初创公司尽调
尽调报告 industrial / robotics Series A 2026-06-11

Hark

资本很充足,但这家 AI 硬件与个性化机器人创业公司仍未完成验证;其 $6 billion Series A 估值领先于已披露的客户、产品和经济性证据。

Hark 创始人履历、资本和算力渠道都很强,但公开记录仍支持回避判断:$6 billion 的 Series A 轮估值已经跑在已披露客户、产品和经济性证据前面。

封面要素

成立时间 01
2025 [CO034]
总部 02
San Jose, California, USA [CO022]
公开客户证明 06
none disclosed [CU013, CV007]

公司概况

Hark 是 Brett Adcock 于 2025 年末创立、位于 San Jose 的私营 AI 与机器人公司。公开材料称,公司正垂直整合自研多模态模型、持久记忆、软件体验以及后续 AI-native 硬件,打造高级个人智能。公司于 2026 年 3 月公开亮相,披露上线前由创始人自筹资金,随后在 2026 年 5 月宣布完成超额认购的 $700 million Series A,投后估值 $6 billion。它尚未披露的内容同样重要:具名客户、定价、产品规格、确认收入和单位经济性都仍未进入公开记录。

官网
hark.ai
创始人
Brett Adcock
创立地点
San Jose, California, USA
总部
San Jose, California, USA
产品
Hark 正在搭建个人智能栈,把多模态 AI 模型、软件体验、智能体工具、持久记忆以及后续 AI-native 硬件设备组合起来,目标是成为人与机器之间的通用接口。
客户
公开证据支持其未来面向消费者和 prosumer 的个人 AI 体验,并可能延伸到企业和工作流场景,但公司尚未公开披露经验证的付费客户细分。
商业模式
可能的变现路径包括付费软件体验、订阅或按用量计费的 AI 服务,以及最终的 AI-native 硬件销售,但公司尚未完整公开定价或收入模型。
阶段
Series A
融资情况
2026 年 5 月完成超额认购的 Series A,融资超过 $700 million,投后估值 $6 billion;此前公司初始搭建期由创始人大约自筹 $100 million。
[CO006, CO007, CO010, CO016, CO017, CO022, CE002, CE005]

执行摘要

主要优势

  • Brett Adcock 给 Hark 带来创始人-市场契合、融资可信度,以及大多数新机器人创业公司拿不到的人才、算力伙伴和战略投资人入口。
  • Hark 披露了 AI 硬件领域最大规模的早期融资之一,资源厚度罕见,足以推进模型加硬件的垂直整合路线。
  • 公开产品愿景不是单一窄口径硬件,而是围绕多模态模型、记忆、智能体和 AI 原生硬件拼出一条连贯路线。

主要风险

  • 公开证据里还没有客户、定价、收入、毛利率或单位经济性,无法用经营基本面支撑当前估值。
  • Hark 要同时自研模型和硬件,难度很高,落地取决于稀缺算力、供应链执行和安全产品交付。
  • 公司高度围绕 Brett Adcock 展开,公开披露的管理层梯队仍然偏薄。
  • 可比 AI 设备和机器人项目显示,热度可能跑在产品-市场契合、安全成熟度,以及消费者或企业可接受 ROI 前面。

未决问题

  • 具名付费客户、部署指标、留存信号,以及 Hark 买方群体不止停留在早期尝鲜的证据。
  • 产品规格、首发市场细节、定价,以及首个商业版本里硬件与软件各自承担的准确角色。
  • Series A 轮背后的已确认收入、烧钱速度、现金跑道、毛利率、算力承诺和股权结构条款。
  • 独立估值证据,例如老股交易、客户支撑的 KPI,或能缩小 Hark $6 billion 定价与证据更充分同业差距的可比公司。

目录

Chapter 01

01公司概况

1.1 身份、产品逻辑与商业模式

Hark 把自己定位为一家人工智能实验室,要打造「世界上最先进的个人智能」,而不是一家狭窄的应用公司。从官网、2026 年 3 月发布稿到 2026 年 5 月融资公告,公司反复讲同一套核心架构:自有基础模型、软件系统和专用硬件一起开发,作为人与机器之间的通用接口。目标产品是多模态且具备智能体能力,整合语音、文本、视觉、上下文感知和持久记忆,让系统主动管理用户数字环境,并最终延伸到物理环境的部分任务。因此,Hark 更接近下一代个人计算平台,而不是传统 SaaS 助手。管理层称,首批软件体验和模型将在 2026 年夏季推出,AI-native 设备随后跟进。但截至本报告日期,Hark 尚未披露客户、定价或变现细节,商业模式应被视为尚未商业化、由投资逻辑驱动,而不是已经被公开牵引力验证。[CO001, CO002, CO003, CO004, CO005, CO006]

FO002: 公司快照逻辑

创始人、资本、算力、模型、硬件与披露缺口如何串起 Hark 投资逻辑。

[CO002, CO004, CO005, CO016, CO023, CO024]

1.2 创始人、领导层与关键人依赖

Hark 与 Brett Adcock 高度绑定,官方发布和第三方报道都把他列为创始人兼 CEO。这段履历很关键,因为投资人显然在承销过往执行力,权重不低于当下产品证据:Adcock 此前创立 Figure AI、联合创办 Archer Aviation,更早还创立 Vettery。Hark 公开材料还重点介绍 Abidur Chowdhury;他曾是 Apple 设计师,参与过 iPhone 和 Mac 项目,如今负责设计,把公司的个人智能逻辑转译成新界面。缺失的信息几乎同样重要。本章审阅的公开记录没有列出更完整的高管梯队、董事会,或 Adcock 角色与 Chowdhury 设计领导之外的独立治理结构。这种集中度带来实质性关键人风险:Adcock 仍公开关联 Figure,而 Figure 商业化说法引发的负面报道可能外溢到 Hark 的可信度。因此,投资人应尽调接班安排、创始人精力分配和治理控制,而不能假设团队深度已经匹配 Series A 的规模。[CO008, CO009, CO013, CO014, CO015, CO029]

领导层和创始人表
人员职位背景创始人-市场匹配 / 职能覆盖关键人物依赖
Brett Adcock创始人兼 CEOHark 创始人;此前创办 Figure AI、共同创办 Archer Aviation,并创办 Vettery。带来深厚融资可信度、AI 硬件雄心和此前建公司履历,明显锚定投资人信心。极高:公开层面 Hark 与 Adcock 绑定,他的注意力也分散在相邻项目上。
Abidur Chowdhury设计负责人前 Apple 设计师,参与 iPhone 和 Mac 项目;被招募来领导 Hark 设计。给 Hark 带来消费硬件和界面可信度,契合 AI 原生设备和环境计算的投资逻辑。高:公开材料将他列为另一个唯一被突出识别的 Hark 领导者。

覆盖不完整,因为已审阅公开材料只点名 Adcock 和 Chowdhury;完整高管名单、董事会名单或继任计划均未公开披露。

[CO008, CO009, CO013, CO014, CO015, CO033]

1.3 融资历史、投资人基础与利益相关方图谱

Hark 的资本形成异常前置。Observer 和 TechCrunch 称,公司于 2025 年末启动,约有 $100 million 来自 Adcock 自有资金。2026 年 5 月 21 日,Hark 宣布完成超额认购的 Series A,融资超过 $700 million,投后估值 $6 billion;本轮由 Parkway Venture Capital 领投,NVIDIA、AMD Ventures、ARK Invest、Brookfield、Greycroft、Intel Capital、Prime Movers Lab、Qualcomm Ventures、Salesforce Ventures、Tamarack Global 和 Align Ventures 参投。股权结构表具有战略意义:它把芯片供应商、风险投资机构和偏企业端的投资人组合在一起,与 Hark 垂直整合 AI 设备的投资逻辑相匹配。Qatalyst Partners 担任交易顾问。仍不透明的是所有权分布。公开材料没有披露 Adcock 自筹资金带来的稀释、投资人持股比例、本轮附带的治理权利,也没有说明是否存在侧函、老股交易或债务工具。结果是一段资本很充足、但尚不透明的融资故事。[CO010, CO016, CO017, CO018, CO019, CO020]

利益相关方或投资人地图
利益相关方角色控制或经济重要性尽调要求
Brett Adcock创始人、CEO 和初始资本提供方提供了大约 $100M 早期资本,并仍是公司背后的核心声誉资产。确认当前持股比例、投票控制、关联方安排,以及在 Hark 和 Figure 之间的时间分配。
Parkway Venture CapitalSeries A 领投方可能是 $700M+ 轮次的锚定治理声音,也是估值预期的关键塑造者。确认董事会席位、保护性条款、清算优先权和跟投储备策略。
NVIDIA投资人和算力供应商战略重要性不止资本,因为 Hark 在 NVIDIA B200 基础设施上训练。澄清供应承诺、定价、排他性,以及算力访问是否与融资里程碑绑定。
Intel Capital投资人增加半导体生态验证和企业分销邻近性。判断 Intel 是否拥有任何商业合作权、信息权或优先访问条款。
AMD Ventures投资人提供替代芯片生态信号,以及相对单一供应商叙事的可选性。询问 AMD 参与是否纯属财务投资,或与未来推理 / 部署硬件计划相连。
Qualcomm Ventures投资人暗示未来 Hark 硬件可能与边缘、移动或可穿戴设备分销相关。核实 Qualcomm 是否拥有路线图可见性、共同开发权或渠道杠杆。
Salesforce Ventures投资人引入企业软件连接性;如果 Hark 扩展到消费工作流之外,这可能很重要。测试 Salesforce 期待的是企业集成、智能体工具,还是纯组合暴露。
ARK Invest / Brookfield / Greycroft / Prime Movers Lab / Tamarack / Align 等投资人财务或主题型共同投资人扩大资本池,但公开材料未披露具体出资规模或权利。索取完整股权结构表、附函,以及按投资人划分的集中度。

地图只反映公开材料中具名的利益相关方;持股比例、老股交易和完整股权结构经济性均未披露。

[CO017, CO018, CO020, CO021, CO023, CO038]

1.4 规模信号、披露边界与运营依赖

公开规模指标很窄,但有意义。发布稿称 Hark 已从 Apple、Meta、Google、Tesla 和领先 AI 实验室组建了超过 45 名研究员、工程师和设计师;5 月融资文章称团队已增长到约 70 人。同一批材料称 Hark 已获得一座新的 NVIDIA B200 数据中心和数千块 GPU,用于支持多模态模型训练,进一步说明算力获取是一阶运营依赖。公开披露还显示公司位于 California 州 San Jose。除此之外,Hark 透露极少:收入、ARR、定价、毛利率、客户数或具名试点用户均未公开。公司转而描述其个人 AI 平台将在 2026 年夏季晚些时候开放早期访问,未来再推出硬件。出于尽调目的,Hark 应被视为一家重金投入、尚未推出产品的公司;人才和基础设施信号很强,但需求生成、转化或商业单位经济性的公开证据很少。[CO011, CO012, CO022, CO023, CO024, CO025]

快照 KPI 表
指标数值 / 状态日期置信度缺口 / 备注
成立20252025已审阅来源未公开确切注册日期和法律实体名称。
总部San Jose, California2026-05-21地点根据官方新闻稿发稿地和二手报道推断,而不是来自已发布地址页面。
阶段Series A / 产品发布前2026-05-21公司尚未披露商业发布完成。
Series A 估值60002026-05-21USD millions;官方投后估值。
外部融资额700+2026-05-21USD millions;官方 Series A 金额表述为 “over $700 million”。
已披露融资总额(含自筹)800+2026-05-21将据报道 $100M 创始人自筹加入 $700M+ Series A;确切总额取决于四舍五入。
发布时团队规模45+2026-03-24发布新闻稿说 “more than 45”,不是精确人数。
Series A 后团队规模~702026-05-21官方帖子使用了 “around 70 people”。
首次软件发布窗口Summer 20262026-05-21未披露确切发布日期或 beta 队列。
客户 / 试点未公开披露2026-06-11公司提到早期访问,但未具名任何客户、试点或 beta 伙伴。
收入 / ARR / 定价未公开披露2026-06-11未发现公开单位经济披露。

各行混合了已披露事实和明确披露缺口;美元数值单位为 USD millions,空缺经济数据代表未知,不是零。

[CO016, CO020, CO022, CO025, CO026, CO034]
FO003: 快照 KPI

截至运行日期的关键成熟度与尽调信号。

商业证据 KPI 是定性指标,因为 Hark 尚未公布客户、收入或定价指标。

[CO012, CO016, CO020, CO025, CO026, CO037]

1.5 里程碑、创始人风险语境,以及后续章节应复用的事实

Hark 很新,因此本章时间线很短,但影响已经不小。公司创立于 2025 年,2026 年 3 月通过发布稿公开亮相,并在 2026 年 5 月转化为 AI 硬件领域最大的 Series A 之一。公开里程碑包括 Chowdhury 加入并担任设计负责人、NVIDIA 支持的算力建设、员工数从约 45 人增至 70 人,以及承诺在 2026 年夏季推出软件。主要反向视角来自创始人关联和行业 hype,而不是 Hark 自身的具体运营失败。TechCrunch 报道称,Adcock 的另一家主要公司 Figure 围绕商业化说法和公开视频透明度受到质疑;更广泛的机器人报道也持续追问,人形机器人和 AI 硬件愿景能多快转化为可持续商业。后续章节因此应把以下事实作为底稿:Hark 是一家私营 Series A 公司,公开估值锚点是投后 $6 billion,已披露外部融资为 $700 million+,产品仍未发布,未来 12 个月是检验资本和履历能否转化为产品市场证据的关键窗口。[CO016, CO020, CO029, CO030, CO031, CO032]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2025-Q4Hark 成立,并以自筹资金走出隐身成立据报道 $100M 自筹Brett Adcock创始人资本在外部资金进入前资助了早期团队建设。
2026-03-24公司作为 AI 实验室公开发布产品发布公告Hark, Brett Adcock建立公开身份和产品投资逻辑。
2026-03-24披露设计负责人和早期团队治理合计点名 45+ 名团队成员Abidur Chowdhury;来自 Apple、Meta、Google、Tesla 的员工首次公开展示领导层深度和招聘策略。
2026-03-24详述垂直全栈策略产品模型 + 软件 + 原生硬件Hark澄清 Hark 打算掌握完整界面栈。
2026-04-01NVIDIA B200 集群计划于 4 月上线合作数千块 GPU / 算力扩张Hark, NVIDIA算力访问成为核心执行依赖。
2026-05-21宣布 Series A融资$700M+,投后 $6BParkway Venture Capital 及财团迄今最大 AI 硬件 Series A 融资之一。
2026-05-21融资后团队规模更新规模大约 70 人Hark显示发布到融资之间快速扩员。
2026-05-21软件平台承诺在 2026 年夏季推出产品计划中 / 尚未发布Hark设定第一个公开产品验证期限。
2026-05-26Forbes 将估值描述为资本和创始人护城河,而非牵引力证明反向外部分析批评Forbes突出估值跑在已披露产品证据之前。
2025-06-06TechCrunch 报道围绕 Figure 和 Adcock 透明度的商业化怀疑反向外部批评报道TechCrunch、Figure AI、Brett Adcock 公开材料创始人相邻信誉风险可能外溢到 Hark 尽调。

时间线在公开来源提供时使用确切日期;2025-Q4 成立和 2026-04 算力时点是根据报道推导的近似窗口,而不是已提交公司记录。

[CO010, CO011, CO016, CO023, CO031, CO034]
FO001: 公司里程碑时间线

从 2025 年隐身成立,到 2026 年 5 月融资,再到近期上线窗口的压缩时间线。

2025 年 10 月成立和 2026 年 7 月软件目标,都是根据来源材料中“2025 年末”和“今年夏天”的措辞推定的近似窗口。

[CO006, CO007, CO010, CO011, CO012, CO016]

1.6 展示材料

Chapter 02

02市场分析

2.1 市场边界与相邻领域

Hark 并未在公开材料中把自己呈现为仓库机器人 OEM。它自己的材料描述的是一套垂直整合的个人智能栈,组合多模态模型、记忆和专用硬件。做市场分析时,相关边界因此比单一机器人品类更宽,但比泛 AI 软件更窄。核心市场包括人形机器人和其他必须在人类既有空间中运行的服务机器人,以及让这些机器在工业和物流场景中有用所需的控制软件、部署服务、界面和数据系统。这是最接近 Hark 的公开市场框架,因为公司谈的是人与机器之间的通用接口,而不是已披露的一支固定功能仓库机器队伍。 同一边界也必须排除重要替代品。固定式工业机器人、AMR 和仓库软件自动化已经解决了同一劳动力与吞吐问题中的重要部分,只是不声称采用通用类人形态。Gartner 明确认为,多功能轮式系统在许多供应链任务中可能优于人形机器人;Amazon 已部署超过百万台机器人,也说明买方已经拥有规模化的非人形选项。因此,定义 Hark 机会的正确方式,是把它看作具身 AI、工业自动化和个人机器人之间的相邻层:如果它证明自己比专用替代方案更有用,就可能作为控制、界面或硬件层参与这些预算。 第二个相邻领域是个人和辅助机器人。Goldman 的乐观情景假设,消费机器人年销量最终可超过 100 万台;WHO 的数据则显示,人口老龄化正在形成巨大的未满足辅助技术需求。这并不等于养老护理或家用机器人今天就是 Hark 已验证的近期市场,但足以说明,个人机器人是一个真实的相邻支出池,不能简单当成科幻。[CM001, CM002, CM003, CM004, CM005, CM006]

市场定义表
细分 / 品类纳入支出排除支出买方 / 付款方与 Hark 的相关性
工业人形机器人机器人硬件、具身 AI 软件、部署、维护已为单一任务定制的固定工业单元工厂运营、自动化工程、制造业资本开支如果 Hark 成为具身控制或界面层,这是直接相邻市场
仓储和物流服务机器人拣选、料箱搬运、线边配送、编排软件仅传统 WMS 支出和人工 staffing供应链运营、履约负责人、物流 capex/opex最可信的近期企业邻近市场,因为买方已经在这里为机器人付费
个人 AI 原生硬件消费级或专业消费者硬件、多模态模型、设备服务没有原生硬件或持久记忆的通用聊天机器人个人用户或企业知识工作预算这是 Hark 今天最清晰表述的品类
辅助 / 个人机器人邻近市场辅助设备、陪伴和独立生活支持系统、无障碍服务没有设备或交互层的通用医疗支出家庭、照护者、医疗系统、公共项目长期邻近市场;未满足需求很大,但尚未证明是 Hark 核心市场
控制 / 界面软件层语音、视觉、记忆、编排、安全和机队界面没有直接人机交互的纯分析OEM、集成商、企业创新团队这是 Hark 公开定位和机器人预算之间最可防守的桥
现状替代品现有 AMR、轮式机械臂、固定机器人、工作流软件n/a同上买方这些替代方案已经吸收今天的自动化预算,并定义竞争基线

边界逻辑结合 Hark 自身定位与 Gartner、McKinsey、IFR、Amazon 和 WHO 证据;替代品被排除在核心市场之外,但保持可见,因为它们争夺同一批工业和个人技术预算。

[CM001, CM002, CM003, CM005, CM006, CM007]

2.2 规模测算视角与预测分散

最强的公开上行情景仍是 Goldman Sachs 的预测:到 2035 年,人形机器人总可用市场(TAM)可能达到 $38 billion,同期出货量达 1.4 million。这一预测很重要,因为它解释了资本为何持续流入具身 AI 和硬件平台。但同一来源也说明,不能用一个头条 TAM 概括这个市场。Goldman 的逻辑依赖大幅降本、更快商业化,以及最终的工业和消费端采用;而 Gartner 的 2026 年报告称,到 2028 年,真正进入制造和供应链生产使用的公司可能少于 20 家。两句话可以同时大方向成立:长期 TAM 很大,同时近期可服务市场很窄。 更受证据约束的规模测算框架,不应是单一数字,而是一组视角。第一,是 2035 年远期人形机器人 TAM。第二,是工业自动化当前装机基础和资本开支基线;IFR 仍预计到 2028 年,每年工业机器人安装量超过 700,000 台,并报告各地区自动化密度差异很大。第三,是相邻的个人和辅助需求池;WHO 预计,到 2050 年,需要辅助产品的人数将超过 3.5 billion,同时也记录了严重的可及性和劳动力缺口。具体到 Hark,公开证据还不足以拆出可信的可服务市场(SAM)或可获取市场(SOM):公司没有披露首个买方细分、定价模型、部署模型,也没有说明首个产品偏消费者、企业,还是机器人控制。结果是一个显然重要、显然可投、但仍过早,无法用一个宽泛估计可信定价的市场。[CM008, CM009, CM010, CM011, CM012, CM013]

TAM / SAM / 规模测算视角表
发布方年份地域数值CAGR / 时点方法置信度局限
Goldman Sachs2024 / 2035全球$38B 人形机器人 TAM;出货 1.4M 台2035 年展望长期可寻址市场和出货预测周期很长;取决于成本下降和商业化
Goldman Sachs2024 / 2030全球>250k 台人形机器人出货,主要面向工业2030 年基准情景工业采用的较近中期基准情景出货量不是收入,且仍取决于准备度
Goldman Sachs2024全球当前单台制造成本区间为 $30k-$150k当前成本区间从当前制造经济性推导的硬件成本视角成本区间不是总拥有成本
Gartner2026 / 2028全球超出实验阶段的 <100 家;投产的 <20 家到 2028 年供应链和制造供应商的商业化上限公司数量视角,不是支出
IFR2026 / 2028全球到 2028 年,年度工业机器人安装量 >700k 台~7% CAGR 2025-2028相邻资本开支预算的工业自动化基线不特指人形机器人
WHO2025 / 2050全球需要辅助产品的人群 >3.5B结构性长期需求个人 / 辅助机器人需求的邻近视角需求不等于可变现机器人支出

该表有意混合市场价值、出货量、商业化和需求视角,因为 Hark 没有可信公开 SAM 或 SOM。它保留了乐观 TAM 预测与狭窄近期生产证据之间的巨大分散。

[CM008, CM009, CM010, CM011, CM016, CM025]
FM001: 市场规模测算视角

在证据约束下拆分市场规模:长期品类 TAM 与更窄的近期可服务市场分开看。

前两层是品类级预测,第三层是商业化上限,第四层是 Hark 的明确尽调缺口,而不是建模数字。

[CM008, CM009, CM010, CM016, CM017, CM044]
FM002: 市场估算区间

公开估算横跨狭窄的近期商业化与宽泛的长期硬件价值,说明单一标题 TAM 容易误导。

前三行是以十亿美元计的市场价值视角;第四行是以千美元计的单机制造成本带,用来说明预测结果为何对成本假设高度敏感。

[CM008, CM011, CM043]

2.3 买方、用户与采用路径

人形和服务机器人近期的购买中心仍在工业和物流运营,而不是泛化的消费者预算。McKinsey 的仓库研究把韧性、安全、吞吐和劳动力挑战作为决策框架,这意味着运营、自动化工程和供应链负责人是天然买方。用户通常是现场主管或工厂流程负责人。付款方一般是工厂或供应链资本开支预算;如果供应商能提供机器人即服务或按拣选付费结构,有时会混入运营开支。采用路径因此高度分阶段:企业从狭窄工作流开始,验证可靠性和安全性,然后才考虑更大规模的机队 rollout。 公开部署证据强化了这一模式。BMW 和 Figure 先聚焦 Spartanburg 的一个车身车间任务,然后再把 physical-AI 试点扩展到其他地方。Agility 和 GXO 把 Digit 定位为产生收入的商业部署,但仍处在边界很清楚的物流工作流内。Apptronik 和 Mercedes 探索的是制造场景中的套件和料箱配送,同样是受控应用。换句话说,早期买方并不是抽象购买一台通用机器人;他们资助的是一个具体、重复、体力强度高、人体工学差或难招人的工作流。 对 Hark 来说,买方问题仍未解决。它的公开材料描述的是通用接口和个人智能,而不是具名工业工作流。因此存在两条可能的采用路径。一条是个人 AI 与硬件路线,之后连接到具身系统。另一条是企业工作流路线,即 Hark 的模型和原生设备成为第三方已部署工业机器人的控制层或交互层。现有证据尚不能证明哪个预算所有者、价格点或商业包装会先跑出来。[CM004, CM005, CM020, CM021, CM032, CM033]

细分 / 买方地图
细分买方用户付款方工作流预算负责人采用触发器
汽车 / 电子制造工厂自动化和车身车间工程产线主管、技术员制造业资本开支重复定位、线边供应、人体工学负担重的任务工厂总经理 / 运营副总裁劳动力质量、人体工学、可重复性、安全
3PL 和电商仓储履约或供应链负责人仓库现场经理、流程工程师物流资本开支或机器人即服务运营开支周转箱搬运、卸货、拣选支持、异常处理供应链 VP / COO吞吐压力、劳动力波动、韧性
工业创新团队企业转型或先进制造团队试点项目经理创新预算具身 AI 和柔性自动化试点评估CTO / 首席转型官需要测试下一代自动化,但不能全面重构
企业知识工作 / 个人 AIIT、创新团队或个人买家专业终端用户技术运营开支或消费者支出个人智能、多模态辅助、设备承载工作流CIO 或终端用户生产力提升和工作流卸载
辅助 / 护理生态医疗系统、公共项目、照护者老年人、残障人士、护理者公共报销、家庭支出、服务机构预算独立生活支持、沟通、移动、监测服务机构 CFO / 家庭人口老龄化和护理可及性缺口
机器人 OEM / 集成合作机器人制造商、系统集成商部署工程师和操作员合作或项目预算控制层、交互层或多模态界面集成产品 / 合作负责人需要更好的人机交互和编排

买家图谱把工业机器人买家与 Hark 仍未定义的商业化路径分开。公开证据目前能支撑工业 / 物流预算;至于 Hark 自身的个人机器人预算,只能算相邻方向,尚未验证。

[CM004, CM005, CM020, CM032, CM034, CM035]
FM003: 买方 / 细分市场与替代品地图

工业买方与个人技术买方的采用路径不同,也会把类人机器人与已规模化替代品比较。

该矩阵强调预算路径与替代品压力,而不是逐行重复表格。Hark 自己的首个买方仍未验证。

[CM005, CM018, CM020, CM034, CM036, CM039]

2.4 增长驱动、替代品与商业约束

增长逻辑很容易讲清楚。Goldman 称,组件可得性已经改善,成本下降快于预期;当工作危险、肮脏、枯燥或难招人时,工业需求最强。McKinsey 解释了物流买方为何在失败后仍继续追求自动化:他们仍需要韧性、安全、准确性和吞吐。IFR 的区域密度和安装量预测确认,工业自动化预算真实存在且仍在增长。WHO 的辅助技术数据则增加了更长期的个人机器人相邻机会,因为人口老龄化和护理缺口会持续扩大对支持独立生活技术的需求。 更难的问题是商业化时点。Gartner 和 IEEE 都认为 hype 跑在了准备度前面。Gartner 称,到 2028 年,多数人形机器人部署仍局限在严格受控环境,并警告多功能轮式机器人往往经济性更好。IEEE 把瓶颈说得更具体:电池取舍、关键运营中约 99.99% 可靠性的要求、平衡型机器人尚未解决的安全要求,以及一个基础需求问题——还没有设施级 killer app 在千台机器人规模上浮现。Rodney Brooks 还补充了更广泛的警示:原型成功不等于规模部署。 这组约束对估值的意义大于头部 TAM。今天的买方在把人形机器人与 Amazon 非人形机队等已验证替代品比较,而不只是与人工比较。如果品类转向高端控制层或界面层,Hark 仍可能受益;但公开证据尚未证明,市场奖励一个宽泛个人智能平台的速度,会快于奖励边界清晰、按工作流定制的自动化。因此,本章同时保留上行和矛盾:资本和预测在加速,但采用仍窄,基准数据仍薄,胜出的形态仍未定。[CM012, CM014, CM018, CM019, CM020, CM023]

增长驱动因素与约束表
驱动因素 / 约束方向时间含义尽调问题
组件成本下降更快,供应选项改善驱动因素近期支撑更多试点,也让 TAM 模型更有野心询问 Hark 硬件物料清单中的哪些部分同样受益于这条成本曲线
工业自动化仍是活跃资本开支类别驱动因素近期IFR 密度和装机增长显示,相邻预算已经存在判断 Hark 是卖进现有自动化预算,还是创造新的预算项
仓库韧性、安全和吞吐需求驱动因素当前即便项目失败,买家仍会继续投自动化找出一个 Hark 能比既有自动化方案改善经济性的工作流
老龄化拉动辅助技术需求驱动因素长期创造大型相邻个人机器人需求池澄清 Hark 路线图是否包含护理、无障碍或陪伴设备用例
需求集中在高度受控环境约束当前至 2028 年近期 SOM 被限制在狭窄工作流,而非通用部署量化 Hark 第一个受控用例和部署标准
多功能机器人通常给仓库带来更好的经济性约束当前人形和具身 AI 入局者必须打败强替代品,不只是替代人工在真实工作流中把 Hark 与轮式、非人形替代方案对标
可靠性、续航和停机容忍度仍是严苛门槛约束当前商业买家不会规模化不能达到正常运行时间门槛的系统向 Hark 或合作伙伴索取正常运行时间、充电、人工介入和维护指标
安全标准和责任模型仍在演进约束近期至中期合规工作可能拖慢在人机共享空间中的部署评估 Hark 是否参与标准制定、安全工具或合作伙伴合规栈

表格把公开需求信号与商业化阻碍放在一起看。如果替代方案、可靠性门槛和标准工作仍未解决,这个品类可以具备战略吸引力,但商业落点仍然很窄。

[CM012, CM018, CM020, CM023, CM025, CM026]
FM004: 采用漏斗或价值链地图

具身 AI 系统要从工作流识别走向规模化部署,买方必须先跨过可靠性、安全和 ROI 门槛。

漏斗值是指数化指标,不是绝对市场数量。它们综合了 Gartner、McKinsey、IEEE 和公开部署案例描述的商业化门槛。

[CM016, CM017, CM021, CM026, CM029, CM030]

2.5 展示材料

Chapter 03

03竞争格局

3.1 Hark 产品简报仍不清晰,却要面对非常清晰的替代选择

Hark 的公开材料描述了一套由多模态 AI 和原生硬件构成的人机通用接口,但尚未披露具体机器人形态、客户工作流、价格点或生产计划。这种模糊性很重要,因为投资人仍必须根据未来 Hark 系统可能要解决的任务来承销公司。工业和物流用例中,参照集合已经很拥挤:人形机器人一侧有 Figure、Tesla Optimus、Agility Robotics、Apptronik 和 Boston Dynamics;另一侧还有更大装机基础的非人形自动化,包括 Symbotic、Locus、GreyOrange、Berkshire Grey 和 Amazon。家庭和个人助理用例中,1X 是最清晰的公开人形相邻公司。Goldman 仍认为长期人形机器人市场很大,但 Gartner 和 IEEE 都指出,近期生产规模仍窄、经济性偏弱,买方经常更偏好每美元吞吐更高的专用机器人。因此,Hark 进入的不是空白品类,而是一组已经定义好的选择:用于灵活劳动力的直接人形机器人、追求运营 ROI 的多功能仓库机器人,以及面向个人辅助的 AI-native 设备概念。[CP001, CP002, CP003, CP005, CP007, CP008]

竞争对手画像表
竞争对手类别规模或融资信号目标细分市场差异化相比 Hark 的局限
Hark参照公司2026 年 Series A,估值 $6B;此前创始人种子轮 $100MAI 原生硬件 / 个人与物理 AI 界面Brett Adcock 旗下整合多模态 AI 与定制硬件愿景没有公开部署、定价、客户或具体机器人工作流证据
Figure直接人形机器人同业2025 年 Series C,估值 $39B制造、物流,最终进入家庭Helix VLA 品牌,加上 BMW 运行数据和 Figure 03 路线图公开经济性和已实现经常性收入仍很薄
Tesla Optimus直接人形机器人同业 / 可能入局者Tesla 内部项目,拥有重大制造资源先做内部工厂自动化,随后拓展更广泛人形机器人野心品牌、制造规模和投资者心智公开证据仍显示延期,工厂实用性也不清晰
Agility Robotics直接物流同业GXO 商业部署,合作伙伴名单扩大仓储、物流、制造Digit 加 Arc 工作流软件和 RaaS 模式公开销量和价格透明度仍有限
1X相邻家庭人形机器人同业OpenAI 支持的家庭人形机器人项目家庭辅助和个人机器人家庭优先的安全与自主叙事距离商业规模仍很远,对仓库买家的相关性较低
Apptronik直接制造同业Apollo 获 Mercedes 商业试点仓库、制造、更广泛工业任务面向人类空间设计的人体尺度机器人公开融资和部署深度不如 Figure 或 Tesla 的头条信息清晰
Boston Dynamics直接企业标杆Atlas 加更广的 Spot / Stretch 软件生态工业物料搬运和汽车本组中最强的公开工业规格和软件生态相比成熟的非人形产品,人形机器人的商业化仍早
Symbotic工业 / 物流替代方案Walmart 支持 APD 开发并承诺部署大型零售商、批发商、食品饮料供应链端到端交钥匙仓库自动化和 AI 软件不是面向人类设计任务的柔性通用人形机器人
Locus Robotics工业 / 物流替代方案大型客户名单和多次生产力案例研究3PL、零售、医疗履约现有仓库中的柔性 AMR 和货到人执行任务范围比宽泛的人形机器人承诺更窄
GreyOrange工业 / 物流替代方案横跨大型商业网络的仓库编排零售和全渠道仓储供应商无关的编排,加 AMR 式自动化更少聚焦通用具身智能
Amazon Robotics现状型在位替代方案2012 年以来部署超过 100 万台机器人Amazon 规模下的内部履约和分拣在拣选、上架、搬运、分拣工作流中跑通的专用系统主要供内部使用且高度专用,不是商售人形机器人平台

各行优先覆盖用户指定的直接人形机器人品牌,以及已经在仓库和工厂解决类似劳动力问题的替代系统。

[CP001, CP003, CP009, CP011, CP013, CP015]
FP001: 竞争定位地图

用序数刻画与 Hark AI 原生机器人野心的直接重叠度,以及公开部署证据。

坐标轴是有证据支撑的序数评分,不是市场份额或收入排名。

[CP001, CP009, CP012, CP013, CP015, CP018]

3.2 直接具身 AI 同行已经发布更强的部署或产品证明

在直接人形机器人同行中,Figure 目前提供了最强的软件雄心、现场证据和资本组合。其公开 Helix 材料把公司定位为板载视觉-语言-动作栈,BMW 更新则展示了 Hark 尚未公开匹配的真实产线运行时间。Tesla 更难判断,因为规模和人才显而易见,但最新可得报道仍描述了一个面临延误、领导层动荡,以及在 Tesla 工厂内用途未解的项目。对物流买方而言,Agility 看起来运营成熟度更高,因为 Digit 已进入 GXO 的付费部署,并配套 Arc 机队软件;Apptronik 则借 Mercedes 验证了「在人类设计空间中运行的人形机器人」叙事。Boston Dynamics 拥有最深的企业机器人积累,包括 Atlas 规格和 Orbit 工作流集成,即便其人形机器人商业化仍处早期。1X 对仓库买方没那么直接,但如果 Hark 更接近 AI-native 个人设备而非工厂劳动力平台,它就很重要。六家同行的共同点是:每一家都已经发布了比 Hark 更具体的产品、合作伙伴或试点证据。[CP009, CP010, CP011, CP012, CP013, CP014]

功能 / 能力矩阵
购买标准HarkFigureTesla OptimusAgility Digit1X NEOApptronik ApolloBoston Dynamics Atlas备注
公开工作流具体性Hark 尚未披露具体工作流;同业通常已经点名工业或家庭用例
工业部署证据部分Figure、Agility、Apptronik 和 Boston 都有合作伙伴或现场叙事;Hark 还没有
仓库软件 / 车队层UnknownUnknownAgility Arc 和 Boston Orbit 是最清晰的公开锁定层
家庭或个人 AI 取向1X 是最清晰的家庭优先同业;Hark 的界面话术也指向消费者相邻方向
AI 栈披露Figure 通过 Helix 给出了最清晰的公开 AI 软件品牌
定价透明度在这组证据中,直接人形机器人基本没有公开标价
人类空间设计优势理论上人形机器人同业都强调能在为人设计的环境中工作
替代方案挤压工业买家常常可以选择 AMR、穿梭车或机器人拣选,而不是人形机器人

单元格反映留存来源能清楚支撑的内容;未知或低往往意味着缺少公开证据,而不一定代表底层能力弱。

[CP001, CP009, CP013, CP015, CP018, CP020]
定价 / 包装对比
公司公开价格或合同模式商业化信号包含能力未知项或折价因素含义
Hark所审阅来源中未找到公开定价或合同模式有限 beta,没有公开客户部署多模态 AI 加原生硬件愿景具体硬件形态、买家和单台经济性仍未披露买家证据太早期,无法直接比较定价
Figure所审阅来源中没有公开标价BMW 产线部署和 Figure 03 过渡人形机器人硬件加 Helix AI 栈已实现 ASP、经常性软件收入和服务经济性未披露产品叙事很强,但商业条款不透明
Tesla Optimus所审阅来源中没有公开标价路线图关注度高,但公开报道仍指向延期Tesla 面向内部工厂工作的人形机器人,并有更广泛野心当前在 Tesla 工厂的实用性和真实量产时间仍有争议入局者很强,但运营证据仍难以承销
Agility Digit通过 GXO 采用 RaaS / 商业部署模式付费现场仓库部署,加上更多协议Digit 机器人、Arc 车队软件和集成服务公开车队规模和定价仍不透明服务主导模式可能降低买家摩擦,并提高切换成本
1X NEO早期访问,而非成熟企业合同模式原型机家庭测试和概念阶段扩张带基础自主能力和遥操作支持的家用机器人商业化时间、价格和安全指标仍不清楚与消费者相邻方向有关,不是当前工业采购标的
Apptronik Apollo所审阅来源中没有公开标价Mercedes 制造试点通用人形机器人,载荷 55 磅留存来源没有充分披露融资深度和重复部署规模切入人类空间制造可信,但仍早
Boston Dynamics Atlas企业产品模式,不公开标价Hyundai 现场测试和更广泛的 Boston Dynamics 软件生态Atlas 人形机器人加 Orbit 集成路径人形机器人收入规模和客户数未披露即便没有标价透明度,企业可信度仍高
Symbotic大型多年期自动化协议Walmart 出资开发,并有部署 400 个 APD 的选择权端到端交钥匙仓库自动化不是通用人形机器人;项目经济性按设施而异说明买家可以签大型合同,而不必押注人形机器人的柔性
Locus Robotics仓库自动化部署模式,有具名案例研究许多客户故事和可量化生产力提升AMR、编排、仪表盘、客户成功动作具体定价不公开,范围也按工作流而定替代方案可以在现有仓库内渐进落地
Amazon Robotics内部专用自动化技术栈已部署超过 1 million 台机器人覆盖拣选、上架、搬运、分拣和触觉感知的专用系统主要内部使用,并未作为第三方平台出售证明买方任务可以拆成多个专用系统,而不是押注单一人形机器人

关键比较点是商业化模式和验证,而不只是标价;多数直接人形机器人厂商仍依赖不透明的企业合同。

[CP002, CP003, CP010, CP011, CP013, CP015]
FP002: 商业取向 / 就绪度地图

压缩呈现直接同行更偏工业还是家庭用例,以及各自公布了多少商业证据。

这些值概括保留公开来源中可读出的商业取向与证据,而不是原始技术能力。

[CP012, CP013, CP016, CP019, CP021, CP023]

3.3 仓库和工厂替代品已经用更高证据解决买方任务

最大的竞争误判,是只把 Hark 与其他人形机器人创业公司比较。在仓库和工厂自动化中,买方并不需要人形机体才能证明采购合理。Symbotic 已经销售端到端仓库自动化,并拥有 Walmart 支持的发展和部署路径,面向数百个加速取货和配送站点。Locus 展示了 robots-to-goods AMR 的吸引力:它能进入现有设施,并有具名客户案例和测得的生产率提升。GreyOrange 把自己定位为覆盖仓库、门店和供应链的编排软件加机器人;Berkshire Grey 则直接自动化件拣选和拖车卸货。Amazon 是最清晰的证明:专用 physical AI 可以吸收大量仓库劳动力,而不必建造通用人形机器人。McKinsey 的仓库自动化综述进一步强化了这一点:运营方已经可以从成熟 AMR、货到人系统、穿梭车,以及降低资本摩擦的收费结构中选择。如果 Hark 想卖进物流或制造,它竞争的是已验证的吞吐系统,而不只是头条人形机器人品牌。[CP024, CP025, CP026, CP027, CP028, CP029]

FP003: 护城河 / 就绪度 KPI

五个压缩指标,总结 Hark 当前竞争风险最高的位置。

[CP003, CP011, CP016, CP031, CP041, CP043]

3.4 在 Hark 证明工作流锁定前,护城河耐久性很弱;仅有硬件雄心还不够

本章最强的反向证据不是竞争对手的丑闻,而是人形机器人有限的生产准备度与结构化工作中充足的替代自动化叠加。Gartner 认为,人形机器人仍昂贵、不成熟,在吞吐和正常运行时间上通常不如多功能机器人;IEEE Spectrum 则认为,真正瓶颈不是制造机器人,而是在规模上找到可靠、安全、经济性成立的需求。这些批评落到 Hark 身上尤其重,因为它的公开记录比直接同行更早期。Hark 有资本、设计人才和创始人可信度,但尚未展示定价、客户胜利、部署指标,或可与 Agility Arc、Boston Dynamics Orbit 相比的软件控制层。这意味着切换成本仍是假设。如果硬件成本继续下降,基础机器人机体存在商品化风险,价值会转向工作流软件、数据、服务和渠道合作。Hark 通向持久差异化的最佳路径,因此不是简单发出一台能力不错的机器人或设备;而是占住一个工作流:直接人形机器人尚未主导,替代品也还不能证明 ROI 更好。[CP007, CP008, CP034, CP035, CP038, CP039]

护城河耐久度 / 竞争风险登记表
护城河主张威胁严重性有证据支撑的理由缓释方式 / 尽调要求
集成 AI 加原生硬件Figure 已经用更强的现场验证讲同一个软件加机器人一体化故事Figure 有 Helix、BMW 运营数据和更高估值,而 Hark 仍没有披露具名落地工作流要求 Hark 展示其界面独有地改善了哪类用户工作流,以及为什么 Figure 不能快速补上
创始人履历和融资势头资金雄厚的同行和 Tesla 能在硬件、数据和人才上花得比 Hark 更多Hark 的 $6B 估值有分量,但 Figure、Tesla 和大型既有厂商仍拥有更多公开部署验证或资本杠杆要求提供招聘、算力和制造计划,说明 Hark 能在哪些环节做到同类最佳,而不只是资金充足
人形机器人在面向人类空间中的灵活性多功能机器人和仓储自动化替代品可能以每美元带来更高吞吐Gartner、McKinsey、Symbotic、Locus、GreyOrange、Berkshire Grey 和 Amazon 都支撑专用自动化逻辑要求给出工作流地图,说明哪些环节必须使用人形机体,而不只是可以接受
潜在个人 AI 和家庭场景相邻性如果 Hark 继续以设备为中心,1X 和其他消费级具身 AI 项目可能挤占叙事1X 明确优先家庭场景;Hark 公开描述个人 AI 和原生硬件,但没有固定工业切入点说明 Hark 优先做消费辅助、企业机器人,还是混合进入市场路径
工作流软件和车队锁定效应Agility Arc 和 Boston Orbit 已经定义了控制层切换成本机体只是企业锁定效应的一部分;集成、地图、故障排查和数据系统会随时间变得更重要在假设存在持久切换成本前,先要求 Hark 说明计划中的控制软件、数据护城河和集成 API
硬件成本下降组件成本下行会让基础机器人机体商品化Goldman 认为组件和物料清单成本会下降,价值因此转向数据、软件和渠道触达验证 Hark 是否有非硬件护城河,例如自有数据、服务流程或独家渠道触达

严重性是基于截至 2026-06-11 公开证据作出的承销判断,不是对最终市场份额的预测。

[CP005, CP007, CP008, CP012, CP016, CP021]

3.5 展示材料

Chapter 04

04财务情况

4.1 收入模式能见度:产品表面已公开,变现尚未公开

Hark 先让产品架构变得可读,却还没有让商业模式变得可读。官方材料持续描述一套垂直整合的个人 AI 平台,组合多模态模型、持久记忆和定制硬件;官网称公司已经在审核 beta 访问申请。隐私政策也从一个狭窄角度透露了财务线索:它提到网站、应用、产品和服务、支付卡信息、交易历史、沙盒任务执行,以及能在第三方页面间行动的浏览器操作员。这些披露说明,Hark 预期会出现涉及账户、交易和服务使用的商业工作流。它们没有展示的是公开价目表、订阅层级、API 费率、硬件价格,或任何未来订单的会计处理。结果是一种少见的清晰与不透明并存:公司明确想把软件和硬件一起变现,但确认收入、客户数、转化,或潜在需求与确认收入之间的区别,仍没有公开证据。因此,承销时不能用融资公告替代收入证明。[CI001, CI002, CI003, CI005, CI006, CI007]

收入流表
收入流机制单位当前数值 / 状态质量尽调要求
Beta 平台访问Hark 个人 AI 软件体验的早期访问账户官网称 Hark 正进入 beta 并审核申请;未披露公开变现方式提供 beta 转化漏斗、付费转化时间,以及访问是免费、邀请制还是基于订金。
软件 / 模型服务使用 Hark 应用、产品、服务和模型订阅或用量单位未知隐私政策和发布材料暗示存在软件服务,但未披露公开价格表或收入确认政策提供 SKU 清单、定价依据,以及订阅、使用费和递延收入的会计处理。
AI 原生硬件设备销售或融资购买专用设备设备硬件公开计划在软件发布后推出,但未披露 ASP、毛利率目标或首发市场提供产品路线图、目标 ASP、BOM、保修准备金假设和渠道模式。
交易 / 操作员工作流用户授权的动作、连接器和交易支持服务交易或席位未知隐私政策提到支付卡和交易历史,暗示可能存在或计划推出可收费工作流,但未公布费率表提供支付架构、抽成率、处理费和任何企业工作流打包方案。
合作伙伴 / 渠道经济与基础设施或分销合作伙伴的商业条款分成或承诺量未知投资方和基础设施伙伴是公开的,但未披露渠道收入、转售条款或最低采购量承诺极低提供已签商业协议、收入分成条款,以及任何硬件或算力预付款义务。

各行区分公开产品界面和实际变现。类似 null 的表述意味着 Hark 未公布标价、合同模板或已确认收入披露。

[CI001, CI002, CI005, CI006, CI013, CI015]
定价 / 变现表
价格 / 单位 / 合同标价 vs 实现价格折扣 / 未知项来源
Hark 平台 beta:无公开标价公开信息只有 beta / 申请访问状态未知 beta 是免费、付费,还是绑定后续订阅转化SI001
Hark 软件 / 模型:无公开订阅或使用费率无公开实现价格数据合同期限、用量计量、递延收入或企业折扣未知SI003
Hark 硬件设备:无公开设备价格或融资条款因尚未公开发布硬件,未披露实现价格ASP、捆绑、补贴、保修或附加销售经济性未知SI004
Humane AI Pin 代理:$699 标价后下调至 $499需求疲弱后,观察到品类价格压缩不能直接外推到 Hark 定价,但说明专用 AI 硬件可能需要大幅重定价SI020
NVIDIA B200 输入成本代理:8+ GPU 集群中每块 GPU MSRP 约 $30k-$40k这是 Hark 成本输入代理,不是 Hark 客户价格Hark 实际采购、托管和利用率条款仍未披露SI022

前三行是 Hark 相关信息,且大多未披露。最后两行是外部代理,用于观察品类定价压力和算力成本,不代表 Hark 实现收入。

[CI040, CI042, CI044, CI049, CI052]
FI001: 收入模型桥

公开证据显示 Hark 可能如何从 beta 访问走向软件和硬件变现,但每一个带价格的环节都未披露。

这座桥接图是定性的,因为公开记录只暴露产品界面和访问流程,没有公布变现条款或已确认收入。

[CI001, CI005, CI006, CI013, CI043, CI044]

4.2 GTM 动作与牵引力代理:beta 兴趣、招聘和投资人辛迪加可见,但销售效率不可见

公开记录支持的是一个发布前的 go-to-market 故事,而不是规模化商业故事。Hark 官网称平台正在进入 beta;招聘页则显示,公司正在 San Jose 总部招聘 AI、工程和设计岗位。3 月,公司称已有超过 45 名研究员、工程师和设计师;到 5 月,独立报道称团队约 70 人,并称新融资将用于招聘、算力和组件。投资人名单本身也是信号:芯片制造商、跨阶段投资人和基础设施级资本都参与本轮,说明 Hark 在大范围产品发布前,正组装供应、可信度和未来潜在分发。但这些都不等于商业牵引力。Hark 没有公开披露客户 logo、合同胜利、付费 beta 席位、发布地域、目标价格点、CAC、销售周期长度或回本周期。即便友好报道也强调,公司在发货产品和公布客户管线之前就完成了 Series A。因此,GTM 读数仅限于需求准备和人才扩张,而不是收入效率。[CI004, CI010, CI011, CI013, CI018, CI019]

单位经济表
指标数值 / null置信度为什么重要尽调要求
已确认收入 / ARRnull收入端验证是任何承销模型最核心的缺失输入提供月度已确认收入、ARR、订单额和递延收入桥表。
付费客户 / 付费账户null判断发布牵引力和收入集中度需要客户数量提供活跃用户、付费账户、销售管线,以及按客户类型划分的集中度。
硬件 ASPnullASP 决定定制硬件能否覆盖 BOM、保修和支持成本按设备提供目标 ASP、融资条款和预期软件套餐附加率。
毛利率null对算力密集且绑定硬件的模式,毛利率路径是关键问题按软件、硬件和公司混合口径提供毛利率。
算力成本代理数千块 B200 GPU;公开 B200 MSRP 代理约为每块 $30k-$40k,功耗 1000W即使不计客户支持或硬件销货成本,训练和推理经济性也可能很重要提供实际 GPU 采购或托管合同、折旧政策和利用率。
云端 B200 代理公开 B200 小时价格示例大约为每 GPU 小时 $2.80-$27.04在 Hark 未披露内部成本数据时,可用于给推理 / 训练运营成本划边界提供每训练小时、推理小时,以及每 token 或请求的实际混合成本。
员工数代理>45 名团队成员在 March;到 May 约 70 名员工收入仍未披露时,薪酬很可能是主要烧钱驱动按职能提供全负荷薪酬、股权激励和招聘计划。
公开融资可见度Form D 总发行额 $1.0B;到 May 已披露 >$700M Series A即使现金消耗仍是私有信息,已融资本是估算现金跑道最清晰的公开输入提供交割时间表、现金余额、受限现金和融资成本。

Null 表示未发现公司层面的公开披露。非 null 数值区间是来自公开来源的明确代理,不应误认为 Hark 披露的单位经济。

[CI010, CI012, CI014, CI019, CI037, CI039]
FI002: 单位经济模型桥接

可见的单位经济模型链条从稀缺算力和硬件工程延伸到未披露的定价与利润率层。

这座桥接图混合了 Hark 自身披露和明确的外部算力代理指标;缺失的私有指标保留为未知,而不是估算。

[CI012, CI020, CI021, CI027, CI042, CI048]

4.3 资本强度与融资依赖:融资可见,内部现金方程不可见

Hark 最清晰的财务披露是已融资本和基础设施雄心,而不是运营经济性。Hark Labs Inc. 2026 年 3 月 SEC Form D 显示,发行总额为 $1.0 billion;截至 2026 年 3 月 10 日首次销售日已售出 $50 million,文件中剩余 $950 million 可供发行,收入区间明确标为「Decline to Disclose」。两个月后,公司宣布完成超额认购的 Series A,融资超过 $700 million,投后估值 $6 billion。这一顺序很关键:公开记录显示融资动能和资产负债表支持,但仍未披露账上现金、月度烧钱速度、现金跑道,或任何债务或项目融资结构。资本强度也比毛利率更容易看见。Hark 3 月发布公告称,一个由数千块 NVIDIA B200 GPU 组成的大型集群将在 4 月上线;公开 B200 定价代理显示,这些系统价格高、耗电大,并且通常只能通过企业合同或稀缺云容量获得。把这一算力足迹与定制硬件开发配在一起,Hark 看起来更像一家算力加设备公司,而不是轻量软件创业公司。因此,Series A 看起来不像是给已验证模型的增长资本,更像是训练、工具、硬件和上市前时间所需的前置融资。[CI012, CI014, CI015, CI018, CI020, CI021]

资本充足性表
指标当前数值 / 状态有来源支撑的含义尽调要求
手头现金null公开来源未披露当前现金余额提供最新现金、现金等价物、受限现金,以及交割后实际收到的款项。
月度烧钱null虽然明显需要算力、招聘和硬件投入,但没有公开烧钱数据提供过去六个月净烧钱,以及产品发布后的预期烧钱。
现金跑道月数null没有现金和烧钱披露,无法推导现金跑道提供基准、下行和上行情景的现金跑道模型。
创始人种子资金TechCrunch 称 Brett Adcock 在 2025 年末向 Hark 注入 $100M 种子资金显示在外部收入验证前已有显著的 Series A 前资本支持提供股权结构表,并说明种子资金仍以现金留存,还是已投入早期资本开支。
Form D 授权发行额SEC Form D 于 2026-03-24 列示 $1.0B 总发行额显示即使在 May 轮次公布前,融资依赖度也已超过常规种子 / Series A 规模提供完整交割时间表,并说明该申报覆盖同一笔 Series A 还是平行证券。
截至 2026-03-10 Form D 已售金额已售 $50M;剩余 $950M;两名投资者显示截至首次销售日,仅申报金额的一小部分已经售出,意味着融资仍在推进提供逐批交割日期和投资者层面的已到账款项。
2026-05-21 披露的 Series A超过 $700M,投后估值 $6B,超额认购证明融资能力强,但不能证明收入质量或自我造血能力提供清算优先权、跟投权、里程碑约束和董事会条款。
计划资金用途招聘、算力 / 组件、模型和下一代硬件资金用途似乎指向资本密集型建设,而不是纯进入市场扩张提供横跨研发、硬件、算力和商业发布的资金用途预算。
下一轮触发条件未公开披露公开来源未指出下一次融资事件所需的收入、使用量或硬件里程碑提供内部融资计划和筹集下一笔资本所需最低指标。
债务 / 项目融资义务未披露公开债务;基础设施承诺仍不透明B200 集群和硬件项目可能涉及预付款、租赁或托管承诺,目前尚未公开提供所有债务、租赁、供应、托管和项目融资义务。

融资时间线被压缩为判断未来充足性所需的融资事实,不是完整历史轮次清单。Null 表示该指标未公开披露。

[CI014, CI018, CI020, CI035, CI038, CI039]
FI003: 财务估算区间

融资进展、团队规模、估值和 B200 成本代理指标有来源支撑的公开边界;收入和现金跑道没有。

这些只是公开边界和代理指标,不是 Hark 报告的收入、烧钱速度或利润率指标。

[CI014, CI019, CI039, CI042, CI051]
FI004: 资本强度 / 现金流图

Hark 财务模型里最可见的是资金用途和融资投入;经营现金回流仍不透明。

单元格标签是证据强度的序数摘要,不是私有经营指标。

[CI004, CI012, CI014, CI020, CI021, CI042]

4.4 财务结论:收入、毛利和硬件经济性缺口仍阻断承销

乐观情景很直接。Hark 融到了一笔异常巨大的资金,从顶级硬件和 AI 项目招来人才,并公开承诺搭建一套垂直整合栈;如果跑通,这套栈可能很难复制。悲观情景同样直接。公开证据仍没有显示确认收入、订单、付费用户数、实际定价、硬件 ASP、BOM、毛利率、质保准备金、烧钱速度或现金跑道。独立报道反复指出,公司很少披露发布市场、目标价格或客户管线;品类历史也提示谨慎:Humane 曾为 AI Pin 融资超过 $230 million,随后 HP 以 $116 million 收购关键资产,设备也被关闭。Hark 的执行可能远好于此前 AI 硬件尝试,但公开投资人还无法把雄心与经济性分开。正确的财务结论因此不是 Hark 没有商业模式;而是公开记录主要证明了融资能力和资本需求,同时让收入质量与毛利路径未被验证。尽调应聚焦三座桥:从 beta 需求到经常性软件收入,从定制硬件到正向单位毛利,以及从头条 Series A 到持久现金充足性。[CI022, CI025, CI027, CI040, CI041, CI044]

公开财务缺口表
缺失的私有指标影响精确尽调路径
已确认收入、ARR 和订单额无法区分融资势头和真实商业牵引力要求提供月度管理账、收入确认政策、订单额瀑布和递延收入明细。
付费用户和客户数量无法分析 CAC、销售管线转化和集中度要求提供包含 beta 用户、付费用户、企业试点和流失的 cohort 文件。
标价、折扣和合同结构阻碍任何实现价格或收入质量分析要求提供当前价目表、已签订单、折扣政策和合作伙伴转售条款。
硬件 ASP、BOM、保修准备金和退货假设阻碍硬件毛利率建模和营运资本分析按 SKU 要求提供 BOM、目标 ASP、保修计提、退货准备金和库存计划。
算力合同条款和利用率阻碍训练 / 推理成本分摊,以及资本开支 vs 运营开支评估要求提供 GPU 采购、托管、折旧、利用率和电力 / 冷却合同。
现金余额、烧钱和现金跑道无法承销偿付能力和融资依赖度要求提供月度现金桥表、董事会烧钱包和现金跑道情景模型。
债务、租赁、供应预付款或项目融资义务可能实质性改变有效现金跑道和稀释需求要求提供所有债务明细、租赁明细和供应商承诺函。
发布定价、首发市场和客户管线无法形成可信的首年收入预测要求提供发布计划、地域、渠道组合、预订或候补名单转化,以及销售管线复盘。

每一行都是公开记录止步于核心承销输入之前的阻断点。

[CI044, CI045, CI049, CI050, CI052, CI053]

4.5 展示材料

Chapter 05

05产品与技术

5.1 产品定义与当前工作流

Hark 的公开叙事对雄心讲得异常清楚,对已交付细节却异常稀薄。公司称其正在打造「高级个人智能」,把多模态模型、持久记忆和定制硬件组合成通用接口;官网也称平台正在进入 beta。判断今天到底有什么,最强证据不是产品演示或定价页,而是官网、融资文章、隐私政策和偏集成的招聘。合在一起,它们暗示了一套工作流:用户向 Hark 输入聊天、语音或文件;Hark 把这些上下文带入智能体会话;系统最终代表用户在外部工具和服务之间行动。 这与传统聊天机器人有实质差异,但并不等于已经公开记录的商业产品。公开来源没有披露设备形态、支持渠道或具名生产客户,官方 beta 页面也仍只是一个申请访问漏斗。结果是,产品故事在客户任务语言上最强——卸下数字任务、管理上下文、主动行动;但在运营细节上仍弱,买方或投资人还无法明确知道今天可用的是什么、哪些只是为后续硬件发布承诺的内容。[CE001, CE002, CE003, CE004, CE005, CE006]

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
个人 AI 平台 beta早期用户 / 候补名单申请者Beta 访问已开放;生产环境范围未披露承诺主动式个人智能,而不是窄聊天机器人需要实时 beta 功能清单、用户数和任务成功指标
Omni 多模态基础模型终端用户通过 Hark 体验使用计划 2026 年夏季推出;无公开模型卡文本、音频和视觉由预训练和后训练岗位共同覆盖需要基准测试、延迟目标、所有权地图和推理拆分
智能体运行时和计算机使用层委托数字任务的用户通过隐私和集成界面得到部分证明沙盒执行、浏览器操作员和工具 schema 暗示有真实执行动作的目标需要支持工具、权限模型、回滚控制和失败率
集成和连接器层连接现有服务的用户招聘证据强;公开产品文档稀少明确招聘覆盖电子邮件、日历、生产力、开发者工具、MCP、OAuth 和 webhook 支持需要连接器清单、API 文档、授权范围和限流处理
AI 原生硬件设备未来设备买家已公开承诺;未披露形态硬件从一开始就与模型和 OS 栈共同设计,而不是后加需要设备类别、BOM 区间、首发市场和 EVT / DVT 时间
嵌入式 OS 和音频平台硬件 / 固件团队招聘证据很强岗位中可以看到 AI 优先的 OS 架构、OTA、安全启动、DSP、麦克风、扬声器和常开监听需要主板架构、SoC 选择、电池目标和热设计空间
隐私和安全控制平面用户、监管方和内部运营运营意图有证据;外部保证较薄专门的隐私和安全工程岗位与政策中关于删除和审核的表述相匹配需要审计、认证、状态报告和事故历史

成熟度标签反映留存公开证据能支撑的内容,而不是 Hark 内部路线图信心。

[CE001, CE002, CE005, CE009, CE011, CE013]
工作流 / 用例表
用户任务当前工作流Hark 解决方案可衡量收益限制
卸下重复性数字行政工作用户通过聊天、语音或文件发出请求Hark 将自己定位为带记忆、能主动行动的智能体助手按公司定位,可能降低心智负担和上下文切换尚无任务成功率或节省时间的公开证明
跨现有工具工作浏览器会话、应用和第三方服务仍然割裂集成层意在连接电子邮件、日历、生产力、通信和开发者工具比独立聊天框拥有更广的行动面支持服务和权限粒度未公开
让 AI 在浏览器中行动用户仍登录本地服务Browser Operator 和沙盒执行暗示可在真实会话中委托任务可能无需从头重建用户工作流,就能实现实用自动化环境隐私和失败处理控制仍未披露
使用多模态交互语音、文本、文件和视觉上下文目前彼此分离Hark 称其模型结合语音、文本、视觉和上下文感知如果延迟和可靠性足够好,界面会更自然尚无模型卡或延迟披露
从软件走向设备原生访问当前 AI 大多存在于手机和聊天应用内Hark 承诺推出从一开始就与其模型共同设计的 AI 原生硬件可能比改造既有设备形成更紧、更低摩擦的交互循环硬件形态、人体工学或分销计划均未公开
用隐私和安全控制运营智能体系统带来数据处理和滥用风险隐私与安全岗位加政策指向删除、审核和事故工作流这是可信通用界面的必要基础无公开外部保证材料

收益仍是有来源支撑的方向性判断,不是已验证的 ROI 或用户采用指标。

[CE002, CE008, CE009, CE010, CE011, CE012]
FE002: 客户工作流 / 运营流程

Hark 设想的工作流从多模态输入和已记住的上下文开始,经过模型和工具路由;眼下通过软件返回结果,未来再接入硬件。

[CE002, CE008, CE009, CE010, CE011, CE024]

5.2 架构、训练与关键依赖

最具体的产品与技术证据来自法律和招聘页面。Hark 的隐私政策描述了一套系统:可接受多模态输入,可与第三方应用协作,可在既有会话中操作浏览器扩展,并可在隔离沙盒中运行智能体任务。招聘板进一步勾勒出图景:Hark 正在招聘预训练、多模态语音和视觉、后训练和 RL、大规模训练基础设施、集成、嵌入式软件、音频固件、隐私工程和 AI 安全岗位。这一组合暗示了一个垂直整合架构,至少有四个可见层:模型训练和数据管线;智能体运行时与连接器层;面向 AI-native 硬件的嵌入式设备栈;以及信任与安全控制平面。 公开证据也揭示了主要依赖。Hark 的基础设施岗位指向 10,000+ GPU 雄心,融资材料称其已锁定 B200 容量,因此算力获取是一阶依赖。连接器广度是另一项依赖:只有 Hark 能让工具使用在电子邮件、日历、浏览器和第三方 API 之间安全、快速、可靠地运行,产品才会变得真正有用。最后,硬件侧依赖 NPI 纪律、合同制造、可靠性测试,以及兼顾功耗和延迟的 OS 设计。这些依赖已经足够可见,可以画出地图;但还不够可见,无法承销为已经解决。[CE007, CE008, CE009, CE010, CE011, CE012]

技术 / 运营架构表
层级 / 流程 / 组件作用依赖风险
数据整理与合成数据循环筛选、去重并扩充多模态训练语料依赖数据权利、标注纪律和质量控制公开证据显示该环节存在,但无法判断数据护城河是否自研或持久
基础模型预训练栈构建核心文本、音频和视觉模型能力依赖分布式训练框架、基准测试和实验速度没有直接的基准测试或模型卡披露
大规模 GPU 基础设施承担预训练和后续推理负载依赖 B200 供应、集群可靠性、网络架构和 SLO 纪律算力集中度和成本都是重要依赖
智能体运行时和沙箱执行委派任务、文件和 shell 级工作流依赖稳健隔离、日志、回滚和滥用检测沙箱薄弱会带来严重安全和信任风险
连接器和工具协议把外部服务映射成可调用的模式和动作依赖 OAuth、webhook、API 标准化和合作伙伴在线稳定性模式漂移、授权过期和不稳定的第三方 API 都可能打断产品
嵌入式 OS 和设备栈协调 BSP、内核、中间件、应用、OTA 和音频子系统依赖芯片选择、电源管理、安全启动和软硬件协同设计公开证据主要来自招聘而非产品,准备度仍不确定
制造和工厂测试把原型转成可重复交付的出货设备依赖代工厂、可靠性测试和产线验证尚无 DVT/PVT 完成或现场可靠性的公开证据
隐私和安全控制平面处理删除、同意、审核和事件响应依赖法律解释、政策工具和生产可观测性控制失效会让环境式助手产品面临生死风险

本表根据法律页面、融资帖和开发者信号来源重建公开技术栈;它不是 Hark 内部架构图。

[CE009, CE010, CE012, CE014, CE015, CE016]
FE001: 产品架构图

公开可见的 Hark 技术栈从多模态用户输入和记忆出发,穿过模型、工具和嵌入式设备平台,运行时外层包着信任控制。

这是基于政策页面、招聘信息和公告重构的公开证据图,而不是 Hark 内部系统图。

[CE009, CE010, CE012, CE014, CE015, CE019]
FE003: 关键依赖图

Hark 要跑通通用界面投资逻辑,同时依赖算力、多模态数据、外部服务触达、设备工程、制造和信任控制。

[CE011, CE015, CE016, CE019, CE021, CE022]

5.3 创始人延续性与路线图可信度

Brett Adcock 在 Figure 的履历与 Hark 技术章节相关,但只能作为语境,不能证明 Hark 继承了 Figure 的实际技术栈。Figure 官方 Helix 和 Helix 02 披露展示了一套很具体的产品哲学:垂直整合智能、高频控制、跨模型层快速迭代,并在产品需要时愿意用学习系统替代手工工程逻辑。TechCrunch 关于 Figure 与 OpenAI 分手的报道,从 Adcock 本人角度强化了同一原则:如果硬件体验依赖核心 AI,就不能把核心 AI 外包。这一哲学与 Hark 自己关于模型、软件和硬件必须从一开始共同设计的说法高度一致。 但公开披露止步于哲学。保留来源没有显示 Hark 复用 Figure 代码、数据集、供应商或运营基础设施。更好的推断应更窄:Hark 受益于一位近期运营过整合 AI 与硬件项目、招募过顶级设计人才,并把相邻系统带入真实运行环境的创始人。这让 Hark 的路线图比随机 stealth startup 更可信,但不能抹去一个事实:Hark 仍处于设备前阶段、披露很轻,并且依赖未来执行,才能把架构意图变成可销售、可信任的平台。[CE016, CE017, CE026, CE027, CE029, CE030]

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
2025 年底Hark 成立,起初由 Brett Adcock 自筹资金已完成暗示公开发布前已有创始人资金支持的孵化Observer;TNW
2026-03-24Hark AI 实验室公开发布,包含多模态模型、个性化记忆、硬件和 B200 集群计划已宣布确立官方全栈产品逻辑,以及先软件、后硬件的第一步顺序Business Wire 发布
2026-03官网和 beta 访问已上线;招聘和隐私页面已发布官方界面已上线说明即便产品细节稀疏,软件 / 法务 / 招聘界面已经存在Hark 官网;招聘;隐私政策
2026-05-20 至 2026-05-21Series A 公布;团队约 70 人;重点提到新的 B200 数据中心已完成融资里程碑买来的是招聘、组件和基础设施时间,而不是 PMF 证明Hark 文章;Business Wire;TechCrunch
2026-06Greenhouse 岗位提到消费电子产品组合、大规模训练、集成、隐私和安全招聘信号已上线暗示架构正在模型、运行时和设备层面落地Greenhouse 职位板和岗位页面
2026 年夏季起AI 平台先行,硬件随后计划中承销应把当前证明视为设备前、路线图占比高Hark 文章;Business Wire Series A

日期只跟踪公开披露的里程碑;不代表内部工程关口已经达成。

[CE005, CE016, CE017, CE018, CE026, CE035]
FE004: 产品成熟度 / 能力图

公开证据最能支撑 Hark 的战略栈和招聘意图;对实时软件使用的支撑较弱,对已出货硬件证明和信任保障的支撑最弱。

[CE006, CE025, CE028, CE035, CE039, CE040]

5.4 信任、安全与披露缺口

Hark 值得认可的一点是,它比许多 stealth hardware 叙事释放了更多信任与安全信号。隐私政策有实质内容,不是占位文本,并且覆盖沙盒执行、浏览器操作员行为、连接器、删除权利和第三方 AI 提供商使用。隐私和安全岗位描述也说明,公司预期会面对围绕数据删除、去标识化、审核分类器、生产监控和事件响应的真实运营负担。这些都是有意义的产品信号,因为一个会读取页面、执行任务并记住用户上下文的通用接口,如果没有这些能力,商业上会失败。 但公开信任包仍不完整。没有保留下来的 trust center、状态页、认证清单、硬件安全报告、公开 SLA 或独立可靠性证据。产品侧更广泛地存在同一张力:Hark 显然已经为一套雄心很大的多模态、智能体、软硬件平台配置团队,但公开记录仍缺少模型卡、基准、设备规格、beta 使用指标和制造细节。出于尽调目的,核心承销结论因此是:对架构严肃性正面,对公开就绪证明负面。Hark 很可能确实在搭建它描述的技术栈;问题在于,外部观察者仍看不到足够信息,无法验证它到底走到了哪一步。[CE022, CE023, CE024, CE025, CE028, CE035]

信任 / 质量 / 合规表
控制 / 认证 / 质量领域状态范围缺口
隐私政策公开且内容具体覆盖账户、提示词、输出、沙箱执行、浏览器操作员、连接器和权利请求不能替代信任中心、审计包或系统架构披露
删除 / DSAR / 去标识化基础设施正在招聘建设Privacy Engineer 职位称 Hark 计划建设 DSAR、删除、留存、令牌化和去标识化系统尚无公开证据证明该栈已上线或通过审计
安全分类器和事件响应正在招聘建设AI Safety 职位称 Hark 计划建设审核分类器、生产监控和滥用缓解没有公开安全分类体系、滥用报告透明度或模型护栏指标
浏览器会话处理部分披露政策称 Hark 可以读取已授权页面,且不存储登录凭证没有公开说明凭证委派、密钥存储或浏览器扩展审核状态
安全启动 / OTA / 配置由硬件岗位暗示OS 架构岗位提到安全启动链、OTA 更新和生产配置没有公开设备安全白皮书或更新政策
外部保证材料保留来源中未公开可见未保留状态页、认证清单、SLA 或硬件安全文档在承销广泛消费者部署前,这是高优先级尽调请求

各行把有直接证据的控制项和仅由招聘暗示的控制项分开;后者仍需要第三方保证。

[CE022, CE023, CE024, CE025, CE037, CE038]
Chapter 06

06客户情况

6.1 买方、用户与付款方结构:公开记录指向先有直接账户,而不是具名企业买方

Hark 的公开页面在用户交互上比客户披露具体得多。官网描述了搭配定制硬件的个人智能系统;隐私政策则描述账户、支付卡信息、交易历史、提示词、输出、第三方应用、连接器、沙盒执行,以及能在用户自己的浏览器会话中行动的 Browser Operator。合在一起,这些披露暗示了一种直接账户关系:第一个可识别买方和付款方很可能是账户持有人,而不是公开具名的企业采购部门。用户看起来也是操作者:个人偏好、数据、浏览器上下文和已连接服务让系统变得有用的那个人。这并不排除未来企业采用,但确实意味着,在实际可得证据中,Hark 当前公开客户故事的结构更偏消费者或 prosumer 主导。即便招聘页也强调 AI、工程和设计,而不是可见的销售、实施或参考客户叙事。对客户尽调而言,这很重要,因为它会把本章从编造企业 logo 拉回更窄、基于证据的解读:Hark 似乎先在为个人或小团队账户构建,任何更广泛的买方层级仍未披露。[CU001, CU002, CU003, CU004, CU005, CU006]

客户分层表
分层可能买方可能用户可能付款方公开证据规模 / 战略价值关键缺口
消费者个人 AI 账户个人成年账户持有人同一名个人同一名个人持卡人官网和隐私政策描述个人智能、账户、银行卡、交易、连接器和浏览器动作公开支撑最直接的分层,因为账户和授权流程写得明确没有公开价格、发布地区、客户数或具名参考
准专业创作者 / 高阶用户工作流个人或小团队所有者执行语音、文件、研究或智能体任务的操作员个人或团队管理员隐私政策提到提示词、文件、智能体会话和第三方应用如果 Hark 解决日常工作流摩擦,可能产生早期高意图使用没有创作者、顾问或小企业部署的公开案例
企业知识工作者团队部门负责人或 IT / 运营发起人使用已连接服务的员工雇主官方材料称 Hark 会跨用户已经依赖的产品和服务运行如果管理、安全和审计控制到位,未来可能扩展 ACV没有披露具名企业试点、采购信号或管理员功能集
开发者 / 集成商账户产品或运营负责人配置工具的开发者或操作员雇主隐私政策提到连接器、沙箱环境和浏览器操作员工具可支撑聊天之外的嵌入式或工作流自动化用例没有 API 文档、开发者定价或实施案例研究
绑定硬件的家庭或团队套装消费者家庭买方或团队预算所有者佩戴或使用 Hark 原生设备的终端用户消费者或雇主发布和融资材料称软件先行、硬件跟随如果硬件成为首选界面,可能提高切换成本没有 SKU 清单、ASP、保修、渠道或附加率证据

各行把公开产品和政策界面暗示的分层,与已有具名客户证明的分层分开。没有部署证据时,战略价值仍是假设。

[CU001, CU003, CU004, CU005, CU006, CU007]
FU001: 客户旅程图

最能被证据支撑的 Hark 旅程,是一条由账户牵引的路径:先从发现进入可信权限,再走向更深扩张。

这是根据 Hark 产品和政策界面推断的定性旅程图。公开信息没有披露转化率、流失率或队列数量。

[CU003, CU004, CU006, CU007, CU015, CU016]

6.2 采用轨迹与具名证明:Hark 公开上仍处部署前,而相邻机器人同行已经发布具名运营参考

关于 Hark 的核心客户事实不是隐藏增长,而是缺少证明。Hark 自己的 3 月发布公告称,软件体验和 AI 模型将在 2026 年夏季到来,AI-native 硬件随后推出;5 月融资文章重复了这一时间。TechCrunch 到 5 月末仍称公司高度保密,Forbes 则明确指出,$700 million Series A 没有可命名客户,也没有可承销的牵引力。在审阅的 Hark 官网、发布材料、融资材料、隐私披露和融资报道中,公司没有公开识别试点客户、生产部署、企业设计伙伴、付费席位数、活跃账户数,或带结果的案例研究。这不能证明没有用户;它只能证明,公开记录尚不支持超出预期发布时间的采用主张。与相邻自动化市场对比时,缺口更清楚。Figure 发布 BMW 运行时间、吞吐和周期时间指标;Agility 和 GXO 描述了一个产生收入的商业化试点后部署;Apptronik 和 Mercedes 命名了具体制造任务;Symbotic 和 Walmart 披露了有条件的部署经济性;Locus 和 DHL 报告在 40 多个站点完成 10 亿次仓库拣选。这些不是 Hark 客户,本章只把它们作为证明门槛基准。但它们有用,因为它们说明,当一家公司拥有具名客户证据时,证据会长什么样。Hark 在公开层面还没到那里。[CU009, CU010, CU011, CU012, CU013, CU014]

客户增长 / 采用轨迹表
指标数值日期来源置信度含义缺失分母
公司公开发布宣布 AI 实验室和未来个人智能平台2026-03-24Business Wire 发布 + Observer显示品类意图和公司可见度,不是客户采用没有用户、账户或合同数量
公开产品时间表软件体验 / 模型目标为 2026 年夏季推出;硬件随后2026-03-24 至 2026-05-21Business Wire 发布 + 官方融资帖 + TechCrunch公开叙事仍面向未来,而不是由部署支撑没有发布市场、发布队列或出货目标
具名 Hark 客户null2026-06-11已审阅 Hark 官方界面 + TechCrunch + Forbes已审阅来源中看不到公开具名客户证明没有设计伙伴名单或参考访谈
公开采用指标null2026-06-11已审阅 Hark 官方界面 + SEC 文件 + 融资报道没有发布活跃账户、付费席位、使用量或复购指标没有 beta、waitlist 或 MAU 的分母
团队规模作为代理3 月 >45;到 5 月约 702026-03 至 2026-05Business Wire 发布 + 官方融资帖 + TechCrunch客户证明可见之前,团队扩张已经可见没有商业、支持和研发员工数拆分
牵引力之前的资本$700M 已披露 Series A;SEC Form D 显示总发行金额 $1.0B2026-03 至 2026-05SEC + Business Wire公开客户披露之前,资产负债表实力已经到位没有收入、预订额或付费转化数据

本表记录公开可观察到的客户动能。Null 表示公开记录没有披露该指标,并不代表该指标在内部一定为零。

[CU009, CU010, CU011, CU012, CU013, CU014]
具名客户证明表
客户 / 参考与 Hark 的关系部署 / 用例生产 vs 试点结果 / 证明质量限制
已审阅来源中未发现公开具名 Hark 客户直接 Hark 证据缺口没有公开试点、生产部署或案例研究披露未披露没有公开 ROI、席位数、活跃使用或参考引语这是截至运行日期、仅限已审阅公开记录的缺失陈述
BMW + Figure可比背景,不是 Hark 客户Spartanburg 活跃装配线上的钣金装载;已讨论向 Leipzig 推广生产部署Figure 报告 90,000+ 个零件、1,250+ 小时运行、30,000+ 辆 X3 车辆,以及 >99% 放置目标可作为运营部署证明基准,但不能证明 Hark 需求
GXO + Agility Robotics可比背景,不是 Hark 客户Digit 在真实物流环境中把周转箱从 cobots 运到输送机概念验证后的商业部署Agility 称该部署产生收入,并受多年协议约束仓库人形机器人用例不同于 Hark 个人 AI 产品
Mercedes-Benz + Apptronik可比背景,不是 Hark 客户Apollo 处理套件配送、周转箱配送和相关制造支持任务商业协议 / 试点阶段具名用例和客户引语显示真实买方和具体工作流仍早于 BMW/Figure 的阶段,且不能直接迁移到 Hark
Walmart + Symbotic可比背景,不是 Hark 客户面向加速取货和配送中心的自动化有条件多年部署协议包含开发资金;若满足标准,最多可部署 400 个 APD仓库自动化经济性与消费者 / 智能体 AI 采用差异很大
DHL + Locus Robotics可比背景,不是 Hark 客户AMR 在 40+ 个 DHL 管理站点拣选规模化生产部署Locus 引用 10 亿次拣选、30-180% 生产率提升和 80% 更低培训时间AMR 证明对证明质量有参考价值,但不能说明 Hark 的终端市场

第一行记录 Hark 自身公开证明缺口。其余行是相邻基准,说明自动化领域披露的客户证明长什么样;它们不是 Hark 客户。

[CU012, CU013, CU021, CU022, CU023, CU024]
FU002: 采用 / 部署漏斗

公开证据支撑一条从公司发布走向未来产品发布的路径,但漏斗在具名部署证明之前断开。

这条流程编码的是证据状态,不是客户数量。它有意停在公开证明消失的位置。

[CU009, CU010, CU011, CU012, CU013, CU014]

6.3 留存、耐久性、扩张和集中度:所有耐久客户指标仍为空,因此承销重点转向信任和工作流摩擦

Hark 的公开材料暗示了几条可能的扩张回路,但还没有客户证据验证。隐私政策显示,公司可能靠账号变现、连接器、可交易的工作流,以及能在第三方服务里执行动作的浏览器操作员来扩张。一旦这些界面变成真实客户行为,Hark 可以在单个账号里拿到更广权限、更高频的自主任务、更高价值订阅,并在后续叠加硬件。但这仍是扩张理论,不是已经观察到的 land-and-expand 动作。公司没有披露 NRR、GRR、logo 留存、流失率、续约条款、合同期限、活跃用户队列、CSAT、NPS,也没有公开客户评论语料能明确对应到这个 Hark 实体。因此,耐久性必须保持为空,不能猜。同样需要谨慎看待集中度。公开证据无法说明 Hark 的首批采用者究竟是少数富裕个人、内部测试者、未披露的设计伙伴,还是更广泛的候补名单,因此无法量化头部客户风险。不过,可能的集中度风险在方向上很清楚:公开发布前,采用面大概率很窄;而产品按描述需要个人数据、支付凭证、已连接服务,以及环境式动作执行权限,信任负担异常高。换句话说,Hark 未来的扩张界面可能很有吸引力,但要让客户长期使用,现实障碍是隐私、权限、工作流可靠性,以及日常效用证明。[CU004, CU005, CU006, CU007, CU016, CU017]

留存 / 重复使用 / 满意度表
指标数值 / null分层置信度尽调要求
净留存率(NRR)null所有 Hark 客户按队列,并按纯软件账户与绑定硬件账户提供 NRR
总留存率(GRR)null所有 Hark 客户提供 GRR,并附降级、流失和收缩细节
Logo 留存 / 续约率null具名账户提供首批商业队列的账户级续约历史和合同条款
重复使用 / 活跃队列深度null消费者或准专业用户账户提供 WAU / MAU、会话频率和任务完成重复率
公开满意度信号null可作为参考的客户提供参考访谈、CSAT、NPS、支持指标和评价质量证据
硬件附加 / 复购行为null未来设备买家提供附加率、替换周期假设、退货率和保修索赔

每一行都刻意保留为 null,因为已审阅来源中没有发现 Hark 特定的公开留存或满意度指标。尽调要求列出补齐缺口所需的具体证据。

[CU017, CU040, CU044]
扩张和集中度风险表
扩张驱动因素集中度风险影响尽调路径
更多账户权限和连接器如果用户不愿授予广泛数据和浏览器访问,激活可能停滞削弱日常效用,并拖慢从初始好奇心向外扩张审查授权漏斗、连接器附加率和隐私相关流失
未来 AI 原生硬件附加软件留存被证明前,硬件渠道或制造集中度可能主导经济性抬高毛利和支持风险,同时遮蔽软件持久性请求设备路线图、渠道模型、附加假设和毛利桥
企业或团队推出没有具名参考账户,采购可能卡在安全、隐私和审计审查推迟更高 ACV 动作,也让企业需求难以承销请求试点名单、安全包、管理控制和实施参考
小规模初始发布队列早期收入可能集中在极少数 beta 用户、合作伙伴或设计客户造成需求信号波动,并削弱参考质量请求大客户集中度、队列组合,以及按发布月份拆分的付费 vs 免费使用
投资者和生态光环战略资本关注可能被误认为有机客户拉力可能夸大真实采用准备度和扩张速度将投资者带来的介绍与可重复管道和有机需求分开

本表聚焦早期客户基础稀疏时,哪些机制会扭曲外界感知的动能,尤其是在公开证明弱于融资可见度时。

[CU016, CU039, CU040, CU041, CU042]
FU003: 客户证明与采购摩擦矩阵

与相邻部署相比,Hark 同时具备最弱的公开证明,以及最高的未解信任和采购摩擦。

单元格是分类标签,概括披露质量和可能的采购摩擦,而不是数值评分。

[CU017, CU024, CU033, CU034, CU039, CU041]

6.4 客户尽调通读:Hark 当前要回答的是需求是否存在,不是客户已经跑通

本章结论因此很窄,但重要。Hark 最终可能卖进一个很大的市场,Brett Adcock 的生态也给公司带来关注度、人才入口和相邻运营语境。但本章里,这些都不能替代客户证据。截至本次报告日期,公开记录支持买方假设、工作流假设,以及来自相邻公司的若干证明基准;但不足以支撑关于 Hark 客户规模、生产部署、推荐质量、留存或续约耐久性的主张。正确的尽调动作不是否认 Hark 可能创造需求,而是要求能把推测转成承销依据的材料:具名试点或生产账号、队列使用数据、定价和支付结构、硬件附着假设、续约数据、头部账号集中度,以及愿意讨论日常价值和隐私顾虑的客户推荐。在证据出现前,客户章节应被视为一张证据缺口地图,并附带谨慎的客群假设。Hark 最强的公开客户信号,是产品围绕持续的账号级使用来设计;最弱的公开客户信号,是还没有任何公开客户具名为产品背书。[CU012, CU013, CU017, CU037, CU038, CU039]

客户证明基准表
证明维度Hark 公开状态相邻基准投资含义
具名客户公开未披露BMW、GXO、Mercedes-Benz、Walmart、DHL 是相邻部署中点名的交易对手方Hark 低于客户章节通常要求的举证门槛
运营指标公开未披露Figure 披露零部件、工时、车辆和周期时间目标;Locus 披露拣选量和生产率Hark 还不能公开声称有结果支撑的采用
部署成熟度公开材料里,软件和硬件仍偏前瞻Agility、Figure 和 Locus 描述的是现场部署,而不只是计划发布时间Hark 仍是部署前承销问题
留存可见度未披露 NRR、GRR、流失率或续约条款已规模化的物流同业会发布重复使用或生产率连续性的叙事Hark 的耐久性必须继续记为 null
采购清晰度未公开合同结构、设计合作伙伴或安全参考客户Symbotic/Walmart 和 Apptronik/Mercedes 发布了具体商业结构或用例细节公开证据还不足以支撑企业扩张承销

这些基准只用于校准可信客户证据通常包含什么,不意味着 Hark 与这些公司存在任何客户关系。

[CU012, CU027, CU028, CU029, CU037, CU038]

6.5 佐证

Chapter 07

07风险

7.1 创始人集中、产品证明缺口与预期风险

目前的 Hark 案例仍更像创始人加投资逻辑,而不是产品加证明。官方材料展示了一项异常激进的计划:垂直整合模型、软件和定制硬件;数千块 Nvidia B200 GPU;2026 年夏季发布软件,随后推出硬件。公司还以 $6 billion 投后估值融资超过 $700 million。但这笔资本之下,公开运营记录很薄。已审阅材料仍未披露具名客户、定价、收入或留存证据,TechCrunch 也明确指出 Hark 透露的信息很少。因此,Brett Adcock 成为核心承销对象。公开来源还显示,Adcock 仍与 Figure 深度绑定,并维持其他活跃项目;在客户证明取代创始人品牌、成为估值主要支撑之前,任何分心、信誉受损或时间线失误,都可能传导到 Hark。[CR001, CR003, CR005, CR007, CR008, CR009]

人员 / 执行风险登记表
角色 / 职能依赖或缺口可能性严重性缓释措施尽调路径
创始人 / CEO(Brett Adcock)在仍公开关联 Figure 和其他事业的同时,还是产品、资本和可信度的中心节点Adcock 有很强的创始人履历,并投入了大量个人资本审查时间分配、董事会监督、授权结构和继任计划
硬件和设计领导层必须在激进时间线下,把高端人才转化为可制造、可支持的消费产品中高Hark 已招募 Apple、Tesla、Meta 等公司的知名硬件老兵要求提供组织架构图、项目里程碑,以及 EVT/DVT/PVT 纪律证据
安全 / 隐私 / 法务领导层规模化前,必须把生物识别、隐私、智能体权限和产品责任控制落到运营中已有政策文字,发布尚未发生询问具名负责人、审查节奏、外部律师支持和上线前审批流程
商业化团队必须足够快地证明付费意愿和激活,才能支撑 $6B 起点中高资本让 Hark 能在产品就绪后积极招聘要求提供 GTM 招聘计划、beta 漏斗、定价测试和上线后 KPI 阈值

人员风险不只在于 Hark 能否招到人才,更在于公司是否有足够深的高级运营能力,把创始人愿景转成安全、可靠、商业上读得懂的发布。

[CR005, CR011, CR012, CR013, CR014, CR048]
FR001: 风险热力图

Hark 最高的剩余风险落在客户证据出现之前:创始人集中度、产品证明缺口和硬件执行彼此叠加。

[CR007, CR008, CR011, CR012, CR016, CR023]

7.2 硬件、制造与供应链执行风险

Hark 不只是交付软件。公司自己的材料称,产品依赖紧密耦合的模型、设备和始终可用的个人界面。这会成倍放大执行负担。公司招募了有经验的硬件运营人才,也签下了大量算力,但除 Nvidia 外,尚未公开披露制造伙伴、认证路径、可靠性数据或组件采购策略。更广泛的市场证据给出的是警示。Figure 自己的 BMW 披露在有意义运行时长之后仍突出一个头部硬件故障点,BMW 也把德国部署描述为试点,而不是成熟的规模化项目。MIT Technology Review、Berkeley、Bain、McKinsey 和 TechCrunch 的独立研究都指向同一个信息:人形机器人和物理 AI 系统在灵巧性、电池、正常运行时间、安全、监管和供应商准备度上,仍有硬瓶颈。因此,Hark 正试图切入曲线最难的部分,而不是容易的部分。[CR004, CR006, CR013, CR014, CR016, CR017]

运营 / 质量 / 安全风险登记表
失效模式可能性严重性缓释成熟度剩余敞口未解决缺口
产品上线前延期,或发布后客户效用偏弱低中;Hark 有资本和可见路线图,但还没有公开客户证据首次发布若延期或表现平庸,会同时打击采用、招聘和估值未公开定价、客户、beta 转化或参与度证据
设备可靠性和可制造性低于预期低中;Hark 已聘用经验丰富的硬件人才,但没有公开 DVT 或认证证据实体设备 bug、散热问题、退货或支持负担,可能很快压垮一家年轻公司未披露制造合作伙伴、可靠性看板或认证时间线
持久记忆和智能体权限引发信任或隐私失效中高低中;已有政策文字,但没有公开审计证据一次涉及旁观者、敏感上下文或未授权行动的高关注失效,就会不成比例地损害信任未公开红队结果、事件指标或安全权限架构
算力和组件瓶颈拖慢模型进展或设备节奏中;Hark 披露了 Nvidia 集群协议如果芯片、组件或集成时间线滑坡,上线日期和产品质量可能同时恶化未公开组件短缺或认证延误的应急方案
炒作跑在产品市场现实前面,重演近期 AI 设备失败中高低;品牌和融资预期领先于市场证据Humane 式的叙事与用户价值错配,可能迅速触发降价、退货和战略重置还没有公开证据显示付费意愿、留存或习惯性使用

本表拆出第一代设备和服务周围的执行与信任风险。只要 Hark 有公开野心和人才信号、却缺少公开运营证据,缓释成熟度就保持保守。

[CR003, CR006, CR008, CR013, CR014, CR016]
FR002: 风险传导图

Hark 的主要风险通过发布时间、信任和融资传导,而不是来自单一孤立失效模式。

[CR003, CR008, CR011, CR021, CR023, CR029]
FR003: 依赖图

设备市场证明出现很久以前,Hark 路线图就已经依赖一整套外部系统和交易对手。

[CR006, CR012, CR018, CR021, CR029, CR037]

7.3 隐私、生物识别与监管风险

第一台消费设备上市之前,Hark 的产品愿景已经扩大了监管暴露面。公司想要一个系统,能在第三方应用和用户工作流中听、说、看、记忆,并主动行动。隐私政策确认,Hark 会收集输入和输出、设备和位置数据、第三方应用内容、沙盒文件、shell 命令、生成的代码和日志;政策还称,一些图像、音频和 avatar 功能可能产生在欧盟和美国州法下可被视为生物识别的数据。这意味着法律暴露不是抽象风险。California 新隐私和 ADMT 规则已经进入倒计时,官方网络安全指引也越来越期待智能体系统按 secure-by-design 部署。由于公开证据缺少 Hark 红队测试、事件处理和独立审计,缓释叙事仍主要停留在主张,而不是证明。[CR002, CR021, CR022, CR023, CR024, CR025]

监管 / 法律风险登记表
规则 / 案件 / 义务司法辖区当前状态可能性严重性缓释措施剩余敞口尽调路径
生物识别、图像、语音和持久记忆隐私义务美国各州;关联欧盟的用例Hark 政策已纳入生物识别处理和同意要求中高政策文字承认生物识别同意、留存和销毁义务公开控制仍停留在描述层面,没有外部审计或生产证据说明这些义务如何在已出货产品中落地要求提供生物识别数据流图、同意 UX、留存时间表、删除流程,以及逐司法辖区上线备忘录
加州隐私、风险评估和 ADMT 规则加州CPPA 规则 2026-01-01 生效;风险评估要求 2026 年开始,ADMT 义务 2027 年开始Hark 已维护一份详细隐私政策,能在大规模上线前按合规要求设计任何会提出实质建议的主动助手,如果通知、风险评估或退出机制薄弱,都可能受到审查获取法律顾问备忘录,将当前产品计划映射到 CPPA 风险评估和 ADMT 要求
智能体 AI 安全部署预期美国联邦 / 企业安全CISA 及相关指南强调谨慎采用、安全部署,以及面向智能体系统的安全内建设计控制如果 Hark 早期优先投入安全工程,仍能在规模化前搭建上线流程个人助手若能执行任务、访问第三方应用并存储持久上下文,一旦权限或边界失效,信任敏感度会很高审查威胁模型、权限模型、红队结果、人工接管设计和供应商管理控制
先进芯片出口管制和尽调规则美国出口管制体系BIS 在 2026 年再次更新先进计算指导,并把 IC 设计商时间线延长到年底中高Hark 已获得算力访问,可以使用成熟供应商芯片政策变化很快,可能拖慢模型训练或未来设备路线图的采购、合规和全球供应选择要求提供芯片采购计划、出口管制合规负责人,以及受限供应下的应急方案
常开 AI 设备的产品责任和青少年安全审查美国消费品和产品责任环境公开层面还没有上线诉讼,但 Hark 瞄准主动设备,同时声明服务不面向 18 岁以下儿童上线前阶段仍有空间收紧披露、年龄门槛和安全边界常开设备会感知环境并代表用户行动,一旦出现误用、旁观者被采集或不安全自动化,责任风险可能被放大消费者发布前,审查年龄门槛、测试协议、事件响应、保险覆盖和产品责任假设

各行按 2026-06-11 可获得的公开证据集所显示的剩余严重性排序。这是一份聚焦最高优先级问题的部分登记表,而不是覆盖所有上线市场的法律备忘录。

[CR021, CR022, CR023, CR024, CR025, CR026]

7.4 市场怀疑、伙伴暴露与资本密集风险

即使 Hark 按时发布,也仍要跨过一个越来越怀疑人形机器人时间线和 AI 设备炒作的市场。最强的公开可比对象并不令人安心。Berkeley、MIT Technology Review、Bain、McKinsey 和 TechCrunch 的专家评论认为,物理 AI 商业化大概率比当前估值所暗示的更慢、更垂直,也更昂贵。Humane AI Pin 提供了消费硬件的警示故事:一个融资充足的 AI 设备叙事,最终变成硬件停产、退货超过销售、电池警告,以及规模小得多的战略资产出售。Hark 还背着伙伴和资本依赖。它需要算力、第三方 AI 基础设施、未来投资人,以及最终还未公开具名的制造交易对手。SEC 材料给出第二个提醒:这是资本密集型建设。Hark Labs 在 2026 年 3 月提交了 $1 billion 豁免发行文件,说明即便在头条 Series A 之后,融资仍然可能非常重要。[CR015, CR018, CR019, CR020, CR029, CR030]

合作伙伴 / 依赖风险登记表
依赖项交易对手方角色集中度失效情景严重性缓释措施剩余敞口
模型训练算力NVIDIA 和先进 GPU 生态训练和服务多模态模型栈算力供应或政策摩擦拖慢模型进展和设备上线Hark 已签下 B200 产能,并获得芯片生态参与方的投资者支持算力集中和先进芯片政策仍带来进度与成本敏感性
设备制造和组件组装未披露的合同制造商和供应商制造、认证并规模化 AI 原生硬件隐性制造瓶颈推迟上线,或造成可靠性与支持失效有经验的硬件招聘可能缩短学习曲线关键交易对手和就绪证据仍未披露,剩余风险依然高
第三方模型和应用生态Third-Party AI Providers 和集成的第三方应用输出生成、工作流触达和用户上下文中高供应商失效、政策变化或边界控制薄弱,会降低产品质量或提高隐私敞口理论上,Hark 可以随着时间推移把更多栈收回自有体系,降低依赖当前隐私政策已经承认第三方 AI 和应用依赖,这会增加控制复杂度
创始人相邻的硬件叙事Figure / BMW 生态物理 AI 执行力的跨公司可信度信号Hark 拿出自身证据前,Figure 部署或商业化若再起争议,会削弱投资者对 Hark 的信心中高两家公司在形式上彼此独立,Hark 可以独立成功公开市场仍可能把 Adcock 的多项事业当成一组相互关联的声誉篮子
未来定价资本新外部投资者和老股交易上线后验证或重置 Hark 的估值预期下一轮融资或清算事件的定价明显低于当前叙事大额初始资本提供了一定现金跑道如果没有客户证据,未来投资者可能用上线证据大幅重估公司

Hark 目前还不像依赖单一客户。其最高合作伙伴敞口在上游:算力、制造、第三方模型基础设施,以及塑造投资者信心的创始人关联叙事生态。

[CR006, CR011, CR012, CR015, CR018, CR019]

7.5 缓释因素、监测指标与终止标准

Hark 并非没有真实缓释因素。公司有资本、有可信的招聘品牌、有清晰可见的技术雄心,也正式承认技术栈必须一起构建,而不是发布后临时拼接。这些都重要。但公开证据留下的未解问题太多,缓释因素还撑不起完整投资逻辑。实际操作中,投资人应把 Hark 当作带闸门的承销项目,而不是纯愿景押注。只有管理层能证明发布准时、隐私和安全控制可信、客户转化可见、制造路径现实,决策才应改善。在此之前,正确姿态是监测一小组可衡量触发器:发布滑坡、隐私或生物识别麻烦、持续硬件或供应延误、Figure 创始人争议外溢,以及任何把估值叙事向下重置的融资或市场信号。如果多个触发器同时出现,投资逻辑就不是延迟,而是破裂。[CR003, CR007, CR014, CR027, CR028, CR035]

缓释与否决标准表
风险可监控触发因素阈值 / 事件行动含义
产品证据缺口上线延期、早期访问转化偏弱,或管理层仍无法披露有意义的用户激活承诺的 2026 年软件窗口后出现任何实质延误,或上线后没有可信早期采用信号将 $6B 标记视为创始人溢价,不再继续承销,直到使用证据出现
隐私 / 生物识别 / 智能体控制失效监管问询、安全事件、旁观者采集争议,或泄露的不安全行为任何已确认的隐私事件,涉及持久记忆、生物识别处理或未授权任务执行将 Hark 重估为监管程度更高的资产;支持更多资本前,要求独立安全审查
硬件和供应链执行认证滑坡、设备可靠性偏低,或可见组件瓶颈硬件发布反复延误、资质里程碑失败,或关键组件短缺未解决下调发布假设,延长烧钱分析,并检验业务是否变得资本效率低下
创始人集中度和声誉外溢新的 Figure 争议、治理压力,或可见的创始人过度分身又一次重大透明度争议,或证据显示 Adcock 精力正在限制 Hark 执行升级董事会和授权尽调;不要假设创始人声誉仍能抵消证据缺口
估值重置 / 资本强度下一轮融资或老股信号明显低于当前预期定价事件或条款清单实质性低于当前估值叙事从新的清算价格重建稀释、招聘和回报假设,而不是沿用标题化的 Series A 故事

这些否决标准刻意设为可观察。Hark 仍可能很快降风险,但前提是上线证据开始取代声誉和资本,成为支撑故事的主要依据。

[CR003, CR007, CR008, CR027, CR028, CR043]
Chapter 08

08估值

8.1 当前价格与公开证明

Hark 的公开价格异常明确,运营证明也异常薄。公司宣布在 2026 年 5 月完成超过 $700 million 的 Series A,投后估值 $6 billion;2026 年 3 月的 Form D 显示,公开宣布前,公司似乎已经启动了一个发行上限 $1.0 billion 的融资流程。这种资本入口很重要。它给 Hark 真实资源去招聘、购买稀缺算力,并比多数发布前 AI 硬件公司更快推出产品。但同一份公开记录仍缺少通常支撑如此高价格的基本承销输入:没有披露收入、定价表、付费客户名单、留存数据或硬件毛利率证据。官方故事仍主要向前看:2026 年夏季晚些时候发布模型,之后推出硬件,以及面向个人智能的垂直整合愿景。 结果是一个市场信号,却没有公开运营账本。投资人可以验证 Hark 融到了这轮钱、谁支持了它、Form D 写了什么,以及公司在同时建设模型和原生硬件。但他们还无法验证用户是否会付费、产品品类是否有粘性,或者硬件路径是否比纯软件界面创造更好经济性。在当前条款下,估值因此是押注能力和雄心,而不是押注已经证明的变现。这个区别正是本章保持价格敏感、而非创始人敏感的核心原因。[CV001, CV002, CV003, CV004, CV005, CV006]

建议摘要表
入场视角建议置信度风险评级估值立场决策含义
当前公开条款(2026 年 5 月)回避昂贵仅凭公开证据,不要按已宣布的 $6B 投后估值投资。
什么会改变判断观察 / 继续研究偏高到合理只有付费采用出现、股权结构条款可接受,或价格实质性下调时,才重新评估。

第一行是当前判断;第二行明确说明,该建议对价格和证据敏感,而不是对公司质量的永久判断。

[CV001, CV037, CV045, CV046, CV047, CV048]
投资逻辑 / 反向逻辑表
论点当前证据什么会改变判断
投资逻辑:Hark 可以靠资金迅速打到品类相关性Series A 融资超过 $700M,超额认购,主要算力和半导体投资者参与。若 2026 年夏季按时发布指标,并证明资本正在转化为产品使用,论点会增强。
投资逻辑:垂直整合可能形成有防御力的界面护城河官方材料持续描述模型、软件、记忆和定制硬件的一体化构建。如果 Hark 证明硬件相较纯软件界面,能实质提升留存、转化或任务完成率,论点会增强。
投资逻辑:品类上行空间真实存在Goldman 对 $38B 人形机器人 TAM 的观点,以及 Figure、1X、Apptronik 的高估值,说明投资者会继续资助赢家。如果 Hark 证明自己属于赢家队列,而不只是炒作队列,论点会增强。
反向逻辑:公司在变现公开前就已被定价未发现公开收入、定价、客户 logo、毛利率或付费 beta 披露。如果 Hark 披露付费使用、留存和客户案例,论点会削弱。
反向逻辑:证据更充分的同业更便宜,或只是略便宜Agility 和 Apptronik 在较低据报估值下已有公开部署证据;Hark 已高于 Apptronik 的据报估值标记。如果 Hark 拿出质量明显优于这些同业的证据,或价格重置,论点会削弱。
反向逻辑:整个品类可能仍被高估Gartner、IEEE、Brooks、McKinsey 和 Humane 的结果都指向同一点:硬件野心可能跑在需求和 ROI 前面。如果 Hark 交付到一个付费意愿、正常运行时间和留存都清晰可见的用例中,论点会削弱。

各行混合投资逻辑和反向逻辑,使决策保持证据敏感。多行取决于尚未公开发生的事件。

[CV001, CV006, CV007, CV024, CV025, CV026]
FV001: 推荐逻辑

规模野心、缺失验证、品类怀疑和当前价格如何共同导向最终建议。

该图为定性分析,但每个节点都有明确主张支撑:融资事实、战略叙事、缺失验证、可比公司区间和反向研究。

[CV001, CV006, CV007, CV032, CV033, CV047]

8.2 可比融资显示,Hark 的定价领先于证据更充分的同行

检验 Hark $6 billion 价格最干净的方法,是对照近期机器人融资和交易,且这些案例的公开记录不止一条融资标题。Figure 给出两个锚点:2024 年 2 月估值 $2.6 billion,Microsoft、Nvidia、OpenAI 和 Bezos 参与;随后在 2025 年 9 月估值 $39 billion,当时它已在披露的软件和生产证明上走得更远。Figure 后来的 BMW 部署展示了可衡量的产线活跃运行时长、搬运零件数量和触达车辆数量。Apptronik 是 2026 年高可信人形机器人初创公司的最近下界:在 TechCrunch 报道大约 $5.3 billion 估值前,公司已有 Mercedes-Benz、GXO、DeepMind 邻近性,以及近 $1 billion 融资。Agility 在 2026 年公开引用中更便宜,约 $2.1 billion 至 $2.15 billion,但它已经有多年期 GXO 协议和更清晰的物流工作流。 因此,Hark 落在一个尴尬的相对位置。尽管披露的商业证明少于任何一家,它仍比 Apptronik 更贵,也远比 Agility 和 Boston Dynamics 已披露的并购基准更贵。唯一明显高于 Hark 的可比对象,是 Figure 2025 年末的领导者估值和 1X 据报道 $10 billion 的融资目标,但两者都不能有力防守今天的价格。Figure 的更高估值背后有多得多的证据;1X 的更高估值只是据报道的目标,不是已关闭融资。这让 Hark 看起来更像一个以溢价购买未来品类领导权的期权,而不是一个已经被公开运营证据验证的价格。[CV011, CV012, CV013, CV014, CV015, CV016]

可比估值表
可比公司指标倍数 / 估值 / 状态相关性局限
HarkSeries A,2026 年 5 月$6.0B 投后估值,融资 >$700M当前承销参考点无公开收入、客户或定价证据
Figure AISeries B,2024 年 2 月$2.6B 投后估值,融资 $675M显示投资者在后续规模化前,为领先人形机器人标的支付了什么价格相对 2026 年已过时,且仍未盈利
Figure AISeries C,2025 年 9 月$39B 投后估值,承诺资金 >$1B品类的高端领导者基准公开产品和部署证据远强于 Hark
ApptronikSeries A 延长轮,2026 年 2 月投后约 $5.3B;Series A 总额 >$935MFigure 之外可信度最高的未上市类人机器人可比公司估值由 TechCrunch 报道,Apptronik 未正式披露
Agility RoboticsSeries C / 2026 年公开参考2025 年约 $400M 融资后约 $2.1B-$2.15B商业落地标杆,工作流证据更清楚公开估值来自二手参考,不是公司最新新闻稿
1XSeries B,2024 年 1 月已完成 $100M 融资;估值未公开披露可作为 Hark 界面愿景下的消费 / 家庭类人机器人参照没有清晰的官方轮次估值
1X据报道的 2025 年目标寻求最多 $1B 融资,目标估值 $10B+体现公众想象力如何给消费级类人机器人上行空间定价仅为目标;不是已完成融资
Boston DynamicsHyundai 收购,2021 年 6 月$1.1B 交易价值本组唯一完全披露的机器人并购估值交易较早,产品和买方语境不同
Humane资产出售,2025 年 2 月融资 >$230M 后以 $116M 出售资产AI 硬件下行情景的反向标杆不是机器人公司,结局反映的是消费硬件失败

Hark 缺少收入数据,无法干净套用公开市场倍数,因此表中混合了私人融资轮、已披露并购交易和一个 AI 硬件反向案例。只有存在公开来源时才引用报道估值;其中有些来自公司披露,有些来自出处清楚的媒体报道。

[CV001, CV011, CV012, CV016, CV019, CV021]
FV002: 估值敏感性

Hark 在不同里程碑验证状态下的示意公允价值中点,不只看公布价格。

数值是分析师估计的公允价值中点,单位为百万美元。它们是基于里程碑的敏感性标记,不是预测,也说明公布的 $6B 价格已假设相当大的执行成功。

[CV024, CV031, CV038, CV039, CV041, CV049]

8.3 概率加权结果仍低于要价

Hark 的上行情景并非凭空想象。Goldman 的品类 TAM 工作说明,为什么投资人愿意为人形机器人或具身 AI 的真正赢家支付很高价格;Hark 自身融资也证明,资本市场愿意资助雄心勃勃的界面加硬件押注。如果 Hark 2026 年夏季模型推出能转化成真实付费使用,如果公司能证明硬件显著改善留存或效用,并且如果它能披露早期经常性经济性,高于今天价格的估值是可能的。问题在于,公开证据尚不足以让外部投资人给这种上行情景赋予高概率。Gartner、IEEE Spectrum、IEEE RAS、Rodney Brooks 和 McKinsey 都指向同一个方向:人形机器人和具身自动化市场仍难以规模化、容易被过度炒作,也容易被已经交付仓储 ROI 的专用替代方案冲击。 因此,情景表保持保守。基准情景假设 Hark 拿到的估值更接近后期证明型同行,但低于当前标记;悲观情景假设发布失误或变现疲弱,让公司暴露在类似其他 AI 硬件失望案例的重估中;乐观情景假设 Hark 成为少数平台赢家之一,把产品雄心转成持久付费意愿。对这些路径加权后,中点仍低于当前 $6 billion 价格。同样重要的是,从当前进入价看,回报曲线并不吸引:即便结果非常好,也只产生有限的总上行;而下行仍严重,因为已融资资本、可能的优先权和未来稀释都排在普通股上行之前。[CV024, CV025, CV026, CV027, CV028, CV029]

乐观 / 基准 / 悲观情景表
情景假设估值 / 回报逻辑关键风险概率信号
乐观2026 年夏季模型发布转化为付费使用;硬件顺利上线;已披露经济性支撑平台叙事;后续投资者把 Hark 重新定价为消费 / 具身 AI 领导者。$8B-$12B 结果;按今天 $6B 入场计算、后续稀释前,毛 MOIC 约 1.3x-2.0x;所需证据要比 Agility 更接近高端 1X / Figure 叙事。消费硬件采用低于预期;留存偏弱;未来仍需要融资。25%
基准Hark 发布产品,但公开证据仍好坏参半;软件价值存在,硬件仍昂贵,市场给 Hark 的估值更接近证据更充分的私营机器人同业。$2.5B-$4.0B 结果;按当前入场计算,毛 MOIC 约 0.4x-0.7x;更接近 Agility/Apptronik 式基准,而不是突破型平台定价。上线延期,定价仍不透明,市场折价处理缺失的变现证据。45%
悲观上线延期或转化不佳;硬件经济性仍不清楚;公司在证明付费意愿前还需要更多资本;品类倍数压缩。$0.75B-$1.5B 结果;毛 MOIC 约 0.1x-0.25x;下行更像估值重置或战略出售,而不是高溢价增长轮。下轮融资保护条款、需求偏弱和替代自动化方案限制恢复价值。30%

情景价值是分析师估计,依据是当前私营机器人可比公司、公开品类怀疑,以及 Hark 已知价格。概率加权中点约 $4.3B,低于当前 $6B 要价。

[CV019, CV024, CV025, CV031, CV038, CV039]
投资逻辑失效与否决触发器表
触发器阈值对投资逻辑的传导行动含义
上线节奏滑坡到 2026 年 Q4,承诺的 2026 年夏季模型仍没有实质性公开发布收入证据出现前支付溢价估值,核心依赖执行力;延迟会削弱这一前提坚决不买;公允价值向悲观区间靠拢
付费采用缺失上线后两个季度内,仍未披露付费用户、客户标识或变现证据当前价格会从产品市场匹配押注变成纯叙事押注维持回避;要求价格重置或数据室证据
硬件护城河缺席原生硬件看起来可有可无,或相较纯软件入口没有差异相比纯模型或应用竞争者,溢价倍数的主要理由被拿掉估值下调至纯软件或失败设备先例附近
品类估值回调类人机器人 / 具身 AI 同行在下一轮融资周期明显重估下行支撑 Hark 当前估值标记的可比倍数会收缩收紧下行假设,要求更深折价
隐藏优先权堆叠Series A 条款显示激进优先权、棘轮条款或大规模老股出售即便企业价值增长,新投资人与普通股股东的上行空间也会被压低除非价格补偿结构性下行,否则回避

每个触发器都能靠公开公告或尽调室监测。资本结构触发器目前无法靠公开证据解决,因此也列入证据缺口。

[CV003, CV007, CV025, CV031, CV039, CV040]
FV003: 估值 / 回报区间

情景估值区间与当前 $6B 进入价格的对比。

所有数值单位均为百万美元,概率加权中点仍低于当前要价。当前要价用相同的低值和高值显示为固定点。

[CV001, CV041, CV042]
FV004: 投资 KPI

面向投委会的核心指标与参考点,用于当前 Hark 决策。

该图混合使用 Hark、同业和下行参考数字。除隐含持股比例直接以百分比表示外,其余单位均为百万美元。

[CV001, CV004, CV010, CV019, CV016, CV031]

8.4 建议、纪律与尽调要求

基于公开证据,正确的投资结论是在当前价格回避。这不是说 Hark 质量低,而是说当前价格已经假设了某种产品市场证明,而公开记录尚未展示这种证明。公司有强投资人、创始人可信度、重算力建设计划,以及如果个人 AI 硬件突破就可能变得重要的叙事。但价格很重要。以 $6 billion 投后估值计算,Hark 已经高于 Apptronik 据报道的 2026 年估值,也远高于 Agility 的 2026 年公开区间,同时在客户、定价和商业化上提供的公开证据更少。 未来估值可能变得可支撑,但只能靠证据或价格变动。最明显的上调因素包括:2026 年夏季推出后披露的付费采用;证明原生硬件实质改善使用或留存;具名商业设计伙伴;透明的股权结构表,显示优先股堆叠没有吃掉太多下行空间。缺少这些数据时,进入纪律应当严格:不要把宣布的估值承销为公允价值,也不要把品类热情混同于 Hark 理应高于证据更充分同行交易的证明。正确的近期姿态是继续研究,但在证明改善或价格重置之前,只能站在场边。[CV037, CV041, CV042, CV044, CV045, CV046]

最终尽调问题表
主题缺失证据重要性负责人或尽调路径
股权结构表与优先权Series A 条款清单、清算优先权、反稀释和期权池处理判定普通股与新投资人的下行风险是否远差于标题估值暗示的水平向公司法律顾问和领投方索取最终交割文件
付费采用2026 年夏季上线后的用户数、付费转化、流失、留存和队列行为区分产品好奇心与可变现的产品市场匹配董事会材料、增长看板和支付处理商摘录
客户证据具名企业或渠道伙伴、设计伙伴和硬件试点经济性验证 Hark 的界面或设备是否解决真实购买问题客户访谈和已签商业协议
单位经济性硬件 BOM、保修准备金、支持负担和软件毛利率桥看原生硬件是在创造价值,还是只是在吃掉资本CFO / 运营复核加供应商报价
独立估值标记支撑 $6B 标记的老股交易、投资人信或第三方估值备忘录检验公布价格是被广泛接受,还是只是融资标题领投方附函复核和基金 LP 材料
竞争差异化Hark 在留存或付费意愿上能跑赢纯软件助手以及证据更充分的机器人同行的证据没有这类证据,相比 Agility / Apptronik 的溢价倍数很难自洽产品演示、基准测试套件和比较用户研究

从公开市场叙事走向可承销的估值意见,至少需要这些材料。若干尽调问题直接对应两个尚未解决的研究问题。

[CV007, CV037, CV043, CV044, CV046, CV049]

8.5 佐证

免责声明

本尽调报告由 AI 研究代理基于截至 2026-06-11 的公开来源生成。报告不构成投资建议。Hark 是一家私营公司,客户、收入、利润率、现金跑道和融资条款等关键投资判断输入仍未在公开记录中披露。

证据索引

结论
编号陈述可信度来源
CO001 Hark’s homepage says the company is building the most advanced personal intelligence in the world. SO001
CO002 Official Hark materials describe the product as a universal interface between humans and machines built from the company’s own AI models and bespoke hardware. SO001, SO002, SO003, SO004
CO003 Hark says its systems combine speech, text, vision, contextual awareness, and persistent memory. SO001, SO002
CO004 Hark says it is designing AI-native hardware devices and agentic computers alongside its software stack. SO001, SO002, SO003, SO004
CO005 Hark’s public thesis is vertically integrated across foundation models, software systems, hardware, and interfaces rather than a single software layer. SO002, SO004, SO017
CO006 Hark says its first AI models and software experiences will be available in summer 2026. SO003, SO004, SO006, SO009
CO007 Hark says AI-native hardware devices will follow after the software launch. SO003, SO004, SO006
CO008 Brett Adcock is publicly identified as Hark’s founder and CEO. SO002, SO004, SO006, SO013
CO009 Public founder biographies say Adcock founded Figure AI and co-founded Archer Aviation before starting Hark. SO012, SO013, SO014, SO022
CO010 Observer and TechCrunch report that Hark started in late 2025 with about $100 million of Adcock’s own capital. SO006, SO007, SO010
CO011 The March 2026 launch release said Hark had more than 45 researchers, engineers, and designers. SO002, SO010
CO012 The May 2026 funding post said Hark’s team had grown to around 70 people. SO003, SO006, SO011
CO013 Hark’s design effort is led by Abidur Chowdhury, a former Apple designer associated with iPhone and Mac products. SO002, SO006, SO007, SO010
CO014 The reviewed public record does not identify a disclosed Hark board or independent directors as of the run date. SO001, SO002, SO003
CO015 The reviewed official materials name Adcock and Chowdhury but do not disclose a broader Hark executive roster. SO001, SO002, SO003
CO016 On May 21, 2026, Hark announced an oversubscribed Series A of over $700 million at a $6 billion post-money valuation. SO003, SO004, SO005, SO006, SO011
CO017 Parkway Venture Capital led Hark’s Series A round. SO003, SO004, SO005, SO006
CO018 The named Series A participants included NVIDIA, AMD Ventures, ARK Invest, Brookfield, Greycroft, Intel Capital, Prime Movers Lab, Qualcomm Ventures, Salesforce Ventures, Tamarack Global, and Align Ventures. SO003, SO004, SO005, SO011, SO021
CO019 Business Wire said Qatalyst Partners provided strategic and financial advice to Hark on the Series A transaction. SO004, SO011
CO020 The minimum disclosed financing committed to Hark is more than $800 million when reported $100 million self-funding is added to the $700 million-plus Series A. SO003, SO006, SO007
CO021 No reviewed public source disclosed investor ownership percentages, secondaries, debt facilities, or cap-table rights for Hark’s financing. SO003, SO004, SO006
CO022 Public releases and coverage place Hark in San Jose, California. SO002, SO004, SO011, SO021
CO023 Hark disclosed NVIDIA B200-based compute infrastructure and a new data center for training its next generation of multimodal models. SO002, SO003, SO006
CO024 Hark says its personal AI platform will work across products and services users already use, acting as a proactive assistant for their digital world. SO003, SO006
CO025 Hark has not publicly named customers, pilots, or beta partners in the sources reviewed for this chapter. SO003, SO004, SO006, SO007
CO026 Hark has not publicly disclosed revenue, ARR, pricing, or customer count as of the run date. SO003, SO004, SO006, SO020
CO027 Figure describes its current product as a general-purpose humanoid robot for everyday home help. SO015, SO025
CO028 TechCrunch reported that Figure chose in-house AI over an OpenAI partnership because Adcock argued embodied AI needs vertical integration of hardware and software. SO017
CO029 TechCrunch reported in June 2025 that Figure faced skepticism around its BMW relationship and had not done a live public demo while Adcock threatened legal action against a publication. SO018
CO030 Forbes wrote in 2024 that Figure had raised major capital but still faced substantial technical and commercialization challenges. SO019
CO031 Forbes argued Hark’s $6 billion valuation is better understood as a moat of capital, compute, talent, and founder track record than as proof of commercial traction. SO020
CO032 Robotics & Automation News described humanoid robotics in early 2026 as fast-moving but still defined by uncertain real-world potential and commercialization limits. SO023
CO033 Because Adcock remains publicly linked to both Hark and Figure, Hark carries material key-person and spillover-reputation risk. SO006, SO007, SO013, SO018
CO034 Hark was founded in 2025 and emerged publicly in March 2026. SO002, SO006, SO013
CO035 The March 2026 launch announcement simultaneously introduced Hark’s personal-AI thesis, Chowdhury’s design role, and NVIDIA-backed compute scaling. SO002, SO007
CO036 By May 2026, Hark had turned from a stealth self-funded lab into one of the largest AI hardware Series A financings in the market. SO003, SO004, SO006, SO021
CO037 Hark’s first public proof window is the promised summer-2026 software release followed by hardware thereafter. SO003, SO004, SO006, SO009
CO038 The Series A syndicate clusters chip suppliers, enterprise software investors, and deep-tech funds around Hark’s vertically integrated AI-device thesis. SO003, SO004, SO005, SO008
CO039 Hark’s public materials repeatedly frame the company as a replacement for today’s chat-interface paradigm and legacy consumer devices. SO001, SO002, SO007
CO040 As of the run date, Hark should be treated as a highly capitalized, pre-product Series A company whose valuation rests more on founder pedigree and resource access than on disclosed commercial traction. SO003, SO004, SO006, SO020
CM001 Hark says it is building the most advanced personal intelligence in the world. SM001
CM002 Hark says its product combines bespoke hardware devices and agentic computers with speech, text, vision, and highly personalized memory. SM001, SM002
CM003 Hark says its hardware and AI are designed to serve as a universal interface between humans and machines. SM002
CM004 Hark says it plans to roll out software experiences and AI models before introducing AI-native hardware devices. SM002
CM005 Public Hark materials support classifying the company as an adjacent embodied-AI and personal-robotics interface play rather than a disclosed robot-OEM vendor today. SM001, SM002, SM003
CM006 For this chapter, the core market boundary includes humanoid robots, service robots that work in human-designed spaces, and the software, integration, and control layers required to deploy them. SM003, SM004, SM005
CM007 Fixed industrial manipulators, AMRs, and pure software automation should be treated as important substitutes rather than the target market itself. SM004, SM005, SM008, SM021
CM008 Goldman Sachs says the total addressable market for humanoid robots could reach $38 billion by 2035, more than six times a prior $6 billion projection. SM003
CM009 Goldman Sachs increased its 2035 humanoid shipment estimate to 1.4 million units. SM003
CM010 Goldman Sachs says its base case is for more than 250,000 humanoid robot shipments in 2030 and that almost all of those shipments would be industrial. SM003
CM011 Goldman Sachs says the manufacturing cost range for humanoid robots fell from about $50,000-$250,000 to about $30,000-$150,000 in the latest update. SM003
CM012 Goldman Sachs attributes the faster commercialization path to cheaper components, more supply-chain options, and improved designs and manufacturing techniques. SM003
CM013 Goldman Sachs says faster cost declines could pull factory applications forward by about one year and consumer applications forward by two to four years versus prior expectations. SM003
CM014 Goldman Sachs says humanoids are especially appealing for dangerous, dirty, and dull tasks and for sectors that do not have enough workers. SM003
CM015 Goldman Sachs says consumer robot sales could exceed one million units annually in just over a decade in its base case. SM003
CM016 Gartner says that through 2028 fewer than 100 companies will move humanoid proofs of concept beyond experimentation and fewer than 20 companies will go live in production for supply-chain and manufacturing use cases. SM004
CM017 Gartner says most humanoid production deployments through 2028 will remain limited to tightly controlled environments rather than dynamic and high-throughput supply-chain operations. SM004
CM018 Gartner says polyfunctional robots with wheels and telescopic arms can deliver higher uptime and lower energy use than humanoids in many supply-chain workflows. SM004
CM019 Gartner identifies technological limitations, integration complexity, high costs, and energy constraints as major barriers to humanoid adoption in supply chain and manufacturing. SM004
CM020 McKinsey says companies fund warehouse automation to improve resilience, speed, reliability, flexibility, safety, space utilization, and throughput while addressing labor challenges. SM008, SM014
CM021 McKinsey says many warehouse automation programs fail because leadership lacks a cohesive vision, misunderstands the technology, or stays internally misaligned on operating principles. SM008
CM022 McKinsey describes a consumer-goods company that spent more than $150 million on an automated warehouse whose advanced automation features were later underused because planning assumptions proved wrong. SM008
CM023 IFR says 54% of annual industrial robot installations worldwide were deployed in China. SM006
CM024 IFR says Western Europe reached 267 robots per 10,000 manufacturing employees in 2024, ahead of North America at 204 and Asia at 131. SM006
CM025 IFR forecasts global industrial robotics installations will surpass 700,000 units in 2028, implying roughly 7% CAGR from 2025 to 2028. SM006
CM026 IEEE says the biggest scaling problem for humanoids is demand because no application has yet been found that requires several thousand robots per facility. SM011, SM012
CM027 IEEE says market requirements for humanoids include battery life, reliability, and safety before meaningful scale is possible. SM011
CM028 IEEE uses Agility Digit as an example of current battery trade-offs: about 90 minutes of runtime and a full recharge in about 9 minutes. SM011
CM029 IEEE says industrial customers often expect roughly 99.99% reliability because even 99% reliability can still create about 5 hours of monthly downtime. SM011, SM012
CM030 IEEE says specific safety standards for dynamically balancing legged robots are still under development. SM011, SM012
CM031 Rodney Brooks argues that turning prototypes into systems deployed at scale is materially harder than proving the underlying idea in demonstrations. SM013
CM032 Figure says its 11-month deployment at BMW Spartanburg loaded more than 90,000 parts, delivered more than 1,250 runtime hours, and contributed to production of more than 30,000 X3 vehicles. SM017
CM033 BMW says its Leipzig physical-AI pilot follows earlier successful humanoid work in Spartanburg and is focused on repetitive, precise, and ergonomically difficult tasks in production. SM018, SM017
CM034 Agility and GXO describe their multiyear Digit agreement as the first formal commercial deployment and first robots-as-a-service deployment of humanoid robots. SM019
CM035 Apptronik says Mercedes-Benz is exploring Apollo for logistics tasks such as delivering assembly kits and totes inside manufacturing facilities. SM020
CM036 Amazon says it has deployed more than one million robots across its operations network since 2012. SM021, SM022
CM037 Amazon says robots across its fulfillment network can reduce handling time by up to 15 seconds per item, highlighting the scale and maturity of non-humanoid warehouse automation. SM022, SM021
CM038 WHO says the need for assistive products is expected to grow to more than 3.5 billion people by 2050. SM023, SM024
CM039 WHO says large gaps in assistive-technology service provision and trained workforce remain and that only one in ten people globally who need assistive products have access to them. SM023, SM025
CM040 WHO says barriers to assistive-technology adoption include high costs, procurement challenges, workforce capacity gaps, fragmented markets, and limited awareness. SM024, SM025
CM041 The Future of Jobs Report 2025 says technological change, demographic shifts, and geoeconomic fragmentation are major forces shaping labor markets through 2030. SM016
CM042 The Future of Jobs Report 2025 synthesizes the views of more than 1,000 global employers representing more than 14 million workers across 55 economies, so it is a broad macro-demand signal rather than a robot-specific market forecast. SM016
CM043 Applying Goldman’s more-than-250,000 industrial shipments base case to its current $30,000-$150,000 per-unit cost range implies a very wide roughly $7.5 billion-$37.5 billion annual hardware revenue envelope before services and software. SM003
CM044 Public evidence does not yet show whether Hark’s first commercial wedge will be a consumer device, an enterprise workflow layer, or an embodied-agent control stack tied to third-party robots.
CP001 Hark publicly describes itself as a universal interface between humans and machines built from multimodal AI and native hardware, but it does not yet disclose a specific robot form factor. SP001, SP002
CP002 Hark plans software and model releases before hardware devices, which indicates a public product posture earlier than the deployment stage already shown by leading humanoid peers. SP002, SP003
CP003 Hark said in 2026 that it raised a $700 million Series A at a $6 billion post-money valuation after Brett Adcock initially funded the company with $100 million of his own capital. SP002, SP003
CP004 TechCrunch reported that Hark had about 70 employees and was buying hardware, product, and AI talent rather than already running public robot deployments. SP003
CP005 Goldman Sachs projected the global humanoid robot market could reach $38 billion by 2035. SP004
CP006 Goldman Sachs also argued that near-term humanoid demand would be predominantly industrial rather than consumer. SP004
CP007 Gartner said in 2026 that fewer than 20 companies are likely to reach production-stage humanoid deployments in manufacturing and supply chain by 2028. SP005
CP008 Gartner argued that polyfunctional robots can outperform humanoids in supply chain use cases on throughput, uptime, energy use, and cost. SP005
CP009 Figure publicly positions Figure 03 and Helix as a tightly integrated humanoid hardware and onboard AI system. SP009, SP010
CP010 Figure said its BMW deployment ran every working day for 11 months, logged more than 1,250 hours of runtime, loaded more than 90,000 parts, and contributed to more than 30,000 X3 vehicles. SP011
CP011 TechCrunch reported that Figure reached a $39 billion valuation in its latest funding round, giving it far more public financial scale than Hark. SP012
CP012 Figure is the closest direct benchmark to Hark because both market integrated AI plus hardware ambition, but Figure already has stronger field and financing proof. SP001, SP010, SP011, SP012
CP013 The retained Electrek guide portrays Tesla Optimus as a strategically important program that is still dealing with redesigns, leadership churn, and unresolved usefulness inside Tesla factories. SP013
CP014 Tesla remains a serious competitive threat because it can pair humanoid ambition with Tesla manufacturing scale and internal factory distribution even before it proves external demand. SP013
CP015 Agility says Digit is designed for logistics work in spaces where people already work and that Arc connects the robot to existing warehouse automation and management systems. SP014
CP016 Agility and GXO said their 2024 agreement was the first formal commercial deployment of humanoid robots and the first Robots as a Service deployment of humanoid robots. SP015
CP017 Agility’s GXO release also said Digit was moving totes from cobots onto conveyors in a live warehouse while being orchestrated through Agility Arc. SP015
CP018 1X positions NEO as a home robot that automates chores, supports early access, and emphasizes gentle and safe interactions in the home. SP016
CP019 TechCrunch said 1X’s Neo Gamma was still in limited in-home testing and remained a long way from commercial scaling and deployment. SP017
CP020 Apptronik says Apollo is a commercial humanoid designed for mass manufacturability, high payloads, safety, and near-term use in warehouses and manufacturing plants. SP018
CP021 Apptronik’s Mercedes announcement said Apollo could lift 55 pounds and that Mercedes would pilot it for manufacturing logistics and assembly-kit delivery tasks. SP019
CP022 Boston Dynamics discloses Atlas as an enterprise-grade humanoid with four hours of battery life, 50 kilograms of instant payload, 30 kilograms of sustained payload, and Orbit integration into enterprise systems. SP020
CP023 Boston Dynamics says Atlas commercialization starts with Hyundai and builds on a broader software and services ecosystem developed across Spot and other mobile robots. SP021
CP024 Symbotic positions itself as end-to-end AI-enabled warehouse automation for retailers, wholesalers, and food and beverage supply chains. SP022
CP025 Symbotic said Walmart funded a development program and committed to purchase and deploy systems for 400 accelerated pickup and delivery centers if performance criteria are met. SP023
CP026 Locus positions itself as a robots to goods and AMR execution platform that can be deployed in existing warehouses without major redesign. SP024
CP027 Locus customer references cite named deployments with productivity gains, including Brother and Saddle Creek improving SKUs picked per hour from 30 to 80 to 100 and Maersk reporting a 300 percent productivity improvement. SP025
CP028 GreyOrange says its orchestration layer coordinates people, robots, and systems and delivers more than 1 million optimizations every minute. SP026
CP029 GreyOrange also markets customer stories such as doubled fulfillment productivity and large pallet-moving AMR deployments, which strengthens the substitute case against humanoids. SP026
CP030 Berkshire Grey markets robotic trailer unloading and piece-picking systems that directly automate repetitive warehouse tasks without a humanoid body. SP027
CP031 About Amazon says Amazon has deployed more than 1 million robots across its operations network since 2012. SP028
CP032 Amazon says specialized systems such as Sequoia, Vulcan, Sparrow, Cardinal, and Proteus already address sorting, stowing, moving, and package-handling workflows at fulfillment-center scale. SP028
CP033 McKinsey says warehouse buyers can already choose from mature automation options such as AMRs, goods to person systems, shuttle systems, and fee structures like pay per pick. SP007
CP034 IEEE Spectrum argues that the humanoid market is still largely hypothetical because demand, battery life, reliability, and safety remain unsolved at scale. SP029
CP035 IEEE Spectrum also argues that building many humanoids may be easier than finding applications that justify several thousand robots per facility. SP029
CP036 Hark therefore competes simultaneously against direct humanoid builders for flexible embodied AI and against warehouse automation substitutes for real buyer budgets. SP001, SP003, SP005, SP022, SP024, SP026, SP028
CP037 Figure and Hark both market integrated AI plus bespoke hardware, but Figure has already published a clearer AI stack and field deployment narrative. SP001, SP010, SP011
CP038 Switching costs in this market come less from the body plan itself and more from workflow software, fleet management, integrations, and service processes. SP014, SP015, SP020, SP021, SP007
CP039 Agility Arc and Boston Dynamics Orbit are evidence that incumbent humanoid vendors are already trying to own the software and systems layer around the robot. SP014, SP015, SP020
CP040 Symbotic, Locus, GreyOrange, Berkshire Grey, and Amazon collectively show that many warehouse and factory tasks can already be automated by non-humanoid systems. SP022, SP023, SP024, SP025, SP026, SP027, SP028
CP041 Because Hark has not disclosed customers, pricing, deployments, or a concrete robot workflow, its current moat is weaker in public evidence than those of Figure, Agility, Apptronik, or Boston Dynamics. SP001, SP003, SP011, SP015, SP019, SP020
CP042 Falling component costs and larger supply chain participation increase the risk that basic humanoid hardware commoditizes, shifting value toward data, software, and channel access. SP004, SP029
CP043 The most credible near-term displacement risk for Hark in industrial settings is specialized automation with proven ROI rather than another general humanoid startup. SP005, SP007, SP022, SP024, SP028
CP044 1X matters strategically because Hark’s public interface rhetoric overlaps more with personal AI hardware than with a narrowly defined warehouse-robot pitch. SP001, SP002, SP016, SP017
CP045 Public pricing transparency remains poor across direct humanoid peers, which makes workflow proof and partner credibility more important than list price in the current comparison set. SP009, SP013, SP014, SP016, SP018, SP020
CP046 Gartner’s 2026 production-stage warning and McKinsey’s menu of mature alternatives imply Hark should be underwritten as pilot-stage optionality rather than near-term scaled deployment. SP005, SP007
CP047 Apptronik and Boston Dynamics both argue that human-form-factor robots matter when facilities are already designed around people, which is one of the few publicly legible pro-humanoid arguments in this set. SP019, SP021
CP048 Even when the humanoid form factor helps, McKinsey, Gartner, and the substitute vendors all suggest many operators will still prefer lower-risk automation that integrates faster and pays back sooner. SP005, SP007, SP022, SP024, SP026, SP027, SP028
CI001 As of the run date, Hark's homepage says the company is entering beta and reviewing applications to join its platform. SI001
CI002 Hark's homepage describes a product strategy built around bespoke native hardware devices paired with end-to-end speech, text, and vision models plus personalized memory. SI001
CI003 Hark says its systems are multimodal, built from scratch, and intended to interact with the world in a natural way. SI001, SI004
CI004 Hark's careers page says the company is hiring across AI, engineering, and design from San Jose, California. SI002
CI005 Hark's privacy policy says the company offers a website, apps, and other products and services. SI003
CI006 Hark's privacy policy says account information includes payment card information and transaction history. SI003
CI007 Hark's privacy policy says sandbox environments can collect files, commands, code, and task execution logs when Hark executes tasks on a user's behalf. SI003
CI008 Hark's privacy policy says its Browser Operator can access user-authorized page content and act inside authenticated browser sessions. SI003
CI009 Hark's March 2026 launch announcement described the company as a new AI lab founded by Brett Adcock. SI004
CI010 Hark said in March 2026 that it had more than 45 researchers, engineers, and designers. SI004
CI011 Hark said in March 2026 that it had recruited talent from Apple, Meta, Google, Tesla, and leading AI labs. SI004
CI012 Hark said in March 2026 that it had signed a deal for a large cluster of thousands of NVIDIA B200 GPUs coming online in April. SI004
CI013 Hark said in March and May 2026 that software experiences and AI models would come first, with AI-native hardware devices following after software launch. SI004, SI005
CI014 A round of more than $700 million at a $6 billion post-money valuation implies new-money ownership of at least about 11.7% before any option-pool or secondary adjustments. SI005, SI006, SI013, SI014
CI015 Hark said the Series A proceeds would accelerate development of advanced personalized intelligence and next-generation hardware. SI005, SI006
CI016 Hark said Qatalyst Partners provided strategic and financial advice on the May 2026 financing. SI005, SI014
CI017 Intel Capital's post repeated the same funding amount, valuation, and broad investor roster as Hark's May 2026 announcement. SI005, SI006
CI018 TechCrunch reported that Brett Adcock launched Hark in late 2025 with $100 million of his own money. SI007, SI011
CI019 Independent May 2026 reporting described Hark as having roughly 70 employees. SI007, SI011, SI012
CI020 TechCrunch reported that Hark planned to use the new funding for recruiting hardware, product design, and AI research talent and for securing compute and components. SI007, SI009
CI021 TechCrunch and Startup Researcher reported that Hark was operating or securing an NVIDIA B200 data center for model development. SI007, SI012
CI022 TechCrunch said there were still more questions than answers about Hark and highlighted privacy discomfort as a challenge for any always-on personal AI assistant. SI007
CI023 The Next Web said Hark closed its Series A roughly two months after emerging from stealth and before shipping a product. SI008
CI024 The Next Web said Hark had not publicly disclosed target price, launch market, or customer pipeline. SI008
CI025 The Next Web described Hark's category as small, expensive, and littered with failures such as Humane AI Pin and Rabbit R1. SI008
CI026 The Next Web said supply allocation is often the binding constraint on AI hardware companies and that Hark's cap table may ease that constraint relative to peers. SI008
CI027 SiliconANGLE warned that if Hark's devices rely on large-scale cloud inference, operating cost could become prohibitively high unless the company uses more efficient on-device execution. SI009
CI028 SiliconANGLE described Hark as developing custom AI models and AI-optimized devices as an alternative to traditional ways of accessing AI services. SI009
CI029 Ventureburn said Hark planned to expand its engineering organization from 70 to 200 researchers and engineers. SI010
CI030 Grey Journal said Hark reached a $6 billion valuation before shipping a single product. SI011
CI031 Grey Journal argued that Hark's investor mix was designed to support a new device category rather than just a single product launch. SI011
CI032 Startup Researcher said all four major U.S. chip makers invested in the same Hark financing round. SI012
CI033 Yahoo Finance and Morningstar both republished Hark's May 2026 financing announcement, corroborating the amount and valuation but not adding financial detail. SI013, SI014
CI034 Morningstar's mirror of the Business Wire release shows the funding announcement was distributed through third-party financial information channels. SI014
CI035 SEC search results and the Hark Labs Inc. filings page show that Hark Labs Inc. is the filer behind CIK 0002117821 and that the company had one Form D on file as of the run date. SI015, SI016
CI036 Hark Labs Inc.'s Form D says the issuer was incorporated within the prior five years and lists 2025 as the year of incorporation. SI018, SI019
CI037 Hark Labs Inc.'s Form D lists the issuer revenue range as Decline to Disclose. SI018, SI019
CI038 Hark Labs Inc.'s Form D lists March 10, 2026 as the date of first sale. SI018, SI019
CI039 Hark Labs Inc.'s Form D lists a $1.0 billion total offering amount, $50 million sold, $950 million remaining, two investors already invested, and zero sales commissions or finders fees. SI017, SI018, SI019
CI040 TechCrunch reported that Humane raised more than $230 million, cut AI Pin pricing from $699 to $499, saw returns outpace sales, and then sold assets to HP for $116 million. SI020, SI021
CI041 HP said its $116 million Humane deal was meant to speed development of devices that orchestrate AI requests both locally and in the cloud. SI021
CI042 Thunder Compute says B200 access in 2026 remains scarce, lists an MSRP-style range of about $30,000 to $40,000 per GPU in 8+ GPU clusters, and notes hourly cloud pricing examples from roughly $2.80 to $27.04 per GPU-hour plus 1000W power draw. SI022
CI043 Hark's official site shows request-based beta access rather than broad commercial availability. SI001
CI044 Across the fetched Hark homepage, careers page, and privacy policy, Hark does not publish a public software subscription price, API price, or hardware price. SI001, SI002, SI003
CI045 Across fetched official Hark surfaces, the May funding announcement, and the March Form D, no public recognized revenue, ARR, paying-customer count, or unit-sales disclosure appears. SI001, SI005, SI018
CI046 Because Hark's public financing disclosures document capital raised rather than customer payments or recognized sales, the Series A and Form D cannot be treated as revenue. SI001, SI005, SI018
CI047 The March Form D and the May financing announcement together imply that Hark's public capital raising progressed materially between March and May 2026 within a financing process larger than the $50 million first-sale snapshot in the filing. SI005, SI017, SI018
CI048 A strategy that combines in-house multimodal models, thousands of B200 GPUs, and bespoke hardware implies a capital profile closer to compute infrastructure plus device development than to a lightweight software startup. SI004, SI007, SI022
CI049 Because Hark has not publicly disclosed hardware ASP, hardware BOM, software pricing, or compute contract terms, its margin path cannot be underwritten from public evidence. SI001, SI003, SI008, SI009, SI022
CI050 Because public sources disclose financing size and use of proceeds but not cash balance, monthly burn, runway, or obligations, financing dependency remains a material diligence issue. SI005, SI018, SI019
CI051 Public evidence supports a source-backed financing visibility range from $50 million sold by March 10, 2026 to more than $700 million disclosed by May 21, 2026, against a $1.0 billion Form D total offering amount. SI005, SI018, SI019
CI052 Independent coverage and official surfaces do not disclose Hark's launch market, target price, or customer pipeline as of the run date. SI001, SI008
CI053 The public record therefore documents product ambition, financing access, and infrastructure buildup much more clearly than it documents revenue quality, margin, or capital adequacy. SI001, SI005, SI018, SI022
CE001 Hark’s public product surface is still a request-access beta rather than a fully documented commercial launch. SE001, SE013
CE002 Hark publicly defines the product as advanced personal intelligence built from multimodal models, personalized memory, and bespoke native hardware. SE001, SE004, SE005
CE003 The company’s core promise is a universal interface between humans and machines rather than a single-purpose app or wearable accessory. SE005, SE006
CE004 Public materials do not yet disclose a concrete device form factor, target price, launch market, or named customer deployments. SE001, SE004, SE007, SE010, SE028
CE005 Hark’s published sequence is software experiences and multimodal models in summer 2026, followed by AI-native hardware later. SE004, SE005, SE006, SE007
CE006 Because Hark has not published pricing, hardware specs, or customer proof, the current public workflow is better evidenced as prelaunch positioning than as an underwritten product release. SE001, SE004, SE007, SE010, SE028
CE007 Hark’s privacy policy covers website, apps, and other services, indicating a broader software surface than the homepage alone reveals. SE003
CE008 The privacy policy says users may interact with Hark through chat, voice, file uploads, and agentic sessions. SE003
CE009 The privacy policy describes Third-Party Apps, Connectors, and a Browser Operator extension that can act inside the user’s local browser session. SE003
CE010 Hark says its task execution runs in isolated sandbox environments that can create files, run shell commands, and generate task logs. SE003
CE011 Hark’s Integration Engineer role says the product is meant to connect to email, calendar, productivity, communication, and developer tools, and to take actions on the user’s behalf. SE020
CE012 The same integrations role explicitly references REST, GraphQL, gRPC, MCP, OAuth flows, webhooks, and execution sandboxing, implying a fairly opinionated tool-use stack. SE020
CE013 Hark’s public job board spans AI foundation models, AI infrastructure, computer-use agents, embedded software, hardware engineering, privacy and security, mobile, and product engineering. SE014, SE015
CE014 The Pretraining role says Hark’s Omni team works across text, audio, and vision with data curation, deduplication, synthetic data generation, distributed training, and internal benchmarks. SE016
CE015 The Large-scale Training role says Hark wants training infrastructure at the scale of 10,000-plus GPUs with job scheduling, fault tolerance, incident response, and network optimization. SE017
CE016 Hark’s own fundraising post says it has secured a new NVIDIA B200 data center to train the next generation of models. SE004, SE006
CE017 TechCrunch reported in May 2026 that Hark had about 70 employees and ran a data center with Nvidia B200 GPUs. SE007
CE018 The Operating System Architect role describes a consumer electronics portfolio that is expected to move from NPI and prototype builds through validation, production ramp, and sustaining engineering. SE018
CE019 The OS Architect role points to an AI-first embedded stack spanning BSP, kernel, middleware, services, application frameworks, secure boot, OTA updates, and provisioning workflows. SE018
CE020 Hark’s Greenhouse roles indicate voice-first consumer hardware ambitions through audio firmware, microphone and speaker integration, DSP, always-on listening paths, and factory audio test workflows. SE015
CE021 The hardware roles also imply manufacturing dependence on contract manufacturers, factory bring-up, reliability validation, thermals, and hardware test infrastructure. SE015
CE022 The Privacy Engineer role indicates Hark expects to build DSAR, deletion, data-retention, de-identification, tokenization, and privacy incident-response systems across live services. SE019
CE023 The AI Safety Engineer role indicates Hark expects multimodal content moderation classifiers, detection pipelines, production monitoring, and abuse mitigation rather than relying on policy copy alone. SE021
CE024 Hark’s Browser Operator language says the company can access authorized page content from logged-in sessions but says it does not store login credentials. SE003
CE025 Hark’s retained official surface is unusually thin for a company claiming a new computing interface: the visible official pages are effectively homepage, careers, privacy, and one funding article. SE013, SE001, SE002, SE003, SE004
CE026 The careers page and launch announcement emphasize a San Jose team drawn from Apple, Meta, Google, Tesla, and leading AI labs. SE002, SE005, SE012
CE027 Independent coverage confirms that Abidur Chowdhury left Apple to lead design at Hark. SE005, SE012
CE028 Hark’s public design language is strongly human-first and ambient, but that branding is not yet matched by published device ergonomics, safety testing, or usability data. SE001, SE002, SE028
CE029 Figure’s official Helix materials show Brett Adcock’s recent technical strategy emphasizing vertically integrated models, hardware, and end-to-end control rather than outsourcing the core intelligence layer. SE022, SE025
CE030 TechCrunch quoted Adcock saying embodied AI at scale requires vertically integrated robot AI and that Figure could not outsource AI any more than hardware. SE025
CE031 Figure’s Helix and Helix 02 articles describe a hierarchy that moved from System 1 and System 2 to System 0, 1, and 2, combining language, vision, touch, and real-time control. SE022, SE023
CE032 Figure says Helix 02 trains a whole-body controller on more than 1,000 hours of human motion data and over 200,000 simulated environments, replacing 109,504 lines of hand-engineered C++. SE023, SE026, SE027
CE033 Figure’s BMW deployment provides evidence that Adcock’s recent teams can take integrated hardware and AI systems into extended real-world runtime rather than staying only in lab demos. SE024, SE027
CE034 No retained source shows Hark reusing Figure code, models, data, or supply-chain relationships, so any transfer thesis remains founder-pattern inference rather than disclosed fact. SE005, SE008, SE025
CE035 Independent coverage repeatedly frames Hark as unusually opaque or pre-product relative to the size of its funding round. SE007, SE010, SE028
CE036 TNW explicitly said Hark had not disclosed headcount, hardware form factor, target price, launch market, or customer pipeline as of late May 2026. SE010
CE037 TechCrunch highlighted privacy discomfort as an unresolved design problem for always-on personal AI hardware that might observe people around the user. SE007
CE038 Taken together, the privacy policy and privacy-engineering role imply that Hark expects significant regulated-data and consent burdens well before public hardware details are clear. SE003, SE019
CE039 The software and agent-runtime side of Hark is publicly more concrete than the hardware side, because legal pages and integration roles describe capabilities in much more detail than any device page does. SE001, SE003, SE020
CE040 Hark’s public product maturity today is strongest in strategic narrative, hiring evidence, and enabling infrastructure, but still weak in shipped specs, reliability disclosure, and third-party trust proof. SE001, SE014, SE016, SE018, SE028
CU001 Hark’s homepage presents the product as a personal-intelligence system paired with bespoke native hardware and multimodal speech, text, and vision models. SU001
CU002 The homepage says Hark wants to bring advanced personal intelligence to everyone in the world, which points to a mass-market framing rather than a named-vertical customer pitch. SU001
CU003 Hark’s privacy policy says the Services include the website, apps, and other products and services, which supports an account-based software surface beyond future hardware. SU002
CU004 The privacy policy says Hark collects account information including payment card information and transaction history. SU002
CU005 The privacy policy says Hark supports prompts and outputs across chat, voice, file uploads, and agentic sessions. SU002
CU006 The privacy policy says users can integrate third-party applications with Hark through connectors. SU002
CU007 The privacy policy says the Browser Operator can read, extract, and act on authorized web pages within a user’s local browser environment. SU002
CU008 The visible careers page emphasizes AI, engineering, and design hiring rather than a public sales or customer-reference narrative. SU003
CU009 Hark’s March 2026 launch release said the company planned to roll out software experiences and AI models in summer 2026 and AI-native hardware soon after. SU005
CU010 Hark’s official May 2026 funding article repeated that the AI platform would be available in summer 2026 and hardware would come next. SU004
CU011 TechCrunch described Hark in late May 2026 as still revealing little about what it was building while expecting first multimodal models that summer. SU007
CU012 Across the reviewed Hark homepage, launch materials, funding materials, and TechCrunch financing coverage, Hark publicly names no customer, pilot, deployment, or case study. SU001, SU004, SU005, SU007
CU013 Forbes wrote that Hark had no customers to name and no traction to underwrite when it raised its Series A. SU009
CU014 Observer said in March 2026 that Hark’s first AI models would arrive in summer 2026 followed shortly by hardware, which places public customer rollout in the future tense. SU008
CU015 Hark’s official materials frame the product around a system that understands and acts for an individual user across existing products and services. SU001, SU004
CU016 Because Hark’s policy surfaces mention payments, connectors, transactions, and browser actions, any future expansion loop is more likely to come from deeper account permissions and future hardware attachment than from already-proven enterprise upsell. SU002, SU004
CU017 No reviewed Hark source discloses NRR, GRR, churn, renewal terms, contract length, or customer satisfaction metrics for this company. SU001, SU002, SU004, SU007
CU018 Hark’s SEC Form D marked the revenue range as decline to disclose. SU011
CU019 Hark’s financing materials and coverage emphasize recruiting, compute, and components rather than customer adoption metrics. SU004, SU006, SU007
CU020 Intel Capital’s post on Hark discussed product timing and infrastructure but did not add named customer proof. SU010
CU021 Figure says its BMW deployment ran 10-hour shifts on every working day during the active assembly-line rollout. SU012
CU022 Figure says the BMW deployment loaded more than 90,000 parts over 1,250-plus runtime hours and contributed to more than 30,000 X3 vehicles. SU012, SU013
CU023 BMW publicly discussed expanding humanoid-robot activity to Leipzig, which indicates that the Figure relationship advanced beyond a one-off demo. SU013
CU024 TechCrunch reported that Figure’s CEO sidestepped questions about BMW deal scale on stage, which shows that even named customer proof can still leave transparency gaps. SU014
CU025 Agility and GXO announced a multi-year agreement to deploy Digit after a late-2023 proof of concept. SU015, SU016
CU026 Agility said Digit was already deployed at a customer site, generating revenue, and solving real-world business problems. SU015
CU027 The Apptronik and Mercedes-Benz announcement described Apollo as Apptronik’s first publicly announced commercial deployment and Mercedes’ first humanoid-robot application, with specific kit-delivery and tote-delivery tasks. SU017
CU028 Symbotic said Walmart committed to purchase and deploy systems for 400 accelerated pickup and delivery centers if performance criteria are achieved. SU018, SU019
CU029 Locus said DHL expanded AMRs across more than 40 sites and reached one billion picks with reported 30 to 180 percent productivity gains and an 80 percent reduction in training time. SU020, SU021, SU022
CU030 Amazon says more than 750,000 robots now work alongside employees across its fulfillment and transportation network. SU023
CU031 Gartner said fewer than 20 companies will go live in production with humanoid robots for supply chain and manufacturing use cases by 2028. SU024
CU032 Gartner said most humanoid production deployments will remain limited to tightly controlled environments rather than dynamic, high-throughput operations. SU024
CU033 IEEE Robotics and Automation Society wrote that the humanoid-robot market is still almost entirely hypothetical and that leading companies have only small numbers of robots in carefully controlled pilots. SU025
CU034 IEEE Robotics and Automation Society reported that customers care intensely about downtime, reliability, safety, and battery life before scaling humanoid systems. SU025
CU035 Rodney Brooks argued that technologies often take much longer to move from proof of concept into scaled deployment than early hype suggests. SU026
CU036 The strongest buyer hypothesis supported by current evidence is that Hark’s first monetizable relationship is a direct account holder using personal or small-team workflows rather than a publicly named enterprise buyer. SU001, SU002, SU007
CU037 Hark’s public adoption trajectory remains pre-deployment because the sources show launch, funding, hiring, and future timing rather than active customer deployments. SU004, SU005, SU007, SU008
CU038 Hark’s public proof bar is materially below the benchmark set by Figure/BMW, Agility/GXO, Apptronik/Mercedes, Symbotic/Walmart, and Locus/DHL because those pairs name counterparties and operating use cases while Hark does not. SU012, SU015, SU017, SU018, SU020, SU001, SU004, SU007
CU039 Because Hark has not disclosed customer count or top-account mix, concentration risk cannot be quantified publicly, but any initial launch cohort is likely to be narrow enough that concentration should be assumed until disproved. SU004, SU006, SU009
CU040 The public materials imply a possible land-and-expand path through more permissions, more connected workflows, and later hardware attachment, but they do not show a realized upsell motion yet. SU002, SU004
CU041 Hark’s described product design creates meaningful adoption friction because customers would need to trust an AI system with personal data, payment-linked accounts, connected services, and browser-level action authority. SU002, SU007
CU042 Gartner, IEEE, and Rodney Brooks collectively suggest that real customer adoption in ambitious automation categories tends to reward narrow, validated workflows instead of broad category promises. SU024, SU025, SU026
CU043 As of the run date, Hark’s public customer proof quality is logo-free and outcome-free because no public source reviewed here provides ROI, seat counts, deployment metrics, or a named reference customer for Hark. SU001, SU004, SU005, SU007, SU009
CU044 Every public retention or satisfaction judgment for Hark must remain null until the company discloses cohorts, renewals, usage depth, or referenceable customer outcomes. SU001, SU002, SU004, SU007
CU045 IIoT World described BMW and Figure as the first real production-validated data point for physical AI and said BMW was expanding deployment to Leipzig in summer 2026. SU027
CU046 VaaSBlock said that the only two verified Western production humanoid deployments in 2026 were Figure at BMW and Agility at GXO, with other vendors still in pilot or demo tiers. SU028
CR001 Hark describes its product as advanced personalized intelligence built across foundation models, software systems, native hardware, and new interfaces. SR001, SR004
CR002 Hark says its system is meant to listen, speak, see, maintain persistent memory, behave proactively, and influence the world around the user. SR001, SR004
CR003 Hark says it plans to roll out software experiences and AI models in summer 2026, with AI-native hardware devices soon after. SR004, SR005, SR007
CR004 Hark's March 2026 launch materials said it had more than 45 researchers, engineers, and designers. SR004
CR005 TechCrunch reported that Hark had about 70 employees by the May 2026 financing and intended to spend new capital on talent, compute, and components. SR007
CR006 Hark said it signed a deal for thousands of Nvidia B200 GPUs to support multimodal pre-training and model post-training. SR004, SR007
CR007 Hark raised over $700 million in Series A financing at a $6 billion post-money valuation on 2026-05-21. SR005, SR006
CR008 Hark's launch and funding materials still describe a roadmap and platform thesis rather than public customer, revenue, or pricing proof. SR004, SR005, SR007
CR009 TechCrunch said Hark launched in late 2025 with $100 million of Brett Adcock's own money. SR007
CR010 Observer said Hark was initially financed entirely by Adcock's own capital before the larger outside round. SR008
CR011 Brett Adcock's biography states that he remains involved with Archer, Figure, and Cover while also starting Hark. SR011
CR012 Observer said Adcock remains CEO of Figure while Hark is a separate company. SR008, SR011
CR013 Hark's careers page says the company is hiring across AI, engineering, and design in San Jose and acknowledges that building Hark will be hard. SR003
CR014 Hark said its hardware team includes veterans from Apple, Tesla, and Meta with mechanical, electrical, firmware, embedded, supply-chain, and operations experience. SR004
CR015 Figure's website says Figure 03 is a general-purpose humanoid robot for the home, showing that Adcock's parallel hardware effort remains active. SR012, SR011
CR016 Figure's BMW deployment post says Figure 02 logged more than 1,250 runtime hours and that the forearm was the top hardware failure point. SR013
CR017 BMW says its German humanoid deployment is a pilot project and frames the robots as supporting people rather than replacing them. SR014
CR018 TechCrunch reported that Figure faced skepticism about whether the BMW relationship was a pilot or commercially valuable and that Adcock did not provide contract specifics. SR015
CR019 TechCrunch reported that Figure had not done a live robot demo at major events even as it publicized videos and targeted roughly 100,000 units within four years. SR015
CR020 TechCrunch reported that Figure was pursuing a $1.5 billion raise at about a $39.5 billion valuation amid commercialization questions. SR015
CR021 Hark's privacy policy says the company collects account data, chat, voice, file, and agentic-session inputs and outputs, third-party app content, device information, browsing data, and location-derived data. SR002
CR022 Hark's privacy policy says it may collect sandbox files, shell commands and outputs, generated code, and task execution logs when acting on a user's behalf. SR002
CR023 Hark's privacy policy says some image, audio, and avatar features may create data that could be considered biometric under EU and some U.S. state laws. SR002
CR024 Hark's privacy policy says it will seek notices and consents required by law for biometric processing and maintain a public retention schedule and destruction guidelines. SR002
CR025 Hark tells users not to provide sensitive personal information through the service. SR002
CR026 Hark says its services are not directed to children under 18. SR002
CR027 CPPA rules approved in September 2025 took effect on 2026-01-01, with risk-assessment obligations starting in 2026 and ADMT requirements beginning in 2027. SR023
CR028 CISA's AI guidance highlights careful adoption of agentic AI services, secure deployment of externally developed AI systems, and secure-by-design principles. SR024
CR029 BIS in May 2026 reiterated license requirements and due-diligence obligations for advanced computing items and extended the approved IC designer timeline through 2026-12-31. SR025
CR030 CHIPS for America says semiconductors are essential to AI and data centers and that the U.S. is spending $50 billion to strengthen domestic semiconductor R&D and manufacturing. SR026
CR031 MIT Technology Review quoted roboticists saying humanoids are mostly not intelligent, lack common sense, require big batteries, and are complex to manufacture. SR016
CR032 MIT Technology Review said safety regulations for humans working alongside humanoids do not yet exist and that adoption is likely drawn out, industry specific, and slow. SR016
CR033 Berkeley roboticist Ken Goldberg said humanoid robots are unlikely to reach the touted capability level within the next two, five, or even ten years. SR017
CR034 Goldberg said dexterity limits and a 100,000-year data gap make real-world robot skill acquisition much slower than LLM progress. SR017
CR035 McKinsey said commercial viability depends on radical cost reduction, better dexterity and mobility, sustained uptime, and fenceless-operation safety. SR019
CR036 McKinsey said typical humanoid bills of materials still range from roughly $30,000 to $150,000 and require significant compression to unlock mass-market demand. SR019
CR037 McKinsey said critical bottlenecks include harmonic drives, roller screws, force and tactile sensing, and permanent magnets with China-heavy processing. SR019
CR038 Bain said most humanoid deployments remain in pilots and still rely heavily on human supervision. SR020
CR039 Bain said most humanoids today operate for only about two hours and that a full eight-hour shift could take up to a decade. SR020
CR040 Bain said commercial success also requires regulatory pathways, safety certification, workforce acceptance, and public trust. SR020
CR041 TechCrunch said Rodney Brooks and other robotics investors and scientists are skeptical about near-term humanoid revenues, use cases, safety, and unit economics. SR021
CR042 The same TechCrunch piece said even Nvidia executives compared humanoid timelines to the slower-than-expected path of self-driving cars. SR021
CR043 SEC records show Hark Labs Inc. is a Delaware corporation with a Form D filed on 2026-03-24 and a first sale date of 2026-03-10. SR027, SR028
CR044 The same Form D shows a $1 billion total offering amount, $50 million sold, and $950 million remaining to be sold at the filing date. SR027, SR028
CR045 TechCrunch reported that Humane discontinued the AI Pin, warned devices would stop functioning at the end of February 2025, and said returns had outpaced sales. SR029
CR046 HP said it acquired Humane's AI platform, talent, and more than 300 patents and patent applications for $116 million. SR029, SR030
CR047 TechCrunch said one unresolved challenge for Hark is giving a personal AI enough life context to be useful without making bystanders uncomfortable or violating privacy. SR007
CR048 Observer said Hark expected headcount to reach 100 in the first half of 2026. SR008
CV001 Using Hark's announced round size of more than $700 million and its $6 billion post-money headline, the financing implies a pre-money valuation of roughly $5.3 billion before later dilution adjustments. SV003, SV004, SV005
CV002 Hark said the round was oversubscribed and led by Parkway Venture Capital, with participation from Nvidia, AMD Ventures, Qualcomm Ventures, Salesforce Ventures, ARK Invest, Brookfield, Greycroft, Intel Capital, Prime Movers Lab, and Tamarack Global. SV003, SV005
CV003 Hark said it plans to roll out its first AI models later in summer 2026 and then introduce AI-native hardware devices designed specifically for those systems. SV002, SV003, SV004
CV004 Hark’s March 2026 Form D showed a total offering amount of $1.0 billion, $50 million sold, and $950 million remaining to be sold, with a first sale date of March 10, 2026. SV006, SV007
CV005 The same Form D marked Hark’s revenue range as “Decline to Disclose.” SV006, SV007
CV006 Hark’s official materials describe a vertically integrated plan spanning foundation models, software systems, native hardware, and new human-machine interfaces. SV001, SV002, SV003
CV007 As of June 11, 2026, Hark’s public materials do not disclose a price list, paid customer logos, ARR, gross margin, or unit economics. SV001, SV003, SV009
CV008 TechCrunch reported that Hark had roughly 70 employees at the time of the Series A announcement. SV004
CV009 Hark’s March 2026 launch announcement said a large cluster of thousands of Nvidia B200 GPUs was coming online in April 2026. SV002
CV010 A $700 million round at a $6 billion post-money implies minimum new-money ownership of about 11.7% before any option-pool or secondary adjustments. SV003
CV011 Figure raised $675 million in February 2024 at a $2.6 billion valuation, with investors including Nvidia, Microsoft, OpenAI, Jeff Bezos, Intel Capital, Parkway, Align Ventures, and ARK Invest. SV010
CV012 Figure announced in September 2025 that it had exceeded $1 billion in committed Series C capital at a $39 billion post-money valuation. SV013, SV014
CV013 Figure’s BMW deployment ran 10-hour shifts, logged more than 1,250 runtime hours, loaded 90,000-plus parts, and contributed to the production of 30,000-plus BMW X3 vehicles. SV011
CV014 Figure publicly discloses embodied-AI software through Helix plus scaling plans for BotQ manufacturing, giving it materially more visible technical and operational proof than Hark has published. SV011, SV012, SV013
CV015 Apptronik said in February 2026 that its reopened Series A totaled more than $935 million and brought total capital raised to nearly $1 billion. SV025, SV026
CV016 TechCrunch reported Apptronik’s post-money valuation at about $5.3 billion in February 2026, roughly triple its initial Series A valuation of around $1.75 billion. SV026, SV027
CV017 Apptronik had already announced a Mercedes-Benz commercial agreement to pilot Apollo in manufacturing facilities before its 2026 valuation step-up. SV028, SV025
CV018 Agility and GXO announced a multi-year agreement in June 2024, giving Agility public deployment evidence for Digit in logistics. SV020
CV019 Public 2026 sources place Agility’s latest valuation around $2.1 billion to $2.15 billion after a roughly $400 million 2025 round. SV021, SV022
CV020 Public.com provides only a secondary-market share-price estimate for Agility rather than a fresh official round price, highlighting remaining opacity in private-company marks. SV023
CV021 1X closed a $100 million Series B in January 2024 and had raised about $125 million in less than a year after adding its earlier Series A. SV016, SV017
CV022 Two September 2025 reports said 1X was seeking up to $1 billion at a valuation of at least $10 billion, but those articles described a fundraising target rather than a completed round. SV018, SV019
CV023 Hyundai completed its acquisition of Boston Dynamics in June 2021, with Hyundai stating that the transaction valued Boston Dynamics at $1.1 billion. SV029, SV030
CV024 Goldman Sachs projects the global humanoid-robot market could reach $38 billion by 2035, supporting a large upside for category leaders that actually commercialize. SV032
CV025 Gartner predicted in January 2026 that fewer than 20 companies will scale humanoid robots to production in manufacturing and supply chain by 2028, while polyfunctional robots win the warehouse sooner. SV031
CV026 IEEE Spectrum argued in 2025 that the bigger bottleneck for humanoids is demand, not supply, and questioned whether any use case requires thousands of robots per facility. SV033
CV027 IEEE RAS said humanoid companies have raised hundreds of millions at billion-dollar valuations built on promises before widespread scaled deployment was proven. SV034
CV028 Rodney Brooks’ published predictions argue that frontier technologies usually take longer to adopt than hype cycles suggest. SV035
CV029 McKinsey says warehouse automation is expanding quickly but too many projects still fail to deliver the expected results, which matters because buyers can choose lower-risk specialized automation instead of humanoids. SV038
CV030 Amazon says it has deployed more than one million robots since 2012 across its operations network, demonstrating that specialized systems already absorb large amounts of warehouse labor without general-purpose humanoids. SV039
CV031 Humane raised more than $230 million for the AI Pin, then sold most assets to HP for $116 million in 2025 while discontinuing the product, creating a cautionary AI-hardware downside benchmark. SV036, SV037
CV032 Hark’s $6 billion post-money is roughly 13% above Apptronik’s reported $5.3 billion post-money despite Apptronik having far more public commercialization evidence. SV003, SV026, SV028
CV033 Hark’s $6 billion price is roughly 2.8 times Agility’s $2.15 billion benchmark even though Agility has publicly named deployment partners and a clearer logistics workflow. SV003, SV020, SV021
CV034 Hark’s $6 billion price is more than five times Boston Dynamics’ disclosed $1.1 billion 2021 acquisition value while offering much less public product and revenue proof. SV003, SV023, SV030
CV035 Hark is cheaper than Figure’s late-2025 $39 billion round and 1X’s reported $10 billion target, but those references also come with more public operating context than Hark has supplied. SV013, SV018, SV019
CV036 Hark’s $6 billion price is about 2.3 times Figure’s February 2024 $2.6 billion financing even though Figure already had a named BMW partnership on the way and later demonstrated real line runtime. SV010, SV011, SV003
CV037 Public evidence supports Hark’s access to capital and ambition, but it does not yet support a revenue- or customer-backed fair value at $6 billion. SV003, SV004, SV007, SV009
CV038 Without new proof, a more defensible near-term Hark valuation anchor is closer to the $2.1 billion to $5.3 billion band spanned by Agility and Apptronik than to the announced $6 billion price. SV021, SV022, SV026
CV039 A bull case above the current valuation requires Hark to turn its summer 2026 model rollout into observable paid adoption and show that native hardware creates a durable interface moat. SV002, SV003, SV004
CV040 Because Hark has not disclosed revenue, pricing, or customer proof, any scenario analysis should discount the current price for evidence risk rather than anchor on founder pedigree alone. SV003, SV004, SV007, SV031
CV041 A scenario range of $0.75 billion to $1.5 billion bear, $2.5 billion to $4.0 billion base, and $8 billion to $12 billion bull yields a probability-weighted midpoint below the current $6 billion price. SV021, SV026, SV031, SV032
CV042 At a $6 billion entry price, even a strong $10 billion outcome produces only about 1.7x gross MOIC before follow-on dilution and time, while base and bear cases destroy capital. SV003, SV018, SV021, SV026
CV043 The visible dilution from the new money is only part of the capital-structure story because Hark has not disclosed liquidation preferences, ratchets, secondary allocations, or option-pool refreshes. SV003, SV006, SV007
CV044 No public secondary trade, investor letter, or independent mark surfaced in this chapter’s research to corroborate Hark’s $6 billion price beyond company and press reports. SV003, SV004, SV005
CV045 At the current price, the appropriate recommendation is avoid because downside asymmetry and missing evidence outweigh the upside optionality available from public information. SV031, SV032, SV036
CV046 The recommendation confidence is medium: the price signal is clear, but cap-table details, monetization, and customer proof remain too opaque for high-confidence precision. SV003, SV006, SV007, SV037
CV047 Hark’s risk rating is high because the company combines prelaunch execution risk, hardware capital intensity, and a category that multiple expert sources say may remain stuck in pilots. SV004, SV025, SV031, SV033, SV034
CV048 Hark’s valuation stance is expensive because the announced price sits above better-evidenced private-robotics comps on the public record. SV021, SV026, SV032, SV037
CV049 A realistic upgrade path is either a materially lower entry price around or below the mid-single-digit billions or new evidence showing paid adoption, retention, and hardware differentiation. SV003, SV038, SV039
CV050 Hark’s privacy policy implies the company expects payment-enabled services, but it still does not disclose actual pricing or realized commercial usage. SV009
来源
编号出版方标题引文
SO001 Hark Hark homepage At Hark, we are building the most advanced personal intelligence in the world.
SO002 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence Hark currently has a team of more than 45 researchers, engineers, and designers from leading AI and technology companies.
SO003 Hark Hark announces $700 million fundraising round We've raised over $700 million in Series A funding at a $6 billion post-money valuation.
SO004 Business Wire Hark Raises $700M Series A at a $6B Valuation The round was oversubscribed and led by Parkway Venture Capital.
SO005 Intel Capital Hark Raises $700M Series A at a $6B Valuation SAN JOSE, Calif.—Hark... has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SO006 TechCrunch Hark raises $700M Series A for its secretive universal AI interface Hark currently has 70 employees and runs a data center with Nvidia B200 GPUs.
SO007 Observer Billionaire Brett Adcock Launches New Startup to Build Personal A.I. For now, Hark is financed entirely by Adcock’s own money: $100 million in personal capital.
SO008 SiliconANGLE Hark raises $700M+ to build personalized intelligence devices
SO009 The Next Web Brett Adcock’s AI hardware startup Hark raises $700m at $6bn valuation According to the BusinessWire announcement in March, Hark intends to release its first multi-modal models this summer.
SO010 RoboHorizon Figure CEO Brett Adcock unveils Hark, a secretive AI hardware firm Hark has already assembled a team of about 45 engineers and designers.
SO011 Yahoo Finance Hark Raises $700M Series A at a $6B Valuation Hark... has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SO012 Brett Adcock Bio | Brett Adcock Official I founded and grew Archer into a leading aerospace company... I founded Figure to give AI a body.
SO013 Wikipedia Brett Adcock In 2025, Adcock founded Hark, a startup working on AI models and next-generation AI hardware devices.
SO014 Forbes Brett Adcock profile Brett Adcock founded robot maker Figure in 2022.
SO015 Figure AI Figure homepage Figure 03 is a general purpose humanoid robot for every day.
SO016 Archer Archer company page
SO017 TechCrunch Figure drops OpenAI in favor of in-house models To solve embodied AI at scale in the real world, you have to vertically integrate robot AI.
SO018 TechCrunch Figure AI CEO skips live demo, sidesteps BMW deal questions onstage at tech conference Figure has recently been the subject of a couple news articles that questioned its progress with marquee customer BMW.
SO019 Forbes This Robot Billionaire From Figure AI Has Big Backers And Big Challenges Armed with more than $750 million in funding, Brett Adcock vows that Figure will become one of the most important businesses in the world. First, he has a lot of work to do.
SO020 Forbes Hark’s $6 Billion Valuation With No Product Actually Makes Sense There’s no product to scale, no customers to name, no traction to underwrite.
SO021 BestStartup Hark Raises $700M Series A to Build Personal AI Hardware On May 21, 2026, San Jose-based AI lab Hark announced it had raised over $700 million in a Series A round at a $6 billion post-money valuation.
SO022 MarketScreener Brett Adcock: positions and network Figure AI was founded in 2022 by Brett Adcock, who has been the CEO since 2022.
SO023 Robotics & Automation News The state of humanoid robotics: from research labs to real-world potential The humanoid robotics field has generated fresh headlines that underscore both the accelerating pace of innovation and the widening range of actors entering the race.
SO024 PitchBook Hark company profile
SO025 Figure AI Figure updates page
SM001 Hark Hark At Hark, we are building the most advanced personal intelligence in the world.
SM002 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence Hark is creating intelligence paired with next-generation hardware designed to serve as a universal interface between humans and machines.
SM003 Goldman Sachs The global market for humanoid robots could reach $38 billion by 2035 The total addressable market for humanoid robots is projected to reach $38 billion by 2035.
SM004 Gartner Gartner Predicts Fewer Than 20 Companies Will Scale Humanoid Robots for Manufacturing and Supply Chain to Production Stage by 2028 Through 2028, less than 100 companies will progress humanoid robot proofs of concept beyond experimentation, with fewer than 20 companies going live in production.
SM005 International Federation of Robotics World Robotics Reports
SM006 Business Wire Top 5 Global Robotics Trends 2026 – International Federation of Robotics Reports Global industrial robotics installations are forecast to surpass 700,000 units in 2028.
SM007 Stanford HAI Artificial Intelligence Index Report 2026
SM008 McKinsey & Company Maximize ROI from warehouse robotics A warehouse automation revolution is underway, but too many projects are not delivering the results.
SM009 MHI Annual Industry Reports
SM010 DHL Trend Reports
SM011 IEEE Spectrum Humanoid Robots: The Scaling Challenge The bigger problem is demand—I don’t think anyone has found an application for humanoids that would require several thousand robots per facility.
SM012 IEEE Robotics and Automation Society Reality Is Ruining the Humanoid Robot Hype
SM013 Rodney Brooks My Dated Predictions Turning them into being deployed at scale is even harder.
SM014 U.S. Bureau of Labor Statistics Table 1. Job openings levels and rates by industry and region, seasonally adjusted - 2026 M04 Results
SM015 U.S. Bureau of Labor Statistics Laborers and Material Movers, Hand: Occupational Outlook Handbook
SM016 World Economic Forum The Future of Jobs Report 2025
SM017 Figure F.02 Contributed to the Production of 30,000 Cars at BMW
SM018 BMW Group First humanoid robot introduced in Plant Leipzig
SM019 Agility Robotics GXO Signs Industry-First Multi-Year Agreement with Agility Robotics
SM020 Apptronik Apptronik and Mercedes-Benz Enter Commercial Agreement
SM021 Amazon Amazon robotics: Meet the robots inside fulfillment centers
SM022 Amazon Supply Chain Services Designed to deliver: the next-level tech behind Amazon’s fulfillment network
SM023 World Health Organization Assistive technology data portal
SM024 World Health Organization Assistive technology
SM025 World Health Organization WHO launches new guide to help shape assistive technology markets
SM026 International Federation of Robotics World Robotics 2025 - Service Robots
SP001 Hark Hark
SP002 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence
SP003 TechCrunch Hark raises $700M Series A for its secretive universal AI interface
SP004 Goldman Sachs The global market for humanoid robots could reach $38 billion by 2035
SP005 Gartner Gartner Predicts Fewer Than 20 Companies Will Scale Humanoid Robots for Manufacturing and Supply Chain to Production Stage by 2028
SP006 International Federation of Robotics World Robotics Reports
SP007 McKinsey & Company Getting warehouse automation right
SP008 Figure Figure
SP009 Figure Figure 03
SP010 Figure Helix
SP011 Figure Production at BMW
SP012 TechCrunch Figure reaches $39B valuation in latest funding round
SP013 Electrek Tesla Optimus
SP014 Agility Robotics Solutions
SP015 Agility Robotics GXO Signs Industry-First Multi-Year Agreement with Agility Robotics
SP016 1X NEO
SP017 TechCrunch Norway's 1X is building a humanoid robot for the home
SP018 Apptronik Apollo
SP019 Apptronik Apptronik and Mercedes-Benz Enter Commercial Agreement
SP020 Boston Dynamics Atlas
SP021 Boston Dynamics Electric new era for Atlas
SP022 Symbotic Symbotic
SP023 Symbotic Symbotic completes acquisition of Walmart's Advanced Systems and Robotics business and signs related commercial agreement
SP024 Locus Robotics Locus Robotics
SP025 Locus Robotics Customers
SP026 GreyOrange GreyOrange
SP027 Berkshire Grey Berkshire Grey
SP028 About Amazon Amazon uses robots that sort, lift, and carry packages—see them in action
SP029 IEEE Spectrum Humanoid Robots: The Scaling Challenge
SI001 Hark Hark homepage Hark is entering beta
SI002 Hark Careers - Hark We’re hiring candidates who are hungry to make their impact across AI, engineering, and design at our headquarters in San Jose, CA.
SI003 Hark Privacy Policy - Hark Hark collects information associated with your account, including ... payment card information, and transaction history.
SI004 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence Hark has recently signed a deal to scale its compute capacity, with a large cluster of thousands of NVIDIA B200 GPUs coming online in April.
SI005 Business Wire Hark Raises $700M Series A at a $6B Valuation Hark ... has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SI006 Intel Capital Hark Raises $700M Series A at a $6B Valuation Hark ... announced it has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SI007 TechCrunch Hark raises $700M Series A for its secretive universal AI interface Still, there are more questions than answers.
SI008 The Next Web Brett Adcock’s AI hardware startup Hark raises $700m at $6bn valuation The category Hark is entering is small, expensive, and littered with failures.
SI009 SiliconANGLE Hark raises $700M+ to build personalized intelligence devices If Hark’s devices are adopted by a significant number of users, the cost of performing cloud-based inference could become prohibitively high.
SI010 Ventureburn Hark Raises $700M to Build Personalized AI Hardware The enormous infusion of capital will also be leveraged to expand Hark’s engineering department from 70 to 200 researchers and engineers.
SI011 Grey Journal Hark Raises 700M Series A at 6B Valuation Adcock priced Hark at $6 billion before a single product had shipped.
SI012 Startup Researcher AI startup Hark raises over 700 million in Series A funding All four major U.S. chip makers backed Hark in the same round.
SI013 Yahoo Finance Hark Raises $700M Series A at a $6B Valuation Hark ... announced it has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SI014 Morningstar Hark Raises $700M Series A at a $6B Valuation Hark ... has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SI015 SEC SEC company search results for Hark Hark Labs Inc.
SI016 SEC Hark Labs Inc. company filings page Notice of Exempt Offering of Securities
SI017 SEC Hark Labs Form D filing detail Form D - Notice of Exempt Offering of Securities
SI018 SEC Hark Labs Form D primary XML Decline to Disclose
SI019 SEC Hark Labs complete Form D submission text ITEM INFORMATION: Rule 506(b) provides a "safer harbor" for a private offering
SI020 TechCrunch Humane’s AI Pin is dead, as HP buys startup assets for $116M Humane’s returns for the AI Pin started outpacing its sales.
SI021 HP HP accelerates AI software investments to transform the future of work The $116 million transaction is expected to close at the end of this month.
SI022 Thunder Compute NVIDIA B200 pricing in 2026 MSRP for the B200 was around $30,000-40,000 in clusters of 8+ GPUs.
SI023 Bloomberg AI hardware startup Hark valued at $6 billion in new funding round
SI024 Publicnow Mirror of Hark funding announcement
SI025 rabbit rabbit r1 homepage unlimited recordings, transcripts & AI summaries — all free, no subscription required
SE001 Hark Hark homepage Hark is entering beta
SE002 Hark Careers - Hark
SE003 Hark Privacy Policy - Hark When Hark executes tasks on your behalf, it operates within isolated sandbox environments (virtual machines).
SE004 Hark Hark announces $700 million fundraising round We're working on an AI platform that will be available this summer.
SE005 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence
SE006 Business Wire Hark Raises $700M Series A at a $6B Valuation
SE007 TechCrunch Hark raises $700M series A for its secretive universal AI interface
SE008 Observer Billionaire Brett Adcock Launches New Startup to Build Personalized AI
SE009 SiliconANGLE Hark raises $700M+ to build personalized intelligence devices
SE010 The Next Web Brett Adcock’s AI hardware startup Hark raises $700m at $6bn valuation
SE011 Intel Capital Hark raises $700M Series A at a $6B valuation
SE012 Mobile World Live iPhone Air designer joins AI start-up
SE013 Hark Hark news page
SE014 Greenhouse Jobs at Hark
SE015 Greenhouse Hark Greenhouse jobs API
SE016 Greenhouse Member of Technical Staff, Pretraining - Hark
SE017 Greenhouse Infrastructure, Large-scale Training - Hark
SE018 Greenhouse Operating System Architect - Hark
SE019 Greenhouse Privacy Engineer - Hark
SE020 Greenhouse Integration Engineer - Hark
SE021 Greenhouse AI Safety Engineer - Hark
SE022 Figure Introducing Helix
SE023 Figure Introducing Helix 02
SE024 Figure Production at BMW
SE025 TechCrunch Figure drops OpenAI in favor of in-house models
SE026 Humanoids Daily From Pixels to Torque: Figure Unveils Helix 02
SE027 Humanoids Daily The End of C++: Brett Adcock on Helix 02
SE028 Forbes Hark’s $6 billion valuation with no product actually makes sense
SU001 Hark Hark homepage
SU002 Hark Privacy Policy Hark collects information associated with your account, including your name, contact information, account credentials, payment card information, and transaction history.
SU003 Hark Careers
SU004 Hark Hark announces USD700 million fundraising round
SU005 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence
SU006 Business Wire Hark Raises $700M Series A at a $6B Valuation
SU007 TechCrunch Hark raises $700M Series A for its secretive 'universal' AI interface
SU008 Observer Billionaire Brett Adcock Launches New Startup to Build Personal A.I.
SU009 Forbes Hark’s $6 Billion Valuation With No Product Actually Makes Sense There’s no product to scale, no customers to name, no traction to underwrite.
SU010 Intel Capital Hark raises $700M Series A at a $6B valuation
SU011 Securities and Exchange Commission Hark Labs Inc. Form D primary document
SU012 Figure Production at BMW
SU013 BMW Group Humanoid robot in Leipzig
SU014 TechCrunch Figure AI CEO skips live demo, sidesteps BMW deal questions on stage at tech conference
SU015 Agility Robotics GXO Signs Industry-First Multi-Year Agreement with Agility Robotics
SU016 Retail Tech Innovation Hub GXO Logistics lays claim to an industry first as it inks multi-year agreement with Agility Robotics
SU017 PR Newswire Apptronik and Mercedes-Benz Enter Commercial Agreement That Will Pilot Apptronik’s Apollo Humanoid Robot in Mercedes-Benz Manufacturing Facilities
SU018 Symbotic Symbotic Completes Acquisition of Walmart’s Advanced Systems and Robotics Business and Signs Related Commercial Agreement
SU019 Symbotic Symbotic to Acquire Walmart’s Advanced Systems and Robotics Business and Sign Related Commercial Agreement
SU020 Locus Robotics One Billion Picks — And the Warehouse Robots Behind Them
SU021 Robotics & Automation News DHL and Locus Robotics reach one billion warehouse picks milestone
SU022 Automated Warehouse DHL Supply Chain completes 1B picks with Locus Robotics
SU023 About Amazon Amazon robotics in fulfillment centers
SU024 Gartner Gartner predicts fewer than 20 companies will scale humanoid robots for manufacturing and supply chain to production stage by 2028
SU025 IEEE Robotics and Automation Society Reality Is Ruining the Humanoid Robot Hype
SU026 Rodney Brooks My dated predictions
SU027 IIoT World Physical AI Deployment ROI: BMW’s 30,000-Car Proof
SU028 VaaSBlock Humanoid Robotics 2026: Figure, Optimus, 1X Commercial Reality
SR001 Hark Hark
SR002 Hark Privacy Policy | Hark
SR003 Hark Careers | Hark
SR004 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence
SR005 Business Wire Hark Raises $700M Series A at a $6B Valuation
SR006 Intel Capital Hark Raises $700M Series A at a $6B Valuation – Intel Capital
SR007 TechCrunch Hark raises $700M Series A for its secretive 'universal' AI interface | TechCrunch
SR008 Observer Flying Car Billionaire Brett Adcock Launches Startup to Build Personal A.I.
SR009 The Next Web Brett Adcock’s AI hardware startup Hark raises $700m at $6bn valuation
SR010 SiliconANGLE Hark raises $700M+ to build ‘personalized intelligence’ devices - SiliconANGLE
SR011 Brett Adcock Official Bio | Brett Adcock Official
SR012 Figure Figure
SR013 Figure F.02 Contributed to the Production of 30,000 Cars at BMW
SR014 BMW Group BMW Group: First humanoid robot introduced in Plant Leipzig
SR015 TechCrunch Figure AI CEO skips live demo, sidesteps BMW deal questions onstage at tech conference | TechCrunch
SR016 MIT Technology Review Why the humanoid workforce is running late
SR017 Berkeley News Are we truly on the verge of the humanoid robot revolution? - Berkeley News
SR018 U.S. News & World Report / AP Humanoid Robots Take Center Stage at Silicon Valley Summit, but Skepticism Remains
SR019 McKinsey & Company Turning humanoid supply chain constraints into billion-dollar wins
SR020 Bain & Company Humanoid Robots: From Demos to Deployment
SR021 TechCrunch The world is just not quite ready for humanoids yet | TechCrunch
SR022 Federal Trade Commission Artificial Intelligence Compliance Plan
SR023 California Privacy Protection Agency California Finalizes Regulations to Strengthen Consumers' Privacy
SR024 CISA Artificial Intelligence | CISA
SR025 Bureau of Industry and Security BIS semiconductors and advanced computing guidance
SR026 CHIPS for America CHIPS for America
SR027 Securities and Exchange Commission EDGAR Search Results
SR028 Securities and Exchange Commission Hark Labs Inc. Form D filing text
SR029 TechCrunch Humane's AI Pin is dead, as HP buys startup's assets for $116M | TechCrunch
SR030 HP HP Accelerates AI Software Investments to Transform the Future of Work
SV001 Hark Hark At Hark, we are building the most advanced personal intelligence in the world.
SV002 Business Wire Hark Launches AI Lab Building Futuristic Interface to Artificial Intelligence The company has recently signed a deal to scale its compute capacity, with a large cluster of thousands of NVIDIA B200 GPUs coming online in April.
SV003 Business Wire Hark Raises $700M Series A at a $6B Valuation Hark ... has raised over $700 million in Series A funding at a $6 billion post-money valuation.
SV004 TechCrunch Hark raises $700M Series A for its secretive 'universal' AI interface The company currently has 70 employees and runs a data center with Nvidia B200 GPUs.
SV005 Intel Capital Hark Raises $700M Series A at a $6B Valuation – Intel Capital The round was oversubscribed and led by Parkway Venture Capital, with participation from NVIDIA ... Intel Capital ...
SV006 SEC Hark Labs Form D primary XML Decline to Disclose
SV007 SEC Hark Labs complete Form D submission text 1000000000 50000000 950000000
SV008 Hark Careers | Hark
SV009 Hark Privacy Policy | Hark Hark collects information associated with your account, including ... payment card information, and transaction history.
SV010 Reuters / Yahoo Finance UPDATE 1-Robotics startup Figure raises $675 mln from Microsoft, Nvidia, OpenAI Figure said on Thursday it raised $675 million ... at a valuation of $2.6 billion.
SV011 Figure F.02 Contributed to the Production of 30,000 Cars at BMW Contributed to the production of 30,000+ X3 vehicles
SV012 Figure Helix | Figure Helix is a generalist humanoid Vision-Language-Action model.
SV013 Figure Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation We have exceeded more than $1 billion in committed capital through our Series C financing round, at a post-money valuation of $39 billion.
SV014 TechCrunch Figure reaches $39B valuation in latest funding round Earlier this year, Figure CEO Brett Adcock claimed that Figure was the most “sought-after” stock on the private market.
SV015 BMW Group First humanoid robot introduced in Plant Leipzig
SV016 1X 1X Secures $100M in Series B Funding 1X ... has raised $100 million in Series B funding.
SV017 EQT Ventures Norwegian Robotics Startup 1X Secures $100M in Series B Funding Led by EQT Ventures Add in 1X’s $25 million Series A funding round last March, and the company has now raised around $125 million in under a year.
SV018 EqualOcean Backed by OpenAI, 1X Technologies Aims for Up to USD 1 Billion Funding, Valuation May Top USD 10 Billion 1X Technologies ... is seeking up to USD 1 billion in new funding, targeting a valuation of at least USD 10 billion.
SV019 Humanoids Daily Report: Humanoid Robotics Firm 1X Seeking Up to $1B at a Valuation of $10B or More A $10 billion valuation would be a significant leap for 1X, which last raised $100 million in a Series B round in January 2024.
SV020 Agility Robotics / GXO GXO Signs Industry-First Multi-Year Agreement with Agility Robotics GXO ... and Agility Robotics ... announced a multi-year agreement.
SV021 ONAMI Agility Robotics Raises $400M on $2.15B valuation Agility Robotics Raises $400M on $2.15B valuation
SV022 TSG Invest Agility Robotics Stock: $2.1B Valuation — Is It a Buy? Agility Robotics has raised approximately $640 million in total funding.
SV023 Public Agility Robotics Valuation, Share Price Estimates & Funding History Agility Robotics's estimated secondary market share price is $60.74 as of May 2026.
SV024 Agility Robotics Agility Robotics Powers the Future of Robotics with NVIDIA We received an investment from NVentures (NVIDIA’s venture capital arm) in our Series C round.
SV025 Apptronik Apptronik Closes Over $935 Million Series A with New $520 Million Extension Round The Series A-X extension round follows a $415 million oversubscribed initial Series A raise in 2025, bringing Apptronik’s total Series A to more than $935 million.
SV026 TechCrunch Humanoid robot startup Apptronik has now raised $935M at a $5B+ valuation TechCrunch separately learned that its post-money valuation is now about $5.3 billion.
SV027 CNBC Apptronik raises $520 million to beat Chinese humanoids, Tesla Optimus to market Apptronik raises $520 million ... at a $5 billion valuation
SV028 Apptronik Apptronik and Mercedes-Benz Enter Commercial Agreement That Will Pilot Apptronik’s Apollo Humanoid Robot in Mercedes-Benz Manufacturing Facilities The partnership represents Apptronik’s first publicly announced commercial deployment of Apollo.
SV029 Boston Dynamics Hyundai Motor Group Acquires Boston Dynamics Hyundai Motor Group acquires a controlling interest in Boston Dynamics from SoftBank.
SV030 Hyundai Motor Group Hyundai Motor Group Completes Acquisition of Boston Dynamics from SoftBank The transaction values Boston Dynamics at $1.1 billion.
SV031 Gartner Gartner Predicts Fewer Than 20 Companies Will Scale Humanoid Robots for Manufacturing and Supply Chain to Production Stage by 2028 Through 2028, less than 100 companies will progress humanoid robot proofs of concept beyond experimentation, with fewer than 20 companies going live in production.
SV032 Goldman Sachs The global market for humanoid robots could reach $38 billion by 2035 The total addressable market for humanoid robots is projected to reach $38 billion by 2035.
SV033 IEEE Spectrum Humanoid Robots: The Scaling Challenge The bigger problem is demand—I don’t think anyone has found an application for humanoids that would require several thousand robots per facility.
SV034 IEEE Robotics and Automation Society Reality Is Ruining the Humanoid Robot Hype Humanoid robotics companies have been consistently promising ... enabling them to raise hundreds of millions of dollars at valuations that run into the billions.
SV035 Rodney Brooks My Dated Predictions People tend to underestimate how long new technologies will take to be adopted.
SV036 TechCrunch Humane's AI Pin is dead, as HP buys startup's assets for $116M The Bay Area startup ... raised more than $230 million to create the device.
SV037 HP HP Accelerates AI Software Investments to Transform the Future of Work The $116 million transaction is expected to close at the end of this month.
SV038 McKinsey & Company Getting warehouse automation right A warehouse automation revolution is underway, but too many projects are not delivering the results.
SV039 Amazon Amazon uses robots that sort, lift, and carry packages—see them in action Amazon has deployed more than 1 million robots across its operations network since 2012.