Deep Robotics
这家未上市的工业足式机器人公司已实现首个盈利年度,并正接近科创板上市
Deep Robotics 已从机器人叙事走到真实工业收入和首次盈利,但私有 / IPO 前状态、资本强度和估值证据噪音较大,使更合适的判断仍是观察而非买入。
封面要素
公司概况
Deep Robotics 是一家杭州起家的工业机器人公司,由浙江大学研究人员朱秋国和李超创办。公司的商业化产品面已覆盖 X 系列四足机器人、Lynx 轮足机器人、Lite3 开发者平台和早期 DR 人形机器人线;公开客户证据最强的是电力巡检,以及 Singapore Power 的隧道巡检部署。到 2026 年运行日,公司仍未上市,但已完成辅导并获得科创板申报受理;公开申报摘要显示,2025 年收入为人民币 3.37 亿元、首次盈利,毛利率超过 50%。尽调约束不在于业务是否真实,而在于股权结构条款、现金 / 现金跑道、客户集中度和经常性收入质量仍明显披露不足。
- 创始人
- Zhu Qiuguo, Li Chao
- 创立地点
- Hangzhou, Zhejiang, China
- 总部
- Hangzhou, Zhejiang, China
- 产品
- X20/X30 工业四足机器人、Lynx 轮足机器人、Lite3 开发者和教育平台,以及早期 DR 人形机器人线,面向巡检、安防、应急响应和复杂工业场景销售。
- 客户
- 公用事业、工业运营商、公共安全和安防团队、应急响应组织,以及科研或开发者用户。
- 商业模式
- 以硬件机器人销售和巡检解决方案为主,辅以载荷、软件和开发者工具、部署集成及相关服务。
- 阶段
- Private, pre-IPO
- 融资情况
- 公开报道显示,公司在 2025 年 7 月完成近 CNY500 million 的 pre-Series C,2025 年 12 月完成超过 CNY500 million 的 Series C,并在 2026 年初完成数亿元人民币的 pre-IPO 轮,但累计融资总额仍未核清。
执行摘要
主要优势
- 2025 年公开申报摘要显示商业化已成立:收入 RMB 337 million、首次盈利、毛利率 52.83%,不再只是原型机故事。
- 公司已拼出覆盖四足、轮足机器人和早期人形产品的多产品栈,最清晰的落地证据来自电力巡检和 Singapore Power 隧道作业流。
- Deep Robotics 已完成股改、辅导并获得 STAR Market 申报受理,同时仍有国内战略投资者支持,退出准备度好于多数私有机器人同行。
主要风险
- 公开估值证据不一致,且高于纪律性低十亿美元区间后显得偏紧:late-2025 私募标记、2026 媒体标记和更高的 $3.5 billion IPO 标题同时流传。
- 下一阶段仍很吃资本;拟募资 RMB 2.503 billion 将投向具身 AI 研发、机器人本体、产业化和制造扩张。
- 收入质量披露仍不足:现金、烧钱速度、客户集中度、实际折扣、经常性服务占比、保修负担和复购队列都未公开。
- 竞争、认证和地缘政治压力可能压缩利润率或 TAM,尤其是在价格 / 性能上面对 Unitree、在海外关键基础设施市场落地时。
未决问题
- 完整股权结构表、清算优先权、VAM 或特殊权利条款,以及 IPO 进程延后时这些权利如何生效
- 客户集中度、复购队列、服务附加、保修负担,以及按产品线拆分的实际折扣
- 现金、月度烧钱速度、营运资金需求,以及现金跑道相对于计划 STAR Market 募资时间和确定性的情况
- 已具名多台人形机器人商业化案例,以及系统级危险场景或安全 / 合规认证
目录
01公司概况
1.1 身份、阶段和商业模式
Deep Robotics 把自己定位成一家围绕危险和工业场景打造的具身 AI 机器人公司,而不是消费机器人故事。官方材料称,公司是总部位于杭州的高科技企业,聚焦四足机器人、人形机器人和机器人核心部件,采用全栈「感知—决策—执行」框架,面向能源、应急响应、工业运营和教育。独立报道与这一叙事一致:公司成立于 2017 年 11 月,仍未上市;现在更适合被理解为一家接近 IPO 的晚期未上市工业机器人公司,而不是早期科研衍生项目。公开资料里可见的商业模式,是向巡检、救援、物流和科研场景销售硬件和解决方案,并在已商业化的四足机器人基础上叠加后续人形机器人工作。核心尽调提醒是披露深度。基本身份和产品范围已经清楚,但审计财务、当前客户数和精确累计融资额仍未在公开口径中核清。[CO001, CO002, CO003, CO005, CO007, CO041]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 / 备注 |
|---|---|---|---|---|
| 总部 | 中国浙江杭州 | 当前 | 高 | 已审阅公开资料包未披露确切运营地址。 |
| 成立日期 | 2017 年 11 月 | 2017-11 | 高 | 已审阅来源未确认确切注册日。 |
| 当前法律状态 | 私有股份有限公司(未上市) | 2025-11-03 | 高 | 截至报告日,未识别到公开招股书或交易所申报文件。 |
| 主营业务 | 工业四足、轮足和人形机器人 | 当前 | 高 | 未披露按产品家族划分的收入结构。 |
| 报道估值 | 约 CNY8B(约 USD1.1B) | 2025-12-25 | 中 | 媒体报道数字,而非交易所验证标记。 |
| 已披露 2025 年融资 | 7 月近 CNY500M,12 月另有 >CNY500M | 2025 | 高 | Pre-IPO 轮仅被描述为数亿元人民币。 |
| 累计融资历史 | 截至 2026 年初报道 8 轮以上 | 2026-01-05 | 中 | 公开信息未对累计融资总额作统一核对。 |
| 2024 年收入披露 | 数亿元人民币;同比增长 >100% | 2025 | 中 | 没有公开的 2024 年经审计收入报表。 |
| 盈利信号 | 管理层称公司未亏损 | 2025 | 中 | 没有公开的经审计毛利率、烧钱或现金流细节。 |
| 团队规模 | 近 400 人 | 2025 | 高 | 未发现截至报告日更新的 2026 年员工数披露。 |
| 2025 年出货目标 | 10,000 台 | 2025 | 中 | 这是管理层目标,不是已报告实际值。 |
| 报道部署宽度 | 超过 600 个行业项目 | 2026-01-05 | 中 | 未披露客户数量和按项目计的收入集中度。 |
快照明确保留缺乏支撑的私有公司指标,而不是用臆造数值填补。
[CO003, CO005, CO007, CO008, CO009, CO010]Deep Robotics 将创始人主导的机器人研发,与工业产品、客户部署、后期资本和境内 IPO 路径连在一起。
[CO001, CO002, CO004, CO006, CO019, CO020]1.2 创始人、公开管理梯队和治理可见度
创始故事确实构成了承销判断的一部分,因为 Deep Robotics 在技术方向和公开叙事上仍明显由创始人主导。朱秋国和李超来自浙江大学机器人生态,公开访谈把朱秋国呈现为战略、商业和使命设定的声音,李超则负责技术商业化、工业适用性和行业定位。朱秋国的副教授身份,以及公司反复强调的全栈自研、专利深度和工业部署纪律,也强化了这种创始人—市场匹配。更不让人安心的是,创始人之外的公开治理层很薄。已审阅来源提到若干面向公众的运营负责人,例如产品和海外销售负责人,但没有给出清晰的董事会名单、独立董事名单,或股改后的完整高管明细。这意味着,领导层深度主要通过公开发言人和客户侧运营人员可见,而不是通过标准的晚期治理披露包可见。[CO003, CO004, CO031, CO032, CO033, CO034]
| 人物 | 当前职务 | 公开背景 | 覆盖范围 / 创始人-市场匹配 | 关键人物依赖 |
|---|---|---|---|---|
| Zhu Qiuguo | 创始人兼 CEO | Zhejiang University 副教授、机器人研究者 | 设定公司使命、融资叙事和商业化优先级;把创始人-市场匹配锚定在足式机器人 | 高——公司主要战略和融资门面 |
| Li Chao | 联合创始人兼 CTO | Zhejiang University 博士;商业化议题上的公开技术发言人 | 掌握技术可信度、产品实用性,以及四足到人形路线图叙事 | 高——产品架构和商业化叙事的核心人物 |
| Peter Deng | 产品经理 / 公开发言人 | 在 2024 年公开产品网络研讨会报道中代表 Deep Robotics | 向外部受众解释 X30 用例、工业部署和降本逻辑 | 中——体现运营人员深度,但不是已披露高管 |
| Hank Cheng | 高级销售经理 / 部署负责人 | 在新加坡 SPock 部署报道中被点名 | 覆盖海外本地化、客户适配和国际执行证明 | 中高——对出口部署和客户证明很重要 |
| Mao Tang | Lynx M20 产品经理 | Lynx M20 发布文章引用其发言 | 代表轮足产品定位和行业解决方案打包 | 中——体现产品线深度,而非企业控制权 |
| Li Chang-bo | 海外销售负责人,韩国市场 | 就海外定价和竞争定位被引用 | 传递国际商业化抱负和性价比信息 | 中——有用的扩张信号,但不是已披露公司高管 |
部分列举聚焦于已审阅材料点名的创始人和公开露面的运营负责人。
[CO003, CO004, CO031, CO032, CO033, CO034]1.3 资本基础、未上市状态和利益相关方地图
Deep Robotics 的融资故事规模不小,但还没达到完整披露级别。公开报道支撑了一条清晰的近期融资序列:2025 年 7 月近 CNY500 million 的 pre-Series C,2025 年 12 月超过 CNY500 million 的 Series C,以及 2026 年初仅被描述为数亿元人民币的 pre-IPO 轮。公司还在 2025 年 11 月完成股份制改制,变为未上市股份公司,并在 12 月 8 日签署辅导协议后,于 12 月 23 日提交 IPO 辅导备案。合在一起看,这些步骤强烈支持一个判断:公司仍未上市,但正在积极准备境内上市。公开记录也支持媒体报道的约 CNY8 billion 估值,但还无法拼出招股书级别的股权图。精确累计融资额、债务或授信敞口,以及任何老股交易历史仍未解决,因此投资者能看清动能和融资方向,却还不能只靠公开文件重建干净的股权结构表或估值标记历史。[CO005, CO006, CO007, CO008, CO009, CO010]
| 利益相关方 | 角色 | 控制权或经济重要性 | 尽调要求 |
|---|---|---|---|
| 投资方:Fortune Capital + CRHC/Guoxin Fund | 2025 年 7 月 Pre-Series C 领投方 | 锚定近 CNY500M 融资,并强化国家关联资本支持 | 确认轮次文件、优先权栈,以及任何董事会或否决权。 |
| 投资方:China Merchants Bank International + China Asset Management | 2025 年 12 月 Series C 共同领投方 | 支持 >CNY500M 融资,并拓宽后期财务投资人基础 | 索取条款清单、获得的所有权,以及任何与 IPO 时间挂钩的结构化权利。 |
| China Telecom / China Unicom 关联基金 | Series C 战略投资人 | 价值可能不止资本,因为它们能打开工业网络和基础设施渠道 | 评估其是被动投资人,还是分销 / 生态伙伴。 |
| 投资方:National AI Industry Fund + JD.com | Pre-IPO 轮领投 / 跟投方 | 传递上市前的国家优先级背书和后期市场可信度 | 核实确切轮次规模、资金用途,以及该轮是否重置估值。 |
| 电网客户:State Grid / China Southern Power Grid | 电力板块旗舰需求锚 | 支撑工业用途投资逻辑和长周期巡检用例 | 索取合同期限、扩张逻辑,以及按部署计的已实现 ROI。 |
| 海外客户:SP Group + Eastern Green Power | 海外部署交易对手 | 提供公开记录中最清晰的具名国际客户证明和本地化案例 | 索取试点到规模化转换条款,以及初始隧道项目后部署是否扩大。 |
| IPO 顾问组:China Securities、Zhong Lun、Pan-China | 上市准备交易对手 | 虽然不是股权持有人,但对 IPO 路径至关重要 | 在招股书提交前确认目标交易所、时间线,以及未结法律或会计工作流。 |
利益相关方地图强调当前经济或尽调相关性最清晰的资本提供方、战略渠道和客户锚点。
[CO006, CO008, CO009, CO010, CO011, CO012]紧凑展示可投资性:哪些公开可见,哪些仍不透明。
KPI 条有意将若干私营公司披露项保留为定性表述,因为未找到经审计的公开数值。
[CO007, CO008, CO009, CO010, CO011, CO013]1.4 产品、客户证据和规模信号
公司概况里最强的一点是,Deep Robotics 没有直接跳进人形机器人营销,而是先在电力巡检和其他高风险环境中做出可用的四足机器人生意。公开证据显示,公司已有一条工业产品阶梯:X20 奠定全天候四足机器人基线,X30 成为出口级巡检旗舰,DR01 在 2024 年完成人形机器人首秀,Lynx M20 在 2025 年 4 月把产品组合扩展到轮足移动,DR02 则在 2025 年 10 月把公司推向工业人形机器人。客户和部署证据在电力基础设施中最强。Singapore Power Group 的 SPock 部署既有独立佐证,也有具体运营细节,包括隧道巡检、定位工作,以及估计每年节省 480 个工时。管理层访谈还指向应急响应、安防巡逻、科研和工业现场的应用深度正在扩大,并提到超过 600 个项目和 2025 年 10,000 台出货目标。即便审计后的客户数和 backlog 披露仍缺席,这仍是一个有分量的商业化信号。[CO017, CO018, CO019, CO020, CO021, CO022]
1.5 里程碑和主要反向张力
里程碑记录让 Deep Robotics 看起来明显比纯概念阶段的人形机器人初创公司成熟。从 2018 年第一代四足机器人和 2020 年电网试点开始,公司推进到产品化工业四足机器人、海外部署、新产品品类、反复融资和正式 IPO 准备。公开来源还显示,2024 年收入同比翻倍以上,员工数达到大约 400 人,管理层也愿意表示公司没有亏损经营。反向张力不是缺少动作,而是缺少完整证明。投资者和行业分析师的公开评论仍在质疑,机器人头部公司能否足够快地放大 ROI 为正的部署,从而支撑当前估值标记;DR02 发布本身也留下了自主性、续航和成本问题。换句话说,Deep Robotics 的商业化进度似乎领先许多人形机器人同行,但公开资料仍支持谨慎读法:真实牵引、真实里程碑、真实 IPO 准备,与不完整的治理和财务披露,以及尚未解决的商业化质疑并存。[CO013, CO014, CO015, CO016, CO017, CO025]
| 日期 | 事件 | 类型 | 金额 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2017-11 | Deep Robotics 在杭州成立 | 成立 | 公司组建 | Zhu Qiuguo;Li Chao | 为后续所有章节建立核心创始基线。 |
| 2018-07 | 报道称获得 Ginkgo Valley Capital 和 Daosheng Capital 天使融资 | 融资 | 天使轮;已审阅英文资料包未披露金额 | 来源:Deep Robotics;Ginkgo Valley Capital;Daosheng Capital | 显示首个原型阶段之后不久就有外部资本进入。 |
| 2018 | 发布国内首款能爬楼梯、自主导航的四足机器人 | 产品 | 第一代 Jueying 里程碑 | Deep Robotics | 标记最早的产品和技术证明点。 |
| 2020 | China Southern Power Grid 试点开启电力巡检商业化拐点 | 合作 | 首个主要工业场景锚 | 来源:Deep Robotics;China Southern Power Grid | 推动公司从实验室机器人走向真实工业采用。 |
| 2022-10 | Fangguang Capital 领投 Series B 融资 | 融资 | Series B 完成;已审阅来源未说明规模 | Deep Robotics;Fangguang Capital 及跟投方 | 支持 2025 年融资激增前的工业规模化。 |
| 2024-08 | DR01 在 World Robot Conference 首发 | 产品 | 首次公开人形机器人亮相 | Deep Robotics | 显示从四足机器人扩展到具身 AI 和人形机器人。 |
| 2024-12 | X30/SPock 部署于新加坡电力隧道 | 规模化 | 海外电力系统中的首款中国四足机器人 | 来源:Deep Robotics;SP Group;Eastern Green Power | 形成公开记录中最清晰的国际部署证明。 |
| 2025-04 | Lynx M20 发布 | 产品 | 新增轮足产品家族 | Deep Robotics | 将平台拓展到极端地形移动用例。 |
| 2025-07-08 | 宣布 Pre-Series C 融资 | 融资 | 融资近 CNY500M | Deep Robotics;Fortune Capital;CRHC/Guoxin 等 | 为产品和人才扩张提供重要成长资本。 |
| 2025-10-09 | DR02 工业人形机器人发布 | 产品 | 声称全球首款 IP66 人形机器人 | Deep Robotics | 推动公司进入户外工业人形机器人品类。 |
| 2025-11-03 | 完成股改,公司变为未上市股份制实体 | 治理 | 法律形式转换 | Deep Robotics | 强烈释放 IPO 准备和治理重组信号。 |
| 2025-11 | Caijing 投资人批评质疑当前机器人是否足够有用 | 负面 | 商业化质疑浮出公开视野 | Caijing 来源;Li Chao;未具名投资人 | 尽管增长动能存在,但引入了关于实用性和 ROI 的真实负面视角。 |
| 2025-12-09 | Series C 融资完成 | 融资 | 融资超过 CNY500M | Deep Robotics;CMB International;China Asset Management;电信基金 | 增加后期资本和战略基础设施支持方。 |
| 2025-12-23 | 12 月 8 日顾问协议后,CSRC 记录 IPO 辅导备案 | 监管 | 辅导流程启动 | 来源:Deep Robotics;China Securities;Zhong Lun;Pan-China | 形成公司迈向内地上市的最清晰标记。 |
| 2026-01 | 报道 Pre-IPO 轮融资 | 融资 | 数亿元人民币;确切规模未披露 | 来源:Deep Robotics;National AI Industry Fund;JD.com | 暗示公司在上市准备阶段仍持续融资。 |
年表混合了确切日期和月份级日期,因为已审阅公开资料包并不支持具体到日。
[CO003, CO006, CO009, CO010, CO011, CO019]Deep Robotics 从高校孵化走向工业部署平台,再进入 IPO 准备期,同时从四足机器人拓展到轮腿式和人形产品。
在已审阅公开报道无法支持精确到日时,采用月份级日期。
[CO003, CO004, CO005, CO006, CO009, CO010]02市场分析
2.1 市场边界、纳入支出和替代方案
不应把 Deep Robotics 按整个机器人经济,甚至整个四足机器人品类来估值。公司自己的公开证据窄得多:工业巡检、地下导航、应急响应,以及围绕难以进入环境的科研部署。这个边界意味着,纳入支出应围绕公用事业和工业运营商的移动巡检硬件、载荷、自主软件和部署服务。它也意味着明确排除项。纯防务 UGV 项目、消费四足机器人、固定传感器和纯无人机预算,应放进相邻或替代池,而不是 Deep 专属 TAM 故事。这个区分很关键,因为现状并非空白;人工巡检、轮式机器人、固定感知和无人机已经覆盖许多巡检任务。只有当楼梯、狭小空间、不规则地形或接触式感知让足式移动明显优于这些替代方案时,Deep Robotics 才能赢。[CM001, CM002, CM003, CM004, CM005, CM039]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 相关性 |
|---|---|---|---|---|
| 公用事业和电网巡检 | 面向变电站、隧道和 UHV 资产的机器人平台、传感器、自主能力、部署服务 | 不含移动巡检的通用智能电网 IT 和固定传感器 | 电网运营商、电厂业主、维护预算 | Deep 核心滩头阵地 |
| 可再生能源场站巡检 | 远程风电场巡逻、异常检测、地形移动、报告工作流 | 风机制造支出和通用无人机预算 | 风电资产运营方和 O&M 团队 | 有实时证明的核心邻近市场 |
| 危险工业巡检 | 认证和正常运行时间很重要的爆炸风险设施巡检 | 不含移动巡检的通用工厂自动化 | 油气、化工和矿业运营方 | 重要但受认证约束 |
| 研究和教育 | 可负担的足式平台、SDK 访问、实验和培训 | 大众玩具和非机器人实验室设备 | 大学、实验室、研发项目 | 有用的出货量层,变现较低 |
| 国防和公共安全 | 任务特定四足机器人和机构采购 | 民用公用事业预算 | 国防机构和公共安全部门 | 邻近市场,不是 Deep 主要证据基础 |
边界有意收窄:聚焦巡检和相邻现场作业,即足式移动能力能改变工作流经济性的场景。
[CM001, CM002, CM003, CM004, CM005, CM028]Deep Robotics 从宽泛的四足机器人 TAM,收窄到更小的巡检主导切片;在这个切片中,腿式移动能力优势更清晰。
[CM001, CM005, CM011, CM034, CM039]2.2 规模测算口径和相互矛盾的估计
公开语料无法支撑一个干净的 Deep Robotics 标题 TAM。最好的大市场数据来自四足机器人市场报告,但这些估计差异很大。Precedence 预计 2026 年市场为 USD 3.71 billion,到 2035 年为 USD 17.53 billion;Emergen 则从 2024 年 USD 1.2 billion 起步,到 2034 年仅达到 USD 4.8 billion。矛盾不代表市场是假的,而是品类定义不同,尤其是防务、物流、科研和巡检是否被打包在一起。Deep Robotics 因此必须用多重口径来测算:广义四足机器人 TAM、中国巨大的机器人安装基础和政策支持,以及由证据支撑、更窄的公用事业和工业运营商已部署巡检机器人切片。投资者应把广义 TAM 与可触达 SAM 之间的差距当作核心尽调问题,而不是舍入误差。这个差距会影响估值,因为不同发布机构能让同一家公司看起来很小众,也能看起来很巨大。[CM006, CM007, CM008, CM009, CM010, CM011]
| 发布方 / 视角 | 年份 | 地理范围 | 数值 | CAGR | 方法论 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|
| Precedence 全球四足机器人 TAM | 2026 | 全球 | 3.71 | 18.84 | 包括国防和工业用途的宽口径四足机器人市场 | 中 | 可能宽于 Deep 实际服务市场 |
| Emergen 全球四足机器人 TAM | 2024 | 全球 | 1.2 | 14.8 | 预测窗口和品类边界不同的宽口径四足机器人市场 | 中 | 基数低得多,造成矛盾的 TAM 叙事 |
| IFR 中国机器人安装视角 | 2024 | 中国 | 295000 | 以工业机器人安装量作为国内自动化代理指标 | 中 | 覆盖所有工业机器人,不只四足机器人 | |
| Deep 公用事业部署视角 | 2025 | 中国 / 新加坡 | 变电站、风电场、隧道和 UHV 场站案例研究 | 中 | 证明 SAM 方向,不证明市场规模 | ||
| ANYbotics 危险行业视角 | 2026 | 全球能源 / 化工 | 借助 ATEX/IECEx 认证证明,切入已认证危险场所巡检细分市场 | 中 | 相关细分市场规模尚未公开拆分 | ||
| 国防预算邻近视角 | 2024 | 美国 | 1.4 | Emergen 引用的 DoD 无人地面系统预算 | 低-中 | 国防支出并非 Deep 当前的主要市场 |
数值混合了美元和设备数量,因为公开材料只能支撑不同视角,不能拼出一套权威的 Deep 专属 TAM/SAM/SOM。
[CM006, CM007, CM008, CM009, CM010, CM011]公开四足机器人市场估算差异很大,因此读区间比选一个 TAM 数字更安全。
各行保留发行方特定的点估算,并统一为一个单位(USD billions),而不是抹平成虚假的共识预测。
[CM006, CM007, CM008, CM010, CM035]2.3 买方、用户、付款方和区域滩头
证据最充分的近期买方是公用事业和能源基础设施运营商,而不是泛泛的「企业」桶。Deep Robotics 自己的案例研究指向变电站、风电场、特高压资产和新加坡电力隧道,这意味着用户包括维护人员、现场巡检员和运营团队。不过,付款方或赞助方更可能是掌握维护、安全和自动化预算的电网运营商、工厂业主或工业运营职能。科研和教育买家遵循另一套逻辑:他们更在意可负担性和实验空间,而不是认证后的正常运行时间。竞争证据进一步把图像拉清楚。ANYbotics 面向危险工业运营商,并在油气领域拥有认证和渠道优势;Ghost 主要服务防务机构;Unitree 则显示中国厂商正在向公用事业巡检收敛,同时也覆盖低成本科研需求。因此,Deep 的公开滩头在能源基础设施最强,中国和周边亚洲的证据强于西方。这个分裂也影响销售动作:公用事业账户需要现场集成和采购耐心,科研买家成交更快,但合同金额小得多。[CM005, CM017, CM018, CM019, CM020, CM021]
| 细分市场 | 买方 | 用户 | 付款方 | 工作流 | 预算负责人 | 采用触发点 |
|---|---|---|---|---|---|---|
| 电网和变电站 | 国家电网运营方和公用事业资产业主 | 巡检技术员和维护班组 | 公用事业运营组织 | 日常巡检、异常检测、应急响应 | 维护 / 自动化预算 | 安全暴露和重复巡检人力 |
| 风电和偏远可再生能源资产 | 风电场运营方 | 现场 O&M 团队 | 资产运营方 | 偏远路线巡检和故障检测 | 运营和可靠性预算 | 恶劣地形和人手短缺 |
| 危险油气和化工场所 | 工业工厂运营方 | 巡检和 HSE 团队 | 工厂业主或场站运营方 | 带认证要求的危险区域巡检 | 安全和可靠性预算 | 需要 ATEX 级安全和正常运行时间 |
| 国防和公共安全 | 国防机构和专业运营方 | 军事或响应单位 | 政府采购机构 | 任务侦察和机构运营 | 项目采购预算 | 受保护的国内安全需求 |
| 研究和教育 | 大学和机器人实验室 | 研究人员和学生 | 实验室 PI 或院系 | 实验、教学和二次开发 | R&D 或教学预算 | 平价平台准入和 SDK 开放度 |
公用事业和工业场站里,买方、用户、付款方分离最明显;研究类买方则把这些角色压缩在一个组织内。
[CM018, CM019, CM020, CM021, CM027, CM028]公用事业、高危行业运营商、国防机构和研究实验室采购理由不同,预算来源也不同。
[CM017, CM020, CM022, CM027, CM028, CM034]2.4 采用驱动因素、约束和估值相关性
当机器人能解决昂贵、危险或重复的巡检工作时,采用逻辑最强。Deep Robotics 声称在变电站和风电场实现了有意义的效率和准确率提升,Emergen 引用自主系统带来的重大巡检成本下降,OSHA 统计也让工业付款方持续面对工人安全压力。中国还通过第十五个五年规划和仍偏低的机器人密度提供政策顺风。但约束对估值同样重要。ANYbotics 的 ATEX 和 IECEx 认证说明,危险场所合规是一条真实的商业护城河。Unitree 公开承认自主性限制,GasGoo 强调量产和可靠性摩擦,Boston Dynamics 的定价加 Orbit 软件也说明企业买家必须同时消化硬件和编排支出。这些约束意味着 Deep Robotics 有理由在公用事业和选定工业场景增长,但除非认证、软件和可靠性一起提升,市场不会按完整 TAM 出清。[CM015, CM016, CM021, CM023, CM024, CM025]
| 驱动因素 / 约束 | 方向 | 时间 | 影响 | 尽调问题 |
|---|---|---|---|---|
| 中国机器人政策和较低机器人密度 | 正向 | 当前至长期 | 扩大国内自动化需求,也让巡检机器人试点更容易获得正当性 | 这部分需求有多少流向移动巡检,而不是工厂机器人? |
| 危险巡检的安全压力 | 正向 | 当前 | 付款方不仅能用人效,也能用减少伤害来论证机器人投入 | 目标客户里,哪个买方真正掌握安全预算? |
| 可量化的巡检 ROI | 正向 | 当前 | Deep 和市场报告都把自主巡检描述为能带来显著效率提升 | 这些收益离开公司撰写的案例后还能持续吗? |
| 危险场所认证门槛 | 负向 | 当前 | ATEX 和 IECEx 要求可能把未认证机器人挡在高价值油气场所之外 | Deep 是否有可核验的认证路线图? |
| 自主能力仍需要人工支持 | 负向 | 当前 | 手动操作和二次开发抬高总成本,也拖慢落地 | 购买后还需要多少工程投入? |
| 制造和可靠性扩张风险 | 负向 | 当前至中期 | 原型成功不等于车队规模下能稳定保持现场正常运行时间 | 是否有质保、MTBF 或车队留存数据? |
| 硬件加软件的企业支出 | 混合 | 当前 | 车队编排和分析软件抬高预算规模,也可能增强粘性 | Deep 目前是否通过经常性软件变现? |
| 区域采购限制 | 负向 | 当前 | 国防和受保护渠道对中国供应商而言结构性更难进入 | 哪些民用市场可能受到外溢限制? |
估值最大的摆动因素不在于机器人技术上是否可行,而在于 Deep 能否把硬件验证与认证、软件和车队可靠性配起来。
[CM013, CM021, CM023, CM024, CM025, CM029]采用通常从高危人工工作开始,进入试点路线;若 ROI 和可靠性站住脚,再接入软件并扩大机队部署。
[CM024, CM025, CM029, CM032, CM037, CM038]2.5 图表
03竞争格局
3.1 格局和竞争赛道
Deep Robotics 面对的是分层竞争场,而不是单一参照对手。在自己的主赛道里,公司把 X30 和 Lynx 定位为面向管线、变电站、隧道和其他危险基础设施环境的工业巡检机器人,并用中国变电站、宁夏风电资产、新加坡电缆隧道和特高压站案例支撑这个定位。Unitree 是最接近的国内硬件同行,因为它同时覆盖民用和工业四足机器人,但其公开材料仍披露了部分栈存在手动操作和二次开发限制。ANYbotics 是最危险的全球工业巡检对手,因为它在电力、能源和公用事业上与 Deep 重叠,同时持有 Deep 尚未公开展示的 ATEX/IECEx 危险区域资质。Boston Dynamics 仍是品类里最知名的高端品牌,但其市场角色是更高成本的平台加软件方案,而不是低成本中国公用事业专家。Ghost Robotics 在公用事业巡检中重要性较低,却提醒我们,四足机器人需求也流向防务和政府预算;中国来源硬件在那里可能面对结构性采购壁垒。替代方法仍然真实存在:室内无人机、固定传感器和现状下的人工巡检仍能解决部分巡检任务,尤其是在不需要完整四足移动能力的场景。[CP001, CP004, CP005, CP006, CP007, CP008]
| 公司 / 方法 | 类别 | 规模 / 融资信号 | 目标买方 | 核心差异化 | 相比 Deep Robotics 的主要短板 |
|---|---|---|---|---|---|
| Deep Robotics | 直接 | 私营;公开规模未披露;2025 年有多项部署案例研究 | 电力公用事业、隧道、变电站、工业巡检 | 公用事业部署验证;X30 四足机器人加 Lynx 轮足选项 | 没有公开定价、收入或危险区域认证证据 |
| Unitree | 直接 / 邻近 | 覆盖广泛民用和工业产品线;B2 和 Go2-W 公开露出 | 开发者、专业消费者、工业运营方、集成商 | 高机动规格、轮足选项、宽硬件梯队 | 公开自主能力免责声明;巡检专用验证少于 Deep |
| Boston Dynamics | 既有高端标杆 | 2020 年起由 Hyundai 控制;品牌领先,配有企业软件栈 | 工业巡检、危险响应、研究 | Spot 载荷能力加 Orbit 车队管理和预配置套装 | 历史价格高,成本驱动的公用事业采用证据较弱 |
| ANYbotics | 直接全球巡检对手 | 据 Yokogawa 2026 年信息,融资超过 $150m,团队 200 名专家 | 油气、化工、电力、采矿、交通 | ATEX/IECEx Zone 1 巡检可信度和全球能源客户背书 | 定价未披露且可能是企业高端价;公开中国验证少于 Deep |
| Ghost Robotics | 邻近 / 受保护渠道 | 25+ 国家安全客户,已售出或签约 1,000+ 台机器人 | 国防、国土安全、公共安全 | 适配美国政府任务采购;开放架构 | 公开材料几乎没有民用公用事业巡检重叠证据 |
| Flyability Elios 3 | 替代方法 | 企业巡检无人机,公开定价未披露 | 受限空间巡检团队和工业资产业主 | 在不可达空间进行室内无人机建图和成像 | 不能替代爬楼接触式巡检或四足载荷任务 |
规模信号综合了截至 2026-06-21 披露的融资、所有权或客户信息。未知财务数据保留描述,不做估算。
[CP013, CP017, CP021, CP025, CP026, CP032]用监管 / 部署准备度和可覆盖买方任务宽度两个维度,对主要竞争者及一种替代方法做序数地图。
轴是有证据支撑的序数评分,不是市场份额或收入估算。
[CP013, CP020, CP025, CP027, CP028, CP034]3.2 能力、定价和买方适配对比
从能力看,当巡检买家更在意地形处理、公用事业工作流和集成感知,而不是品牌或防务血统时,Deep Robotics 最强。X30 规格表强调 IP67 防护、多传感器融合和 2.5-4 小时任务,工业巡检页面还加入热成像、声学和机械臂选项,能清晰映射到公用事业和工厂维护工作流。Lynx M20 具有战略意义,因为它显示 Deep 并不执着于纯四足机器人:当速度和走廊效率重要时,公司正转向轮足设计。Unitree B2 是更显眼、进攻性更强的通用硬件挑战者,拥有更高的标称速度、载重和轮足可选性,但其关于手动操作的公开免责声明削弱了它对完全无人值守巡检的论证。Boston Dynamics 依靠 Orbit 和预配置包,仍是托管巡检软件的标杆;不过 Spot 历史定价也解释了为什么许多买家会先测试更便宜或按报价销售的亚洲替代方案。ANYbotics 通过认证后的 Zone 1 就绪度赢得证据最充分的危险环境位置,而 Ghost Robotics 更贴合防务任务,不是民用能源资产。这个市场的公开定价披露仍然很薄,因此可执行的比较重点不在精确标价,而在于哪些厂商最适合具体买方任务、监管环境和自主性预期。[CP002, CP003, CP004, CP010, CP011, CP012]
| 购买标准 | Deep Robotics | Unitree | Boston Dynamics | ANYbotics | Ghost Robotics | Flyability Elios 3 |
|---|---|---|---|---|---|---|
| 变电站和隧道巡检适配度 | 高 | 中 | 中 | 中 | 低 | 中 |
| 危险 / Ex 区准备度 | 未知 / 公开材料未显示 | 低 / 公开材料未显示 | 低 / 有危险响应套装,但此处未引用 Ex 认证 | 高 / ATEX-IECEx Zone 1 | 低 | 低 |
| 国防 / 国家安全采购适配度 | 低 | 低 | 中 | 低 | 高 | 低 |
| 开阔区域空中覆盖 | 低 | 低 | 低 | 低 | 低 | 高 |
| 轮足或速度优先机动选项 | 中 / Lynx | 高 / B2 轮足选项 | 低 | 低 | 低 | 低 |
| 车队软件和打包工作流证据 | 中 | 低 | 高 | 中 | 中 | 低 |
| 真实工业部署的公开验证 | 高 | 中 | 中 | 高 | 民用巡检为低 | 中 |
高 / 中 / 低是基于保留证据作出的编辑判断,不是实验室基准。未知表示保留的公开材料不足以支撑更强结论。
[CP002, CP003, CP010, CP011, CP014, CP020]| 产品 | 获取模式 | 公开价格证据 | 公开显示的内含能力 | 尽调影响 |
|---|---|---|---|---|
| Deep Robotics X30 | 企业 / 报价驱动 | 未公开披露 | 四足硬件、多传感器巡检、自主定位 | 需要直接报价和支持范围,才能比较 TCO |
| Deep Robotics Lynx M20 | 预订 / 企业 | 未公开披露 | 轮足机动、热插拔电池、聚焦工业地形 | 可能是对速度效率质疑的战略回应 |
| Unitree B2 | 产品 / 报价驱动,带商店页面露出 | 抓取的官方页面未保留可读标价 | 高速工业四足、重载荷、可选轮足形态 | 可能压低高端同业价格,但公开条款仍不透明 |
| Boston Dynamics Spot | 企业套装 | 历史公开基准:2020 年基础款 $74,500,另加配件 | 14kg 载荷、预配置套装、Orbit 车队软件 | 为 ROI 审查设定高端成本参照点 |
| ANYmal X | 企业 / 报价驱动 | 未公开披露 | ATEX-IECEx Zone 1 巡检机器人,带能源市场背书 | 认证护城河可能支撑高端定价 |
| Ghost Vision 60 | 政府合同 / 企业 | 未公开披露 | 国防任务载荷、开放架构、坚固平台 | 不是干净的民用同类价格比较对象 |
| Flyability Elios 3 | 企业 / 报价驱动 | 未公开披露 | 室内无人机建图和受限空间巡检 | 空中进入优于地面机动的场景里,它是有用替代方案 |
巡检机器人公开标价普遍稀缺。未知项是有意保留,应该通过报价、服务条款和软件包尽调解决,而不是猜测。
[CP004, CP011, CP012, CP014, CP015, CP020]序数矩阵展示哪些替代方案最适合特定巡检环境和采购语境。
数值是锚定引用来源的定性评估。公开来源无法支撑更强判断时,保留「未知」。
[CP003, CP010, CP014, CP020, CP024, CP032]3.3 切换成本、替代方案和渠道进入
从公开证据看,Deep Robotics 似乎没有硬性合同锁定;切换成本更多来自运营和场景。一旦工厂或公用事业客户完成路线建图、调好巡检载荷、把热成像 / 声学读数接入工作流,并训练员工使用某个机器人家族,更换厂商就会产生摩擦——但这不是某些企业软件市场里的排他数据护城河。因此,多供应商并用是可信的。买家可以在变电站试点一款四足机器人,在危险石化区域部署另一款,同时仍用无人机或固定传感器处理开阔区域或连续监测任务。最强替代方案不是另一只机器狗,而是混合栈:例如 Flyability 的 Elios 3 这类用于室内 / 狭小空间成像的无人机、用于全天候告警的固定传感器、ABB 这类常规工业自动化厂商提供的更广巡检和测试自动化,以及用于低频或低复杂度路线的人工班组。当任务需要楼梯、隧道、杂乱环境、接触式感知或无需重度场地改造的频繁重新部署时,Deep 有优势。若买方需要认证防爆准入、公开证明的美国防务资格,或高度成熟的机队软件,公司则较弱。国际上,新加坡和 GITEX 显示出出口意图,但渠道力量看起来仍比 ANYbotics 所称的区域企业覆盖、或 Ghost Robotics 所称的政府基础更早期。[CP003, CP007, CP008, CP020, CP021, CP024]
| 护城河主张 | 威胁 | 威胁为何可信 | 严重性 | 缓释 / 尽调问题 |
|---|---|---|---|---|
| 公用事业部署 know-how | Unitree 和其他中国硬件挑战者追上现场验证 | Deep 公开护城河是部署,不是封闭标准 | 高 | 索取重复部署、续约数据和场站级工作流粘性 |
| 通过 Lynx 做轮足适配 | 某些通道里,纯四足经济性仍偏弱 | Deep 自己也在释放信号:一种运动形态不适合所有场站 | 中 | 验证哪些任务真正需要腿,而不是轮足混合形态 |
| 国际扩张可选性 | ANYbotics 以更强危险场所资质进入亚洲和中东 | Yokogawa 明确把亚洲和中东列为 2026 年增长重点 | 高 | 按垂直行业和区域测试管线重叠,尤其是 GCC 能源 |
| 中国制造规模 | 品类商品化压缩利润率 | Gasgoo 认为,仅靠制造实力已不再构成稳固护城河 | 高 | 要求提供毛利率桥和制造缺陷 / 正常运行时间数据 |
| 巡检工作流采用 | 无人机、固定传感器和人工巡检组成的替代栈仍然够用 | 混合方法巡检在窄任务上可能胜过任何单一机器人 | 中 | 建模任务级挂载率,不要假设机器人完全替代 |
| 进入受保护采购池 | 美国国防和国土渠道偏向本土供应商 | Ghost 展示了一条 Deep 结构性难以进入的大赛道 | 中 | 承销时把商业公用事业 TAM 与受保护国防 TAM 分开 |
严重性反映承销相关性,不代表损失确定性。该表聚焦未来一个刷新周期内可能改变定价权、渠道准入或护城河耐久度的风险。
[CP021, CP025, CP031, CP035, CP037, CP038]针对最可能决定下一轮刷新周期竞争耐久性的维度,为 Deep Robotics 做启发式评分卡。
分数是依据保留证据得出的 0-10 编辑启发式判断,不是外部评级。
[CP005, CP007, CP020, CP021, CP031, CP035]3.4 护城河持久性和反向证据
反向证据并不是说 Deep Robotics 弱,而是说这个品类越来越难防守。Gasgoo 2026 年 1 月对 MirrorMe Apollo 发布的报道,是最重要的反证信号,因为它认为,稳定制造和单点创新本身已不再构成持久护城河;真正重要的是迭代与工业验证之间的反馈循环。这一点与 Deep 相关,因为它今天最好的差异化恰恰是公用事业里的部署反馈循环,而不是公开可见的认证、软件或采购垄断。Unitree 可以从下方用灵活硬件和低端可及性施压,ANYbotics 可以从上方用危险区域信任和全球能源参考案例施压,Boston Dynamics 也能继续定义软件驱动巡检的高端标杆。同时,Ghost Robotics 显示,美国国家安全采购这一大需求池,无论技术优劣,可能仍相对不向中国来源供应商开放。因此,Deep 的轮足 Lynx 应对在战略上是正面的:它说明公司在适应现场现实,而不是过度押注单一运动形态。尽管如此,未解决事项对尽调很重要:当前公开定价、装机基础透明度、ATEX/IECEx 状态,以及公司的海外活动有多少是可重复的渠道增长,多少只是展示驱动的市场进入。[CP015, CP021, CP025, CP028, CP031, CP034]
04财务情况
4.1 收入模式和定价信号现在更像工业化、产品化业务,但实际定价仍不透明
公开证据现在显示,公司已有真实收入引擎,而不是概念阶段机器人故事。通过 Sina、CnTechPost 和 Humanoids Daily 审阅的招股书摘要都指向同一幅 2025 年图像:Deep Robotics 实现人民币 3.37 亿元收入,具身机器人贡献了核心业务收入的 95% 以上,产品组合由工业绝影 X 系列、较低价的 Lite 线,以及新推出的 Lynx M20 轮足机器人带动。因此,公司的变现面看起来以硬件优先、企业客户驱动为主,部件和服务存在,但经济重要性低于机器人本体销售。这一点很关键,因为承销判断应从出货量、ASP 和各产品家族毛利率开始,而不是从经常性 SaaS 指标开始。 定价只以局部形式可见。最清晰的公开信号,是 2025 年招股书中的 ASP:X 系列约人民币 28.75 万元,Lite 约人民币 2.77 万元,Lynx 约人民币 19.76 万元,单台 DR 人形机器人销售为人民币 82.3 万元。管理层和第三方报道也把 Deep Robotics 描述为显著低于 Boston Dynamics 和部分韩国同行,但这些是定位主张,不是实际合同经济。公开记录仍未披露折扣表、软件附加率、维护定价,或项目收入中服务与机器人硬件各占多少。[CI001, CI003, CI004, CI005, CI006, CI007]
| 收入流 | 机制 | 单位 | 当前数值 / 状态 | 质量 | 尽调问题 |
|---|---|---|---|---|---|
| Jueying X 系列 | 面向电力、巡检、应急和安防工作流销售企业级四足硬件 | 机器人台数 | 2025 年 681 台,收入 RMB 196 million;ASP 约 RMB 287.5k | 产品级 ASP 和销量质量高 | 拆分直接硬件收入与打包软件、集成和维护。 |
| Jueying Lite 系列 | 面向教育、研究和较轻商业场景销售低价四足机器人 | 机器人台数 | 2025 年 1,850 台,收入 RMB 51.3 million;ASP 约 RMB 27.7k | 产品级 ASP 和销量置信度高 | 提供终端客户在高校、研究实验室和商业买家之间的构成。 |
| Lynx M20 系列 | 面向狭窄或崎岖工业环境的轮足硬件销售 | 机器人台数 | 2025 年售出 377 台,收入 RMB 74.5 million;ASP 约 RMB 197.6k | 中,因 2025 年为上市首年 | 提供 Lynx 合同绑定的在手订单、复购率和部署服务。 |
| DR 人形系列 | 早期工业人形机器人硬件销售 | 机器人台数 | 2025 年披露 1 台,金额 RMB 823k;商业化仍属探索 | 低,因样本量只有一笔销售 | 提供已签管线、试点转量产转化率,以及单次部署支持成本。 |
| 组件和配套服务 | 组件和配套服务等非本体收入 | 项目或服务包 | 招股书摘要中有出现,但经济权重低于具身机器人收入 | 低,因公开拆分不完整 | 量化服务收入占比、维护附着率和经常性支持毛利率。 |
招股书摘要披露了较强的产品线 ASP,但没有披露实际折扣,也没有拆出机器人本体、软件、集成和服务收入的精确构成。
[CI003, CI004, CI005, CI006, CI007, CI013]| SKU 或信号 | 价格 / 单位 / 合同 | 标价与实际价格 | 折扣 / 未知项 | 含义 |
|---|---|---|---|---|
| Jueying X 系列 | 2025 年单台 ASP 约 RMB 287.5k | 来自申报摘要的实际 ASP,不是公开标价卡 | 折扣、传感器套装定价或服务增量未公开 | 企业巡检机器人已经能承接高客单价工业预算。 |
| Jueying Lite 系列 | 2025 年单台 ASP 约 RMB 27.7k | 来自申报摘要的实际 ASP | 区域定价或渠道毛利未公开 | 低价 SKU 拉开销量,但可能压低混合单台毛利。 |
| Lynx M20 | 2025 年单台 ASP 约 RMB 197.6k | 上市首年申报摘要披露的实际 ASP | 预订价格、支持费和部署收费未公开 | 轮足产品不必等人形机器人成熟,也能把平均合同额抬到 Lite 之上。 |
| DR 人形机器人 | 2025 年一笔销售,金额 RMB 823k | 单笔披露交易观察到的实际价格 | 复购性、自主能力包或服务经济性没有证据 | 人形机器人定价目前只是一个证明点,还不是可承销的收入流。 |
| 相对市场定位 | 管理层称,同等性能机器人价格为 Boston Dynamics 的三分之一到二分之一;韩国销售称比 Boston Dynamics 和 Rainbow Robotics 便宜 20-30% | 相对定位主张,不是客户合同价格表 | 没有按竞争对手调整后的 BOM、质保或折扣数据 | 高端工业定价眼下看似守得住,但相对价格仍需合同级证据。 |
Deep Robotics 没有发布标准化企业价目表,本章只能依靠实际 ASP、管理层定位和第三方价格对比。
[CI004, CI005, CI006, CI007, CI023, CI024]Deep Robotics 先把工业巡检和高危作业场景转化为机器人本体收入;服务和软件在公开记录中仍大多未披露。
这座桥接图只做定性判断,因为公开证据披露的是产品系列 ASP 和利润率点位,而不是从硬件到服务、维护再到净留存毛利的完整收入瀑布。
[CI003, CI004, CI005, CI006, CI007, CI013]4.2 GTM 动作围绕危险巡检、由解决方案驱动,ROI 代理指标强,但单位经济缺口很大
已审阅来源一致把 Deep Robotics 描述为企业解决方案卖方,而不是消费机器人厂商。官方页面强调工业巡检、应急响应和危险环境部署;管理层访谈把电力巡检称为切入口,因为中国约有 20,000 座变电站,巡检工作重复且规则明确;Singapore Power 这类海外证据点也显示,同样模式是在关键任务基础设施中做定制化部署。这在财务上重要,因为公司似乎靠运营可靠性、环境韧性和部署支持赢得业务,而不是靠低摩擦的自助下单。 公开 ROI 信号强于许多机器人初创公司,但仍不完整。官方案例研究声称巡检准确率 98%,全站巡检不到 35 分钟、人工超过一小时,工作量减少 50% 以上,并在某个变电站场景下降低约 70% 运营成本。Singapore Power 预计每年节省超过 480 个工时。李超访谈还给出一个有用的商业测试:客户买一次只说明产品可用;如果带来 100、300 或 500 台复购,才算「好用」。即便如此,公开记录仍未给出 CAC、回本周期、流失率、NRR、服务附加、保修负担或安装毛利,因此收入质量判断是方向性正面,而不是已完全承销。[CI014, CI016, CI017, CI018, CI019, CI020]
| 指标 | 数值或状态 | 置信度 | 重要性 | 尽调要求 |
|---|---|---|---|---|
| 综合毛利率 | 2025 年 52.83% | 高 | 说明工业硬件业务已经能支撑有吸引力的单台经济性。 | 按硬件、软件、服务和地区对齐毛利率。 |
| X 系列产品毛利率 | 2025 年 54.35% | 高 | 说明旗舰工业 SKU 并未以明显亏损引流价出售。 | 提供质保准备、现场服务负担和传感器套装构成。 |
| Lite 系列产品毛利率 | 2025 年 38.52% | 高 | 低成本 SKU 可能拉动销量,同时稀释混合毛利。 | 提供教育买家与商业买家的构成以及渠道毛利。 |
| Lynx M20 产品毛利率 | 2025 年 57.50% | 中 | 如果上市首年定价能维持,新轮足产品有望具备吸引力。 | 说明 2025 年是否包含上市补贴或异常有利的产品组合。 |
| 变电站部署 ROI 代理指标 | 一个官方案例显示,人工工作量削减 >50%,运营成本约下降 70% | 中 | 客户 ROI 是判断回本和复购的最佳公开代理指标。 | 提供已签署的回本模型和实际合同续约数据。 |
| 巡检效率代理指标 | 全站巡检不到 35 分钟,人工超过 1 小时;准确率 98% | 中 | 效率和准确率提升可支撑危险场景中的溢价定价。 | 提供部署前后工时、避免停机时间和缺陷预防经济性。 |
| 海外劳动力套利信号 | 新加坡部署预计每年节省 >480 个人工小时 | 中 | 劳动力成本高的市场具备国际付费意愿。 | 提供分国家 ROI 和本地支持成本结构。 |
| CAC、回本期、NRR 和流失率 | 未公开披露 | 低 | 这些指标决定毛利能否转化为高效增长。 | 按细分市场提交 cohort 数据、续约曲线和销售效率指标。 |
公开证据支持产品毛利率和客户结果代理指标,但不足以支撑完整的 SaaS 式或项目制单位经济性。Null 是真实尽调缺口,不是零。
[CI008, CI017, CI019, CI020, CI021, CI051]公开记录把客户痛点和可复现的 ROI 信号连了起来,但还没有给出完整增长效率承保所需的 CAC 与回本指标。
该图只使用公开流程和 ROI 证据;最后一个节点刻意留缺,因为经审计的销售效率指标并未公开。
[CI016, CI017, CI018, CI019, CI020, CI021]4.3 2025 年盈利是真实的,但人形机器人扩张和制造规模仍让融资依赖成为核心
相比大多数未上市机器人公司,Deep Robotics 进入本章时已经有更好的公开财务证据。2025 年申报摘要显示,公司首次盈利,毛利率为 52.83%,海外收入贡献也有分量。近期私募融资看起来同样可观:2025 年 7 月约人民币 5 亿元,2025 年 12 月超过人民币 5 亿元,以及 2026 年初据报道单独完成的数亿元 pre-IPO 轮。截至运行日,公司仍未上市,但科创板申请已获受理,计划 IPO 将募集约人民币 25.03 亿元,用于具身 AI 模型、新机器人本体、产业化和基地建设。 承销问题不是 Deep Robotics 能否融资,而是当前利润率能否在没有新资本、也没有有利产品组合的情况下持续。公开报道反复把下一阶段增长与人形机器人研发、制造扩张和持续招人绑定。基于招股书的报道称,2025 年研发已消耗人民币 8430 万元,约占收入 25%;公司自身也提示,快速技术迭代可能导致盈利波动。由于手头现金、烧钱速度、债务和现金跑道仍未披露,更稳妥的结论是,2025 年利润证明了工业四足机器人的可行性,但并没有消除未来融资依赖。[CI002, CI008, CI009, CI010, CI026, CI027]
| 项目 | 公开状态 | 证据 | 承销解读 | 尽调要求 |
|---|---|---|---|---|
| 2025 年盈利 | 2024 年净亏损后,2025 年净利润 RMB 28.68 million | Sina、CnTechPost 和 Humanoids Daily 的招股书摘要 | 确认工业四足业务在 IPO 前已经跨过盈利线。 | 提供月度利润桥和非经常性项目。 |
| 2025 年 7 月 Pre-Series C | 募资约 RMB 500 million | Yicai 及管理层访谈 | 申报前已有大额私募资本到位,可能用于产线扩张和招聘。 | 提供交割日期、工具条款和投后估值。 |
| 2025 年 12 月 Series C | 募资超过 RMB 500 million | Yicai 关于电信相关战略投资方的报道 | 战略资本拓宽工业渠道,但也可能提高政策相关预期。 | 提供清算条款、董事会权利和任何商业附带协议。 |
| 2026 年初 Pre-IPO 融资 | 据报道募资数亿元人民币 | 36Kr / Tian Tian IPO 报道 | Series C 之后,公司似乎仍寻求过桥资本。 | 确认准确金额、投资方名单,以及资金是否仍不受限制。 |
| 计划中的科创板 IPO | 计划募资 RMB 2.503 billion;申请于 2026 年 5 月获受理 | 申报页面及多份招股书摘要 | IPO 是可见的最大融资环节,也是下一阶段 capex / R&D 的核心。 | 如果市场窗口滑动,提供预计时间表、备选融资计划和最低现金阈值。 |
| 募资用途 | 具身算法 / 模型 R&D、机器人本体 / 解决方案 R&D、产业化和制造 / 基地建设 | 招股书摘要 | 资金投向长期能力建设,而不只是销售扩张。 | 按项目提供阶段门、里程碑预算和盈亏平衡预期。 |
| 现金、烧钱、跑道、债务和项目融资 | 未公开披露 | 已审阅来源均未给出现余额、月度烧钱或杠杆安排 | 这是得出清晰资本充足性结论的主要障碍。 | 提交当前现金、12 个月烧钱计划、债务契约和下行情形流动性案例。 |
历史融资轮次时间线放在公司概览;本表聚焦从最新私募轮到仍待完成的 2026 年 IPO 之间,资本是否足够。
[CI002, CI026, CI027, CI028, CI029, CI030]可见融资范围已经覆盖多轮规模不小的私募融资和一笔拟议的大额 IPO,但这些融资之间的现金余额仍未披露。
固定点数值用低值等于高值呈现,因为公开记录给出的是披露金额,不是模型区间;这些数字之间未披露的现金桥接,是关键缺失输入。
[CI001, CI010, CI026, CI027, CI031]Deep Robotics 已经证明工业四足机器人可以赚钱,但下一阶段的大部分资本压力落在人形机器人研发和产能扩张上,而不是新增销售人手。
该矩阵只做定性判断,因为公开来源披露了融资意图和产品经济性,但没有完整现金流量表或月度流动性情况。
[CI008, CI031, CI032, CI037, CI040, CI046]4.4 财务结论:商业上可信的工业机器人生意,但还不是完整、可尽调闭环的案例
正面逻辑对一家未上市机器人公司来说异常具体。Deep Robotics 现在有招股书支撑的收入、产品级 ASP 披露、改善中的毛利率、首次盈利、可见海外收入,以及真实基础设施场景里的客户 ROI 故事。这些信号不小:它们把公司和那些仍几乎完全依赖演示、科研销售或叙事融资的机器人厂商区分开。 负面逻辑在于,剩余盲点恰好决定 2025 年盈利能否复利。公开来源没有披露现金余额、月度烧钱速度、债务额度、营运资本压力、backlog 质量、按收入划分的客户集中度、实际折扣,或服务与保修经济性。反向市场来源还提醒,人形机器人竞赛正在变成资金储备之战;管理层自身也强调价格压力和人才挖角。因此,本章结论是平衡的:Deep Robotics 今天看起来像一家拥有可防守需求的真实工业硬件公司,但前瞻承销仍取决于流动性、集中度,以及利润率能否撑过从四足机器人转向更重资本人形项目的私人证据。[CI002, CI015, CI037, CI040, CI041, CI042]
| 缺失的私有指标 | 影响 | 精确尽调路径 |
|---|---|---|
| 按月现金余额 | 如果 IPO 时间滑动,无法检验流动性。 | 要求提供 2025 年 1 月至最新月份的月度现金余额,并附银行对账单。 |
| 月度烧钱和跑道 | 盈利仍可能掩盖重 capex 或营运资金流出。 | 要求提供月度经营现金流、capex 和情景跑道模型。 |
| 债务、担保和项目融资义务 | 隐性杠杆会削弱表面上的毛利叙事。 | 要求提供债务明细、担保、质押资产和契约包。 |
| 按收入计的客户集中度 | 国家电网集中度可能让收入质量不如表面增长那样耐久。 | 要求提供前 10 大客户、收入占比、续约日期和公共部门敞口拆分。 |
| 实际折扣和投标定价 | 如果大客户拿到大幅让利,ASP 可能高估可持续定价。 | 要求提供投标中标表、折扣区间和按客户 cohort 划分的毛利率。 |
| 服务和维护附着率 | 只看硬件毛利,无法判断生命周期价值或现场服务负担。 | 要求提供年度维护收入、备件毛利和质保准备历史。 |
| 在手订单和管线转化 | 有已签在手订单可见度,出货增长才更容易承销。 | 要求提供已预订 backlog、管线阶段和取消历史。 |
| 人形机器人试点经济性 | 一笔 DR02 销售不能证明商业可复制。 | 要求提供试点经济性、部署成本和人形机器人的目标毛利爬坡。 |
要把积极的商业叙事变成经完整尽调的财务承销案例,至少需要补齐这些私有数据。
[CI025, CI037, CI045]4.5 图表
05产品与技术
5.1 产品线真实存在、按工作流分层,简报里的「L20」标签与公开产品面不匹配
公开产品面已经足够宽,可以把 Deep Robotics 按平台公司承销,而不只是单一四足机器人厂商。官方页面和 2025 年末发布显示出四条清晰赛道:面向工业四足巡检的 X 系列、面向更狭窄或更复杂地形轮足作业的 Lynx 系列、面向教育和开发者使用的 Lite3 家族,以及面向工业具身 AI 扩张的 DR 人形机器人线。在这套产品栈中,X20 仍是更轻量的工业四足机器人,X30 是更坚固的旗舰,Lynx M20 是更重型的工业轮足机器人,Lynx S10 是面向狭窄空间的小型轮足机器人,Lite3 是开放科研平台,DR01 是展会阶段的人形机器人演示机,DR02 则是第一个看起来像商业人形机器人方案的产品页面。 一个简报假设经不起来源审阅:没有任何已审阅的官方页面、FAQ、GitHub 页面或第三方报道提到「L20」型号。反复出现的名称是 X20 和 Lynx M20,因此最可辩护的解释是简报指的是其中之一。工作流分割也有意义。公用事业和隧道运营商购买 X 系列或 Lynx 机器人用于巡检和缺陷检测;应急团队购买的是地形进入能力加气体、热成像和影像载荷搭载;研究人员购买 Lite3,因为它开放 ROS、SDK 和 sim-to-real 工具。这种分层让产品组合比泛化的「机器狗」故事更容易映射到真实预算和部署动作。[CE001, CE002, CE003, CE004, CE005, CE006]
| 产品 | 买家或用户 | 工作流角色 | 成熟度或状态 | 差异化 | 尽调缺口 |
|---|---|---|---|---|---|
| X20 | 公用事业、安防、工业 O&M 团队 | 危险但可重复路线中的核心四足巡检 | 产品页仍在;已成型的工业 SKU | IP66、模块化载荷接口、自主巡检定位 | 与 Lite3 和 M20 相比,公开 SDK 和车队管理细节偏少。 |
| X30 | 公用事业、隧道运营方、应急和安防团队 | 面向更恶劣地形和黑暗环境的坚固旗舰四足机器人 | 旗舰产品页仍在,并有具名部署 | IP67、暗环境感知、45° 爬楼梯、快速换电 | 公开软件栈停留在功能层描述,没有部署手册。 |
| Lynx M20 | 公用事业、救援团队、狭窄或崎岖地形中的工业运营方 | 轮足自主巡检和响应平台 | 2025 年推出,案例研究支持在增加 | 33 kg、单步越障 80 cm、双 96 线 LiDAR、OTA 和可选充电 | 需要来自生产账号的车队在线率、故障率和服务附着证据。 |
| Lynx S10 | 轻量巡检、巡逻、救援和教育用户 | 面向狭窄空间的小型轮足进入机器人 | 较新的公开 SKU,有规格页,但部署证据较少 | ≤20 kg、最高速度 8 m/s、越障 50 cm、IP66 | 已审阅材料中尚未公开具名客户背书。 |
| Lite3 系列 | 高校、实验室、开发者、应用研究团队 | 开放开发和算法实验平台 | 研究端界面成熟,有多个版本和公开工具 | ROS1/ROS2、SDK/API 访问、基于 Jetson 的感知选项、RL 工作流 | 工业部署证据弱于 X30 和 Lynx M20。 |
| DR01 | R&D 和展示受众 | 用于展示具身 AI 能力的人形机器人样机 | 首秀阶段的展会平台 | J60 和 J100 自研关节、楼梯和平衡演示 | 未审阅到公开部署、定价或支持包。 |
| DR02 | 工业巡逻、工厂和户外作业买家 | 公开端首个工业化人形机器人产品 | 产品页已上线;商业化仍早 | IP66、-20°C 至 55°C、模块化快拆四肢、全天候定位 | 公开证据仍缺少具名生产部署和公开 SDK 材料。 |
矩阵只反映截至 runDate 已审阅的公开界面;简报中的 “L20” 标签按错误处理,因为官方材料展示的是 X20 和 Lynx M20。
[CE002, CE003, CE004, CE005, CE006, CE007]| 用户任务 | 当前工作流 | 公司方案 | 可量化收益 | 限制 |
|---|---|---|---|---|
| 特高压和变电站巡检 | 人工巡检在危险地形中检查仪表、开关、热状态和局部放电信号 | X30 和 Lynx M20 搭载摄像头、热成像巡检、LiDAR 和自主导航,进入难抵达路线 | 官方案例提到检测准确率 >95%、识别准确率 96.5%、O&M 成本降低约 70%、效率提升 >50% | 指标来自公司主张,站点级成本基线未公开。 |
| 地下隧道巡检 | 工人步行穿过长距离电缆隧道,发现裂缝、渗水、泄漏和热异常 | X30 SPock 在 Singapore Power 隧道中巡线、捕捉缺陷并生成报告 | SP Group 预计扩大部署,每年节省约 480 个人工小时 | 公开证据未量化机器人停机、误报或维护人工。 |
| 安防巡逻 | 保安巡逻大型园区或市场综合体,视野不完整,响应速度不稳定 | 机器狗增加扩音、识别、数据采集、跟随和连续路线巡逻 | 西宁警方和义乌市场案例展示 24/7 巡逻定位和风险检测 | 采购、隐私和公共空间运营规则仍记录较少。 |
| 应急救援和消防 | 团队在不稳定区域需要地形进入能力、气体 / 热感知,并降低人员暴露 | Lynx M20 和 X30 结合机动性、双光成像、气体传感器,有时还协同无人机 | 官方演练证据显示空地协同和远程危险感知 | 相比公开复盘绩效数据,营销更强调场景广度。 |
| 教育和开发者研究 | 实验室需要可用 SDK 的四足机器人,用于运动、感知和 sim-to-real 实验 | Lite3 版本开放 ROS、SDK、运动 / 感知开发和 RL 部署资源 | 多份官方文档和 GitHub 仓库显示真实的二次开发界面 | 这是公开工具链最深的一条线,但不等同于大规模工业车队证据。 |
收益指标混合了官方案例数字和一项独立报道的新加坡节省人工数据;读者不应假设所有工作流都有相同 ROI。
[CE008, CE011, CE012, CE013, CE014, CE015]核心商业动作从危险巡检或响应路线切入,先证明自主数据采集可行;只有集成和支持顶得住,才会继续扩张。
[CE003, CE004, CE008, CE011, CE012, CE013]5.2 核心架构是从感知到执行的全栈系统,最具体的公开证据在开发者栈
Deep Robotics 最可信的技术主张,不是某个单一速度或载荷数字,而是公司通过产品页面、GitHub 仓库和 2025 年后续公告展现出的栈一致性。公司介绍和 Series C 发布使用同一套「感知—决策—行动」框架,DeepVLA 1.0 则加入更明确的软件层:开放词汇语义点云、大语言模型命令解析,以及用于多楼层门到门导航的拓扑路网。硬件侧,公开页面呈现出一致模式:多模态感知、机载计算、模块化载荷安装和加固型运动能力。Lynx M20 是最清楚的例子,因为其官方页面把双工业处理器、双 96 线 LiDAR、双广角相机、Ethernet、热插拔电池和 OTA 升级放在同一个 SKU 中发布。 开发者表面强于许多未上市机器人公司。DeepRoboticsLab 发布 MJCF/URDF/USD 机器人模型、基于 IsaacLab 的 Lite3 和 M20 强化学习环境、ROS2 部署 SDK,以及 J60 模块的 CAN 级关节控制。美国站上的 Lite3 文档走得更远,描述了 ROS1/ROS2 支持、Motion SDK 访问、LiDAR SLAM 示例、基于 Jetson 的感知主机,以及 sim-to-real 强化学习工作流。这对尽调很重要,因为它把产品从封闭设备变成客户、大学和集成商可以扩展的平台。限制在于范围:开放生态在 Lite3 和较新平台上最深,X20/X30/DR02 则更多停留在营销页面可见,而不是 SDK 可见。[CE011, CE012, CE013, CE014, CE015, CE016]
| 层级或组件 | 角色 | 公开证据 | 关键依赖 | 主要风险 |
|---|---|---|---|---|
| 机动与机械 | 在轮式机器人和人类难以进入的地形中,足式和轮足运动提供进入能力 | X20、X30、Lynx M20、Lynx S10、DR01 和 DR02 页面都强调楼梯、斜坡和障碍处理 | 关节功率密度、步态算法和坚固底盘耐久性 | 真实现场耐久性可能偏离实验室规格表。 |
| 感知栈 | 摄像头、LiDAR、气体传感器和热成像载荷生成巡检与救援所需的场景理解 | X30 暗环境感知、Lynx M20 双 96 线 LiDAR,以及 Lite3 感知手册均已公开 | 传感器校准、环境鲁棒性和载荷集成 | 公司案例之外,公开基准测试很薄。 |
| 机载计算 | 嵌入式处理器在机器人近端运行导航、感知和控制 | Lynx M20 公布双工业处理器;Lite3 高配版本公布 RK3588 加 Jetson Xavier NX | 热预算、功耗和软件优化 | 公开文档没有展示车队级故障模式或计算利用率。 |
| 自主和导航软件 | 规划、避障、地形识别、语义导航和报告生成,把硬件转化为工作流价值 | 公司页面描述感知-决策-行动架构和 DeepVLA 1.0 语义导航 | 高质量建图、状态估计和站点级调优 | DeepVLA 的生产表现仍是公司主张,尚无独立基准。 |
| 开发者和集成层 | SDK、URDF/MJCF/USD 模型、ROS 桥和 RL 环境让用户扩展机器人或搭建内部工具 | GitHub 暴露 rl_training、sdk_deploy、deep_robotics_model 和 Deep_Motor_SDK | 仓库维护、文档质量和按 SKU 的模型覆盖 | Lite3 和 M20 的公开工具链远深于完整商用车队。 |
| 服务和生命周期层 | 热插拔电池、可选自主充电、模块化快拆四肢和远程 / OTA 更新支撑长期部署 | Lynx M20 和 DR02 页面明确把这些功能定位为运营使能项 | 备件、支持地域和受训现场服务覆盖 | 仅限中国大陆的售后条款和较宽的质保排除,约束外部信任。 |
架构各行区分了已明确记录的内容和仍需客户尽调的内容;本次审阅的公开来源没有披露完整车队管理或云端控制手册。
[CE011, CE016, CE017, CE018, CE019, CE020]Deep Robotics 的公开技术栈从面向具体流程的机器人本体出发,向上延伸到感知、导航软件、开发者工具和全生命周期运营。
该技术栈由产品页面、GitHub 文档和 Series C 新闻稿拼出;公开来源没有披露完整内部云架构或闭源车队管理器。
[CE010, CE016, CE017, CE018, CE021, CE022]公开成熟度最高的是四足巡检,最低的是人形机器人商业化;开发者开放度则集中在 Lite3 和 M20。
[CE006, CE007, CE008, CE022, CE037, CE038]5.3 部署证据在公用事业巡检中最强,但信任和支持控制落后于运动能力叙事
当工作流狭窄、重复且危险时,产品最容易被相信。公用事业和隧道巡检符合这个模式,已审阅案例显示 Deep Robotics 反复围绕地形进入、数据采集和减少人员暴露来销售。对一家未上市机器人厂商来说,变电站证据异常具体:官方页面声称覆盖近 30 座 China Southern Power Grid 变电站,成本降低约 70%,效率提升超过 50%,巡检检测准确率超过 95%,并在宁夏沙漠部署中达到 96.5% 识别准确率。Singapore Power 的 SPock 部署给出第二个、更国际化的工作流证据点:X30 巡逻一条 40 km 电力隧道,检测渗漏和热异常,并触发了更大规模部署计划。 信任和质量信号真实存在,但不均匀。硬件加固很具体,工业线具有 IP66 或 IP67 等级,DR02 的模块化快拆设计也被明确定位为运维特性。与此同时,除防护等级和可靠性营销主张外,公开合规证据很薄。支持和隐私页面制造了真实尽调旗帜:售后条款仅面向中国大陆,恶劣环境和未授权改装的保修排除很宽,隐私政策允许收集日志、时间、位置、设备配置和 IP 数据,并在法律要求时向主管机关披露。换句话说,运动栈看起来比公开信任、安全和全球支持包走得更远。[CE027, CE028, CE029, CE030, CE031, CE032]
| 控制或信号 | 状态 | 范围 | 证据 | 缺口或风险 |
|---|---|---|---|---|
| 防护和环境等级 | 已公开记录 | X20 IP66、X30 IP67、Lynx S10 IP66、Lynx M20 IP66、DR02 IP66,高端型号覆盖较宽的 -20°C 至 55°C 区间 | 官方产品页和产品文章 | 评级有帮助,但公开材料没有展示认证机构和测试报告。 |
| 模块化可维护性 | DR02 有公开文档,Lynx M20 仅部分公开 | DR02 采用快拆四肢;Lynx M20 支持热插拔电池和可选自主充电 | 产品页和公用事业案例研究 | 未公开 MTBF、备件 SLA 或现场服务指标。 |
| 可靠性声明 | 证据混合 | Lynx S10 称机身通过严格可靠性测试;公用事业案例声称检测准确率 >95% | 产品页及官方案例研究 | 未公布测试协议、样本量和审计方法。 |
| 数据和隐私处理 | 有公开文档,但口径较宽 | 可收集日志、时间、位置、设备配置和 IP 地址,并可在法律要求下披露 | 官方隐私条款 | 跨境基础设施买家很可能会要求更严格的合同控制。 |
| 保修和运行边界 | 有公开文档,且限制较多 | 保修排除自行改装、非官方维修、强干扰、水 / 碎屑进入、化学损伤,以及部分极端使用场景 | 官方支持条款和 Lite3 免责声明页 | 营销用例延伸到危险场景,而条款对其中一部分作出免责。 |
| 全球服务覆盖 | 文档较弱 | 已审阅的售后政策明确称仅适用于中国大陆 | 官方支持条款 | 海外机队支持、备件库存和升级路径仍需尽调。 |
本表只看公开的运行控制证据;未发布 ISO 式证书或网络安全证明,被视为证据缺口,而非确认的负面事实。
[CE027, CE028, CE029, CE030, CE031, CE032]只有感知、软件、支持和治理依赖一起打通,Deep Robotics 才能把运动性能转成持久价值。
[CE017, CE021, CE022, CE031, CE032, CE033]5.4 从四足机器人走向具身导航和人形机器人的路线图可信,但核心巡检之外的商业化证据很快变薄
官方表面的 2025 年路线图显示的是真实出货进展,而不是模糊的未来承诺。Deep Robotics 在 4 月发布 Lynx M20,年中扩展公用事业和救援案例研究,10 月发布 DR02 作为工业人形机器人,并在 12 月 Series C 公告中把产品、DeepVLA 1.0 和具身 AI 试产基地串成一个扩张叙事。同一公告还称,新资本将投入更多研发、更多人才,以及持续的四足机器人加人形机器人开发,这意味着公司并不把人形机器人视为副业。 主要技术风险不是机器人能否移动,而是 Deep Robotics 能否把强运动和感知基础转化成可持续、可支持、可治理,并能跨多个地区和产品家族量产的系统。在这方面,DR02 仍处早期:公开证据集中在产品页面和展会演示,而不是具名生产部署或公开 SDK。DeepVLA 1.0 具有战略重要性,但已审阅语料仍主要把它呈现为公司声称的使能层,而不是公开基准测试过的现场系统。并且,虽然公司显然在思考机器狗、无人机和固定传感器的集群作业,开源和公开文档对 Lite3 和 M20 的支撑明显强于整个商用机队。因此,技术论点是正面的,但尽调至少应像审运动能力一样,严格承销可支持性、数据治理和部署可重复性。[CE017, CE031, CE035, CE036, CE037, CE038]
| 日期或阶段 | 功能或里程碑 | 状态 | 含义 | 来源信号 |
|---|---|---|---|---|
| 2024 World Robot Conference | DR01 人形机器人首秀,搭载自研 J60 和 J100 关节 | 已完成演示里程碑 | 说明公司在公开工业产品落地前,已从四足机器人推进到具身 AI 叙事 | 官方转载及 SCMP 报道 |
| April 2025 | Lynx M20 发布 | 已发布 | 新增轮足工业品类,而不只是刷新 X 系列参数 | 官方发布文章和产品页 |
| Mid-2025 | 围绕 Lynx M20 和 X30 扩展公用事业、救援场景 | 已在市场 | 说明公司在 UHV、救援和安防流程中复用同一移动平台 | 官方应用文章 |
| October 2025 | DR02 工业人形机器人及具身 AI 中试基地 | 已公开宣布 | 表明人形机器人正从展会概念进入工业产品路线图 | 官方 Series C 摘要和 DR02 页面 |
| 2025 年 12 月以后 | DeepVLA 1.0、更多研发招聘,以及继续投入四足机器人和人形机器人 | 公司表述的前瞻路线图 | 暗示软件和人形机器人层是战略重点,但公开部署指标仍然偏薄 | 官方 Series C 摘要 |
各行混合了已出货产品和公司前瞻表述;在客户侧部署指标公开之前,DeepVLA 和人形机器人放量应按路线图方向理解。
[CE007, CE017, CE035, CE036, CE038, CE039]5.5 图表
06客户情况
6.1 客户分层真实存在,但各垂直领域的证据质量差异很大
Deep Robotics 并不是卖给单一、同质的买方类别。公开记录至少支持四个实际客户桶。第一,公用事业和电网运营商购买四足和轮足机器人,用于隧道、变电站、电缆廊和危险巡检;这是公司具名证据最具体、运营结果最清楚的地方。第二,公共安全和安防买家使用机器人进行巡逻、火灾隐患检测、周界监控和事件响应,具名参考包括 Xining Public Security Bureau、Yiwu International Trade City 和 Singapore 的 Certis industrial park。第三,应急响应组织和消防队出现在演练和场景验证中,这证明了恶劣环境下的产品—市场匹配,但不一定证明经常性收入。第四,大学和研究实验室把 Lite3 及相关开放平台用于教育和研发工具。关键尽调区分不在于这些细分是否存在,而在于哪些有独立佐证的运营部署证据,哪些只是公司精选案例、类似合作伙伴的展示,或 logo 与触达范围营销。[CU001, CU002, CU010, CU014, CU021, CU032]
| 客群 | 买方 / 用户 / 付款方 | 主要用例 | 最强公开证明 | 战略价值 | 关键限制 |
|---|---|---|---|---|---|
| 公用事业和电网运营商 | 公用事业或电网运营商付款;巡检和维护团队在隧道、变电站、电缆廊道使用机器人 | 危险巡检、缺陷检测、第一响应和数字化运维流程 | SP Group 的 SPock 部署,以及中国南方电网和国家电网相关巡检叙事 | 公开材料中质量最高的工业证明,ROI 逻辑也最清晰 | 未披露按公用事业客户划分的收入集中度 |
| 公共安全和安防运营商 | 公安局、市场运营方或产业园安防运营方付款;巡逻人员和控制中心使用机器人 | 巡逻、火灾隐患检测、预警和闭环干预 | 西宁市公安局、义乌国际商贸城,以及新加坡 Certis 产业园 | 说明平台不止能做巡检,也能支撑运营级巡逻流程 | 多数证据由公司发布或分发,而非客户自行撰写 |
| 应急响应和消防救援团队 | 消防队、应急主管部门或演练组织方可能是买方;救援指挥员和一线团队是最终用户 | 灾害现场侦察、有毒气体感知、空地协同,以及防汛或消防演练 | 浙江消防总队联合演练,以及 2022 年以来的消防应急解决方案材料 | 适合恶劣环境,传感器载荷也契合需求 | 公开证明主要是演练和场景验证,不是经常性合同披露 |
| 科研和教育买家 | 大学、实验室或研发团队付款;学生、研究人员和工程师使用平台 | 开放机器人研究、自主导航、强化学习和教学实验室 | Lite3 引用中点名清华、北大、Edinburgh 和 UCL | 分散客户基础,把销量扩展到高 ASP 公用事业之外 | 公开证据未披露续费、机队规模或各机构收入贡献 |
| 新兴国际场景买家 | 各站点客户身份不同;当地运营方或合作伙伴很可能付款,一线员工使用机器人 | 仓储安防、野生动物监测和极端环境巡检 | 北美仓储园区、可可西里保护和 NEOM 户外巡检系统 | 说明产品可以跨地域、跨流程类型落地 | 这些证明大多只有官方口径,合同经济性和独立验证都偏薄 |
分层把买方类型和流程拆开,因为公用事业、巡逻 / 安防运营商、应急组织和大学购买机器人,是为了解决不同任务。
[CU001, CU002, CU010, CU014, CU021, CU032]公用事业和安防客户似乎沿着一条高接触路径推进:先选问题,再做本地化和运营验证,之后才扩大部署。
[CU002, CU004, CU026, CU027, CU039, CU040]6.2 具名客户证据在公用事业中最强,Singapore Power 是锚点案例
支撑最好的客户证据是 Singapore Power Group。官方 SPock 案例研究和多篇独立行业文章讲的是同一条核心故事:X30 巡检 SP Group 地下输电隧道网络的部分区段,执行热成像和视觉巡检任务,需要现场特定定位工作;如果推广扩大,预计每年节省约 480 个人工巡检小时。China Southern Power Grid 是下一个最重要的具名账户,因为 Deep Robotics 称其机器人现在服务近 30 座变电站,并带来有意义的成本和效率改善,但这一规模仍由公司材料披露,而不是客户文件披露。更广的 State Grid 关系可信且越来越重要,尤其考虑到 2026 年公用事业采购计划和 Jiemian 关于此前框架协议的报道;但这些信号在证据质量上仍比 SP Group 低一层。公用事业之外,安防、保护、仓储和沙特户外巡检都有具名部署,但多数仍是官方单方证明,商业深度披露更少。[CU003, CU004, CU005, CU006, CU007, CU008]
| 指标或信号 | 数值 / 状态 | 日期 | 来源质量 | 含义 | 缺失分母 |
|---|---|---|---|---|---|
| SP Group 隧道部署 | 运营试点,并计划更大范围铺开 | 2024-2025 | 高 | Deep Robotics 从演示跨到海外关键任务基础设施使用的最佳证明 | 未披露合同金额或机队规模 |
| SP Group 节省人力声明 | 每年约节省 480 个小时的人工巡检 | 2025 | 高 | 给出了本章最清晰的公开客户结果指标 | 未披露前后成本基线或回本周期 |
| 中国南方电网运行规模 | 近 30 座变电站;声称运维成本降低约 70%,效率提升超过 50% | 2025 | 中 | 暗示国内公用事业用量已放大,而非单站点测试 | 仍是公司报告的部署和 ROI 数字 |
| 国家电网框架信号 | 界面报道称,2023 年框架交易覆盖 1,000+ 座变电站,并到 2025 年完成 600+ 个部署项目 | 2026 年报道回顾 2023-2025 年活动 | 中 | 指向除 SP Group 个案外更广的公用事业嵌入 | 二手报道;本次未审阅公开合同文本 |
| 公司披露的客户触达 | 中国 26 个省 / 市的 100+ 客户 | 2024 | 中 | 说明客户版图在中国境内已在拓宽 | 不是当前经审计客户数 |
| 公司披露的全球触达 | 44 个国家 / 地区和 600+ 个行业场景 | 2025-2026 | 中 | 表明地域和流程广度超过少数展示站点 | 场景数量不等于唯一付费账户数 |
| 2026 年公用事业采购背景 | 国家电网 2026 年 8,500 台机器人计划中,包含 5,000 台四足巡检机器人 | 2026 | 中 | 如果 Deep Robotics 保持竞争力,可服务的电网需求池会扩大 | 未披露 Deep Robotics 获得的份额 |
最强采用证明来自具名部署和已披露的流程结果,而不是公司报告的宽口径触达指标。
[CU003, CU004, CU005, CU006, CU016, CU017]| 客户或账户 | 客群 | 部署或用例 | 生产 vs 试点 | 结果或证明质量 | 限制 |
|---|---|---|---|---|---|
| Singapore Power Group (SP Group) 客户 | 公用事业基础设施 | X30 SPock 在地下输电网络中做隧道巡检 | 运营试点,并计划更大范围铺开 | 具名部署获得独立佐证,包含本地化细节和每年节省 480 小时人力的声明 | 未披露合同金额、机器人数量和续约条款 |
| China Southern Power Grid | 公用事业基础设施 | 四足机器人部署到近 30 座变电站,执行智能巡检 | 生产部署 | 官方规模和 ROI 声明指向有意义的国内使用 | 独立的变电站级佐证有限 |
| Xining Public Security Bureau | 公共安全 | 自主巡逻,带扩音、识别和数据采集 | 运营部署 | 官方 Series C 部署摘要中的具名政府用户 | 公开证明大多由公司发布 |
| Yiwu International Trade City | 安防运营 | X30 移动安防前哨,用于巡逻、火灾隐患检测和异常预警 | 运营部署 | 具体流程描述支持真实使用,而不是单纯 logo | 没有公开商业条款或客户证言 |
| Certis industrial park, Singapore 项目 | 安防运营 | 全天候巡逻、违规预警和充电桩监控 | 运营部署 | 新加坡闭环巡逻使用有清晰运营叙事 | 证据是公司分发的新闻稿,不是独立客户声音 |
| Zhejiang Fire Brigade | 应急响应 | X30 机器人与无人机联合救援演练 | 演练 / 验证,不是生产收入证明 | 平台能在恶劣救援条件下工作,证据较好 | 不能证明经常性采购或留存 |
| NEOM, Saudi Arabia | 极端环境巡检 | 沙滩条件下的全自动户外巡检系统 | 公司声称的运营部署 | 若属实,场景指向性很强 | 公开证明仅来自官方,本次未获独立验证 |
| Hoh Xil conservation project 项目 | 环境监测 | 高海拔藏羚羊观测使用的野生动物监测机器人 | 场景部署 | 说明机器人可支撑专门的一线场景 | 商业结构和收入重要性未知 |
本表刻意区分类生产部署、演练证据、场景证明和只有官方口径的声明。
[CU003, CU004, CU006, CU007, CU008, CU009]Deep Robotics 现有客户基础从公用事业和巡逻用例出发,延伸到更宽的场景营销;但离开公用事业后,证据密度会下降。
[CU016, CU017, CU021, CU032, CU039, CU045]具名账户同时带有流程细节和独立佐证时,证据质量最强;章节里只有 logo 或公司发布的场景主张时,证据最弱。
[CU018, CU019, CU020, CU037, CU038, CU045]6.3 科研和教育采用是真实的,但不应过度解读 logo 和触达指标
Deep Robotics 还有第二条客户动作,和工业巡检很不一样。Lite3 及相关开放平台面向大学、实验室和开发者团队,这些客户看重模块化、SDK 访问、传感器开放性和更低价格。官方和第三方来源反复提到 Tsinghua、Peking University、University of Edinburgh 和 University College London,足以把学术采用视为真实。关于 Tencent 和 Alibaba 的引用较弱:它们显示平台兴趣或研发使用,而不是足以让投资者推断稳定企业收入的生产部署细节。对公司宏大的顶线触达声明也应保持同样谨慎。100+ 客户、覆盖 26 个中国省份、44 个国家和地区、600+ 行业场景这类说法,是有用的方向性采用信号,但不等于运行日付费客户数,也不能告诉我们有多少账户是活跃客户、复购客户或收入重要客户。这是本章在已验证具名证据和营销广度之间划出的主线。[CU014, CU015, CU016, CU017, CU030, CU033]
| 信号类型 | 示例 | 能证明什么 | 不能证明什么 |
|---|---|---|---|
| 经独立佐证的具名部署 | SP Group SPock 隧道巡检 | 真实运营用例、流程细节和结果代理指标 | 完整收入规模、机器人数量或续约条款 |
| 具名官方部署 | 具名部署:China Southern Power Grid、Xining Public Security Bureau、Yiwu International Trade City、Certis | 公司有具体具名场景引用,不只是泛泛的 logo 墙 | 独立客户证言或合同经济性 |
| 演练或场景验证 | Zhejiang Fire Brigade 联合演练 | 产品在恶劣真实条件下的适配性 | 经常性生产部署或可持续客户收入 |
| 科研和教育用户引用 | 清华、北大、Edinburgh、UCL,以及围绕 Lite3 提到的 Tencent 和 Alibaba | 平台在教育和研发场景中被真实采用 | 生产部署深度或当前单账户支出 |
| 无支撑的宽口径规模声明 | 100+ 客户、44 个国家、600+ 场景 | 方向性广度和营销触达 | 经审计的运行日付费客户数、活跃账户数或集中度画像 |
本图表是本章的控制项,避免把 logo 和场景数量过度解读为经审计的客户指标。
[CU015, CU016, CU017, CU020, CU037, CU038]6.4 公用事业客户耐久性看起来有希望,但留存、集中度和政策风险仍披露不足
客户结论是建设性的,但还没有完全承销。正面读法是,多项部署显然已经越过演示阶段剧场:SP Group 把 SPock 运营化进关键任务隧道工作流,China Southern Power Grid 据称扩展到多座变电站,公共安全 / 安防场景也显示机器人能处理静态巡检以外的任务。负面读法是,Deep Robotics 仍未披露能把使用证据转化为收入质量证据的指标。公开没有 NRR、GRR、流失、续约、队列或头部客户收入拆分。国际客户耐久性还取决于本地化和支持强度:新加坡案例需要数月适配,公开的中国大陆售后条款也无法清楚说明海外 SLA 强度。最后,美国议员 2026 年推动限制中国制造四足和人形机器人进入联邦场景,并不是当前 Deep Robotics 的客户流失,但它确实是未来在美国公共部门扩张的品类级阻碍。[CU022, CU024, CU025, CU026, CU027, CU028]
| 指标或代理指标 | 当前公开解读 | 置信度 | 重要性 | 尽调问题 |
|---|---|---|---|---|
| NRR / GRR / 流失 | 未公开披露 | 中 | 公开记录中没有直接的耐久性指标 | 索取按主要垂直行业划分的账户级留存和 cohort 数据 |
| 重复部署代理 | SP Group 计划试点后更大范围铺开;中国南方电网和国家电网相关公用事业叙事暗示多站点延续 | 中 | 公用事业看起来最可能重复使用并扩张 | 确认重复订单量,以及首次试点到后续采购之间的时间 |
| 流程嵌入 | SP Group 巡检员继续用缺陷图像和模拟场景训练 SPock | 高 | 表明机器人已进入运营流程,而不是摆在货架上的演示 | 索取使用小时数、缺陷检测精度随时间变化,以及操作员采用指标 |
| 海外支持可见度 | 中国公开售后条款仅限大陆,优先远程诊断,且常依赖客户寄送 | 中 | 国际耐久性取决于现场支持和 SLA 强度,而这些仍不透明 | 索取海外服务覆盖、备件物流和正常运行时间 SLA |
| 学术复用 | Lite3 被大学反复引用,DEEP 称 ICRA 2026 带来更多欧洲试点兴趣 | 中 | 暗示除公用事业外,还存在长尾客户动线 | 量化年度教育机销量、重复采购机构和分销商贡献 |
本章能看到重复使用代理信号,但看不到足以支撑留存判断的收入质量数据。
[CU014, CU022, CU025, CU026, CU027, CU028]| 风险或扩张杠杆 | 当前证据 | 影响 | 含义 | 尽调路径 |
|---|---|---|---|---|
| 公用事业垂直集中 | 具名证明最集中在公用事业和电网巡检 | 高 | 单一垂直行业 PMF 强,如果其他行业仍处早期,可能演变成收入集中 | 索取按垂直行业划分的收入结构和前十大客户敞口 |
| 国际本地化负担 | 新加坡部署需要数月的传感器、算法和导航本地化 | 中高 | 新账户扩张可能服务占比高,速度也慢于机器人出货标题所暗示的节奏 | 询问新站点部署毛利率和本地化人力投入 |
| 海外证明只有官方口径 | NEOM、北美仓储园区和可可西里均由公司公开宣传,第三方客户验证有限 | 中 | 扩张叙事真实存在,但可融资性不如 SP Group | 索取已签客户推荐和后续订单历史 |
| 海外监管或采购摩擦 | 2026 年美国立法提案瞄准联邦场景中的中国四足机器人和人形机器人 | 中 | 可能限制美国公共部门和关键基础设施扩张 | 跟踪未来规则制定是否点名 Deep Robotics 或其分销商 |
| 支持和数据治理尽调 | 公开条款突出大陆限定售后政策,并显示在法律要求下可能向主管部门披露运营数据 | 中 | 敏感基础设施客户可能要求更强的合同保障 | 索取国际 DPA 条款、审计权和海外支持结构 |
客户上行空间可见,但最大的商业未解问题仍是集中度、本地化成本,以及国际账户真正能留多久。
[CU021, CU027, CU028, CU029, CU035, CU036]6.5 图表
07风险
7.1 监管和法律风险现在是真实的市场准入约束,而不只是理论上的地缘政治阴影
Deep Robotics 最严峻的外部风险,不是某一条已经点名公司的出口禁令,而是美国采购、供应链和技术管制规则越织越密, 在直接列名出现之前就可能压窄可服务市场。FY2025 NDAA 尤其关键,因为它覆盖的正是 Deep 对外销售的组件和系统类型: 依赖 LiDAR 的巡检系统,以及来自受覆盖外国国家的无人地面车辆。与此同时,2026 年 6 月的 1260H 通知显示, 四足机器人近邻 Unitree 已经从抽象审查进入美国军事企业正式认定。Deep 自身定位又放大了敞口。官方材料反复瞄准变电站、 隧道、气体检测、应急响应和公共安全流程,而这些场景恰恰是公共买家、国防相邻承包商和关键基础设施运营方提高供应商筛查的地方。 实际含义是,承销 Deep 时应把它看作一家海外最佳机会来自民用工业客户的公司,这些客户愿意围绕中国硬件做本地化; 不能把它看成能无摩擦进入西方公共部门或国防预算的供应商。第二个法律与商业约束是认证。ANYbotics 现在宣传已具备 ATEX/IECEx Zone 1 准入,并借助 Yokogawa 渠道进入能源和危险行业;保留的 Deep 公开资料显示其可选防爆传感, 但没有系统级证书。这并不排除 Deep 进入公用事业和隧道场景,它在这些场景已经有强证明;但在要求最高的危险流程环境里, 这划出了一条清晰边界。因此,正确的尽调姿态是默认监管和认证会同时塑造产品市场契合度、渠道设计和估值,而不是把它们当成下游合规任务。[CR001, CR002, CR003, CR004, CR005, CR006]
| 风险 | 司法辖区 / 规则 | 当前触发点 | 可能性 | 严重性 | 缓释成熟度 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| 美国 DoD 对中国来源机器人的采购和供应链排除 | 美国 NDAA 第 805 条 / 1260H 框架 | 对 1260H 相关实体,直接 DoD 合同限制自 2026 年 6 月开始,供应链限制自 2027 年 6 月开始 | 中 | 高 | 低 | 高 | 梳理所有美国联邦或国防承包商收入路径,并要求管理层提供受限方和联邦销售政策。 |
| LiDAR 与受限 UGV 规则压缩公共部门 TAM | Public Law 118-159 第 164 和 1078 条 | FY2025 NDAA 明确研究或限制 LiDAR 系统和受限外国 UGV | 中 | 高 | 低 | 高 | 按地区审查物料清单,并确认是否有任何产品变体能在不触及受限 LiDAR 风险的情况下销售。 |
| 先进算力出口管制收紧,迫使重设计或拖慢交付 | EAR / BIS 2026 先进算力管制 | 获批 IC 设计商和先进逻辑管制机制到 2026 年仍有效 | 中 | 高 | 低 | 中高 | 获取硬件 BOM、算力路线图,以及非美国边缘 AI 模块的备用供应商方案。 |
| 危险场景认证落后于已获认证的同业 | 全球油气、化工和 Ex 区域 | 留存材料显示 ANYmal X 已通过 ATEX/IECEx 认证,而 Deep 未见等效系统证书 | 高 | 高 | 低 | 高 | 按场地类别索取任何 ATEX、IECEx 或等效证书、测试报告和部署案例。 |
| 军民两用与公共安全观感挡住敏感买家 | 美国、欧洲、MENA 关键基础设施 | Deep 主打安防、巡逻和应急场景,而行业对中国来源机器人的信任筛查正在收紧 | 中 | 高 | 中低 | 中高 | 审查网络架构、数据本地化选项、本地部署证明,以及第三方安全评估。 |
各行按严重程度排序,结合已生效法律文本、正式通知和已观察到的产品 / 认证定位,而不是假设性制裁。
[CR001, CR002, CR003, CR004, CR005, CR006]热力图式矩阵,对最直接影响 Deep Robotics 市场准入、利润率韧性和融资路径的六类风险排序。
评估为定性判断,并锚定证据;它概括章节中有来源支撑的观点,而不暗示统计精度。
[CR006, CR009, CR018, CR025, CR028, CR039]7.2 运营证明可信,但集中在依赖伙伴、信任和现场执行的定制化公用事业部署
相比许多机器人初创公司,Deep 的真实运营证据更强。公开记录显示,它的巡检平台技术能力扎实:X30 针对严苛环境做了加固, 变电站案例声称带来可观的准确率和人力节省,新加坡 SPock 部署也不只是演示话术,而是进入了真实隧道巡逻流程,并有可衡量的时间节省。 这很重要,因为它证明公司能在真实基础设施里解决昂贵的巡检问题,而不只是交付研究硬件。但同一组证据也暴露了运营风险画像。 Deep 最好的引用案例,是需要路线建图、定位、软件打磨和深度流程集成的定制化公用事业与变电站部署。新加坡这个海外旗舰项目经历了数月安全评估和本地调校; 这是严肃部署应有的样子,投资人乐见其成,但也正是它让扩张更依赖服务,比纯硬件收入更难规模化。 公开证明也很集中。具名引用主要来自国家关联或公用事业式客户;即便最强的海外证明,也仍是伙伴推动的试点, 不是跨多个民营运营方反复落地的广泛客户群。这种集中度又被依赖风险放大。海外交付已经依靠 EGP 等伙伴, 高信任危险巡检渠道则正被认证更强、工业软件联盟更深的竞争对手锁住。另一个运营复杂点是网络信任:尤其在同业硬件安全担忧浮出水面之后, 客户对中国来源机器人进入数据敏感场所更谨慎。直白说,Deep 目前的护城河是严苛公用事业环境里的部署 know-how, 但这条护城河有多深,取决于服务团队、伙伴质量,以及逐个客户通过信任筛查的能力。[CR012, CR013, CR014, CR015, CR016, CR017]
| 失效模式 | 已观察信号 | 可能性 | 严重性 | 缓释成熟度 | 剩余暴露 | 未解决缺口 |
|---|---|---|---|---|---|---|
| 严苛场景下,参考部署之外的正常运行时间不达标 | 官方案例显示其在变电站和隧道表现强劲,但公开证据集中在少数具名环境 | 中 | 高 | 中 | 中 | 需要按产品族提供逐站点正常运行时间、故障率和平均修复时间数据。 |
| 国际落地仍重服务、速度慢 | 新加坡项目耗费数月做安全评估、本地化和软件打磨 | 高 | 中高 | 中 | 中高 | 需要部署周期、扣除本地化后的毛利率,以及跨国家可复制性的队列数据。 |
| 网络或数据信任异议挡住关键基础设施客户 | 行业评论显示,数据敏感场地不愿部署中国四足机器人 | 中 | 高 | 低 | 高 | 需要本地部署架构、SBOM、渗透测试结果和客户安全审批。 |
| 误检或安全事故削弱巡检 ROI | Deep 的价值主张建立在高准确率、自主巡逻和减少人员进入危险区域之上 | 中 | 高 | 中 | 中 | 需要事故记录、假阳性和假阴性率,以及质保或责任流程。 |
| 多产品 QA 负担增加支持和现场复杂度 | Deep 现在支持四足、轮足和人形平台,以及定制载荷 | 中 | 中高 | 中低 | 中高 | 需要产品支持组织架构、备件模型和按地区划分的现场服务人员配置。 |
本登记表强调可部署性和信任,因为 Deep 最强的公开证明来自关键任务巡检,而不是消费级或实验室用途。
[CR012, CR013, CR014, CR015, CR016, CR019]| 依赖项 | 交易对手或系统 | 角色 | 集中度 | 失效场景 | 严重性 | 缓释措施 | 剩余暴露 |
|---|---|---|---|---|---|---|---|
| 具国资关联的公用事业案例 | State Grid、Southern Power Grid、SP Group 类账户 | 锚定证明、部署学习和公开可信度 | 高 | 公用事业账户预算或采购变化,会让当前证明集不再容易迁移 | 高 | 扩大具名民营部门和非中国案例 | 高 |
| 海外伙伴执行 | EGP 和本地集成商 | 本地化、客户触达、部署支持 | 中高 | 伙伴表现不佳会拖慢海外推广,或把 Deep 无法直接配员承担的服务义务留给公司 | 中高 | 标准化实施手册,并认证更多区域伙伴 | 中高 |
| 边缘算力与传感器供应链 | Jetson 级算力、LiDAR、热成像、气体载荷 | 核心自主能力和巡检表现 | 中 | 出口管制或组件变化迫使重设计、延迟发货或抬高成本 | 高 | 在可行处双源化算力,并认证非美国替代方案 | 中高 |
| 危险行业渠道和软件生态 | ANYbotics + Yokogawa 及类似竞争者 | 争夺能源、金属和 Ex 区域账户 | 中 | 竞争对手提供认证叠加一体化软件和服务网络,Deep 因此失去最高信任度的工业客户 | 高 | 先在公用事业和隧道胜出,再补上认证和软件缺口 | 高 |
各行聚焦留存来源集能看到的依赖:客户证明、本地化伙伴、组件栈和竞争对手渠道力量。
[CR008, CR018, CR021, CR022, CR023, CR035]梳理 Deep Robotics 产品与可规模化海外巡检业务之间的外部依赖。
依赖项聚焦外部可见的卡点——客户证明、合作伙伴、供应商、认证和服务能力——而不是未公开的供应商关系。
[CR018, CR022, CR023, CR036, CR037, CR038]7.3 财务模型已经转向盈利,但下一阶段同时面对价格压力、资本强度和异常拥挤的 2026 融资窗口
正面的财务事实确实存在。多份基于招股书的摘要都指向同一组结论:Deep 2025 年收入约人民币 3.37 亿元, 毛利率超过 52%,并首次实现盈利。与很多私有机器人公司可见的公开证据相比,这明显更强。承销难题在于下一步。 让公司更可投资的每一条来源,也让它显得更吃资本。公司现在同时推进 Lynx 家族、国际现场部署和 DR02 人形机器人项目; 招股书相关报道还称,下一轮融资将用于具身 AI 算法、大模型、新机器人本体和制造扩张。换句话说, 今天已经盈利的四足机器人业务,被要求为明天更宽的机器人平台融资。 竞争让这场平衡更难。Unitree 仍是这个品类的价格和出货量参照,头部硬件指标更强,单位规模也深得多。 SemiAnalysis 和其他保留来源显示,Deep 相对 Unitree 仍有溢价,但没有 Unitree 的出货基础,也没有消费或学术飞轮。 同时,中国机器人 IPO 队列正在拥挤:同业也在冲向公开或准公开流动性事件,行业评论反复回到同一个未解问题——量产经济性。 结果是一组更像硬件成长公司、而非纯软件故事的风险叠加:毛利率可能压缩,营运资本可能膨胀;如果 IPO 窗口滑走, 或人形机器人投入跑得比工业业务现金生成更快,估值叙事会迅速松动。[CR024, CR025, CR026, CR027, CR028, CR029]
| 风险驱动因素 | 当前信号 | 可能性 | 严重性 | 缓释成熟度 | 剩余暴露 | 投资含义 |
|---|---|---|---|---|---|---|
| 2025 年盈利随着人形投入上升而逆转 | 首个盈利年份真实存在,但招股书口径来源提醒,研发和迭代可能压迫利润率 | 中 | 高 | 中低 | 高 | 没有产品组合和成本桥接时,不应把 2025 年利润率按稳态承销。 |
| 价格战压缩工业四足利润率 | Unitree 在重叠硬件赛道里同时拥有更大出货规模、更强账面参数和更低价格 | 高 | 高 | 低 | 高 | 采用保守利润率假设,并要求证明软件、服务或认证差异化。 |
| 营运资本和制造资本开支超过已披露现金 | IPO 和私募轮融资款被指定用于算法、机器人本体和制造扩张 | 中高 | 高 | 低 | 高 | 把融资视为核心依赖,而不是锦上添花的增长加速器。 |
| 同业拥挤下,IPO 窗口推迟或估值下修 | Deep、Unitree 和其他中国机器人公司都在测试公开或准公开流动性路径 | 中 | 高 | 中低 | 中高 | 如果 2026-2027 年 STAR 流程或同业 IPO 交易带恶化,应对融资假设打折。 |
本表区分当前财务证明和前瞻承销风险;今天的利润基础来自工业四足,明天的支出则越来越偏向人形和制造。
[CR024, CR025, CR027, CR028, CR029, CR031]展示监管、竞争和执行风险如何传导到客户准入、毛利率、融资依赖和估值。
该图突出保留公开证据中信号最强的传导路径,而不是穷尽所有可能的二阶影响。
[CR006, CR008, CR010, CR025, CR028, CR029]7.4 人才与执行风险卡在组织宽度:如果 Deep 不能像加产品一样快速补上合规、服务和资本纪律,投资逻辑就会断裂
Deep 的执行挑战不止产品工程。公司材料把它描述成一家全栈机器人组织,横跨核心部件、四足机器人、人形机器人、制造、销售和服务。 这种宽度是战略抱负,也是人才和控制问题,因为公司必须同时扩张合规、现场支持、认证工作、伙伴管理和资本配置。 目前最强的证据仍落在公用事业和严苛巡检路线里的工业四足机器人。Deep 越是从这条主线走向国际渠道建设、危险行业认证和人形机器人商业化, 瓶颈就越会从技术可能性转向管理带宽。 因此,最有用的投资人框架是触发器驱动。有些风险如果保持有边界就可管理:只要客户多元化改善,服务密集型海外铺开可以接受; 只要四足机器人毛利健康、融资通道保持打开,人形机器人扩张可以接受;只要 Deep 聚焦于本地部署和民用流程已经足够的市场, 国别来源筛查也可以容忍。但如果这些缓释措施没有成熟,投资逻辑就会断裂。杀伤条件并不玄妙,而是具体事件: 没有认证路径、电力公用事业之外没有多元化、价格压力下毛利率可见下滑,或在研发强度仍高时 IPO 和融资延迟。 在管理层证明组织能像工业化机器人本体一样快速工业化控制体系和渠道之前,执行风险都应被视为支撑估值的关键约束之一。[CR017, CR030, CR031, CR035, CR036, CR037]
| 角色或职能 | 依赖或缺口 | 可能性 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| 跨产品线的高管带宽 | 管理层必须同时放大四足、Lynx、DR02、制造和国际服务 | 中 | 高 | 让人形投入按里程碑放行,并优先推进核心巡检赛道 | 索取产品 P&L 归属、阶段门流程和项目复盘节奏。 |
| 具身 AI 与现场机器人领域的人才留存 | 研发强度仍高,公司仍需要资本和招聘来支撑扩张 | 中高 | 中高 | 用融资锁定关键工程和现场服务人才 | 索取研发和部署团队的流失率、招聘计划和薪酬结构。 |
| 合规与认证负责人 | 监管、网络和危险场地文档公开成熟度仍不足 | 高 | 高 | 更大规模海外扩张前,招聘或提拔专职合规与认证负责人 | 审查组织架构、认证预算和出口管制责任归属。 |
| 国际现场服务梯队 | 伙伴主导的部署看起来仍人员偏薄,且高度定制化 | 中 | 高 | 搭建标准化培训、备件体系和区域支持枢纽 | 索取服务级别指标、伙伴评分卡和按地区划分的装机支持覆盖。 |
这里的执行风险是组织问题,不只是技术问题:公司已经展现产品能力,但规模化仍需要控制体系、服务深度和合规肌肉。
[CR030, CR031, CR036, CR037, CR038, CR039]| 风险 | 可监控触发项 | 阈值或事件 | 行动含义 |
|---|---|---|---|
| 危险场地准入仍未验证 | 认证进展 | 下一轮危险行业推进前,仍没有系统级 ATEX、IECEx 或等效证书,也没有已提交路线图 | 将油气和 Ex 区域上行排除在范围外,并主要把 Deep 按公用事业与隧道公司承销。 |
| 原产地审查扩散到商业买家 | 被拦账户证据 | 主要美国、欧盟或盟友关键基础设施买家以合规或网络理由拒绝 Deep,且两个季度内没有可行补救方案 | 削减海外 TAM,并提高估值所需安全边际。 |
| 竞争下利润率韧性失效 | 毛利率和 ASP 趋势 | 综合毛利率显著跌破 40%,或 ASP 下滑且没有软件或服务附着抵消 | 转向更低的硬件倍数框架,并下调对规模经济的信心。 |
| 客户集中成为结构性问题 | 具名客户多元化 | 下一次融资事件前,没有新的非公用事业锚定账户,也没有海外重复量产部署出现 | 把当前案例质量视为集中式证明,而不是广泛市场验证。 |
| 人形扩张跑在融资能力前面 | 现金和融资路径 | 研发和制造支出加速,同时 IPO 节奏推迟或过桥融资变得不确定 | 停止承销人形可选性,只按核心工业四足给公司估值。 |
| 执行控制落后于范围扩张 | 组织搭建 | 更大规模国际推广前,仍看不到合规负责人、认证计划或区域现场服务建设 | 推迟投资,或要求按里程碑进入,直到运营控制成熟。 |
触发项有意设计成可监控且与投资论点相连,而不是泛泛的红旗;每一项都会改变收入质量、融资风险或可信估值框架。
[CR006, CR010, CR035, CR036, CR039, CR041]7.5 展示项
08估值
8.1 Deep Robotics 已深度进入 IPO 流程,但仍是私有公司,估值记录依旧嘈杂
Deep Robotics 已不再是寻找第一场市场定价事件的概念期机器人初创公司。截至本次报告日,公司已经完成 IPO 辅导, STAR Market 申请已获受理,招股书摘要也在多家中英文媒体流传。这很重要,因为它给投资人的运营证据,远多于多数私有机器人公司。 同时,上述步骤都不等于公司已经上市。截至 2026-06-21,Deep Robotics 仍处于上市前状态:私有、申报推进中, 真正流动性还取决于最终审核、定价和市场窗口。因此,估值线索是混杂的,不是干净的。Caixin 在 2025 年 12 月把公司描述为约人民币 80 亿元; Humanoids Daily 随后称业务估值超过人民币 100 亿元;Pandaily 则描述潜在上市后估值约 $3.5 billion。 这些不是等价参照。有些是当前私有或 IPO 前标记,有些是申报期价值的媒体摘要,还有些是完成公开交易后的标题式预期。 谨慎的承销姿态,是把可观察事实与愿景切开:公司显然奔着 IPO 去,但被充分证据支撑的估值区间,明显窄于最具宣传性的标题所暗示的范围。[CV001, CV002, CV003, CV004, CV005, CV040]
| 维度 | 取值 | 支撑证据 | 决策含义 |
|---|---|---|---|
| 建议 | 跟踪 | 运营证明真实存在,但公开估值证据区间过宽、营销色彩过重,不足以给出干净的买入判断 | 保持接触,但必须有明确价格纪律和更深入尽调 |
| 信心 | 中 | 申报期财务证据强于多数私人机器人同业,但估值标记互相冲突,条款经济仍不透明 | 对公司质量保持中等信心,对 headline 估值保持低信心 |
| 风险评级 | 高 | 人形资本开支、地缘政治、竞争和股权结构不确定性,都可能快速压缩价值 | 把它视为需要主动监控的持仓,而不是被动叙事押注 |
| 估值立场 | 高于低十亿美元区间则偏贵;接近 $3.5B headline 则偏投机 | RMB 80-100 亿元仍可论证;最激进的 IPO headline 缺乏当前证明支撑 | 没有新证据时,不追逐亢奋的 IPO 前定价 |
这是私募市场建议。尽管 IPO 流程活跃,截至运行日期,公司仍是私人公司。
[CV001, CV040, CV041, CV042, CV043, CV047]| 论点类型 | 论点 | 什么会改变判断 |
|---|---|---|
| 投资论点 | Deep Robotics 已经证明,工业四足能在危险环境中创造真实收入、真实利润和真实部署 ROI。 | 披露复购队列、服务附着和客户集中度数据,证明 2025 年不是一次性利润率尖峰。 |
| 投资论点 | 公司比多数机器人初创企业更接近流动性,因为辅导已经完成,STAR Market 申请也已获受理。 | 提供最终审批里程碑、承销商价格指引,以及上市后流通盘 / 流动性假设。 |
| 投资论点 | 如果 2026 年执行继续强劲,工业证明叠加早期人形可选性,可以支撑较普通硬件可比公司的溢价。 | 展示多台人形销售,而不只是单笔高 ASP 证明交易,同时维持高毛利率。 |
| 反论点 | 公开估值证据不一致,可能混杂了私募估值、媒体预期和上市后愿景式结果。 | 发布更收敛的当前投后估值参考,或有投行背书的定价区间。 |
| 反论点 | 在人形商业化和软件变现上,Deep Robotics 仍明显早于估值最高的机器人可比公司。 | 证明人形、软件或服务能对经常性经济产生实质贡献,而不只是贡献故事价值。 |
| 反论点 | 地缘政治和市场准入风险意味着,中国工业机器人发行人相对西方或已上市中国同业可能应该折价,而不是溢价。 | 提供在地缘政治不那么敏感的渠道中,海外企业需求具备韧性的证据,并说明出口算力韧性。 |
本章有意保持价格敏感:公司质量和估值支撑分开判断。
[CV001, CV006, CV007, CV018, CV031, CV032]从未上市状态、经营验证、估值模糊度和资本强度出发,推导出跟踪建议的决策流。
该图服务于决策,而非穷尽所有因素;它展示最可能改变建议的变量。
[CV001, CV006, CV007, CV040, CV043, CV047]8.2 商业证明真实,但可比倍数仍要求入口纪律,而不是动量定价
相比通用工业硬件倍数,愿意为 Deep Robotics 支付显著溢价的最强理由,是它已经跨过了许多机器人同业尚未跨过的门槛。 招股书衍生报道显示,公司 2025 年收入人民币 3.37 亿元、净利润人民币 2868 万元、毛利率 52.83%, 具身机器人收入占比较高,并在电力巡检和新加坡公用事业隧道有可见的工业部署证明。这些是严肃信号,不是演示剧场。 但估值必须放到其他机器人标记已经能买到什么的背景里看。Unitree 是最接近的私有可比公司,因为它同样立足中国、 有足式机器人传承,也在 IPO 路径上;但 Unitree 收入已越过人民币 10 亿元,私有估值仍被讨论在约人民币 120 亿元, IPO 目标则激进得多,约 $7 billion。UBTECH 提供了一个正在交易的中国公开市场基准,2025 年营业额约人民币 20 亿元, 市值达数十亿美元;Boston Dynamics 和 Skild AI 则显示,公开市场和战略投资人会给品类领导者或软件优先平台分配多少可选性。 Deep Robotics 在规模和宽度上低于这些天花板。它 2025 年卖出一台人形机器人,四足机器人和 Lynx 有实质牵引, 工业切入点很强;但它也面对不利市场来源强调的同样价格压力、地缘政治折价和人形机器人资本开支负担。 这种组合支持相对无差异硬件的溢价,但不支持开空白支票。[CV006, CV007, CV008, CV009, CV010, CV011]
| 可比对象 | 指标 | 倍数 / 估值 / 状态 | 可比意义 | 局限 |
|---|---|---|---|---|
| Deep Robotics 当前私募 / 媒体走廊 | 2025 年收入与报道私募估值标记对比 | Caixin 式 80 亿元人民币约 ~$1.1B(约 22x 销售额),到 100 亿元人民币以上即 >$1.5B(约 28x 销售额) | 最适合锚定当前进入纪律,因为它使用公司自己披露的收入基础 | 仍把 2025 年末私募语境与 2026 年媒体摘要混在一起,而不是已签署定价轮 |
| Deep Robotics 目标 IPO 标题 | 2025 年收入与上市后预期 | 约 $3.5B 标题估值(约 70x 过往销售额) | 展示部分媒体报道押注的上行叙事 | 对一家 2025 年只卖出一台人形机器人的公司来说,显著高于当前证据能支撑的水平 |
| Unitree 从私募到 IPO 的路径 | 2025 年收入与私募 / IPO 估值 | Series C 后估值 >120 亿元人民币,目标科创板 IPO 估值约 $7B,对应 2025 年收入约 $248M | 中国腿式机器人中最接近的可比公司,规模更强,人形叙事也更宽 | 目标 IPO 估值不等于最终定价,且已包含市场乐观预期 |
| UBTECH 上市基准 | 2025 年营业额与市值 | 2025 年营业额 RMB 2.001B,估值参考约 $7B | 提供一个已上市中国基准,观察公开市场如何给机器人规模定价 | 业务组合比 Deep Robotics 更宽,市值参考也不是运行日的即时报价 |
| Boston Dynamics 战略估值天花板 | 最新战略估值参考 | 分析师估计约 $25B;Hyundai 2021 年收购价为 $1.1B | 展示有全球工业母公司背书时,品类龙头能吸引多少期权价值 | 对一家聚焦巡检的中国私营发行人来说,不是干净的直接可比 |
| Skild AI 软件溢价 | 最新融资估值 | 机器人基础模型公司 Series B 估值约 $4.5B | 说明当前市场相对硬件优先玩家,仍在过度奖励软件和 AI 期权价值 | 软件优先的 AI 公司不能作为 Deep Robotics 的同类倍数 |
保留公开、私募和目标估值的混合口径,是因为这些就是投资者会遇到的真实市场背景。衍生销售倍数在可得时采用已披露的 2025 年收入。
[CV023, CV024, CV025, CV026, CV027, CV029]条形图展示 Deep Robotics 与选定可比公司在当前、目标和战略估值参照下的聚集方式。
数值来自保留来源,按百万美元四舍五入,混合了私募估值、公开市场价值和目标 IPO 参照。
[CV002, CV003, CV004, CV024, CV025, CV027]IC 风格评分卡,覆盖最可能决定 Deep Robotics 上市前是否值得高价进入的维度。
分数是基于保留证据给出的 0-10 编辑判断,不是外部评级。
[CV006, CV007, CV018, CV031, CV034, CV047]8.3 证据支持“跟踪”建议:纪律性私有标记附近可买,激进 IPO 标题价下偏投机
核心估值问题不是 Deep Robotics 抽象上是不是一家好公司,而是当前可观察价格走廊是否还留有有吸引力的风险调整回报空间。 按 2025 年约 $49.5-49.6 million 的追踪收入,2025 年 12 月 Caixin 式人民币 80 亿元标记已经意味着约 22 倍销售额。 如果按超过人民币 100 亿元理解,公司会被推向约 28 倍销售额;尽管规模明显更小、人形机器人证明也少得多, 估值却已接近 Unitree 式区域。$3.5 billion 的上市后标题价则意味着约 70 倍追踪销售额, 更接近软件式或投机性人形机器人可选性定价,而不是证据支撑的工业机器人定价。这并不意味着公司不可投资, 而是估值需要场景纪律。熊市情形下,硬件倍数压缩、IPO 延迟或毛利率正常化,都可能把公允价值拖到近期私有标题价之下。 基准情形下,持续工业增长加上可信但仍早期的人形机器人可选性,可以支撑低到中段 $1 billions 区间。 牛市情形下,Deep Robotics 不仅需要继续兑现四足机器人执行,还需要 2026 年持续增长、稳定高毛利率, 并拿出真实证据证明人形机器人能成为可重复业务,而不只是叙事延伸。这就是建议为跟踪而非买入的原因。 证据足以让投资人保持兴趣,但还不足以在没有更锋利入口价或更深尽调的情况下承销最激进标记。[CV040, CV041, CV042, CV043, CV044, CV045]
| 情景 | 假设 | 估值 / 回报逻辑 | 关键风险 | 概率信号 |
|---|---|---|---|---|
| 熊 | IPO 节奏推迟,工业增长回归常态,人形商业化仍停留在叙事层面,利润率从 2025 年高位回落。 | 可支撑股权价值降至约 $450M-$900M;若按 2025 年末或更高估值进入,将面临明确下行。 | 倍数压缩、出口管制摩擦、Unitree 竞争和公开市场疲劳。 | 如果 2026 年增长显著放缓,或利润率在资本开支仍高时回落,则该情景可信。 |
| 基准 | 工业四足和 Lynx 继续增长,利润保持为正但波动较大,人形可选性仍处早期而非改变格局。 | 可支撑股权价值大致集中在 $800M-$1.6B;只有接近近期媒体估值低端时,纪律性参与才说得通。 | 需要在不依赖软件式重估的情况下维持执行。 | 最贴合当前可观察证据集。 |
| 牛 | Deep Robotics 维持高速增长,毛利率接近当前水平,IPO 顺利过会,并把人形做成可复制的商业平台。 | 可支撑价值达到约 $1.8B-$3.0B,对最激进的 $3.5B headline 只能形成部分支撑。 | 需要 2026 年在规模、利润率和新形态牵引力上拿出异常强的证明。 | 可能成立,但取决于尚未公开的证据。 |
情景区间是由公开证据推导的可支撑价值护栏,不是对最终 IPO 价格的预测。
[CV041, CV042, CV043, CV044, CV045, CV046]在噪音较大的头条估值走廊下,Deep Robotics 可被公开证据支撑的价值区间。
这些是公开证据可支撑的区间,并非断言最终 IPO 必须按哪个价格定价。
[CV040, CV041, CV042, CV043, CV044, CV045]8.4 退出准备度好于多数私有同业,但公开市场准备度不等于估值准备度
Deep Robotics 看起来比许多私有机器人公司更接近退出,因为它已经完成辅导、STAR Market 申报获受理, 通过招股书摘要披露了 2025 年盈利,并且资本用途计划与下一阶段路线图相匹配。这比依赖风险投资路演材料或未定价传闻轮要扎实得多。 即便如此,IPO 路径不等于一份已经尽调完整的投资案例。BigGo 的招股书摘要在这里尤其重要:它称如果 IPO 流程停滞, 投资人估值调整权可能重新出现,这意味着优先权和治理条款仍可能扭曲普通股经济性。其他未解问题同样现实。 公开记录仍未显示准确的当前投后估值、优先权栈、清算保护、按收入计的客户集中度、复购 cohort、质保负担、 服务绑定率,或 2025 年的毛利率画像能否经受更广泛的人形机器人推进。监管和地缘政治摩擦也仍然重要。 美国采购和出口管制体系到 2026 年继续收紧;同时,拥挤的机器人 IPO 日程和资金充足的 AI 软件挑战者,让全球竞争进一步升温。 因此,最终投资问题是运营性的:买方能否拿到足够的私有尽调,确认 Deep Robotics 不只是具备 IPO 能力, 而且值得在市场给出公开价格之前支付溢价?在答案更清楚之前,正确姿态是主动跟踪、定义杀伤触发器,并要求证据补上股权结构和收入质量缺口。[CV032, CV033, CV034, CV035, CV048, CV049]
| 触发因素 | 阈值 | 对投资论点的传导 | 行动含义 |
|---|---|---|---|
| IPO 进程停滞或大幅降价 | 辅导 / 受理文件推进未能转化为市场化发行,或最终定价较媒体预期大幅折价 | 证实当前标题估值高估了市场胃口或股权结构质量 | 从跟踪下调为回避,直到出现新的干净入场价格 |
| 2026 年增长放缓,同时人形机器人投入上升 | 相比 2025 年增长明显转弱,且没有软件或人形机器人收入证据抵消 | 打破公司可用复合增长的工业基本盘为期权价值供血的判断 | 转入熊市情形承销,并要求硬性下行保护 |
| 毛利率显著低于 2025 年水平 | 从 52.83% 持续回落,且没有产品组合或规模解释 | 暗示 2025 年盈利能力只是阶段性,或高度依赖产品组合 | 按更低硬件倍数重新定价 |
| 出口管制或采购壁垒收紧 | 新限制削弱进入海外关键基础设施或先进算力路径的能力 | 推高折现率,并压缩现实的海外 TAM | 假设退出倍数更低、国际渗透更慢 |
| 人形机器人商业化仍可忽略 | IPO 进程推进后,仍无有意义的多台人形机器人收入或重复部署 | 拿掉激进估值中内含的期权溢价 | 主要按工业四足和轮腿式硬件厂商给公司估值 |
每个触发因素都应能在当前到任何定价交易之间被监测,而不只是描述性指标。
[CV016, CV031, CV032, CV034, CV035, CV043]| 主题 | 缺失证据 | 重要性 | 负责人或尽调路径 |
|---|---|---|---|
| 当前股权结构与优先权栈 | 最新投后估值、清算优先权、反稀释条款,以及任何投资者附带协议 | 即使标题估值看似不变,入场经济性也可能剧烈变化 | 定价任何一轮前,要求提供完整股权结构、融资文件和律师审查 |
| VAM / 特殊股东权利 | 若 IPO 进程停滞,受影响的触发条件、经济条款和相关方 | 招股书阶段摘要已经提示,IPO 失败时投资者权利可能恢复 | 获取已签署的股东协议和承销商 / 法律备忘录 |
| 收入质量 | 经常性服务占比、质保负担、实际折扣,以及按产品家族划分的复购队列 | 2025 年利润率和利润只有在规模化与支持义务下守得住才有意义 | 要求产品线 P&L、队列留存和服务附着分析 |
| 客户集中度 | 国家电网、南方电网、海外公用事业和其他大客户之间的收入集中度 | 集中的工业基本盘会在上市时同时放大上行和下行 | 要求管理层提供前十大客户明细和集中度说明 |
| 人形机器人商业化 | 除 2025 年披露的一台销量外,管线、已签订单、单位经济性和附着率 | 激进估值情形隐含为人形机器人上行付费 | 要求已签管线明细和制造就绪度审查 |
| IPO 准备度与价格发现 | 承销商反馈、指示性定价区间、流通股、基石投资者胃口和上市时间表 | 私募估值与市场化 IPO 之间,差距可能达到数亿美元 | 直接做投行尽调,并与公开同业交易窗口比较 |
这些问题优先围绕最能快速改变投资判断或可接受入场价格的证据。
[CV005, CV016, CV049, CV050, CV051, CV052]8.5 展示项
免责声明
本报告是基于截至 2026-06-21 可得公开信息生成的 AI 辅助尽调摘要,不构成投资建议。Deep Robotics 目前释放出的公开经营信号强度,在私营机器人公司里并不常见,但在投入资本前,估值逻辑和下行保护仍需要对股权结构条款、客户经济性、现金跑道和 IPO 执行做直接尽调。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Deep Robotics describes itself as a Hangzhou-based embodied-AI robotics company spanning quadruped robots, humanoid robots, and core robotic components. | 中 | SO020 |
| CO002 | Official materials say Deep Robotics built a full-stack “Perception–Decision–Execution” framework for energy, emergency response, industrial operations, and education scenarios. | 中 | SO020 |
| CO003 | Deep Robotics was founded in November 2017. | 中 | SO011, SO013 |
| CO004 | Zhu Qiuguo and Li Chao founded Deep Robotics after working in the Zhejiang University robotics ecosystem. | 高 | SO011, SO013, SO014 |
| CO005 | On 2025-11-03 the company changed into a non-listed joint-stock entity under the Hangzhou Deep Robotics Technology Co., Ltd. name. | 中 | SO005, SO012, SO013 |
| CO006 | Deep Robotics filed for IPO tutoring on 2025-12-23 after signing advisory agreements on 2025-12-08 with China Securities, Zhong Lun, and Pan-China Certified Public Accountants. | 高 | SO001, SO007 |
| CO007 | The reviewed public record supports that Deep Robotics remained private and in pre-IPO preparation as of 2026-06-21 rather than already listed. | 中 | SO001, SO004, SO005 |
| CO008 | Public reporting around the IPO tutoring event valued Deep Robotics at about CNY8 billion, or roughly USD1.1 billion. | 高 | SO001, SO007 |
| CO009 | Deep Robotics announced a nearly CNY500 million pre-Series C financing in July 2025 led by Fortune Capital and CRHC/Guoxin-linked capital with additional state-backed investors. | 高 | SO002, SO011, SO013 |
| CO010 | Deep Robotics announced a more-than-CNY500 million Series C in December 2025 co-led by China Merchants Bank International and China Asset Management, with China Telecom and China Unicom affiliated funds participating. | 高 | SO003, SO006, SO013 |
| CO011 | 36Kr reported that Deep Robotics completed a pre-IPO financing round of hundreds of millions of renminbi in early 2026 led by the National AI Industry Fund with JD.com following on. | 中 | SO013 |
| CO012 | The public financing record supports at least eight financing rounds through early 2026, but it does not reconcile a precise lifetime total raised. | 中 | SO011, SO013 |
| CO013 | Management interviews said 2024 revenue increased by more than 100% year over year versus 2023. | 中 | SO011, SO012 |
| CO014 | The public revenue disclosure remains qualitative at “several hundred million yuan” rather than an audited figure. | 中 | SO005, SO012 |
| CO015 | Li Chao publicly said Deep Robotics is “not in the red,” implying at least breakeven operating status by late 2025. | 中 | SO012 |
| CO016 | Public reporting in 2025 described the Deep Robotics team as nearly 400 people. | 中 | SO011, SO013 |
| CO017 | Management said 2025 shipments could reach 10,000 units, roughly three to five times 2024 volume. | 中 | SO005, SO011, SO012 |
| CO018 | 36Kr reported that Deep Robotics had already implemented more than 600 industry projects across power, emergency, security, education, tunnels, and entertainment. | 中 | SO013 |
| CO019 | China Southern Power Grid became the first major industrial application anchor, turning Deep Robotics from a lab-driven project into a power-inspection vendor. | 中 | SO011, SO012, SO013 |
| CO020 | Public sources name state power utilities such as State Grid and China Southern Power Grid among Deep Robotics customer anchors. | 中 | SO006, SO007 |
| CO021 | The X30 deployment at Singapore Power Group made Deep Robotics the first Chinese quadruped vendor deployed in an overseas power system. | 高 | SO008, SO025 |
| CO022 | SP Group estimated the SPock deployment would save about 480 manual inspection hours per year. | 高 | SO008, SO025 |
| CO023 | The X30 was launched on 2023-10-10 for power stations, factories, utility corridors, and rescue work with a 4 m/s top speed, 45° climbing angle, and 20 cm obstacle capability. | 中 | SO008, SO011 |
| CO024 | Official X20 materials describe an IP66 industrial quadruped with a 53 kg body, 15 km range, 4 m/s speed, and 20 kg payload/endurance figure for harsh inspection work. | 中 | SO015, SO022 |
| CO025 | Deep Robotics debuted DR01 at the 2024 World Robot Conference as its first humanoid robot. | 高 | SO002, SO011, SO021 |
| CO026 | Deep Robotics launched the Lynx M20 in April 2025 as a wheeled-legged robot for complex terrains and hazardous industrial environments. | 高 | SO002, SO003, SO023, SO026 |
| CO027 | Official Lynx M20 materials cite 33 kg weight, 15 kg payload, 2.5-hour loaded runtime, 80 cm single-step capability, IP66 protection, and a -20°C to 55°C operating range. | 高 | SO023, SO026 |
| CO028 | Deep Robotics launched the DR02 on 2025-10-09/10 as the world’s first IP66 full-body industrial humanoid with 175 cm height, 65 kg weight, and 275 TOPS computing power. | 高 | SO009, SO010, SO024 |
| CO029 | Public DR02 materials say the robot operates from -20°C to 55°C, walks at 1.5 m/s with a 4 m/s peak, and can handle 10 kg per arm. | 高 | SO009, SO010, SO024 |
| CO030 | Independent coverage said Deep Robotics had not disclosed DR02 autonomy, battery-life or operating-duration details, or price at launch. | 中 | SO010, SO016 |
| CO031 | 36Kr reported that Deep Robotics held more than 100 authorized patents, including more than 20 invention patents. | 中 | SO011 |
| CO032 | Zhu Qiuguo is an associate professor at Zhejiang University and both founders emerged from the same university laboratory. | 高 | SO011, SO014 |
| CO033 | Deep Robotics has publicly argued that its quadrupeds are materially cheaper than Western incumbents, with claims ranging from one-third-to-one-half of Boston Dynamics costs to 20%-30% cheaper than Korean peers. | 中 | SO011, SO016 |
| CO034 | Named public operators beyond the founders include product and deployment spokespeople Peter Deng, Hank Cheng, Mao Tang, and Li Chang-bo. | 中 | SO015, SO016, SO025, SO026 |
| CO035 | The reviewed public pack does not disclose a clean current board roster, independent-director list, or fully itemized post-reform executive lineup. | 中 | SO005, SO012, SO020 |
| CO036 | A quoted investor in Caijing argued the embodied-robotics industry is selling products to customers who know current robots are not useful enough yet. | 中 | SO012 |
| CO037 | Sector analysis positioned Deep Robotics below Unitree and AgiBot and argued the robotics IPO wave still has to prove ROI, cost discipline, and real-world utility at scale. | 中 | SO018, SO019 |
| CO038 | Exact lifetime capital raised, detailed ownership, debt or credit facilities, and any secondary-sale history remain undisclosed in the reviewed public pack. | 中 | SO001, SO013, SO020 |
| CO039 | The public pack supports meaningful commercial traction but not a full disclosure set, because revenue range, headcount, shipment target, and product breadth are public while audited financials and customer count are not. | 中 | SO005, SO011, SO012 |
| CO040 | The 2025 shareholding reform is widely interpreted as IPO preparation even though management publicly described it as driven by development needs. | 中 | SO005, SO012, SO013 |
| CO041 | The visible business model is industrial robot and solution sales into inspection, emergency, logistics, research, and related hazardous-work scenarios rather than a scaled consumer robotics business. | 中 | SO011, SO015, SO020 |
| CO042 | By 2023 Deep Robotics had already sold Lite3 units into more than a dozen countries and named universities such as Tsinghua, Peking, Edinburgh, and UCL as users. | 中 | SO017 |
| CM001 | Deep Robotics' evidenced market is narrower than generic robotics and centers on legged or wheel-legged robots for industrial inspection, underground navigation, emergency response, and research deployments. | 高 | SM011, SM012, SM017 |
| CM002 | Included spend should cover mobile inspection hardware, autonomy software, sensor payloads, and deployment integration for difficult industrial sites. | 高 | SM011, SM024 |
| CM003 | Excluded spend should include general industrial robots, defense-only UGV programs, consumer quadrupeds, and drone-only inspection budgets. | 中 | SM001, SM010, SM018 |
| CM004 | Manual patrols, fixed sensors, wheeled robots, and drones remain the main status-quo substitutes for the inspection jobs Deep Robotics targets. | 中 | SM002, SM015, SM018 |
| CM005 | Deep Robotics' public deployments cluster in power substations, wind farms, power cable tunnels, and UHV grid assets. | 高 | SM013, SM014, SM015, SM017 |
| CM006 | Precedence Research forecasts the global quadruped robot market at USD 3.71 billion in 2026, reaching USD 17.53 billion by 2035 at an 18.84% CAGR. | 中 | SM001 |
| CM007 | Emergen Research frames the market much lower, at USD 1.2 billion in 2024 and USD 4.8 billion by 2034 with a 14.8% CAGR. | 中 | SM002 |
| CM008 | The gap between Precedence and Emergen shows that public quadruped TAM estimates vary materially by publisher, time window, and category boundary. | 中 | SM001, SM002 |
| CM009 | Published analyst reports make defense the largest explicit application bucket, which means industrial inspection should not be treated as the whole quadruped market. | 中 | SM001, SM002 |
| CM010 | Precedence says North America held 46% of market value in 2025 while Asia Pacific is the fastest-growing region. | 中 | SM001 |
| CM011 | Deep Robotics' practical serviceable market is a subset of the broad quadruped TAM, focused on industrial inspection workflows rather than all legged-robot uses. | 高 | SM001, SM011, SM023 |
| CM012 | China installed 295,000 industrial robots in 2024, representing 54% of global demand. | 中 | SM003 |
| CM013 | China's 15th Five-Year Plan for 2026-2030 places robotics and AI physical applications at the center of industrial policy. | 中 | SM003 |
| CM014 | China robot density remains 166 per 10,000 manufacturing workers, well below South Korea and Germany, indicating further adoption headroom. | 中 | SM003 |
| CM015 | Deep Robotics claims more than 50% efficiency gain and 98% inspection accuracy in power-substation deployments. | 中 | SM013 |
| CM016 | Deep Robotics claims 96.5% detection accuracy in autonomous wind-farm inspections in Ningxia. | 中 | SM014 |
| CM017 | Deep Robotics' Singapore tunnel deployment shows the company can localize a legged inspection robot for an overseas utility environment. | 中 | SM015 |
| CM018 | In utility deployments, the everyday user is likely the maintenance or field inspection team rather than the executive budget owner. | 高 | SM013, SM015, SM017 |
| CM019 | Research and education buyers behave differently from industrial buyers because they prioritize affordable platforms and experimentation over enterprise uptime guarantees. | 中 | SM011, SM018 |
| CM020 | ANYbotics explicitly targets oil and gas, chemicals, power and utilities, mining and metals, and transport operators. | 高 | SM007, SM023 |
| CM021 | ANYmal X is marketed as ATEX and IECEx Zone 1 certified and IP67 rated, creating a real certification barrier in hazardous industrial sites. | 高 | SM006, SM008 |
| CM022 | ANYbotics said in February 2026 that it is expanding further into energy markets across Asia and the Middle East with Yokogawa. | 中 | SM006 |
| CM023 | Boston Dynamics put Spot on the market at a USD 74,500 base price, with an additional USD 34,570 autonomy package and USD 12,000 annual warranty plan. | 中 | SM004 |
| CM024 | Boston Dynamics now layers Orbit fleet software on top of the robot, turning enterprise inspection purchases into combined hardware-plus-software budgets. | 高 | SM009, SM024 |
| CM025 | Unitree warns that some Go2-W functions still require manual operation or secondary development because of current technological and computing-power limits. | 中 | SM018 |
| CM026 | Unitree positions B2 as an industrial quadruped with 6 m/s speed and 40 kg sustained load, reinforcing a broad Chinese price-performance ladder. | 中 | SM019 |
| CM027 | Unitree markets dedicated power inspection solutions, showing that utility budgets are contested by multiple Chinese legged-robot vendors. | 高 | SM019, SM025 |
| CM028 | Ghost Robotics is positioned mainly for defense, public safety, and agency customers rather than mainstream industrial inspection buyers. | 中 | SM010 |
| CM029 | OSHA keeps workplace-injury and fatality statistics visible, which supports a safety-justified budget case for removing humans from hazardous inspections. | 高 | SM002, SM021 |
| CM030 | Emergen cites up to 60% pipeline-inspection cost reduction from autonomous robotic systems, supporting an efficiency ROI pitch for industrial payers. | 中 | SM002 |
| CM031 | Emergen cites 340% growth in robotics-related patent applications from 2019 to 2023, signaling rapid innovation pressure across the category. | 中 | SM002 |
| CM032 | GasGoo says moving from stable prototypes to mass-produced, consistent industrial products remains a stubborn challenge across robotics. | 中 | SM020 |
| CM033 | Low-cost Chinese platforms increase price pressure, but autonomy disclaimers and missing hazardous certifications suggest price is not the only adoption variable. | 中 | SM018, SM019, SM021 |
| CM034 | The strongest evidence-backed buyer cluster for Deep Robotics today is grid and energy infrastructure operators in China and Singapore. | 高 | SM013, SM014, SM015, SM017 |
| CM035 | No reviewed public source isolates a clean Deep-specific SAM or SOM, so investors must rely on constrained sizing lenses rather than a settled market number. | 中 | SM001, SM002, SM003 |
| CM036 | The reviewed public source set does not verify that Deep Robotics holds ATEX or IECEx hazardous-site certification. | 低 | SM011, SM012, SM013, SM014, SM015, SM016, SM017 |
| CM037 | Premium Western competitors sell a combined robot-plus-orchestration stack, whereas Chinese vendors foreground hardware capability and price ladders in public materials. | 高 | SM009, SM019, SM024, SM025 |
| CM038 | A plausible adoption path starts with a pilot on repetitive or hazardous routes, followed by workflow tuning, software integration, and only then broader fleet rollout. | 高 | SM009, SM013, SM024 |
| CM039 | Confined underground spaces and stair-heavy industrial sites favor legged robots over drones or purely wheeled systems. | 高 | SM008, SM015, SM017 |
| CM040 | Western defense procurement is structurally separated from Deep Robotics' likely market because Ghost and related domestic suppliers focus on that protected channel. | 中 | SM002, SM010 |
| CP001 | Deep Robotics positions X30 and its industrial-inspection stack around substations, pipelines, tunnels, confined spaces, and other hazardous industrial sites. | 高 | SP001, SP002 |
| CP002 | The X30 product page claims IP67 protection, a -20°C to +55°C operating range, 2.5-4 hours of continuous operation, at least 10 km patrol range, and industrial stair-climbing up to 45°. | 中 | SP001 |
| CP003 | Deep Robotics says its industrial-inspection offer adds modular thermal, acoustic, and robotic-arm payloads plus autonomous route execution for hazardous industrial environments. | 中 | SP002 |
| CP004 | Lynx M20 is Deep Robotics’ wheel-legged response for industrial terrain, with claims of 80 cm step climbing, 50 cm passage width, 15 kg effective payload, and IP66 resilience. | 中 | SP003 |
| CP005 | Deep Robotics’ substation case study claims more than 50% manual-workload reduction, about 70% lower operating costs, and 98% overall inspection accuracy after deployment. | 中 | SP004 |
| CP006 | Deep Robotics says its Ningxia wind-farm project reached 96.5% recognition accuracy across more than 200 inspection missions between August 2024 and February 2025. | 中 | SP005 |
| CP007 | Deep Robotics says X30 is operating in Singapore’s underground power cable network with SP Group, making it the company’s clearest public foreign utility reference. | 中 | SP006 |
| CP008 | Deep Robotics used GITEX Global 2025 in Dubai to market Lynx M20 and X30 for security, inspection, and emergency-response use cases. | 中 | SP007 |
| CP009 | Deep Robotics frames UHV substations as a niche with extreme safety and layout challenges that justify wheel-legged inspection robots. | 中 | SP008 |
| CP010 | Unitree’s Go2-W page explicitly says some functions must still be implemented through manual operation or secondary development because of current technology and computing limits. | 中 | SP009 |
| CP011 | Unitree’s B2 product page claims more than 6 m/s top speed, 120 kg standing load, more than 40 kg continuous walking load, IP67 protection, and more than 20 km unloaded endurance. | 中 | SP010 |
| CP012 | Unitree presents B2 as an optionally wheeled-leg industrial platform with quick-change battery support and autonomous charging options. | 高 | SP010, SP011 |
| CP013 | Unitree spans both civilian and industrial quadruped segments, making it broader in hardware coverage than Deep Robotics’ more inspection-specific public posture. | 高 | SP009, SP010 |
| CP014 | Boston Dynamics positions Spot as an inspection, research, and hazardous-response robot with 14 kg payload capacity and Orbit fleet-management software. | 高 | SP012, SP013 |
| CP015 | IEEE Spectrum reported in 2020 that Spot launched at $74,500 base price, with a $34,570 enhanced-autonomy package and a $12,000 annual CARE plan. | 中 | SP013 |
| CP016 | IEEE Spectrum argued that Boston Dynamics no longer wholly owns the quadruped space because more affordable Asian platforms were already emerging. | 中 | SP013 |
| CP017 | IEEE Spectrum reported that Hyundai bought about 80% of Boston Dynamics at a roughly $1.1 billion valuation and that Boston had sold only hundreds of robots by late 2020. | 中 | SP014 |
| CP018 | ANYbotics says it was founded in 2016 as an ETH Zurich spin-off and achieved first sales in 2017 with named customers including Petronas, Shell, and BASF. | 中 | SP015 |
| CP019 | ANYbotics positions ANYmal for oil and gas, chemicals, power and utilities, mining and metals, and transport—verticals that materially overlap with Deep Robotics’ industrial-inspection thesis. | 高 | SP015, SP016 |
| CP020 | ANYmal X is publicly presented as an ATEX and IECEx Zone 1, IP67-rated inspection robot for explosive industrial environments. | 高 | SP017, SP018 |
| CP021 | Yokogawa said in February 2026 that ANYbotics had raised more than $150 million, employed a 200-expert team, and was expanding further into Asia and the Middle East. | 中 | SP018 |
| CP022 | ANYbotics’ published certification and energy-market references create a stronger publicly visible hazardous-environment trust moat than Deep Robotics currently shows. | 中 | SP017, SP018 |
| CP023 | Ghost Robotics markets itself broadly to public safety, defense, and agency customers rather than to utility-inspection buyers. | 中 | SP019 |
| CP024 | Vision 60 is presented as an IP67, modular, open-architecture quadruped with more than 3 hours continuous walking or 20-plus hours standby endurance. | 中 | SP020 |
| CP025 | Ghost Robotics’ defense page claims ISR, EOD, and CBRN mission support, plus more than 25 national-security customers and 1,000-plus robots sold or contracted. | 中 | SP021 |
| CP026 | Ghost Robotics is better understood as a procurement-protected defense specialist than as a direct civilian inspection competitor to Deep Robotics. | 高 | SP019, SP021 |
| CP027 | Precedence Research estimates the global quadruped robot market at USD 3.12 billion in 2025 and USD 17.53 billion by 2035, implying an 18.84% CAGR. | 中 | SP022 |
| CP028 | Precedence Research says North America held 46% of quadruped market share in 2025 while Asia Pacific is expected to grow fastest through 2035. | 中 | SP022 |
| CP029 | Emergen Research says defense represents about 38% of global demand and notes a USD 1.4 billion U.S. Department of Defense allocation for unmanned ground systems in fiscal 2024. | 中 | SP023 |
| CP030 | IFR reported that China installed 295,000 industrial robots in 2024, equal to 54% global share, and that China’s 15th Five-Year Plan puts robotics at the heart of its industrial system. | 中 | SP024 |
| CP031 | Gasgoo argued in January 2026 that moving from lab prototypes to mass-produced, consistent industrial robots remains a stubborn challenge and that manufacturing strength alone no longer builds a solid moat. | 中 | SP025 |
| CP032 | Flyability positions Elios 3 as an indoor and confined-space inspection drone with LiDAR mapping, making drones a credible substitute for some inspection routes. | 中 | SP026 |
| CP033 | Quadrupeds retain an advantage over drones and fixed sensors when the job requires stair-climbing, underground routing, contact inspection, or frequent route changes without site modification. | 中 | SP002, SP026 |
| CP034 | Deep Robotics has stronger public proof in utility and tunnel deployments than Unitree but weaker public hazardous-area certification than ANYbotics. | 中 | SP004, SP006, SP017 |
| CP035 | Deep Robotics’ most visible moat today is deployment-specific know-how in utilities and critical infrastructure rather than a publicly visible regulatory monopoly or exclusive channel. | 中 | SP004, SP006, SP008, SP025 |
| CP036 | Switching costs in inspection robotics look operational rather than contractual because public materials emphasize SDKs, route setup, payload integrations, and fleet tools more than exclusivity clauses. | 中 | SP002, SP010, SP012, SP020 |
| CP037 | Multi-homing is credible because buyers can split tasks across quadrupeds, drones, sensors, and manual teams without any one vendor visibly controlling the whole inspection stack. | 中 | SP002, SP020, SP026 |
| CP038 | Deep Robotics’ international expansion is real but early: Singapore deployment and Dubai event presence are meaningful, yet they do not equal the protected channel access Ghost has or the energy-market references ANYbotics claims. | 中 | SP006, SP007, SP018, SP021 |
| CP039 | Deep Robotics faces a two-sided squeeze: Unitree can pressure from below on accessible hardware breadth, while ANYbotics and Boston Dynamics pressure from above on certification and software trust. | 中 | SP010, SP012, SP017, SP018 |
| CP040 | Public pricing, revenue, and installed-base transparency remain insufficient to underwrite Deep Robotics’ competitive economics cleanly from open sources alone. | 低 | |
| CP041 | Deep Robotics’ inspection materials mention hazardous industrial zones and gas-detection use cases, but the retained public evidence did not show an ATEX or IECEx credential for Deep itself. | 中 | SP002, SP017 |
| CP042 | Boston Dynamics is strongest when the buyer values managed inspection workflows and fleet software more than lowest hardware entry cost. | 中 | SP012, SP013 |
| CP043 | ANYbotics and Deep Robotics overlap most directly in power, utilities, and broader energy inspection—especially as both surface Asia or Middle East growth narratives. | 中 | SP016, SP018, SP007 |
| CP044 | Unitree’s own autonomy disclaimer weakens its trust case for buyers who want minimum-integrator, lights-out inspection deployment. | 中 | SP009, SP002 |
| CP045 | The U.S. government-defense quadruped lane appears structurally harder for Deep Robotics to access than for Ghost Robotics because Ghost already markets directly into that buyer class. | 中 | SP021, SP023 |
| CP046 | The named field leaders in retained sources are Deep Robotics, Unitree, Boston Dynamics, ANYbotics, and Ghost Robotics, which confirms a fragmented landscape rather than a single dominant direct incumbent. | 中 | SP022 |
| CP047 | Lynx M20 and the UHV case study show Deep Robotics is already adapting toward wheel-legged operating economics instead of insisting that every industrial site wants a pure quadruped. | 高 | SP003, SP008 |
| CP048 | Product-scope breadth differs materially across the set: Deep and ANYbotics emphasize industrial inspection, Boston combines robot and software platform, Unitree emphasizes broad hardware coverage, Ghost emphasizes defense, and Flyability is an aerial substitute. | 中 | SP001, SP010, SP012, SP016, SP021, SP026 |
| CP049 | ABB Robotics illustrates a substitute path where buyers automate inspection and testing tasks with broader industrial-automation portfolios rather than mobile quadrupeds. | 中 | SP027 |
| CI001 | Deep Robotics' filing summaries put 2025 revenue at RMB 337 million, up from RMB 103 million in 2024 and RMB 50.11 million in 2023. | 中 | SI007, SI008 |
| CI002 | Deep Robotics reported first-time profitability in 2025 with RMB 28.68 million of net profit after a 2024 net loss. | 中 | SI007, SI008, SI009 |
| CI003 | Embodied intelligent robots contributed about RMB 322.38 million, or 95.69%, of core business revenue in 2025. | 中 | SI007 |
| CI004 | The Jueying X series sold 681 units in 2025 for RMB 196 million of revenue at an average selling price of about RMB 287.5 thousand. | 中 | SI007 |
| CI005 | The Jueying Lite series sold 1,850 units in 2025 for RMB 51.3 million of revenue at an average selling price of about RMB 27.7 thousand. | 中 | SI007 |
| CI006 | The Lynx M series sold 377 units in 2025 for RMB 74.49 million of revenue at an average selling price of about RMB 197.6 thousand. | 中 | SI007 |
| CI007 | Deep Robotics disclosed one DR humanoid sale in 2025 at RMB 823 thousand, versus three humanoid sales in 2024 at RMB 390.4 thousand each and negative gross margin. | 中 | SI007 |
| CI008 | Company-wide gross margin reached 52.83% in 2025, up materially from the early commercialization phase. | 中 | SI008, SI009 |
| CI009 | Overseas revenue reached CNY 61 million in 2025, representing just over 18% of total income. | 中 | SI009 |
| CI010 | Deep Robotics spent RMB 84.3 million on R&D in 2025, or about 25% of revenue. | 中 | SI009 |
| CI011 | Annual robot sales volume rose from 529 units in 2023 to 3,936 units in 2025. | 中 | SI008 |
| CI012 | Before the filing, management said 2024 revenue grew more than 100% year over year and targeted 10,000 robot shipments in 2025. | 中 | SI012 |
| CI013 | Public sources show Deep Robotics sells quadrupeds, wheel-legged robots, humanoids, components, and supporting services, with robot-body revenue dominating the mix. | 中 | SI001, SI007 |
| CI014 | Deep Robotics positions its products around hazardous industrial workflows such as power inspection, industrial maintenance, security patrol, and emergency response. | 中 | SI001, SI025 |
| CI015 | CnTechPost reported that nearly 80% of Deep Robotics' revenue comes from complex industrial scenarios rather than consumer markets. | 中 | SI008 |
| CI016 | Li Chao said the power-inspection opportunity is repeatable because China has about 20,000 substations with highly repetitive rules-based inspection work. | 中 | SI013 |
| CI017 | Li Chao said a robot is merely usable when a customer places an order, but is truly good-to-use when it drives repeat purchases of 100, 300, or 500 units. | 中 | SI013 |
| CI018 | Deep Robotics says a substation can be modeled and route-planned within one to two days, with 500-point configuration completed in two to four weeks. | 中 | SI004 |
| CI019 | In one official substation case study, the X30 achieved 98% inspection accuracy and completed a full-station patrol in under 35 minutes versus more than one hour manually. | 中 | SI004 |
| CI020 | The same substation case study claims more than 50% reduction in manual workload and about 70% reduction in operating costs. | 中 | SI004 |
| CI021 | Deep Robotics' Singapore deployment is expected to save SP Group more than 480 manual inspection hours per year. | 中 | SI005, SI015 |
| CI022 | TechNode reported that the Singapore deployment required three to four months of safety assessments before it went live. | 中 | SI015 |
| CI023 | Zhu Qiuguo said Deep Robotics' equal-performance robot dogs cost only one-third to one-half as much as Boston Dynamics. | 中 | SI012 |
| CI024 | Interesting Engineering quoted Deep Robotics' Korean sales lead saying the company's products are 20-30% cheaper on average than Boston Dynamics and Rainbow Robotics. | 中 | SI018 |
| CI025 | Independent coverage says Deep Robotics has not publicly disclosed DR02 autonomy, operating duration, or cost. | 中 | SI018, SI019 |
| CI026 | Yicai reported a July 2025 financing of CNY 500 million led by Fortune Venture Capital and CRHC Fund to strengthen quadrupeds and accelerate humanoid development. | 中 | SI010 |
| CI027 | Yicai reported a December 2025 Series C of more than CNY 500 million led by China Merchants Bank International and China Asset Management, with China Telecom and China Unicom funds also participating. | 中 | SI011 |
| CI028 | 36Kr's IPO-history article says Deep Robotics had completed eight financing rounds and then added a pre-IPO round of hundreds of millions of yuan led by the National AI Industry Fund with JD.com follow-on investment. | 中 | SI014 |
| CI029 | Newo reported that Deep Robotics filed for IPO tutoring on December 23, 2025 after signing its advisory agreement on December 8, 2025. | 中 | SI020 |
| CI030 | The Eastmoney filing page for Hangzhou Deep Robotics contains a prospectus-draft PDF link, and multiple May 2026 reports say the STAR Market application had already been accepted. | 中 | SI006, SI007, SI008 |
| CI031 | Deep Robotics plans to raise RMB 2.503 billion in the STAR Market IPO. | 中 | SI007, SI008, SI009 |
| CI032 | Public prospectus summaries say the IPO proceeds are earmarked for embodied-AI algorithm and model R&D, robot body and solution R&D, industrialization, and manufacturing-base construction. | 中 | SI007, SI008, SI009 |
| CI033 | The filing summaries say founder Zhu Qiuguo directly owns 15.61% of Deep Robotics and controls 31.05% of voting rights including the employee platform. | 中 | SI007, SI009 |
| CI034 | The prospectus summaries list multiple state and industry funds among large shareholders, indicating a capital base with meaningful state-linked participation. | 中 | SI007, SI014 |
| CI035 | Deep Robotics remained a private company as of 2026-06-21 because its STAR Market application had only been accepted, not completed into a public listing. | 中 | SI006, SI008 |
| CI036 | Management said recent funding would be used for quadruped production-line expansion, humanoid R&D, and recruitment of high-end talent. | 中 | SI010, SI012 |
| CI037 | The company warned in the prospectus, as summarized by CnTechPost, that continued R&D investment and rapid technological iteration could make short-term profitability volatile. | 中 | SI008 |
| CI038 | Tutoring-stage coverage in late 2025 put Deep Robotics' private valuation at roughly CNY 8 billion, or about USD 1.1 billion. | 中 | SI020 |
| CI039 | Post-filing 2026 coverage describes Deep Robotics as worth more than CNY 10 billion, or north of USD 1.5 billion. | 中 | SI009 |
| CI040 | Humanoids Daily argues that the current humanoid race is defined by multi-billion-dollar war chests needed for mass manufacturing and AI development. | 中 | SI021 |
| CI041 | Humanoids Daily says Unitree is a rare profitable robotics peer and is targeting up to a USD 7 billion IPO valuation, setting a tougher commercialization benchmark for Deep Robotics. | 中 | SI021, SI022 |
| CI042 | SemiAnalysis places the quadruped hardware market under the lead of Boston Dynamics and Unitree, with Deep Robotics and ANYbotics behind the top two and Deep Robotics at roughly 600 deployments in 40-plus countries. | 中 | SI023 |
| CI043 | SemiAnalysis says some oil-and-gas and other data-sensitive buyers resist Chinese-origin quadrupeds on security grounds, which can limit export mix quality for Deep Robotics. | 中 | SI023 |
| CI044 | Li Chao identified price cuts and talent poaching from new entrants as live commercial pressure points on established robotics startups. | 中 | SI013 |
| CI045 | No reviewed public source discloses cash on hand, monthly burn, runway, debt, customer concentration by revenue, realized discounting, CAC, payback, NRR, or warranty burden. | 中 | SI006, SI007, SI008, SI009, SI013 |
| CI046 | Deep Robotics' current revenue quality is still driven mainly by industrial quadrupeds and wheel-legged robots rather than by humanoids. | 中 | SI007, SI008, SI017 |
| CI047 | Because humanoids remain exploratory while the company plans a large R&D and manufacturing build-out, the strong 2025 margin profile may not hold if future mix shifts toward capital-heavier programs. | 中 | SI007, SI009, SI021 |
| CI048 | Deep Robotics' official and management narratives frame robot adoption around hazardous-work replacement and uptime gains rather than around pure labor substitution. | 中 | SI004, SI013 |
| CI049 | Li Chao said overseas buyers such as Singapore are more willing to adopt robots because local labor costs are materially higher than in China. | 中 | SI013 |
| CI050 | Company and press-release sources say Deep Robotics holds a leading position in B2B quadruped applications, but the underlying IDC methodology is not publicly reproduced in the reviewed material. | 低 | SI012, SI024 |
| CI051 | In a June 2025 UHV-substation case study, Deep Robotics said multiple Lynx M20 units can coordinate across substations covering more than 500 acres while maintaining defect-detection accuracy above 95%. | 中 | SI026 |
| CI052 | In a February 2025 wind-farm project, Deep Robotics said the X30 completed more than 200 inspection missions across 349 checkpoints with 96.5% recognition accuracy. | 中 | SI027 |
| CI053 | BigGo Finance reported that investor valuation-adjustment rights would automatically revive if the IPO process stalls, creating a capital-structure risk if public financing is delayed. | 中 | SI028 |
| CI054 | 100EC reported that Deep Robotics plans to issue at least 82.98 million shares, at no less than 18% of post-issue share count, to fund four embodied-intelligence projects. | 中 | SI029 |
| CE001 | Deep Robotics publicly presents itself as a developer of quadruped robots, humanoid robots, and core robotic components rather than a single-product vendor. | 中 | SE001, SE015 |
| CE002 | X20 is positioned as an industrial quadruped for patrol inspection with IP66 protection, modular peripherals, and autonomous inspection functions. | 中 | SE002 |
| CE003 | X30 is the more rugged flagship quadruped, with IP67 protection, -20°C to 55°C operation, 45° stair climbing, and quick battery replacement. | 中 | SE003, SE030 |
| CE004 | Lynx M20 is a distinct wheel-legged industrial platform with 33 kg weight, 15 kg effective payload, 80 cm single-step clearance, dual 96-line LiDAR, and 45° slope capability. | 中 | SE004, SE008 |
| CE005 | Lynx S10 is the smaller wheel-legged SKU, weighing at most 20 kg while claiming 8 m/s speed, 50 cm obstacle clearance, and IP66 protection for narrow-space work. | 中 | SE005 |
| CE006 | DR02 is publicly framed as an industrial humanoid product with IP66 protection, -20°C to 55°C operation, and modular quick-detach limbs for lower O&M burden. | 中 | SE006, SE015 |
| CE007 | DR01 remains best understood as a debut-stage humanoid demonstrator built around proprietary J60 and J100 joints rather than a publicly proven production deployment. | 中 | SE014, SE025, SE027 |
| CE008 | Lite3 is explicitly positioned for academic research, robotics education, and reinforcement-learning experimentation with open SDK and API access. | 中 | SE016, SE023 |
| CE009 | No reviewed official Deep Robotics source names an L20 SKU, while official surfaces repeatedly expose X20 and Lynx M20 as the relevant product names. | 中 | SE002, SE004, SE016 |
| CE010 | The portfolio is segmented into X-series industrial quadrupeds, Lynx wheel-legged robots, Lite3 research robots, and the DR humanoid line. | 中 | SE001, SE015, SE016 |
| CE011 | Deep Robotics uses a public "perception-decision-action" framework to describe its full-stack technical system. | 中 | SE001, SE015 |
| CE012 | DeepVLA 1.0 combines open-vocabulary semantic point clouds, large language models, and a topological road network for door-to-door multi-floor navigation. | 中 | SE015 |
| CE013 | Lynx M20 publishes the clearest onboard architecture, including dual industrial processors, dual 96-line LiDARs, dual wide-angle cameras, Gigabit Ethernet, and OTA upgrades. | 中 | SE004 |
| CE014 | X20's public feature set includes precise navigation, dynamic obstacle avoidance, terrain recognition, and human-robot interaction. | 中 | SE002 |
| CE015 | X30's integrated perception is marketed as able to navigate and operate in darkness, strong light, flicker, and even without any light source. | 中 | SE003 |
| CE016 | Lite3's public stack includes ROS1 and ROS2 support, motion SDK access, perception-development manuals, LiDAR SLAM examples, and sim-to-real reinforcement-learning workflows. | 中 | SE016, SE019, SE023 |
| CE017 | The DeepRoboticsLab GitHub organization exposes at least 20 repositories spanning models, RL training, ROS bridges, lidar secondary development, and joint SDK tooling. | 中 | SE017 |
| CE018 | The rl_training repository provides IsaacLab-based reinforcement-learning environments for Lite3 and M20 plus ONNX export and distributed training workflows. | 中 | SE018 |
| CE019 | The sdk_deploy repository documents sim-to-sim and sim-to-real deployment for Lite3 and M20, including a Lite3 UDP-to-ROS2 transfer layer. | 中 | SE019 |
| CE020 | The deep_robotics_model repository publishes MJCF, URDF, and USD models for Lite3, M20, X30, and DR02 variants. | 中 | SE020 |
| CE021 | The Deep_Motor_SDK repository exposes CAN-based J60 control and debugging utilities but explicitly limits compilation and runtime support to Linux. | 中 | SE021 |
| CE022 | Lite3 is sold in at least four public variants—Basic, Venture, Pro, and LiDAR—with different perception, payload, and endurance trade-offs. | 中 | SE022, SE023 |
| CE023 | Utility substation inspection is the best-documented production workflow, with robots collecting meter, thermal, and equipment-state data in dangerous terrain. | 中 | SE009, SE010, SE015 |
| CE024 | Deep Robotics says its robot dogs now operate in nearly thirty China Southern Power Grid substations with about 70 percent O&M cost reduction and more than 50 percent efficiency improvement. | 中 | SE015 |
| CE025 | Official UHV case studies claim more than 95 percent inspection-detection accuracy and 96.5 percent recognition accuracy in a Ningxia Gobi deployment. | 中 | SE009, SE010 |
| CE026 | The Singapore Power SPock deployment shows X30 patrolling a 40 km underground power tunnel, detecting defects, and being considered for wider-scale rollout. | 中 | SE013, SE028 |
| CE027 | Emergency-response case studies position Lynx M20 and X30 as mobile sensor carriers for gas, thermal, and visual intelligence in dangerous zones. | 中 | SE011, SE012 |
| CE028 | Security-patrol cases bundle loudspeaker, identification, data collection, and following functions into the robot workflow. | 中 | SE015 |
| CE029 | Product ruggedization is specifically tiered across the line, with X20 at IP66, X30 at IP67, Lynx S10 at IP66, Lynx M20 at IP66, and DR02 at IP66. | 中 | SE002, SE003, SE004, SE005, SE006 |
| CE030 | DR02's modular quick-detach design is positioned as an O&M feature intended to reduce downtime and maintenance cost. | 中 | SE006 |
| CE031 | Lynx M20 combines hot-swappable batteries and optional autonomous charging, but some public features such as omnidirectional obstacle avoidance and point-cloud surround view are still flagged as OTA-enabled. | 中 | SE004, SE009, SE010 |
| CE032 | The public support terms exclude warranty coverage for unauthorized modification, unofficial repair, water or debris ingress, chemical damage, strong interference, and some extreme-environment use. | 中 | SE007, SE023 |
| CE033 | The reviewed after-sales terms explicitly say they apply only in mainland China, leaving overseas service coverage undocumented on the public surface. | 中 | SE007 |
| CE034 | The public privacy policy allows collection of logs, time, location, device configuration, and IP address information and allows disclosure under legal or government requirements. | 中 | SE007, SE023 |
| CE035 | The 2025 product roadmap visibly advanced from Lynx M20 launch to DR02 launch, DeepVLA 1.0 release, and the creation of an embodied-AI pilot production base. | 中 | SE008, SE015 |
| CE036 | Management says Series C funds will continue to increase R&D spending, expand production capacity, hire more talent, and deepen both quadruped and humanoid development. | 中 | SE015 |
| CE037 | The public developer ecosystem is materially deeper for Lite3 and M20 than for X20, X30, and DR02. | 中 | SE017, SE018, SE019, SE020, SE021, SE023 |
| CE038 | DR02 commercialization remains early because the reviewed corpus shows product claims and expo-style positioning but no named production deployment or public SDK package. | 中 | SE006, SE014, SE025 |
| CE039 | Deep Robotics is publicly steering toward clustered operations that combine robot dogs with drones and fixed sensing for broader autonomous coverage. | 中 | SE010, SE012 |
| CE040 | The main publicly visible product risk is supportability and governance rather than locomotion, because service geography, data handling, and warranty boundaries are documented more weakly than motion capability. | 中 | SE007, SE021, SE015 |
| CE041 | Independent coverage also corroborates that Lynx is a wheel-legged robot capable of steep inclines and bipedal-like maneuvers, reinforcing the mechanical distinction from the X-series. | 中 | SE026, SE030 |
| CE042 | Founder-linked locomotion research on terrain reconstruction shows a learning-based path to precise foothold selection using onboard depth images and proprioception rather than heavyweight external mapping. | 中 | SE024 |
| CE043 | Deep Robotics' portfolio segmentation follows workflow differences more than robot size alone, with industrial inspection products separated from research and education products. | 中 | SE002, SE003, SE004, SE023 |
| CU001 | Deep Robotics' public customer base spans utilities, public-safety and security operators, emergency-response organizations, and research or education buyers. | 中 | SU001, SU007, SU008 |
| CU002 | Utility accounts appear to buy for hazardous inspection workflows while frontline inspectors and maintenance teams are the day-to-day users. | 中 | SU002, SU003, SU016 |
| CU003 | Deep Robotics' X30 robot nicknamed SPock is operational in Singapore Power Group's underground power transmission cable network. | 中 | SU002, SU010, SU011, SU012, SU013 |
| CU004 | The SP Group deployment is described as a pilot with wider-scale rollout planned rather than a fully disclosed network-wide rollout. | 中 | SU002, SU010, SU011, SU012, SU013 |
| CU005 | Public sources tie the SP Group deployment to an expected saving of about 480 manual inspection hours per year. | 中 | SU002, SU010, SU011, SU012, SU013 |
| CU006 | Deep Robotics says its robots now serve nearly 30 China Southern Power Grid substations and have reduced O&M costs by about 70% while improving efficiency by over 50%. | 中 | SU007 |
| CU007 | The Chengzhong Branch of the Xining Public Security Bureau is a named deployment using Deep Robotics' patrol solution with loudspeaker, identification, and data-collection functions. | 中 | SU007 |
| CU008 | Yiwu International Trade City is a named security deployment where Deep Robotics says X30 serves as a mobile security outpost for patrol, fire-hazard identification, and abnormal-activity warning. | 中 | SU007 |
| CU009 | Certis industrial park in Singapore is a named security deployment, but the current public proof is a company-distributed press release rather than independent customer testimony. | 中 | SU014 |
| CU010 | Zhejiang Fire Brigade is verified in the public record as a joint-drill or scenario-validation user rather than as a clearly disclosed recurring production customer. | 中 | SU005, SU025 |
| CU011 | Deep Robotics says the Jueying X30 established Saudi Arabia's first fully automated outdoor inspection system at NEOM, but this deployment is not independently corroborated in the chapter source pack. | 中 | SU007 |
| CU012 | The company's reported North America storage-park deployment with HawkRobo is also official-only proof and should not be treated as independently verified customer traction. | 中 | SU007 |
| CU013 | Hoh Xil wildlife monitoring is a named application proof point, but the public record does not disclose whether it is a commercial contract, a public-agency program, or a research collaboration. | 中 | SU007 |
| CU014 | Official and third-party sources consistently cite Tsinghua, Peking University, the University of Edinburgh, and UCL as Lite-series research or education users. | 中 | SU008, SU018, SU019, SU020 |
| CU015 | References to Tencent and Alibaba show R&D-oriented platform adoption or interest around Lite3, not production-style industrial deployment with disclosed outcomes. | 中 | SU018, SU019 |
| CU016 | Deep Robotics' 100+ clients in 26 Chinese provinces and cities is a company-claimed reach metric rather than a current audited customer-count disclosure. | 中 | SU001 |
| CU017 | Later official communications expand the breadth story to 44 countries and regions and 600+ industry scenarios, but those metrics still do not equal a run-date paying-customer count. | 中 | SU006, SU007 |
| CU018 | SP Group is the chapter's strongest named-customer proof because it combines a named account, workflow detail, localization detail, and independent corroboration. | 高 | SU002, SU010, SU011, SU012, SU013 |
| CU019 | China Southern Power Grid is the next-strongest named account, but its public proof quality is lower than SP Group because scale and ROI claims are still company-reported. | 中 | SU007 |
| CU020 | Xining, Yiwu, NEOM, Hoh Xil, and the North America storage park should be treated as official-only named proofs rather than independently validated commercial wins. | 中 | SU007 |
| CU021 | The named-customer corpus is still concentrated in utilities and public-sector or security workflows rather than a broad private-enterprise base. | 中 | SU001, SU007, SU016 |
| CU022 | The best public repeat-usage proxies are utility-specific, including China Southern Power Grid's multi-substation disclosure, Jiemian's State Grid framework reporting, and SP Group's planned wider rollout. | 中 | SU007, SU016, SU002, SU010 |
| CU023 | Jiemian reported that State Grid signed a 2023 framework deal with Deep Robotics covering more than 1,000 substations and that deployment projects exceeded 600 by 2025. | 中 | SU016 |
| CU024 | State Grid's 2026 procurement plan for 5,000 quadruped inspection robots supports strong future demand in the utility segment but does not disclose any committed Deep Robotics share. | 中 | SU016, SU017 |
| CU025 | Deep Robotics does not publicly disclose NRR, GRR, churn, renewal rates, or top-customer concentration metrics in the chapter source pack. | 中 | SU001, SU006, SU007, SU008 |
| CU026 | SP Group's workflow appears operationally embedded because inspectors continue training SPock with defect images and simulated scenarios after deployment. | 中 | SU011, SU012, SU013 |
| CU027 | The Singapore deployment required months of safety assessment and changes to sensors, algorithms, and navigation logic, showing that customer expansion is services-heavy rather than plug-and-play. | 中 | SU010, SU011, SU012, SU013 |
| CU028 | The public CN after-sales terms are explicitly mainland-China-only, remote-diagnosis first, and often require the customer to bear shipping or paid-service costs outside free-warranty conditions. | 中 | SU009 |
| CU029 | Deep Robotics' published privacy terms say the company may collect usage logs, time, location, device configuration, and IP-address data and may disclose information when required by legal or public-authority processes. | 中 | SU009 |
| CU030 | The chapter did not identify independent customer-satisfaction datasets or complaint platforms that would let an investor benchmark Deep Robotics' retention like a SaaS or review-led business. | 中 | SU008, SU009 |
| CU031 | Independent evidence in the public source pack is strongest for utilities and materially weaker for security, emergency-response, and extreme-environment customers. | 中 | SU010, SU011, SU012, SU013, SU014 |
| CU032 | Research and education buyers purchase Lite3 for openness, SDK access, modularity, and affordability, while utilities purchase ruggedized workflow automation. | 中 | SU003, SU008, SU018, SU021 |
| CU033 | Generation Robots positions Lite3 primarily for universities, laboratories, engineering schools, and R&D teams, reinforcing that the product serves a long-tail developer market. | 中 | SU021 |
| CU034 | The mixed customer base reduces dependence on a single buyer archetype, but public sources do not disclose revenue mix by utilities, security, education, or other segments. | 中 | SU001, SU007, SU008 |
| CU035 | A 2026 U.S. policy push to restrict Chinese-made quadruped and humanoid robots in federal or covered settings creates a real category-level customer-acquisition risk for Chinese vendors. | 中 | SU022, SU023, SU024 |
| CU036 | The U.S. policy risk is category-level rather than company-specific and does not prove a current Deep Robotics customer loss as of the run date. | 中 | SU022, SU023, SU024 |
| CU037 | Verified named deployments materially outweigh pure logo references in this chapter, but only SP Group combines named customer identity, independent corroboration, and a disclosed operating outcome. | 中 | SU002, SU010, SU011, SU012, SU013, SU007 |
| CU038 | University and enterprise names such as UCL, Edinburgh, Tencent, and Alibaba are useful adoption proof for Lite3 but do not establish production deployment scale, retention, or revenue quality. | 中 | SU018, SU019, SU020 |
| CU039 | Deep Robotics' customer story is credible on real-world adoption but still underdisclosed on exact paying-customer count, contract sizes, and revenue concentration. | 中 | SU001, SU006, SU007, SU016 |
| CU040 | The public evidence supports Deep Robotics more strongly as a workflow-localization vendor in utilities and patrol/security than as a horizontally adopted global robotics platform. | 中 | SU001, SU002, SU007, SU008 |
| CU041 | The Certis case shows the security solution can handle closed-loop intervention and charging-pile monitoring rather than only passive patrol. | 中 | SU014 |
| CU042 | The UHV and emergency case studies prove terrain access, sensor payload, and hazard response more strongly than they prove named-customer commercial depth. | 中 | SU003, SU004, SU005, SU025 |
| CU043 | Deep Robotics says ICRA 2026 generated European university and lab discussions about joint projects and pilot programs, but those are pipeline signals rather than booked customer disclosures. | 中 | SU008 |
| CU044 | Publicly named deployment references span Asia, the Middle East, North America, and Europe-linked academia, but independently corroborated commercial wins outside Singapore remain scarce. | 中 | SU007, SU010, SU011, SU012, SU013, SU018, SU019 |
| CU045 | The public corpus supports a clear proof ladder in which China Southern and SP Group look closest to scaled operations, Certis, Xining, Yiwu, and NEOM look operational but less disclosed, and Zhejiang Fire Brigade remains drill evidence. | 中 | SU002, SU007, SU014, SU005, SU025 |
| CR001 | Public Law 118-159 section 164 prohibits the Department of Defense from operating or procuring covered LiDAR technology and systems that incorporate it from covered foreign countries, including China, effective June 30, 2026. | 中 | SR016, SR017 |
| CR002 | Public Law 118-159 section 1078 requires a study of covered foreign unmanned ground vehicle systems and allows a procurement or operation ban to spring into effect one year after an adverse recommendation. | 中 | SR016, SR017 |
| CR003 | Covington's FY2025 NDAA analysis says section 805 of the FY2024 NDAA bars DoD from contracting with 1260H-listed companies from June 2026 and bars items containing their products or components in June 2027. | 中 | SR016 |
| CR004 | The June 10, 2026 Department of Defense notice published on GovInfo names Hangzhou Yushu Technology Co., Ltd. (Unitree) as a Chinese military company under section 1260H. | 中 | SR019 |
| CR005 | Deep's X30, industrial-inspection, and UHV marketing all emphasize LiDAR, thermal, gas-detection, and autonomous inspection workflows in critical-infrastructure environments. | 中 | SR001, SR002, SR006 |
| CR006 | Because Deep sells Chinese-origin inspection robots into critical-infrastructure workflows while U.S. rules are tightening around Chinese LiDAR and unmanned ground systems, overseas public-sector and defense-adjacent TAM can narrow before any Deep-specific designation exists. | 中 | SR002, SR016, SR017, SR019 |
| CR007 | The April 9, 2026 BIS final rule extended the trigger date and application deadline for approved IC designer status to December 31, 2026 under the Export Administration Regulations. | 中 | SR018 |
| CR008 | SemiAnalysis says advanced quadruped deployments increasingly depend on Nvidia Jetson-class edge compute, while Deep's own materials market AI-driven multi-sensor autonomy, making future compute-control tightening a product and export risk. | 中 | SR014, SR001, SR002 |
| CR009 | ANYbotics publicly markets ANYmal X as ATEX and IECEx Zone 1 certified, but the retained Deep public materials did not disclose a comparable system-level ATEX or IECEx credential for Deep platforms. | 中 | SR024, SR001, SR002 |
| CR010 | That certification gap leaves Deep better positioned for substations, tunnels, and non-Ex industrial sites than for the highest-trust hazardous-process plants that ANYbotics and Yokogawa are targeting together. | 中 | SR024, SR025, SR002 |
| CR011 | Deep's GITEX and AccessWire materials explicitly market security, public safety, emergency response, and public-security patrol use cases alongside industrial inspection. | 中 | SR005, SR013 |
| CR012 | The X30 official product page lists IP67 ingress protection, 2.5 to 4 hours of operation, and LiDAR plus infrared multi-sensor inspection capability. | 中 | SR001 |
| CR013 | Deep's industrial-inspection page describes route mapping, autonomous patrols, automatic return and charging, SCADA/API integration, digital-twin workflows, and custom payload support. | 中 | SR002 |
| CR014 | In its Zhejiang substation reference, Deep claims 98% overall inspection accuracy, a full-station inspection time under 35 minutes, more than 50% manual-work reduction, and about 70% operating-cost reduction. | 中 | SR003 |
| CR015 | Official and TechNode Singapore coverage say SPock patrols a 40-kilometer tunnel and can save more than 480 manual-inspection hours annually after a three-to-four-month safety and localization cycle. | 中 | SR004, SR012 |
| CR016 | Deep's Singapore deployment required months of local testing and algorithm, sensor, and navigation refinement, showing that foreign utility rollouts are service-heavy rather than plug-and-play. | 中 | SR004, SR012 |
| CR017 | Deep's Singapore and South Korea article says the company has served more than 100 clients in 26 Chinese provinces and cities, but only a small number of identified overseas industry projects are publicly visible. | 中 | SR011 |
| CR018 | The public named-reference set clusters around State Grid, China Southern Power Grid, power tunnels, wind farms, and related state-linked infrastructure accounts. | 中 | SR003, SR006, SR011, SR031 |
| CR019 | The Lynx M20 and GITEX materials position Deep for UHV substations, gas-leak checks, perimeter patrols, smart-city security, and emergency firefighting. | 中 | SR006, SR005 |
| CR020 | The UHV materials highlight specialized operational requirements such as 50-centimeter trench crossing, multi-robot coordination, autonomous charging, and predictive reporting, implying site-criticality drives deployment complexity. | 中 | SR006 |
| CR021 | Ghost Robotics markets Vision 60 for defense, public safety, and perimeter security, so U.S. and allied defense lanes already have a local incumbent with a better procurement fit than a Chinese-origin quadruped vendor. | 中 | SR027, SR019 |
| CR022 | Yokogawa and ANYbotics are integrating robot-management software for energy, power, metals, and hazardous areas, showing a rival channel stack that combines certification, software, and global industrial reach. | 中 | SR025, SR024 |
| CR023 | ANYbotics says it has hundreds of industry leaders as customers plus more than $150 million of funding and a 200-expert team, raising the bar for Deep in high-trust hazardous inspection deployments. | 中 | SR025 |
| CR024 | Unitree's B2 official page advertises 6 m/s speed, more than 4 hours of walking with a 20-kilogram load, and more than 5 hours unloaded, showing how a lower-cost domestic rival can pressure Deep on raw hardware benchmarks. | 中 | SR026 |
| CR025 | SemiAnalysis says Deep is priced more than 30% above comparable industrial Unitree offerings while Unitree holds about 70% of global quadruped volume and roughly 10 times the next competitor's 2023 shipments. | 中 | SR014 |
| CR026 | The Unitree research document characterizes Deep as strong in Chinese power-grid inspection but still behind Unitree in consumer, academic, and humanoid scale, reinforcing the risk that the price leader can move upmarket. | 中 | SR029, SR014 |
| CR027 | CnTechPost, Sina, and Humanoids Daily all point to Deep's 2025 revenue near RMB 337 million, gross margin above 52.8%, and first profitable year. | 中 | SR020, SR021, SR030 |
| CR028 | Prospectus-linked reporting says rapid technological iteration and continued R&D expansion could cause Deep's profitability to fluctuate even after the 2025 inflection. | 中 | SR020, SR030 |
| CR029 | Humanoids Daily says planned IPO proceeds are earmarked for embodied-AI algorithms, large models, new robot bodies, and manufacturing expansion, so future capital needs sit more in humanoids than in the mature quadruped line. | 中 | SR030 |
| CR030 | DR02 official materials show Deep is now pursuing an IP66 all-weather industrial humanoid with modular quick-replacement design, materially broadening product scope beyond quadrupeds and wheel-legged inspection robots. | 中 | SR009 |
| CR031 | Humanoids Daily reports Deep spent about 25% of revenue on R&D in 2025 and kept roughly 40% of employees in research roles, confirming that the company remains R&D-heavy despite profitability. | 中 | SR030 |
| CR032 | Caixin says Deep started IPO tutoring in December 2025, aimed for review completion by mid-2026, and was competing for investor attention with Unitree and AgiBot in the same period. | 中 | SR031, SR022 |
| CR033 | The separate Humanoids Daily IPO-rush piece argues that leading Chinese robotics candidates still have to prove mass-production economics and clear ROI, so the 2026 listing window is crowded but not fully underwritten. | 中 | SR022 |
| CR034 | The Robot Report's AW 2026 coverage shows Chinese humanoid rivals, Unitree, and Boston Dynamics-backed Hyundai all pushing into Korea's factory-automation budget at the same time. | 中 | SR023 |
| CR035 | State Grid-style utility concentration is a strategic strength today but also makes public proof less portable if Chinese SOE procurement priorities or grid budgets change. | 中 | SR011, SR031, SR003 |
| CR036 | The Singapore deployment runs through local partner EGP and required localization work, so overseas scaling depends on partner capability and field-service execution rather than on product performance alone. | 中 | SR011, SR004, SR012 |
| CR037 | Deep's company page says it simultaneously develops quadrupeds, humanoids, core components, manufacturing, sales, and service under one full-stack framework. | 中 | SR010 |
| CR038 | That breadth means the company must execute hardware reliability, AI autonomy, field deployment, and compliance at once, increasing organizational complexity compared with a single-product inspection vendor. | 中 | SR010, SR009, SR002 |
| CR039 | Caixin and Humanoids Daily both show that Deep still relies on new capital to fund manufacturing and humanoid expansion, so management bandwidth and talent retention remain gating resources rather than solved problems. | 中 | SR031, SR030 |
| CR040 | SemiAnalysis reports that data-sensitive operators in oil and gas, utilities, and semiconductor fabs have become wary of Chinese quadrupeds after hardware-security concerns emerged at a peer, creating a trust hurdle beyond pure specs. | 中 | SR014 |
| CR041 | Deep's inspection pages market hazardous zones and optional explosion-proof sensors, but the retained public pack still lacks a system-level certificate or detailed cyber-control disclosure that would let an investor treat those mitigations as mature. | 中 | SR002, SR024, SR005 |
| CR042 | The 1260H notice and FY2025 NDAA text show that U.S. screening criteria are now formalized, justified, and updated at least annually, so an indirect compliance issue can become a procurement blocker quickly. | 中 | SR019, SR017 |
| CR043 | Caixin and Humanoids Daily both tie Deep's financing path to public markets and continuing private rounds, so any IPO delay would directly pressure the cash plan for humanoid and manufacturing expansion. | 中 | SR031, SR030 |
| CR044 | The combination of premium positioning versus Unitree, service-heavy deployments, and missing public concentration data means Deep's 2025 margin profile is not yet proven durable at international scale. | 中 | SR014, SR020, SR021 |
| CR045 | No retained public source directly confirms whether Deep itself appears on a U.S. restricted-party list, and no retained public source provides a system-level ATEX or IECEx certificate for a current Deep platform. | 中 | |
| CV001 | As of 2026-06-21, Deep Robotics remains private even though it has completed IPO tutoring and has an accepted STAR Market filing. | 中 | SV001, SV014 |
| CV002 | Caixin described Deep Robotics at about RMB 8 billion (about $1.1 billion) when the company began IPO tutoring in December 2025. | 低 | SV002 |
| CV003 | Humanoids Daily later framed Deep Robotics at north of CNY 10 billion (about $1.5 billion) in May 2026 filing-era coverage. | 低 | SV027 |
| CV004 | Pandaily reported a post-listing valuation expectation around $3.5 billion alongside the planned 2.503 billion yuan IPO raise. | 低 | SV016 |
| CV005 | BigGo said prospectus terms could reinstate investor VAM rights if the IPO process stalls, creating cap-table and governance overhang risk. | 中 | SV031 |
| CV006 | Prospectus-based sources consistently put Deep Robotics 2025 revenue at RMB 337 million. | 中 | SV001, SV015, SV027, SV031 |
| CV007 | The same filing-era sources say 2025 was Deep Robotics’ first profitable year, with net profit around RMB 28.68 million. | 中 | SV001, SV015, SV027, SV031 |
| CV008 | Deep Robotics’ 2025 gross margin reached 52.83% in filing-era reporting. | 中 | SV001, SV015, SV027, SV031 |
| CV009 | Deep Robotics spent roughly RMB 84.3 million on R&D in 2025, or about one quarter of revenue. | 中 | SV015, SV027, SV031 |
| CV010 | Humanoids Daily said overseas revenue reached roughly CNY 61 million, or just over 18% of total income. | 低 | SV027 |
| CV011 | Deep Robotics’ 2023-2025 revenue path of RMB 50 million, RMB 103 million, and RMB 337 million implies a very steep recent growth curve. | 中 | SV015, SV031 |
| CV012 | More than 95% of 2025 core business income came from embodied intelligent robot products. | 中 | SV027, SV031 |
| CV013 | Filing-era summaries say industry applications contributed roughly 80% of revenue and that Deep Robotics ranked first globally in quadruped industry-application revenue in 2025. | 中 | SV015, SV031, SV032 |
| CV014 | The Jueying X series sold 681 units in 2025, generated about RMB 196 million of revenue, and reached roughly RMB 287.5 thousand ASP with 54.35% gross margin. | 中 | SV031 |
| CV015 | The Lynx M series sold 377 units in its 2025 launch year and generated about RMB 74.49 million of revenue with 57.50% gross margin. | 中 | SV031 |
| CV016 | Humanoid commercialization was still embryonic in 2025 because filing-era sources disclosed only one sold DR humanoid unit. | 中 | SV031 |
| CV017 | That one 2025 DR humanoid sale carried an ASP around RMB 823 thousand, showing price potential but not repeatable demand. | 中 | SV031 |
| CV018 | Deep Robotics has real field proof in power inspection and overseas utility tunnels rather than only lab or expo evidence. | 中 | SV005, SV006, SV007 |
| CV019 | The Singapore Power deployment is described as the first Chinese industrial quadruped used in an overseas power system and as saving roughly 480 labor hours per year. | 中 | SV005, SV006 |
| CV020 | Deep Robotics claims more than 50% inspection-efficiency improvement and around 70% operating-cost reduction in a substation use case. | 低 | SV007 |
| CV021 | The official industrial-inspection page positions Deep Robotics for hazardous environments such as pipelines, tunnels, plants, and confined spaces. | 中 | SV008 |
| CV022 | The IDC number-one business-application market-share claim exists in a company-linked press release but is not independently verified in the retained source set. | 低 | SV009 |
| CV023 | Unitree had crossed RMB 1 billion of annual revenue and was described as profitable since 2020 in 2025-2026 reporting. | 中 | SV018, SV012 |
| CV024 | Unitree’s 2025 Series C financing was described at nearly RMB 700 million with a post-money valuation above RMB 12 billion. | 中 | SV018 |
| CV025 | The humanoid-atlas tracker lists Unitree with about $248 million of 2025 revenue and a targeted roughly $7 billion STAR Market IPO. | 中 | SV019 |
| CV026 | ET Net shows UBTECH with 2025 turnover of RMB 2,000,999 thousand and attributable loss of RMB 703,191 thousand. | 中 | SV026 |
| CV027 | The same atlas tracker values UBTECH at about $7 billion, giving Deep Robotics a public Chinese robotics benchmark that is much larger in revenue scale. | 中 | SV019, SV026 |
| CV028 | The retained ANYbotics evidence shows more than $130 million of total funding by December 2024, framed around US expansion and onboard GPU integration rather than a clean current valuation. | 中 | SV020 |
| CV029 | The atlas tracker puts Boston Dynamics at about a $25 billion latest valuation estimate and notes possible Nasdaq plans around 2027, making it a strategic ceiling rather than a direct operating comp. | 低 | SV019 |
| CV030 | Yahoo Finance reported Skild AI at about a $4.5 billion Series B valuation, highlighting how software-first robotics narratives attract richer pricing than hardware-first inspection vendors. | 中 | SV021 |
| CV031 | Humanoids Daily’s valuation-chasm analysis argues that top robotics valuations are currently driven by war-chest financing and future optionality more than by present earnings. | 中 | SV011 |
| CV032 | China’s 2026 robotics IPO window is crowded, with Unitree, Aelos, and AgiBot all pushing toward public-market financing paths. | 中 | SV012, SV022 |
| CV033 | AW 2026 showed Unitree, AgiBot, and Leju competing visibly in Korea while Hyundai backed Boston Dynamics and a reported $6 billion robotics hub, underscoring export-market competition. | 中 | SV022 |
| CV034 | FY2025 NDAA commentary indicates US policymakers were still tightening scrutiny around Chinese-origin technology procurement, which is a structural discount factor for Deep Robotics in sensitive markets. | 中 | SV023 |
| CV035 | BIS extended the approved IC designer application deadline to December 31, 2026, showing that advanced-compute export-control rules remained active and fluid during the run year. | 中 | SV024 |
| CV036 | SemiAnalysis describes the quadruped market as led by Boston Dynamics and Unitree, with Deep Robotics and ANYbotics behind them. | 中 | SV010 |
| CV037 | The same SemiAnalysis report says some buyers reject Chinese-origin robots over security concerns, a risk that can compress Deep Robotics’ overseas multiple even if product performance is strong. | 中 | SV010 |
| CV038 | The financial_agent research note pegs Deep Robotics’ last reported valuation around $700 million and places the company well behind Unitree on consumer and humanoid scale. | 低 | SV017 |
| CV039 | That same note argues hardware gross margins in the high-twenties to low-thirties are thin relative to double-digit sales multiples, reinforcing caution on premium pricing. | 中 | SV017 |
| CV040 | Public valuation evidence is noisy because retained sources describe at least three different Deep Robotics markers: about RMB 8 billion, north of RMB 10 billion, and an aspirational ~$3.5 billion post-listing outcome. | 中 | SV002, SV016, SV027 |
| CV041 | Using 2025 revenue around $49.5-$49.6 million, the late-2025 RMB 8 billion mark implies roughly 22x trailing sales. | 中 | SV002, SV031 |
| CV042 | Using the same revenue base, a north-of-RMB-10-billion mark implies roughly 28x trailing sales, close to Unitree-style IPO territory. | 中 | SV027, SV031, SV019 |
| CV043 | A ~$3.5 billion post-listing valuation would imply roughly 70x trailing sales, which public evidence does not support for a company with only one sold humanoid unit in 2025. | 中 | SV016, SV031 |
| CV044 | A base-case supportable range around $800 million to $1.6 billion better fits the current combination of real industrial proof, still-early humanoid optionality, and valuation uncertainty. | 低 | SV002, SV027, SV031, SV019 |
| CV045 | A bear range around $450 million to $900 million is plausible if IPO timing slips, hardware multiples compress, or 2025 gross margin proves unusually favorable. | 低 | SV010, SV017, SV023, SV024 |
| CV046 | A bull range around $1.8 billion to $3.0 billion requires sustained hyper-growth, stable high margins, smooth IPO execution, and real humanoid commercialization. | 低 | SV011, SV016, SV031 |
| CV047 | The evidence supports a track recommendation rather than buy because company quality looks real but price support weakens sharply above the low-$1 billions. | 中 | SV002, SV016, SV027, SV031 |
| CV048 | Deep Robotics is more exit-ready than many private robotics peers because it has completed tutoring, has an accepted filing, and can point to 2025 profitability. | 中 | SV001, SV014, SV031 |
| CV049 | The planned 2.503 billion yuan raise is mainly aimed at embodied algorithms, models, robot-body R&D, industrialization, and base construction, which signals that the next phase remains capital intensive. | 中 | SV001, SV031, SV032 |
| CV050 | Deep Robotics’ investor and customer mix offers domestic strategic support, but state-linked backers and infrastructure ties can intensify geopolitical screening abroad. | 中 | SV004, SV029, SV023 |
| CV051 | The biggest remaining diligence gap is quality of revenue: recurring service share, warranty burden, realized discounting, and repeat-order cohorts are still not public. | 中 | |
| CV052 | A second crucial gap is cap-table visibility, because public summaries flag VAM and special-rights risk but do not disclose the full preference stack or liquidation waterfall. | 中 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Caixin Global | Deep Robotics Starts IPO Tutoring for Mainland Listing | The China Securities Regulatory Commission said Wednesday that the company filed for IPO tutoring on Dec. 23. |
| SO002 | Yicai Global | China’s Deep Robotics Raises USD70 Million to Fund Android R&D | Deep Robotics said the Chinese startup has raised CNY500 million (USD69.7 million). |
| SO003 | Yicai Global | China’s Deep Robotics Raises USD70.7 Million to Beef Up Embodied AI Research, Production | The Series C fundraiser was co-led by China Merchants Bank International and China Asset Management. |
| SO004 | Pandaily | DEEP Robotics Begins A-Share IPO Process, Becomes Third “Hangzhou Six Dragons” Company to Pursue Listing | |
| SO005 | Humanoids Daily | Deep Robotics Joins IPO Rush, Capping Quadruped Success with Humanoid Ambition | |
| SO006 | Humanoids Daily | Deep Robotics Secures $68M in Series C to Fuel Humanoid and Embodied AI Push | |
| SO007 | Newo | China’s Deep Robotics Joins IPO Race: Non-Human Workers Gear Up for Capital Markets | |
| SO008 | TechNode | Deep Robotics delivers robot SPock for Singapore Power Group | SP Group estimates that SPock’s efficiency in underground utility corridor inspections will save 480 hours of manual inspection time annually. |
| SO009 | Newsfile Corp / China Newswire | The World’s First, DEEP Robotics Unveils All-Weather Humanoid Robot to Reshape Industry Applications | As the world's first humanoid robot with an IP66 protection rating and capable of adapting to all-weather outdoor operations. |
| SO010 | KR-Asia | Deep Robotics unveils DR02, the world’s first IP66-rated humanoid robot | Key details regarding its autonomy, battery life, and cost remain undisclosed. |
| SO011 | 36Kr English / China Entrepreneur | DeepRobotics’ robotic dogs are special forces for high-risk scenarios | In November 2017, when the company was founded, it was autumn. |
| SO012 | 36Kr English / Caijing | Building robots is hard — DeepRobotics CTO Li Chao on commercialization, competition, and industry hype | It's not in the red. |
| SO013 | 36Kr English / Tian Tian IPO | Zhejiang University doctors join hands, and the robotic dog is heading for IPO | After eight rounds of financing, it has become the third company among the "Six Little Dragons of Hangzhou" to strive for an IPO. |
| SO014 | Yahoo Finance / GlobeNewswire / Zhejiang University | From DeepSeek to DEEP Robotics: How China’s Zhejiang University Cultivates the Tech Leaders of Tomorrow | Qiuguo Zhu is confirmed as an associate professor at ZJU. |
| SO015 | Interesting Engineering | Deep Robotics plan bots for the dark — and a giant one to help at home | Deep Robotics was first established in 2017 in China’s city of Hangzhou. |
| SO016 | Interesting Engineering | China unveils world’s first humanoid robot that resists dust, rain, heat | Deep Robotics has yet to reveal details about autonomy, operational duration, or the cost of their newest invention. |
| SO017 | The Robot Report | DEEP Robotics launches Lite3 quadruped robot with advanced front flipping mobility | Lite3 has been sold to more than a dozen countries. |
| SO018 | Humanoids Daily | The Great Valuation Chasm: A 2025 Guide to the Humanoid Robotics Capital Race | Deep Robotics' implied valuation places it in Tier 3: Billion-Dollar Contenders. |
| SO019 | Humanoids Daily | China’s Humanoid Sector Sees IPO Rush as Unitree, Aelos, and AgiBot Vie for Market Lead | The race is less about which company lists first and more about which can solve cost, scale, and real-world utility. |
| SO020 | DEEP Robotics | DEEP Robotics company overview | DEEP Robotics has built a full-stack “Perception–Decision–Execution” technology framework. |
| SO021 | DEEP Robotics / South China Morning Post reprint | China’s own Tesla Optimus? Beijing’s ambitions in humanoid robots in full display at expo | Deep Robotics introduced the DR01, the company’s first humanoid robot. |
| SO022 | DEEP Robotics | X20: The Ultimate Quadruped Bot series for Industrial Use | The X20 quadruped robot solution is for industrial patrol inspection under extreme working conditions. |
| SO023 | DEEP Robotics | LYNX M20 product page | The LYNX M20 series represents the world's first wheeled-legged robot built specifically for challenging terrains. |
| SO024 | DEEP Robotics | DR02 product page | The DR02 provides reliable and efficient intelligent solutions for various industries. |
| SO025 | DEEP Robotics | DEEP Robotics’ X30 robot dog pioneers smart inspection in Singapore power cable tunnel | SP Group plans to apply SPock on a wider scale. |
| SO026 | DEEP Robotics | DEEP Robotics debuts LYNX M20, sets new standard for industry-grade legged robotics | Today launched the industry-grade LYNX M20 series. |
| SM001 | Precedence Research | Quadruped Robot Market Size to Hit USD 17.53 Billion by 2035 | |
| SM002 | Emergen Research | Quadruped Robot Market Size, Share - Industry Analysis & Growth | |
| SM003 | International Federation of Robotics | US Robot Industry Returns to Double Digit Growth | China installed 295,000 industrial robots in 2024, accounting for 54% of global demand. |
| SM004 | IEEE Spectrum | Boston Dynamics' Robot Dog Price and Availability | Boston Dynamics is now selling Spot for US$74,500. |
| SM005 | IEEE Spectrum | Hyundai Buys Boston Dynamics for Nearly $1 Billion. Now What? | |
| SM006 | Yokogawa Electric Corporation | Yokogawa and ANYbotics to Integrate OpreX Robot Management Core Software with ANYmal Robotic Inspection Solutions | |
| SM007 | ANYbotics | About | |
| SM008 | ANYbotics | ANYmal X | |
| SM009 | Boston Dynamics | Spot | Boston Dynamics | |
| SM010 | Ghost Robotics | Ghost Robotics | Robots That Feel the World | |
| SM011 | DEEP Robotics | DEEP Robotics – Global Leader in Quadruped Robots | |
| SM012 | DEEP Robotics | DEEP Robotics introduces LYNX M20, a new benchmark for industrial-grade wheel-legged robots | |
| SM013 | DEEP Robotics | Over 50% Efficiency Gain: DEEP Robotics’ Quadruped Robot Becomes the “Super Employee” of Power Substations | The robot achieves over 50% efficiency gain and 98% inspection accuracy in power substations. |
| SM014 | DEEP Robotics | DEEP Robotics’ Quadruped Robot Delivers 96.5% Detection Accuracy in Autonomous Wind Farm Inspections Across the Gobi Desert | The X30 Pro delivered 96.5% detection accuracy in autonomous wind farm inspections. |
| SM015 | DEEP Robotics | DEEP Robotics’ X30 robot dog leads smart inspection in Singapore’s power cable tunnels | |
| SM016 | DEEP Robotics | DEEP Robotics Showcases Intelligent Quadruped Robots for Security and Inspection at GITEX GLOBAL 2025 | |
| SM017 | DEEP Robotics | DEEP Robotics’ Wheel-Legged Robot Tackles UHV Substation “Three Highs” Challenge | Manual operations face the three highs: high safety risks, highly complex environments, and low efficiency in precision detection. |
| SM018 | Unitree Robotics | Unitree Go2-W Driving All Terrain | Due to current technological and computing power limitations, certain functions must be implemented via manual operation or secondary development. |
| SM019 | Unitree Robotics | Robot Dog B2_Inspection Robot_B2 Robot Dog Advantage Usage | |
| SM020 | GasGoo AutoNews | MirrorMe Technology Officially Launches Quadruped Robot "Apollo" | Transforming stable lab prototypes into mass-produced, consistent industrial products remains a stubborn challenge across the robotics sector. |
| SM021 | Occupational Safety and Health Administration | Commonly Used Statistics | Occupational Safety and Health Administration | |
| SM022 | ANYbotics | ANYmal | |
| SM023 | ANYbotics | ANYbotics | Autonomous Legged Robots for Industrial Inspection | |
| SM024 | Boston Dynamics | Orbit Robot Fleet Management Software | Boston Dynamics | |
| SM025 | Unitree Robotics | 宇树科技—全球四足机器人行业开创者 | |
| SP001 | DEEP Robotics | DEEP Robotics X30 | |
| SP002 | DEEP Robotics | Industrial Inspection | |
| SP003 | DEEP Robotics | DEEP Robotics introduces LYNX M20, a new benchmark for industrial-grade wheel-legged robots | |
| SP004 | DEEP Robotics | Over 50% Efficiency Gain: DEEP Robotics’ Quadruped Robot Becomes the “Super Employee” of Power Substations | |
| SP005 | DEEP Robotics | DEEP Robotics’ Quadruped Robot Delivers 96.5% Detection Accuracy in Autonomous Wind Farm Inspections Across the Gobi Desert | |
| SP006 | DEEP Robotics | DEEP Robotics’ X30 robot dog leads smart inspection in Singapore’s power cable tunnels | |
| SP007 | DEEP Robotics | DEEP Robotics Showcases Intelligent Quadruped Robots for Security and Inspection at GITEX GLOBAL 2025 | |
| SP008 | DEEP Robotics | DEEP Robotics’ Wheel-Legged Robot Tackles UHV Substation “Three Highs” Challenge | |
| SP009 | Unitree Robotics | Unitree Go2-W Driving All Terrain | Due to current technological and computing power limitations, certain functions must be implemented via manual operation or secondary development. |
| SP010 | Unitree Robotics | Robot Dog B2_Inspection Robot_B2 Robot Dog Advantage Usage | |
| SP011 | Unitree Robotics Store | Unitree B2 | |
| SP012 | Boston Dynamics | Spot | |
| SP013 | IEEE Spectrum | Boston Dynamics’ Spot Robot Dog Now Available for $74,500 | For a long time, Boston Dynamics totally owned the quadruped space. Now, they’re one company among many… with much more affordable platforms coming out of Asia. |
| SP014 | IEEE Spectrum | Hyundai Buys Boston Dynamics—What Does It Mean? | |
| SP015 | ANYbotics | About | |
| SP016 | ANYbotics | ANYmal | |
| SP017 | ANYbotics | ANYmal X | Ex-proof up to Zone 1 IIB; IP67 water and dust ingress protection; designed for explosive atmospheres. |
| SP018 | Yokogawa | Yokogawa and ANYbotics to Integrate OpreX Robot Management Core Software with ANYmal Robotic Inspection Solutions | |
| SP019 | Ghost Robotics | Ghost Robotics | Robots That Feel the World | |
| SP020 | Ghost Robotics | Vision 60 | |
| SP021 | Ghost Robotics | Defense | |
| SP022 | Precedence Research | Quadruped Robot Market Size to Hit USD 17.53 Billion by 2035 | |
| SP023 | Emergen Research | Quadruped Robot Market Size, Share - Industry Analysis & Growth | |
| SP024 | International Federation of Robotics | US Robot Industry Returns to Double Digit Growth | |
| SP025 | GasGoo AutoNews | MirrorMe Technology Officially Launches Quadruped Robot "Apollo" | Transforming stable lab prototypes into mass-produced, consistent industrial products remains a stubborn challenge across the robotics sector. |
| SP026 | Flyability | Elios 3 - Digitizing the inaccessible | |
| SP027 | ABB | Robotics | ABB | |
| SI001 | DEEP Robotics | DEEP Robotics - Pioneering Innovation & Application | DEEP Robotics is a high-tech enterprise with a global focus, specializing in the research, development, manufacturing, sales, and service of quadruped robots, humanoid robots, and core robotic components. |
| SI002 | DEEP Robotics | DEEP Robotics – Global Leader in Quadruped Robots | |
| SI003 | DEEP Robotics | DEEP Robotics X30 | The X30 delivers up to 25% longer battery life than previous models, enabling extended inspection routes and prolonged field missions without interruption. |
| SI004 | DEEP Robotics | Over 50% Efficiency Gain: DEEP Robotics’ Quadruped Robot Becomes the “Super Employee” of Power Substations | Automated data collection and report generation have reduced manual workload and turnaround time by over 50%, while operational costs decreased by approximately 70%. |
| SI005 | DEEP Robotics | DEEP Robotics’ X30 robot dog leads smart inspection in Singapore’s power cable tunnels | Early results suggest the robot will help SP Group save more than 480 hours of manual inspection time each year. |
| SI006 | Eastmoney Data Center | 杭州云深处科技股份有限公司新股上会详细概况 | 点击查看招股说明书(申报稿)pdf |
| SI007 | Sina Finance / 经理人网 | 云深处科创板IPO获受理,四足机器人收入排名全球第二,2025年已实现扭亏为盈 | 2023年至2025年,云深处营业收入分别为5011万元、1.03亿元和3.37亿元,2025年首次实现扭亏为盈。 |
| SI008 | CnTechPost | Deep Robotics files for Shanghai listing following Unitree | In 2025, Deep Robotics' total revenue surged 227% to 337 million yuan and comprehensive gross margin improved to 52.83%. |
| SI009 | Humanoids Daily | Deep Robotics Files for $367M STAR Market IPO Following First Profitable Year | Deep Robotics also demonstrated growing international traction, with overseas revenue reaching CNY 61 million, representing just over 18 percent of its total income. |
| SI010 | Yicai Global | China’s Deep Robotics Raises USD70 Million to Fund Android R&D | The proceeds of Deep Robotics’ latest financing round will be used to strengthen its leading position in the field of quadruped robots, accelerate the development of new products, including humanoid robots, and recruit top talent. |
| SI011 | Yicai Global | China’s Deep Robotics Raises USD70.7 Million to Beef Up Embodied AI Research, Production | The latest fundraiser provides substantial capital for R&D, production expansion, and market development. |
| SI012 | 36Kr English | Exclusive Interview with ZHU Qiuguo, CEO of Cloudminds: Despite Being One of "Six Rising Stars in Hangzhou", We're Still a Small Company | The cost of its robotic dogs with the same performance is only one-third to one-half of that of Boston Dynamics. |
| SI013 | 36Kr English | Dialogue with LI Chao of DeepCloudAI: The Commercialization Path of Robots | If a customer buys the product and then makes a repeat purchase, such as 100 units, 300 units, or 500 units, it means the product is good-to-use. |
| SI014 | 36Kr English | Six Rising Stars in Hangzhou Set to Go Public Again | The latest round of financing took place recently. DeepClouds completed a Pre-IPO round of financing of hundreds of millions of yuan, led by the National Artificial Intelligence Industry Fund, with follow-on investment from JD.com. |
| SI015 | TechNode | Deep Robotics Delivers X30 Quadruped Robot 'SPock' for Singapore Power Inspections | The project required three to four months of safety assessments and could save 480 hours of manual inspection time annually. |
| SI016 | DEEP Robotics | DEEP Robotics introduces LYNX M20, a new benchmark for industrial-grade wheel-legged robots | Global pre-orders for the LYNX M20 are now open, marking a significant milestone in advancing industrial robotics worldwide. |
| SI017 | DEEP Robotics | Deep Robotics DR02 | |
| SI018 | Interesting Engineering | China unveils world’s first humanoid robot that resists dust, rain, heat | Deep Robotics products are 20-30% cheaper on average without much difference in technology from Boston Dynamics and Rainbow Robotics. |
| SI019 | KR-Asia | Deep Robotics unveils DR02, the world’s first IP66-rated humanoid robot | Key details regarding DR02's autonomy, battery life, and cost remain undisclosed. |
| SI020 | Newo | China’s Deep Robotics Joins IPO Race: Non-Human Workers Gear Up for Capital Markets | On December 23, 2025, Hangzhou-based Deep Robotics officially filed for IPO tutoring with China’s securities regulator. |
| SI021 | Humanoids Daily | The Great Valuation Chasm: A 2025 Guide to the Humanoid Robotics Capital Race | Valuations are not based on current earnings but on investor confidence in a company's ability to fund the multi-billion-dollar journey to mass production. |
| SI022 | Humanoids Daily | China's Humanoid Sector Sees IPO Rush as Unitree, Aelos, and AgiBot Vie for Market Lead | All three companies face the same fundamental challenge: proving they can scale mass production at a manageable cost and identify application scenarios with a clear return on investment. |
| SI023 | SemiAnalysis | Quadruped State of The Market - Unitree, Boston Dynamics, ANYbotics, DEEP Robotics, and The Rising Application Ecosystem | |
| SI024 | Digital Journal / Access Newswire | DEEP Robotics Leads the Industrial Robot Dog Race, Demonstrating Strong Capabilities from Power Grid Inspection to Plateau Scientific Exploration | According to an IDC report, DEEP Robotics has maintained the first-place market share in Business Application. |
| SI025 | DEEP Robotics | Industrial Inspection - DEEP Robotics | The X30 and Lynx M20 can be customized with sensor payloads, software modifications, and integrations with SCADA and facility management systems. |
| SI026 | DEEP Robotics | DEEP Robotics’ Wheel-Legged Robot Tackles UHV Substation “Three Highs” Challenge | In sprawling substations covering 500+ acres, multiple LYNX M20 units can work in sync. |
| SI027 | DEEP Robotics | DEEP Robotics’ Quadruped Robot Delivers 96.5% Detection Accuracy in Autonomous Wind Farm Inspections Across the Gobi Desert | Deep Robotics achieved 96.5% recognition accuracy across over 200 inspection missions at an unmanned wind power substation. |
| SI028 | BigGo Finance | Deep Robotics Files for STAR Market IPO: Swings to Profit Last Year, Quadrupedal Robot Industry Application Revenue Tops Global Rankings | If the IPO process hits a roadblock, valuation adjustment mechanism agreements enjoyed by investors will automatically be reinstated. |
| SI029 | 100EC | 云深处科技科创板IPO申请获上交所受理 拟募资 25.03 亿元布局具身智能 | The filing says Deep Robotics plans to issue no fewer than 82.98 million shares and raise RMB 2.503 billion for four embodied-intelligence projects. |
| SE001 | DEEP Robotics | DEEP Robotics company page | |
| SE002 | DEEP Robotics | X20: The Ultimate Quadruped Bot series for Industrial Use | |
| SE003 | DEEP Robotics | The X30 Series Power Robot Dogs with Advanced Functions | |
| SE004 | DEEP Robotics | LYNX M20 product page | |
| SE005 | DEEP Robotics | LYNX S10 product page | |
| SE006 | DEEP Robotics | DR02 product page | |
| SE007 | DEEP Robotics | Service support, warranty, privacy policy, and disclaimer terms | |
| SE008 | DEEP Robotics | DEEP Robotics debuts LYNX M20, sets new standard for industry-grade legged robotics | |
| SE009 | DEEP Robotics | DEEP Robotics' Wheel-legged Robot Helps Solve the Three Highs Challenge in UHV Substations | |
| SE010 | DEEP Robotics | LYNX M20 Empowers UHV Substation Intelligent Inspection | |
| SE011 | DEEP Robotics | LYNX M20 all-terrain robot brings breakthrough solutions for emergency rescue | |
| SE012 | DEEP Robotics | LYNX M20 in emergency rescue | |
| SE013 | DEEP Robotics | SPock — DEEP Robotics' X30 in Singapore Power Group's underground power transmission network | |
| SE014 | DEEP Robotics | China's own Tesla Optimus? Beijing's ambitions in humanoid robots in full display at expo | |
| SE015 | DEEP Robotics | DEEP Robotics completes over 500 million RMB Series C financing to accelerate breakthroughs in core embodied AI technologies and industry deployment | |
| SE016 | DEEP Robotics | Can students and researchers program DEEP Robotics quadruped robots? | |
| SE017 | GitHub | DEEPRoboticsLab organization page | |
| SE018 | GitHub | DeepRoboticsLab rl_training repository | |
| SE019 | GitHub | DeepRoboticsLab sdk_deploy repository | |
| SE020 | GitHub | DeepRoboticsLab deep_robotics_model repository | |
| SE021 | GitHub | DeepRoboticsLab Deep_Motor_SDK repository | |
| SE022 | DEEP Robotics | What models are available and how do they differ? | |
| SE023 | DEEP Robotics | DEEP Robotics Lite 3 | |
| SE024 | arXiv | Walking with Terrain Reconstruction: Learning to Traverse Risky Sparse Footholds | |
| SE025 | South China Morning Post | China's ambitions in humanoid robots in full display at expo | |
| SE026 | Newsweek | China unveils quadruped robot that can cross rough terrain at high speeds | |
| SE027 | Sina Finance | Beyond Unitree, Hangzhou also hides a leading robot-dog company | |
| SE028 | TechNode | Deep Robotics delivers robot SPock for Singapore Power Group | |
| SE029 | The Robot Report | DEEP Robotics launches Lite3 quadruped robot with advanced front flipping mobility | |
| SE030 | Interesting Engineering | DEEP Robotics' robot dog conquers snow, streams, mountains with ease | |
| SU001 | DEEP Robotics | DEEP Robotics Lands Applications in Singapore and South Korea! Will Quadruped Robots Experience an Application Boom? | |
| SU002 | DEEP Robotics | SPock — DEEP Robotics' X30 in Singapore Power Group's Underground Power Transmission Network | |
| SU003 | DEEP Robotics | DEEP Robotics' Wheel-legged Robot Helps Solve the Three Highs Challenge in UHV Substations | |
| SU004 | DEEP Robotics | LYNX M20 Empowers UHV Substation Intelligent Inspection — New Breakthrough in Complex Terrain | |
| SU005 | DEEP Robotics | Technology Empowers Rescue! The Wheel-Legged Robot of DEEP Robotics Makes Emergency Rescue More Efficient and Safer | |
| SU006 | DEEP Robotics | UK Embassy Delegation Visits DEEP Robotics – Advancing UK-China Tech Collaboration | |
| SU007 | DEEP Robotics | DEEP Robotics Completes Over 500 Million RMB Series C Financing to Accelerate Breakthroughs in Core Embodied AI Technologies and Industry Deployment | |
| SU008 | DEEP Robotics | DEEP Robotics Showcases Open Robotics Platforms at ICRA 2026, Expanding Opportunities for Research and Education | |
| SU009 | DEEP Robotics | Service Support — Terms and Policies | |
| SU010 | TechNode | Deep Robotics delivers X30 quadruped robot SPock for Singapore power inspections | |
| SU011 | Robotics 24/7 | DEEP Robotics deploys X30 robot dog to inspect Singapore power cable tunnel | |
| SU012 | Automation.com | DEEP Robotics' X30 Robot Dog Pioneers Smart Inspection in Singapore Power Cable Tunnels | |
| SU013 | Robotics & Automation News | Deep Robotics' robot dog pioneers smart inspection in Singapore power cable tunnel | |
| SU014 | ACCESS Newswire | Deep Robotics' Robot Dog Security Patrol Solution Deployed in Singapore, Setting a New Benchmark for Smart Security | |
| SU015 | Robotics & Automation News | Deep Robotics launches wheel-legged robot to transform UHV substation inspections | |
| SU016 | Jiemian News | China's power grid operators are preparing a large-scale rollout of robots | |
| SU017 | Humanoids Daily | State Grid's $995 Million Bet: 8,500 Robots to Spearhead Deployment Year One in Energy | |
| SU018 | The Robot Report | DEEP Robotics launches Lite3 quadruped robot with advanced front flipping mobility | |
| SU019 | Wevolver | Quadruped robot with advanced front flipping mobility | |
| SU020 | JobToRob | Lite3: DEEP Robotics' Next-Gen Quadruped Robot | |
| SU021 | Generation Robots | Deep Robotics Lite 3 vs Unitree Go2: two quadrupeds, two visions | |
| SU022 | Select Committee on China, U.S. House of Representatives | Moolenaar, Obernolte, McClellan Introduce Legislation to Ban Dangerous Chinese Robots | |
| SU023 | Foundation for Defense of Democracies | As Chinese Robotics Industry Surges, Senate Considers Limited Federal Procurement Ban | |
| SU024 | The AI Insider | Report: U.S. lawmakers to introduce American Security Robotics Act to ban federal agencies from buying Chinese humanoid robots | |
| SU025 | DEEP Robotics | LYNX M20 in Emergency Rescue: Technological Innovation Empowers Urban Emergency Response | |
| SR001 | DEEP Robotics | DEEP Robotics X30 | |
| SR002 | DEEP Robotics | Industrial Inspection | |
| SR003 | DEEP Robotics | Over 50% Efficiency Gain: DEEP Robotics’ Quadruped Robot Becomes the “Super Employee” of Power Substations | The robot achieves over 50% efficiency gain and 98% inspection accuracy in power substations. |
| SR004 | DEEP Robotics | DEEP Robotics’ X30 robot dog leads smart inspection in Singapore’s power cable tunnels | Early results suggest the robot will help SP Group save more than 480 hours of manual inspection time each year. |
| SR005 | DEEP Robotics | DEEP Robotics Showcases Intelligent Quadruped Robots for Security and Inspection at GITEX GLOBAL 2025 | The system covers the full operational chain — perception, communication, decision-making, and execution — tailored for applications in energy infrastructure inspection, smart city security, and emergency firefighting. |
| SR006 | DEEP Robotics | DEEP Robotics’ Wheel-Legged Robot Tackles UHV Substation “Three Highs” Challenge | The M20 supports a 15kg payload ... and its stabilization system ensures detection accuracy above 95%. |
| SR007 | DEEP Robotics | DEEP Robotics’ Quadruped Robot Delivers 96.5% Detection Accuracy in Autonomous Wind Farm Inspections Across the Gobi Desert | |
| SR008 | DEEP Robotics | DEEP Robotics introduces LYNX M20, a new benchmark for industrial-grade wheel-legged robots | |
| SR009 | DEEP Robotics | Deep Robotics DR02 | As the world's first humanoid robot with an IP66 protection rating capable of all-weather outdoor operations, the DR02 provides reliable and efficient intelligent solutions for various industries. |
| SR010 | DEEP Robotics | About DEEP Robotics | |
| SR011 | DEEP Robotics | DEEP Robotics Lands Applications in Singapore and South Korea! Will Quadruped Robots Experience an Application Boom? | Public information shows that DEEP Robotics' industrial-grade quadruped robot has served more than 100 clients in 26 provinces and cities in China. |
| SR012 | TechNode | Deep Robotics Delivers X30 Quadruped Robot 'SPock' for Singapore Power Inspections | The project required three to four months of safety assessments and could save 480 hours of manual inspection time annually. |
| SR013 | Digital Journal / AccessWire | Deep Robotics leads industrial robot market as applications multiply across power, emergency and public security | |
| SR014 | SemiAnalysis | Quadruped State of The Market - Unitree, Boston Dynamics, ANYbotics, DEEP Robotics, and The Rising Application Ecosystem | The quadruped hardware market currently is dominated by two players: Boston Dynamics and Unitree, while DEEP Robotics and ANYBotics lag behind. |
| SR015 | bobjiang / awesome-robots | Deep research for Deep Robotics | 42 manufacturers now producing quadruped robots ... Unitree's aggressive pricing strategy ... creating downward price pressure. |
| SR016 | Inside Government Contracts | FY2025 NDAA: Congressional Efforts to Bolster U.S. Resilience Against Chinese Tech and Influence | Beginning in June 2026, [DoD] prohibits ... procuring items from entities on the [1260H] list ... and beginning in June 2027, prohibits ... products or components made by 1260H listed entities. |
| SR017 | U.S. Congress / Government Publishing Office | Public Law 118-159 — National Defense Authorization Act for Fiscal Year 2025 | |
| SR018 | Government Publishing Office | Federal Register, Volume 91 Issue 68 — Extension of Authorized Integrated Circuit (IC) Designer Status and Application Deadline To Become an Approved IC Designer | The new date is December 31, 2026. |
| SR019 | Government Publishing Office / Department of Defense | Federal Register, Volume 91 Issue 111 — Notice of Availability of Designation of Chinese Military Companies | Hangzhou Yushu Technology Co., Ltd. (Unitree) ... is a military-civil fusion contributor. |
| SR020 | CnTechPost | Deep Robotics files for Shanghai listing following Unitree | In 2025, Deep Robotics' total revenue surged 227% to 337 million yuan and comprehensive gross margin improved to 52.83%. |
| SR021 | Sina Finance / 经理人网 | 云深处科创板IPO获受理,四足机器人收入排名全球第二,2025年已实现扭亏为盈 | 2023年至2025年,云深处营业收入分别为5011万元、1.03亿元和3.37亿元,2025年首次实现扭亏为盈。 |
| SR022 | Humanoids Daily | China's Humanoid Sector Sees IPO Rush as Unitree, Aelos, and AgiBot Vie for Market Lead | All three companies face the same fundamental challenge: proving they can scale mass production at a manageable cost and identify application scenarios with a clear return on investment. |
| SR023 | The Robot Report | AW 2026 features Korea humanoid debuts as industry seeks digital transformation | Hyundai ... reportedly plans to invest $6 billion in a robotics, data, and energy hub near Seoul. |
| SR024 | ANYbotics | ANYmal X | ATEX & IECEx certified up to Zone 1 IIB. |
| SR025 | Yokogawa Electric Corporation | Yokogawa and ANYbotics to Integrate OpreX Robot Management Core Software with ANYmal Robotic Inspection Solutions | ANYbotics ... [is] trusted by hundreds of industry leaders ... supported by over $150m in funding ... and a 200-expert team. |
| SR026 | Unitree Robotics | Unitree B2 | continuous walking with 20kg load can last for greater than 4 hours ... unloaded continuous walking endurance greater than 5 hours |
| SR027 | Ghost Robotics | Ghost Robotics | Robots That Feel the World | From the battlefield to the oil field, Ghost Robotics empowers defense and commercial teams with cutting-edge solutions for the most critical missions. |
| SR028 | 36Kr English | Unitree Technology: Targeting a $7 Billion IPO Valuation, Sparkling the Industry? | The financing amount was close to 700 million yuan, and the post-investment valuation exceeded 12 billion yuan. |
| SR029 | dadachundan / financial_agent | Unitree Research Document | |
| SR030 | Humanoids Daily | Deep Robotics Files for $367M STAR Market IPO Following First Profitable Year | The company turned a profit for the first time last year ... gross margin widened significantly ... to 52.8 percent in 2025. |
| SR031 | Caixin Global | Deep Robotics Starts IPO Tutoring for Mainland Listing | Deep Robotics, valued at about 8 billion yuan ($1.1 billion), began IPO tutoring in December 2025, aiming for review completion by June 2026. |
| SV001 | CnTechPost | Deep Robotics files for Shanghai listing following Unitree | In 2025, Deep Robotics' total revenue surged 227% to 337 million yuan. |
| SV002 | Caixin Global | Deep Robotics Starts IPO Tutoring for Mainland Listing | Deep Robotics, valued at about 8 billion yuan ($1.1 billion), began IPO tutoring in December 2025. |
| SV003 | Yicai Global | China’s Deep Robotics Raises USD70 Million to Fund Android R&D | Deep Robotics said the Chinese startup has raised CNY500 million (USD69.7 million). |
| SV004 | Yicai Global | China’s Deep Robotics Raises USD70.7 Million to Beef Up Embodied AI Research, Production | Deep Robotics has secured more than CNY500 million (USD70.7 million) to support the Chinese startup’s plans to scale up research, production, and widespread industrial deployment. |
| SV005 | TechNode | Deep Robotics delivers robot SPock for Singapore Power Group | SP Group estimates the robot can save more than 480 hours of inspection work annually. |
| SV006 | DEEP Robotics | DEEP Robotics’ X30 robot dog leads smart inspection in Singapore’s power cable tunnels | The solution marks the first Chinese industrial quadruped deployed in an overseas power system. |
| SV007 | DEEP Robotics | Over 50% Efficiency Gain: DEEP Robotics’ Quadruped Robot Becomes the “Super Employee” of Power Substations | Over 50% Efficiency Gain: DEEP Robotics’ Quadruped Robot Becomes the “Super Employee” of Power Substations. |
| SV008 | DEEP Robotics | Industrial Inspection | Deploy an autonomous quadruped robot engineered for safe, precise inspections across all industrial sites—from manufacturing floors and chemical plants to pipelines, tunnels, and confined spaces. |
| SV009 | Digital Journal / AccessWire | DEEP Robotics Leads the Industrial Robot Dog Race, Demonstrating Strong Capabilities from Power Grid Inspection to Plateau Scientific Exploration | DEEP Robotics was recognized by global market intelligence firm IDC as the world’s top company by market share in the business application segment of quadruped robots. |
| SV010 | SemiAnalysis | Quadruped State of The Market - Unitree, Boston Dynamics, ANYbotics, DEEP Robotics, and The Rising Application Ecosystem | The quadruped hardware market currently is dominated by two players: Boston Dynamics and Unitree, while DEEP Robotics and ANYBotics lag behind. |
| SV011 | Humanoids Daily | The Great Valuation Chasm: A 2025 Guide to the Humanoid Robotics Capital Race | As of late 2025, a clear “great valuation chasm” has emerged. |
| SV012 | Humanoids Daily | China’s Humanoid Sector Sees IPO Rush as Unitree, Aelos, and AgiBot Vie for Market Lead | Unitree appears to be at the forefront of commercialization. Founder Wang Xingxing claimed in June that the company’s annual revenue had exceeded 1 billion RMB. |
| SV013 | Pandaily | DEEP Robotics Begins A-Share IPO Process, Becomes Third “Hangzhou Six Dragons” Company to Pursue Listing | |
| SV014 | KrASIA | Deep Robotics completes IPO tutoring, moving closer to Star Market listing | Deep Robotics has completed IPO tutoring for a prospective listing on Shanghai’s Star Market and is expected to submit its prospectus at a later date. |
| SV015 | 36Kr / IPO Zaozhidao | Yun Shen Chu's IPO Application on STAR Market Accepted: Revenue in Quadruped Robot Field Surpasses Boston Dynamics, Turns Profitable | From 2023 to 2025, Deep Robotics' revenues were 50 million yuan, 103 million yuan, and 337 million yuan respectively. |
| SV016 | Pandaily | DEEP Robotics Files for STAR Market IPO, Seeks ~$3.5B Valuation | The company plans to raise approximately 2.503 billion yuan in the IPO, with its post-listing valuation expected to reach around 3.5 billion US dollars. |
| SV017 | dadachundan / financial_agent | Unitree Research Document | Deep Robotics (云深处): last reported valuation ~ USD 700 million. |
| SV018 | 36Kr | Wang Xingxing: Unitree Technology's Annual Revenue Exceeds 1 Billion Yuan | The financing amount was close to 700 million yuan, and the post-investment valuation exceeded 12 billion yuan. |
| SV019 | humanoid-atlas raw data | funding.ts | Unitree ... targeting ~$610M raise ... revenue2025M: 248 ... valuationM: 1700. |
| SV020 | Robotnik raw audit | take_cumulative_funding_audit.md | ANYbotics ... >$130M total raised ... Qualcomm's lead suggests edge-compute integration thesis. |
| SV021 | Yahoo Finance | Nvidia And Samsung Back $4.5B Robotics Startup Skild AI With $35M As SoftBank And Jeff Bezos Drive Push Into Consumer Robots | The Series B funding round ... values Skild AI at approximately $4.5 billion. |
| SV022 | The Robot Report | AW 2026 features Korea humanoid debuts as industry seeks digital transformation | Agibot X2, Unitree G1, and Leju Kuavo 4th Generation Pro humanoid robots make their debuts in South Korea at AW 2026. |
| SV023 | Inside Government Contracts | FY2025 NDAA: Congressional Efforts to Bolster U.S. Resilience Against Chinese Tech and Influence | FY2025 NDAA: Congressional Efforts to Bolster U.S. Resilience Against Chinese Tech and Influence. |
| SV024 | Government Publishing Office / BIS | Federal Register, Volume 91 Issue 68 (Thursday, April 9, 2026) | The new date is December 31, 2026. |
| SV025 | UBTECH Robotics | Financial Reports | UBTECH Robotics | Financial Reports | UBTECH Robotics |
| SV026 | ET Net | 9880.HK UBTECH ROBOTICS - Company Information - Income Statement Overview | 2025/12 - Final ... Turnover 2,000,999 ... Profit / (Loss) Attributable to Shareholders (703,191). |
| SV027 | Humanoids Daily | Deep Robotics Files for $367M STAR Market IPO Following First Profitable Year | The company currently boasts a valuation north of CNY 10 billion ($1.5 billion). |
| SV028 | Humanoids Daily | Deep Robotics Joins IPO Rush, Capping Quadruped Success with Humanoid Ambition | Like Unitree, Deep Robotics is leveraging a successful and revenue-generating quadruped business to fund its expansion into the more capital-intensive humanoid market. |
| SV029 | Humanoids Daily | Deep Robotics Secures $68M in Series C to Fuel Humanoid and ‘Embodied AI’ Push | Deep Robotics has secured over 500 million yuan (approximately $68 million) in Series C financing. |
| SV030 | Newo | China’s Deep Robotics Joins IPO Race: Non-Human Workers Gear Up for Capital Markets | |
| SV031 | BigGo Finance | Deep Robotics Files for STAR Market IPO: Swings to Profit Last Year, Quadrupedal Robot Industry Application Revenue Tops Global Rankings — BigGo Finance | If the company’s IPO process hits a roadblock, the terms of the valuation adjustment mechanism (VAM) agreements enjoyed by investors will automatically be reinstated. |
| SV032 | 100EC | 云深处科技科创板IPO申请获上交所受理 拟募资 25.03 亿元布局具身智能 | 公司计划募集资金 25.03 亿元。 |
| SV033 | Eastmoney Data Center | 杭州云深处科技股份有限公司新股上会详细概况 |