初创公司尽调
尽调报告 Robotics / Hardware (edge AI chips and developer platform) Series B 2026-07-04

D-Robotics

D-Robotics 已经拿出可信的生态牵引力:开发者超过 100,000、截至 B2 轮累计融资 $370M;但其独角兽估值押注的是出货动能,而不是已披露收入。在 NVIDIA Jetson 和 Qualcomm Robotics 已经立起强势平台标尺的市场里,这一点仍需验证。

D-Robotics 已拼出可信的开发者生态深度、继承 Horizon 的工程基因,并拿到 $370M 资本,押注机器人边缘 AI 芯片平台;但独角兽估值要求投资人先为尚未被公开证据证实的收入故事定价。

封面要素

从 Horizon Robotics 分拆 01
January 2024 [CO024]
A 轮融资 02
100 USD M [CI016]
B 轮合计 03
270 USD M [CO025]
开发者社区 05
100000 developers [CO032]
出货量同比增长 06
180 pct [CO031]

公司概况

D-Robotics(地瓜机器人)是一家总部位于北京的边缘 AI 芯片与机器人基础设施公司,2024 年 1 月从 Horizon Robotics 的 AIoT 部门分拆出来。公司打造自研 Brain Processing Unit(BPU)、RDK 硬件开发套件(X5 为 10 TOPS;S100 为 80–128 TOPS)、TogetheROS.Bot 机器人操作系统,以及 NodeHub 开放应用市场。它的市场进入策略押注开发者生态:注册开发者超过 100,000、GitHub 仓库 229 个、开源算法超过 200 个,并以 OEM 伙伴计划切入人形机器人、陪伴机器人和自动化机械。公司在 2025 年 5 月完成 $100M A 轮,2026 年初合计完成 $270M B 轮(B1 + B2),投资方包括 Didi Global、Meituan Long-Z Fund、Prosperity7 Ventures、GL Ventures、Vertex Growth 和 5Y Capital。D-Robotics 被列入 2025 中国独角兽观察报告新增独角兽。最近披露期内,出货量同比约增长 180%,客户数约增长 200%。

官网
d-robotics.cc
成立时间
2024-01-01
创始人
Horizon Robotics (spin-out)
创立地点
Beijing, China
总部
Beijing, China
产品
核心产品分三层:(1)边缘 AI 芯片——BPU Nash 架构,在 S100 平台提供 80–128 TOPS,在 X5 提供 10 TOPS,覆盖 5–128 TOPS 性能区间;(2)开发者套件——RDK X5 和 RDK S100 开发板,支持 Ubuntu 22.04、兼容 ROS2,并提供多传感器 IO;(3)软件栈——TogetheROS.Bot(兼容 ROS2 的 RTOS,支持零拷贝通信和硬件加速 CV)、NodeHub(开放应用中心)以及 200+ 算法模型库。
客户
面向人形、陪伴、服务和物流机器人的机器人 OEM;嵌入式 AI 系统集成商;基于 BPU 开发者平台搭建项目的高校与创业开发者。
商业模式
硬件销售(芯片 + 开发套件)、潜在的软件授权和云开发者服务,以及 OEM 在量产机器人中采用 BPU 芯片后形成的生态收入模式。
阶段
Series B (B1 + B2 completed April 2026)
融资情况
A 轮:$100M(2025 年 5 月);B1 轮:$120M(2026 年 3 月),来自 Didi Global、Meituan Long-Z Fund、BAIC Capital、Xilian Capital 和 Joyoun;B2 轮:$150M(2026 年 4 月),来自 Didi Global、Prosperity7 Ventures、GL Ventures、Vertex Growth 和 5Y Capital。截至 runDate,累计融资约 $370M。
[CO024, CO025, CO026, CO027, CI016, CI017, CI018, CI019]

执行摘要

主要优势

  • 自研 BPU Nash 架构(S100 为 80–128 TOPS)让机器人可在端侧推理,不必依赖 NVIDIA 或 Qualcomm,同时带来差异化和利润率潜力。
  • Horizon Robotics 背景给拆分团队带来 IP 深度、成熟 BPU 芯片经验和大规模工程人才池。
  • 100,000+ 注册开发者、229 个 GitHub 仓库和 200+ 开源算法组成开发者生态,为 OEM 设计选型沉淀真实切换成本护城河。
  • 出货量同比 +180%、客户数 +200%,在收入数字披露前已显示真实商业牵引力。
  • Didi、Meituan、Prosperity7、GL Ventures 等强机构背书验证平台逻辑,也降低近期融资风险。

主要风险

  • NVIDIA Jetson(AGX Orin 最高 275 TOPS)、Qualcomm 机器人 SoC 和 Ambarella CV5 平台都是资金更充足的全球对手,且已拥有 OEM 关系和全球软件生态。
  • 收入、毛利率和烧钱速度完全未披露,投资人无法用具体财务指标承销 B 轮估值。
  • 中美出口管制收紧(BIS 2026 年 5 月指引)可能间接限制 D-Robotics 的芯片供应链、晶圆厂关系和国际客户触达。
  • 公司公开产品线仍主要面向开发套件和创客市场;能否大规模转化为量产 OEM 设计定点尚未确认。
  • Cambricon $125B 市值说明 AI 芯片倍数可以高度投机;若 D-Robotics 达成收入里程碑前市场情绪转向,估值风险会放大。

未决问题

  • 审计或管理层口径财务数据——收入、ARR、毛利率、烧钱速度和 B2 轮后的现金跑道。
  • 经确认的投后估值数字;独角兽标签来自二手 Baidu Baike 引用,而非一手投资人文件。
  • 具名 OEM 量产客户名单,用以确认 BPU 芯片已进入规模出货的商用机器人产品设计。
  • 晶圆厂、封装伙伴和供应链细节,用以评估出口管制暴露。
  • D-Robotics 拆分实体的管理团队姓名、背景和股权归属,需与 Horizon Robotics 母公司领导层分开披露。

目录

Chapter 01

01公司概览

1.1 身份定位、产品边界与机器人技术栈

D-Robotics 给自己的定位不是单板供应商,而是机器人基础设施公司。当前英文官网和产品页把开发者硬件、操作系统或中间件层、应用示例和部署工具都收进 RDK 体系。虽然命名还在演进,公开硬件叙事已经清楚:RDK X5 主打 10 TOPS 机器人与边缘智能开发板,RDK S100 则定位更高端的具身智能平台,围绕 BPU Nash 以及 CPU+BPU+MCU 的感知、决策、控制分工搭建。软件层也比典型设备厂商的支持页厚得多。TogetheROS.Bot、NodeHub、RDK Studio 和公开文档仓库显示,D-Robotics 想把开发者工作流从模型适配、传感器集成一路管到基于 ROS2 的机器人应用开发。这种组合与公司公开定位一致:为更小形态机器人提供低功耗边缘推理方案,替代更笨重的通用 GPU 栈。[CO001, CO002, CO003, CO004, CO005, CO006]

D-Robotics 快照 KPI 表
指标数值 / 状态日期置信度缺口 / 注意事项
定位机器人开发基础设施:硬件 + 软件 + 应用2026官方表述很宽,未绑定正式细分市场分类
分拆状态从 Horizon Robotics AIoT 事业部分拆2024-01独立新闻报道披露,非英文官网披露
Series B 融资B1 $120M + B2 $150M = 约 $270M2026-03 to 2026-04已审阅来源未直接披露估值
员工规模信号LinkedIn 显示 51-200 名员工2026-07-04LinkedIn 区间只具方向性,不是审计口径人数
开发者触点229 个公开 GitHub 仓库2026-07-04仓库数量来自 GitHub 观察,不是公司审计
社区下限第三方报道称 100,000+ 名用户 / 开发者2026-04已审阅来源未支撑更高的 200,000+ 说法
RDK X5 算力10 TOPS2026官方产品说法
RDK S100 算力BPU Nash 上 80/128 TOPS2026官方产品说法,并有零售渠道交叉印证
公开收入披露未公开披露2026无已审阅来源拆分硬件与 SDK / 软件收入

混合了官方产品说法、第三方融资报道和生态信号;未获支撑的估值与收入数字保留为未解决问题,而不是推断填补。

[CO001, CO006, CO008, CO012, CO016, CO024]
FO002: 公司快照逻辑

D-Robotics 将芯片、中间件、开发者工具和生态项目串成一个机器人平台叙事。

[CO001, CO003, CO008, CO010, CO013, CO014]

1.2 领导谱系、开发者触点与渠道存在感

公开材料对公司源流的证据强于正式治理。D-Robotics 没有在英文网站披露完整管理层页面或董事会名单,但独立报道和 Nash BPU 品牌上的技术延续性,都强烈指向它与 Horizon Robotics 的关系。Horizon 自己的英文介绍页将 Dr. Yu Kai 标为创始人兼 CEO,Yicai 则把 D-Robotics 描述为 2024 年 1 月从 Horizon 的 AIoT 部门分拆出来的公司。开发者足迹更容易核验。GitHub 显示它有一个大型公开组织,仓库数以百计,并明确包含 RDK OS、模型库、摄像头和机器人栈项目。围绕 TogetheROS.Bot 和 NodeHub 的文档说明,D-Robotics 面向的是生态开发者和 OEM 伙伴,而不只是内部参考设计。LinkedIn 提供了中等规模团队信号:员工区间为 51-200,关注者数千;Discord、YouTube 和社区页等公开社区触点,也显示公司持续招募布道者、创客和创业团队。这些信号不能证明市场主导地位,但确实显示出一套有意推进的国际化开发者获客动作。[CO010, CO011, CO012, CO013, CO014, CO015]

领导层与创始人表
人物 / 实体角色或关系公开证据重要性关键人物依赖
Dr. Yu KaiHorizon Robotics 创始人兼 CEO;D-Robotics 的母公司脉络锚点Horizon 关于页面将其列为创始人兼 CEO;Yicai 将 D-Robotics 与 Horizon 分拆脉络相连锚定分拆背后的技术和资本血统
Horizon Robotics AIoT 事业部分拆前的母公司业务线Yicai 称 D-Robotics 于 2024 年 1 月从该事业部分拆解释 D-Robotics 为何继承芯片加工具链 DNA
D-Robotics 生态开发者外部构建者社区,而非高管团队社区、GitHub 和文档显示公司依赖生态采用开发者驱动是平台战略核心
公开治理名单已审阅英文公开界面未披露已审阅资料未发现公开董事会或投票权细节带来控制权和治理不确定性

公开资料更能证明脉络,而不是现任 D-Robotics 具名高管;董事会和治理仍未披露。

[CO021, CO022, CO024, CO037]
利益相关方或投资者地图
利益相关方角色控制权或经济重要性尽调问题
Horizon Robotics技术和脉络母体解释分拆起源、BPU 脉络和 Yu Kai 关联厘清 IP 转让、共享人员以及任何持续交叉持股
Didi Global已披露 B1 和 B2 投资方同时出现在两轮,释放战略信心信号确认持股比例以及任何商业合作权利
Prosperity7 Ventures已披露 B2 投资方国际成长型投资者参与厘清董事席位、信息权和跟投意愿
Vertex Growth已披露 B2 投资方Temasek 相关成长资本增加信号价值确认持股比例和治理权利包
Meituan Long-Z Fund已披露 B1 投资方为早期 Series B 财团加入中国战略互联网资本厘清参与是纯财务性质,还是绑定商业合作
BAIC Capital / Xilian Capital已披露 B1 投资方代表财团中的国内产业资本和金融资本还原确切出资规模和战略价值
Joyoung Family Office / 5Y Capital 投资方家族办公室和风险投资参与方表明战略投资者组合较宽,而不是单一领投叙事索取完整 cap table 和轮次机制,包括任何老股转让

投资者行只反映已审阅 2026 年报道中明确点名的各方;未披露投资者和持股比例仍是开放尽调事项。由于持股比例和未披露投资者仍为私有信息,覆盖并不完整。

[CO024, CO025, CO026, CO027]

1.3 资本形成、分拆经济性与已验证里程碑

材料中最清晰的第三方证据来自 2026 年融资序列。Yicai 和 CXO Digitalpulse 均报道,D-Robotics 先完成 $120M B1 轮,随后完成 $150M B2 轮,使 B 轮融资约达 $270M。Yicai 披露的投资方名单也比英文官网更细:B2 包括 Didi Global、Prosperity7 Ventures、GL Ventures、Vertex Growth 和 5Y Capital,B1 包括 Didi Global、Meituan Long-Z Fund、BAIC Capital、Xilian Capital 和 Joyoung Family Office。这让资本故事站得住,但仍不完整。用户简报中的约 $1.5B 估值和 200,000+ 开发者数字,并未被本次审阅来源直接支持,因此暂时不应当作已验证的头部事实。其他公开里程碑证据较健康:2024 年分拆、RDK S100 于 2026 年 1 月在日本零售渠道上架、RDK X5 于 2026 年 3 月被第三方硬件采用,以及 2026 年在 embedded world 和 ICML 的活动节奏,都显示公司正快速从分拆身份进入生态建设和渠道扩张。[CO019, CO024, CO025, CO026, CO027, CO029]

里程碑表
日期事件类型金额 / 状态参与方含义
2015-06Horizon Robotics 成立,Yu Kai 被列为创始人兼 CEO创立母公司成立Horizon Robotics建立后来 D-Robotics 分拆背后的母公司脉络
2024-01D-Robotics 从 Horizon AIoT 事业部分拆治理分拆已报道D-Robotics, Horizon Robotics形成独立、聚焦机器人的法律和运营叙事
2024Horizon 称 Nash BPU 在母公司侧发布产品技术里程碑Horizon Robotics帮助解释后来 D-Robotics 的 BPU Nash 品牌
2026-01-20RDK S100 由日本 Switch Science 上架规模$499 MSRP / ¥102,300 日本本地零售价Switch Science显示国际转售渠道和早期市场可得性
2026-03Series B1 已报道融资$120MDidi, Meituan Long-Z, BAIC Capital, Xilian, Joyoung Family Office 等投资方确认投资者愿意押注具身 AI 基础设施
2026-03-25CNX 记录 LooperRobotics 产品使用 RDK X5合作第三方产品证明LooperRobotics, D-Robotics证明外部生态采用
2026-04-08Series B2 已报道;Series B 总额达到约 $270M融资$150M / ~$270M Series B 总额Didi, Prosperity7, GL Ventures, Vertex Growth, 5Y 等投资方验证分拆后的持续融资动能
2026-04Yicai 报道 100+ 款机器人产品和 100k+ 开发者社区规模已报道增长指标D-Robotics支撑生态宽度,但尚未审计确认
2026-07ICML 交流活动和官方渠道可见的持续社区推进合作生态持续扩张D-Robotics, ZODA, 开发者显示 2026 年仍在继续全球外联

这条时间线只保留已审阅来源直接支撑的有日期里程碑;估值、收入和治理事件在获得支撑前继续排除。由于内部发布、招聘事件和估值里程碑并不完全公开,覆盖仍然不完整。

[CO019, CO021, CO022, CO023, CO024, CO025]
FO001: 公司里程碑时间线

已验证里程碑显示,D-Robotics 很快从 Horizon 血统转向面向国际市场的机器人平台。

[CO019, CO021, CO023, CO024, CO025, CO034]
FO003: 牵引与披露 KPI

公开 KPI 同时包含牵引信号和尚未解决的披露项,凸显尽调信心仍被哪些问题压住。

人员、社区和融资指标混合了官方口径、平台可观测数据和第三方报道值; 缺乏支撑的估值说法已刻意剔除。

[CO006, CO008, CO012, CO016, CO025, CO032]

1.4 风险信号与披露缺口

最大的尽调问题不是缺少动作,而是披露不均。公开证据确认了 2026 年融资大幅跃升和可信的产品栈,但收入构成、软件授权经济性、董事会控制权,以及总部或法律实体足迹的精确情况,仍留下重大盲区。独立报道还显示,更广义的机器人资本市场正在升温并变得更扭曲。Yicai 2026 年 4 月的私募股权分析称,资本正涌入机器人等硬科技赛道,同时部分投资人警告估值可能已经处在周期高点,热门轮次的入场还可能要求异常的下一轮承诺或保证金。该信号并不意味着 D-Robotics 自身有特定问题,但说明公司正在一个融资条款可能快过基本面的市场里扩张。实际含义是,投资人可以把产品和生态牵引力视为真实存在,但在管理层提供更干净的一手数据室前,应暂缓判断估值公允性、收入质量和治理水平。[CO016, CO024, CO025, CO033, CO037, CO038]

Chapter 02

02市场分析

2.1 市场边界、纳入支出与相邻品类

不能把 D-Robotics 当成向单一垂直行业销售成品机器人的公司来分析。它的公开触点显示的是一套机器人开发与边缘算力栈:开发者硬件、具身机器人算力平台、中间件和部署工具。这把公司放在几个支出池的交汇处。最宽的相邻池是 Cloud FinOps 和云成本优化,因为买方越来越在意算力效率、工作负载放置和业务价值问责。更窄但运营上更相似的池是 Kubernetes 成本管理和集群级自动化,节点供给、成本归因和工作负载效率已经是明确预算项。机器人专属层则叠在这些抽象之上:具身机器人、工业机器人、家庭机器人和移动机器人都需要本地感知、控制、联网和带安全意识的算力。因此,相关市场应包括边缘 AI 芯片、机器人中间件和开发者基础设施,同时排除通用 ERP 采购软件,以及不直接购买算力或开发工具的终端机器人服务收入。这个边界逻辑很关键,否则任何机器人或云算力统计都会显得可以被公司全部触达。[CM001, CM002, CM003, CM004, CM005, CM006]

市场定义表
细分 / 类别纳入支出排除支出主要买方 / 付款方相关性
机器人开发基础设施开发者套件、机器人 OS、参考算法、部署工具、边缘推理板卡成品机器人劳务收入,或没有机器人栈的通用 IT 硬件平台工程、机器人 R&D、OEM 项目D-Robotics 的核心市场视角
Cloud FinOps / 云成本优化成本管理、优化、成本回溯、计算资源责任归属通用企业财务软件和非技术采购工具FinOps 负责人、财务、基础设施负责人仅作为宽口径外层 TAM 边界
Kubernetes 成本管理节点配置、工作负载效率、集群归因、规格优化非容器应用监控和通用可观测性平台工程和 SRE最接近的软件类比
具身 / 人形 / 移动机器人计算感知、决策、控制、低功耗推理、本地 AI 硬件不采购基础设施的机器人运营服务机器人工程和 AI 基础设施团队与 D-Robotics 直接相关的高增长邻近市场
云原生 / 开源控制Autopilot、Karpenter、捆绑式成本管理、标准编排第三方专用机器人平台现有云平台团队重要替代品,也会吸走预算

该表将 D-Robotics 定义为机器人应用之下的基础设施;它有意排除终端服务型机器人收入,以及缺少机器人或计算优化内容的通用企业软件。

[CM001, CM003, CM004, CM006, CM007, CM016]
FM003: 买方 / 客群地图

采购通常从机器人或平台运营方起步,再扩展到财务和 AI 基础设施负责人,并在每一步都要与原生控制工具竞争。

[CM019, CM020, CM021, CM022, CM023, CM024]

2.2 规模测算视角与受约束的 TAM/SAM/SOM

公开市场规模只有在视角明确时才有用。MarketsandMarkets 给出宽口径 Cloud FinOps 框架,The Business Research Company 提供窄得多的 Kubernetes 成本管理视角,Verified Market Reports 则给出单独的云成本优化预测。没有一个口径能完美映射到 D-Robotics,因为它一部分落在软件效率里,一部分落在机器人专属边缘算力里。这意味着,直接拿一个大标题数字会夸大确定性。更好的方法是把宽口径报告当作外层 TAM 边界,再把可服务市场收窄到那些除成本和编排功能外,还需要本地机器人推理、实时控制或全栈开发工具的买方。这样得到的 2026 SAM 区间更克制但更可信,大致在低个位数十亿美元;第三方供应商可拿到的 SOM 还会更小,因为原生工具和既有平台套件会吃掉一部分预算。[CM008, CM009, CM010, CM011, CM012, CM033]

TAM/SAM/SOM 或规模测算视角表
发布方 / 视角年份 / 周期地域数值 / 区间(USD B)CAGR方法置信度局限
MarketsandMarkets Cloud FinOps2025-2030全球14.88 → 26.91页面摘录未给出宽口径 Cloud FinOps 市场页面对 D-Robotics 本身过宽
The Business Research Company 的 Kubernetes 成本管理市场2025-2030全球1.75 → 2.23 → 5.78摘录未给出更窄的软件效率市场报告排除机器人专用边缘计算层
Verified Market Reports 的 CCMO 市场2026-2034全球9.2 → 35.414.1%宽口径云成本管理与优化快照可能重叠多个邻近类别,方法透明度较弱
受约束的 D-Robotics 相关 SAM(作者估算)2026全球2.0 → 4.0n/a以 Kubernetes 效率视角和具身机器人计算邻近市场为锚没有公开来源隔离这个精确重叠类别
专用机器人计算优化器的第三方 SOM(作者估算)2026-2030全球0.3 → 0.8n/a假设原生工具和既有厂商保留有意义份额对替代和采用速度高度敏感

公开报告定义的类别差异很大,因此本章保留多个互不兼容的视角,而不是强行给出一个标题式 TAM。

[CM008, CM009, CM010, CM011, CM012, CM033]
FM001: 收窄后的市场口径

这张图说明,D-Robotics 实际只覆盖更大云和机器人支出池中经过筛选的一部分。

[CM008, CM009, CM010, CM012, CM033, CM034]
FM002: 市场估算区间

研究机构估算跨度很大,因为测量的品类和时间范围不同。

中点只是多年市场报告的分析锚点,不是厂商发布的年度数字。最后一行是作者估算,而非第三方市场数字。

[CM008, CM009, CM010, CM011, CM033, CM039]

2.3 买方、用户、付款方与采用路径

D-Robotics 的买方地图天然是跨职能的。平台工程、SRE 和机器人软件团队最像一线使用者,因为他们管理部署、节点利用率、传感器集成和机器人应用性能。财务与 FinOps 团队同样重要,因为效率和费用回显也是立项理由的一部分,尤其当机器人项目横跨云、边缘和混合环境时。随着 D-Robotics 推进具身智能,AI 基础设施负责人也会成为关键利益相关者,因为加速器效率、散热上限和软硬件兼容性会决定一个设计能否成立。因此,采用路径通常从原型或开发者项目开始,进入真实机器人形态上的评估;一旦客户能证明性能、可靠性和单位经济性,再放大到量产预算。官网上创客与企业用户的分层,也符合这条从实验到部署的路径。[CM005, CM019, CM020, CM022, CM023, CM024]

细分 / 买方地图
细分主要用户付款方 / 预算负责人工作流采用触发D-Robotics 为何重要
机器人原型 / 创客实验室开发者和机器人爱好者创新预算或创始人预算原型组装、测试、边缘推理实验需要易上手的机器人栈RDK 套件、示例和社区降低上手摩擦
平台工程 / SRE集群和基础设施运维方基础设施工程预算资源调度、利用率、节点经济性容器成本或复杂度变得重要优化和控制层开始和预算相关
FinOps / 财务FinOps 从业者和云财务分析师中央云预算或共享服务预算成本回溯、预测、政策、责任归属支出可见性和负责人缺口出现硬件效率和成本责任开始相互绑定
AI 基础设施 / 机器人自主团队感知、规划和自主工程师高级 R&D 或 AI 平台预算传感器融合、模型部署、运行时性能需要低功耗本地推理专用机器人计算可能跑赢通用技术栈
OEM / 量产机器人项目机器人产品团队和系统工程师项目或产品线预算从参考设计走向量产部署原型证明 ROI 和可靠性全栈比单一芯片规格更重要

用户和付款方往往是不同角色;对原型友好的开发者能打开入口,但规模化预算通常掌握在中央基础设施、AI 平台或 OEM 项目负责人手中。

[CM019, CM020, CM022, CM023, CM024, CM025]
FM004: 采用漏斗或价值链地图

机器人计算销售通常先从广泛开发者认知收窄,只有完成技术和经济性验证后,才进入量产机器人项目。

[CM019, CM024, CM025, CM040]

2.4 增长驱动与采用约束

最强的增长驱动并不是某一种机器人形态,而是三个问题的汇合:云原生复杂度、对高效推理的需求上升,以及物理系统中实时安全控制的必要性。Renesas、Ambarella、TI 和 MediaTek 都在强调低功耗、视觉 AI、连接能力或可安全落地的运动控制中的某种组合,说明市场扩张来自机器人现在每瓦都需要更多本地智能。与此同时,多重约束会限制 D-Robotics 这类供应商能抓住多少机会。GKE Autopilot 和 Karpenter 等原生工具会削掉部分第三方需求。OpenShift 成本管理等平台套件能在既有环境中提供可视化和费用回显。公开来源也仍无法精确拆出中国专属边缘机器人 SAM。结果是:这个市场有清晰的结构性增长,也有替代风险、品类边界模糊和真实的测量不确定性。[CM013, CM014, CM015, CM017, CM018, CM026]

增长驱动因素和约束表
驱动因素 / 约束方向时点影响尽调追问
云原生与 Kubernetes 复杂度驱动当前 / 持续支撑优化与自动化需求衡量客户仍有多少集群支出靠人工管理
低功耗边缘 AI 与散热上限驱动当前 / 持续利好为端侧推理优化的芯片和技术栈对比偏 GPU 替代方案的每瓦性能
人形与工业机器人需求驱动2026 年起抬高对实时控制、安全与传感器融合的要求确认 D-Robotics 最强的是 AMR、人形机器人还是其他细分场景
云原生替代方案约束当前 / 持续Autopilot 和 Karpenter 降低部分客户的采购紧迫性与托管服务和开源默认方案做基准对比
平台套件成本可视化约束当前 / 持续现有套件可满足成本归因和内部成本展示需求厘清买家在什么条件下仍需要专门的机器人平台供应商
品类边界模糊,TAM 可比项偏弱约束持续容易夸大市场规模和估值逻辑保留区间式 TAM 表述,避免单点确定性
数据清洗与打标负担约束当前 / 持续即使预算压力真实,也会拖慢价值兑现评估完整 ROI 兑现前客户需要投入多少实施工作

几个因素具有双刃剑属性:复杂度和边缘 AI 增长创造需求,也会加重实施负担,并强化更简单原生替代方案的吸引力。

[CM013, CM014, CM015, CM026, CM027, CM028]
Chapter 03

03竞争格局

3.1 竞争版图与方案类型

分析 D-Robotics 时,应把它同几类解决同一机器人算力任务的替代方案比较。公司自己的公开触点描述的是一个完整栈:开发板、基于 BPU 的推理、兼容 ROS 的中间件、文档和开发者上手路径。因此,它最接近的竞争对象不只是边缘 AI 芯片。NVIDIA 代表物理 AI 和人形机器人项目的高端既有路径,Jetson Orin 覆盖宽模块梯队,Jetson Thor 则走向更大内存和 Transformer 级推理负载。Qualcomm 通过嵌入式参考平台竞争,把 AI、连接能力和对 ROS 友好的开发套件打包在一起。Ambarella、Renesas、TI 和 MediaTek 从不同角度切入:低功耗视觉 SoC、工业机器人控制与安全、传感器融合和电机控制子系统,或通用边缘 AI 模块。最后,一些买方可以在共享基础设施问题上沿用 Autopilot、Karpenter 和既有 FinOps 工具,再自行集成机器人组件。关键尽调点在于,D-Robotics 卖的是碎片化市场中的一体化机器人算力路径,而不只是打一场单一基准测试竞赛。[CP001, CP002, CP005, CP009, CP010, CP011]

竞争对手画像表
竞争对手 / 类别品类规模 / 融资信号目标细分市场差异化局限
D-Robotics直接技术栈供应商2026 年 Series B 轮总额约 $270M;分拆信息披露仍有限AMR、人形机器人、边缘 AI 开发者、机器人 OEM低功耗 BPU 板卡,加 ROS 兼容中间件和开发者工具公开基准、定价和治理披露仍有限
NVIDIA Jetson既有物理 AI 平台全球既有玩家,Jetson 产品梯队宽,Thor 路线图清晰人形机器人、自主机器、高端边缘 AI已审阅对象中,可见算力覆盖最宽,物理 AI 软件栈最丰富对成本敏感的小型机器人项目来说,可能配置过剩或功耗偏高
Qualcomm Robotics RB 系列嵌入式参考平台竞品嵌入式平台谱系生命周期长;RB5 公开开发套件价格可见AMR、无人机、嵌入式机器人、IoT连接能力、对 ROS 友好的 Linux 支持,以及紧凑参考套件定位已审阅公开材料中,机器人生态牵引力不如 NVIDIA 明显
Ambarella以视觉为中心的低功耗竞品公开 AIoT 与机器人 SoC 目录消费机器人、AGV、机器视觉、物流强调低功耗 CVflow 计算机视觉和 AI 每瓦性能端到端机器人中间件叙事窄于 D-Robotics 或 NVIDIA
Renesas工业机器人既有玩家工业机器人和人形机器人解决方案组合较宽工业、移动、协作、家用、人形机器人视觉 AI 加电机控制、安全、工业网络和 ROS 2 就绪能力更偏子系统和工业技术栈,而不是单一开发套件叙事
Texas Instruments子系统与安全既有玩家控制与模拟业务根基深;公开人形机器人思想领导力人形与工业机器人设计方传感器融合、确定性通信、电源和电机控制积累深已审阅来源没有把它定位成单一机器人计算平台赢家
MediaTek Genio相邻边缘 AI 平台覆盖较广的 IoT 边缘平台,有多款伙伴设备智能设备、工业 HMI、生产力与安全系统通用边缘 AI 模块可能削弱专用平台需求已审阅来源中,机器人原生软件栈表述不够明确
现状 / 内部自建替代方案现有集群、云和工程预算基础设施团队和 OEM 在内部解决部分需求组合 Autopilot、Karpenter、FinOps 工具和自定义集成,避免供应商锁定单靠它无法解决机器人本地计算、运动控制或机器人中间件

行覆盖直接、既有、相邻和替代类别中最显眼的公开替代方案,而非全球每一家机器人半导体供应商。

[CP001, CP004, CP010, CP012, CP015, CP017]
FP001: 竞争定位图

从已审阅的格局看,竞争按堆栈掌控和性能野心分层:D-Robotics 位于创客友好型机器人集成和工业边缘 AI 之间,NVIDIA 则锚定高性能一端。

坐标轴使用有证据支撑的序数评分,而不是来源支撑的绝对数值。

[CP011, CP014, CP017, CP020, CP021, CP026]

3.2 竞品画像与买方决策标准

买方决策不只看 TOPS。D-Robotics 能讲出连贯的机器人故事,因为 RDK X5、RDK S100、TogetheROS.Bot、GitHub 文档、NodeHub 和社区触点共同压缩了从评估到跑通机器人演示的时间。对重视中间件、示例代码、传感器包和小板卡形态的开发者来说,这个栈有意义。NVIDIA 赢在光谱的另一端:更宽的 Jetson 梯队,以及面向人形机器人、智能体 AI 和重型多传感器负载设计的 Thor 路线图。Qualcomm 处在移动基因嵌入式 AI 与机器人参考套件之间;Ambarella 强调低功耗计算机视觉,Renesas 强调视觉加电机控制、联网和安全,TI 强调系统级人形机器人需求,MediaTek 则把 Genio 放在成本敏感的 IoT 和边缘 AI 设备中。实际采购时,买方会比较每瓦性能、ROS 与 SDK 成熟度、电机控制相邻能力、摄像头和传感器支持、价格透明度、伙伴可得性,以及必须由自己承担多少定制集成。[CP003, CP004, CP005, CP006, CP007, CP009]

功能 / 能力矩阵
采购标准D-RoboticsNVIDIAQualcomm RB工业技术栈(Ambarella / Renesas / TI)现状 / 内部自建
低功耗端侧推理
机器人中间件 / ROS 封装
电机控制 / 功能安全邻接能力
开放开发者接口和示例
前沿模型规模 / 内存余量
全球 OEM 渠道引力n/a
云 / 基础设施自动化替代价值
已审阅公开材料中的定价透明度

评分是基于已审阅公开证据的序位总结,并非经基准归一化的测量;未知或证据较弱的领域刻意不提升到“高”。

[CP005, CP009, CP010, CP014, CP016, CP018]
定价 / 打包对比
供应商 / 产品公开价格或合同模式包含能力未知项或折扣注意事项影响
D-Robotics RDK S100 开发套件日本零售价 ¥102,30080 TOPS BPU Nash 板卡,聚焦机器人开发,配件生态仅有板卡零售价;无公开 SDK 授权或量产芯片价格对开发者入门验证有用,但不足以证明量产经济性
D-Robotics RDK X5已审阅材料中未清晰披露10 TOPS 机器人 / 边缘 AI 套件,带软件生态本章材料未直接审阅到公开价格限制入门成本的同口径对比
Qualcomm RB5 核心 / 视觉套件$485 核心套件;$695 视觉套件15 TOPS 机器人开发套件,支持 Linux、Ubuntu、ROS 2 和连接能力较早参考平台快照;量产价格不可见说明 Qualcomm 历史上曾以透明开发套件价格参与竞争
Qualcomm RB3 Gen 2公开套件打包可见;已审阅来源没有给出清晰价格摘录12 TOPS QCS6490 平台,配摄像头和传感器无法从已审阅材料确定当前套件价格评估阶段有竞争力,但不足以支撑 TCO 论证
NVIDIA Jetson Orin / Thor公开产品页更强调性能和伙伴,而非价格Jetson 梯队覆盖面广,从较小 Orin 模块到面向人形机器人的 Thor定价需要在已审阅核心页面之外查询伙伴或渠道渠道覆盖宽度可能比标价更重要
现状工具Autopilot 按用量计价,Karpenter 开源,FinOps 工具包开源无需专门机器人供应商即可获得基础设施自动化和成本控制不覆盖机器人本地计算和控制层替代方案试用成本低,可能推迟专门供应商采用

已审阅公开材料在开发套件定价上远比在量产模块、SDK 或企业合同经济性上更充分。

[CP012, CP013, CP022, CP023, CP024, CP032]
FP002: 功能宽度 / 能力地图

D-Robotics 在中间件、示例和低功耗机器人套件的组合上最强;竞争对手则分别主导原始性能或工业子系统宽度。

这是从产品页和文档提炼出的定性热力图,不是经过基准归一化的测试数据。

[CP005, CP006, CP010, CP016, CP018, CP019]

3.3 切换成本、分销力量与替代品

只有当机器人团队已经押注某个技术栈之后,切换成本才会上升。采购早期,多栈并行是可行的,因为开发者可以混用 ROS2 工作流、云工具、传感器供应商和参考套件,同时保持应用逻辑可迁移。之后,成本会围绕开发板 I/O 假设、加速器专属模型工具、中间件包、调试习惯和渠道关系上升。这种动态有利于拥有广泛全球渠道和已知 OEM 信誉的既有玩家。NVIDIA 在高端最有生态引力,Renesas 和 TI 等工业供应商则受益于长期控制、安全和联网关系。D-Robotics 也有对冲因素:开放 GitHub 触点、兼容 ROS 的中间件、开发者示例、社区上手路径,以及日本零售和第三方产品采用带来的至少一些外部渠道证明。但现状替代品仍然真实存在。基础设施团队可以先用 Autopilot、Karpenter 或内部自建路径解决部分运营问题,直到机器人出货量或延迟需求逼近临界点,再决定是否引入新的专用平台供应商。[CP006, CP007, CP022, CP023, CP024, CP025]

3.4 护城河耐久性、商品化与替代风险

D-Robotics 护城河最强的版本不是公开基准测试的绝对领先,而是一体化可达性。当买方需要低功耗本地推理、机器人中间件、示例和从套件到部署机器人更快的路径时,D-Robotics 相比纯芯片供应商或纯云替代方案会显得不同。这个优势真实存在,但很脆弱。公开竞品证据显示,多家供应商现在都把 AI 加速与软件工具链、长生命周期支持或工业参考设计打包,削弱了“硬件 + SDK”基础卖点的独特性。NVIDIA 也用软件丰富、高性能的物理 AI 平台抬高买方预期上限,Renesas 和 TI 则把买方注意力拉向完整系统安全和电机控制集成。同样重要的是,本次审阅的公开材料没有提供同口径基准测试、TCO 研究或量产价格矩阵,证明 D-Robotics 能在不同负载中取胜。因此,投资人应把护城河视为有潜力但尚未充分证明;基准测试、渠道和商品化风险仍需重点尽调。[CP027, CP028, CP029, CP031, CP033, CP034]

护城河耐久性 / 竞争风险台账
护城河主张威胁严重性重要性缓解措施 / 尽调追问
芯片 + 中间件 + 示例一体化竞品 SDK 商品化多个竞争对手已把 AI 芯片与工具链、参考设计打包索取客户赢单 / 输单原因,以及软件层绑定率数据
低功耗机器人聚焦NVIDIA 形成的既有性能上限高端人形机器人买家可能以 Thor 级能力为锚,而非小形态效率按工作负载、瓦特、延迟和 BOM 档位索取基准数据
开放开发者接口量产买家仍偏好既有渠道GitHub 开源热度不保证 OEM 转化或支持能力按原型、试点、量产阶段审阅项目管线
区域 / 渠道扩张全球分销劣势伙伴和零售验证存在,但既有 OEM 渠道仍更强索取已签渠道、设计导入和地区收入组合数据
ROS 兼容集成内部自建仍可行成熟团队可组合开源基础设施和混合硬件栈衡量相较内部替代方案的价值兑现时间和切换摩擦
公开产品动能基准和定价不透明没有清晰公开矩阵证明 D-Robotics 在各类工作负载的 TCO 或性能上胜出在 NDA 下获取价格表、基准方法和客户背书

多数风险是执行和证明风险,而不是产品缺乏竞争力的品类层面证据。

[CP029, CP031, CP033, CP034, CP035, CP037]
FP003: 护城河 / 准备度 KPI

竞争档案中,集成和生态信号最强;公开基准和量产价格证明最弱。

分数是基于证据广度的 1-5 分析判断,不是经审计的公司 KPI。

[CP025, CP029, CP033, CP034, CP035, CP037]
Chapter 04

04财务

4.1 收入模型与变现姿态

D-Robotics 公开呈现出的商业姿态,更像赋能型平台供应商,而不是机器人 OEM。最清晰的外部描述来自 36Kr:公司不打算制造机器人本体,而是想成为许多机器人厂商共用的软硬件底座。这个定位与已抓取的官方触点一致:公司把 RDK X5 和 RDK S100 推为开发与算力平台,发布 TogetheROS.Bot 手册,维护大型 GitHub 组织,并运营围绕 BPU 的模型库。合在一起看,可见栈暗示了三层可变现空间:硬件套件和算力模块、软件 / 工具链赋能,以及伙伴 / 开发者生态服务。 缺失的一环是价格兑现。公开页面展示了产品能力和上手材料,但没有给出标价、折扣区间、附加销售率或经常性软件合同条款。因此,最站得住的财务解读是结构性的,而不是定量的:D-Robotics 看起来通过开发者入口和下游平台采用变现,但公开来源尚未说明收入中有多少来自开发板销售、芯片 / 模块项目或软件挂钩服务。[CI001, CI002, CI003, CI004, CI005, CI006]

收入流表
收入流机制公开状态收入质量尽调追问
RDK 开发套件向开发者和机器人团队销售 RDK X5 / RDK S 系列开发硬件产品页和教程可见;实际销量未披露硬件收入,可能偏交易型且依赖发布节奏按板卡系列披露出货量、ASP 和复购率
嵌入式计算芯片 / 模块向下游机器人制造商销售平台计算能力S100 和 BPU Nash 算力定位可见;商业设计导入数量未披露潜在价值高,但项目收入不透明披露活跃设计导入、出货爬坡和模块 / 芯片定价基础
软件 / 工具链赋能TogetheROS.Bot、模型库工作流、ISP 训练、工具链手册公开赋能界面丰富;合同模式未披露可能提升粘性,但变现形式不清晰厘清哪些工具免费、打包、付费支持或按企业授权收费
伙伴生态服务联合营销、集成支持、生态入驻、赛事伙伴和活动动态可见;服务收入未披露更像间接变现和需求生成,而非清晰 ARR披露伙伴返利、认证费和联合开发经济性
教育 / 培训项目课程交付、比赛和校企合作高参与度代理指标可见;收入捕捉不清晰漏斗信号好,但直接收入可见度弱披露培训由补贴、付费还是渠道支持

公开证据支持多个变现界面,但没有来源披露收入流层面的收入组合或实际定价。

[CI001, CI002, CI003, CI004, CI007, CI010]
定价 / 变现表
产品公开标价可见?观察到的证据影响
RDK X5 套件产品页展示规格和接口,但没有结账价格开发者能了解能力,投资人却无法推断 ASP
RDK S100 平台产品页营销具身 AI 计算和架构,但没有商业条款高端平台价值主张可读,变现仍未公开
工具链 / 模型库资产GitHub 和手册展示了丰富工具界面,但没有公开授权价目表软件可能免费、打包,或作为战略获客线索,而非独立收入
培训课程课程 API 显示 lookCount 和 learnCount,而非票价或订阅价格社区牵引力可衡量;收入转化不可见

本表记录披露可见度,而非隐含价值。没有价格本身就是尽调信号。

[CI005, CI006, CI009, CI010, CI024]
FI001: 收入模型桥

公开可见界面显示,D-Robotics 把平台 R&D 转化为硬件销售、软件赋能和合作伙伴带动的下游采用,而不是直接赚机器人本体收入。

[CI001, CI002, CI007, CI012, CI025]

4.2 市场进入动作与销售效率代理指标

最强的公开销售效率证据不是已确认收入,而是生态强度。D-Robotics 维护着 229 个仓库的 GitHub 组织、带端到端部署示例的模型库、已有可观浏览量的课程目录、近期竞赛,以及一张覆盖教育机器人、工业相机、陪伴机器人和双臂具身系统的增长伙伴名单。这些不能替代 CAC 或回本周期数据,但确实暗示了一条开发者主导漏斗,可以降低售前摩擦并扩大产品发现面。伙伴动态也说明,D-Robotics 用外部硬件厂商和集成商做下游分销,而不是自己制造所有终端产品。 同一模式的反面也能在支持触点上看到。近期论坛主题仍包括 CAN FD 问题、掉帧报告、工具下载链接缺失和模型转换精度投诉。再叠加 CSDN 上的外部上手教程,证据指向一个仍在成熟中的赋能负担。D-Robotics 可能用较低的直接销售强度,换来更高的文档、支持和生态投入成本。[CI008, CI009, CI010, CI011, CI012, CI013]

单位经济性表
指标公开数值重要性当前置信度尽调追问
混合 ASP / ACV需要区分板卡销售经济性与嵌入式项目经济性按系列提供硬件 ASP,并按客群提供软件绑定率
毛利率判断“芯片 + 支持”模式能否持续的关键按硬件、软件和服务提供毛利率
CAC / 回收期生态驱动 GTM 可能用支持 / 社区成本替代销售成本提供付费获客、渠道和社区来源客户的成本数据
单个活跃开发者支持成本论坛和教程密度暗示入门支持负担不轻提供工单量、文档维护人员数和支持 SLA 成本
单个活跃伙伴收入只有收入集中或能规模化时,伙伴广度才有意义提供活跃伙伴数、头部伙伴收入和返利结构

抓取到的公开来源均未披露这些单元格;空值表示不可获得,不是零。

[CI010, CI012, CI014, CI015, CI021, CI031]
FI004: 单位经济性桥

从开发者发现到下游部署,公开漏斗能看清;但经济转化点没有公开。

这是定性桥接,因为 CAC、回本期、转化率或留存指标都没有公开披露。

[CI008, CI010, CI012, CI013, CI014, CI015]

4.3 融资历史与资本充足性

可访问的公开来源足以证明近期融资动能,但不足以干净地判断资本充足性。Baidu Baike 记录了 2025 年 5 月 $100M A 轮、2026 年 3 月 $120M B1 轮和 2026 年 4 月 $150M B2 轮,仅 B 轮资本就达到 $270M。公开来源也把 D-Robotics 描述为新兴独角兽,这与用户简报中的后期增长框架方向一致,但已抓取页面没有发布明确的投后估值表、投资方名单,也没有说明融资款如何转化为资产负债表现金续航。 这个缺口很关键,因为 D-Robotics 正在搭建全栈机器人平台:芯片、套件、软件手册、培训内容、社区支持和伙伴计划。这些动作都意味着持续的研发、生态和支持开支。缺少账上现金、烧钱速度、库存和毛利率披露时,审慎结论只能是:近期 B 轮降低了融资紧迫性,但没有把下一轮融资依赖从投资假设中拿掉。[CI016, CI017, CI018, CI019, CI020, CI024]

资本充足性表
项目公开证据置信度含义
A 轮2025 年 5 月:Baike 称 $100M确认平台扩张在 B 轮前已需要重资本投入
B1 轮2026 年 3 月:Baike 称 $120M显示进入 2026 年时公司仍需要新资金
B2 轮2026 年 4 月:Baike 称 $150M;B 轮合计 $270M缓解短期融资压力,但没有披露资金可支撑期
现金 / 资金消耗 / 可支撑期公开资料无法支撑资金可支撑期判断
募资用途从芯片、套件、软件、培训和生态触达等方向推断资金大概率同时投向研发和采用基础设施,而不只是芯片流片

融资金额来自二手资料;现金余额、资金消耗和可支撑期仍未披露。

[CI016, CI017, CI018, CI020, CI024, CI034]
FI003: 资本强度 / 现金流地图

资本负担看起来分散在芯片、套件、软件、培训和支持上,而不是集中在单一成本池。

[CI007, CI010, CI011, CI014, CI024, CI032]

4.4 公开牵引力与私有指标缺口

公开记录对生态牵引力的证明远强于对收入质量的证明。Baidu Baike 引述称,到 2025 年末,平台已有超过 100 款产品、超过 100 家伙伴、100,000 名开发者,并与 500 多所学校合作。社区 API 和课程目录显示上线后确有活动,开发者案例流也确认第三方在机械臂、四足机器人、智能健身、服务机器人和安全系统上有真实构建。这些指标有用,说明平台正在被尝试、教授和集成。 它们不能证明收入耐久性。已抓取的公开来源没有披露合同收入、ARR、毛利率、流失率、客户集中度或产品级收入结构。即便伙伴覆盖很宽,也不能直接转化为多元化,因为没有公开的收入拆分能按客户类型、细分市场或地理区域划分。结果是一个熟悉的私有硬件平台问题:D-Robotics 看起来商业活动活跃,但数据室之外,收入质量仍不透明。[CI021, CI022, CI023, CI031, CI033]

公开财务缺口表
缺失项目对分析的影响确切尽调路径
收入 / ARR / 增长缺少这项,估值和销售效率分析只能停留在结构层面按产品线索取月度收入和年化收入规模
按收入流划分的毛利率无法判断套件、芯片或支持服务是否能盈利扩张按板卡、芯片 / 模组、软件 / 支持类别索取产品 P&L
现金 / 资金消耗 / 可支撑期无法评估下一轮融资触发点或资本充足性索取月度现金桥和未来十二个月运营现金消耗
B1/B2 投资方名单和条款无法判断信号质量、清算顺位或治理增补索取已签署融资摘要,列明投资方名称和证券条款
客户与伙伴集中度无法把生态广度转化为收入韧性索取前 10 大客户、前 10 大伙伴和按细分市场划分的收入结构

这些是把平台故事落成可投资财务模型所需的最低数据室材料。

[CI020, CI021, CI030, CI031, CI033, CI035]

4.5 财务结论与投资判断卡点

从结构看,D-Robotics 有几个正面因素。它卖向宽口径机器人赋能层,而不是单一设备利基;它围绕硬件有可见的软件和社区资产;它也已经募集足够多的后期私募资本,支撑 2026 年继续扩张。NVIDIA Jetson 等公开可比平台也说明,边缘 AI 供应商可以靠打包算力、开发套件和软件生态捕获价值,而不只是卖芯片。 但投资判断卡点很重。公开证据没有揭示实际成交价格、芯片与套件收入结构、软件变现、烧钱速度、现金续航、客户集中度或单位经济性。2026 年 B 轮投资方身份无法从可访问页面完整恢复,精确估值仍是二手来源估计,而不是公开申报事实。财务上,D-Robotics 更像一个资金充裕但尚未披露关键数据的增长平台:定性模型能看懂,定量可投资性仍未解决。[CI026, CI027, CI028, CI029, CI030, CI033]

FI002: 财务估算区间

抓取到的来源没有正式披露营收、烧钱速度和估值,因此只能给出方向性的公开区间。

找到的公开硬数字只有融资轮次金额;估值和资金跑道仍只是估算性的背景标记。

[CI016, CI017, CI018, CI020, CI024, CI030]
Chapter 05

05产品与技术

5.1 产品组合与开发者工作流

D-Robotics 的公开栈足够宽,支撑的是可信平台叙事,而不是单板故事。RDK X5 被呈现为机器人开发套件,具备较强通用 I/O、Ubuntu 22.04 支持、有线和无线连接,以及创客到原型开发者都熟悉的工作流。RDK S100 在栈中的位置更高,被定位为具身智能算控平台,并明确使用“大脑 + 小脑”表述,把应用算力与运动控制配对。在这些硬件层周围,D-Robotics 公开了机器人软件手册、GitHub 组织、面向 BPU 部署的模型库,以及帮助开发者从首次启动走到模型执行的培训触点。 这形成了一条可见工作流:选择硬件、烧录并配置开发板,使用机器人中间件和示例,把模型移植或量化为 BPU 兼容格式,然后把开发板集成进伙伴或自建机器人。这是一条连贯的开发者旅程,不过仍面向能够承受配置工作和社区驱动排障的建设者。[CE001, CE004, CE005, CE010, CE012, CE021]

产品模块 / 资产矩阵
模块 / 资产主要用户状态 / 成熟度差异化尽调缺口
RDK X3入门创客、教育场景、低成本机器人集成商已公开发布并有文档四核 A53 + 双核 Bernoulli BPU,5 TOPS,4K 编解码,载板形态未公开 BOM、定价或生命周期停售日期
RDK X5机器人开发者、创客、集成商已公开发布并有文档中文页面显示 I/O 丰富,支持 Ubuntu 22.04、CAN FD、PoE、Wi-Fi 6/BT 5.4未公开 BOM、定价或生命周期窗口
RDK S100具身 / 人形机器人开发者2026 年已公开推广CPU+BPU+MCU 拆分,搭载 BPU Nash,并强调实时控制未公开基准测试表或已出货设计定点名单
RDK S600更高阶具身智能 / 端云开发者DDC 2025 发布;目标 2026 年初推出,尚未看到出货证据560 TOPS INT8 算力,主打端云协同;宣称运行 Pi-Zero / Qwen2.5-VL-7B 时相对主流平台有 2.2x+ 优势发布后未确认独立基准测试、出货日期或价格
TogetheROS.Bot面向 ROS2 的机器人开发者公开手册已上线把硬件接到 SLAM、语音控制和 ROS2 演示未公开企业支持矩阵
NodeHub希望直接使用预构建节点 / 算法、而非自建流程的开发者公开目录已上线宣称一分钟部署节点;项目代码在 GitHub 开源;算法和演示数量从 100+(X3 时期)到 200+(S100/DDC-2025 时期)未公开目录规模审计、版本策略或维护 SLA
RDK Studio需要用一个桌面工具管理多块板卡 / 主机的开发者公开工具和文档已上线可刷写 / 管理 RDK X3、X5、S100/S100P;通过 SSH 连接 Jetson、Raspberry Pi、Rockchip 主机;支持 “OpenClaw” 部署路径除 GitHub 硬件支持说明外,公开文档偏薄;没有采用率或可靠性指标
RDK 模型库面向 BPU 部署的 AI 开发者活跃公开仓库覆盖多类模型的端到端转换和验证示例没有正式成功率、延迟或维护 SLA
DGP / 伙伴赋能创业公司和下游机器人厂商公开项目页活跃明确强调 MVP 加速和生态支持商业条款和支持范围未披露

成熟度来自在线文档、仓库和课程维护推断,不来自经审计的产品生命周期披露。RDK S600 仍处发布阶段,尚未确认出货。

[CE001, CE005, CE010, CE012, CE021, CE027]
工作流 / 使用场景表
用户任务当前工作流D-Robotics 层可衡量收益限制
原型化视觉机器人板卡设置、摄像头接入、模型部署、应用测试RDK X5 + 模型库 + 入门文档加快支持板卡上的首次部署仍需要熟悉刷机、联网和 Linux
搭建具身控制栈在计算资源之间拆分感知、规划和关节控制RDK S100 架构公开叙事称其更适合具身负载未发布独立延迟基准测试
增加 ROS2 机器人功能安装中间件并运行打包演示TogetheROS.Bot 手册减少 SLAM 和机器人演示的样板工作生产级加固范围不清楚
将 ONNX/CV 模型移植到 BPU转换、量化、验证并运行示例rdk_model_zoo / 工具链缩短推理启动调试,并给出参考模式不支持的算子可能回落到 CPU
加速创业公司 MVP使用伙伴 / DGP 支持和预构建栈层DGP 项目 + 公开文档可能缩短机器人 MVP 周期商业支持条款未公开

收益停留在工作流层面且面向公开叙事;没有来源发布量化的转换耗时节省或工程工时节省。

[CE001, CE007, CE010, CE012, CE015, CE021]
FE002: 客户工作流 / 操作流程

抓取到的公开工作流从开发板启动走到模型部署,再进入下游机器人集成。

[CE001, CE010, CE012, CE016, CE026]

5.2 RDK X3 到 S600 的硬件梯队与平台工具(NodeHub、RDK Studio)

公开目录呈现的是分层硬件梯队,而不是单一开发板。RDK X3 是 D-Robotics 的入门层——一个四核 Cortex-A53 模块,配备双核 Bernoulli 架构 BPU,额定 5 TOPS,支持 4K 编解码,载板形态兼容通用机器人底盘。RDK X5 位于其上,推理预算约翻倍(10 TOPS),无线 I/O 也更丰富。RDK S100 进一步抬高上限,采用本章其他部分描述的 CPU+BPU Nash+MCU 分工;D-Robotics 还公开宣布了作为梯队顶端的 RDK S600:一款 560-TOPS(INT8)算力平台,在 DDC 2025 大会上发布,面向具身智能负载,目标 2026 年初上市,并明确围绕边云协同而不是纯端侧算力来定位。 两个工具触点包裹着这条硬件梯队。NodeHub 是 D-Robotics 公开的“智能机器人应用中心”——一个即取即用节点和算法示例目录(视觉巡线、室内服务机器人导航、机械臂演示),公司称大约一分钟即可跑通部署,项目源码也公开发布在 GitHub。RDK X3 和 RDK S100 营销页都直接引用了这个目录深度,从 X3 发布时的“100+ 开源算法和应用”,发展到 S100/DDC-2025 阶段的“200+ 开源算法和应用演示”。RDK Studio 是第二个触点:一款“AI 原生工作台”桌面工具,可烧录和管理 RDK X3、X5、S100/S100P 开发板,也能通过 SSH 连接通用 Linux 主机、NVIDIA Jetson、Raspberry Pi 和 Rockchip 开发板,提供范围更窄的 RDK 专属功能,并提供一条受 SSH 或网络可达性约束的“OpenClaw”端侧部署路径。合在一起,硬件梯队和工具强化了同一套平台论点:D-Robotics 卖的是一段可逐级上探的算力范围,以及配套部署和算法工具,而不是一块旗舰板。[CE036, CE037, CE038, CE039, CE040, CE041]

5.3 架构与算控设计

已抓取来源中最有辨识度的技术主张是 S100 架构。D-Robotics 不只是营销更多 TOPS,而是描述了一种横跨 CPU、BPU Nash 和 MCU 资源的算力分工,让感知、规划和关节级控制不必都挤在同一个处理器预算里。公开文本称,BPU Nash 承担 80/128 TOPS,支持超过 160 个 ONNX 算子,并针对 CNN 和 Transformer 优化;Arm Cortex-R52+ MCU 资源则处理低延迟执行。相比之下,RDK X5 更像经典边缘 AI 开发板,提供丰富 I/O 和无线接口用于原型开发。 软件架构补上了这套分工。TogetheROS.Bot 展示安装、ROS2 包使用、SLAM 和语音控制演示,模型库则记录转换、量化、推理和验证流水线。结果是一张相当具体的公开架构图:底层是异构硬件,中间是机器人中间件,上层是以 BPU 为中心的模型工具。[CE002, CE003, CE006, CE007, CE008, CE010]

技术 / 运营架构表
层 / 组件角色依赖风险
RDK X5 板卡硬件用于原型开发的边缘计算和传感器 I/O板级支持包、驱动、文档社区仍出现硬件设置和接口问题
RDK S100 CPU+BPU+MCU 拆分拆分具身感知与控制BPU Nash + MCU 协同公开文档描述架构,但没有实测控制延迟
RDK S600 端云工具链端侧 BPU 算力配合云端数据闭环、仿真和智能体式开发工具DDC 2025 发布的云服务和自然语言智能体工具仍处发布阶段;尚无公开 GA 日期、定价或独立基准测试
TogetheROS.Bot面向 ROS2 的中间件和演示层包维护和文档新鲜度未公开企业支持合同或兼容性矩阵
NodeHub 算法中心用于快速组装机器人的预构建节点 / 算法目录托管在 GitHub 的节点源码、目录维护未发布目录审计、版本管理或长期维护承诺
RDK Studio覆盖 RDK 产品梯队和部分第三方主机的桌面刷写 / 管理工作台SSH/Type-C 设备可达性、逐板卡刷写流程除 GitHub 说明外,公开文档偏薄;没有独立可靠性证据
模型库 / 工具链模型转换、量化、推理、验证算子支持和工具版本不支持的算子可能回落到 CPU,改变性能
社区培训和支持入门和问题解决触点课程更新、论坛维护、外部教程对支持依赖过高会拖慢机构采用

这张架构表结合官方文档和公开仓库证据;风险单元格标出正式披露仍缺的内容。

[CE006, CE008, CE010, CE012, CE015, CE025]
FE001: 产品架构图

D-Robotics 的公开架构把硬件异构性、机器人中间件和 BPU 部署工具打包成一个开发者栈。

[CE005, CE006, CE010, CE012, CE027, CE031]

5.4 成熟度、部署与发布证据

公开成熟度证据最强的地方在文档、分支、课程和持续生态维护。rdk_x5 分支被明确标为 RDK X5 的主交付分支,S100 硬件文档页则说明内容在 2026 年 6 月迁移到新的文档中心,而不是被弃置。课程流显示有专门的 X5 ISP 和 S100 ISP 培训,也有较早的 YOLOv5 和工具链课程,并且互动量不低。这就是一个活着的开发者平台应有的样子:不像消费级电器那样打磨完美,但在持续维护。 但公开成熟度不等于企业级资质。当前最好的证据是版本迁移、培训刷新、示例广度和使用案例,而不是正式认证包或基准测试记分卡。买方可以推断 D-Robotics 在积极维护平台;尽调团队仍无法仅凭已抓取页面推断确定性性能边界或生命周期保证。[CE009, CE013, CE016, CE017, CE018, CE032]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
2024-09RDK X5 公开发布背景历史发布,仍在活跃标志其转向面向广泛机器人品类的千元以下开发套件定位36Kr
2024-11X5 ISP 培训在线课程材料显示发布后继续投入开发者教育高校课程动态
2025-10S100 ISP 培训在线课程材料显示平台推出后已有 S100 专属赋能内容高校课程动态
2026-06-10S100 文档迁移到新文档中心文档迁移已完成表明仍在维护并刷新信息架构S100 硬件文档页
2026-07活跃论坛和近期竞赛内容持续平台仍在社区里被教学、调试和扩展论坛 + 活动动态

公开路线图信号来自文档更新、培训发布和生态活动推断,不来自正式前瞻产品路线图材料。

[CE009, CE016, CE018, CE023, CE025, CE032]
FE004: 产品成熟度 / 能力地图

公开证据在面向开发者的能力宽度上最强,在生产级验证披露上最弱。

[CE009, CE012, CE015, CE025, CE028, CE029]

5.5 差异化与生态触达

D-Robotics 的差异化不在某个单一头部参数,而在周边生态密度。GitHub 组织公开了 229 个仓库,模型库覆盖从分类到 OCR 和多模态任务,DGP 页面明确要压缩机器人 MVP 时间,合作伙伴动态把平台连接到传感器、教育机器人、陪伴机器人和具身系统。开发者案例流进一步证明,第三方正在这个栈之上做出真实作品,而不只是阅读营销页。 这并不意味着 D-Robotics 是唯一尝试打包模块、开发套件、软件和伙伴的公司;NVIDIA Jetson 显然是这条全栈边缘 AI 路线的标尺。但已抓取的 2026 年来源确实支持一个有差异的本地论点:D-Robotics 正把机器人原生、BPU 优先的平台推入具身和教育场景,并拥有可见的中文社区密度和伙伴宽度。[CE011, CE014, CE019, CE020, CE021, CE023]

信任 / 质量 / 合规表
控制 / 质量信号状态范围缺口
在线硬件文档和手册板卡设置、ROS2、演示、硬件介绍没有可下载验证报告或审计包
有版本的培训入口X5 ISP、S100 ISP、YOLOv5、工具链培训互动量不是生产可靠性的证明
开放仓库和样例BPU 部署示例和按板卡区分的分支未抓取到正式支持 SLA 或安全公告节奏
活跃论坛维护近期 X5 和 S100 主题下的故障排查社区活动也暴露未解决摩擦
伙伴证据机器人、传感器、摄像头、健身系统伙伴存在不等于认证互操作

平台呈现持续维护信号,但公开信任 / 合规证据仍以开发者为中心,还不到采购级。

[CE011, CE016, CE019, CE025, CE028, CE029]
FE003: 关键依赖地图

除芯片性能外,D-Robotics 还依赖社区、合作伙伴和面向特定开发板的工具。

[CE019, CE020, CE021, CE025, CE034, CE035]

5.6 技术风险、信任缺口与尽调问题

公开负面证据是实际使用层面的,而不是根本性的。近期论坛主题显示,用户仍会遇到 CAN FD 问题、掉帧、GPIO 接触问题、gdc-tool 下载链接缺失和模型转换精度疑问。首次上手仍需要外部教程。这些对一个活跃开发者平台来说正常,但很重要,因为它们暴露了集成成本仍会回流到客户身上的位置。它们也暗示,D-Robotics 的产品成熟度仍部分依赖社区排障,而不完全是开箱即用体验。 更大的信任缺口是披露完整性。已抓取来源没有披露正式的正常运行时间承诺、安全证明、可下载审计包,也没有一张工业买方能直接塞进供应商记分卡的清晰基准测试表。D-Robotics 对严肃试点和伙伴主导建设看起来很强。若要支撑量产级技术尽调,它仍需要在可靠性、认证范围和确定性性能上拿出更深的公开证明。[CE015, CE025, CE026, CE028, CE029, CE033]

Chapter 06

06客户

6.1 细分:消费级机器人 OEM、DGP 创业公司与机构型人形机器人买方

D-Robotics 的公开客户记录横跨五个不同买方触点,而不是一个清晰的企业理想客户画像。公司自己的关于页面把核心商业关系定义为覆盖人形机器人、陪伴机器人、扫地机器人和割草机器人的 OEM 合作;CEO Wang Cong 对 36Kr 表示,当前出货量仍依赖大众市场品类——扫地机、割草机和陪伴机器人——而不是具身智能或人形机器人订单,后者被他界定为更早期、周期更长的押注。叠加在这个 OEM 基础之上的是 D-Robotics Gravity Program(DGP):公司称该加速器已经支持 200 多家早期机器人公司,产出 100 多个机器人产品品类;其开发者门户 API 还独立列出 9 个实名合作伙伴,覆盖教育机器人、可穿戴设备、高尔夫球童机器人和智能健身。第三个触点是 Beijing Innovation Center of Humanoid Robotics 这类机构型人形机器人买方,该机构正在围绕 D-Robotics 的 Xuri S600 芯片联合开发 Tiangong 3.0 人形机器人。第四个触点是 Yicai 和 D-Robotics 自己的 myCobot 博文都提到的 100,000+ 全球开发者社区;第五个触点则是 Insta360 支持的 Antigravity 团队等消费级无人机和影像伙伴。每个触点的买方—用户—付款方逻辑都不同:OEM 采购芯片并把成本传导给终端消费者;DGP 成员实际以购买硬件以及股权 / 注意力资源付费,而不只是现金;机构型人形机器人买方共同投入工程时间;开发者社区则多以免费或低价硬件为主,并不会成为具名商业账户。这种宽度在战略上是正面的,因为它降低了对单一机器人品类的依赖,但也意味着“客户”必须拆开理解:社区门户上的开发者案例投稿绝大多数是未付费的学术或创客项目,而不是商业买方;D-Robotics 自己“全球数百万用户”的表述,也部分借用了 Narwal 等下游 OEM 品牌的规模,后者独立报道的 450 million 名用户属于 Narwal,而不是 D-Robotics 自己可归因账户。[CU001, CU003, CU004, CU005, CU006, CU023]

客户细分表
细分市场买方 / 用户 / 付款方主要使用场景规模 / 证明状态收入或战略价值缺口
消费级机器人 OEM(扫地、割草、陪伴)OEM 厂商采购芯片 / SDK;家庭终端用户使用成品用于自主清洁、割草和陪伴机器人的嵌入式计算公开案例点名 Narwal 的 Xiaoyao 002 与 Hengbot 的 Sirius;CEO 称这是近期收入底盘公司表述中主要已披露的出货量驱动项未披露各 OEM 的出货量或收入拆分
DGP 早期机器人创业公司创业团队加入加速器;D-Robotics 提供硬件折扣和技术支持在 RDK 板卡上孵化、原型开发和产品开发DGP 项目声称支持 200+ 家早期公司;开发者门户具名 9 个合作伙伴未来 OEM / 芯片收入与生态锁定的漏斗未公开 DGP 成员转化为付费且出货客户的比例
消费无人机 / 影像合作伙伴Insta360 支持的 Antigravity 团队采购并集成 Sunrise 5 芯片;终端消费者购买成品无人机Antigravity A1 360 度无人机中的 AI 视觉感知与避障China Biz Insider 称这是消费无人机首个商业化应用;Insta360 自有产品页未提及 D-Robotics把 TAM 从地面机器人拓到空中影像芯片归属来自第三方报道,不是 Insta360 自有营销材料
人形 / 具身智能机构Beijing Innovation Center of Humanoid Robotics 联合设计并采购 Xuri S600 芯片;工业 / 商业终端用户将使用 Tiangong 3.0面向制造、物流和复杂 3D 环境的全尺寸通用人形机器人Gasgoo 和 en.shuziqushi.com 称,量产交付计划在 2026 年下半年具身 AI 芯片能力的战略证明,但截至报告生成日尚未贡献收入量产时间表尚未兑现;没有已交付台数
全球开发者 / 创客社区个人开发者和学生购买 RDK 套件或使用免费档;部分人成为 DGP 传播者教育、爱好者机器人、竞赛和开源项目Yicai 与 D-Robotics 自有博客称开发者社区 100,000+;展示了数十个学生项目未来付费客户和社区生成内容的漏斗社区规模不等于商业客户数
四足陪伴机器人(工业 / 消费)Vbot 等 OEM 为陪伴型四足机器人采购芯片AI 驱动的四足陪伴机器人截至报告生成日,仅一篇 36Kr/KrASIA 资料点名把品类敞口从扫地机、割草机和无人机进一步分散单一来源说法,未找到独立佐证

细分市场按公开证据强弱排序;「缺口」记录尽调团队仍无法从公开资料验证的内容。

[CU001, CU003, CU004, CU005, CU013, CU016]
DGP 合作伙伴目录
伙伴(本地名称 / 英文)产品名称产品品类证据类型来源
乐聚机器人(Leju Robotics)Aelos Embodied 教育机器人教育机器人官方开发者门户伙伴列表D-Robotics 合作伙伴 API
湾侧科技(Wance Technology)LT 系列双目安全传感器工业相机 / 安全感知官方开发者门户伙伴列表D-Robotics 合作伙伴 API
思博威视(Sibo Weishi)羽毛球精彩片段捕捉 AI 相机智能体育官方开发者门户伙伴列表D-Robotics 合作伙伴 API
恒之未来(Hengbot)Sirius 四足陪伴机器人AI 陪伴机器人官方开发者门户伙伴列表,加 D-Robotics 专门博客发布文章D-Robotics 合作伙伴 API;D-Robotics 博客
求之科技(Qiuzhi Technology)MMK2 移动双臂机器人具身 / 工业机器人官方开发者门户伙伴列表D-Robotics 合作伙伴 API
AI GuidedGUIDI AI 路径引导可穿戴设备可穿戴设备官方开发者门户伙伴列表D-Robotics 合作伙伴 API
沛远智能(Peiyuan Intelligent)Caddy Trek 高尔夫球童机器人消费级机器人官方开发者门户伙伴列表D-Robotics 合作伙伴 API
复睿智行(Fu Rui Zhixing)“Super sensor” 智能配件智能配件官方开发者门户伙伴列表D-Robotics 合作伙伴 API
FITURE数字健身 / 智能运动测试解决方案智能健身官方开发者门户伙伴列表D-Robotics 合作伙伴 API

数据直接来自 D-Robotics 自有开发者门户合作伙伴 API,于运行日期抓取;产品和品类标签译自中文 API 载荷,反映 D-Robotics 自有 DGP 目录,并非经过独立审计的出货或收入数据。

[CU023, CU036]
FU001: 客户旅程地图

D-Robotics 的公开客户旅程从 DGP 发现和 RDK 硬件试用开始,进入联合开发、具名发布、社区传播,极少走到已披露的量产或复购。

这是基于公开博客文章、媒体报道和开发者门户数据重建的商业化路径,不是厂商发布的 CRM 漏斗。

[CU001, CU007, CU020, CU022, CU024, CU027]

6.2 Hengbot、Narwal、Insta360 与 Tiangong 3.0 的实名客户证据和采用轨迹

最强的公开客户证据来自两个联名产品发布。D-Robotics 自己的博客点名 Hengbot 的 Sirius 机器狗运行一颗最高 5 TOPS 的 RDK X3 AI 头部模组,这一规格也能在 D-Robotics 的 RDK X3 产品页独立匹配;Hengbot 自己的主页显示,截至运行日,该活动已获得 1,150 名支持者和 $912,768 众筹金额——不过原始活动的两个公开索引 Kickstarter URL 都返回 HTTP 404,因此众筹一手记录本身无法独立恢复。Elephant Robotics 的 myCobot 280 RDK X5 是第二个围绕 D-Robotics RDK X5 开发板打造并商业销售的实名产品。沿证据强度再往下一步,KrASIA 转载的 36Kr 报道称,D-Robotics 已与 60 多家供应链玩家合作,具体点名 Narwal 的 Xiaoyao 002 扫地机器人、Insta360 的 Antigravity A1 全景无人机和一款 Vbot 四足陪伴机器人。Narwal 和 Insta360 各自在官网独立确认了真实且庞大的商业足迹(分别为 450 million 名用户和全球产品线),但两个官网本身都没有点名 D-Robotics,因此两者的芯片归因目前依赖第三方媒体,而不是买方自己的材料。Vbot 提及更弱:它只是单一媒体引用,未找到佐证来源或产品页。2026 年最重要的进展是 Tiangong 3.0,这是一款全尺寸人形机器人,由 D-Robotics 与 Beijing Innovation Center of Humanoid Robotics 围绕 Xuri/Sunrise S600 芯片共同工程开发;Gasgoo 和独立媒体 en.shuziqushi.com(引用 Jiemian News)都确认,量产和交付计划在 2026 年下半年进行,成品机器大规模量产后成本预计下降超过 50%。合在一起,采用轨迹证据显示的是:两个成熟联名消费产品、两个仅由媒体确认的 OEM 关系、一个单一来源且证据薄弱的关系,以及一个尚未产生收入但证据较充分的人形机器人设计导入。投资人不应把这组梯度压缩成一个“具名客户”标题数字。[CU007, CU008, CU009, CU010, CU011, CU012]

客户增长 / 采用轨迹表
指标日期来源置信度含义缺失分母
出货量同比增长~180%FY2025(2026 年报道)KrASIA/36Kr;Yicai Global 等来源确认出货量从低基数快速放大未披露绝对出货台数
客户基数同比增长~200%FY2025(2026 年报道)KrASIA/36Kr;Yicai Global 等来源说明买方名单在扩张,不只是单个客户加深采购未披露绝对客户数和「客户」定义
DGP 支持的早期公司200+截至 DGP 博文发布(2026 年)D-Robotics 官方博客显示加速器漏斗较大没有全部 200+ 家公司的具名名单
基于平台打造的机器人产品品类100+截至 DGP 博文发布(2026 年)D-Robotics 官方博客;Yicai Global(100+ 款机器人产品)覆盖多类机器人平台,适用性广“品类”口径与逐个点名的出货 SKU 未核对一致
开发者社区规模100,000+2026Yicai Global;D-Robotics myCobot 博客漏斗顶部开发者社区规模大活跃开发者与仅注册开发者未区分
Hengbot Sirius 众筹$912,768 筹资 / 1,150 名支持者截至运行日期(2026-07-04)Hengbot 官方首页证明消费者愿意为 D-Robotics 驱动的产品付费支持者到已发货单元的转化未披露
Tiangong 3.0 量产时间H2 2026(计划)2026 年 6 月报道Gasgoo;en.shuziqushi.com 等来源首个点名且与 D-Robotics 芯片绑定的人形机器人量产承诺实际单元销量和交付日期尚未公开
DGP 媒体伙伴网络100+ 家媒体伙伴截至 DGP 博客发布时(2026)D-Robotics 官方博客营销触达,不是商业验证未提供点名伙伴名单

百分比和数量来自 D-Robotics 自报,或来自已点名新闻媒体的报道;均未经过独立审计,因此“置信度”反映交叉验证的广度,而非确定性。

[CU001, CU002, CU009, CU020, CU021, CU030]
点名客户验证表
客户 / 伙伴细分市场部署 / 用例量产 / 试点结果限制
Hengbot(Sirius 机器狗)消费级陪伴机器人面向 1kg 可编程机器狗的 RDK X3 驱动 AI 头部量产 / 出货:Kickstarter 已众筹,目标 2025 年秋季全面上市,官网显示支持者资金仍在实时累计1,150 名支持者,筹资 $912,768;CEO 引述称机器人动作逼真原 Kickstarter 活动 URL 在运行日期返回 HTTP 404;未披露单元发货量
Elephant Robotics(myCobot 280 RDK X5)机械臂教育 / 机械臂硬件配备 RDK X5 控制板的 6-DOF 机械臂,用于 AI 视觉和 LLM 开发量产:产品线已商业销售;公司业务覆盖 51 个国家D-Robotics 博客将其定位为旗舰级联名教育和开发者产品未披露 RDK X5 版本的单元销量或收入
Narwal(Xiaoyao 002 扫地机器人)消费级家用清洁机器人D-Robotics 提供 AI 驱动的双目感知,用于避障量产:Narwal 是已独立规模化的品牌,其官网称覆盖 30+ 个国家、拥有 450M+ 用户为商业成熟的扫地机器人产品线加入 AI 感知能力Narwal 官网没有点名 D-Robotics;归因来自 36Kr/KrASIA 报道
Antigravity / Insta360 支持团队(Antigravity A1 无人机)消费级无人机 / 影像Sunrise 5 芯片驱动 360 度全景无人机的 AI 视觉感知和避障预发布 / 早期量产:计划 2025 年 Q4 试销,目标 2026 年 1 月全球发布据 China Biz Insider,这是 D-Robotics 在消费级无人机的首个商业订单Insta360 官方产品页未提及 D-Robotics 或该芯片
Vbot(未命名四足陪伴机器人)消费级 / 工业四足机器人D-Robotics 芯片集成进四足陪伴机器人不明确:单篇媒体提及中仅描述为“working with”在头部联名发布之外,拓宽品类覆盖单一来源说法(36Kr/KrASIA);未找到独立佐证或产品页
Beijing Innovation Center of Humanoid Robotics(Tiangong 3.0)项目人形机器人 / 具身智能;工业-商业Xuri/Sunrise S600 芯片驱动全尺寸通用人形机器人预量产:计划 H2 2026 量产并交付en.shuziqushi.com 称,规模化后整机成本预计下降超过 50%截至运行日期尚未量产;无已交付单元数量
DGP 合作伙伴群体(Leju Robotics、AI Guided、Caddy Trek、FITURE 等)教育、可穿戴、消费级和健身机器人混合D-Robotics 自有开发者门户伙伴目录列出 9 家公司混合:该目录未披露各伙伴是量产还是试点证明头部联名发布之外的生态广度目录未提供各公司的出货量、收入或合作期限细节

各行按公开证据强度排序,从最强且独立佐证最多的 Hengbot、Elephant Robotics,到最弱的 Vbot;量产 / 试点状态根据所引来源推断,并非来自 D-Robotics 披露的客户登记表。

[CU007, CU011, CU013, CU016, CU017, CU019]
FU002: 采用 / 部署漏斗

公开证据从宽泛生态说法很快收窄到少数经独立佐证的具名部署,披露留存数据则为零。

这个漏斗衡量的是证据质量和独立佐证,不是 D-Robotics 的内部销售管线;计数来自本章引用的来源。

[CU001, CU003, CU017, CU020, CU033]
FU003: 客户证明矩阵

Hengbot 和 Elephant Robotics 证据最强:D-Robotics 自有博客、产品页与合作方主页能对齐;Vbot 最弱,只靠一次媒体提及。

矩阵单元格是截至报告生成日对公开证据质量的定性评级;评分衡量证据可见度,而不是每段关系的潜在商业价值。

[CU007, CU013, CU014, CU017, CU018, CU019]

6.3 扩张飞轮:DGP 项目、开发者社区与布道者网络

D-Robotics 运营着一套刻意设计的漏斗顶部扩张机制,独立于具名 OEM 订单。Robotics Dream Keeper Challenge 提供四阶段晋级(Explorer、Builder、Creator、Core Developer),包括导师指导、$1,000 大奖和购买 RDK X5 的现金返利,明确要把爱好者开发者转化为重复购买硬件的用户。公司还举办了与 Make: Magazine 编辑 David Groom 联合推广的线上“RDK X5 Show & Tell”活动,并送出一台 Elephant Robotics myCobot 280 机械臂来拉动参与;其开发者门户 API 显示 2026 年项目仍在活跃,包括一项 Smart Healthcare 挑战赛道,嵌入全国大学生智能汽车竞赛,报名开放到 2026 年 8 月。D-Robotics 社区页点名了五位代表品牌参与活动和线上传播的“机器人布道者”,其中包括 Frank Fu,被描述为 Navbot CEO。不过,这段关系也说明采用质量上的一个重要差别:NavBot 自己的主页把其四足和 STEM 机器人产品营销为“NVIDIA Jetson Powered”,公开产品信息从未提到 D-Robotics 或 RDK 硬件。因此,Fu/Navbot 联系更像社区触达或布道者关系,而不是 Navbot 在自有产品中出货 D-Robotics 芯片的确认证据。D-Robotics 自己门户上的开发者案例展示,在更大范围内强化了同样的谨慎判断:列出的几十个项目绝大多数是面向竞赛或个人学习的学生和爱好者投稿,而不是付费商业部署。因此,这个扩张飞轮是真实且活跃的社区建设机制,但它是未来商业关系的管线,不能替代具名、带收入客户证据。[CU024, CU025, CU026, CU027, CU028, CU029]

6.4 留存不透明、集中度风险与负面信号

本次审阅未发现任何官方或独立 2026 年来源披露 D-Robotics 任一客户或伙伴细分的净收入留存、毛留存、续约率或流失率,也没有任何来源披露客户集中度指标,例如最大具名账户占出货量或收入的比例。这种不透明很重要,因为 D-Robotics 自己的增长标题——出货量约增长 180%、客户数增长 200%,并由 KrASIA/36Kr 和 Yicai Global 交叉佐证——并不能说明增长是在约 60 个具名或报道过的供应链伙伴中广泛发生,还是集中在扫地机器人等少数高出货量 OEM 品类中。独立行业评论提供了本章最清晰的负面信号:China Biz Insider 对中国具身 AI 独角兽群体的分析把 D-Robotics 描述为“从芯片层进攻问题”,这是一种无论哪个应用层人形机器人赢家出现都能捕获价值的芯片级策略;但同一分析也把大多数群体公司——包括 D-Robotics——置于 18–24 个月现金续航上,并预计行业将在 2027 至 2028 年之间面临清算,同时指出 D-Robotics 自己的 2026 年 4 月融资轮估值约为 RMB 10.8 billion,却没有披露客户收入集中度。额外集中度风险来自地理:本章讨论的几乎每一个具名客户或伙伴——Hengbot、Narwal、Vbot 和 Beijing Innovation Center——都位于中国具身 AI 生态内,因此即便 OEM 品类组合保持多元,国内政策或出口管制变化也可能对客户基础产生不成比例的影响。在 D-Robotics 或具名伙伴披露留存、续约或集中度数字之前,已报告增长率的耐久性——以及 200,000 名开发者漏斗顶部与少数可独立佐证商业部署之间的落差——仍是本章最核心的未解尽调问题。[CU031, CU032, CU033, CU034, CU035, CU039]

留存 / 重复使用 / 满意度表
指标数值 / null细分市场置信度尽调问题
净收入留存率(NRR)null全部细分市场n/a(未披露)向 D-Robotics 索取 OEM / DGP 细分市场的队列级收入留存
毛收入留存率(GRR)null全部细分市场n/a(未披露)索取按客户统计、剔除流失影响的留存数据
续约 / 复购率nullOEM 芯片客户(Narwal、Hengbot 及类似客户)n/a(未披露)向 D-Robotics 或已点名 OEM 索取复购节奏和合同续约数据
众筹支持者到已发货单元转化率nullHengbot Siriusn/a(未披露)向 Hengbot 索取众筹后履约数据
DGP 成员毕业后转化为付费客户的比例nullDGP 加速器队列n/a(未披露)向 D-Robotics 索取 DGP 队列转化跟踪
开发者到付费客户转化率null100,000+ 开发者社区n/a(未披露)索取付费 RDK 单元销量占注册开发者的比例
点名客户满意度 / NPSnull全部点名伙伴n/a(未披露)索取客户满意度调研或开放客户访谈

所有指标均为 null,因为审阅的官方或独立 2026 年来源均未披露任何 D-Robotics 客户细分市场的留存、续约或满意度数据;null 表示缺乏证据,不是报告值为零。

[CU033, CU039]
扩张与集中度风险表
扩张驱动因素集中度风险影响尽调路径
DGP 加速器漏斗(声称 200+ 家初创公司)漏斗依赖 D-Robotics 自有硬件折扣和员工投入;转化为付费客户尚未验证若没有匹配收入,可能夸大外界看到的管线广度索取 DGP 到出货客户的转化指标
消费级 OEM 多元化(扫地机、割草机、陪伴机器人、无人机)CEO 称出货量仍依赖少数大众市场品类,而非具身智能若任一 OEM 品类(如扫地机器人)放缓,芯片出货增长可能停住索取按品类拆分的出货结构
人形机器人 / 具身智能可选性(Tiangong 3.0)截至运行日期,处于收入前阶段,只有一个点名人形机器人设计订单战略上行空间大,但已交付单元证据为零;若 H2 2026 目标延后,时间线风险上升跟踪 Tiangong 3.0 交付里程碑与 H2 2026 目标的偏差
全球开发者社区(100,000+)社区规模是领先指标,不是集中度对冲,因为展示的开发者案例大多是未付费的学术或创客项目存在用社区指标夸大商业广度的风险索取社区成员中付费 RDK 客户占比
中国地域和监管集中度几乎所有点名客户和伙伴(Hengbot 的制造联系、Narwal、Vbot、Beijing Innovation Center)都在中国具身 AI 生态内出口管制或国内政策变化可能对客户基础造成不成比例影响索取客户地理收入结构和出口敞口披露
单一来源伙伴说法(Vbot)至少一个点名伙伴(Vbot)只有单篇媒体提及,没有独立佐证若纳入无法核验的合作关系,会夸大真实点名客户数量为 Vbot 寻找第二个独立来源或直接公司确认

集中度风险基于本章其他部分使用的同一批公开来源推断;D-Robotics 自身没有披露集中度百分比。

[CU005, CU019, CU029, CU031, CU032, CU035]

6.5 展示要点

Chapter 07

07风险

7.1 技术差异化与竞争替代风险

核心技术风险在于,D-Robotics 想在机器人计算市场建立标准,但这个市场里,最显眼的既有巨头已经同时覆盖芯片和软件控制平面。NVIDIA 的 Jetson 产品梯队已经从 67 TOPS 的 Orin Nano、275 TOPS 的 AGX Orin,延伸到配 128 GB 内存、2070 FP4 TFLOPS 的 Thor;JetPack 7 又把 CUDA、TensorRT、Isaac ROS、OTA、安全和云原生部署工具打包在一起。机器人团队拿到的是一套全球更容易识别、也更容易解释的软件界面;D-Robotics 目前公开的基准测试级证据,还没有打出同等清晰度。 D-Robotics 自己当然有真实技术栈:X3、X5、S100、NodeHub 和 DGP 拼出了一条完整开发路径。但公开证据读起来仍更像生态驱动的差异化,而不是全球性能经济性更优的硬证明。Qualcomm 仍在投入 RB3 Gen 2 和 RB5 机器人路径,Ambarella 用每瓦性能叙事切入低功耗机器人感知,Kneron 的资金也足以继续推进边缘 AI 叙事。风险不是 D-Robotics 没有产品;风险在于,在 D-Robotics 拿出更清晰的基准测试、软件和部署优势之前,本土市场之外的买家可能仍会默认选择全球更标准化的技术栈。[CR002, CR003, CR004, CR005, CR006, CR007]

运营 / 质量 / 安全风险登记表
故障模式公开证据可能性严重性缓释成熟度剩余敞口
与 Jetson 级既有厂商的基准差距公开产品页缺少基准级跨供应商验证中低
板卡和工具链现场支持摩擦论坛显示 CAN FD、GPIO、丢帧、MCU 和转换问题中高中高
SDK / 软件栈追赶负担JetPack 展示出更大的量产软件栈中低
资本密集型生态维护文档、论坛、套件、模型库和伙伴支持都要一起扩张
整个人形机器人 / 具身 AI 商业化放缓或泡沫修正TechCrunch、CB Insights 来源报道和中国独角兽队列报道都提示:同业初创公司人形机器人资本过热,灵巧度 / 安全性受到质疑,现金跑道只有 18-24 个月中高

运营风险结合了直接公开问题证据,以及基于官方产品和软件界面复杂度作出的推断。

[CR004, CR005, CR006, CR022, CR023, CR031]
FR001: 风险热力图

基于现有公开证据,按发生可能性、影响、缓释成熟度和剩余敞口梳理 D-Robotics 的主要风险。

高 / 中 / 低分组是作者基于留存公开证据的综合判断,不是精确统计概率。

[CR004, CR005, CR006, CR011, CR017, CR021]

7.2 地缘政治、出口管制和供应链风险

地缘政治是这份材料里最清晰的非公司特定风险之一。BIS 2026 年 5 月 31 日指引明确表示,先进计算物项流向总部位于 Country Group D:5 或 Macau 的实体时,即便该实体位于这些地区之外,仍需要许可证。WilmerHale 2025 年分析补充说,AI Diffusion Rule 可能已被暂停,但围绕中国相关 AI 出口、支持活动和转运筛查的执法姿态变得更敏感,而不是更宽松。另一份 WilmerHale 备忘录显示,晶圆厂和 OSAT 封测厂现在面临更重的尽调和审批负担,且这些负担会在 2026 年进一步加硬。 对 D-Robotics 来说,公开资料并未证明发生过直接制裁事件。风险更隐蔽,也因此更重要:如果与中国相关的机器人和边缘 AI 项目更难获得服务,全球供应商即便没有看到直接点名公司的禁令,也可能放慢工具、芯片、封装或支持交付。现有证据也没有足够清楚披露 D-Robotics 的代工或封装链,无法量化台湾或其他外部卡点敞口。因此,任何长期承保判断都要保留一层有分量的地缘政治压力。 管制体系还在叠加,并非静止不动。BIS 2025 年 3 月将 80 家实体加入 Entity List,理由是这些实体参与与中国军工复合体相关的先进 AI、超级计算和高性能芯片活动;此前 2024 年 12 月的一揽子规则中,Torres Trade Law 称已新增 140 家实体和 16 项 Footnote-5 指定,并引入新的 Foreign Direct Product 规则和许可证例外。Congressional Research Service 2025 年 9 月更新把这放在美国多年努力的一部分:拖慢中国提出的 2030 半导体自给目标。BIS 自己的 2024 年 12 月晶圆代工规则也要求封装厂在出货某些先进芯片前,核验晶体管数量,或持有 Approved/Authorized IC-designer 或 OSAT 身份。TrendForce 数据另行显示,美国对允许进入中国的芯片有意维持一到两代(「N-1」/「N-2」)性能差距;与此同时,晶圆产能和 HBM 瓶颈仍在约束中国 AI 芯片厂商,即便其国内市场份额预计 2026 年接近 50%。这些来源都没有直接点名 D-Robotics,但每一轮新规则都会扩大需要额外筛查的交易对手、封装伙伴和工具供应商范围;这种累积效应,比任何单一规则更构成长期风险。[CR012, CR013, CR014, CR015, CR016, CR017]

监管 / 法律风险登记表
风险司法辖区 / 规则状态可能性严重性缓释措施剩余敞口尽调路径
先进计算出口许可证收紧美国 EAR / BIS D:5 指引截至 2026-05-31 有效保持零部件 / 工具映射更新;分散供应商获取晶圆代工 / 封装 / EDA 依赖地图
以支持或转运理论执行 AI 出口管制美国 BIS 指引 / 法律解释执法姿态升高中高交易尽调和经销商筛查审查客户地域、最终用途控制和服务依赖
晶圆制造商 / OSAT 合规负担美国 BIS 晶圆制造商规则已生效,2026 年门槛更紧尽可能使用获批或低风险伙伴中高识别哪些制造伙伴面临 3A090 相关义务
美国实体清单持续加码,瞄准中国 AI / 先进计算活动美国 BIS 实体清单(2024 年 12 月 + 2025 年 3 月轮次)有效且扩围根据实体清单更新筛查投资者、客户和供应商确认 D-Robotics 已知投资者、经销商或组件供应商均未出现在当前或待定实体清单轮次中
晶圆代工 / OSAT 转运尽调和晶体管数量验证美国 BIS 晶圆代工规则(2024 年 12 月生效,2026 年收紧)已生效,合规负担上升中高封装交由获批 / 授权 IC 设计方或 OSAT 伙伴处理中高识别哪些制造和封装伙伴在 BIS 晶圆代工规则下持有获批 / 授权状态

各行概括了可能约束供应、服务或供应商意愿的公开法律和监管动态,即使没有针对 D-Robotics 的特定行动也会产生影响。

[CR012, CR013, CR014, CR015, CR016, CR017]
FR002: 风险传导图

展示地缘政治、产品和支持风险如何传导为采用放慢、成本上升和融资结果转弱。

[CR011, CR017, CR021, CR023, CR030, CR031]

7.3 具身 AI 商业化疑虑、泡沫风险和中国产业政策敞口

中国自己的机器人生态正被两股力量同时拉扯。IFR 2026 年 5 月关于中国第十五个五年规划的简报称,机器人被放在国家产业战略核心;中国已拥有约 2 million 台工业机器人装机基础,约为日本存量的 4.5 倍,并且按 World Robotics 2025 Report 口径,占 2025 年全球工业机器人安装量的 54%。这对国内机器人计算供应商是一个可信顺风,但也有反作用:国家主导的产业政策往往会把支持和需求牵引集中到官方认定的「国家队」平台;公开记录尚未显示 D-Robotics 是这轮推进里的优先受益者,还是只是众多争夺同一政策相关需求的国内供应商之一。 叠在政策顺风之上的,是公司无法控制的需求时点风险:人形机器人和具身 AI 商业化疑虑。TechCrunch 转述机器人学家 Rodney Brooks 的批评,认为当前人形机器人路径建立在关于灵巧度和安全物理约束的「纯粹幻想式思维」之上,并预测真正的人形机器人实用性到来时间会比今天资本市场假设得更久。另据 Robotics & Automation News 报道的 CB Insights 数据,人形机器人在最近一个季度已经拿下所有 AI 垂直赛道中单一类别最高的 VC 交易量,同时投资者也明确警告泡沫,并呼吁坚持「收入优先」纪律。ChinaBizInsider 2026 年中统计至少 25 家中国具身 AI 独角兽,其中 15 家仅在 2026 年上半年诞生,合计吸收 RMB46 billion (~US$6.39 billion) 已披露融资;每家公司现金跑道只有 18-24 个月。这意味着 2027-2028 年很可能出现一波整合、降价轮或关停。 这直接关系到 D-Robotics,因为它的 RDK S100 / BPU Nash 产品线明确面向具身和人形机器人客户。如果人形机器人投资周期在商业化放量前降温,D-Robotics 可触达的近期需求可能比其更宽的边缘 AI 和 AMR 业务所暗示的收缩得更快;公开记录也没有披露当前出货或在手订单中,有多少绑定人形机器人需求,多少来自更耐久的工业、AMR 或行车记录仪需求。另一个独立问题是,未发现任何公开标准机构或监管方在机器人安全、互操作性或认证语境中点名 D-Robotics;这留下的是一个待尽调事项,而不是已确认的合规缺口。[CR047, CR048, CR049, CR050, CR051, CR052]

7.4 市场集中、渠道依赖和支持风险

市场扩张风险的核心,不是 D-Robotics 能否在中国之外卖出产品,而是它当前拥有怎样的国际足迹。经销商页面令人鼓舞,因为它显示公司在多个亚洲和欧洲国家有可见转售覆盖。但同一页面也是一个警示:足迹仍主要由经销商驱动,配件页面也把外设履约导向本地分销商。这是非常正常的早期渠道设计,但弱于直接现场团队、具名全球 OEM 部署和已发布的国际客户背书。 支持面也强化了这一谨慎判断。2026 年 7 月初的公开论坛活动显示,围绕 CAN FD、GPIO、掉帧、模型转换和 S100 启动行为存在真实问题。这证明产品确实有人在用,但也意味着运营复杂度已经公开显现。再叠加公开上手界面浓厚的教育和开发者导向,结论是一家公司拥有健康的生态活力,但国际变现深度仍不确定。投资者因此应把可见社区参与和已证明的海外收入多元化分开看。[CR019, CR020, CR021, CR022, CR023, CR024]

伙伴 / 依赖风险登记表
依赖交易对手 / 类别角色集中度信号失败场景严重性缓释措施剩余敞口
全球经销商网络第三方经销商国际履约和触达足迹可见,但由经销商主导海外转化弱或库存可得性差中高增加直接 OEM 客户背书和本地支持中高
芯片 / 工具既有厂商NVIDIA / Qualcomm / Ambarella 生态吸力标准竞争目标品类中较高开发者默认选更知名技术栈证明工作负载层面的成本或延迟优势
母公司光环Horizon Robotics 品牌 / 血统信任与技术传承认知有意义但未量化分拆后未能补上继承而来的信誉中高拿出独立客户战绩和经济性中高
开发者社区论坛 / GitHub / NodeHub 用户采用漏斗与反馈闭环覆盖广,但变现能见度低使用未转成可持续收入发布可复用客户案例经济性

主要依赖风险不在单一供应商,而在几件事叠加:经销商拉动扩张、大型对手的生态吸力,以及社区活跃能否转成收入仍不确定。

[CR007, CR008, CR009, CR010, CR019, CR020]
FR003: 依赖关系图

梳理 D-Robotics 若要把开发者牵引力转成可持续机器人收入,必须处理的外部生态和关系。

[CR019, CR020, CR021, CR024, CR025, CR026]

7.5 分拆执行和财务模型风险

执行风险仍然偏高,因为 D-Robotics 不是一家从零起步的软件公司;它是近期分拆出来的公司,要同时商业化芯片、板卡、中间件、工具、社区支持和伙伴赋能。母公司的规模对比会让人冷静。Horizon Robotics 现在披露数十亿 RMB 收入、数百万终端用户和数千万芯片出货,而 D-Robotics 仍在证明这套聚焦机器人的技术栈能独立撑起经济性。这不等于存在隐藏依赖,但确实意味着管理层必须把继承来的品牌信任,替换成自己的可复制执行能力。 财务上,公开记录证明了融资动能,但没有证明自我造血的经济性。Yicai 和 CXO 都报道了 2026 年分两步完成的 $270 million Series B,但同一批公开记录仍未披露收入、毛利率、现金、烧钱速度或跑道。这个平台在结构上资本开支很重,因为产品开发、生态赋能和支持要一起扩张。在这样竞争激烈的市场里,即便资金充足,公司也可能比投资者预期更快再次依赖融资。管理层拿出干净的变现、成本和跑道披露之前,财务模型应被视为风险放大器,而不是稳定器。[CR025, CR026, CR027, CR028, CR029, CR030]

人员 / 执行风险清单
角色 / 职能依赖或缺口可能性严重性缓释动作尽调路径
分拆后领导层执行需要用独立战绩和治理证据替代母公司光环中高用新增资金和清晰的产品迭代节奏支撑要求组织架构图、GTM 归属和决策权地图
产品 / 支持组织板卡、SDK、文档和论坛都要持续维护社区反馈闭环可见要求支持团队人数、SLA 指标和缺陷趋势
国际销售打法公开足迹更多由伙伴带动,而不是企业直销驱动中高中高用分销商播种市场要求地区拆分、本地支持计划和 OEM 管线
财务 / 规划纪律未公开烧钱速度、现金续航或利润率近期融资降低短期压力要求月度现金消耗、毛利率桥接和下一轮融资触发条件

执行风险偏高:公司在公开运营披露有限的情况下,同时扩产品、支持和销售。

[CR024, CR025, CR026, CR027, CR028, CR029]
缓释与否决条件表
风险可监控触发点阈值 / 事件行动含义
出口管制卡点新 BIS 规则 / 供应商通知 / 发货受阻关键工具、部件或服务新增许可要求,或许可申请被拒立即重做供应链测算,并收紧估值假设
国际变现失败具名海外 OEM 定点或收入拆分到下一轮融资周期仍无可信的中国以外量产客户证据将全球扩张论点视为已破裂
支持质量恶化论坛问题反复出现,缺陷类别未解决同类板卡 / 工具链故障跨多个版本持续存在假设支持成本更高、转化更慢
资本依赖回潮商业化证明出现前就融资收入 / 利润率透明度改善前就需要新一轮融资下调投资测算倍数,并要求更强条款
具身 AI 泡沫退潮中国具身 AI 独角兽群体出现一波下调估值融资、关停或被迫并购在 12 个月窗口内,跟踪的约 25 家 RMB10bn+ 具身 AI 创业公司中,至少两家失败、合并或下调估值融资重新评估 D-Robotics 面向人形机器人的 S100 / BPU Nash 管线需求假设,并相应下调增长情景

这些是投资人监控触发点,不是公司承诺;每一个都标志着一个有吸引力的平台故事开始变成明显更弱的投资测算案例。

[CR017, CR021, CR023, CR028, CR030, CR032]

7.6 图表

Chapter 08

08估值

8.1 投资论点与反论点

D-Robotics 的正面估值逻辑不难理解。公开来源显示,它不是一次性的开发板项目,而是真实的机器人平台:公司销售计算套件、开发者工具和生态资产,同时踩中边缘 AI 与具身机器人两条强叙事。2026 年 B1 和 B2 融资也证明,投资者愿意用有分量的资本支持这条叙事。如果 D-Robotics 能把平台宽度转化为 AMR 或人形机器人相邻场景里的耐久软硬件标准,今天的私募估值回头看仍可能显得保守。 反论点是,公开证据现在仍更像一个生态故事,而不是一台商业机器。已审阅资料没有披露收入、毛利率、烧钱速度或客户集中度。Ambarella 和 Horizon 这些有真实估值的公开可比公司,也有 D-Robotics 目前拿不出来的申报级透明度。与此同时,Yicai 关于 2026 资本周期的报道警告,中国机器人和硬科技估值可能已经偏热。实际含义是,D-Robotics 相比普通硬件创业公司应享有真实平台溢价,但在已披露经济性追上叙事之前,这个溢价应被封顶。[CV001, CV002, CV003, CV004, CV014, CV015]

投资论点 / 反论点表
维度多头论点空头反论点
平台范围芯片 + 套件 + 软件有机会成为机器人标准公开证据更偏生态,经济性证据不足
资本$270M B 轮体现投资人真实信心资本能掩盖不透明的单位经济性,不能解决它
市场时点具身机器人和边缘 AI 都是热门品类热门品类也会带来泡沫定价和拥挤竞争
可比公司如果披露成熟,公开边缘 AI 可比公司显示大上行空间同一批可比公司拥有 D-Robotics 缺少的公开透明度
中国位置国内机器人生态可加快迭代地缘政治折价和出口摩擦会压缩价值
国际扩张经销商覆盖可播种全球需求经销商主导的覆盖弱于直接 OEM 证明

核心争论不在 D-Robotics 是否有意思,而在当前信息是否足以支持在公开经济性出现前支付溢价。

[CV001, CV004, CV014, CV015, CV019, CV020]
FV001: 投资建议逻辑

展示平台宽度、披露缺口、资本周期风险和公开可比对象背景如何共同导向跟踪 / 公允判断。

流程为定性梳理,反映本章决策逻辑,不是数学评分模型。

[CV001, CV015, CV017, CV020, CV021, CV022]

8.2 建议、置信度和出价纪律

我们将 D-Robotics 评级为「跟踪」,置信度中等,估值立场为「合理」。这不是「公司不好、回避」的结论。公司拥有足够的产品宽度、融资能力和机器人赛道窗口,值得留在主动观察名单上。但它也还不是一个清晰的「买入」,因为价格无法和公开单位经济性挂钩。在半导体加系统业务里,这比纯软件更重要,因为产品、支持和生态投入会一起扩张。 因此,出价纪律应按里程碑推进。只有在 D-Robotics 证明有国际说服力的客户采用、软件变现以及围绕烧钱和毛利率的更干净披露之后,投资者才应愿意支付更高价格。在此之前,「合理」是最平衡的立场:说便宜过于慷慨,因为平台范围和已募集资本已经支撑了有意义溢价;说过高又过于严厉,因为公司相较未来可能追赶的公开平台,规模仍明显更小、证明也更少。[CV016, CV017, CV018, CV021, CV022, CV033]

投资建议摘要表
维度判断依据
建议跟踪观察战略上行真实存在;定价精度不足
置信度公开可比公司有参考价值,但私募估值未直接披露
风险评级技术、地缘政治、执行和不透明性叠加
估值立场合理溢价有道理,但领军者溢价没有
综合评分6.2 / 10平台有潜力,投资测算基础不完整
入场纪律等待里程碑证明只有客户和财务证据改善后才支付溢价

建议基于里程碑,因为公开证据无法支撑传统收入倍数投资测算模型。

[CV021, CV022, CV032, CV033, CV040]
FV004: 投资 KPI

仅基于公开证据提炼 D-Robotics 的核心可投性信号。

KPI 概括可投性,而非经营表现;多项关键财务数据和投资条款仍未公开披露。

[CV001, CV015, CV021, CV022, CV040]

8.3 融资背景、进入纪律和融资条款风险

最站得住脚的融资结论很清楚,但并不完整。D-Robotics 在 2026 年募集了足够可见的资本,近期不会显得资金不足;单凭这一点,它已经区别于许多更早期的私有边缘 AI 供应商。与此同时,公开记录仍未揭示这些轮次的确切投后估值、优先权结构或下行保护。也就是说,投资者被要求对头条式动能作出反应,却缺少判断普通股进入经济性是否真正有吸引力所需的保护。 公开可比公司在这里主要是纪律工具。Ambarella 和 Horizon 证明,投资者最终会奖励边缘 AI 平台,但前提是这些平台接受申报级披露和公开市场审视。D-Robotics 还没到这一步。因此,进入纪律应聚焦轮次结构、烧钱与跑道的关系,以及下一轮融资更可能来自证明,还是来自希望。没有这些答案,合理估值区间比任何单点估值更有用。[CV001, CV002, CV008, CV015, CV017, CV018]

8.4 乐观、基准和悲观情景

D-Robotics 仍然披露很轻,因此情景分析比点估值更好用。悲观情景下,公司仍是有前景的技术平台,但在收入质量或出口链韧性令人信服之前就必须再次融资;这指向更接近低于 $1.1B 的结果。基准情景下,D-Robotics 持续把生态牵引转化为进展,保持融资渠道,并逐步改善变现证明,但仍没有完全摆脱不透明;这支撑约 $1.2-1.8B 区间。乐观情景下,公司证明中国之外的量产客户,展现真实软件或模块变现杠杆,并降低地缘政治折价;届时 $2.2-3.0B 区间会更可信。 关键不是精确到小数点,而是路径依赖。今天的价值更多由未来里程碑驱动,而不是当前财务产出。因此,「合理」立场可以和有意义的上行空间并存,前提是投资者对下一个估值档位必须满足什么条件保持纪律。[CV023, CV024, CV025, CV029, CV030, CV031]

牛 / 基准 / 熊情景表
情景概率关键假设商业里程碑隐含价值
~30%在经济性可见前还需要更多资本;出口或渠道摩擦恶化没有清晰的国际量产证明$0.8-1.1B
基准~45%生态牵引逐步转化;不透明性只部分改善选择性设计定点,以及更清晰的变现$1.2-1.8B
~25%中国以外 OEM 定点、软件附加率和供应链信心都改善平台不只停留在开发套件,开始可复制$2.2-3.0B
当前立场n/a最佳公开证据区间,而不是精确估值重新评级前等待更硬证据合理

区间是作者估计,锚定融资证据、公开可比公司和风险折价,而不是已披露收入倍数。

[CV023, CV024, CV025, CV033]
FV002: 估值敏感性

以十亿美元计的示意性锚点,覆盖 D-Robotics 和部分可比参照。

D-Robotics 柱形为作者估算;Ambarella 和 Kneron 柱形是留存来源里的公开参照点,其中 Kneron 用已披露累计融资表示,而不是已披露当前估值。

[CV007, CV012, CV023, CV024, CV025]
FV003: 估值 / 回报区间

按情景给出 D-Robotics 估值区间,以十亿美元计。

D-Robotics 区间为作者估算,依据公开可比对象、融资证据和风险折价,而不是已披露当前估值。

[CV007, CV023, CV024, CV025, CV033]

8.5 可比公司组和估值背景

可比公司组提示应谨慎,但不支持直接否定。Ambarella 是功能上最干净的公开可比公司,因为它是一家已披露的边缘 AI 芯片公司,目前估值约 $3.43B。Horizon 是更具战略意义的上限锚点,因为它是已上市的母平台,规模大得多,当前市值为 HK$68.2B。Cambricon 是可比组里的警示标签:当前公开市场数据把它推到极高水平,因此它能说明 AI 芯片市场情绪,却不能作为保守估值模板。Kneron 是更干净的私有可比点,因为它披露融资历史并经营全栈边缘 AI,但从可见已融资本看,它仍比 D-Robotics 更早期、规模更小。 合在一起,可比公司组说明 D-Robotics 应高于简单早期硬件叙事,但低于一家充分披露的公开边缘 AI 平台。这正是「合理」成为正确中间地带的原因。公司比只有炒作的创业公司更有实质,但公开证明还不足以支撑龙头倍数。[CV005, CV006, CV007, CV008, CV009, CV010]

可比估值表
可比对象类型估值 / 状态证明强度适用性局限
D-Robotics标的公司近期 B 轮融资 $270M;确切投后估值未验证产品与融资信号强,公开经济性弱直接投资测算对象当前精确估值未获公开确认
Kneron非上市边缘 AI 可比公司$97M B 轮;总融资 >$190M非上市融资历史已披露最接近的非上市全栈边缘 AI 类比融资额不等于当前估值
Ambarella公开边缘 AI 可比公司$3.43B 市值公开申报文件和市场数据功能上最接近的公开芯片可比公司公开市场流动性和披露溢价
Horizon Robotics公开母公司 / 平台锚HK$68.20B 市值已上市、有披露、规模更大的平台有用的上界和母公司背景锚终端市场组合和成熟度不同
Cambricon中国公开 AI 芯片可比公司CNY 850.08B / ~$125.60B 市值信号公开市场热度信号显示 AI 芯片情绪能热到什么程度过度亢奋、覆盖过宽,不适合作为基准情景可比公司

这个可比组合混合了一个干净的公开功能可比、一个母公司平台锚、一个动量很强的中国公开可比,以及一个私募融资类比,因为不存在完美一一对应的公开机器人芯片可比公司。

[CV005, CV007, CV009, CV010, CV012, CV026]

8.6 退出准备度、投资逻辑失效触发项和最终尽调问题

今天要承保一条清晰的公开市场式退出路径还太早。更可能的近期结果是另一轮私募融资、战略合作路径,或在 D-Robotics 披露可复制客户经济性之后再推进 IPO。避免买贵的最好办法,是盯住真正会打破投资逻辑的事项。如果公司在收入质量可见前又融资,如果国际客户证明无法从偏中国市场的 基础向外拓宽,或者出口管制收紧实质性改变供应商行为,合理估值区间就应下调,而不是上调。 最后的尽调问题直接而且要求很高:确切轮次条款、完整 cap table、产品线毛利率、烧钱和跑道、出口链依赖,以及按地域和用例划分的客户集中度。没有这些答案,即便战略故事仍有吸引力,公司也过于不透明,无法给出买入判断。因此,「跟踪」是正确的操作立场:保持贴近、快速更新,并且只在硬数据追上时才准备重估。[CV029, CV030, CV031, CV032, CV036, CV037]

论点击穿与否决触发点表
触发点阈值对论点的传导行动含义
证明前融资收入质量或利润率可见前就需要新一轮融资说明资本强度跑在证明前面将区间移向熊情景
无国际量产定点到下一轮融资周期仍无可信海外 OEM 证明全球标准论点弱化下调溢价和退出假设
出口管制冲击关键供应商或工具路径受到实质性约束地缘政治折价扩大重新测算整个可比公司组合
支持负担持续板卡 / 工具链摩擦跨版本仍可见软件护城河论点弱化假设长期利润率更低
私募估值跑过已披露证明披露未追上,估值却开始接近公开可比公司区间合理变成偏贵不追轮次

这些具体事件最可能击穿当前公允价值立场。

[CV029, CV030, CV031, CV037]
最终尽调清单表
主题缺失证据重要性负责人或尽调路径
轮次条款确切估值、优先权、清算顺位、按比例认购权决定真实入场经济性公司 / 法律顾问
现金续航现金余额、月度现金消耗、下一轮融资时间决定下行风险和议价能力公司 / 财务
利润率按芯片、套件和软件 / 服务层拆分的毛利率决定平台价值是复利还是流失公司 / 财务
客户质量具名量产客户、留存、按地区的集中度决定生态牵引是否转成可持续收入公司 / GTM
供应链代工、封装、EDA 与出口链依赖决定地缘政治折价和韧性公司 / 运营
软件变现付费支持、工具许可证或附加经济性决定平台广度是否配得上溢价倍数公司 / 产品

没有这些输入,公允价值只能表达为区间,而不是精确的当前估值。

[CV032, CV036, CV038, CV039]

8.7 图表

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。D-Robotics 是未上市公司,未披露财务数据;关键估值、收入和团队事实仍未公开,任何投资决策前都应向管理层和一手文件直接核验。

证据索引

结论
编号陈述可信度来源
CO001 D-Robotics publicly positions itself as a provider of robotics development infrastructure that combines hardware, operating-system layers, and applications. SO001, SO002
CO002 The company mission on its English site is to expand the intelligent capabilities of robots. SO001, SO002
CO003 The RDK brand is presented as an all-in-one developer kit for robotics and edge intelligence development. SO001, SO004
CO004 D-Robotics markets more than 200 open-source algorithms and application examples for faster robotics development. SO001, SO005
CO005 The homepage frames the product line as covering roughly 5 TOPS to 128 TOPS of proprietary BPU inference performance. SO001, SO005
CO006 RDK X5 is marketed as a 10 TOPS development kit for robotics and edge intelligence with support for advanced models such as Transformer, RWKV, and stereo perception. SO004, SO026
CO007 The RDK X5 documentation indicates Ubuntu 22.04 system support and PoE-capable gigabit Ethernet on the board. SO004, SO015
CO008 RDK S100 is marketed around a next-generation BPU Nash architecture with 80 or 128 TOPS of on-device inference and support for more than 160 ONNX operators. SO005, SO016, SO027
CO009 RDK S100 is positioned as a dual-brain CPU+BPU+MCU architecture for perception, decision, and real-time motion control. SO005, SO027
CO010 TogetheROS.Bot is described as a robot operating system for robot manufacturers and ecosystem developers running on the RDK platform. SO017, SO018
CO011 TogetheROS.Bot keeps ROS2 Foxy/Humble API compatibility and adds zero-copy communication, accelerated codec and CV libraries, and packaged algorithms. SO017
CO012 The D-Robotics GitHub organization showed 229 repositories at the time of access. SO018
CO013 The public GitHub organization includes RDK OS image tooling, model-zoo, camera, and robotics repositories, indicating an active open developer surface. SO018, SO019
CO014 NodeHub is presented as an intelligent robotics application center with project source code shared on GitHub. SO012
CO015 The community page shows an evangelist program and named contributors from multiple countries, indicating international community-building rather than a China-only developer motion. SO003, SO009
CO016 LinkedIn lists D-Robotics as a privately held computer hardware manufacturing company with a 51-200 employee size band. SO020
CO017 The public LinkedIn page showed 2,863 followers and about 20 visible employee profiles on the access date. SO020
CO018 The About page says D-Robotics collaborates with global OEMs and partners in humanoid robots, companion robots, robotic vacuum cleaners, and robotic lawn mowers. SO002
CO019 Official 2026 channels show active outbound ecosystem development through embedded world participation, ICML networking, and a startup gravity program. SO007, SO008, SO009
CO020 The DGP post says D-Robotics offers startups hardware, software, and ecosystem support as part of its gravity program. SO009, SO007
CO021 Horizon Robotics states it was incorporated in 2015 and frames itself as a horizontal platform for robotics. SO023
CO022 Horizon Robotics identifies Dr. Yu Kai as founder and CEO on its English About page. SO023
CO023 Horizon Robotics says it launched the Nash BPU in 2024, and D-Robotics now markets BPU Nash in the RDK S100 line. SO023, SO005
CO024 Yicai reports that D-Robotics was spun off from Horizon Robotics' AIoT division in January 2024. SO024, SO025
CO025 Yicai and CXO both report a $120 million Series B1 followed by a $150 million Series B2 in 2026, for about $270 million of Series B financing. SO024, SO025
CO026 Yicai names B2 investors including Didi Global, Prosperity7 Ventures, GL Ventures, Vertex Growth, and 5Y Capital. SO024
CO027 Yicai says B1 participants included Didi Global, Meituan Long-Z Fund, BAIC Capital, Xilian Capital, and Joyoung Family Office. SO024
CO028 Independent 2026 reporting frames D-Robotics as building embodied-AI infrastructure through integrated computing platforms, development tools, and hardware-software systems. SO024, SO025
CO029 Yicai reported an April 2026 product stack that included the RDK S600 platform and embodied-AI model systems HoloBrain and HoloMotion. SO024
CO030 The current English site instead emphasizes RDK X5 and RDK S100, implying the publicly described product portfolio is evolving rapidly across hardware generations and branding layers. SO001, SO024
CO031 Yicai says shipments rose 180 percent year over year and the customer base grew 200 percent in the latest reported year. SO024, SO025
CO032 Yicai and CXO both say the company has launched more than 100 robot products and built a developer community of more than 100,000 users. SO024, SO025
CO033 The public evidence reviewed supports a 100,000-plus community floor, but not the higher 200,000-plus registered-developer claim in the user brief. SO024, SO025, SO020
CO034 CNX Software shows third-party hardware vendors already embedding the RDK X5 into 2026 robotics products such as LooperRobotics' Insight 9 spatial AI camera. SO026
CO035 Switch Science listed the RDK S100 Developer Kit for sale in Japan at a local retail price of ¥102,300 and a seller-listed MSRP of $499 excluding tax. SO027
CO036 Switch Science lists the Japanese release date for the RDK S100 Developer Kit as 2026-01-20. SO027
CO037 Public materials reviewed do not disclose D-Robotics' board composition, voting control, or detailed governance rights. SO002, SO020, SO024
CO038 Public materials reviewed do not disclose D-Robotics' revenue, revenue mix, or SDK licensing contribution. SO001, SO024
CO039 Headquarters disclosure is still opaque in the public file: Yicai calls D-Robotics Shenzhen-based, while the English official surfaces reviewed do not state a headquarters city. SO001, SO020, SO024
CO040 Yicai's private-equity analysis says robotics capital is surging, but warns that valuations in popular sectors may already be at a cyclical peak where bubble risk becomes more evident. SO028
CO041 The same Yicai analysis says some hot-sector companies now require next-round commitments or even deposits to access investor lists, showing how competitive financing conditions can distort diligence. SO028
CM001 D-Robotics positions itself around robotics and edge-intelligence development infrastructure rather than finished robots-as-a-service. SM001, SM003
CM002 RDK S100 is explicitly marketed for embodied robots and on-device inference, showing that D-Robotics targets compute and control layers inside robot systems. SM002
CM003 TogetheROS.Bot indicates D-Robotics also targets software and middleware budgets linked to robot development, not only silicon attach. SM004
CM004 FinOps Foundation defines FinOps as a cultural practice and operating model for managing and optimizing technology value with shared ownership. SM006, SM007
CM005 Microsoft defines FinOps as a framework combining financial management, cloud engineering, and operations to align cloud spending with business value. SM008, SM009
CM006 Because D-Robotics sells robotics compute and software infrastructure, the relevant boundary overlaps cloud-cost optimization, Kubernetes efficiency, and dedicated robotics edge-compute platforms rather than all robotics spending. SM001, SM004, SM006, SM018
CM007 The relevant boundary should exclude generic ERP procurement spend and exclude robot operator service revenue that does not purchase compute or control infrastructure. SM006, SM014
CM008 MarketsandMarkets says the global cloud FinOps market is projected to rise from about USD 14.88 billion in 2025 to USD 26.91 billion by 2030. SM017
CM009 The Business Research Company says Kubernetes cost management reached USD 1.75 billion in 2025 and is expected to reach USD 2.23 billion in 2026 and USD 5.78 billion by 2030. SM018
CM010 Verified Market Reports sizes cloud cost management and optimization at USD 9.2 billion in 2026 and USD 35.4 billion in 2034 with a 14.1 percent CAGR. SM019
CM011 These research-house estimates are not directly interchangeable because they describe different category boundaries from broad Cloud FinOps to narrower Kubernetes cost management. SM017, SM018, SM019
CM012 A practical D-Robotics-relevant serviceable market is narrower than broad Cloud FinOps yet wider than strict Kubernetes cost management because robotics stacks also need on-device vision, control, and middleware. SM001, SM002, SM018, SM023
CM013 CNCF describes 2024 cloud native adoption as approaching a decade of code, cloud, and change, reinforcing that the underlying container and platform market remains broad and active. SM015
CM014 StartUs Insights identifies embodied AI and robotics automation as major 2026 trends, supporting continued strategic attention to robotics infrastructure. SM020
CM015 TI says the humanoid robot market could reach roughly USD 5 trillion by 2050, illustrating why capital is chasing enabling compute and sensing platforms even before near-term volumes are mature. SM025
CM016 Renesas frames robotics demand around precise control, real-time processing, safety, and connectivity across household, humanoid, industrial, and mobile robots. SM023
CM017 Ambarella highlights consumer robots, autonomous delivery, machine vision, inventory logistics, and smart-home applications as target robotics segments. SM021, SM022
CM018 MediaTek markets Genio through edge-AI examples such as industrial HMI, workplace safety, and manufacturing productivity, indicating adjacent buyer demand for low-power embedded AI platforms. SM026
CM019 D-Robotics itself separates makers from business customers on its site, implying at least two budget motions: experimentation and production deployment. SM001
CM020 Google says GKE Autopilot is designed for admins, architects, and operators and lets Google manage nodes, scaling, security, and default resource settings. SM013
CM021 Karpenter is an open-source node lifecycle management project built to improve the efficiency and cost of Kubernetes workloads by provisioning nodes to meet pod requirements. SM010, SM011
CM022 Red Hat says cost management for OpenShift is intended for both IT and financial stakeholders and maps cluster spend to business priorities across hybrid and multicloud environments. SM014
CM023 Microsoft says FinOps depends on collaboration among engineering, finance, and business teams, so the user and payer are inherently split roles. SM008, SM009
CM024 Taken together, the public sources imply platform engineering or SRE teams are the most common hands-on users, while finance, FinOps leaders, and central infrastructure budgets often control or co-own spend. SM008, SM013, SM014
CM025 AI infrastructure owners become additional buyers when edge vision, LLM, VLM, or embodied-control workloads enter the design, because GPU or accelerator efficiency becomes part of the purchase decision. SM002, SM025, SM026
CM026 A major demand driver is persistent Kubernetes and cloud-native complexity, which keeps waste, overprovisioning, and operational friction visible enough to justify optimization budgets. SM015, SM014, SM011
CM027 Another demand driver is the shift toward low-power edge AI in robots, where performance-per-watt and thermal limits matter more than generic datacenter throughput. SM002, SM022, SM023
CM028 Humanoid, industrial, and mobile robots increase demand for real-time control, multi-sensor fusion, and safety-ready compute stacks, which expands the relevance of specialized robotics semiconductors. SM023, SM024, SM025
CM029 Native substitutes such as GKE Autopilot and Karpenter can solve meaningful parts of infrastructure optimization without paying a separate third-party vendor. SM013, SM011
CM030 Red Hat and Microsoft materials show that visibility, tagging, and showback can also be addressed inside existing cloud or platform-management stacks, which can compress third-party differentiation. SM008, SM014
CM031 Market-definition ambiguity is itself a constraint because wide Cloud FinOps or robotics narratives can overstate the near-term SAM for a company that still depends on specific robot compute deployments. SM017, SM018, SM019
CM032 Hybrid and multicloud data hygiene remains a practical adoption constraint because cost-management value depends on mapping spend back to projects, clusters, and cost centers. SM012, SM014
CM033 A constrained 2026 D-Robotics-relevant SAM of roughly USD 2-4 billion is defensible as an author estimate anchored on Kubernetes cost management plus embodied-AI and robotics edge-compute adjacency. SM018, SM019, SM023
CM034 A realistic third-party SOM for cross-cloud robotics compute optimizers is materially smaller, perhaps around USD 0.3-0.8 billion, because native tools and broader platform suites will keep part of the spend in-house. SM013, SM018, SM019
CM035 Official D-Robotics materials suggest the company is trying to monetize the enabling layers beneath robots—compute, middleware, examples, and tools—rather than final robot labor outcomes. SM001, SM004
CM036 Renesas, Ambarella, TI, and MediaTek all market robotics or edge-AI capabilities, showing that the comparison set for D-Robotics is wider than Nvidia alone and spans vision, control, connectivity, and power ecosystems. SM021, SM023, SM025, SM026
CM037 The market opportunity is strongest where buyers need low-power on-device inference, real-time control, and a full developer stack rather than only a dashboard for cloud bills. SM002, SM004, SM023
CM038 Public sources reviewed do not isolate a China-specific robotics edge-compute TAM, so any country-level SAM remains a material evidence gap. SM017, SM018, SM019
CM039 The official and analyst evidence does not support a clean single-number TAM for D-Robotics; a layered range is more honest than one broad headline market size. SM017, SM018, SM019
CM040 Buyer adoption will likely require a workflow from prototype development to production deployment, which is why D-Robotics emphasizes documentation, middleware, and reference examples alongside hardware. SM001, SM003, SM004
CP001 D-Robotics presents a full robotics-compute stack rather than only a chip SKU, combining boards, inference silicon, middleware, and deployment tools. SP001, SP003, SP004
CP002 The public D-Robotics surfaces connect hardware, robot OS, docs, community, and examples into one onboarding path for external developers. SP001, SP008, SP009, SP010
CP003 RDK X5 is marketed as a 10 TOPS development kit for robotics and edge intelligence. SP002
CP004 RDK S100 is positioned as a BPU Nash-based robotics platform, with public evidence for 80 TOPS retail hardware and an upper-family narrative reaching 128 TOPS. SP003, SP026
CP005 TogetheROS.Bot extends ROS2 Foxy and Humble with zero-copy communication, sensor support, codec and CV libraries, model inference helpers, and packaged algorithms. SP004
CP006 The public GitHub organization and documentation repository show that D-Robotics maintains an open developer surface rather than a closed appliance model. SP005, SP006
CP007 The community, NodeHub, and RDK Studio pages indicate a deliberate external-builder motion, not only internal reference design support. SP008, SP009, SP010
CP008 Horizon Robotics lineage and Yu Kai association help explain why D-Robotics enters the market with real chip and platform credibility, but they do not eliminate current execution risk. SP001, SP007, SP027
CP009 NVIDIA publicly offers a Jetson Orin ladder spanning roughly 34 TOPS to 275 TOPS across Nano, NX, and AGX variants. SP011, SP028
CP010 NVIDIA Jetson Thor is marketed for physical AI and humanoid robotics with up to 2,070 FP4 TFLOPS, 128 GB memory, and a 40-130 W envelope. SP012
CP011 NVIDIA competes on top-end performance breadth and software-stack ambition, not only on low-power embedded inference. SP011, SP012, SP028
CP012 Qualcomm Robotics RB5 combined a 15 TOPS QRB5165 robotics processor with Linux, Ubuntu, ROS 2 support, optional 5G, and public dev-kit pricing at $485 and $695. SP013, SP030
CP013 Qualcomm RB3 Gen 2 uses a QCS6490-based platform with 12 TOPS AI performance and a long-lifecycle embedded posture for robotics, IoT, and embedded applications. SP014
CP014 Qualcomm competes as an embedded and connectivity-heavy robotics reference-platform alternative, sitting between smartphone-derived AI compute and robotics dev kits. SP013, SP014, SP030
CP015 Ambarella explicitly targets consumer robots, AGVs, warehouse logistics, and industrial machine vision in its AIoT robotics materials. SP015, SP016
CP016 Ambarella markets low-power CVflow computer vision and AI performance-per-watt as core differentiation for robotics-related SoCs. SP015, SP016, SP029
CP017 Renesas positions itself as a broad robotics supplier spanning AI vision, motor control, power, networking, safety, and ROS 2-ready compute. SP017, SP018
CP018 Renesas claims edge inferencing up to 80 TOPS at 5 W and explicitly highlights ROS 2-ready MPUs and reference designs. SP017
CP019 Renesas humanoid materials emphasize multi-axis motor control, system compute, power, sensor integration, and functional safety as a system-level package. SP018
CP020 TI argues that humanoid robots need multi-sensor fusion, more than 50 degrees of freedom in advanced designs, and deterministic communications and safety-aware motor-control coordination. SP019
CP021 MediaTek Genio competes as a general edge AI and IoT platform used across industrial, signage, control-panel, and productivity devices rather than as a robotics-only stack. SP020
CP022 GKE Autopilot is a meaningful status-quo substitute for infrastructure management because Google operates nodes, scaling, security, and worker-node configuration in managed mode. SP021
CP023 Karpenter is a meaningful open-source substitute for infrastructure optimization because it automatically provisions right-sized nodes and consolidates under-utilized capacity. SP022
CP024 Microsoft's FinOps toolkit shows that some buyers can improve cost visibility and governance inside existing tooling rather than adopting a specialized robotics platform vendor. SP024
CP025 External adoption and channel proof exist: CNX reported a LooperRobotics product built on RDK X5 with ROS 2 support, and Switch Science retailed the RDK S100 in Japan at a visible price point. SP025, SP026
CP026 The competitive field spans direct stack vendors, high-end incumbents, embedded reference platforms, industrial subsystem suppliers, and status-quo internal-build paths. SP011, SP012, SP014, SP017, SP021, SP022
CP027 D-Robotics is most likely to win when a buyer needs low-power local inference, robotics middleware, and an accessible dev-kit-to-demo path at the same time. SP001, SP003, SP004, SP009, SP010
CP028 D-Robotics is more vulnerable when buyers prioritize frontier model scale, massive memory, or the largest physical-AI software ecosystem, where NVIDIA sets a higher public ceiling. SP011, SP012
CP029 Distribution power currently favors incumbents and industrial suppliers with broad OEM, safety, and networking relationships over a newer spinout still proving global channel depth. SP017, SP019, SP020, SP025
CP030 Multi-homing is plausible because builders can combine ROS-compatible middleware, mixed hardware, cloud services, and separate sensor or control suppliers before full standardization. SP004, SP017, SP019, SP021
CP031 Switching costs rise after teams tune model deployment, board interfaces, middleware packages, debugging workflows, and support relationships around one stack. SP004, SP013, SP014, SP017
CP032 Public pricing transparency is limited across the field: the reviewed file shows retail or dev-kit pricing for only selected offers, not a production-grade pricing matrix. SP013, SP026
CP033 D-Robotics' moat appears stronger in integration and ecosystem packaging than in publicly demonstrated raw benchmark leadership. SP001, SP004, SP005, SP006, SP025
CP034 Commoditization risk is real because multiple rivals now ship AI silicon with toolchains, long-lifecycle support, industrial references, or edge-AI software stories. SP014, SP016, SP017, SP020
CP035 The reviewed public file does not provide an independent apples-to-apples benchmark or TCO matrix proving D-Robotics beats the field across workloads. SP002, SP003, SP011, SP012, SP017
CP036 Robotics demand is broadening across industrial, logistics, consumer, and humanoid categories, which means D-Robotics will likely face new entrants and adjacent rivals rather than a static peer set. SP015, SP017, SP018, SP020, SP023
CP037 D-Robotics' open-source and documentation surface likely lowers initial adoption friction for developers relative to more opaque platform approaches. SP004, SP005, SP006, SP009
CP038 Open developer friendliness alone may not overcome incumbent ecosystem gravity at production scale, especially when buyers care about long support windows, distribution, and top-end software ecosystems. SP011, SP017, SP019
CI001 D-Robotics publicly frames itself as a robot soft/hardware base supplier rather than a robot-body manufacturer. SI001, SI004
CI002 The public stack spans smart-compute chips, RDK developer kits, robot software manuals, and model-deployment assets. SI003, SI004, SI005, SI006
CI003 36Kr described RDK X5 as a developer kit aimed at broad robot categories such as lawn mowers, vacuums, and quadrupeds. SI001
CI004 36Kr described RDK S100 as a higher-compute embodied-intelligence platform for perception, understanding, and decision workloads. SI001
CI005 The RDK X5 product page exposes a hardware bill of materials and interfaces but no public checkout or list price. SI003
CI006 The RDK S100 product page exposes architecture and compute information but no public price, margin, or contract terms. SI004
CI007 The TogetheROS.Bot manual shows D-Robotics also invests in software onboarding and application-layer tooling around the hardware. SI005
CI008 The public GitHub organization shows 229 repositories, supporting a broad engineering and open-ecosystem footprint. SI007
CI009 The RDK model zoo provides end-to-end model conversion, quantization, inference, and validation examples for BPU deployment. SI006
CI010 The community course catalog shows thousands of look-counts for robotics and toolchain courses, which is a self-serve demand proxy rather than a revenue disclosure. SI008, SI009
CI011 The recent-activity feed shows D-Robotics still funds contests and outreach programs, implying ongoing ecosystem-development spend in 2026. SI010
CI012 The cooperative-partner feed lists education robots, AI companion robots, industrial cameras, and mobile dual-arm systems, indicating partner-led route-to-market breadth. SI011
CI013 The developer-case feed shows third parties building manipulation, quadruped, logistics, and safety applications on the platform, implying customer-acquisition leverage from community projects. SI012
CI014 The forum latest feed shows active problem-solving around CAN FD, dropped frames, tool-download issues, and model-conversion precision, implying support load and onboarding friction. SI013
CI015 Third-party tutorials on CSDN still focus on first-boot, image flashing, and remote login, suggesting onboarding remains material enough to spawn external guides. SI014, SI015
CI016 Baidu Baike states D-Robotics completed a $100 million A round on 2025-05-28. SI002
CI017 Baidu Baike states D-Robotics completed a $120 million B1 round on 2026-03-16. SI002
CI018 Baidu Baike states D-Robotics completed a $150 million B2 round on 2026-04-08 and brought cumulative B-round funding to $270 million. SI002
CI019 Baidu Baike describes D-Robotics as a newly added unicorn in a 2025 China unicorn observation report released in 2026. SI002
CI020 The fetched public sources do not disclose post-B2 cash on hand, monthly burn, or runway. SI001, SI002, SI003, SI004
CI021 The fetched public sources do not disclose recognized revenue, ARR, gross margin, or EBITDA. SI001, SI002, SI003, SI004, SI005
CI022 Baidu Baike states the company had over 100 listed products, over 100 upstream and downstream partners, and 100,000 developers by December 2025. SI002
CI023 Baidu Baike states D-Robotics had established talent-development cooperation with more than 500 schools by December 2025. SI002
CI024 The recent-course feeds show dedicated X5 and S100 ISP training programs with thousands or hundreds of public view counts, indicating continuing post-launch enablement costs. SI009
CI025 The D-Robotics platform therefore appears to monetize through a mix of hardware, software enablement, and ecosystem support rather than through a single chip-only contract model. SI001, SI003, SI004, SI005, SI006, SI009
CI026 Ambarella and NVIDIA both maintain formal investor-relations and filing surfaces, underscoring how little public financial disclosure exists for D-Robotics by comparison. SI016, SI018
CI027 NVIDIA’s Jetson lineup illustrates that edge-AI vendors often monetize a full platform of modules, developer kits, and software stacks rather than raw silicon alone. SI020
CI028 Ambarella’s IR surfaces show listed edge-AI vendors are judged on periodic filings and financial cadence that private D-Robotics does not yet provide publicly. SI016, SI017
CI029 Leju’s official site and 2026 Tencent coverage show embodied-robot customers and partners are themselves still early commercial businesses, which weakens near-term downstream spending visibility. SI022, SI023
CI030 The fetched public sources do not identify the B1 or B2 investor roster with enough specificity for underwriting. SI002, SI024
CI031 Because no public customer-concentration figures are disclosed, partner breadth cannot be translated into revenue diversification. SI011, SI012
CI032 Forum evidence of toolchain, I/O, and conversion problems suggests D-Robotics still bears nontrivial support and product-hardening cost after launch. SI013, SI014, SI015
CI033 Public evidence supports strong ecosystem breadth but not revenue quality because usage, course views, and repository count do not convert directly into contracted revenue. SI007, SI008, SI009, SI012
CI034 The B-round raises reduce near-term financing risk but do not eliminate next-round dependency because cash, burn, and margin data remain undisclosed. SI002, SI003, SI004
CI035 D-Robotics therefore remains a capital-dependent growth story in public evidence, not a self-underwritten profitability case. SI001, SI002, SI021
CE001 The RDK X5 product page exposes a developer-board workflow centered on cameras, displays, USB, Ethernet, CAN FD, and Ubuntu 22.04 support. SE001, SE002, SE003
CE002 RDK X5 publicly lists 4x USB 3.0 host ports, a USB 2.0 device port, dual MIPI CSI, HDMI, MIPI DSI, and PoE-capable gigabit Ethernet. SE001, SE002
CE003 The Chinese RDK X5 page additionally advertises Wi-Fi 6 and Bluetooth 5.4 for wireless connectivity. SE002
CE004 The hardware docs position RDK X5 as a board for multimedia, deep-learning, and multi-sensor application development and testing. SE003
CE005 The RDK S100 page presents a “big brain + small brain” architecture that combines CPU, BPU, and MCU resources for robotics. SE004, SE018
CE006 The public S100 page states the platform uses 6x Arm Cortex-A78AE CPU cores and a BPU Nash architecture. SE004, SE017
CE007 The public S100 page states the BPU Nash supports 80/128 TOPS and 160+ ONNX standard operators optimized for CNN and Transformer workloads. SE004, SE008
CE008 The S100 page states 4x Arm Cortex-R52+ MCU resources are used for high-frame-rate, low-latency joint real-time control. SE004, SE021
CE009 The archived S100 hardware-doc page says the content moved to a new document center in June 2026, implying an actively maintained docs migration. SE005
CE010 The TogetheROS.Bot manual exposes installation, ROS2 package usage, SLAM mapping, and voice-controlled car demos as part of the public robotics software layer. SE006, SE007
CE011 The GitHub organization page describes one repo as the entry point for building TogetheROS.Bot and shows 229 repositories at fetch time. SE007, SE023
CE012 The rdk_model_zoo repository exposes BPU-ready workflows from PyTorch/ONNX through quantization, inference, post-processing, and validation. SE008, SE009
CE013 The rdk_x5 branch is the primary delivery branch for RDK X5 and recommends RDK OS >= 3.5.0 based on Ubuntu 22.04 and TROS-Humble. SE009
CE014 The model zoo covers classification, detection, segmentation, pose estimation, OCR, and multi-modal models. SE008, SE009
CE015 The model zoo explicitly states that non-quantizable or BPU-unsupported operators can fall back to CPU execution. SE009
CE016 The recommended-course feed shows dedicated X5 ISP and S100 ISP training with 3,198 and 1,432 public look counts respectively. SE010, SE011
CE017 The college-home feed also shows YOLOv5 deployment and toolchain courses with multi-hundred to multi-thousand engagement counts. SE012
CE018 The recent-activity feed shows D-Robotics still organizes robotics competitions and ecosystem education programs in 2026. SE019
CE019 The cooperative-partner feed links the platform to education robots, AI companion robots, industrial cameras, dual-arm robots, radar-based sensing, and digital fitness solutions. SE020
CE020 The developer-case feeds show third-party builds in safety manipulators, dual-arm robots, quadrupeds, home robots, and SLAM systems on the broader D-Robotics/Horizon stack. SE021, SE022
CE021 The DGP page says D-Robotics wants to shorten robot MVP construction time by offering full-stack enablement to creators. SE023
CE022 Leju’s company page and Tencent’s 2026 coverage confirm that at least one visible D-Robotics partner operates in the still-unprofitable humanoid/education robot market. SE024, SE025
CE023 36Kr says D-Robotics’ commercial strategy is to cover broad robot categories first so they contribute baseline revenue and influence. SE018
CE024 Baidu Baike says D-Robotics had more than 100 products, more than 100 partners, and 100,000 developers by late 2025. SE017
CE025 The forum latest feed contains live complaints on CAN FD, dropped frames, bin conversion precision, missing downloads, GPIO contact issues, and model-tooling questions. SE013
CE026 The CSDN onboarding posts still emphasize image flashing, Wi-Fi setup, SSH, and sample-run validation, which is typical of a developer-first rather than appliance-like product. SE015, SE016
CE027 The public stack therefore contains four visible layers: compute hardware, robotics middleware, model-deployment tooling, and training/community support. SE001, SE004, SE006, SE008, SE010
CE028 The public docs do not expose formal enterprise-grade SLA, uptime guarantees, or downloadable audit packages for institutional diligence. SE001, SE004, SE006, SE013
CE029 The products look mature enough for serious prototyping and pilot integration, but not fully disclosure-complete for industrial qualification. SE001, SE004, SE006, SE013, SE016
CE030 NVIDIA Jetson’s module family illustrates the benchmark D-Robotics is competing against: a tightly coupled stack of modules, kits, software, and partners. SE026
CE031 RDK X5 exposes a classic edge-AI board shape with broad I/O, while S100 pushes toward embodied-robotics compute/control specialization. SE001, SE004
CE032 The public feeds show documentation upkeep in 2026 through course refreshes, doc migration, and active forum moderation rather than through a formal roadmap deck. SE009, SE010, SE011, SE013
CE033 Because no verified public benchmark sheet was fetched, the strongest external performance proof remains examples, tutorials, and community deployments rather than audited throughput claims. SE003, SE008, SE013, SE015
CE034 The presence of partner modules and developer cases shows real ecosystem usage, but it also implies dependency on third-party hardware, sensors, and robot-body builders. SE019, SE020, SE021, SE023
CE035 The fetched public sources consistently support D-Robotics as a capable developer-centric robotics platform with expanding embodied-robot reach and unfinished production-grade disclosure. SE004, SE006, SE007, SE013, SE023
CE036 RDK X3 is D-Robotics' entry-tier board, publicly specified with a quad-core Cortex-A53 CPU and a dual-core Bernoulli-architecture BPU rated at 5 TOPS. SE028, SE029, SE030
CE037 The RDK X3 product page advertises "100+ Open Source Algorithms and Applications" delivered through the NodeHub catalog to accelerate solution deployment. SE028
CE038 D-Robotics has publicly announced RDK S600, a compute platform unveiled at the DDC 2025 conference and targeted for an early-2026 launch, positioned above RDK S100 as the top of the hardware ladder. SE031, SE032
CE039 RDK S600 is described as delivering 560 TOPS of INT8 compute and is framed around edge-cloud collaboration rather than pure on-device inference. SE031, SE032
CE040 Coverage of the RDK S600 announcement cites a cloud-side platform offering "hundreds of ready-to-deploy robotics algorithms" alongside device-side BPU refinement, consistent with growth from the 100+ algorithm figure cited at RDK X3 launch toward a 200+ figure by the S100/DDC-2025 period. SE031, SE032
CE041 NodeHub is D-Robotics' public "intelligent robotics application center," offering a catalog of nodes and application demos (line patrol/obstacle avoidance, indoor service robot, robot arm) with project source published on GitHub. SE033, SE028
CE042 RDK Studio is publicly positioned as an "AI-Native Workbench for Robotics" that flashes and manages RDK X3, RDK X5, and RDK S100/S100P boards, and can additionally reach generic Linux hosts, NVIDIA Jetson, Raspberry Pi, and Rockchip devices over SSH for a narrower set of RDK-specific functions. SE034, SE035
CE043 RDK Studio's documented "OpenClaw" on-device deployment path requires SSH or network reachability, and board-specific model files (.hbm) remain tied to a specific RDK generation and are not interchangeable across boards. SE035
CU001 D-Robotics' DGP accelerator states it has supported more than 200 early-stage robotics companies worldwide and that more than 100 robotic product categories have been built on its platform. SU001, SU015
CU002 D-Robotics reported roughly 180% year-on-year growth in product shipments and about 200% growth in its customer base over the prior year. SU009, SU015
CU003 36Kr reporting carried by KrASIA states D-Robotics has partnered with more than 60 companies across the robotics supply chain. SU009
CU004 D-Robotics' official About page states the company has established collaborations with premier global OEMs across humanoid robots, companion robots, robotic vacuum cleaners, and robotic lawn mowers, and describes reaching millions of users worldwide without giving a verifiable figure. SU018
CU005 D-Robotics CEO Wang Cong told 36Kr that the company's near-term shipment volume relies on mass-market categories such as robot vacuums, lawn mowers, and companion robots, while embodied-intelligence and humanoid applications remain an earlier-stage, longer-horizon bet. SU016
CU006 In the 36Kr interview, Wang Cong named embodied-intelligence and research partners including Xingdong Jiyuan, Zhuji Dongli, Qiuzhi Technology, Tsinghua University's AIR Institute, and RealMan as collaborators. SU016
CU007 Hengbot's Sirius robotic dog integrates a D-Robotics RDK X3 AI head delivering up to 5 TOPS of edge computing power, positioned by D-Robotics as a flagship DGP partnership case. SU002, SU022
CU008 Hengbot launched Sirius on Kickstarter with 3,000+ interested beta testers and 600+ pre-orders ahead of its 2025 launch, per D-Robotics' own blog post. SU002
CU009 As of the run date, Hengbot's official homepage shows its crowdfunding campaign for Sirius raised $912,768 from 1,150 backers. SU004
CU010 Both publicly indexed Kickstarter campaign URLs for Hengbot's Sirius returned HTTP 404 on the run date, so the original crowdfunding page is not independently recoverable from its source. SU026
CU011 Elephant Robotics' myCobot 280 RDK X5 is a 6-DOF robotic arm that embeds an RDK X5 board with up to 10 TOPS of AI compute, marketed by D-Robotics as a budget-focused developer and education product. SU003
CU012 D-Robotics' blog post on the myCobot 280 RDK X5 states Elephant Robotics, founded in 2016, has expanded to 51 countries and regions across manufacturing, commercial, research, healthcare, logistics, and education verticals. SU003
CU013 36Kr reporting carried by KrASIA states D-Robotics supports Narwal's Xiaoyao 002 robot vacuum by supplying AI-powered stereo perception for obstacle detection. SU009
CU014 Narwal's own homepage states the company has entered more than 30 countries and markets and serves more than 450 million global users, independent evidence that the Xiaoyao 002 relationship sits inside a commercially scaled robot-vacuum brand rather than a small pilot account. SU005
CU015 An independent robotics-unicorn ranking (CNMRA) lists Narwal among home and commercial cleaning-robot makers whose category is 'ensuring stable commercial revenue streams,' corroborating that the Narwal channel is a commercially mature buyer rather than a speculative one. SU021
CU016 36Kr reporting carried by KrASIA states D-Robotics enabled Insta360's launch of the Antigravity A1, billed as the world's first panoramic drone. SU009
CU017 Independent reporting (China Biz Insider) states the Antigravity A1, from a team backed by Insta360 (Shenzhen Arashi Vision), is powered by D-Robotics' new-generation Sunrise 5 chip delivering up to 10 TOPS with a heterogeneous BPU/CPU/GPU/DSP architecture, marking D-Robotics' first commercial application in consumer drones. SU012, SU013
CU018 Insta360's own official blog post announcing the Antigravity A1 quotes co-founder Max Richter but does not itself name D-Robotics or its chip anywhere in the page, so the chip relationship is confirmed only by third-party reporting rather than by Insta360's own marketing copy. SU013
CU019 36Kr reporting carried by KrASIA states D-Robotics is working with Vbot on a quadruped companion robot, but unlike Hengbot's multi-source, dedicated-launch corroboration, no additional independent source or product page confirms the Vbot relationship as of the run date. SU009
CU020 Beijing Innovation Center of Humanoid Robotics and D-Robotics jointly developed the full-size humanoid Tiangong 3.0, which is powered by D-Robotics' Xuri S600 embodied-AI compute chip and is scheduled to begin mass production and delivery in the second half of 2026. SU010
CU021 Independent coverage (en.shuziqushi.com, citing Jiemian News) corroborates that Tiangong 3.0 uses D-Robotics' Sunrise S600 chip and is set for mass delivery in late 2026, and adds that the projected total machine cost of Tiangong 3.0 will fall by more than 50% at scale. SU011
CU022 Per Gasgoo, D-Robotics and Beijing Innovation Center engineers compressed Tiangong 3.0's native large model to less than 20% of its original size after more than a year of joint technical debugging, to achieve precise adaptation with the Xuri S600 chip. SU010
CU023 D-Robotics' own developer-portal cooperative-partner API lists nine named DGP member companies as of the run date, spanning education robots (Leju Robotics), AI companion robots (Hengbot), embodied dual-arm robots (Qiuzhi Technology), wearable path-guidance (AI Guided), golf-caddy robots (Caddy Trek), and smart fitness (FITURE). SU017
CU024 D-Robotics' community page lists five named robotics evangelists, including Frank Fu, described as CEO of Navbot, alongside a webmaster, a Robot Maker CEO, a robotics software engineer, and a high-school student. SU008
CU025 NavBot's own homepage markets its quadruped and STEM robotics products as NVIDIA Jetson powered and does not mention D-Robotics or RDK hardware anywhere in its public product messaging, indicating the Frank Fu evangelist relationship is a community-outreach tie rather than confirmed proof that NavBot ships D-Robotics silicon. SU007, SU008
CU026 D-Robotics hosted a live 'RDK X5 Show & Tell' YouTube event co-promoted with Make: Magazine editor David Groom, giving away an Elephant Robotics myCobot 280 arm and an RDK X5 kit to drive developer engagement. SU019
CU027 D-Robotics' Robotics Dream Keeper Challenge offers a four-stage progression path (Explorer, Builder, Creator, Core Developer) with mentoring, a $1,000 grand prize, and cash rebates toward RDK X5 purchases, an expansion mechanism aimed at converting hobbyist developers into repeat RDK buyers. SU020
CU028 D-Robotics' developer portal lists a 2026-dated 'Smart Healthcare' challenge track within the 21st National University Student Intelligent Vehicle Competition, open for registration through August 2026, showing ongoing 2026 community programming beyond one-off launch events. SU024
CU029 D-Robotics' developer-case showcase lists dozens of student and hobbyist projects built on RDK X3/X5 boards, such as university-team robotic arms, quadrupeds, and security robots, which demonstrate ecosystem breadth but are unpaid academic or maker submissions rather than named commercial customers. SU023
CU030 Yicai reported that D-Robotics has built a developer community exceeding 100,000 users across Asia-Pacific, Europe, and North America, a figure corroborated by D-Robotics' own myCobot blog post, though the company did not disclose shipment volumes or a revenue breakdown by segment or customer. SU015, SU003
CU031 An independent embodied-AI unicorn analysis (China Biz Insider) frames D-Robotics as attacking commercialization from the silicon layer, supplying compute chips and operating systems to robot manufacturers so that it captures value regardless of which application-layer winner emerges. SU014
CU032 The same China Biz Insider analysis places most Chinese embodied-AI cohort companies, including chip-level suppliers like D-Robotics, on 18-to-24-month cash runways with an expected industry reckoning between 2027 and 2028, and notes D-Robotics' April 2026 B2 round valued it at roughly RMB 10.8 billion without disclosing customer-revenue concentration. SU014
CU033 No public source reviewed discloses D-Robotics' revenue concentration by customer, net revenue retention, gross or logo retention, or renewal rates for any named partner (Hengbot, Narwal, Insta360/Antigravity, Vbot, or the Beijing Innovation Center), leaving the durability of the reported 200% customer-base growth unverifiable at the account level.
CU034 D-Robotics attributes the '100+ robotic product categories' and '200+ early-stage robotics companies supported' figures only to the DGP accelerator umbrella, without a public registry naming every company or reconciling categories against shipped-in-volume SKUs. SU001
CU035 Wang Cong's 36Kr commentary implies humanoid and embodied-intelligence deals such as Tiangong 3.0 are strategically important but not yet D-Robotics' shipment-volume driver, since he states shipment scale still depends on vacuum, lawn-mower, and companion-robot categories rather than humanoid or embodied intelligence. SU016
CU036 D-Robotics' cooperative-partner API and its dedicated Hengbot blog post both independently name Hengbot as a DGP member building the Sirius robotic dog, giving the Sirius case two separate official-channel confirmations of an active, ongoing partnership. SU002, SU017
CU037 D-Robotics' RDK X3 product page states the RDK X3 series delivers up to 5 TOPS of edge inference designed to integrate with generic products, the same compute tier D-Robotics' Hengbot blog post attributes to the Sirius robotic dog's AI head. SU022, SU002
CU038 D-Robotics' developer-college API lists ongoing 2026 course and competition content, such as smart-car competition preparation livestreams, targeted at the same student and maker audience its evangelist and Dream Keeper programs court, suggesting the community-expansion funnel is an actively maintained channel. SU025
CU039 Across all fetched official and independent sources, none disclose a customer-concentration metric such as the share of shipments or revenue tied to the top named partner, leaving open whether D-Robotics' commercial base is genuinely diversified across the roughly 60 named or reported supply-chain partners or effectively concentrated in a handful of high-volume OEM categories.
CU040 D-Robotics' own About page markets reaching millions of users worldwide through OEM partners, a claim directionally consistent with Narwal's independently reported 450 million global users, but that downstream user base is owned and reported by Narwal, not D-Robotics, so the headline figure cannot be attributed to D-Robotics-specific accounts. SU018, SU005
CR001 D-Robotics is a post-Horizon robotics-infrastructure spinout whose public positioning emphasizes robot compute, developer kits, and middleware rather than robot bodies. SR026, SR017
CR002 The public D-Robotics stack spans RDK X5, RDK S100, RDK X3, NodeHub, and DGP ecosystem tooling rather than a single discrete chip product. SR018, SR019, SR014, SR021, SR020
CR003 The RDK X3 page claims 100+ robot development kits and 100+ open-source algorithms, indicating meaningful ecosystem breadth but also a large support surface. SR014
CR004 NVIDIA's current Jetson lineup spans up to 67 TOPS on Orin Nano, 157 TOPS on Orin NX, 275 TOPS on AGX Orin, and 2070 FP4 TFLOPS with 128 GB on Jetson Thor. SR004
CR005 NVIDIA presents JetPack 7 as a robotics-ready software stack with real-time kernel support, CUDA, TensorRT, Isaac ROS, OTA, security, and cloud-native tooling. SR005
CR006 Reviewed D-Robotics public materials do not provide an apples-to-apples benchmark showing BPU Nash clearly outperforming Jetson-class alternatives on globally relevant robotics workloads. SR018, SR019, SR004
CR007 Qualcomm is still actively investing in robotics developer surfaces, including RB3 Gen 2 promotion at MIT Reality Hack 2025 and Edge Impulse support. SR006, SR007
CR008 Qualcomm also keeps a dedicated robotics hardware posture through the QCS6490 family and the RB5 platform press kit. SR008, SR009
CR009 Ambarella markets its robotics SoCs as higher computer-vision performance per watt than GPUs and FPGAs, with low thermal envelope and multi-sensor support. SR010
CR010 Kneron says it has raised $190 million and combines proprietary AI hardware, software, and an open development ecosystem, showing private edge-AI rivals remain funded and active. SR011
CR011 D-Robotics' competitive risk is therefore less about one direct benchmark race than about competing against multiple incumbents that each bring stronger visible software, channel, or capital advantages. SR005, SR007, SR010, SR011
CR012 BIS guidance dated May 31, 2026 says a license is still required for advanced computing items destined for entities headquartered in Country Group D:5 or Macau, even if the entity is located outside those jurisdictions. SR001
CR013 WilmerHale says the AI Diffusion Rule was paused, but BIS simultaneously issued new guidance that elevates risk for AI-related exports and services. SR002
CR014 WilmerHale says the May 2025 BIS guidance can trigger licensing obligations when exporters know AI models or advanced computing ICs may support China or other D:5 parties. SR002
CR015 WilmerHale's February 2025 fabricator note says new BIS rules added technical and know-your-customer due diligence burdens for front-end fabricators and OSATs serving advanced IC programs. SR003
CR016 The same fabricator note says that after April 13, 2026, designers need tighter approval status to retain authorized treatment, making China-linked supply chains more compliance-sensitive. SR003
CR017 Export-control tightening therefore threatens D-Robotics indirectly through vendor caution, red-flag screening, and service-provider reluctance even without a public sanctions event aimed specifically at D-Robotics. SR001, SR002, SR003
CR018 The public file still does not disclose D-Robotics' foundry, packaging, or Taiwan-related manufacturing exposure in a way that lets investors quantify geopolitically driven supply interruption risk. SR018, SR019, SR001
CR019 D-Robotics' distributor page shows international availability through external resellers in Japan, South Korea, India, Southeast Asia, Singapore, Taiwan, Germany, France, and Spain. SR012
CR020 The accessories page says official RDK accessories are available from local distributors, implying the company depends on channel partners even for peripheral availability. SR013
CR021 The public international footprint is broader than domestic-only sales, but it is still visibly reseller-led rather than built around disclosed overseas subsidiaries or named global OEM programs. SR012, SR013
CR022 Forum latest.json shows recurring 2026 issues involving CAN FD, GPIO, dropped frames, model-conversion precision, S100 MCU boot, and camera-stitching topics. SR016
CR023 The same forum activity is evidence of real ongoing usage, but also of continuing hardening and support work that can drag on operating leverage. SR016
CR024 D-Robotics' public surfaces still lean heavily on developer, maker, and education-style onboarding rather than a disclosed roster of global production OEM accounts. SR017, SR020, SR021, SR016
CR025 Horizon Robotics' investor-relations page shows the parent at a much larger operating scale, with 2025 revenue of RMB 3.76 billion and cumulative Journey chip shipments above 10 million. SR015
CR026 The same Horizon page says the parent works with 27 automakers across 42 brands and serves more than 6 million car owners, underlining the scale gap between parent and spinout. SR015
CR027 The scale gap means D-Robotics still carries execution risk as a recent spinout that must prove it can stand independently on go-to-market, support, and capital formation. SR026, SR017, SR015
CR028 Yicai and CXO Digital Pulse both report a 2026 Series B1 of $120 million and a B2 of $150 million, taking B-round capital to about $270 million. SR028, SR029
CR029 Baidu Baike and other secondary references describe D-Robotics as unicorn-scale, but the exact post-money valuation still lacks filing-grade public confirmation in the reviewed source set. SR030, SR028
CR030 Public sources reviewed still do not disclose D-Robotics revenue, gross margin, cash balance, burn rate, or runway. SR028, SR029, SR030
CR031 The company is simultaneously funding chip programs, developer kits, documentation, model-zoo software, community support, and partner enablement, which is structurally more capital intensive than a pure software model. SR018, SR019, SR025, SR016, SR020
CR032 That cost structure can absorb fresh capital quickly even after a $270 million B-round step-up if product hardening, support, and partner seeding keep scaling together. SR028, SR019, SR016
CR033 The GitHub organization and model zoo show a meaningful open-source surface, but that surface is a developer-interest signal rather than proof of high-margin enterprise monetization. SR024, SR025
CR034 D-Robotics' market risk is amplified because its public opportunity areas—AMRs, humanoids, and edge AI—are the same categories where NVIDIA, Qualcomm, Ambarella, Kneron, and Cambricon remain active. SR004, SR008, SR010, SR011, SR015
CR035 NVIDIA's software stack raises the competition bar beyond TOPS by bundling CUDA, TensorRT, Isaac ROS, OTA, and security into a single production path. SR005
CR036 Qualcomm's ongoing robotics kit promotion suggests that mobile-edge incumbents still treat robotics as a live strategic adjacency rather than a legacy niche. SR006, SR007
CR037 Ambarella's low-power vision positioning creates pressure exactly where D-Robotics argues edge efficiency and robotics readiness should matter most. SR010
CR038 The most visible mitigations today are a broad developer surface, an emerging international reseller footprint, and fresh capital that reduces immediate financing pressure. SR012, SR014, SR028, SR024
CR039 The main thesis-break triggers are failure to win non-China production customers, tighter export-control chokepoints on needed silicon or tooling, persistent field-support friction, or the need for new capital before commercialization proof matures. SR012, SR001, SR016, SR028
CR040 Based on public evidence alone, D-Robotics still carries medium-high to high residual exposure across technology differentiation, geopolitics, execution, and financial transparency. SR004, SR001, SR016, SR028
CR041 BIS added 80 entities to the Entity List in March 2025 for advanced-AI, supercomputing, and high-performance-chip activity tied to China's military-industrial complex, on top of prior 2022-2024 rounds. SR031
CR042 BIS's December 2024 rule imposes a broader license requirement on foundries and packaging companies exporting certain advanced chips unless the exporter uses a trusted Approved/Authorized IC designer or OSAT that verifies transistor counts. SR032
CR043 Torres Trade Law says the December 2024 Entity List Updates Rule added 140 entities and 16 Footnote-5 designations, alongside a companion interim rule adding new Foreign Direct Product rules and license exceptions. SR034
CR044 The Congressional Research Service's September 2025 report frames US export controls as a multi-year effort to slow China's stated 2030 goal of world-leading semiconductor self-sufficiency across the integrated-circuit supply chain. SR033
CR045 The 2022-2025 export-control rounds are cumulative rather than one-off, progressively expanding the population of entities, fabricators, and OSATs subject to license and due-diligence requirements that any China-based robotics-compute supplier's partners could fall under. SR031, SR032, SR033, SR034
CR046 TrendForce projects the US will maintain a one-to-two-generation ("N-1"/"N-2") performance gap on AI chips it permits into China, while wafer-capacity and HBM bottlenecks continue to constrain Chinese AI-chip makers even as domestic market share is projected to reach roughly 50% in 2026. SR039
CR047 IFR's May 2026 release says China's 15th Five-Year Plan places robotics at the heart of national industrial strategy, with an installed industrial-robot base of about 2 million units (~4.5x Japan's) and 54% of 2025's global industrial robot installations occurring in China. SR035
CR048 State-directed robotics industrial policy in China can concentrate demand-pull and preferential support toward officially favored national-champion platforms, and the public record does not show whether D-Robotics is positioned as a preferred beneficiary or merely one of many competing domestic vendors. SR035
CR049 TechCrunch reports that roboticist Rodney Brooks calls current humanoid-robot dexterity approaches "pure fantasy thinking," citing unresolved touch-sensing and fall-safety physics problems that push real commercialization timelines well beyond current investor expectations. SR036
CR050 CB Insights data reported by Robotics & Automation News shows industrial humanoid robotics captured the most single-category VC deal volume of any AI vertical in a recent quarter, alongside explicit investor warnings that AI hype is fueling a humanoid-robotics bubble. SR037
CR051 ChinaBizInsider reports at least 25 Chinese embodied-AI startups now carry valuations above RMB 10 billion (~US$1.39 billion), with 15 crossing that threshold in the first half of 2026 alone and combined disclosed funding of RMB 46 billion (~US$6.39 billion). SR038
CR052 The same ChinaBizInsider report says most of those startups carry only 18-24 months of cash runway, implying a probable consolidation, down-round, or shutdown wave across the sector in 2027-2028. SR038
CR053 D-Robotics markets its RDK S100 and BPU Nash line explicitly toward embodied and humanoid robotics customers, so a sector-wide humanoid-investment slowdown or capital pullback would plausibly reduce near-term demand for its highest-end silicon faster than its broader AMR and edge-AI business would suggest. SR019, SR038, SR036
CR054 The reviewed public record does not disclose what share of D-Robotics' shipments or sales pipeline is tied to humanoid/embodied-AI customers versus AMR, dashcam, or other edge categories, making the size of its bubble exposure unquantifiable from public evidence. SR019, SR038
CR055 No public standards body, regulator, or certification page in the reviewed source set names D-Robotics in a robotics safety, interoperability, or compliance-certification context, leaving that as an open diligence question rather than a confirmed gap. SR035, SR019
CR056 The layered export-control tightening (BIS/CRS/Torres) and the humanoid-bubble/China-industrial-policy evidence together indicate D-Robotics faces both a supply-side chokepoint risk and a correlated demand-side commercialization-timing risk, rather than two independent exposures. SR031, SR032, SR033, SR034, SR035, SR038
CV001 Yicai and CXO both report a $120 million Series B1 followed by a $150 million Series B2 in 2026, implying roughly $270 million of B-round capital for D-Robotics. SV013, SV014
CV002 Reviewed public sources still do not publish a filing-grade post-money valuation for D-Robotics' 2026 financing rounds. SV013, SV014, SV015
CV003 Baidu Baike describes D-Robotics as a newly added unicorn, giving directional evidence that the company sits in the billion-dollar class without proving the exact current mark. SV015
CV004 Yicai's 2026 private-equity analysis says capital is flooding hard-tech sectors such as robotics while some investors warn valuations may already be at cyclical peaks. SV023
CV005 StockAnalysis shows Horizon Robotics trading on the Hong Kong exchange with a market cap of HK$68.20 billion as of July 3, 2026. SV004
CV006 The same StockAnalysis page reports Horizon Robotics trailing-twelve-month revenue of CNY 4.18 billion, highlighting a disclosure and scale gap versus D-Robotics. SV004
CV007 CompaniesMarketCap shows Ambarella at a July 2026 market cap of $3.43 billion. SV001
CV008 SEC EDGAR shows Ambarella filed a 2026 10-K on March 23, 2026, reinforcing that the public comp benefits from filing-quality disclosure that D-Robotics lacks. SV008
CV009 Google Finance shows Cambricon at a market cap of roughly CNY 850.08 billion in early July 2026. SV003
CV010 CompaniesMarketCap simultaneously shows Cambricon at about $125.60 billion, making clear that current public China AI-chip comps are far above late-stage private robotics-startup ranges. SV002
CV011 Cambricon is therefore a useful ceiling-reference for AI-chip enthusiasm, but not a clean base-case underwriting anchor for D-Robotics. SV003, SV002
CV012 Kneron disclosed a $48 million Series B in 2022, a further $49 million in 2023 that brought total Series B financing to $97 million, and over $190 million total funding. SV006, SV005, SV024
CV013 Kneron also disclosed an earlier $18 million Series A1 round in 2018, giving a fuller private-capital history for a full-stack edge-AI comparison point. SV007
CV014 D-Robotics should be valued as a platform candidate rather than a generic board vendor because public materials show chips, development kits, model-zoo software, and ecosystem tooling together. SV017, SV018, SV019, SV021
CV015 However, the same public record still does not disclose revenue, ARR, gross margin, burn, or runway, so valuation cannot be underwritten the way public comps can. SV013, SV014, SV015
CV016 Because revenue is undisclosed, the right framework is milestone-based valuation rather than a simple revenue-multiple model. SV013, SV001, SV004
CV017 Public filing and market-data sources for Ambarella and Horizon highlight just how large the disclosure gap is between listed comps and D-Robotics. SV008, SV004, SV012
CV018 That disclosure gap requires a private-company discount to any public comp-derived fair value. SV008, SV004, SV001
CV019 D-Robotics still deserves some premium over smaller AI-hardware startups because it already shows a multi-layer robotics stack and recent late-stage capital access. SV018, SV019, SV021, SV013
CV020 US-China export-control friction is a legitimate valuation discount because it can affect tools, suppliers, services, and global customer comfort even without a company-specific sanctions event. SV025, SV026
CV021 A track recommendation with medium confidence best fits the current evidence because upside is real while price precision remains low. SV013, SV001, SV004
CV022 The right valuation stance is fair rather than attractive because recent funding, platform breadth, and robotics timing offset, but do not overcome, the disclosure gap and risk discount. SV013, SV023, SV001
CV023 A reasonable bear-case value range is roughly $0.8-1.1 billion if financing returns before revenue proof, export controls worsen, or overseas commercialization stalls. SV023, SV025, SV013
CV024 A reasonable base-case value range is roughly $1.2-1.8 billion if D-Robotics converts ecosystem traction gradually but still enters the next stage with opaque financial disclosure. SV013, SV001, SV024
CV025 A reasonable bull-case range is roughly $2.2-3.0 billion if D-Robotics proves non-China OEM wins, clearer software monetization, and a more globally durable supply chain. SV018, SV019, SV001
CV026 Ambarella is the cleanest public functional comp because it is a disclosed edge-AI chip company with a mid-single-digit-billion market cap rather than an auto-stack or momentum-heavy China AI platform. SV001, SV008
CV027 Horizon is more useful as an upper-bound parent-platform anchor than as a direct multiple comp because its public business is larger, broader, and more disclosed. SV004, SV030
CV028 Kneron is the cleaner private comparison point because it is also a full-stack edge-AI company, but its disclosed financing history places it earlier and smaller than D-Robotics' latest capital raised. SV006, SV005, SV024, SV013
CV029 The most important thesis-break trigger for valuation is another financing event before revenue quality, margin, and customer-concentration proof improve. SV013, SV023
CV030 A second thesis-break trigger is failure to show internationally relevant production customers beyond a China-heavy developer and reseller base. SV017, SV018, SV019
CV031 A third thesis-break trigger is a material export-control or vendor-compliance shock that changes the cost or availability of critical tooling or silicon. SV025, SV026
CV032 The minimum diligence package before paying above fair value is exact round terms, cap table, burn, runway, gross margin by product line, and customer concentration by geography. SV013, SV008, SV004
CV033 Public evidence supports only a fair-value band, not a precise current mark, because the key term-sheet and operating-metric inputs remain undisclosed. SV013, SV014, SV015
CV034 Parent Horizon's current public market cap of HK$68.20 billion places it several turns of proof above where a robotics-focused spinout like D-Robotics should sit today. SV004
CV035 D-Robotics' recent $270 million financing does justify a premium to Kneron-style earlier private funding references, but not a premium equal to disclosed public edge-AI platforms. SV013, SV005, SV001
CV036 Moving the stance from fair to attractive would require proof of revenue quality, gross-margin durability, and cleaner global customer diversification. SV004, SV008, SV017
CV037 Moving the stance from fair to stretched would require evidence that the private mark is already pricing D-Robotics near public-comp levels without disclosure or revenue proof. SV001, SV004, SV023
CV038 The exact round preference stack, liquidation terms, and investor-protection structure remain missing from public evidence.
CV039 Named customer economics, retention, and product-line gross margins remain missing from public evidence.
CV040 Overall, D-Robotics looks investable enough to track closely, but not disclosed enough to justify a buy call ahead of clearer commercial proof; fair is the balanced stance. SV013, SV001, SV004, SV024
来源
编号出版方标题引文
SO001 D-Robotics D-Robotics homepage
SO002 D-Robotics About D-Robotics | Contact Us
SO003 D-Robotics Join the D-Robotics Community | Collaborate & Innovate
SO004 D-Robotics RDK X5 Development Kit - 10 TOPS Computing
SO005 D-Robotics RDK S100 Development Kit - Advanced AI Computing Platform
SO006 D-Robotics The official documentation for the RDK series
SO007 D-Robotics Revolutionize Your Robotics Journey with D-Robotics
SO008 D-Robotics D-Robotics Returns to embedded world 2026 with the Theme "Intelligence Embedded, Robotics Reborn" - D-Robotics
SO009 D-Robotics Robotics Startups: It's Time to Join the DGP Program and Shape the Future with D-Robotics - D-Robotics
SO010 D-Robotics From RDK X5 to Your Robot: Join Make: & D-Robotics' Live Show & Tell - D-Robotics
SO011 D-Robotics Building a Voice-Controlled Robot Using OpenAI Realtime API: A Full-Link Implementation from RDK X5 to ES02 - D-Robotics
SO012 D-Robotics NodeHub Intelligent Robotics Application Center - Developer Platform for Robot Enthusiasts
SO013 D-Robotics RDK Studio — AI-Native Workbench for Robotics
SO014 D-Robotics D-Robotics RDK Suite FAQ
SO015 D-Robotics RDK X5 hardware introduction
SO016 D-Robotics D-Robotics RDK Kit | RDK DOC
SO017 D-Robotics Introduction to TogetheROS.Bot | RDK DOC
SO018 GitHub D-Robotics organization on GitHub
SO019 GitHub GitHub - D-Robotics/rdk_doc
SO020 LinkedIn D-Robotics | LinkedIn
SO021 Discord 🤖D-Robotics Community
SO022 YouTube D-Robotics - YouTube
SO023 Horizon Robotics About Us_Smart Driving Technology_Smart Driving Company-HorizonRobotics
SO024 Yicai Global China's D-Robotics Raises USD150 Million in New Funding, Series B Total Hits USD270 Million
SO025 CXO Digitalpulse D-Robotics raises $150 million to expand embodied AI and global robotics ecosystem
SO026 CNX Software LooperRobotics Insight 9 standalone spatial AI camera features D-Robotics RDK X5 SoC, supports ROS 2 (Crowdfunding) - CNX Software
SO027 Switch Science RDK S100 Developer Kit
SO028 Yicai Global China’s Private Equity Divide Deepens as Capital Floods Hard Tech, Flees Once-Hot Sectors
SM001 D-Robotics D-Robotics homepage
SM002 D-Robotics RDK S100 Development Kit - Advanced AI Computing Platform
SM003 D-Robotics RDK X5 Development Kit - 10 TOPS Computing
SM004 D-Robotics Introduction to TogetheROS.Bot | RDK DOC
SM005 Horizon Robotics About Us_Smart Driving Technology_Smart Driving Company-HorizonRobotics
SM006 FinOps Foundation FinOps Foundation - What is FinOps?
SM007 FinOps Foundation FinOps Framework Overview
SM008 Microsoft Learn What is FinOps? - Cloud Computing
SM009 Microsoft Learn FinOps documentation - Cloud Computing
SM010 Karpenter Karpenter
SM011 Karpenter Documentation
SM012 Google Cloud Well-Architected Framework: Cost optimization pillar
SM013 Google Cloud GKE Autopilot overview
SM014 Red Hat Cost management for Kubernetes on Red Hat OpenShift
SM015 CNCF Cloud Native 2024: Approaching a Decade of Code, Cloud, and Change
SM016 CNCF Reports
SM017 MarketsandMarkets Cloud FinOps Market report page
SM018 The Business Research Company Kubernetes Cost Management Market Share, Size Report 2026
SM019 Verified Market Reports Global Cloud Cost Management and Optimization Market Size, Industry Share, Growth Trends & Forecast 2026-2034
SM020 StartUs Insights Top 10 Robotics Trends [2026] | StartUs Insights
SM021 Ambarella Computer Vision Camera SoCs for AIoT, Industrial & Robotics
SM022 Ambarella AIoT, Industrial & Robotics
SM023 Renesas Robotics Solutions - AI Vision & Precision Motor Control
SM024 Renesas Humanoid Robots
SM025 Texas Instruments What will it take to bring humanoid robots into the real world?
SM026 MediaTek MediaTek Genio | IoT Platform for Smart Devices
SP001 D-Robotics D-Robotics | Expand the Intelligent Capabilities of Robot
SP002 D-Robotics RDK X5 Development Kit - 10 TOPS Computing
SP003 D-Robotics RDK S100 Development Kit - Advanced AI Computing Platform
SP004 D-Robotics Introduction to TogetheROS.Bot | RDK DOC
SP005 GitHub D-Robotics
SP006 GitHub GitHub - D-Robotics/rdk_doc
SP007 Horizon Robotics About Us_Smart Driving Technology_Smart Driving Company-HorizonRobotics
SP008 D-Robotics Join the D-Robotics Community | Collaborate & Innovate
SP009 D-Robotics D-Robotics Developer Community
SP010 D-Robotics D-Robotics Developer Community
SP011 NVIDIA NVIDIA Jetson AGX Orin
SP012 NVIDIA NVIDIA Jetson Thor
SP013 CNX Software Qualcomm Robotics RB5 Platform Targets the Development of 5G and AI-Enabled Robots - CNX Software
SP014 CNX Software Qualcomm RB3 Gen 2 Platform with Qualcomm QCS6490 AI SoC targets robotics, IoT and embedded applications - CNX Software
SP015 Ambarella Computer Vision Camera SoCs for AIoT, Industrial & Robotics | Ambarella
SP016 Ambarella AIoT, Industrial & Robotics | Ambarella
SP017 Renesas Robotics Solutions - AI Vision & Precision Motor Control
SP018 Renesas Humanoid Robots
SP019 Texas Instruments What will it take to bring humanoid robots into the real world? Read our experts’ insights | TI.com
SP020 MediaTek MediaTek Genio | IoT Platform for Smart Devices
SP021 Google Cloud GKE Autopilot overview
SP022 Karpenter Karpenter
SP023 StartUs Insights Top 10 Robotics Trends [2026] | StartUs Insights
SP024 Microsoft Learn FinOps toolkit overview - Cloud Computing
SP025 CNX Software LooperRobotics Insight 9 standalone spatial AI camera features D-Robotics RDK X5 SoC, supports ROS 2 (Crowdfunding) - CNX Software
SP026 Switch Science RDK S100 Developer Kit
SP027 Yicai Global China's D-Robotics Raises USD150 Million in New Funding, Series B Total Hits USD270 Million
SP028 NVIDIA NVIDIA Embedded Systems for Next-Gen Autonomous Machines
SP029 Ambarella Advanced Imaging Chipset Technology for Intelligent Cameras | Ambarella
SP030 CNX Software Qualcomm QRB5165 system-on-module is made for robotics applications - CNX Software
SI001 36Kr 从地平线拆分后,「地瓜机器人」不做本体,想做具身智能“卖水人” 我们不做本体,这是很明确的。
SI002 Baidu Baike 地瓜机器人 2026年3月16日,地瓜机器人宣布近期完成1.2亿美元B1轮融资。
SI003 D-Robotics RDK X5 产品页
SI004 D-Robotics RDK S100 产品页
SI005 D-Robotics TogetheROS.Bot 用户手册
SI006 D-Robotics GitHub rdk_model_zoo
SI007 D-Robotics GitHub D-Robotics organization Showing 10 of 229 repositories.
SI008 D-Robotics Community 课程首页接口
SI009 D-Robotics Community 推荐课程接口
SI010 D-Robotics Community 近期活动接口
SI011 D-Robotics Community 合作伙伴接口
SI012 D-Robotics Community 开发者案例接口
SI013 D-Robotics Forum latest.json
SI014 CSDN RDK X5超新手入门教程:从系统烧录到yolov5物体识别
SI015 CSDN RDK X5 从零上手:烧录系统镜像 + Wi-Fi 配网 + SSH 远程登录 完整教程
SI016 Ambarella Investor Relations Annual Reports and Proxies
SI017 Ambarella Investor Relations SEC Filings
SI018 NVIDIA Investor Relations Annual Reports and Proxies
SI019 NVIDIA Investor Relations SEC Filings
SI020 NVIDIA Jetson Modules
SI021 Ambarella Homepage
SI022 Leju Robotics 乐聚官网
SI023 腾讯新闻 乐聚智能以价换量策略如何影响行业竞争格局?
SI024 知乎专栏 20天连融两轮、B轮累计近20亿,地瓜机器人凭什么?
SI025 D-Robotics Docs RDK X5 hardware introduction
SE001 D-Robotics RDK X5 English product page
SE002 D-Robotics RDK X5 product page
SE003 D-Robotics Docs RDK X5 hardware introduction
SE004 D-Robotics RDK S100 product page
SE005 D-Robotics Docs RDK S100 hardware introduction
SE006 D-Robotics Docs TogetheROS.Bot manual
SE007 D-Robotics GitHub D-Robotics organization
SE008 D-Robotics GitHub rdk_model_zoo
SE009 D-Robotics GitHub rdk_model_zoo rdk_x5 branch
SE010 D-Robotics Community Recommended course feed
SE011 D-Robotics Community Latest course feed
SE012 D-Robotics Community College home feed
SE013 D-Robotics Forum latest.json
SE014 D-Robotics Forum forum homepage
SE015 CSDN RDK X5超新手入门教程:从系统烧录到yolov5物体识别
SE016 CSDN RDK X5 从零上手:烧录系统镜像 + Wi-Fi 配网 + SSH 远程登录 完整教程
SE017 Baidu Baike 地瓜机器人
SE018 36Kr 从地平线拆分后,「地瓜机器人」不做本体,想做具身智能“卖水人”
SE019 D-Robotics Community Recent activity feed
SE020 D-Robotics Community Cooperative partner feed
SE021 D-Robotics Community Developer cases feed
SE022 D-Robotics Community Featured developer cases
SE023 D-Robotics DGP program page
SE024 Leju Robotics Company introduction
SE025 腾讯新闻 乐聚智能以价换量策略如何影响行业竞争格局?
SE026 NVIDIA Jetson Modules
SE027 知乎专栏 20天连融两轮、B轮累计近20亿,地瓜机器人凭什么?
SE028 D-Robotics RDK X3 product page
SE029 D-Robotics RDK X3 developer portal page
SE030 Waveshare D-Robotics RDK X3 module spec sheet
SE031 KrASIA D-Robotics unveils upgrades to tighten the loop between cloud and edge robotics
SE032 Robotics Observer D-Robotics announces upgrades that tighten the loop from cloud to edge robotics
SE033 D-Robotics NodeHub intelligent robotics application center
SE034 D-Robotics RDK Studio — AI-Native Workbench for Robotics
SE035 D-Robotics GitHub RDK Studio doc: supported hardware RDK Studio 对 RDK 系列开发板支持最完整,同时也允许通过 SSH 接入普通 Linux 主机、Jetson、Raspberry Pi、Rockchip 等设备。
SU001 D-Robotics Robotics Startups, It's Time to Join the DGP Program and Shape the Future with D-Robotics We've witnessed over 100+ robotic product categories built on our platform and have supported 200+ early-stage robotics companies worldwide.
SU002 D-Robotics Hengbot Launches Sirius on Kickstarter: A Programmable Robotic Dog with D-Robotics' RDK X3 as Its Brain for Consumers an AI head powered by D-Robotics' RDK X3, delivering up to 5 TOPS of edge computing power
SU003 D-Robotics myCobot 280 RDK X5: A 6 DOF Robotic Arm with RDK X5 Power The RDK X5 boasts an outstanding community ecosystem, bringing together nearly 100,000 developers.
SU004 Hengbot Hengbot - Official Homepage $912,768 USD, a milestone in robotic companionship, 1150 Backers
SU005 Narwal Narwal - Official Homepage Today, Narwal has entered over 30 markets ... 30+ Countries and Markets, 450M+ Global Users
SU006 Insta360 Insta360 - Official Homepage
SU007 NavBot NavBot - Official Homepage NVIDIA Jetson Powered - Edge AI on every platform
SU008 D-Robotics D-Robotics Community - Robotics Evangelists We Collaborate with Robotics Evangelists Worldwide: Frank Fu, CEO of Navbot
SU009 KrASIA D-Robotics Unveils Upgrades to Tighten the Loop Between Cloud and Edge Robotics the company has partnered with more than 60 players across the robotics supply chain ... supporting Narwal on the Xiaoyao 002 robot vacuum ... enabling Insta360's launch of ... the Antigravity A1; and working with Vbot on a quadruped companion robot.
SU010 Gasgoo Beijing Innovation Center of Humanoid Robotics' Tiangong 3.0 to Enter Mass Delivery in H2 2026 powered by D-Robotics's Xuri S600 high-computing chip for embodied intelligence
SU011 Shuziqushi Tiangong 3.0 Humanoid Robot Set for Mass Delivery in Late 2026 Using D-Robotics Chip the total machine cost of Tiangong 3.0 is projected to fall by more than 50%
SU012 China Biz Insider D-Robotics Enters Consumer Drone Market with AI Chip in New Antigravity A1 powered by D-Robotics' new-generation Sunrise 5 intelligent computing chip
SU013 Insta360 Antigravity Launches A1, 8K 360 Drone Antigravity A1 is unlike anything we've helped bring to market, and it's set to revolutionize drone flight
SU014 China Biz Insider 15 Embodied AI Unicorns in 6 Months: China's Robot Race Hits a Reality Check D-Robotics ... is attacking the problem from the silicon layer, supplying compute chips and operating systems to robot manufacturers ... if embodied AI scales, D-Robotics captures value regardless of which application-layer winner emerges.
SU015 Yicai Global China's D-Robotics Raises USD150 Million in New Funding, Series B Total Hits USD270 Million shipments jumped 180 percent last year from the previous year, while its customer base grew 200 percent ... developer community exceeding 100,000 users
SU016 36Kr 36Kr Interview with D-Robotics CEO Wang Cong on Chip Strategy and Robot Categories 地瓜机器人跑量不靠人形和具身智能,靠的是扫地机、割草机,这些产品可以帮助我们把机器人的基本盘、影响力建立起来
SU017 D-Robotics D-Robotics Developer Portal - Cooperative Partner Directory (API) 9 named cooperative partners including Leju Robotics (乐聚机器人) and Hengbot (恒之未来)
SU018 D-Robotics About D-Robotics collaborations with premier global OEMs and partners across diverse sectors including humanoid robots, companion robots, robotic vacuum cleaners, and robotic lawn mowers ... millions of users worldwide
SU019 D-Robotics From RDK X5 to Your Robot: Join Make: & D-Robotics Live Show & Tell
SU020 D-Robotics Robotics Dream Keeper Challenge: Newbie-Friendly, Win $1,000
SU021 CNMRA Global Unicorn Index 2026: 32 Chinese Robotics Companies Make the Ranking Narwal ... ensuring stable commercial revenue streams
SU022 D-Robotics D-Robotics RDK X3
SU023 D-Robotics D-Robotics Developer Portal - Developer Case Showcase (API)
SU024 D-Robotics D-Robotics Developer Portal - Recent Community Activity (API) 全国大学生智能汽车竞赛地瓜机器人创意组智慧医疗挑战赛报名通知 ... activityEndTime 2026-08-31
SU025 D-Robotics D-Robotics Developer Portal - College/Course Home (API)
SU026 Kickstarter Sirius - Hengbot Robotic Dog Campaign Page
SR001 Bureau of Industry and Security Guidance Regarding Enforcement of License Requirements for Advanced Computing Items for Entities Headquartered in Country Group D:5 and Macau A license requirement continues to apply ... to all destinations outside the United States for these advanced computing items when such items are for entities headquartered in, or whose ultimate parent company is headquartered in, Country Group D:5 or Macau.
SR002 WilmerHale US Export Controls on AI Diffusion Officially Paused But New Guidance Elevates Risk for AI Related Exports The newly issued guidance documents suggest that ... BIS will likely increase enforcement of AI-related exports under still-applicable authorities.
SR003 WilmerHale BIS Issues Export Control Compliance Requirements for Semiconductor Fabricators After April 13, 2026, a company must meet all three requirements and have also submitted an application to become an “approved” IC designer.
SR004 NVIDIA Jetson Modules, Support, Ecosystem, and Lineup Jetson Thor™ series modules ... deliver up to 2070 FP4 TFLOPS of AI compute and 128 GB of memory.
SR005 NVIDIA NVIDIA JetPack Software Stack JetPack 7 ... is purpose-built to enable cutting-edge robotics and generative AI at the edge.
SR006 Qualcomm Qualcomm RB3 Gen 2 development kit at MIT Reality Hack Qualcomm RB3 Gen 2 development kit at MIT Reality Hack.
SR007 Qualcomm Powering IOT developers with edge AI: Qualcomm RB3 Gen 2 Kit is now supported in the Edge Impulse platform Qualcomm RB3 Gen 2 Kit is now supported in the Edge Impulse platform.
SR008 Qualcomm Qualcomm Dragonwing QCS6490 Qualcomm Dragonwing™ QCS6490.
SR009 Qualcomm Qualcomm Robotics RB5 Platform Press Kit Qualcomm Robotics RB5 Platform Press Kit.
SR010 Ambarella The next generation of AI-enabled robotics Compared to existing GPUs and FPGAs, Ambarella SoCs offer higher computer vision performance per watt.
SR011 Kneron About us | Kneron – Full Stack Edge AI Backed by leading global investors ... with $190 million in funding, Kneron is driving the next generation of edge intelligence.
SR012 D-Robotics Find D-Robotics Distributors | Robotics Solutions Near You Find RDKs for sale worldwide through our trusted distributors.
SR013 D-Robotics Advanced Robotics Hardware and Accessories Official RDK accessories are available from our local distributors.
SR014 D-Robotics RDK X3 Development Kit - 5 TOPS AI Computing 100+ Robot Development Kits Accelerate Product Assembly.
SR015 Horizon Robotics Investor Relations 2025年公司实现营收达37.6亿元人民币.
SR016 D-Robotics Forum latest.json
SR017 D-Robotics About D-Robotics | Contact Us
SR018 D-Robotics RDK X5 product page
SR019 D-Robotics RDK S100 product page
SR020 D-Robotics DGP program page
SR021 D-Robotics D-Robotics Developer Community
SR022 D-Robotics RDK X5 hardware introduction
SR023 D-Robotics D-Robotics RDK Kit | RDK DOC
SR024 D-Robotics GitHub D-Robotics organization
SR025 D-Robotics GitHub rdk_model_zoo
SR026 36Kr 从地平线拆分后,「地瓜机器人」不做本体,想做具身智能“卖水人”
SR027 Horizon Robotics About Us_Smart Driving Technology_Smart Driving Company-HorizonRobotics
SR028 Yicai Global China's D-Robotics Raises USD150 Million in New Funding, Series B Total Hits USD270 Million
SR029 CXO Digitalpulse D-Robotics raises $150 million to expand embodied AI and global robotics ecosystem
SR030 Baidu Baike 地瓜机器人
SR031 Bureau of Industry and Security Commerce Further Restricts China's Artificial Intelligence and Advanced Computing Capabilities BIS added 80 entities to the Entity List from China ... for engaging in the development of advanced AI, supercomputers, and high-performance AI chips for China-based end-users with close ties to the country's military-industrial complex.
SR032 Bureau of Industry and Security Commerce Strengthens Restrictions on Advanced Computing Semiconductors to Enhance Foundry Due Diligence and Prevent Diversion Impose a broader license requirement for foundries and packaging companies seeking to export certain advanced chips, unless one of three conditions is met ... a trusted "Approved" or "Authorized" integrated circuit (IC) designer ... an "Approved" outsourced semiconductor assembly and test services (OSAT) company.
SR033 Congressional Research Service U.S. Export Controls and China: Advanced Semiconductors (R48642) In 2014, the PRC government issued a national semiconductor industrial policy with the stated goal of establishing a world-leading semiconductor industry in all areas of the integrated circuit supply chain by 2030.
SR034 Torres Trade Law New Rules Further Restrict China's Access to Semiconductor Technology One final rule (the "Entity List Updates Rule") adds 140 entities to the BIS Entity List and assigns 16 entities the new Footnote 5 designation ... an interim final rule makes several changes to the Export Administration Regulations, including adding new Foreign Direct Product (FDP) rules.
SR035 International Federation of Robotics (IFR) China Makes AI-Powered Robots Core of National Strategy China has launched its 15th Five-Year Plan by placing robotics at the heart of its modern industrial system ... China's manufacturing industry already has an operational stock of around 2 million units -- approximately 4.5 times more than the global no. 2, Japan. 54% of annual industrial robots installed worldwide were deployed in China.
SR036 TechCrunch Famed roboticist says humanoid robot bubble is doomed to burst Brooks ... is particularly skeptical of companies ... trying to teach robots dexterity by showing them videos of humans doing tasks. In a new essay, he calls this approach "pure fantasy thinking."
SR037 Robotics & Automation News Investors warn AI hype is fueling a bubble in humanoid robotics Data from CB Insights shows that investors' attention inside the AI market is shifting rapidly toward industrial humanoid robotics ... the flood of AI capital is pushing robotics toward a speculative zone, with too many startups promising breakthroughs without commercial evidence.
SR038 ChinaBizInsider 15 Embodied AI Unicorns in 6 Months: China's Robot Race Hits a Reality Check At least 25 Chinese embodied intelligence startups now carry valuations above RMB 10 billion (US$1.39 billion) ... most startups in the cohort carry cash runways of only 18 to 24 months ... a reckoning ... is likely to arrive between 2027 and 2028.
SR039 TrendForce China's CSPs and OEMs likely to Actively Procure H200, but Domestic AI Chip Development Continues to Accelerate China's domestic AI chip market share is projected to reach 50% in 2026 ... Development among Chinese AI chip makers and CSP-designed ASICs continues to be limited by upstream bottlenecks, such as restricted wafer capacity and HBM access.
SV001 CompaniesMarketCap Ambarella (AMBA) - Market capitalization As of July 2026 Ambarella has a market cap of $3.43 Billion USD.
SV002 CompaniesMarketCap Cambricon Technologies (688256.SS) - Market capitalization As of July 2026 Cambricon Technologies has a market cap of $125.60 Billion USD.
SV003 Google Finance Cambricon Technologies Corp Ltd (688256) Stock Price & News Mkt. cap 850.08B.
SV004 StockAnalysis Horizon Robotics (HKG:9660) Stock Price & Overview Market Cap 68.20B -30.2%.
SV005 Kneron Kneron Announces $49M in Additional Funding, Bringing Total Series B to $97M Kneron today announced $49M in strategic funding ... bringing their total Series B financing to $97M.
SV006 Kneron Kneron Closes B Round at $48 Million Leading full-stack edge AI company Kneron today announces new series B funding of $48 million.
SV007 Kneron Kneron Receives US$18 Million Series A1 Financing Kneron ... announced today the completion of their series A1 financing of US$18 million.
SV008 Securities and Exchange Commission EDGAR Search Results for Ambarella 10-K filings 10-K ... Filing Date 2026-03-23.
SV009 Hong Kong Exchanges and Clearing Listed Company Information Title Search Listed Company Information Title Search.
SV010 Google Finance Ambarella Inc (AMBA) Stock Price & News Ambarella Inc (AMBA) Stock Price & News.
SV011 Google Finance Horizon Robotics (9660) Stock Price & News Horizon Robotics (9660) Stock Price & News.
SV012 Horizon Robotics Announcements and Circulars 公告及通函.
SV013 Yicai Global China's D-Robotics Raises USD150 Million in New Funding, Series B Total Hits USD270 Million
SV014 CXO Digitalpulse D-Robotics raises $150 million to expand embodied AI and global robotics ecosystem
SV015 Baidu Baike 地瓜机器人
SV016 36Kr 从地平线拆分后,「地瓜机器人」不做本体,想做具身智能“卖水人”
SV017 D-Robotics About D-Robotics | Contact Us
SV018 D-Robotics RDK X5 product page
SV019 D-Robotics RDK S100 product page
SV020 D-Robotics GitHub D-Robotics organization
SV021 D-Robotics GitHub rdk_model_zoo
SV022 Horizon Robotics About Us_Smart Driving Technology_Smart Driving Company-HorizonRobotics
SV023 Yicai Global China’s Private Equity Divide Deepens as Capital Floods Hard Tech, Flees Once-Hot Sectors
SV024 Kneron About us | Kneron – Full Stack Edge AI Backed by leading global investors ... with $190 million in funding, Kneron is driving the next generation of edge intelligence.
SV025 Bureau of Industry and Security Guidance Regarding Enforcement of License Requirements for Advanced Computing Items for Entities Headquartered in Country Group D:5 and Macau A license requirement continues to apply ... to all destinations outside the United States for these advanced computing items when such items are for entities headquartered in, or whose ultimate parent company is headquartered in, Country Group D:5 or Macau.
SV026 WilmerHale US Export Controls on AI Diffusion Officially Paused But New Guidance Elevates Risk for AI Related Exports The newly issued guidance documents suggest that ... BIS will likely increase enforcement of AI-related exports under still-applicable authorities.
SV027 NVIDIA Jetson Modules, Support, Ecosystem, and Lineup Jetson Thor™ series modules ... deliver up to 2070 FP4 TFLOPS of AI compute and 128 GB of memory.
SV028 NVIDIA NVIDIA JetPack Software Stack JetPack 7 ... is purpose-built to enable cutting-edge robotics and generative AI at the edge.
SV029 Ambarella The next generation of AI-enabled robotics Compared to existing GPUs and FPGAs, Ambarella SoCs offer higher computer vision performance per watt.
SV030 Horizon Robotics Investor Relations 2025年公司实现营收达37.6亿元人民币.