mahjong gameplay PsiBot's R1 demonstrated long-horizon dexterous manipulation, a capability it markets to logistics and manufacturing customers.
SU018 MERICS Embodied AI: China's ambitious path to transform its robotics industry Chinese embodied-AI deployments rely on site-specific trials and remain too expensive, limiting broad, arm's-length customer adoption.
SU019 IDC Humanoid Robotics Commercialization in 2026 More than 85% of 2025 humanoid deployments were non-productive, underscoring that deployment does not equal sustained productive customer usage.
SU020 China Biz Insider Morgan Stanley Raises China Humanoid Robot 2026 Forecast to 50,000 Units Morgan Stanley found only 23% of prospective industrial buyers satisfied with current humanoid products even as it raised its 2026 China shipment forecast to 50,000 units.
SU021 Curionic BYD/AgiBot vs Unitree vs UBTech — China humanoid robot comparison 2026 China's humanoid makers compete for the same logistics and industrial buyers, where deployment scale and reliability determine customer wins.
SU022 The Next Web China's PsiBot hits $1.48 billion valuation on world-model bet PsiBot's licensing model aims to spread its embodied-AI brain across many robot makers' bodies rather than sell directly to end-customers.
SU023 The Bamboo Works PsiBot's $280 million fundraising bets on the brains behind embodied AI PsiBot's backers include GL Ventures and Lanchi Ventures alongside state-linked co-investors, betting on licensing the intelligence layer rather than selling hardware.
SU024 Inforcapital PsiBot company profile PsiBot's customers are robot developers licensing its Psi model for logistics and manufacturing manipulation tasks.
SU025 Neodrop PsiBot embodied-AI unicorn analysis PsiBot positions its Psi model as an intelligence layer licensed to robot makers, compounding a proprietary dexterous-manipulation dataset from deployments.
SR001 A&O Shearman Key amendments to China's Cybersecurity Law The amended Cybersecurity Law, effective 1 January 2026, raises the general administrative-fine cap from RMB 1 million to RMB 10 million, integrates AI ethics and risk-monitoring obligations, and expands extraterritorial reach to any overseas activity that endangers China's cybersecurity.
SR002 Rimon Law China AI Regulatory Developments — July 2026 Analysis China's 2026 AI governance agenda folds generative-AI, agent, and robotics risk into existing data and cybersecurity law pending comprehensive AI legislation.
SR003 China Law Translate What China's National Intelligence Law Says, And Why it Doesn't Matter Article 7's cooperation obligation sits in the general provisions, mirrors duties in other laws, and has no enforcement mechanism; yet it would be difficult for any Chinese citizen or company to resist a direct state security request, and courts cannot be relied on for a remedy.
SR004 US Department of Homeland Security Data Security Business Advisory: Risks and Considerations for Businesses Using Data Services and Equipment from Firms Linked to the PRC Chinese laws, including the National Intelligence Law, may compel firms linked to the PRC to provide the Chinese government access to data, creating risk for businesses using such firms' data services and equipment.
SR005 Chambers and Partners Cybersecurity 2026 — China (Global Practice Guides) China's Cybersecurity Law, Data Security Law and PIPL impose data classification and a cross-border-transfer regime requiring a CAC security assessment, standard contractual clauses, or certification.
SR006 Klea Legal China Data Laws 2026: Key Changes for Businesses From 1 January 2026 the amended Cybersecurity Law raises penalties to RMB 10 million, while cross-border transfers require a CAC assessment, SCCs, or certification and biennial audits for large-scale personal-data handlers.
SR007 Humanoid.guide Humanoid Robot Supply Chain Report 2026–2027 — Tier 1 and Tier 2 Precision reducers and harmonic drives remain a technology bottleneck where foreign suppliers set the standard, even as China's rising local capacity reduces import dependency for motors, actuators and batteries.
SR008 The Board Strategic Impact of NVIDIA Export Controls on China US export controls on advanced Nvidia chips remain a supply-chain tail risk for players reliant on US compute, pushing Chinese robotics firms toward domestic alternatives that trail the frontier.
SR009 Robotics Center of Silicon Valley State of Robotics 2026 — China China controls an estimated 63–70% of the global humanoid supply chain, with component suppliers clustered within a short logistics radius, though high-end precision reducers and frontier compute remain partial dependencies.
SR010 CNBC Humanoid robots touted as next trillion-dollar AI investment opportunity Analysts project a multi-billion-dollar humanoid niche growing toward a $200 billion market by 2035, fuelling record funding rounds and eye-popping valuations across the embodied-AI cohort.
SR011 Voxos.ai The State of Embodied Intelligence: Robotics in 2026 Most VLA benchmark demonstrations remain highly controlled, with claims of broad generalization and robust autonomy largely unverified by third parties and a persistent gap between demos and reliable deployment.
SR012 RoboZaps Humanoid Robot Industry Report 2026 Over a dozen humanoid robots are commercially available amid record funding, but the sector faces growing skepticism over unverified benchmarks and ambitious labor-displacement claims.
SR013 The Straits Times China's Psibot becomes latest AI start-up to hit US$1 billion value Psibot, backed by Chery and Lens Technology, reached a US$1.48 billion valuation while licensing embodied-AI models, with no disclosed revenue anchoring the figure.
SR014 MERICS Embodied AI: China's ambitious path to transform its robotics industry Chinese embodied-AI deployments rely on site-specific trials, remain too expensive with costs needing to fall by half or more, and still look to US research for VLA breakthroughs.
SR015 IDC Humanoid Robotics Commercialization in 2026 More than 85% of 2025 humanoid deployments were non-productive, underscoring that deployment does not equal sustained productive usage.
SR016 Tech Times Chery Bets on Psibot's VLA Platform as China Births Another Embodied AI Unicorn Psibot licenses its VLA platform to robot makers and relies on Nvidia compute and foreign precision components common across China's humanoid stack.
SR017 Tech Times Unitree Robotics Nears Shanghai IPO — Profitable Actuators Exposed, China's Spy Law China's national intelligence and data laws expose Chinese robotics firms to compelled-cooperation and data-access concerns that constrain foreign sales and heighten geopolitical scrutiny.
SR018 Gasgoo Auto News Embodied AI tech firm Psibot closes new funding round Psibot, founded in 2024, closed a new strategic round to scale deployment, remaining a very young organization scaling data operations and OEM relationships.
SR019 Psibot (Lingchu Intelligence) About Us — PsiBot PsiBot licenses its Psi model to robot developers and targets China's largest dexterous-hand dataset with one million hours of data in 2026, designing structure and motion range while outsourcing component production.
SR020 Psibot (Lingchu Intelligence) PsiBot news — Psi-R2 / Psi-W0 and data reserve PsiBot reports capability milestones and a 100,000-hour data reserve underpinning its models, with headline demonstrations that are company-reported.
SR021 AInvest Morgan Stanley Doubled China's 2026 Humanoid Robot Forecast — FOMO or Real Commercialization? The question remains whether doubled humanoid forecasts reflect real commercialization or FOMO ahead of a catalyst window, a bubble signal for richly valued embodied-AI startups.
SR022 Curionic BYD/AgiBot vs Unitree vs UBTech — China humanoid robot comparison 2026 China's humanoid makers compete intensely for the same buyers and for scarce embodied-AI talent, raising execution and retention risk for newer entrants.
SR023 The Next Web China's PsiBot hits $1.48 billion valuation on world-model bet PsiBot's $1.48 billion valuation rests on a world-model and licensing bet rather than disclosed revenue, exposing it to re-rating if the technology bet underperforms.
SR024 The Bamboo Works PsiBot's $280 million fundraising bets on the brains behind embodied AI PsiBot's backers include GL Ventures and Lanchi Ventures alongside state-linked co-investors, funding a capital-intensive embodied-AI bet on the intelligence layer.
SR025 AI Robotic Daily Is the embodied AI market a bubble? Roughly 22 embodied-AI unicorns were minted in 2026 amid record capital inflows, a classic bubble signal raising the odds of valuation compression.
SR026 China Biz Insider Morgan Stanley Raises China Humanoid Robot 2026 Forecast to 50,000 Units A Morgan Stanley survey found only 23% of prospective industrial buyers satisfied with current humanoid products even as the 2026 China shipment forecast was raised to 50,000 units.
SR027 Sahm Capital PsiBot's $280M fundraising signals China's bet on embodied AI PsiBot has raised around US$300 million to date, a finite runway for a capital-intensive embodied-AI developer likely to require further raises.
SR028 Inforcapital PsiBot company profile PsiBot's revenue path runs through robot-developer licensees rather than diversified arm's-length customers.
SR029 Humanoids Daily Inside Unitree's prospectus — revenue climbs and profits dip ahead of STAR Market IPO Even a leading Chinese humanoid maker's prospectus shows thin, volatile profitability, a caution for pre-revenue peers carrying unicorn valuations.
SR030 NextBigFuture Figure AI humanoid robots valued at $39 billion Global embodied-AI valuations have reached extreme levels, exemplified by Figure AI's reported $39 billion figure, framing the multiple-compression risk for the cohort.
SV001 AI Funding Tracker Embodied AI valuations and rounds — 2026 Embodied-AI valuations in 2026 span from early-stage Chinese unicorns near US$1.4 billion to global leaders such as Figure AI and Physical Intelligence in the tens of billions, with most priced on forward potential rather than revenue.
SV002 China Biz Insider China's new embodied-AI unicorns — a reality check on runways At least 25 Chinese embodied-intelligence startups now carry valuations above RMB 10 billion, 15 of them in the first half of 2026, but most carry cash runways of only 18 to 24 months, implying a reckoning between 2027 and 2028.
SV003 China Biz Insider Unitree's IPO and the valuation reckoning for China robotics Unitree's STAR Market listing implies a market capitalization of RMB 42 billion, with CCB International projecting RMB 109 billion including brand premium — a price-to-sales multiple of 32x — while DEEP Robotics implies roughly 41x, resetting how the primary market prices pre-revenue robotics companies.
SV004 Embodied Global 15 embodied-AI unicorns crowned in H1 2026 Fifteen Chinese embodied-AI startups reached unicorn status in the first half of 2026, spanning full-stack integrators and software-centric brain developers.
SV005 Embodied Global Humanoid robot funding report — H1 2026 Global humanoid and embodied-AI funding accelerated through the first half of 2026, with software-centric platforms commanding premium valuations relative to hardware-only makers.
SV006 Yicai Global WAIC 2026 embodied-AI exhibitors carry over US$14.7B aggregate valuation Embodied-AI firms exhibiting at the 2026 World Artificial Intelligence Conference carried an aggregate valuation exceeding US$14.7 billion, underscoring the concentration of capital in the sector.
SV007 AI in China Unitree Robotics IPO — coverage Unitree Robotics cleared China's securities regulator for a STAR Market listing, opening a domestic public-market path that peers including embodied-AI startups are expected to follow.
SV008 Robotics Center China robotics market — sizing and outlook China's robotics market is projected to expand rapidly through the late 2020s, with embodied-AI and humanoid segments drawing the largest share of venture and strategic capital.
SV009 Reportify Unitree Robotics STAR Market listing filing Unitree's listing filing discloses profitability and deliveries of more than 5,500 robots in 2025, providing an auditable public reference point for pricing Chinese robotics peers.
SV010 The Straits Times China's Psibot becomes latest AI startup to hit US$1 billion value Psibot's latest round valued the Chinese embodied-AI startup at US$1.48 billion, making it one of the newest unicorns in a crowded field, on funding led by Chery Automobile and Lens Technology.
SV011 Tech Times Chery bets on Psibot's VLA platform as China births another embodied-AI unicorn Chery Automobile led a roughly US$100 million round in Psibot, whose Vision-Language-Action platform licenses an AI "brain" to robotics manufacturers rather than building its own hardware.
SV012 Gasgoo Auto News Embodied-AI tech firm Psibot closes new funding round Psibot closed a new financing round backed by automotive and manufacturing strategics, adding to more than US$280 million raised across prior rounds.
SV013 The Next Web Psibot raises at US$1.48B on world-model robotics bet Psibot's new financing set a post-money valuation of US$1.48 billion, extending a wave of billion-dollar valuations for Chinese embodied-AI developers.
SV014 Bamboo Works Psibot secures about US$280M in embodied-AI financing Psibot had raised on the order of US$280 million before its unicorn round, an unusually large sum for a company founded only in 2024.
SV015 Sahm Capital Chinese embodied-AI firm Psibot raises around US$280 million Psibot's cumulative fundraising approached US$280 million ahead of its 2026 unicorn round.
SV016 The AI Insider Physical Intelligence in talks to raise above US$11B valuation Physical Intelligence was reported to be in talks to raise new funding at a valuation exceeding US$11 billion, underscoring the premiums attached to software-centric embodied-AI platforms.
SV017 NextBigFuture Figure AI reaches about US$39B valuation Figure AI reached a valuation of about US$39 billion in 2025, a roughly fifteenfold increase in about 18 months, illustrating the paper premiums attached to leading humanoid platforms.
SV018 Tech Market Briefs Unitree pre-IPO financing and valuation trajectory Unitree's pre-IPO financing rounds lifted its valuation ahead of a STAR Market listing that analysts value in the tens of billions of yuan.
SV019 Tech Market Briefs Figure AI funding and humanoid valuation context Figure AI's funding history frames the upper end of humanoid-robot valuations, with leading US platforms priced far above their Chinese counterparts.
SV020 Humanoids Daily Unitree prospectus — deliveries and profitability Unitree's prospectus shows it shipped more than 5,500 robots in 2025 and reached profitability, while peers such as DEEP Robotics and Leju disclose far smaller revenue and, in Leju's case, continuing losses.
SV021 HTX Insights Decoding Unitree's prospectus economics Unitree's prospectus economics give investors a rare audited window into robotics unit economics, raising the bar for pre-revenue startups seeking comparable valuations.
SV022 Yicai Global Unitree Robotics raises about US$608M ahead of listing Unitree Robotics raised about US$608 million in pre-IPO financing, valuing the company ahead of a STAR Market listing that will provide the sector's first large auditable public benchmark.
SV023 China Biz Insider Unitree IPO economics and the China robotics valuation debate Once Unitree, DEEP, and Leju trade publicly, primary-market investors gain three granular reference points that reward delivery and profitability, narrowing financing windows for pre-revenue startups.
SV024 AInvest Morgan Stanley warns of FOMO-driven embodied-AI valuations Morgan Stanley cautioned that fear-of-missing-out dynamics are inflating embodied-AI valuations well ahead of commercial proof, with only a minority of prospective industrial buyers satisfied with current products.
SV025 MERICS China's embodied-AI push and its structural constraints MERICS judges many Chinese embodied-AI deployments to be site-specific and expensive, and notes regulatory and data-governance constraints that limit international demand for Chinese robotics vendors.
SV026 IDC Most embodied-AI deployments remain non-productive pilots IDC estimates that more than 85% of humanoid and embodied-AI deployments in 2025 remained non-productive pilots rather than revenue-generating production systems.
SV027 Embodied Global US embodied-AI funding tracker — global comparison US embodied-AI leaders including Figure, Physical Intelligence, Skild AI, and Apptronik command valuations from several billion to tens of billions of dollars, setting the global comparison band.
SV028 Psibot Psibot company overview Psibot builds a Vision-Language-Action platform and dexterous-manipulation data pipeline that it licenses to robotics manufacturers, positioning itself as the intelligence layer of the embodied-AI stack.
SV029 CNBC The trillion-dollar humanoid-robot race Investors are racing into humanoid robotics on expectations of a trillion-dollar market, funding both US and Chinese developers well ahead of meaningful revenue.
SV030 Curionic Comparing China's embodied-AI unicorns China's embodied-AI unicorns range from full-stack humanoid makers to software-only brain developers, with valuations clustering above RMB 10 billion despite little disclosed revenue.