XDOF
面向物理 AI 的机器人数据基础设施
XDOF 已显露真实的早期需求,也拿出了具象的机器人数据成果;但业务仍是重资本的服务与软件结合体,收入能否持续、利润率和安全管控能否站住脚,都尚未验证。
覆盖范围与披露说明
公开资料最能证实 XDOF 的机器人数据工作流、早期客户进展和数据集成果;收入能否持续、客户能否留存以及产品安全管控,仍缺少充分证据。
封面要素
公司概况
XDOF 提供机器人数据基础设施,由 UC Berkeley GELLO 项目的研究者创立。公司把遥操作、人体动作采集、标注、交付工具和评测流程整合起来,为前沿 AI 实验室和机器人开发商采集物理世界训练数据。
- 官网
- xdof.ai
- 创始人
- Philipp Wu, Fred Shentu, Nemo Jin
- 创立地点
- Berkeley, California
- 总部
- San Mateo, California
- 产品
- 面向具身智能系统的外包机器人数据服务,覆盖遥操作、数据标注、数据集工具和评测。
- 客户
- 前沿 AI 实验室和机器人开发商
- 商业模式
- 托管数据采集服务、软件工具辅助的数据交付,以及机器人工作流相关的系统销售或许可授权。
- 阶段
- Series A
- 融资情况
- 已完成 $70M Series A 轮融资,据报道正就约 $1.2B 的 Series B 估值展开洽谈。
执行摘要
主要优势
- 切中具身 AI 数据瓶颈
- 据报道,早期客户进展与付费需求已显现
- ABC-130K 提供了明确的技术实证
主要风险
- 收入持续性与客户留存尚无独立验证
- 运营离不开大量人力和硬件投入
- 治理安排和后续融资仍不明朗
未决问题
- 经审计的收入构成、留存率与利润率数据
- 具名客户及合同期限的核验材料
- 产品安全与客户数据管控证据
目录
01公司概况
1.1 企业定位、公司架构与核心使命
XDOF, Inc. 于 2024 年 10 月成立,源自加利福尼亚州 Berkeley 的学术研究,主要管理办公室位于加利福尼亚州 San Mateo [CO001, CO013]。具身智能发展的核心瓶颈是数据:数字基础模型可利用海量互联网文本和图像,通用实体机器人却缺少真实交互数据 [CO005]。XDOF 自研低成本遥操作工具和人工标注系统,搭建端到端数据流水线,替机器人制造商和前沿 AI 实验室采集物理数据 [CO005, CO006]。公司结合机器人遥操作与第一人称视角穿戴传感器,采集高保真的物理操作示范,帮助前沿实验室训练多任务机器人基础模型,也让它们不必承担自建数据采集设施的高额资本开支 [CO005, CO006]。
| 指标 | 数值 / 状态 | 日期 / 截至时间 | 置信度 | 尽调风险 / 信息缺口 |
|---|---|---|---|---|
| 投后估值 | Series B 正洽谈 ~$1.2B 估值,尚未交割;Series A 估值未公开 | 2026-09 | 中 | 条款和估值仍可能变化;A 轮估值未披露 |
| 累计股权融资 | $70.0M | 2026-06 | 高 | 不含未经确认的种子轮融资估算(部分数据库显示累计融资 $78M) |
| 年化收入运行率 | 据报道接近 $50M | 2026-09 | 低 | 未披露经审计的 GAAP 收入或合同期限;收入可能集中于试点项目 |
| 客户数 | ~20 家活跃付费客户 | 2026-06 | 高 | 客户名称大多未公开;客户流失率和 ACV 未知 |
| 核心团队员工数 | ~60 名全职员工 | 2026-06 | 中 | 不含国际遥操作和传感器采集承包人员 |
| 总部地点 | 加利福尼亚州圣马特奥(创立于加利福尼亚州伯克利) | 2026-09 | 高 | 管理办公室位于圣马特奥,现场运营正向全球扩展 |
指标来自 Series A 披露、公司公告和经核实的科技媒体报道;缺少依据的非公开指标列为尽调缺口。
[CO001, CO002, CO004, CO008, CO009, CO010, CO013]1.2 创始人背景、技术领导力与公司治理
XDOF 于 2024 年 10 月成立,由 UC Berkeley 机器人研究者 Philipp Wu、Fred Shentu 联合创办,Nemo Jin 是技术联合创始人 [CO001]。Philipp Wu 曾在 UC Berkeley 攻读博士,研究机器人如何从大规模示范数据中学习;他认为,缺少开源、多任务的物理操作数据,是具身智能发展的关键障碍 [CO001, CO006]。Wu 与首席技术官 Fred Shentu 共同开发了开源、低成本遥操作接口 GELLO,降低了采集灵巧操作轨迹的成本门槛 [CO006]。XDOF 将这项学术成果商业化并继续拓展 [CO006]。但公司尚未披露正式治理架构;XDOF 和 Series A 投资方都没有公开董事席位分配、独立董事安排或创始人投票权条款,潜在投资者仍需补做重要的治理尽调 [CO014]。
| 人员 | 职务 | 背景 | 创始团队与市场的契合度及职责覆盖 | 关键人物依赖度 |
|---|---|---|---|---|
| Philipp Wu | 联合创始人兼首席执行官 | UC Berkeley 博士研究员,研究机器人如何从大规模物理数据集中学习;GELLO 遥操作项目共同创建者。 | 高:低成本机器人操作数据基础设施的技术先驱;主导技术方向,并负责前沿 AI 实验室客户关系。 | 关键:公司的核心研究者和对外代表;物理 AI 数据供应链愿景的主要提出者。 |
| Fred Shentu | 联合创始人兼首席技术官 | UC Berkeley 机器人研究者;开源 GELLO 遥操作系统共同创建者、GELLO 开创性论文共同作者。 | 高:核心系统工程师,负责适配不同机器人硬件的机械、固件和遥操作架构。 | 高:硬件接口、低延迟遥操作工具和传感器流水线的核心架构负责人。 |
| Nemo Jin | 联合创始人 | 2024 年 10 月公司成立时的联合创始人;参与早期运营和工程架构。 | 中:兼具运营和技术职责,支持早期平台基础设施建设与部署。 | 中:承担工程职能;随着技术团队扩大,公司对个人的依赖有所分散。 |
此处列出经核实的创始团队;COO、CFO 等其他高管任命尚未公开宣布。
[CO001, CO006, CO010]1.3 股权融资历程、估值演进与资方财团
XDOF 一直低调研发,直到 2026 年 6 月 17 日才宣布完成 $70 million Series A;这是目前公开记录中公司唯一一轮机构股权融资 [CO002]。本轮投资方包括 Thrive Capital、Andreessen Horowitz (a16z)、Lux Capital、Spark Capital 和 WndrCo [CO003]。独立风投分析指出,5 家一线机构共同参与首轮融资也带来后续融资风险:没有哪一家持有主导性股权,也没有单一投资方承担确保后续资金到位的受托责任 [CO011]。公开亮相后不到 3 个月,风投机构的兴趣便推动 XDOF 进入 Series B 后期谈判;拟议估值约 $1.2 billion,由 8VC 领投 [CO004]。但截至 2026 年 9 月,交易尚未完成,条款仍在谈判;在商业指标得到验证前,估值仍可能大幅波动 [CO004, CO014]。
| 利益相关方 | 角色 | 控制权 / 经济影响 | 尽调问题 |
|---|---|---|---|
| Thrive Capital | A 轮投资方 | 主要机构投资者;一线 AI 投资方(投资过 OpenAI、Stripe),可带来信誉和客户资源。 | 核实清算优先权、董事会席位和后续跟投承诺。 |
| Andreessen Horowitz (a16z) | A 轮投资方 | 领先的风险投资机构,重点投资机器人、AI 基础设施和生物物理系统。 | 核查信息权、跟投权和战略客户引荐条款。 |
| Lux Capital | A 轮投资方 | 专注前沿技术和深科技,深耕物理科学与机器人领域的风险投资基金。 | 确认按比例跟投权、技术顾问职责和老股交易权。 |
| Spark Capital | A 轮投资方 | 知名软件和企业科技投资机构,参与 Series A 联合投资。 | 评估联合表决协议,以及各方在后续融资中的利益是否一致。 |
| WndrCo | A 轮投资方 | 科技与媒体控股投资公司,由 Jeffrey Katzenberg 创立,参与种子轮和 Series A 联合投资。 | 确认经济权益比例、战略媒体或合作角色,以及董事会观察员席位。 |
此处列出已披露的 XDOF Series A 机构投资方;未列入个人天使投资人及未披露的种子轮参与者。
[CO002, CO003, CO011, CO014]1.4 运营里程碑、数据集分发与业务规模
2026 年 6 月公开亮相后,XDOF 与 UC Berkeley AI Research 实验室联合发布开源操作数据集 ABC-130K,收录 195 项双臂操作任务、超过 130,000 条轨迹,提升了公司在具身智能社区的技术公信力 [CO007]。公司称,发布初期约有 20 家活跃付费客户,其中包括数家未具名的前沿 AI 实验室 [CO008]。据报道,截至 2026 年 9 月,公司年化收入运行率接近 $50 million,核心团队约 60 人 [CO009, CO010]。为承接增长的客户需求,XDOF 已在北美、拉美和东南亚招聘,扩充远程遥操作员及现场数据采集员队伍 [CO010]。
XDOF 从 UC Berkeley 学术研究起步,随后走出隐身、完成 Series A 融资并拓展商业化。
[CO001, CO002, CO004, CO006, CO007, CO008, CO010, CO013]1.5 尽调评估、治理风险与未决缺口
XDOF 位于基础模型与实体机器人发展的交汇点,战略位置有利,但仍有几项重要尽调问题待解 [CO009, CO012]。最突出的是,据报道接近 $50 million 的年化收入只是第三方媒体估算,尚无经审计的 GAAP 财务数据、客户留存率或毛利率披露 [CO009]。公司需要在全球管理分散的遥操作员和硬件传感器,运营成本负担沉重;若定价能力减弱,单位经济效益可能恶化 [CO011, CO012]。此外,董事会治理、投票权安排和领投方后续出资承诺都不透明;评估后期股权估值倍数前,投资者必须完成法律核查 [CO011, CO014]。
1.6 重点图表
02市场分析
2.1 具身智能基础设施与市场边界界定
物理交互基础模型兴起,正在改变商用机器人软件市场。前沿 AI 实验室擅长训练网络规模的数字语言模型,但实体机器人需要密集、多模态的交互数据,公开互联网无法提供这类数据。这也划定了具身智能数据基础设施的市场范围:外包遥操作系统、第一人称人体传感器数据流,以及标准化操作轨迹标注。通用文本抓取、传统机器人硬件制造和确定性工业可编程逻辑控制器(PLC)工程不在其中。XDOF 等专业供应商为实体机器人搭建端到端数据供应链,让基础 AI 实验室和人形机器人开发商不必投入大量资金、精力自建物理数据采集体系。[CM001, CM002, CM005, CM006, CM015]
| 细分市场 | 纳入的支出 | 不纳入的支出 | 买家 / 付款方 | 对 XDOF 的意义 |
|---|---|---|---|---|
| 前沿 AI 机器人基础设施 | 外包遥操作设备、高自由度传感器采集、多模态轨迹标注 | 数字文本抓取、通用网络爬取、大语言模型人类反馈强化学习 | 前沿 AI 实验室和基础机器人开发商 | 核心市场;定制数据集采集,客单价高 |
| 工业自动化与制造 | 特定工厂装配示范、缺陷操作数据集、定制机械臂工具 | 标准工业 PLC 编程、硬件资本开支、机床租赁 | 一线工业设备制造商和汽车整车厂 | 次级扩张市场;定制需求较少 |
| 人形与服务机器人 | 双臂操作轨迹、家务任务传感器数据流、第一人称视角动作采集 | 现成执行器部件、电池组、底盘制造 | 人形机器人平台和服务机器人初创公司 | 直接切入的市场;对通用操作数据有迫切需求 |
| 家用消费级自动化 | 家务任务轨迹库、洗碗和洗衣示范数据流 | 消费级家电制造、零售分销、保修服务 | 消费机器人初创公司和家用自动化开发商 | 长期探索性 TAM;生产部署仍处起步阶段 |
此分类划定物理训练数据基础设施与更广义机器人硬件、数字软件支出的边界;纳入的支出类别截至 2026 年已有采购证据。
[CM001, CM002, CM006, CM009]2.2 不同口径下的可触达市场规模
估算物理机器人训练基础设施的市场,不能把宏观预测和企业眼下的实际预算混为一谈。自上而下的预测认为,未来 10 年,随着人形和服务机器人进入商业物流与消费场景,更广义的具身 AI 训练和自动化市场可能扩大至 $12.5B 至 $18.0B。近期尽调则应聚焦规模较小的可服务市场(SAM):$1.8B 至 $2.5B,代表前沿 AI 实验室和资金充足的机器人基础模型团队当前的数据采购预算。近期可获取市场(SOM)的年化合同需求约为 $150M 至 $250M。据报道,XDOF 公开亮相后 3 个月内,年化收入运行率就接近 $50 million,显示其迅速拿下客户预算;但投资者仍需衡量巨额首轮融资带来的高估值门槛。[CM003, CM004, CM008, CM011, CM014, CM016]
| 市场层级 | 基准年份 | 覆盖区域 | 市场规模 | 增速 | 测算依据 | 置信度 | 主要尽调局限 |
|---|---|---|---|---|---|---|---|
| 全球物理 AI 数据 TAM | 2026 | 全球 | $12.5B - $18.0B | 38.5% YoY | 结合具身 AI 硬件落地进度,自上而下估算物理 AI 数据供应链 | 低 | 把探索性家用自动化支出与企业机器人支出合并 |
| 前沿机器人训练 SAM | 2026 | 北美与欧洲 | $1.8B - $2.5B | 52.0% YoY | 以 ~40 家活跃基础模型开发商的前沿 AI 实验室数据采购预算为基础,自下而上测算 | 中 | 假设企业会持续选择外包,而不是自建专用采集队伍 |
| 可触达远程操作 SOM | 2026 | 美国及国际枢纽 | $150M - $250M | 65.0% YoY | 根据 ~50 家合格目标客户的供应商合同规模($1M-$5M/yr),估算近期可获取份额 | 中 | 受人工操作员扩张瓶颈及遥操作硬件供应能力限制 |
| XDOF 当前商业化进展 | 2026 | 全球 | ~$50M ARR 运行率 | 快速增长 / 走出隐身模式 | 据报道,该年化收入运行率来自 ~20 个活跃企业及前沿实验室客户账户 | 高 | 非上市公司的运行率在 Series B 融资洽谈期间据报道披露;数据未经审计 |
数值综合了截至 2026 年 Q3 的市场测算和据报道的私募融资运行率;此处将宽泛的自上而下 TAM 与近期可服务市场区分开来。
[CM003, CM004, CM008]从宏观机器人自动化支出逐层收敛至近期可服务的遥操作收入,估算物理 AI 数据基础设施市场。
各层宽度仅表示层级,不代表数值比例。
顶层 TAM 代表物理 AI 训练数据的宏观预测;中层 SAM 和 SOM 反映基于活跃研究团队与平均合同价值的自下而上测算。
[CM004, CM008, CM014, CM016]2.3 客户类型、采购决策与预算来源
专业物理数据采集的主要买家有两类:前沿 AI 基础模型实验室,以及自主研发人形机器人的企业;工业自动化和物流履约运营商则是新兴的次级客群。实验室通常由具身智能负责人(Head of Physical AI)或研究副总裁(VP of Research)推动采购,使用集中管理的基础模型研发与训练预算。商用人形机器人开发商的预算通常由工程副总裁(VP of Engineering)或自主系统副总裁(VP of Autonomy)掌握;他们希望加快灵巧操作能力的开发,又不耽误机械硬件进度。合同通常包括多任务轨迹交付许可和持续的平台支持,结合按量交付数据与企业软件订阅。合作能否深入,还取决于服务等级协议(SLA)能否保证交付、传感器标定是否严格,以及任务规格能否适配客户各自的机器人构型。[CM002, CM003, CM009, CM010, CM013]
| 客户类别 | 目标买家 | 最终用户 | 付款方 | 主要工作流 | 预算负责人 | 采购动因 |
|---|---|---|---|---|---|---|
| 前沿 AI 基础模型实验室 | 物理 AI 副总裁 / 机器人研究负责人 | 研究科学家和基础模型训练工程师 | 中央 AI 研究与算力预算 | 多任务双臂操作训练和策略预训练 | 首席技术官 / 研究副总裁 | 下一代基础模型缺少高质量操作轨迹 |
| 商业化人形机器人开发商 | 自主系统副总裁 / 操作算法负责人 | 机器人软件与控制算法开发人员 | 自主系统工程部 | 遥操作运动示范与灵巧末端执行器训练 | 工程副总裁 | 内部实体硬件设备难以扩容,无法采集数百万小时的任务数据 |
| 工业 OEM 自动化部门 | 先进制造 / 自动化总监 | 工厂系统集成工程师和自动化专家 | 工程资本开支和先进制造预算 | 精密装配与多品类分拣优化 | 工厂运营部门 / 制造副总裁 | 传统启发式机器人自动化故障率高、运行不稳定 |
| 物流与履约运营商 | 履约机器人负责人 / 供应链自动化负责人 | 仓储机器人操作员、机器人车队管理工程师 | 运营技术创新预算 | 采集第一人称视角下的拣货、打包和包裹分拣轨迹 | 首席运营官 / 供应链副总裁 | 不必针对每个 SKU 单独编程,就能处理形态各异的商品 |
采购模式综合行业招聘动向、职位描述和客户画像数据判断;预算归属按企业部署物理 AI 的常见架构归纳。
[CM002, CM009, CM010, CM013]2.4 采用驱动力、结构性瓶颈与尽调风险
高质量物理操作数据极度稀缺,是通用人形机器人落地面临的最大技术障碍,也推动了市场需求。但行业顺风并不能消除落地约束和尽调风险。Altis 的独立研究指出,一些企业买家认为第三方遥操作数据不如内部采集的数据,也认为不同采集服务商之间切换成本不高。扩大物理数据采集还要管理庞大的分布式操作员队伍;在拉美和东南亚设点会增加运营复杂度、硬件维护开支和跨国用工管理难度。Mecka AI、Encord 等资金充足的对手正在进入,Scale AI 等数字标注龙头也在跨界扩张。若 XDOF 建不成专有的数据质量和合成能力壁垒,标准化示范采集就可能陷入低价竞争。[CM001, CM007, CM011, CM012, CM015]
2.5 重点图表
03竞争格局
3.1 竞争格局总览与同业分层
XDOF 面对三类竞争者:具身智能数据服务商、数字人工数据标注龙头,以及前沿 AI 实验室的内部采集团队。Mecka AI、Encord 等初创公司专攻机器人数据流水线和多模态标注。Scale AI(2024 年以 $13.8 billion 估值融资 $1 billion)、Micro1 等数字数据标注公司资金雄厚,也在从大语言模型数据拓展至实体机器人场景。风投支持的人才平台 Mercor 据报道在 2025 年底估值达到 $10 billion,反映投资者对人工数据众包的兴趣。对 XDOF 威胁最大的替代方案仍是基础模型开发者自己采集数据:这些团队通常先部署自有遥操作设备,再评估外部数据供应商。[CP001, CP002, CP003, CP004, CP005]
| 竞争对手 | 类别 | 规模 / 融资 | 目标细分市场 | 核心差异化 | 战略局限 |
|---|---|---|---|---|---|
| XDOF | 物理 AI 数据基础设施 | $70M A 轮(累计融资 ~$78M);~$50M ARR(2026 年) | 前沿 AI 实验室与人形机器人开发商 | GELLO 遥操作技术渊源、ABC-130K 双臂数据集和全球采集团队 | 运营扩张仍处早期,客户切换成本不高 |
| Mecka AI | 机器人数据服务商 | 风险融资金额未披露(早期阶段) | 机器人研究团队与具身 AI 开发商 | 专做实体机器人的演示数据采集 | 资本规模小于 XDOF,公开数据集的验证案例也更少 |
| Scale AI | 企业 AI 数据平台 | 融资 $1B,估值 $13.8B(2024 年 5 月) | 企业基础模型实验室与政府 AI 项目 | 资金充足、企业 GTM 成熟,标注工具链完善 | 过去主要做数字 LLM 和视觉数据,而非机器人遥操作硬件 |
| 实验室自建(DIY) | 内部自用遥操作 | 前沿实验室自有研发预算 | 内部基础模型研究与自研机器人项目 | 知识产权完全自有,传感器设备可定制,轨迹采集也能按需设计 | 运营开销高,还会分散管理团队投入核心模型的精力,资金需求也大 |
规模和融资数据综合截至 2026 年已披露的风投轮次及据报道的商业年化收入;非上市竞争对手的财务条款仍属机密。
[CP001, CP003, CP004, CP009, CP011, CP012]3.2 技术能力矩阵与数据采集差异化
具身智能模型需要结构化的真实交互数据,公开网络上无法抓取。XDOF 的差异化来自 UC Berkeley GELLO 项目:公司把低成本遥操作机械臂与人体第一人称穿戴传感器结合起来,采集日常居家操作数据。这套技术也支撑了 XDOF 与 UC Berkeley AI 研究实验室联合发布 ABC-130K,收录 195 项双臂操作任务、超过 130,000 条轨迹。数字数据标注龙头则主要提供文本和视觉标注软件,没有整合硬件遥操作。为维持数据产能,XDOF 在美国、墨西哥和印度尼西亚运营跨国采集与视频审核网络,并执行结构化质控;规模较小的同类初创公司仍主要在本地运营。[CP006, CP007, CP008, CP012]
| 能力领域 | 评估标准 | XDOF | Scale AI / Micro1 | Mecka AI / Encord | 实验室内部 DIY |
|---|---|---|---|---|---|
| 硬件遥操作 | 低成本远程机械臂控制设备 | 已展示(GELLO 架构、双臂控制) | 受限(主要提供软件标注工作流) | 已展示(实体机器人数据采集设备) | 已展示(实验室定制自研设备) |
| 第一人称视角数据采集 | 采集员佩戴身体传感器记录日常家务 | 已展示(传感器用于折叠、铺平和操作任务) | 未见公开证据(主要处理数字屏幕 / 文本任务) | 部分具备(多模态采集能力刚起步) | 已展示(由内部研究人员采集数据) |
| 公开基准数据集 | 经公开审计的多任务机器人操作数据集 | 已展示(ABC-130K,与 UC Berkeley AI Research 合作) | 未见公开证据(数据集为商业专有) | 受限(研究任务基准规模较小) | 未见公开证据(实验室专有数据集未公开) |
| 全球操作员网络 | 分布式国际数据采集与质检团队 | 已展示(在美国、墨西哥和印度尼西亚开展部署) | 已展示(规模庞大的全球数字众包网络) | 未见公开证据(精品采集网络仍在起步阶段) | 未见公开证据(团队集中在本土研究实验室) |
能力评级依据公开数据集、技术文档和运营岗位披露;“未见公开证据”表示现有公开文档没有佐证。
[CP001, CP003, CP006, CP007, CP008, CP012]3.3 销售组合与单位经济
机器人数据服务通常采用高客单价企业合同,而非标准化的自助订阅。据报道,2026 年年中 XDOF 约有 20 家客户,年化收入接近 $50 million;据此推算,前沿 AI 实验室客户的平均年合同额(ACV)约为 $2.5 million。公司通过企业直销打包销售软件许可、硬件和持续支持服务,并按功能和轨迹复杂度分级定价。前沿实验室外包数据供应链,可省去招聘、管理全球遥操作团队所需的高额资本开支和运营精力。不过,客户仍高度集中;单条轨迹或每小时遥操作的单位经济效益,则取决于定制谈判的企业主服务协议(MSA)。[CP009, CP010, CP012, CP013]
| 供应商 / 交付模式 | 商业定价结构 | 包含功能 | 折扣与未知项 | 战略启示 |
|---|---|---|---|---|
| XDOF 企业合同 | 企业直签订阅与定制项目收费 | 端到端遥操作、第一人称视角数据采集和标注流水线 | 单条轨迹和操作员小时的具体费率未公开 | 初期 ACV 较高(约 20 个账户对应 ~$50M ARR);客户可能严格审视价格 |
| Scale AI 企业级解决方案 | 按用量分档收费,并授权使用企业平台 | 标注工具、数据整理和质量核验流程 | 机器人专用硬件采集仍按项目定价,未公开报价 | 企业采购方议价能力较强,但专用硬件数据成本更高 |
| 实验室自营 | 实验室内部划拨运营支出(OpEx)和硬件资本支出(CapEx)预算 | 工程团队直接监督、传感器定制集成,知识产权留在内部 | 工程师全成本、操作员招聘和硬件折旧往往没有计入 | 适合早期研发;轨迹达到数百万条后,成本效益转差 |
定价结构根据据报道的 ARR、客户数和公开商业模式归纳;具体合同价目表和阶梯用量折扣仍属商业机密。
[CP003, CP004, CP009, CP010, CP012]3.4 护城河能否持久
XDOF 能否守住长期护城河,取决于能否提高现场采集效率,同时抵御服务商品化。行业分析师的独立专家访谈显示,前沿实验室会确保不同供应商的数据可以互通,因此更换机器人数据流水线的成本不高。资金充足的传统标注巨头也有能力借助雄厚资产负债表和现成的大客户采购渠道,在需求成熟时迅速进入物理数据采集市场。XDOF 凭借 Berkeley 学术渊源和早期进入前沿实验室取得先发优势,但基础模型开发者也可能扩大合成数据模拟,用来补充遥操作示范。要保持差异化,XDOF 必须持续提供高质量轨迹、自研第一人称传感器采集工具,并签下有保护力的企业级服务协议。[CP001, CP002, CP003, CP011, CP014]
比较物理 AI 数据供应商、数字标注老牌企业和内部自建方案的硬件专业度与数据基础设施规模。
| 条目 | X (2.5–8.5) | Y (3–9.5) | 说明 |
|---|---|---|---|
| 1. XDOF | 8.5 | 7 | 具备极高的物理机器人遥操作专业化程度,专用数据基础设施扩张迅速。 |
| 2. Scale AI | 3 | 9.5 | 数字数据平台体量庞大、资产负债表实力雄厚,但在物理硬件遥操作上的积累较浅。 |
| 3. Mecka AI | 8 | 4 | 专注真实场景机器人数据采集,融资规模仍处早期阶段。 |
| 4. Encord | 5 | 5.5 | 多模态 AI 数据标注工具链,在 $60M 风险投资支持下拓展物理 AI 业务。 |
| 5. Micro1 | 2.5 | 6 | 人工数据平台从 LLM 扩展至其他领域,业务仍以远程数字人才和软件任务为主。 |
| 6. 实验室内部 DIY | 7.5 | 3 | 围绕特定专有模型具备深厚技术专长,但受内部研发团队产能限制。 |
坐标代表基于 2026 年已披露数据集规模、硬件遥操作能力及据报道风险融资规模的 1–10 分序数评分。
[CP001, CP002, CP003, CP011, CP014]3.5 重点图表
04财务分析
4.1 收入模式、定价梯度与商业合同机制
XDOF 通过设备直销、软件许可和持续托管数据采集服务,组合销售具身机器人训练基础设施。客户购买遥操作设备,并订阅用于流水线编排、数据摄入和处理的软件。产品按功能分档,自研高自由度操作硬件的价格明显高于标准遥操作套件。虽然公司宣传覆盖制造、物流和家用自动化,但目前企业收入主要来自为前沿机器学习研究机构定制的数据生成合同。[CI001, CI002, CI003]
| 收入流 | 变现方式 | 交付单元 | 当前状态 | 收入质量 | 尽调核实要点 |
|---|---|---|---|---|---|
| 硬件远程操作设备 | 直接销售基于 GELLO 的控制臂和传感采集硬件 | 按台采购 | 正在商业化 | 一次性硬件交易利润率 | 核查物料清单(BOM)毛利率和硬件保修负债准备金 |
| 软件平台许可 | 用于编排数据流水线和遥测的自研软件许可 | 年度或多年期订阅 | 正在商业化 | 高毛利的经常性软件收入 | 核实软件独立 ARR 与捆绑专业服务收入如何划分 |
| 定制数据采集服务 | 为特定操作任务提供人工遥操作和第一人称动作捕捉 | 按轨迹小时计费,或按任务数据集里程碑交付 | 核心收入驱动因素 | 服务交付毛利率浮动,受人力成本影响 | 按客户队列核算每项采集任务的实际毛利率 |
| 支持与持续维护 | 日常运营维护、流水线重新标定和传感器支持 | 年度服务等级协议(SLA) | 正在商业化 | 经常性维护收入的利润率 | 核查 SLA 履约成本、客户续约率和支持团队配置 |
收入分类综合 Preqin 公司档案和公开披露的数据采集信息;硬件销售与数据集服务各自贡献多少收入仍未公开。
[CI001, CI002, CI003, CI004]| 产品层级 | 定价模式 | 据报道的挂牌条款 | 实际定价机制 | 折扣与未知项 | 数据来源参考 |
|---|---|---|---|---|---|
| 核心 GELLO 硬件与工具 | 直接硬件销售 | 按机械臂功能分档定价 | 挂牌价随执行器精度而变 | 批量采购硬件的折扣未披露 | Preqin 档案 |
| 企业数据集合同 | 基于里程碑或交付量的许可 | 定制企业合同($1M-$5M ACV) | 约 20 个账户的实际 ACV 平均约 ~$2.5M | 针对特定实验室的折扣与采购量让利未披露 | Value Add VC、TechCrunch |
| 平台订阅与数据接入 | 按年软件订阅 | 按席位数或接入机器人数量分档收取许可费 | 与数据集采集承诺捆绑 | 未公开平台单独订阅的挂牌价 | Preqin 公司档案 |
| 托管式数据整理与质检 | 按工时和材料计费,或按固定交付收费 | 按通过验证的演示轨迹计价 | 须满足严格的客户准确率 SLA | 返工罚款和审计缺陷扣款均未知 | Built In 招聘记录 |
官方价目表受保密协议保护,须私下协商;相关数据以据报道的年化收入总额除以已披露客户数计算。
[CI002, CI005, CI006, CI007]4.2 商业落地进展、客户集中度与信息不对称
据公开报道,XDOF 公开亮相约 3 个月后的 2026 年 9 月,年化收入已接近 $50 million。收入主要来自约 20 个活跃客户账户,折合平均每户年化合同额接近 $2.5 million。快速获客说明行业迫切需要真实物理操作训练数据,但也使客户集中风险突出。少数前沿 AI 实验室贡献了绝大部分账单;任何一家客户流失、转为自建或改变研究方向,都会直接威胁总收入稳定性。公司未披露私人财务报表、详细净营收留存率(NRR)或各期续约队列。[CI005, CI006, CI009, CI015]
4.3 成本结构、可变人工密集度与单位经济拖累
纯基础模型和软件企业的边际分发成本接近于零,XDOF 的物理数据供应链却高度依赖人力。人工遥操作员需要远程操控 GELLO 机械臂,穿戴传感器的采集员还要亲自完成物理任务,因此运营开支几乎随采集量线性增长。招聘信息显示,公司正扩大在印度尼西亚、墨西哥等低成本地区的数据采集和视频质检中心。跨地区用工降低了直接人力成本,但遥操作硬件维护、高吞吐量云端数据处理和质检审核仍会把毛利率压到传统软件企业水平以下。[CI004, CI008, CI012, CI014]
| 单位经济指标 | 据报 / 估计值 | 评估置信度 | 战略意义 | 投资尽调要核查的事项 |
|---|---|---|---|---|
| 单账户平均收入(ARPU) | 年化 ~$2.5M | 中 | 显示前沿 AI 研究实验室愿意支付较高价格 | 获取逐客户 ACV 分布及合同期限明细 |
| 综合毛利率 | 未披露 | 低 | 判断估值倍数处于 20 多倍中段是否合理,毛利率至关重要 | 索取经审计的营业成本,核对遥操作员薪酬和算力支出 |
| 获客成本(CAC) | 未披露 | 低 | 看 XDOF 在规模小、专业度高的企业买方市场中获客效率如何 | 核查高管销售薪酬、差旅预算,以及创始人主导销售的成本分摊 |
| 净留存率(NRR) | 未披露 | 低 | 衡量数据集采购承诺和软件用量能否随时间扩大 | 检查过去 12 个月的扩张队列和客户流失记录 |
所有为空或未披露的指标,反映公司结束隐身状态仅 3 个月、仍为非上市风投企业的情况;估算以据报道约 20 个活跃客户账户和 $50M 收入运行率为参照。
[CI005, CI006, CI008, CI015]4.4 资金储备、烧钱速度与估值支撑
XDOF 于 2026 年 6 月完成 $70 million Series A,投资方包括 Thrive Capital、Andreessen Horowitz、Lux Capital 和 Spark Capital,为公司补充了主要资金储备。收入迅速增长后,投资人主动递来投资意向书(term sheet);据报道,8VC 正领投推进后期 Series B 谈判,估值约 $1.2 billion。按约 $50 million 年化收入计算,拟议估值对应 20 多倍市销率。要支撑这一高倍数,投资者必须相信 XDOF 不只是周期性的人机协作服务商,而是不可或缺的专有数据基础设施。Series A 后公司现金跑道看起来尚可,但持续扩张全球线下设施会增加资本需求。[CI007, CI010, CI011, CI013]
4.5 重点图表
05产品与技术
5.1 提供机器人数据服务,而非单一机器人产品
目前最清楚的公开描述是:XDOF 为具身智能提供托管数据系统,不是成熟通用机器人的成品。多篇独立报道都把遥操作、人体穿戴传感器、标注和评测列为交付流程。另有报道把数据采集分为三类:在客户目标机器人上采集示范、用低成本遥操作系统采集更通用的数据,以及以第一人称视角记录日常人体活动。招聘信息显示,公司还在开发供客户自助获取数据、探索数据集的产品界面;人形机器人工程岗位则暗示公司也在积累硬件和控制能力。Preqin 进一步称 XDOF 销售面向家用和工业任务的机器人系统及基础模型,但更可靠的工作流报道并未印证这一说法。招聘信息反映计划开展的工作,不等于已验证的商用产品目录;产品成熟度和销售组合仍有待尽调。[CE001, CE002, CE003, CE006, CE015]
| 模块或产品形态 | 主要用户 | 公开材料所能证实的现状 | 差异化假设 | 尽调缺口 |
|---|---|---|---|---|
| 目标机器人遥操作数据 | 机器人实验室或机器人开发者 | 独立报道将其列为价值最高的数据采集层级 | 演示数据直接对应客户计划使用的硬件 | 尚无任务层面的质量、验收、吞吐量或实际成交价证据 |
| 低成本遥操作采集 | 基础模型与机器人研发团队 | 被描述为覆盖更广、适用性更强的数据采集层级 | 有望降低采集成本,同时覆盖更多任务 | 尚无公开基准衡量数据能否迁移到不同机器人本体 |
| 第一人称视角人体动作采集 | 用人类活动数据训练模型的团队 | 独立报道提到,采集员佩戴传感器记录日常任务 | 无需动用目标机器人,也能补充人体动作和物理环境数据 | 传感器套件、同步精度、知情同意管理和数据集权利均未披露 |
| 数据与研究平台 | 客户研究人员与内部数据团队 | 近期重新发布的招聘信息提到自助获取、可视化和探索工具 | 有机会把数据采集采购、数据集查看和交付串起来 | 是否可用、租户模式、API、定价和活跃使用情况均无文档记录 |
| 机器人系统与基础模型 | 家庭及工业自动化用户 | Preqin 公司档案提到这一方向,但尚无其他材料证实产品已交付 | 有望让 XDOF 从数据供应延伸到模型或机器人产品 | 产品名称、可用性、部署情况和收入机制都缺乏证据 |
现状依据截至 2026-09-28 的公开描述和招聘信息;这不是经核实的 SKU 清单,也不是对产品能否量产的评估。
[CE002, CE003, CE006, CE015]5.2 融合现场物理采集与数据交付的运营架构
XDOF 的核心价值主张,是把线下采集和数据基础设施接起来。公开描述的流程从客户提出训练需求开始,经过机器人或人工示范采集,再进入数据摄入、标注、质检、数据集工具、交付和评测。招聘岗位要求数据运营、工程、算法和标注团队协作,并明确负责准确率、交付周期、服务等级协议(SLA)、成本和问题解决。资料也提到现场团队负责部署、维护硬件,并把现场经验转化为标准操作规程(SOP)。这些信息只能说明运营架构如何设计,不能证明公司已达到对外承诺的服务水平;指定的公开资料没有运行时间记录、交付质量分布或客户验收数据。[CE001, CE004, CE007, CE008]
| 用户要完成的工作 | 当前限制 | XDOF 所述方案 | 公开可量化的收益 | 局限性 |
|---|---|---|---|---|
| 把模型训练需求落到数据采集任务上 | 物理任务和机器人本体各异,交付规范需要定制 | 客户对接团队把需求转成采集要求,并安排数据交付顺序 | 尚无公开数据说明需求转为验收需要多久,或转化率如何 | |
| 生成观测—动作演示数据 | 公开互联网缺少可比的机器人动作数据 | 由远程遥操作和佩戴传感器的采集员获取演示数据 | 据报道,ABC-130K 收录了覆盖 195 项任务的 130,000-plus 个片段或轨迹 | 开源数据集无法证明其覆盖特定客户任务,也无法证明数据质量 |
| 处理采集数据并把好质量关 | 原始演示数据需要同步处理、标注和审核 | 依靠数据接入算力、标注流水线、质量标准、评分和根因分析 | 尚无公开的错误率分布、返工率或独立质量基准 | |
| 交付数据集并持续改进 | 客户需要可用数据,也需要模型评估反馈 | 用数据集工具交付数据、开展质检和问题分流,再依据评估结果迭代 | 尚无公开的服务等级达标情况、模型性能提升或续约证据 |
收益栏的 null 表示指定材料中找不到可比的公开指标,不代表收益为零。
[CE001, CE004, CE007, CE009]5.3 数据基础设施宽度可见,生产级规模未获验证
招聘材料显示,XDOF 涉及摄入计算、工作流编排、内部研究流程、多租户数据湖管理、大规模数仓工具、后端流水线、可视化工具和端云硬件集成。ABC-130K 是目前最有力的公开技术成果:多篇独立报道显示,该数据集覆盖 195 项任务、至少 130,000 条双臂操作片段或轨迹;一篇详尽报道还提到同步公开了训练代码、模型、仿真基础设施和评测资源。这证明 XDOF 能构建并开放大型科研数据集,却不能证明其宣传的 EB 级基础设施远景、多租户隔离能力、生产环境时延,或付费客户的模型性能确有提升。[CE004, CE005, CE009, CE010]
| 层级或流程 | 作用 | 实施证据 | 关键依赖 | 主要风险 |
|---|---|---|---|---|
| 机器人与感知 | 采集机器人动作、人体运动和环境观测数据 | 多篇报道提到遥操作;招聘信息还出现人形机器人和感知研发岗位 | 可靠的硬件、校准、传感器和操作员 | 针对特定机器人本体采集的数据,可能难以迁移到其他本体 |
| 数据采集运营 | 管理采集场地和外勤团队,审核数据并建立可重复的标准流程 | 招聘岗位职责包括部署和维护硬件、评分、审核及标准作业流程(SOP) | 需要分布式团队、场地合作伙伴和统一的质量控制 | 人力和硬件投入都很重,扩张成本可能高且难以预测 |
| 数据平台 | 接入、编排、存储和标注数据,并支持探索与交付 | 近期招聘描述明确列出数据接入算力、编排、数据湖、数据仓库和可视化工具 | 关键依赖包括存储成本效益、租户隔离、数据血缘和研究人员工具 | 招聘信息不能证明平台已在生产环境以 EB 级规模运行 |
| 信任与治理 | 保护信息、管控访问并尊重数据权利 | 官网隐私政策列出了安全措施和个人隐私权利 | 需要数据集专属合同、管控措施、审计证据和事件处置流程 | 官网隐私政策本身不能证明产品或机器人数据安全 |
本表根据媒体报道与公司招聘披露重构;未经证实的生产能力主张均已明确限定。
[CE004, CE005, CE008, CE012]公开资料呈现一条从任务定义、物理采集到数据运营、交付和评估的垂直流程;ABC-130K 是其中最清晰的公开成果。
- 1. 客户任务定义
- 将目标任务、机器人本体、数据格式和验收要求转化为交付技术规范。
- 条目
- 需求制定与任务排序 — Described in customer-delivery job reposts; no public cycle-time metric.
- 2. 物理演示采集
- 采集目标机器人演示数据、低成本遥操作数据,或第一视角的人体运动数据。
- 条目
- 机器人、传感器、操作员与场地 — Collection modes are reported; sensor and hardware specifications remain undisclosed.
- 3. 数据接入与编排
- 把采集数据送入处理、存储和研究人员工作流。
- 条目
- 算力、数据管线与多租户存储 — Scope comes from hiring text rather than an independently tested production system.
- 4. 标注与质量控制
- 完成数据标注,并落实审核规范、评分、校准和根因分析。
- 条目
- 质量标准与纠错闭环 — No public distribution of error, rework, or customer-acceptance rates.
- 5. 数据集交付与工具链
- 提供数据集、可视化与探索工具,并处理客户问题。
- 条目
- 自助式与托管式交付 — Availability, packaging, API surface, and realized service levels are not public.
- 6. 评估与反馈
- 利用模型评估和客户反馈指导后续针对性采集与持续改进。
- 条目
- ABC-130K 验证标杆 — Independent sources report 130,000-plus trajectories or episodes across 195 tasks.
本图整合已披露的工作流环节,并非经核实的内部系统架构;图示顺序不代表实际延迟或已达到生产就绪状态。
[CE002, CE004, CE009, CE010, CE011]5.4 差异化壁垒取决于交付执行、数据质量与安全信任
如果 XDOF 能建立护城河,关键可能在于采集硬件、运营规范、标注和评测形成的反馈闭环,创始团队据报道积累的 GELLO 遥操作经验也能助其一臂之力。但这种模式执行难度很高:采集依赖专业设备、熟练操作员、传感器校准、设备维护和跨站点稳定质控;成熟的大客户也可能选择在内部保留战略性数据业务。指定资料中唯一直接涉及信任机制的文件,是 XDOF 官网隐私政策。政策说明了网站访问和联络数据的保护措施及用户权利,却没有规定机器人遥测、客户数据集、模型文件、访问隔离、应急响应或第三方认证的安全管控。因此,不能据此断言公司从未发生安全事故。[CE008, CE011, CE012, CE013, CE014]
5.5 重点图表
06客户分析
6.1 客户客群分布与采购决策背景
XDOF 的公开定位和第三方报道都指向企业客户,而不是已验证的消费业务。Built In 发布的公司简介称,XDOF 为机器人开发商提供基础设施;TechCrunch 报道称,其数据流水线、采集工具和标注系统面向难以自建这些能力的前沿 AI 实验室和机器人公司。Preqin 把目标客群扩展至家庭用户以及制造、物流企业,但没有列出这些领域的实际部署。因此,可用于投资判断的证据主要来自前沿实验室和机器人开发商;消费、制造和物流只能视为公司宣称的目标市场,不能当作已验证的客户群。[CU001, CU002, CU007, CU008]
| 客群细分 | 买方 / 使用方 / 付款方 | 使用场景 | 规模证据 | 战略价值 | 证据缺口 |
|---|---|---|---|---|---|
| 前沿 AI 实验室 | AI 研究和机器人团队;企业付款方未披露 | 外包采集并标注物理世界训练数据 | 据报道约 20 家客户中有数家前沿 AI 实验室 | 目前公开证据最能支持的核心客群 | 实验室均未具名,生产部署情况和支出也未披露 |
| 机器人公司 | 机器人研发团队;企业付款方未披露 | 数据流水线、采集工具和标注系统 | 与 XDOF 的基础设施定位直接契合 | 尚无具名买方、部署情况、成效或合同金额 | |
| 制造与物流企业 | 运营和自动化团队;付款方未披露 | 用机器人辅助制造和物流作业 | 有潜力拓展的工业客群 | Preqin 只将其列为目标市场,尚无材料证实其已成为客户 | |
| 家庭住户 | 家庭用户和购买者 | 自动化处理家务 | 长期潜在终端市场 | 现有留存来源未能证实 XDOF 已在家庭场景部署或售出产品 |
各来源对目标客群的描述详略不一;截至 2026-09-28,指定预抓取来源未提供规模证据,因此以 null 标示。
[CU001, CU002, CU003, CU007]6.2 客户数量与客户验证
最明确的客户采用指标是约 20 家。据 TechCrunch 报道,XDOF 称已与 20 家客户合作,其中包括数家前沿 AI 实验室;Altis 在融资消息发布时独立标注约 20 家活跃付费客户。TechCrunch 还报道称,截至 2026 年 9 月,年化收入接近 $50 million。这是年化收入,不是 ARR,也不是经审计确认的收入;现有资料没有说明统计周期、经常性收入占比或客户口径。更重要的是,留存证据没有列出任何具名商业客户。现有记录能证明早期需求和付费活动,却无法区分生产部署与试点,也无法验证客户成效或客户推荐价值。[CU003, CU004, CU005, CU006, CU014, CU015]
| 指标 | 据报数值 | 统计周期 | 信源依据 | 置信度 | 投资启示 | 缺少分母 / 待补充口径 |
|---|---|---|---|---|---|---|
| 客户数 | 约 20 家 | 2026 年融资公告期 | TechCrunch 援引公司说法与 Altis 市场研究 | 中 | 早期付费需求的证据 | 具名客户账户、合作起始日期、生产状态及客户结构 |
| 活跃付费客户数 | 约 20 家 | 融资公告;已留存的 Altis 页面未注明确切日期 | Altis 将该指标标注为活跃付费 | 中 | 比客户 Logo 或销售线索数量更能证明需求,但未经独立审计 | 活跃客户的认定标准、合同期限及客户队列留存率 |
| 年化收入 | 接近 $50 million | 据 2026-09-04 报道 | Yahoo Finance 转载的 TechCrunch 报道 | 中 | 商业化提速的迹象 | 测算周期、经常性收入占比、已确认收入及客户分布 |
据报道数值未经审计。年化收入按据报道口径保留,未改写为 ARR。
[CU003, CU004, CU005]6.3 收入持续性、客群拓展与客户集中度
公开资料显示了可能的业务扩展方向,但扩展仍未经证实。Preqin 称 XDOF 同时销售设备系统、软件许可、长期服务和软件更新订阅;Tech Funding News 则称,公司通过遥操作、人工采集员和标注系统交付,并计划在全球扩充操作员网络。客户的数据需求若不断扩大,这些服务可能加深客户关系,但资料没有披露续约率、合同期限、总留存率(GRR)、净留存率(NRR)、流失率、客户队列或增购收入。客户数约为 20 家,也不足以说明头部客户贡献了多少收入。Tech Funding News 明确提出该采集模式能否大规模扩张仍是未解问题;一份融资分析也警告,巨额首轮融资抬高了后续里程碑要求,却没有带来客户层面的商业验证。因此,收入持续性和客户集中度仍是重要尽调缺口。[CU009, CU010, CU011, CU012, CU013, CU016]
据报道的交付环节为客户账户扩张提供了合理基础,但续约和客户集中度仍缺公开证据。
阶段 1-5 汇总了据报道的工作流程;阶段 6 只是潜在机制,并非经验证的转化或续约路径。
[CU008, CU009, CU010, CU011, CU012]6.4 重点图表
07风险分析
7.1 主要风险与评估基调
最大风险是业务扩张速度与验证进度不匹配。据报道,XDOF 完成 $70 million Series A 后约 3 个月就开始洽谈新一轮融资,年化收入接近 $50 million,并计划在全球扩充遥操作员和穿戴传感器采集员网络。这说明市场有需求、公司能接触资本,却不能证明收入具有经常性、现金可用、跑道充足或运营杠杆已改善。在管理层提供经审计的收入结构、毛利率、烧钱速度、可用现金及与里程碑挂钩的招聘计划前,融资和扩张执行风险应评为“高影响度、中等发生概率”。另一项高影响度风险是数据安全与隐私:XDOF 的官方政策承认,没有任何数据传输或存储系统能做到绝对安全。风险热力图根据已披露事实和信息缺口排列剩余风险,并不表示安全事故已经发生。[CR003, CR006, CR007, CR008, CR010, CR013, CR016]
按影响程度和发生概率梳理重大风险敞口;指定证据不足以估算概率时,保留未知项。
这里只作定性归类;目前没有事件频率、损失分布、经审计的集中度数据或公司风险评分。
[CR002, CR003, CR007, CR009, CR010, CR012, CR015, CR016]7.2 法律合规、监管要求与安全管控
指定证据中,XDOF 官网隐私政策是唯一一手法律管控文件。政策覆盖网站访问和通信数据,说明加州相关权利,列出总法律顾问、安全联系人和数据保护官,并允许把数据传到用户所在辖区以外。这些政策层面的措施有意义,但不能证明实际执行有效,也不能证明公司具备客户数据管控、采集员知情同意规范、合同数据处理条款、商业保险、安全认证或监管审批。政策还允许向服务商和特定商业伙伴披露数据,扩大了第三方管控风险。现有资料不足以推断公司没有安全事故或诉讼记录:资料库没有监管记录、法院案卷、资质状态历史、第三方认证报告或安全事故登记。投资审批前必须调取这些一手材料,不能把公开政策当作内控有效的证明。[CR001, CR002, CR003, CR004, CR005, CR014]
| 风险或法定义务 | 司法管辖区 | 证据现状 | 发生概率 | 严重程度 | 已披露缓解措施 | 残留风险敞口 | 尽调排查路径 |
|---|---|---|---|---|---|---|---|
| 个人信息权利及请求响应 | 加利福尼亚州 / 美国 | 隐私政策说明了访问、删除、数据可携性及身份核验流程 | 中 | 高 | 总法律顾问(General Counsel)联系方式与陈述的请求处理流程 | 中;实际运行有效性未经证实 | 审查请求日志、响应测试记录、数据留存时间表以及投诉历史 |
| 个人数据跨境传输 | 欧洲经济区(EEA)及其他非美国司法管辖区 | 隐私政策称,数据可能传输至美国及其他国家 | 中 | 高 | 政策所述安全保障措施、合规声明及数据保护官(DPO)联系方式 | 中高;跨境传输法律机制及子处理方未披露 | 审查 DPA、SCC、跨境传输影响评估及子处理方清单 |
| 安全、数据泄露及客户数据义务 | 多司法管辖区 | 政策声明了安全保护措施,但并未保证绝对安全 | 未知 | 高 | 政策称已采取物理、电子及管理层面的保护措施 | 未知;未见认证、渗透测试或安全事件记录 | 获取 SOC 报告、渗透测试、安全事件日志、网络安全保险及客户安全补充协议 |
部分风险登记基于 XDOF 2026 年 6 月的网站隐私政策以及对其数据供应链模式的独立描述;未引用具体诉讼或安全事件并不等同于不存在此类风险。
[CR001, CR002, CR003, CR004, CR005, CR014]7.3 运营模式风险与交付质量敞口
XDOF 正用采集工具、标注系统、机器人遥操作和人体动作捕捉搭建外包数据供应链。业务涉及人员培训、知情同意、任务执行一致性、硬件可用率、标注质量和客户数据安全传输,任何环节出问题都可能影响交付。公开报道显示,公司约有 60 名员工和 20 家客户,并计划在全球扩招采集员;运营规模可能比现有内控体系扩张得更快。指定证据没有披露客户身份、合同条款、验收标准、数据集缺陷率、返工率、服务水平或商业保险。真正的否决信号不是偶发的一次质量失误,而是客户反复验收失败、返工增加、重大安全事故,或采集规模增长却没有带来毛利率改善和客户复购。[CR003, CR009, CR010, CR011, CR012, CR014]
| 失效模式 | 发生概率 | 严重程度 | 缓解措施成熟度 | 残留风险敞口 | 监控指标 | 尚未解决的缺口 |
|---|---|---|---|---|---|---|
| 采集员网络扩张快于培训和质控能力建设 | 中 | 高 | 未披露 | 高 | 验收通过率、返工率及采集员通过认证所需时间 | 缺少分批次的质量数据或采集员生产力数据 |
| 遥操作、传感器或标注流水线中断 | 中 | 高 | 除网站常规安全表述外未披露 | 高 | 正常运行时间、会话失败次数、数据丢失率及恢复耗时 | 缺少服务等级协议、灾难恢复或客户安全证据 |
| 客户需求变化快于工具复用能力的形成 | 中 | 高 | 平台与标注系统仅有文字描述,未经独立测试 | 中高 | 定制工作占比、单项目毛利率及复购率 | 客户合同、验收标准及具体项目经济模型属于非公开信息 |
| 资金充裕的竞争对手挤压差异化优势与定价空间 | 中 | 中高 | 指定抓取文本未显示客户存在持久切换成本 | 中高 | 竞标胜率、实际成交价格水平、续约情况及被替换频率 | 缺少质量、价格与切换成本的对比研究材料 |
发生概率和残余风险评级是分析归类,依据已披露的运营机制与明确证据缺口,并非实际故障记录。
[CR003, CR009, CR010, CR011, CR012, CR014, CR016]7.4 外部依赖、融资脆弱性与逻辑否决触发条件
公开资料显示,XDOF 依赖少数主要但未具名的客户、线下采集员、技术和分析服务商、高校合作伙伴,以及潜在投资方。据报道,公司约有 20 家客户,其中包括数家前沿 AI 实验室;但资料没有披露收入集中度、续约条款、生产部署与试点的比例,也没有说明客户是否更换供应商。XDOF 与 UC Berkeley AI Research 实验室合作开发数据集,但相关证据没有公开知识产权归属、独家性、合作期限或商业化权益。据报道,Series B 条款仍未敲定,因此高估值融资消息不等于现金已到账或融资确定。以下情况将推翻投资逻辑:主要客户流失且无法及时补上;无法证明数据集来源和采集员知情同意;发生重大安全或生产事故;遭遇不利知识产权诉讼或监管处罚;或融资扩张计划依赖缺乏可信单位经济数据支撑的增长假设。[CR002, CR007, CR009, CR013, CR014, CR015, CR016]
| 依赖项 | 合作对手方 | 职责定位 | 集中度证据 | 失效场景 | 严重程度 | 缓解措施 | 残留风险敞口 |
|---|---|---|---|---|---|---|---|
| 客户需求与续约 | 约 20 家客户,其中数家据称属于前沿 AI 实验室 | 数据采集与训练数据服务的买方 | 收入集中度与客户身份未披露 | 客户暂停一个或多个大型项目、改为内部采集,或拒收数据 | 高 | 据报道的客户覆盖面,有待验证 | 高(在核查合同与集中度证据前) |
| 科研与数据集合作 | UC Berkeley AI Research Lab | 据报道的 ABC 数据集联合发布方 | 所有权与独家协议条款未披露 | 合作关系、权利归属或研究合作能否延续发生变化 | 中高 | 现有研究合作关系 | 中(有待知识产权及商业化协议确认) |
| 后续资本 | 潜在 Series B 投资者,据报道包括 8VC | 为 Series A 之后的扩张提供资金 | 据报道条款尚未最终确定且金额未披露 | 本轮未能交割,或条款附带限制性经济条件 | 高 | 此前 $70M 的 Series A | 中高(缺少现金余额、烧钱速度、现金跑道及已签署条款) |
| 数据与分析供应商 | 未具名的服务商、防欺诈服务商与分析服务商 | 处理网站与交互信息 | 子处理方清单及其关键程度未披露 | 供应商数据泄露、服务中断或跨境合规失效 | 高 | 合同中的服务商及其披露的安全保障 | 中高(有待供应商清单及控制证据确认) |
集中度一栏仍是证据缺口:公开信息虽指出依赖对象,却未披露合同占比、排他性、收入贡献或替换所需时间。
[CR002, CR006, CR007, CR009, CR014, CR015]7.5 重点图表
08估值分析
8.1 估值演变背景与历史资本结构
XDOF 于 2026 年 6 月公开亮相,宣布完成 $70.0 million Series A 股权融资,由 Thrive Capital 和 Spark Capital 共同领投,Andreessen Horowitz、Lux Capital 与 WndrCo 跟投。若计入早期种子轮,公司累计获得的风险投资约为 $78 million。公开亮相不到 3 个月,XDOF 就在 2026 年 9 月开始洽谈新一轮融资;据市场消息,拟以约 $1.2 billion 估值融资的 Series B 由 8VC 领投。融资节奏加快,背后是商业进展迅速:据报道,公司约有 20 家企业客户,主要是头部前沿 AI 实验室和机器人开发商,年化收入接近 $50 million。但交易条款尚未敲定,最终投前估值、总稀释比例和清算优先权负担仍不确定。[CV001, CV002, CV003, CV004, CV005, CV006]
| 维度 | 评估 | 证据基础 | 决策含义 |
|---|---|---|---|
| 投资建议 | 观察 / 观望 | 收入运行率增长迅速,但 Series B 估值尚未证实,且资本消耗高 | 专有遥操作业务的单位经济性得到验证前,不应按未经核实的 $1.2B 要价投资 |
| 置信度 | 中 | 有文件记录的 $70M-$78M 资本基础和 20 家实验室客户,但 $50M 收入运行率未经审计 | 承诺出资前,核验客户收入留存情况与合同持续性 |
| 风险评级 | 高 | 遥操作员网络资本密集,若融资受阻,现金跑道可能较短 | 要求清算优先权提供下行保护,并按里程碑分期出资 |
| 估值立场 | 偏高 | 据报道,Series B 目标估值为 $1.2B;以 $50M 收入运行率计算,隐含约 24x | 设定入场估值上限,或以经核验的经常性利润率与毛利率里程碑为交割前提 |
截至 2026 年 9 月,评级和估值判断综合考虑了据报道的 $70M Series A 融资、传闻中的 $1.2B Series B 洽谈,以及未经证实的 $50M 收入运行率说法。
[CV001, CV004, CV006, CV008, CV010]8.2 投资逻辑与反方观点
看好 XDOF 的核心逻辑是,训练机器人和具身智能基础模型所需的物理交互数据极其稀缺。大语言模型可利用公开互联网的海量 Token,通用人形机器人却需要直接从物理环境中采集高保真传感器和轨迹数据。XDOF 将 UC Berkeley GELLO 遥操作架构商业化,并开源发布包含 130,000 个操作片段的 ABC-130K 基准数据集,初步成为前沿实验室的专业数据工厂。反方观点则强调其运营短板:采集物理示范需要分布式遥操作员和穿戴动捕设备的采集人员,成本结构因此高度依赖人力,毛利率也明显低于纯软件业务。首轮融资由多家风投共同参与,也带来治理隐患;股权分散可能削弱领投方的责任感,后续融资受阻时,公司将面临流动性风险。[CV003, CV007, CV010, CV011, CV013]
| 视角 | 投资论点 | 反方反驳论点 | 何种情况会改变观点 |
|---|---|---|---|
| 物理数据瓶颈 | 机器人缺少互联网级训练语料;XDOF 的真实场景遥操作管线承担着物理世界版 Scale AI 的角色 | 物理采集无法像网络抓取那样低成本扩张;劳动密集型操作员网络限制了毛利率上限 | 若有证据表明合成数据可迁移,或自动化、自监督采集能减少人工操作时长 |
| 先发规模 | ABC-130K 开源数据集与 UC Berkeley 学术背景确立了标准双臂操作基准 | 开源基准模型降低了采集方案门槛,一线实验室可在内部复刻设备 | 与一线前沿 AI 实验室签署排他性协议或多年期专有数据供应协议 |
| 收入增速 | 据报道,走出隐身模式后 3 个月内年化收入接近 $50M,显示实验室需求极为迫切 | 早期收入只是预训练初期的一次性项目预算,并非持久的经常性软件 ARR | 披露客户群留存、净收入留存率(NRR)超过 120% 以及多年期订阅承诺 |
| 资本结构 | Thrive、Spark、a16z 与 Lux 的支持带来一线机构背书,也为后续融资提供资本储备 | 投资人持股分散;若下一轮融资延迟,没有一家领投机构持有足够份额来确保过桥资金 | 若有机构领投方拿出定价条款清单,并保障至少 24 个月现金跑道 |
核心论点与反方论点反映了物理数据采集经济效益、约 20 家前沿实验室客户的集中度风险以及银团结构。
[CV003, CV007, CV010, CV011, CV013]8.3 情景分析、对标估值与回报倍数
评估 XDOF,需要参考更广泛的 AI 和数据基础设施一级市场,并建立严格的情景模型。若按传闻中的 $1.2 billion 估值和未经验证的 $50 million 年化收入运行率计算,隐含远期市销率约为 24x。这个倍数接近知名 AI 基础设施公司 Scale AI(估值 $13.8 billion)和 Mercor(估值 $10 billion),而同样布局数据流水线的具身智能公司 Encord 已融资 $60 million。基准情景(概率 55%)假设 XDOF 以 $1.0B 至 $1.2B 完成 Series B,年化收入稳定在 $50M 至 $65M,毛利率达到 50%,投资回报取决于执行情况。乐观情景(概率 20%)假设遥操作效率提升至半自主水平,年化收入升至 $100M+。悲观情景(概率 25%)则假设现有收入只是不可持续的一次性项目支出;60 多名全职员工和全球遥操作团队带来的高额烧钱,可能迫使公司接受结构化降估值融资(down-round)。[CV002, CV004, CV006, CV008, CV009, CV012]
| 情景 | 运营假设 | 估值 / 回报逻辑 | 关键风险与下行触发条件 | 概率信号 |
|---|---|---|---|---|
| 乐观情景 | 年化收入增至 $100M+;遥操作网络转向人机协同自动化;前沿实验室客户从 20 家增至 60+ 家 | 下一次流动性事件估值达到 $2.5B-$3.0B;凭借物理 AI 赛道领先地位,估值倍数扩大 2.0x-2.5x | 客户把遥操作收回自营;毛利率降至 40% 以下 | 20% |
| 基准情景 | Series B 以 $1.0B-$1.2B 交割;年化收入稳定在 $50M-$65M、毛利率为 50%;签有 20-30 份活跃企业合同 | 维持 $1.0B-$1.2B 账面估值;回报取决于运营执行,以及收入能否稳步转化为 ARR | 领投机构之间的银团谈判停滞;招聘和烧钱持续消耗现金储备 | 55% |
| 悲观情景 | Series B 未能以 $1.2B 交割;收入最终被证实来自一次性项目支出;60+ 名员工及遥操作员带来的高额开支耗尽 $70M 资本 | 降轮融资或以 $300M-$500M 进行结构化资本重组;高级清算优先权造成严重稀释 | 前沿实验室转向仿真或扩散策略;遥操作人工成本耗尽现金 | 25% |
情景概率与回报逻辑是作者基于一级市场机器人可比公司所作的建模,假设历史融资为 $78M、员工约 60 人。
[CV002, CV004, CV006, CV008, CV012]8.4 投资建议立场、关键尽调诉求与逻辑否决线
综合现有证据,暂定建议为“跟踪 / 保持关注(Track / Hold)”,置信度中等,风险评级高。XDOF 商业进展迅速,也得到一线风投支持,显示市场对具身机器人数据需求旺盛;但若按未经确认的 $1.2 billion 估值投资,投资者将承担估值倍数收缩和下行风险。公开资料尚未证实 $50 million 年化收入运行率究竟来自可持续的经常性 ARR,还是初期模型训练的一次性数据采集项目。尽调应重点核查经审计的客户留存数据、扣除遥操作员薪酬后的净毛利率,以及潜在 Series B 领投方的确定性出资承诺函。若核心客户转为自建遥操作采集、毛利率跌破 35%,或未能在现金跑道缩短至 12 个月前锁定机构领投方,应立即改为“回避(Avoid)”。[CV008, CV010, CV011, CV014]
从物理数据市场瓶颈出发,经收入扩张、人工成本压力和估值倍数,最终落到「观察」建议。
| 节点 / 连接 | 节点 / 起点 | 终点 | 说明 |
|---|---|---|---|
| 节点 1 | 物理数据瓶颈 [n1] | 前沿 AI 实验室缺少互联网级规模的真实世界机器人演示数据 | |
| 节点 2 | 收入运行率快速攀升 [n2] | 20 家实验室客户贡献的年化收入接近 $50M,推动风投机构需求升温 | |
| 节点 3 | 遥操作人力成本 [n3] | 人工遥操作员和 60+ 名员工让公司持续烧钱 | |
| 节点 4 | 多基金联合融资 [n4] | 多家基金分摊投资份额,后续轮次没有单一投资方全额负责 | |
| 节点 5 | 估值倍数偏高 [n5] | 据传 Series B 目标估值为 $1.2B,相当于早期收入约 24x | |
| 节点 6 | 决策:观察 / 暂缓投资 [n6] | 估值尚未核实,毛利率和客户流失数据也待审计;审计完成前暂不投资 | |
| 连接 1 | 物理数据瓶颈 [n1] | 收入运行率快速攀升 [n2] | |
| 连接 2 | 收入运行率快速攀升 [n2] | 遥操作人力成本 [n3] | |
| 连接 3 | 遥操作人力成本 [n3] | 多基金联合融资 [n4] | |
| 连接 4 | 多基金联合融资 [n4] | 估值倍数偏高 [n5] | |
| 连接 5 | 估值倍数偏高 [n5] | 决策:观察 / 暂缓投资 [n6] |
8.5 重点图表
免责声明
本报告基于公开证据撰写,仅为尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;作出投资决定前,应直接向管理层求证,并查阅一手文件核实。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | XDOF, Inc. was founded in October 2024 by UC Berkeley researchers Philipp Wu, Fred Shentu, and Nemo Jin as a robotics infrastructure company headquartered in San Mateo, California. | 高 | SO001, SO005, SO007, SO008 |
| CO002 | XDOF emerged from stealth on June 17, 2026, announcing a $70 million Series A funding round that represents the company's entire lifetime institutional equity financing to date. | 高 | SO001, SO002, SO003, SO004 |
| CO003 | The $70 million Series A round was participated in by venture capital firms Thrive Capital, Andreessen Horowitz, Lux Capital, Spark Capital, and WndrCo. | 高 | SO001, SO002, SO003, SO005 |
| CO004 | As of September 2026, XDOF is in late-stage discussions to raise a Series B funding round at a post-money valuation of approximately $1.2 billion reportedly led by 8VC, though final terms remain unclosed. | 高 | SO001, SO004, SO005 |
| CO005 | XDOF develops data pipelines, remote teleoperation hardware systems, and annotation tooling to provide physical-world training data for general-purpose robot foundation models. | 中 | SO001, SO002, SO006 |
| CO006 | The technological foundation of XDOF originated from GELLO, an open-source, low-cost robotic arm teleoperation rig created by Philipp Wu and Fred Shentu at UC Berkeley. | 中 | SO001, SO005, SO006 |
| CO007 | Alongside its Series A emergence in June 2026, XDOF released ABC-130K, an open-source bimanual manipulation dataset containing over 130,000 trajectories across 195 manipulation tasks in collaboration with UC Berkeley AI Research lab. | 高 | SO001, SO002, SO005, SO006 |
| CO008 | XDOF reported serving approximately 20 active paying customers at the time of its Series A announcement, including several unnamed frontier artificial intelligence laboratories. | 高 | SO001, SO002, SO005, SO006 |
| CO009 | Third-party press reports indicated XDOF's annualized revenue was approaching $50 million as of September 2026, though the company does not publicly disclose GAAP financial statements or verified recurring revenue metrics. | 中 | SO001, SO002, SO004, SO005 |
| CO010 | XDOF's core team size was documented at approximately 60 employees at its June 2026 funding emergence, with active global hiring across North America, Latin America, and Southeast Asia. | 中 | SO005, SO006, SO009 |
| CO011 | Venture analysts note that XDOF's syndication of five brand-name venture funds in a single large first round creates follow-on financing risk because no single fund owns enough equity to bear sole stewardship for future capital rounds. | 中 | SO003 |
| CO012 | A major operational and capital risk facing XDOF is the potential structural burn of scaling a worldwide human teleoperation and sensor-collection workforce before robot data unit economics are verified. | 中 | SO002, SO003 |
| CO013 | XDOF, Inc. maintains its official registered corporate privacy and compliance documentation under California jurisdiction, listing its primary domain as xdof.ai. | 中 | SO008 |
| CO014 | Neither XDOF nor its investors have publicly disclosed the formal composition of the company's board of directors, observer rights, or founder voting control thresholds. | 中 | SO003 |
| CM001 | Physical robotics models face a foundational data bottleneck because real-world robotic interaction data cannot be scraped from the open internet like text for large language models. | 高 | SM003, SM004 |
| CM002 | XDOF positions itself as an outsourced physical data supply chain for robotics, providing teleoperation tools, egocentric sensor collection, and annotation pipelines to frontier AI laboratories. | 高 | SM003, SM004 |
| CM003 | The company reported serving approximately 20 active paying customers by June 2026, comprising primarily frontier artificial intelligence laboratories and humanoid robotics developers. | 高 | SM003, SM008 |
| CM004 | Reports in September 2026 indicated XDOF was experiencing rapid commercial demand with annualized revenue approaching $50 million, prompting late-stage discussions for Series B financing at a $1.2 billion valuation. | 高 | SM003, SM004 |
| CM005 | Industry analysts and investors characterize the emergence of specialized physical AI data providers as analogous to data-labeling platforms like Scale AI and Mercor in digital language models. | 高 | SM003, SM004 |
| CM006 | XDOF released the open-source ABC-130K dataset in collaboration with the UC Berkeley AI Research Lab, containing 130,000 trajectories across 195 bimanual manipulation tasks. | 高 | SM003, SM008 |
| CM007 | Independent expert interviews conducted by Altis Research revealed that some robotic AI buyers evaluated third-party teleoperation data quality below in-house captures and cited modest switching costs. | 中 | SM005 |
| CM008 | Altis Research models the addressable market for robotics data infrastructure bottom-up across distinct TAM, SAM, and SOM layers rather than relying on top-down macroeconomic sizing figures. | 中 | SM005 |
| CM009 | Preqin identifies XDOF's target customer base as spanning both commercial businesses requiring robotics assistance for manufacturing and logistics, as well as residential automation use cases. | 中 | SM006 |
| CM010 | Software licensing, tiered pricing models based on functionality, and long-term service partnerships for ongoing support represent primary commercial monetization mechanisms in robotics infrastructure profiles. | 中 | SM006 |
| CM011 | Startup fundraising analysts note that large first rounds like XDOF's $70 million Series A commit companies to aggressive growth milestones and high valuation hurdles within condensed operating timeframes. | 中 | SM002 |
| CM012 | Job openings indicate XDOF is establishing international data collection hubs and field operations in Latin America and Indonesia to recruit teleoperators and egocentric video collectors. | 中 | SM007 |
| CM013 | Technical recruiting requirements for humanoid engineers at XDOF include developing control algorithms for dexterous manipulation, locomotion, and sensor integration in dynamic environments. | 中 | SM009 |
| CM014 | XDOF's commercial offering is structured around a multi-tier data pyramid combining teleoperated robot episodes, wearable egocentric human demonstrations, and simulated task environments. | 中 | SM008 |
| CM015 | Direct competitors and adjacent data platform providers chasing the physical AI data bottleneck include Mecka AI, Encord, Scale AI, and Micro1. | 高 | SM003, SM004 |
| CM016 | Bottom-up market sizing models structure physical AI data demand into a multi-tiered pyramid, narrowing from broader automation spend down to immediate teleoperation and trajectory capture contracts. | 中 | SM005, SM008 |
| CP001 | Primary peers in physical robotics data collection include Mecka AI, alongside digital human-data platforms Scale AI and Micro1. | 高 | SP001, SP003 |
| CP002 | Encord raised $60 million to develop annotation and data pipeline tools addressing the physical AI data deficit. | 中 | SP004 |
| CP003 | Scale AI raised $1 billion at a $13.8 billion valuation in May 2024, representing the dominant incumbent in digital AI data annotation expanding into robotics. | 中 | SP004 |
| CP004 | XDOF closed a $70 million Series A in June 2026 backed by Thrive Capital, Andreessen Horowitz, Lux Capital, and Spark Capital, bringing total capital to roughly $78 million. | 高 | SP003, SP004, SP005 |
| CP005 | Mercor grew from a $100 million valuation at $2 billion in February 2025 to a reported $10 billion valuation by October 2025 in digital human-data sourcing. | 中 | SP004 |
| CP006 | XDOF released the open-source ABC-130K dataset with UC Berkeley's AI Research Lab, comprising over 130,000 trajectories across 195 bimanual manipulation tasks. | 中 | SP002, SP004 |
| CP007 | Physical AI data collection requires paired remote teleoperation rigs and human egocentric body-sensor capture, differing fundamentally from text and web image scraping. | 高 | SP001, SP003, SP004 |
| CP008 | XDOF operates a distributed global data collection and annotation network spanning operations in California, Mexico, and Indonesia with dedicated quality control audits. | 高 | SP006, SP007 |
| CP009 | XDOF serves approximately 20 paying customers, including several frontier AI foundation model laboratories, with annualized revenue approaching $50 million. | 高 | SP001, SP003, SP004 |
| CP010 | XDOF generates revenue through direct enterprise contracts, software licenses, unit sales, and ongoing support partnerships, with tiered pricing based on functionality. | 中 | SP006 |
| CP011 | Sector expert research across 21 buyer interviews found that customer switching costs between robotics data collection pipelines are modest. | 中 | SP002 |
| CP012 | Frontier AI laboratories and robotics enterprises evaluate build-versus-buy tradeoffs between outsourced data collection vendors and internal proprietary teleoperation teams. | 高 | SP001, SP003 |
| CP013 | XDOF maintains physical and administrative data security safeguards while retaining customer and operational personal information under commercial agreements and regulatory frameworks. | 中 | SP009 |
| CP014 | Robotics foundation model developers face commoditization and displacement risks as standardized benchmark datasets and synthetic simulation pipelines advance. | 中 | SP002, SP004 |
| CI001 | XDOF monetizes robotics training data through direct sales of robotic systems and associated software licenses, alongside ongoing service and support partnerships. | 中 | SI003 |
| CI002 | XDOF structures product pricing on tiered functionality where advanced systems command premium pricing. | 中 | SI003 |
| CI003 | XDOF's business model targets B2B commercial entities for manufacturing and logistics as well as residential automation use cases. | 中 | SI003 |
| CI004 | XDOF deploys capital toward building data pipelines, collection tools, and annotation systems while expanding data collection and processing personnel. | 中 | SI003 |
| CI005 | XDOF achieved annualized revenue approaching $50 million roughly three months after emerging from stealth in mid-2026. | 高 | SI001, SI005 |
| CI006 | XDOF provides robotics training data to approximately 20 active customers, including several unnamed frontier AI labs. | 高 | SI001, SI005, SI006, SI007 |
| CI007 | XDOF is reportedly negotiating an unclosed Series B funding round at a valuation of approximately $1.2 billion led by 8VC. | 高 | SI001, SI002, SI005, SI006 |
| CI008 | Human data collection businesses historically experience thinner gross margins than software companies because operator and sensor equipment expenses scale linearly with collection volume. | 中 | SI001 |
| CI009 | XDOF faces customer concentration risk where the loss of two or three top frontier AI lab accounts could significantly impair revenue run-rate. | 中 | SI001 |
| CI010 | XDOF raised $70 million in Series A funding in June 2026 from Thrive Capital, Andreessen Horowitz, Lux Capital, and Spark Capital. | 高 | SI001, SI004, SI005, SI006 |
| CI011 | XDOF had raised a total of $70 million in documented institutional capital as of August 2026, representing its primary available balance. | 中 | SI004, SI007 |
| CI012 | XDOF relies on a network of remote teleoperators using GELLO rigs and body-sensor collectors recording manipulation tasks like folding clothes and flattening boxes. | 高 | SI001, SI005, SI006 |
| CI013 | A $1.2 billion valuation against approximately $50 million in annualized revenue implies an enterprise revenue multiple in the mid-20s. | 中 | SI001 |
| CI014 | XDOF is hiring international data operations and quality control personnel across regions including Indonesia and Mexico to expand collection capacity. | 中 | SI008 |
| CI015 | Detailed customer acquisition costs, gross margin breakdowns, and contract renewal rates are not publicly disclosed by XDOF. | 中 | SI001, SI004 |
| CE001 | Independent reporting describes XDOF as an outsourced robotics-data supply chain combining collection tools, data pipelines, annotation, and evaluation support. | 中 | SE001, SE005, SE007 |
| CE002 | XDOF's reported collection workflow combines remote robot teleoperation with sensor-wearing humans who capture everyday physical activity. | 中 | SE001, SE005, SE006 |
| CE003 | One independent account describes three collection tiers: data from the customer's target robot, lower-cost teleoperated systems, and egocentric human activity. | 中 | SE001 |
| CE004 | The publicly described operating chain spans collection, ingestion, annotation, dataset tooling, delivery, and evaluation feedback. | 中 | SE001, SE003, SE005, SE007 |
| CE005 | Company-issued hiring text calls for ingestion compute, orchestration, internal developer workflows, multi-tenant data-lake management, and exabyte-scale warehouse tooling. | 中 | SE003 |
| CE006 | A reposted XDOF humanoid-engineering specification calls for mechanical, electrical, and software design plus control, perception, simulation, validation, safety, and reliability work. | 中 | SE002 |
| CE007 | Company-issued role descriptions assign delivery teams responsibility for specifications, quality, accuracy, service levels, schedules, costs, issue resolution, and customer onboarding. | 中 | SE003 |
| CE008 | XDOF's described collection network depends on physical venues, deployed hardware, maintenance, field teams, audits, and reusable operating procedures. | 中 | SE003, SE006 |
| CE009 | Independent sources report that ABC-130K contains at least 130,000 bimanual manipulation episodes or trajectories across 195 tasks. | 中 | SE001, SE006, SE007 |
| CE010 | One detailed account reports that the ABC release includes training code, models, simulation infrastructure, and evaluation resources. | 中 | SE001 |
| CE011 | Independent reporting connects XDOF's origin to the founders' GELLO low-cost teleoperation research and describes a feedback loop from collection through evaluation and targeted improvement. | 中 | SE001, SE005, SE006 |
| CE012 | XDOF's privacy policy states that it uses physical, electronic, and administrative safeguards for personal information collected through its website and company communications. | 中 | SE008 |
| CE013 | The assigned evidence does not establish product-security certifications, customer-dataset controls, uptime performance, or an incident history for XDOF's robotics-data platform. | 低 | |
| CE014 | Independent analysis identifies labor and hardware cost, variable dataset quality, and possible customer insourcing as scaling risks for XDOF's operating model. | 中 | SE001, SE004 |
| CE015 | Preqin's profile characterizes XDOF as developing robotics systems and foundation models for household and industrial automation, in tension with reporting that presents the company primarily as a data supplier. | 低 | SE009 |
| CU001 | XDOF presents itself as an infrastructure partner that supplies tools, data, and services to robotics builders. | 中 | SU002 |
| CU002 | Third-party coverage identifies frontier AI labs and robotics companies as intended users of XDOF's data infrastructure. | 中 | SU004 |
| CU003 | TechCrunch reported that XDOF said it was working with 20 customers, including several frontier AI labs. | 中 | SU004 |
| CU004 | Altis described XDOF as having about 20 active paying customers at the funding announcement. | 中 | SU007 |
| CU005 | TechCrunch reported annualized revenue approaching $50 million in September 2026. | 中 | SU004, SU005 |
| CU006 | The retained customer evidence identifies no commercial customer by name and instead refers to unnamed frontier AI labs. | 中 | SU004, SU005, SU007 |
| CU007 | Preqin describes residential users and businesses requiring manufacturing or logistics automation as XDOF target segments. | 低 | SU008 |
| CU008 | Preqin describes XDOF revenue as coming primarily from direct robotic-system sales and associated software licenses. | 低 | SU008 |
| CU009 | Preqin describes long-term service relationships and software-update subscriptions as additional commercial structures. | 低 | SU008 |
| CU010 | XDOF's reported delivery model combines teleoperation tools, human data collectors, and annotation systems. | 中 | SU005 |
| CU011 | Tech Funding News reported plans to expand XDOF's network of teleoperators and sensor-wearing data collectors globally. | 中 | SU005 |
| CU012 | Tech Funding News identified whether XDOF's data-collection model can scale as an unanswered question. | 中 | SU005 |
| CU013 | Startup Fundraising cautioned that XDOF's large first round creates a growth obligation rather than validating customer durability. | 中 | SU003 |
| CU014 | Altis says its XDOF research draws on customers, competitors, and former employees, but its retained public page does not expose the underlying interview evidence. | 中 | SU007 |
| CU015 | The retained aVenture fetch contains only a page title and supplies no substantive customer evidence. | 中 | SU006 |
| CU016 | Vaia's company-issued job reposts include robotics data-platform user-experience and infrastructure engineering roles. | 中 | SU001 |
| CR001 | XDOF's June 2026 privacy policy covers website, mobile-site, social-media, and communications-derived personal information. | 中 | SR003 |
| CR002 | XDOF says it may disclose information to related companies and service providers for business, professional, or technical support. | 中 | SR003 |
| CR003 | XDOF says it maintains safeguards but cannot ensure or warrant complete security for Internet transmission or data storage. | 中 | SR003 |
| CR004 | XDOF's policy contemplates transferring personal information outside a user's jurisdiction, including to the United States. | 中 | SR003 |
| CR005 | XDOF's policy describes California privacy rights and identifies General Counsel, Security Team, and Data Protection Officer contacts. | 中 | SR003 |
| CR006 | Multiple independent sources report that XDOF closed a $70 million Series A in June 2026. | 高 | SR001, SR002, SR004, SR005, SR006, SR008 |
| CR007 | Independent reporting described proposed Series B terms as non-final and did not establish the amount to be raised. | 高 | SR001, SR005, SR006 |
| CR008 | Independent reports said XDOF's annualized revenue was approaching $50 million, without establishing audited revenue, ARR, or cash generation. | 中 | SR001, SR005, SR006 |
| CR009 | Independent sources report about 20 XDOF customers, including several unnamed frontier AI labs. | 中 | SR001, SR005, SR006, SR008 |
| CR010 | XDOF reportedly plans to expand worldwide teams of remote teleoperators and sensor-wearing data collectors. | 中 | SR001, SR005, SR006 |
| CR011 | Altis and Tech Funding News each describe XDOF as having roughly 60 employees around the financing period. | 中 | SR006, SR008 |
| CR012 | Tracxn lists XDOF among 46 active competitors, including 19 funded competitors. | 中 | SR002 |
| CR013 | An independent fundraising analysis treats XDOF's unusually large first round as an execution obligation and notes that valuation, terms, and board composition are undisclosed. | 中 | SR004 |
| CR014 | Independent reporting describes XDOF as an outsourced provider of data pipelines, collection tools, and annotation systems for robotics customers. | 中 | SR001, SR005, SR006, SR008 |
| CR015 | Independent reports say XDOF is partnering with UC Berkeley's AI Research Lab on the ABC robot-training dataset. | 中 | SR001, SR005 |
| CR016 | The combination of rapid financing, reported growth, and planned collector-network expansion makes execution discipline the leading residual risk. | 中 | SR004, SR005, SR006 |
| CV001 | XDOF raised a $70.0 million Series A equity financing round in June 2026 led by Thrive Capital and Spark Capital. | 中 | SV001, SV002 |
| CV002 | Total venture funding raised by XDOF stands at approximately $78 million when accounting for an earlier seed round prior to its Series A. | 中 | SV003, SV007 |
| CV003 | The Series A investor syndicate includes Thrive Capital, Spark Capital, Andreessen Horowitz, Lux Capital, and WndrCo. | 中 | SV002, SV003, SV004 |
| CV004 | In September 2026, XDOF reportedly entered late-stage discussions to raise Series B funding at a valuation of approximately $1.2 billion led by 8VC. | 高 | SV003, SV006, SV007 |
| CV005 | XDOF's reported $1.2 billion valuation discussion emerged less than three months after the company exited stealth mode in June 2026. | 高 | SV003, SV006 |
| CV006 | Venture capital interest in a rapid Series B was reportedly accelerated by XDOF's annualized revenue approaching $50 million. | 高 | SV006, SV007 |
| CV007 | XDOF serves approximately 20 active customers, primarily comprising frontier AI laboratories and robotics development firms. | 高 | SV003, SV006, SV008 |
| CV008 | At an unconfirmed $1.2 billion valuation against a $50 million annualized revenue run-rate, XDOF's implied multiple stands at roughly 24x. | 中 | SV006, SV007 |
| CV009 | Private physical and digital AI data benchmarks include Scale AI at $13.8 billion valuation, Mercor at $10 billion, and Encord raising $60 million. | 中 | SV007 |
| CV010 | A $70 million first round creates structural pressure by committing the company to an aggressive valuation hurdle that must be justified by commercial milestones. | 中 | SV005 |
| CV011 | Co-investor syndicate structures where brand-name funds split initial allocations dilute lead responsibility, creating follow-on financing risk if subsequent rounds stall. | 中 | SV005 |
| CV012 | XDOF operates a team of approximately 60 employees alongside global human operators steering robots remotely and wearing motion-capture sensors. | 高 | SV003, SV006, SV007, SV008 |
| CV013 | XDOF partnered with UC Berkeley's AI Research lab and MIT to release the ABC-130K open-source bimanual manipulation dataset comprising 130,000 trajectories across 195 tasks. | 中 | SV002, SV003, SV008 |
| CV014 | Investment recommendation is designated as Track / Hold due to stretched headline valuation multiples, unverified gross margin durability, and labor-intensive teleoperation overhead. | 中 | SV005, SV006, SV007 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | TechCrunch | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | TechCrunch | XDOF was co-founded by UC Berkeley researchers Philipp Wu (CEO) and Fred Shentu (CTO) in 2024. TechCrunch reported on the startup’s $70 million Series A in June, with participation from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. |
| SO002 | toflow.ai | What Funding Has XDOF Raised? Full Round History & Key Investors | XDOF is a robotics infrastructure company that builds the data pipelines, teleoperation tooling, and annotation systems used to train robot foundation models. |
| SO003 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | Competitive first rounds often end with several brand-name funds sharing allocation. It is a strong signal, with one practical consequence: no single fund owns enough to feel fully responsible for the next round. |
| SO004 | Yahoo Finance | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | XDOF was co-founded by UC Berkeley researchers Philipp Wu (CEO) and Fred Shentu (CTO) in 2024. TechCrunch reported on the startup's $70 million Series A in June, with participation from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. |
| SO005 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A — TFN | Wu, Shentu and a third co-founder, Nemo Jin, launched XDOF in October 2024. |
| SO006 | Altis Research | XDOF — Robotics Data Infrastructure | Altis Research | Founded by UC Berkeley researchers behind the GELLO teleoperation project, the company emerged from stealth in June 2026 with a $70M raise and the largest open-source bimanual robot manipulation dataset. |
| SO007 | Preqin | xdof, Inc. Asset Profile | Preqin | Based in San Mateo, US, founded in 2024 by its co-founders Philipp Wu, Fred Shentu, and Nemo Jin. |
| SO008 | XDOF, Inc. | Privacy Policy | XDOF | This Privacy Policy describes how xdof, Inc. ("Company," "we," "us," or "our") collects, uses, and discloses information associated with an identified or identifiable individual ("Personal Information") and what choices you have around this activity. |
| SO009 | Built In | XDOF Jobs + Careers | Built In | Own end-to-end robotics data collection and annotation projects, translating client training needs into delivery specifications. |
| SM001 | XDOF | XDOF | |
| SM002 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | Large first rounds buy time to build something expensive, and commit you to a valuation you must grow into. xdof's $70M Series A is its entire lifetime capital. |
| SM003 | Yahoo Finance | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | Unlike LLMs, which initially trained on the entirety of the internet, physical robots don't have an equivalent real-world dataset to draw from, making data collection a critical bottleneck to building general-purpose machines. |
| SM004 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A — TFN | XDOF is effectively a data supply chain for robotics. Frontier AI labs know how to build language models; almost none of them have built the infrastructure to collect physical-world training data at scale, and XDOF is betting they’d rather pay someone else to do it. |
| SM005 | Altis Research | XDOF — Robotics Data Infrastructure | Altis Research | XDOF TAM ... How large XDOF's addressable market is, sized bottom-up as TAM, SAM and SOM rather than taken from a top-down analyst figure. |
| SM006 | Preqin | xdof, Inc. Asset Profile | Preqin | xdof targets a diverse customer base, which includes both residential users seeking automation solutions and businesses that require robust robotic assistance for manufacturing and logistics. |
| SM007 | Built In | XDOF Jobs + Careers | Built In | Leads the build-out and operation of XDOF’s LATAM robotics data-collection network in Mexico or Brazil. Responsibilities include expanding collection venues, sourcing partnerships, recruiting and managing field teams, deploying and maintaining hardware, optimizing operational efficiency, monitoring data quality, and converting market learnings into reusable SOPs and playbooks. |
| SM008 | Crypto Briefing | XDOF in talks for Series B funding at $1.2B valuation just months after exiting stealth | Founded in October 2024 by Philipp Wu (CEO), Yide (Fred) Shentu, and Nemo Jin, XDOF structures its offerings around what it calls a “data pyramid.” XDOF currently employs around 60 people and counts approximately 20 customers, mostly frontier AI labs and robotics firms. |
| SM009 | CV Rocket | Member of Technical Staff, Humanoid Engineer — xDOF at Thrive Capital portfolio, San Mateo, CA, USA | Design, prototype, and test humanoid robotic systems, including mechanical, electrical, and software components. ... Develop and implement control algorithms for dexterous manipulation, locomotion, and human-robot interaction. |
| SP001 | Masternode AI | XDOF in talks for $1.2B Series B just 3 months after stealth exit | Competitors include Mecka AI, as well as human-data platforms expanding beyond LLMs such as Scale AI and Micro1. |
| SP002 | Altis Research | XDOF — Robotics Data Infrastructure | Altis Research | This Altis research is built from our knowledge graph of sector expert calls, synthesized with public sources. |
| SP003 | Yahoo Finance / TechCrunch | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | Other startups attempting to collect real-world data for robot training include Mecka AI, as well as human-data platforms expanding beyond LLMs, such as Scale AI and Micro1. |
| SP004 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A — TFN | XDOF, along with rivals like Mecka AI and Encord, which raised $60 million to chase the same physical-AI data problem, is trying to solve something harder: there’s no internet-scale archive of the physical world to scrape. |
| SP005 | Startup Fundraising Library | The $70M Series A: What xdof's First Round Shows | Large first rounds buy time to build something expensive, and commit you to a valuation you must grow into. xdof's $70M Series A is its entire lifetime capital. |
| SP006 | Preqin | xdof, Inc. Asset Profile | Preqin | The company will use June 2026 funding to develop data pipelines, data collection tools, and annotation systems for robot training, while also expanding teams responsible for data collection and processing. |
| SP007 | Built In | XDOF Jobs + Careers | Built In | Own end-to-end robotics data collection and annotation projects, translating client training needs into delivery specifications. Coordinate international teams across data operations, engineering, algorithms, and annotation, while managing schedules, quality, accuracy, SLAs, costs, and issue resolution. |
| SP008 | CV Rocket / Thrive Capital Portfolio | Member of Technical Staff, Humanoid Engineer — xDOF at Thrive Capital portfolio, San Mateo, CA, USA | Design, prototype, and test humanoid robotic systems, including mechanical, electrical, and software components. |
| SP009 | XDOF | Privacy Policy | XDOF | This Privacy Policy describes how xdof, Inc. ("Company," "we," "us," or "our") collects, uses, and discloses information associated with an identified or identifiable individual ("Personal Information") and what choices you have around this activity. |
| SI001 | Value Add VC | XDOF Valuation 2026: $1.2B Series B Talks, 3 Months Out of Stealth | Human-data-collection companies have historically carried thinner margins than software businesses because the underlying cost — paying people to perform, wear sensors, or teleoperate equipment — scales roughly linearly with volume, unlike a software product that gets cheaper to serve per unit as it grows. |
| SI002 | The Tech Buzz | XDOF in Talks for Series B at $1.2B Valuation | Your source for technology news, insights, and analysis. |
| SI003 | Preqin | xdof, Inc. Asset Profile | Preqin | xdof generates revenue primarily through direct sales of its robotic systems and associated software licenses. The business model includes a structure where customers can purchase individual units or engage in long-term partnerships for ongoing service and support. |
| SI004 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | The most common failure after a large first round is hiring to the balance rather than to the milestone. Twelve months later, burn is structural and options are gone. |
| SI005 | Yahoo Finance | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | XDOF wasn't planning to raise again so soon after that round. But the company's rapid growth — with annualized revenue approaching $50 million — prompted VCs to approach it about a new round, the people said. |
| SI006 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A — TFN | It has around 60 employees and about 20 customers, including several frontier AI labs it won't name. |
| SI007 | Altis Research | XDOF — Robotics Data Infrastructure | Altis Research | XDOF builds the data-collection tools, pipelines, and annotation systems that frontier labs rely on to train robot foundation models. Founded by UC Berkeley researchers behind the GELLO teleoperation project, the company emerged from stealth in June 2026 with a $70M raise and the largest open-source bimanual robot manipulation dataset. |
| SI008 | Built In | XDOF Jobs + Careers | Built In | Own daily accounting operations, including general ledger, payables, receivables, fixed assets, payroll accounting, and reconciliations. Lead monthly, quarterly, and annual close processes; prepare financial statements and management reports; ensure GAAP, tax, and internal-control compliance; coordinate audits; investigate variances; support budgeting and forecasting; and improve accounting policies, systems, and processes in a global technology environment. |
| SI009 | XDOF | Privacy Policy | XDOF | xdof, Inc. ATTN: General Counsel 1875 South Grant Street, Suite 550 San Mateo, CA 94402 |
| SE001 | TAO Media | What Is XDOF? The Robot Data Startup Building Infrastructure For Physical AI | XDOF is trying to build that missing layer. The company collects and organizes real-world teleoperation data, develops data pipelines and annotation systems, and supports evaluation workflows for robotics companies and frontier AI labs working on physical AI. |
| SE002 | XDOF job posting republished by CV Rocket | Member of Technical Staff, Humanoid Engineer — xDOF | Design, prototype, and test humanoid robotic systems, including mechanical, electrical, and software components. |
| SE003 | XDOF job postings republished by Built In | XDOF Jobs + Careers | Own the end-to-end user experience for XDOF's data and research platform. Build self-serve data acquisition flows, commercialize internal tooling with infrastructure, design visualizers and exploration tools for large datasets, and collaborate closely with product and design while operating production web services. |
| SE004 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | Large first rounds buy time to build something expensive, and commit you to a valuation you must grow into. xdof's $70M Series A is its entire lifetime capital. Treat the size as an obligation with a deadline, not as validation. |
| SE005 | Yahoo Finance / TechCrunch | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | The startup aims to build the data pipelines, collection tools, and annotation systems that frontier AI labs and robotics companies can't easily build themselves, essentially acting as an outsourced data-supply chain for the robotics industry. |
| SE006 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A | Its platform combines teleoperation tools, human data collectors and annotation systems: some people remotely pilot robotic arms to demonstrate tasks, others wear body sensors to record how humans move through everyday chores like folding laundry. |
| SE007 | Altis Research | XDOF — Robotics Data Infrastructure | XDOF builds the data-collection tools, pipelines, and annotation systems that frontier labs rely on to train robot foundation models. |
| SE008 | XDOF | Privacy Policy | We maintain commercially reasonable and appropriate measures to secure collected Personal Information, including physical, electronic, and administrative safeguards. However, no data transmission over the Internet, nor any data storage system, is 100% secure. |
| SE009 | Preqin | xdof, Inc. Asset Profile | xdof specializes in the development of advanced robotics systems aimed at automating tasks that are typically mundane or undesirable for humans, such as dishwashing, laundry, and electronics assembly. The company’s core product offerings focus on robotics foundation models that leverage recent advances in generative modeling technology to enhance the capabilities of machines in performing household and industrial tasks. |
| SE010 | aVenture | XDOF Products, Competitors and Risks - aVenture | XDOF Products, Competitors and Risks - aVenture |
| SU001 | Vaia | xdof.ai – Vaia – Talents | Product Engineer, Robotics Data Platform UX |
| SU002 | Built In | XDOF Careers, Perks + Culture | XDOF is the infrastructure partner for the world's most ambitious robotics builders, developing the tools, data, and services that accelerate the future of physical AI. |
| SU003 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | Large first rounds buy time to build something expensive, and commit you to a valuation you must grow into. xdof's $70M Series A is its entire lifetime capital. Treat the size as an obligation with a deadline, not as validation. |
| SU004 | Yahoo Finance / TechCrunch | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | XDOF previously told TechCrunch that it is already working with 20 customers, including several frontier AI labs. |
| SU005 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A — TFN | Whether XDOF’s data-collection model scales as fast as the checks investors are reportedly willing to write is the open question nobody, including XDOF and 8VC, who aren’t talking yet, has answered. |
| SU006 | aVenture | XDOF Products, Competitors and Risks - aVenture | XDOF Products, Competitors and Risks - aVenture |
| SU007 | Altis Research | XDOF — Robotics Data Infrastructure | Altis Research | Customer count at funding announcementabout 20 active paying customers |
| SU008 | Preqin | xdof, Inc. Asset Profile | Preqin | xdof generates revenue primarily through direct sales of its robotic systems and associated software licenses. |
| SR001 | SME Business Review | XDOF Nears $50 Million in Annualized Revenue as Demand for Robot Training Data Grows | The financing terms are not final, and the amount being raised has not been disclosed. |
| SR002 | Tracxn | Xdof | Xdof has raised $70M in funding |
| SR003 | XDOF | Privacy Policy | However, no data transmission over the Internet, nor any data storage system, is 100% secure. While we strive to protect your information, security cannot be ensured or warranted. |
| SR004 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | Large first rounds buy time to build something expensive, and commit you to a valuation you must grow into. xdof's $70M Series A is its entire lifetime capital. Treat the size as an obligation with a deadline, not as validation. |
| SR005 | TechCrunch | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | The terms of the deal are not final and could still change. |
| SR006 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A | Whether XDOF’s data-collection model scales as fast as the checks investors are reportedly willing to write is the open question nobody, including XDOF and 8VC, who aren’t talking yet, has answered. |
| SR007 | aVenture | XDOF Products, Competitors and Risks | XDOF Products, Competitors and Risks - aVenture |
| SR008 | Altis Research | XDOF — Robotics Data Infrastructure | XDOF builds the data-collection tools, pipelines, and annotation systems that frontier labs rely on to train robot foundation models. |
| SV001 | FundedIQ | xdof: Funding, Investors & Team (Sep 2026) | FundedIQ | xdof has raised $70.0M across 1 funding round on record, the most recent an undisclosed round of $70.0M announced in Jun 2026. |
| SV002 | InforCapital | XDOF - Robotics, $70M Raised - Investors & Rounds | InforCapital | The company reached Seed / Series A stage in June 2026 with a $70M round led by Thrive Capital, Spark Capital. XDOF's investor base includes 5 institutional funds, with Thrive Capital leading the most recent equity round. |
| SV003 | Crypto Briefing | XDOF in talks for Series B funding at $1.2B valuation just months after exiting stealth | XDOF closed a $70M Series A just two to three months ago, backed by Thrive Capital, Spark Capital, Andreessen Horowitz, Lux Capital, and WndrCo. Total funding to date sits around $78M when you include an earlier seed round. |
| SV004 | Fundz | xdof raises $70 Million funding round (June 2026) | XDOF, a company building data infrastructure for robotics, has emerged from stealth with $70 million in new funding. The capital will be used to build out data pipelines, collection tools, and annotation systems. |
| SV005 | Startup Fundraising | The $70M Series A: What xdof's First Round Shows | Large first rounds buy time to build something expensive, and commit you to a valuation you must grow into. xdof's $70M Series A is its entire lifetime capital. Treat the size as an obligation with a deadline, not as validation. |
| SV006 | Yahoo Finance | XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation | Less than three months after emerging from stealth, XDOF, a startup that collects real-world teleoperation data for training general-purpose robots, is in late-stage talks to raise a Series B at a valuation of about $1.2 billion valuation led by 8VC, several people with knowledge of the deal said. |
| SV007 | Tech Funding News | XDOF in talks for new funding at $1.2B valuation months after $70M Series A — TFN | XDOF is in late-stage talks to raise a Series B at a valuation of about $1.2 billion, with 8VC reportedly set to lead the round, according to people familiar with the deal cited by TechCrunch. |
| SV008 | Altis Research | XDOF — Robotics Data Infrastructure | Altis Research | XDOF builds the data-collection tools, pipelines, and annotation systems that frontier labs rely on to train robot foundation models. Founded by UC Berkeley researchers behind the GELLO teleoperation project, the company emerged from stealth in June 2026 with a $70M raise and the largest open-source bimanual robot manipulation dataset. |