初创公司尽调
尽调报告 Robotics / embodied AI / world models / physical AGI Private, 2026 Series B2-stage unicorn 2026-06-24

GigaAI

关于 GigaAI(极佳视界)的尽调简报:一家位于北京的物理 AGI 与机器人独角兽

GigaAI 是一家技术野心很强的北京 physical-AGI 公司,2026 年融资通道强,也有可信的早期部署信号;但公开证据还不足以像评估披露更充分的机器人同行那样,有把握地支撑其独角兽估值。

封面要素

报道的投后估值 05
1500 USD millions (derived from >CNY10B public round coverage) [CO029, CO030, CV004, CV005]
旗舰产品 06
GigaBrain / GigaWorld, Maker H01, SeeLight S1 [CO035, CO037, CO038]
具名部署信号 07
FAW Tooling manufacturing workflow, Hubei innovation-center shipment, Longsheng 1,000-robot plan, ~100 SeeLight S1 orders [CO033, CO034, CO039, CO040]
运营网站 08
gigaai.cc gigaai.com is parked [CO001, CO002]
员工人数 10
[CO045]

公司概况

GigaAI 是一家未上市的北京具身 AI 公司,由黄冠于 2023 年创立。公司把自己定位为全栈物理 AGI 平台,组合了世界生成与世界行动模型(GigaWorld、GigaBrain、DriveDreamer)、数据采集硬件以及自研机器人,面向工业和家庭场景。公开报道显示融资节奏罕见:2026 年 3 月 Pre-B 轮约 CNY1B,4 月 B1 轮近 CNY1.5B,6 月 B2 轮 CNY1B,估值推升至 CNY10B 以上,也让公司成为中国最新一批具身 AI 独角兽之一。公开资料在技术和融资里程碑上远多于经营经济性,因此仍应把公司视为高潜力但高度不透明的私营发行人。

官网
www.gigaai.cc
成立时间
2023-06-01
创始人
Huang Guan, Zheng Zhu
创立地点
Beijing, China
总部
Beijing, China
产品
公司开发一套物理 AI 基础设施和产品:GigaWorld 与 DriveDreamer 世界模型系统、GigaBrain 具身基础模型、数据采集硬件、Maker H01 工业机器人,以及 SeeLight 家用机器人系列。
客户
目标客户包括汽车和工业制造商、仓储 / 物流运营商、可能授权模型 / API 能力的机器人或平台合作伙伴,以及 SeeLight 系列面向的家庭服务或养老渠道;已披露证据在中国最强。
商业模式
工业和家用机器人的硬件与部署项目以报价制为主,并叠加可接入合作伙伴部署的模型 / API、授权、仿真或数据系统服务。
阶段
Private, late-stage Series B / unicorn
融资情况
2026 年融资有充分公开来源支撑:3 月 Pre-B 轮约 CNY1B、4 月 B1 轮近 CNY1.5B、6 月 B2 轮 CNY1B,估值超过 CNY10B;具体股权结构变化和优先权条款仍未公开。
[CO003, CO004, CO005, CO006, CO009, CO017, CO027, CO028]

执行摘要

主要优势

  • 公司讲出的不是单一演示产品,而是一条少见的全栈 physical-AGI 路线,覆盖世界模型、具身基础模型、数据系统和自研机器人。
  • 作为一家 2023 年成立的创业公司,资本通道少见地强:公开报道支持其 2026 年三个月内融资 RMB 3.5 billion,估值超过 RMB 10 billion。
  • 工业和家庭场景已有具名部署证据,包括 FAW Tooling、湖北人形机器人创新中心、Longsheng 1,000 台机器人计划,以及约 100 台 SeeLight S1 订单。
  • 创始人与市场匹配度清楚:Huang Guan 的背景把计算机视觉、自动驾驶、具身 AI 和商业化串在一起。

主要风险

  • 公开披露远远落后于估值:已审阅来源没有给出收入、毛利率、ARR、员工数、客户集中度或融资条款透明度。
  • 商业化证据仍弱于融资叙事,因为多数公开证据来自订单、试点和已宣布部署,而不是经审计收入或可重复的生产采用。
  • physical-AI 训练和部署仍依赖稀缺先进芯片与全球基础设施,公司因此暴露在算力、出口管制和地缘政治风险下。
  • 在机器人达到工业级在线率要求前,仿真和世界模型主张还要跨过真实的 sim-to-real、可靠性和安全门槛。
  • 治理和关键人集中度是实质风险:公开叙事由 Huang Guan 和少数可见技术骨干主导,董事会构成仍未披露。

未决问题

  • 已审阅公开材料仍未披露截至 2026 年的收入、毛利率、积压订单、烧钱速度和现金跑道数据。
  • 精确董事会构成、投票控制权、清算优先权,以及 2026 年融资中是否含老股转让,都没有公开。
  • 客户集中度、试点 / 订单向付费量产的转化,以及保修 / 支持经济性仍不清楚。
  • 当前员工数,以及可见创始人科学家核心之外的组织深度,尚未得到公开验证。
  • 关于规模化现场条件下可靠性、安全性和 sim-to-real 表现的独立基准证据仍很少。

目录

Chapter 01

01公司概览

1.1 身份、域名控制与商业模式

最扎实的基础身份画像,在公司名称、所在地和产品叙事上内部一致,但英文域名并不一致。活跃网站 gigaai.cc 指向一个标题为「极佳科技」的极简页面,而 gigaai.com 是挂牌待售的停放域名,并非公司官网。这不削弱运营主体本身,但对自然会先查 .com 的交易对手来说,是一个真实的尽调提示。独立财经和行业报道一致把业务识别为北京极佳视界科技有限公司 / GigaAI,2023 年成立,总部在北京。公司的经营假设不止一个机器人 SKU:公司反复描述的是一套把世界模型、具身基础模型、自有机器人本体和泛化场景合在一起的栈。落到商业化,就是两条仍在公开视野中验证的变现路径:通过软件、API 或授权向合作伙伴交付软件或基础模型服务,以及围绕 Maker 工业线和 SeeLight 家庭线的硬件加部署项目。公开来源支撑战略叙事,但还没有披露稳定收入、ARR 或客户数量等指标,投资人无法据此把叙事动能和经常性商业规模拆开。[CO001, CO002, CO003, CO004, CO011, CO012]

GigaAI 快照 KPI 表
指标数值 / 状态日期信心缺口 / 注意事项
运营网站gigaai.cc 可访问;页面标题为「极佳科技」2026-06-24网站在线但内容单薄;没有更丰富的 JS 渲染时,主要内容无法直接看到
英文品牌 .com 域名gigaai.com 已停放并挂牌出售2026-06-24给非中国交易对手带来命名 / 信任模糊性
法律 / 运营主体北京极佳视界科技有限公司 / GigaAI2026本轮未从可检索的官方政府结果中独立拉取法律登记字段
总部中国北京2026公开报道支持北京;确切园区 / 实体地址未通过实时登记摘录重新核验
成立时间20232023有些媒体具体提到 2023 年 6 月,但年份级支持比精确月份更干净
阶段2026 年 B2 后期私有具身 AI 独角兽2026-06阶段由融资节奏和估值推断,不是公司正式标签
支撑最充分的估值超过 RMB 10B / 约 USD 1.5B2026-04 to 2026-06USD 数字基于媒体 / 换算,不是备案口径
支撑最充分的近期融资额三个月内 RMB 3.5B(2026 年 3–6 月)2026-06-15这不等于精确的累计融资额
最新披露轮次B2,RMB 1B2026-06轮次结构、董事会条款及任何老股成分未披露
产品家族产品 / 模型:GigaWorld、GigaBrain、Maker H01、SeeLight S12025-2026公开来源强调明星产品;完整 SKU 图谱仍更宽
部署信号Maker 已开始交付;100 台 SeeLight S1 订单;1,000 台机器人龙盛计划2026多数规模说法来自公司经媒体披露,而非经审计客户披露
收入 / ARR / 员工数2026-06-24已审阅公开来源没有披露收入、ARR 或当前员工数

本表只保留公开且有来源支撑的指标。Null 表示已审阅公开材料不可得,不是零。

[CO001, CO002, CO003, CO017, CO018, CO027]
FO002: GigaAI 公司快照逻辑

身份、产品、资本和部署闭环,都压在一个小型创始团队和「世界模型优先」的商业化命题上。

[CO001, CO002, CO011, CO012, CO013, CO016]

1.2 创始人、领导层与治理可见度

创始人集中度是公开资料里最清楚的特征之一。几乎每篇融资和人物报道都会出现黄冠,身份是创始人兼 CEO,履历横跨清华自动化、Horizon Robotics、PhiGent / 鉴智机器人、Microsoft Research Asia 和 Samsung China Research。对一家做世界模型和机器人的公司来说,这让创始人与市场的匹配度格外可读:他的履历把计算机视觉、自动驾驶、具身 AI 和工业化连在一起。公开可见的第二梯队领导层更窄,但仍有意义。投资人和人物报道提到郑珠为联合创始人兼首席科学家,孙少岩为联合创始人且有 Alibaba Cloud 和 Horizon 经历,毛继明为有 Baidu Apollo 背景的工程负责人。郑珠个人主页独立确认其 GigaAI 头衔,这让领导层画像不只停留在融资宣传文案。仍然不透明的是治理。已审阅来源披露了大型机构投资人联盟,却没有公布董事会构成、投票控制、保护性条款,也没有说明近期融资是否重塑了创始人控制权。因此,尽调里的关键人物风险是真实存在的:公开叙事、融资通道和技术路线图可信度,都显著集中在黄冠和一个小型技术班底身上。[CO005, CO006, CO007, CO008, CO009, CO010]

领导层和创始人表
人员职位背景创始人市场匹配 / 覆盖关键人依赖
Huang Guan创始人兼 CEO清华自动化博士;曾任 Horizon 视觉感知负责人;此前在 PhiGent、Microsoft Research Asia、Samsung China 任职结合世界模型研究、自动驾驶语境和商业化叙事
Zheng Zhu联合创始人兼首席科学家先后为 CASIA 博士和清华博士后;个人网站称其为 GigaAI 联合创始人兼首席科学家增强 CV / 机器人领域的研究可信度和 benchmark 领导力
Sun Shaoyan联合创始人 / 产品运营负责人投资媒体将其与 Alibaba Cloud 和 Horizon 数据闭环角色联系起来打通产品、云和工业部署职能
Mao Jiming工程负责人 / VP投资媒体将其与 Baidu Apollo 仿真和工业工程角色联系起来把研究产出连接到可部署系统和制造工作流

行覆盖公开具名的高级创始团队,而不是完整组织架构或董事会名单。

[CO005, CO006, CO007, CO008, CO009, CO046]

1.3 融资历史、估值与资本形成

GigaAI 的融资故事,是它在后续章节值得关注的核心原因。公开报道显示,公司从种子轮进入 2024 年天使融资,2025 年完成 Pre-A 和 Pre-A+ 轮,A1 轮由 Huawei Hubble 和 Huakong Fund 支持,并披露了 RMB 200 million 的 A2 轮。2026 年节奏明显加速。已审阅来源在 3 月近 RMB 1 billion 的 Pre-B 轮、4 月近 RMB 1.5 billion 的 B1 轮、6 月 RMB 1 billion 的 B2 轮上基本一致。随后第一财经将三个月融资概括为 RMB 3.5 billion,约 USD 518 million,多家中文媒体称估值已超过 RMB 10 billion;Crunchbase 也将 GigaAI 记录为 2026 年 4 月 Series B 后估值约 USD 1.5 billion 的独角兽。因此,资本底座无疑很大,但仍未完全透明。有些报道列出了每轮的长投资人名单,但一名 B1 参与方仍只被描述为「知名科技巨头」,公开来源也没有披露准确董事会权利、持股结果,或是否存在重大老股转让。实际结论是,估值和融资速度有充分支撑;但股权表机制仍是待追问事项,而非已解决事实。[CO019, CO020, CO021, CO022, CO023, CO024]

利益相关方或投资人图谱
利益相关方角色控制 / 经济重要性尽调问题
Huawei HubbleA1 投资人说明在 2026 年加速前,一家中国大型科技投资方已支持公司确认当前持股,以及任何董事会或信息权
Huakong FundA1 / A2 复投方复投资本提供方,可能对相邻轮次有影响力厘清 pro-rata 权利和治理位置
CICC Capital 和国资平台财团Pre-B / B 轮复投资本基础暗示深厚机构支持,以及可能的政策 / 地方产业协同梳理确切所有权、董事席位和任何地方产业条件
Lion Partners Capital 和 China-Belgium Direct Equity Investment Fund具名 B2 投资人锚定最新一轮的国际品牌化和国资关联调性核验分配规模,以及跟投权是否重要
Wanxiang Qianchao 和汽车产业资本产业投资人 / 生态参与方可能塑造汽车制造部署机会区分纯财务持股与商业采购影响
未具名“知名科技巨头”出现在 B1 报道中未披露 B1 参与方可能对估值信号和商业分发很重要在承销战略价值前,取得具名确认和条款
Longsheng Technology 与 FAW Tooling / Alibaba Cloud商业交易对手,而非股权支持方即便不是股东,也对真实部署证明很重要核验合同规模、部署里程碑和续约经济性

这张图谱混合了主要已披露融资方和少数商业上重要的利益相关方,因为公开 cap table 透明度不完整。

[CO020, CO022, CO023, CO024, CO025, CO026]
FO003: GigaAI 快照 KPI

公开可支撑的 KPI 更强调资本、估值和部署信号;基础经营指标仍然缺位。

RMB 和 USD 数值混合了原始报道与媒体换算。数量指标是公司报告的部署信号,不是经审计的财务 KPI。

[CO017, CO018, CO027, CO028, CO029, CO035]

1.4 里程碑、公开指标与风险信号

公司的里程碑曲线,在技术和部署叙事上强于经审计经营数据。研究代码库和论文显示,公司从自动驾驶世界模型 DriveDreamer,推进到作为具身数据引擎的 GigaWorld-0,再进入 GigaBrain 和实体机器人产品。到 2026 年初,GigaAI 已宣称 Maker H01 大规模交付、首批发往湖北人形机器人创新中心、与 FAW Tooling 和 Alibaba Cloud 的制造流程,以及与 Longsheng Technology 三年部署 1,000 台 Maker 系列机器人的计划。家庭端野心也更具体:多方来源称 SeeLight S1 约有 100 个订单,并计划在 2026 年第三季度扩大交付或运营。问题在于,公开经营指标仍落后于故事。已审阅材料没有提供经验证的收入、ARR 或员工人数,客户证据也更多来自部署、订单和模型基准,而不是现金生成。本轮检索中最尖锐的外部质疑来自 Tencent News 评论,它提醒世界模型热潮中估值膨胀、真实世界泛化和商业化仍是开放问题。已审阅来源中没有出现具体诉讼或处罚,但没有公开负面案卷并不等于法律尽调已经完成。[CO017, CO018, CO033, CO034, CO035, CO037]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2023-09DriveDreamer 世界模型项目发布产品研究发布GigaAI 研究团队显示公司先通过自动驾驶世界模型建立可信度,再进入具身机器人
2024-09Angel 和 Angel+ 融资披露融资近 RMB 50MBAIC Investment、MiraclePlus、People’s Capital 等在更大战略轮之前建立早期外部支持
2025-08Pre-A 和 Pre-A+ 轮披露融资数亿 RMBGuozhong Capital、Zifeng、CICC Capital、Guangzhou Investment 等标志公司从早期研究初创公司转向有机构支持的规模化阶段
2025-11A1 轮披露融资RMB 100M 级轮次Huawei Hubble、Huakong Fund增加战略投资人验证和生态触达
2025-12A2 轮披露融资RMB 200MFortune Capital、Huakong 和跟投方在 2026 年融资激增前完成 2025 年资本爬坡
2025-11Maker H01 发布产品原生工业机器人首秀GigaAI为具身模型商业化建立硬件锚点
2026-03Pre-B 轮完成融资近 RMB 1B产业、国资和财务投资人确认公司已进入超大轮次区间
2026-04B1 轮浮出,估值跨过 RMB 10B融资近 RMB 1.5B;估值 > RMB 10B未具名科技巨头,以及基金和产业投资人推动 GigaAI 进入独角兽区间
2026-04FAW Tooling / Alibaba Cloud 制造工作流公布合作工厂部署参与方:GigaAI、FAW Tooling、Alibaba Cloud提供最好的公开工业用例证明之一
2026-06B2 轮完成融资RMB 1BLion Partners、China-Belgium Fund、Wanxiang Qianchao 等把 3–4 月融资爆发延伸为三轮 RMB 3.5B 浪潮
2026-06Longsheng 计划公布规模3 年内 1,000 台 Maker 系列机器人GigaAI、Longsheng Technology形成一个规模大但仍面向未来的工业部署承诺
2026-04Tencent 评论质疑世界模型估值耐久性反向行业怀疑,不是指控Tencent News / iHeima 评论给原本偏宣传的融资叙事加入公开制衡

这条年表是本章唯一的公开记录时间线。已审阅来源未可靠披露具体日期时,日期采用月份级。

[CO020, CO021, CO022, CO023, CO024, CO025]
FO001: GigaAI 公司里程碑时间线

公开叙事从世界模型研究出发,延伸到具身机器人、大额融资,以及最早可见的家庭和工业部署。

[CO020, CO024, CO025, CO026, CO027, CO028]

1.5 图表

Chapter 02

02市场分析

2.1 市场边界:GigaAI 实际卖向哪里

对 GigaAI 而言,相关市场边界不是泛 AI 软件,也不是所有机器人硬件。可用边界从已经为感知、操作或流程适配付费的物理自动化环境开始,再收窄到帮助机器人在混乱真实任务中感知、规划和泛化的软件与系统层。公开证据支持边界内四个桶:制造业工业机器人升级、物流和仓储机器人、零售或药房等特定服务流程,以及卖给机器人 OEM 和集成商的世界模型或仿真基础设施。边界外的东西同样重要。通用办公室 copilot、纯聊天界面、没有学习闭环的固定功能工厂机械、缺乏预算验证的消费级家庭机器人愿景,都不应计入 GigaAI 近端需求。多份反向来源强调,人形本体本身不是市场;很多工作仍更适合轮式、固定底座或任务专用机器人。这一点很重要,因为 GigaAI 的产品逻辑看起来更接近基础设施和部署层,而不是大众家庭机器人供应商。实际市场由既有自动化支出、集成痛点和运营方为更好泛化付费的意愿决定,而不是银行报告里最大的标题 TAM。[CM001, CM002, CM012, CM044, CM045, CM051]

市场定义表
细分 / 品类纳入支出排除支出买方 / 付款方与 GigaAI 的相关性
工业机器人升级附着在制造机器人和单元上的感知、操作、规划和软件层没有自适应软件或学习闭环的纯固定功能机械工厂自动化、制造工程、capex 负责人近期相关性最高,因为预算已经存在,ROI 会对照劳动力、吞吐量和换线成本衡量
物流和仓储机器人面向配送中心的分拣、拣选、AMR 编排、仿真和感知栈不触及物理工作流的通用仓库 IT3PL 运营商、包裹承运商、零售配送运营相关性高,因为仓库已经采集数据、跑试点,并重视吞吐提升
零售 / 药房 / 护理服务机器人用于店内、药房或辅助任务中重复工作、且工作流明确的机器人仅靠新奇感销售的消费者陪伴机器人门店运营、医疗健康运营商、辅助生活预算选择性相关;预算更窄,但部分中国试点显示真实部署
世界模型 / 仿真 / 工具基础设施出售给机器人 OEM 或集成商的仿真、合成数据、ROS / AI 中间件、模型部署、验证和再训练工具未用于机器人部署的通用企业 AI copilots 或开发者工具机器人 OEM 产品团队、集成商、平台工程具有战略重要性,因为即便人形机器人尚未规模化,它也能跨机队和不同形态扩展
大众消费者家庭人形机器人近期市场只应纳入有限研究或试点需求假设所有家庭劳动今天都可服务化为支出消费者、保险方、养老生态近期相关性低,因为用户偏好、安全、隐私和支付意愿仍缺乏充分验证

边界逻辑把可部署机器人预算与通用 AI 或投机性家庭 TAM 分开,避免本章夸大 GigaAI 可触达市场。

[CM001, CM002, CM012, CM044, CM045, CM049]

2.2 规模测算视角:真实自动化底座 vs 推测性人形上行空间

公开市场规模口径分歧很大,因为它们衡量的不是同一件事。IFR 数据显示,自动化底座已经庞大且具体:2023 年全球机器人密度创纪录,中国已掌控大部分新增工业装机,中国制造业装机基数以百万台计。这是 GigaAI 这类能力可以卖入的基础层。相比之下,TBRC 等宽口径具身 AI 报告把 2025-2026 年市场描述为几十亿美元级,MarketsandMarkets 等更窄的人形机器人报告则从更小基数出发预测更快增长。银行研究把时间线拉得更长:Goldman 的保守情景在未来十年仍只有个位数十亿美元级,而 Morgan Stanley 和 BofA 则在品类走向大规模采用的前提下,建模出巨大的长期出货或收入结果。这些数字不该被平均成一个整洁 TAM。它们更像一架梯子:当前的自动化装机基数,更近的具身 AI 软件和系统层,再往上才是更具推测性的通用人形机会。尽调时,对 GigaAI 最重要的市场数字,是机器人 OEM、集成商和运营方已经愿意付费以降低设置时间、改善感知或提高任务泛化的可服务切片。公开来源尚未清晰隔离这一层,因此 SAM 和 SOM 仍受证据约束,而不是模型可精确测算。[CM003, CM004, CM006, CM011, CM013, CM016]

TAM / SAM / SOM 或规模测算视角表
发布方年份 / 时间范围地理范围数值CAGR / 增长方法 / 边界信心局限
The Business Research Company2025 / 2026 / 2030全球具身 AI 市场:$3.22B(2025)、$3.8B(2026)、$7.24B(2030)到 2026 年 18.1%;到 2030 年 17.5%广义具身 AI,包括机器人、外骨骼、自主系统和智能家电有用的宽品类,但太宽,无法隔离 GigaAI 的软件楔子或人形机器人特定支出
MarketsandMarkets2025-2030美国人形机器人市场:$857.9M(2025)到 $4.601B(2030)39.9% CAGR仅人形机器人、单一国家预测比具身 AI 更窄且有地理限定;不是直接的全球 TAM
MarketsandMarkets2025-2030韩国人形机器人市场:$112.6M(2025)到 $583.5M(2030)39.0% CAGR仅人形机器人、单一国家预测展示亚洲增长,但仍缺少中国公开价值细节
Goldman Sachs(CNBC 引述)10–15 年基准;2035 蓝天情景全球10–15 年基准情景约 $6B;2035 年蓝天情景最高 $154B未以 CAGR 表述围绕制造和养老劳动力替代来界定人形机器人市场情景跨度极大,并取决于成本、接受度和用例突破
Morgan Stanley(CNBC 引述)2030 / 2040 / 2050全球人形机器人保有量:40k(2030)、8M(2040)、63M(2050)未以 CAGR 表述与劳动力短缺和具身 AI 进展挂钩的装机基数预测保有量预测不是收入,且远高于当前部署
Morgan Stanley(CNBC 引述)2050全球人形机器人收入:$4.7T;约 1B units长期情景一体化人形机器人价值链的长期收入视角预测周期极长,且采用了激进的渗透假设
Bank of America Institute(研究机构)2025 / 2026 / 2030 / 2035 / 2060全球出货量:20k(2025)、90k(2026)、1.2M(2030)、10M(2035);到 2060 年装机基数 3B到 2035 年出货量 CAGR 为 86%基于 AI 成熟度、硬件成本下降和人口结构的人形机器人出货量与装机基数模型可用于判断曲线形态,但仍是模型,不是经审计的需求

这些行有意混合广义具身 AI、人形机器人、出货量和收入等视角,保留彼此矛盾的估算,而不是把它们压成一个看似精确的 TAM。

[CM011, CM013, CM014, CM015, CM016, CM017]
FM001: 市场规模测算视角

可用市场从庞大的自动化存量底座开始,收窄到个位数十亿美元的具身 AI 层,再到更窄的人形机器人预算,最后才是更具投机性的长期牛市情景。

数值有来源支撑,但刻意混合单位和时间维度,用来展示相关性如何收窄,而不是暗示可以做算术相加。

[CM006, CM011, CM013, CM016, CM018, CM044]
FM002: 市场估算区间

具身 AI 和人形机器人市场价值的公开估算从个位数十亿美元拉到万亿美元级,根源在于边界和时间跨度完全不同。

所有数值均为十亿美元。各项比较的是引用来源中的目标年份市场价值,而非同一个统一模型。

[CM011, CM013, CM016, CM018, CM048]

2.3 买方、用户与付款方分层

最清楚的买方不是家庭,而是有可衡量人工、正常运行时间和换线问题的企业与机器人公司。在制造业,用户可能是产线、维护团队或工厂工程师,但付款方通常是工厂自动化、资本开支或运营预算。物流场景里,用户是分拣、拣选和仓库运营人员,付款方则是 3PL、包裹或零售运营负责人,他们会把机器人支出同人员流失、吞吐量和工伤成本对比。零售或药房等服务场景中,用户体验可能面向消费者,但采用仍取决于运营方能否用人工节约或服务一致性为机器人编预算。还有一个分层对 GigaAI 更重要:机器人 OEM 和集成商,它们把仿真、感知或世界模型基础设施作为许多部署的赋能层来购买。这些买方掌握数据流水线、模型再训练闭环和部署工具,因此比购买单台机器人的终端用户更像自然的基础设施客户。中国专项报道显示,物流和便利零售属于最早的商业化路径之一,因为这些场景流程密集、任务高频,且数据尾气好。家用人形机器人仍远没那么可承销,因为付费意愿、安全容忍度和真实用户偏好都更不确定。[CM035, CM042, CM045, CM046, CM049]

细分市场 / 买方地图
细分市场买方用户付款方 / 预算负责人工作流采用触发点
汽车 / 电子制造工厂自动化负责人、产线集成商操作员、维护人员、工艺工程师工厂资本开支与运营预算搬运、装配、上下料、检测用工缺口、换线时间缩短、稼动率和精度提升
3PL / 包裹 / 仓储运营副总裁、仓库工程团队、机器人负责人拣货员、分拣员工、主管运营与网络生产率预算拣选、分拣、托盘移动、动态编排吞吐量、工伤减少、劳动力流失、季节性高峰
零售 / 药房服务门店运营总监、创新团队门店员工、药剂师门店用工与服务质量预算商品取放、补货、重复性服务任务持续存在的重复劳动与服务一致性
医疗 / 养老护理护理机构运营方、医院运营、设备采购护士、护理员、患者、住户护理预算、资本采购,有时还包括保险方或公共项目辅助搬运、物流、监测劳动力老龄化、护理人员短缺、符合责任要求的辅助工作流
机器人 OEM / 集成商产品总经理、自治系统负责人、平台工程团队机器人开发者、部署工程师研发、平台和产品预算仿真、数据生成、模型验证、部署工具需要缩短集成周期,并在多个客户部署中复用能力

这张地图区分终端用户和预算负责人,因为 GigaAI 可能向 OEM 或集成商销售基础设施,即便机器人的终端用户是工厂或仓库运营方。

[CM035, CM042, CM045, CM046, CM049]
FM003: 买方 / 细分市场地图

这个矩阵看的是变现准备度,而不只是买方身份:机器人平台买家和物流运营商得分高于家庭或护理场景,因为它们同时具备清晰预算、可复用数据闭环和既有自动化基础设施。

[CM035, CM044, CM045, CM046, CM049, CM051]

2.4 增长驱动、采用约束与中国特有节奏

最强需求驱动来自劳动力稀缺、政策支持和集成摩擦下降。IFR、WEF 和 BofA 都指出,结构性劳动力短缺、劳动力老龄化和制造业回流压力,是企业持续自动化的原因。中国还多了一种加速器:国家政策、深厚零部件基础、高机器人装机量和本土供应商份额提升。技术侧,近端最可信的价值不是神奇自治,而是降低设置成本、改善感知、让操作更灵活,并通过仿真和数据闭环加快迭代。BCG 的框架在这里有用:Level 2 和 Level 3 现在就创造价值,Level 5 推理和宽口径世界模型仍偏愿景。这一区分对 GigaAI 的市场时机很关键。最佳近端采用场景,是既有车队、仓库或工厂先吸收更好软件,再购买数千台通用人形机器人。约束同样实质。反向来源反复点出灵巧度、仿真到现实的数据鸿沟、可靠性、电池寿命、安全标准,以及大型场站人形需求证据不足。多方还认为,平面工厂往往更适合轮式或任务专用系统,而不是腿式人形。因此,采用路径大概率分阶段:先工业和物流部署,再更广泛服务场景,最后才是家用人形。对 GigaAI 来说,今天可投资的市场是已在自动化环境里的升级周期,尤其在中国和亚太,而不是一个普适机器人劳动力替代故事。[CM007, CM008, CM022, CM023, CM024, CM029]

增长驱动因素与约束表
驱动因素 / 约束方向时点含义尽调问题
劳动力短缺与劳动力老龄化驱动当前至 2035 年以后支撑制造、物流和护理自动化的 ROI量化 GigaAI 目标行业中哪些岗位存在可衡量的空缺或加班痛点
中国「十五五」规划与战略性新兴产业支持驱动当前至 2030 年在中国为机器人和具身 AI 带来政策背书和采购拉动要求提供与 GigaAI 有关的项目、补贴或试点区的直接证据
中国既有自动化基础驱动当前庞大的机器人基数和本土供应商深度,降低升级采用摩擦梳理装机基数中哪些部分真正能消化 GigaAI 产品
仿真、世界模型和 VLA 进展驱动当前至中期在完全自治到来前,改善感知、规划和再训练经济性要求提供部署证据,证明这些工具能减少集成时间或错误率
机器人部署设置与再工程成本成本下降时构成驱动当前如果软件削减设置成本,更多变化型工作流就能自动化验证 GigaAI 是否减少工程工时、夹具成本或换线停机时间
灵巧操作与 sim-to-real 瓶颈约束当前至长期限制通用操作能力,也拖慢向狭窄工作流之外扩张要求提供真实环境中的任务级基准,而不只是仿真或演示
可靠性、安全与标准负担约束当前工业买方要求极高稼动率和清晰的安全失效行为从生产试点获取稼动率、MTBF、干预率和安全论证证据
需求发现与资本开支信任爬坡约束当前至中期目前很少有场站能证明上千台人形机器人合理;分阶段采用更现实询问试点转生产的转化率、回本周期以及续约 / 扩张率

驱动因素和约束都绑定到时间,因为 GigaAI 的可投资窗口更取决于部署节奏和 ROI 证明,而不是某个自上而下的 TAM 数字。

[CM007, CM022, CM023, CM024, CM029, CM030]
FM004: 采用漏斗或价值链地图

具身 AI 采用通常从广泛战略兴趣开始,逐步收窄到少数生产部署;这些部署必须跨过安全、可靠性和 ROI 门槛。

[CM030, CM031, CM035, CM043, CM045, CM046]

2.5 图表

Chapter 03

03竞争对手

3.1 格局:直接同业、既有巨头与替代路径

GigaAI 处在具身 AI 市场,但竞争对手不只是其他中国人形机器人创业公司。其公开产品线横跨轮式工业机器人 Maker H01 和轮式家用机器人 SeeLight S1,因此买方不仅会把它同 EngineAI、AgiBot、Unitree 等中国直接同业比较,也会同 Figure、Physical Intelligence 等西方通用项目,以及 FANUC、KUKA、ABB 等传统工业自动化厂商比较。这一区分很重要,因为在行业尚未收敛到家庭、工厂或模型层软件哪一块会创造最大近端价值池之前,GigaAI 就在要求买方采用新的机器人形态。直接同业最强之处,恰是 GigaAI 在公开材料中最弱之处:Unitree 已经有公开商城和全球研究分发信号,EngineAI 的制造规模主张和 Luxshare 支持的供应链故事更响,AgiBot 则公开了更多数据集和开发者界面证据。替代集合也异常强。FANUC、KUKA 和 ABB 已经用广泛产品目录和成熟支持模式服务物料搬运、焊接、装配、码垛和检测。对许多工业任务来说,GigaAI 的实际替代品根本不是另一台人形机器人,而是采购团队已经知道如何部署和维护的低风险固定式或协作式自动化。[CP101, CP103, CP107, CP111, CP113, CP122]

竞争者画像表
公司类型公开规模 / 融资信号主要目标产品 / 套餐信号对 GigaAI 的主要竞争含义
GigaAI直接同业2026 年融资 CNY3.5B;估值超过 CNY10B;约 100 台 SeeLight 订单;1,000 台 Longsheng 框架家庭服务 + 工业搬运SeeLight S1 与 Maker H01,均为轮式具身 AI 机器人势头强,但渠道宽度仍较集中
EngineAI中国直接同业声称具备 10,000 台交付能力;$200M Series B;估值 >RMB10B工业人形机器人和商业服务T800 加更广的 PM01/SA02/JS01 产品线;制造优先套餐制造规模和 Luxshare 入口,可能让 EngineAI 在工业可信度上跑赢 GigaAI
AgiBot中国直接同业商业化量产叙事,加上 AGIBOT World 和活跃 GitHub 仓库通用具身机器人公开数据集、Omnihand SDK、VLA 工具、多条机器人产品线开发者开放度和数据护城河看起来强于 GigaAI 的公开足迹
Unitree中国直接同业 / 价格扰动者入选 Nvidia 带来全球研究分发信号研究、开发者和更广泛的人形机器人采用官方 G1 页面列出 US$13.5K 价格公开定价降低摩擦,并挤压高端同业叙事
Figure西方前沿 OEM新闻页面称 Series C 超过 $1B、投后估值 $39B,并提及 BMW 生产里程碑家庭 + 商业通用机器人Figure 03 围绕 Helix 和 BotQ 规模打造全栈垂直整合可能成为持久护城河叙事的标尺
Physical Intelligence模型层相邻竞争者2025 年融资 $600M;2026 年另有 $1B 洽谈;无商业化时间表机器人无关的基础模型层运行时和 VLA 平台,而非单一机器人 SKU如果模型层胜出,硬件差异化会被压缩
FANUC / KUKA / ABB既有替代者自动化品类和服务支持已成熟选择成熟任务专用自动化的工厂覆盖焊接、装配、码垛、机床上下料和协作机器人等的广泛目录在人形机器人跨过 ROI 与安全门槛前,既有厂商可能先赢下采购

规模和融资单元格只使用抓取来源中的公开披露或公开报道;空白经济指标是有意保持未披露,而不是估算。

[CP104, CP106, CP107, CP109, CP111, CP113]
FP101: 竞争定位地图

按公开分发能力和公开展示的部署证据,对竞品集合做序数映射。

坐标轴是有证据支撑的序数评分,不是实测市场份额:x 轴近似公开分发 / 渠道强度,y 轴近似公开展示的部署证据。

[CP106, CP107, CP109, CP111, CP113, CP114]

3.2 直接同业画像:能力、包装与定价姿态

GigaAI 的公开证据支撑一条双线策略:SeeLight S1 面向家庭服务,Maker H01 面向工业搬运,两款产品都更依赖轮式移动,而不是完全腿式行走。这让公司比纯研究玩家更具体,但也意味着比较集合很杂。EngineAI 围绕 T800 批量交付、79 个检测点和 10,000 台能力主张推进工业化叙事;AgiBot 把商业化量产信息同 AGIBOT World 和活跃公开代码库捆在一起;Unitree 是最清楚的价格颠覆者,因为其官方 G1 页面公开列出 US$13.5K 价格和硬件配置;Figure 用 Helix、家庭使用和 BotQ 制造规模来包装 Figure 03;Physical Intelligence 则并不主要卖机器人单机,而是卖一个意图控制任意机器人的模型层。因此,定价只能部分比较。Unitree 是本来源集合中唯一有清晰公开标价的直接同业。GigaAI、EngineAI、AgiBot、Figure 和 Physical Intelligence 的产品或平台叙事,都强于已实现商业定价披露。这很关键,因为在不成熟市场里,公开价格透明度本身就是竞争武器:它降低购买摩擦,扩大开发者触达,也给供应商更清楚的降本故事。[CP102, CP103, CP109, CP110, CP111, CP112]

功能 / 能力矩阵
购买标准GigaAIEngineAIAgiBotUnitreeFigure / PI
移动形态轮式家庭 + 轮式工业机器人足式工业人形机器人,加陪伴 / 四足产品通用人形机器人组合足式人形机器人,公开研究 / 开发者吸引力较强Figure = 足式人形机器人;PI = 跨机器人模型层
操作硬件双 7-DOF 机械臂;Maker H01 每臂最大载荷 5 kgT800 工业人形机器人,制造叙事强调 QA公开仓库可见 Omnihand / 具身操作生态G1 页面列出可选灵巧手配置Figure 03 强调重新设计的手部以及触觉 / 柔顺感知
家庭定位SeeLight S1 明确面向家庭不是公开核心信息在抓取来源中不是主要公开定位在抓取来源中不是官方主叙事Figure 03 明确定位家庭;PI 演示家居清洁泛化
工业定位Maker H01 在 FAW 和 Longsheng 有部署T800 强调大规模交付与工厂场景商业化量产叙事在这组来源中,研究 / 分发信号强于具名工业部署Figure 在完成家庭安全重新设计后也瞄准商业用途
开发者开放度抓取来源中公开开发者界面有限抓取来源中只有网站信息这组样本中公开数据集和 GitHub 足迹最强公开产品页和研究分发信号PI 开源了 pi0;Figure 公开研究 / 新闻叙事
公开定价抓取的一手来源中没有公开标价抓取的一手来源中没有公开标价抓取的一手来源中没有公开标价G1 页面列出 US$13.5KFigure 和 PI 披露能力,不披露标价
供应链 / 制造信号资本加具名试点客户Luxshare 支持的融资和 12,000 sqm 基地有规模和生态信号,但抓取样本中工厂细节较少零售 / 分发信心和品牌触达Figure 重建供应链并建设 BotQ;PI 将资金投向数据和合作伙伴
商业化时间线清晰度试点订单和计划交付大规模交付叙事量产叙事立即可见的价格与配置购买信号Figure 的规模化叙事;PI 明确缺少商业化时间表

标为缺少公开标价或公开界面有限的单元格,反映抓取的官方 URL 或直接报道 URL 中可见的信息;这不是断言私下企业条款不存在。

[CP102, CP103, CP107, CP111, CP112, CP113]
定价 / 打包对比
公司公开价格 / 合同模式公开包含内容未知项含义
GigaAI试点优先;抓取的一手来源中没有公开标价SeeLight S1 订单、Maker H01 规格、家庭 + 工业路线图实际成交价、保修、支持、折扣、服务人员配置已有商业证明,但价格发现仍需要直接销售
EngineAI报价式 / 抓取的一手来源中未披露T800 制造、QA 和更广机器人产品线T800 实际合同价格和支持条款竞争点在工业可信度,而不是透明定价
AgiBot报价式 / 抓取的一手来源中未披露商业化量产、数据集、GitHub、Omnihand/VLA 生态机器人实际价格和服务经济性可能在价格比较开始前,先赢下技术心智
Unitree官方 G1 页面:US$13.5K公开尺寸、DOF 范围和安全 / 保修披露按地区划分的批量折扣和售后服务同业中最显眼的价格锚;压缩高端叙事
Figure抓取的一手来源中没有公开标价Figure 03 能力栈、家庭安全设计、BotQ 制造商业单机经济性和 BMW 合同条款打包方式是全栈且有愿景,但买方预算规划不透明
Physical Intelligence软件 / 运行时模式;无机器人单机标价通用 VLA 平台、开源 pi0 历史、合作伙伴授权模式、部署价格、商业化时点可能以跨机器人层变现,而不是按硬件单台收费
FANUC / KUKA / ABB集成商 / 报价驱动大装机基数、应用专用机器人、服务网络逐任务系统定价随集成范围变化既有厂商能用成熟任务打包来支撑 ROI,而不是押注人形机器人可选性

这张表有意区分标价和套餐可见度。抓取的一手来源中,只有 Unitree 有清晰公开标价;其他行应理解为报价驱动或未披露,而不是零价格。

[CP113, CP119, CP122, CP127, CP139, CP140]
FP102: 功能广度 / 能力地图

这张高层热力图展示各竞品相对 GigaAI 的强项,维度包括定价、开放度、家庭场景准备度和制造深度。

[CP102, CP106, CP107, CP111, CP112, CP113]

3.3 分发、供应能力与切换成本

具身 AI 近端最耐久的竞争优势,不是原始灵巧度参数,而是渠道触达、零部件杠杆和装机基数学习。GigaAI 确实有早期分发证据:第一财经报道约 100 个 SeeLight S1 订单、与 FAW Tooling Die Technology 的工厂部署,以及与 Longsheng Technology 的三年 1,000 台 Maker 协议。但这看起来仍是集中式,而非广覆盖。Unitree 把面向零售的公开定价同 Nvidia 研究者系统选择带来的全球研究分发信号结合起来。EngineAI 借 Luxshare 和专属深圳基地,拿出了更强的制造和供应链深度公开证据。AgiBot 的 GitHub 和数据集足迹让开发者与研究者看得懂其工具,形成另一种分发。Figure 则通过垂直整合、BMW 证明点和 BotQ 产线以不同方式构建渠道力。在这个背景下,GigaAI 的切换成本可能真实存在,但仍在形成。一旦机器人接入安全规则、流程软件、任务特定数据采集和员工培训,替换就会痛苦。但市场仍足够早,多栈并用仍可行:许多客户可能还在比较试点,而不是押注一个长期机器人栈。这让首批部署具有战略重要性,但也意味着包括 GigaAI 在内的任何玩家,都不能仅凭试点假设自己已有耐久锁定。[CP106, CP114, CP128, CP129, CP130, CP131]

护城河耐久性 / 竞争风险登记表
护城河或风险向量GigaAI 公开位置主要威胁耐久性重要性
新近资本2026 年融资 CNY3.5B,估值超过 CNY10B整个品类资本充裕融资有助于扩张数据和硬件,但资本本身并不稀缺,难以单独构成护城河
家庭试点叙事SeeLight S1 订单和武汉试点叙事Figure 和 PI 也在宣称家庭泛化进展家庭机器人可能是大市场,但商业化时点仍不确定
工业试点关系FAW 和 Longsheng 提供具名工厂证明EngineAI 和既有厂商可用更强制造能力或装机基数可信度反击具名客户有价值,但尚未形成广泛渠道主导
价格位置无公开标价,因此价值叙事买方读不清Unitree G1 以 US$13.5K 设定锚点定价不透明会伤害开发者和价格敏感型试点买方
开发者 / 数据护城河有品牌模型,但抓取来源中的公开开发者界面有限AgiBot GitHub + 数据集;PI 开源和模型层策略低至中开放生态能更快积累采用和训练数据
制造规模试点规模商业化,一手来源中未公开工厂产能EngineAI 声称 10,000 台;Figure BotQ 产线 12,000 台 / 年规模影响降本、可靠性学习和采购信心
行业炒作风险身处融资密集但争议很大的品类Berkeley、MIT Review 和 KrASIA 都质疑时间线与商业化质量外部风险高如果采用持续缓慢,即便优秀运营者也可能被重估

耐久性评级是定性且由证据驱动,而不是数值预测;它们只反映截至 runDate 可由抓取来源支持的内容。

[CP104, CP106, CP127, CP128, CP129, CP130]
FP103: 护城河 / 准备度 KPI

用紧凑指标呈现最清晰的公开基准,这些基准正在塑造 GigaAI 的竞争位置。

[CP104, CP106, CP107, CP109, CP113, CP117]

3.4 护城河耐久性与反向证据

GigaAI 确实具备可信护城河的材料:新鲜资本、具名工业客户、家用试点叙事,以及 GigaWorld、GigaBrain 等内部品牌模型。但每一条护城河向量都有可见的反重力。品类里资本很充裕,因此融资本身并不耐久。公开试点胜利重要,但 EngineAI 和 Figure 也在放大制造故事,AgiBot 与 Physical Intelligence 则投资模型和数据飞轮,这些飞轮可能比 GigaAI 的产品特定优势更快穿越硬件代际。最强反向证据来自整个行业。Berkeley 研究者认为灵巧操作数据缺口巨大,未来两年、五年甚至十年都不应期待实用的人类水平操作。MIT Technology Review 报道人形机器人仍缺常识,采用大概率缓慢且行业特定,同时也提出 Figure 与 BMW 关系到底代表多少真实生产证明的问题。KrASIA 补上融资警告:投资人自己也越来越把具身 AI 描述为模糊,商业化处于低挡位,估值容易受泡沫动态影响。对 GigaAI 的含义很直接。公司也许很适合继续推进试点,但竞争耐久性仍取决于它能否比价格更优、更开放或分发更强的对手更快把资本和演示转成可重复的现场表现。[CP104, CP105, CP131, CP132, CP135, CP136]

Chapter 04

04财务

4.1 收入模式与定价姿态

公开记录中的 GigaAI 不是单线硬件供应商。最清楚的备案式和公司关联证据显示,其业务范围在法律上可以覆盖软件开发、行业应用系统、机器人制造与销售、设备租赁、安装维护、数据服务和进出口。2026 年 6 月独立报道进一步把变现架构说清:一条线是通过软件、API 和授权向合作伙伴出售基础模型能力;另一条线是用 SeeLight 和 Maker 系列把模型栈带入家庭和工业环境的自有产品。这意味着公司原则上可以同时变现经常性软件层和资本设备层,并围绕两者提供部署服务。缺失的是实际组合。已审阅官方或独立来源都没有披露当前收入(如有)中模型服务与机器人硬件分别占多少。定价透明度同样很薄。GigaAI 未在官方界面发布标价、公开结账流程或标准化套餐层级;媒体报道描述的是订单、试点和合同,而不是已发布商业条款。实际读法是报价制模式:工业端协商式企业销售,消费端早期项目式家庭推出,模型或仿真基础设施则很可能采用定制商业条款。这种姿态对工业机器人很正常,但它显著限制了外部人士今天推断实际 ASP、折扣或贡献毛利的能力。[CI001, CI006, CI012, CI028, CI031, CI032]

收入来源表
来源机制单位当前价值 / 状态质量尽调问题
Maker 工业机器人向工厂和物流工作流销售或部署 Maker 系列机器人,可能附带集成和维护按机器人 / 项目具名部署加 1,000 台 Longsheng 框架;未披露确认收入提供已签采购订单积压、实际 ASP、验收里程碑、保修准备金和服务附加率
SeeLight 家庭机器人家庭机器人硬件,以及未来可能的支持或服务方案按家庭单元 / 服务包公开报道约 100 个订单;公开试点铺开仍早于经验证的商业收入披露低-中披露订金条款、付费转化、安装成本、支持负担,以及流失 / 退货假设
基础模型服务面向使用 GigaBrain / GigaWorld 能力的合作伙伴的软件、API 和授权按 API、授权或企业合同2026 年 6 月报道明确提到,但该业务线没有公开合同金额或客户数量展示客户标识、合同 ACV、按用量计费条款和续约结构
DriveDreamer 仿真基础设施出售给 OEM、Tier 1 或自动驾驶团队的世界模型仿真器 / 驾驶基础设施按企业合同 / 部署公开称服务 30 多家车企或自动驾驶客户,但价格未披露披露平均合同规模、期限长度、部署范围,以及经常性与项目制收入占比
租赁 / RaaS / 部署服务围绕硬件落地的机器人租赁、安装、维护和现场服务,属于可能路径按月 / 站点 / SLA申报范围允许租赁和安装,但没有直接公开证据显示 GigaAI 已有活跃 RaaS 合同确认是否有活跃客户不是以资本开支购买,而是通过租赁、服务订阅或按用量计费付款

各行把已有公开证据支撑的变现路径与仅属可能的路径拆开。“当前价值 / 状态”只反映公开依据,不代表管理层内部假设。

[CI001, CI006, CI007, CI008, CI009, CI028]
定价 / 变现表
产品 / 对标项价格 / 单位 / 合同标价与实际价格折扣 / 未知项来源 / 含义
GigaAI Maker H01null未发现 GigaAI 公开标价实际 ASP、安装费和服务包未知官方与媒体信息更支持询价销售,而非目录价
GigaAI SeeLight S1null未发现 2026 年公开标价有公开订单数,但付款条款和试点补贴未知家庭端铺开真实存在,但商业变现条款仍未披露
SeeLight S1 外部目标2027 年 6 月前 < RMB100,000目标 / 第三方报道指引,不是当前实际成交价据称试点家庭免费获得设备;目标价可能不含支持成本可作为家庭可负担性的上行锚点,但用于承销还太早
Maker H01 外部锚点~USD160,000仅为第三方目录估计未经公司确认;配置、服务和折扣未知仅用于粗略框定企业部署 GMV
Unitree G1 官方对标US$13.5K,税费和运费另计基础竞品 SKU 的公开标价运费、关税和 EDU 定制另收费;研究版本仍需联系销售说明该品类可以做到价格透明,即便 GigaAI 尚未采用
工业机器人已部署单元代理USD80K-400K 总单元成本基准成本区间,不是 GigaAI 标价对工装、安全、集成和现场设计高度敏感支持这样一个判断:工业端实际定价以询价和项目为主

本表有意把公司特定的沉默与可比锚点放在一起,方便读者区分公开标价与推断出的企业项目经济性。空值表示未发现已审阅公开价格。

[CI012, CI013, CI014, CI016, CI017, CI020]
FI001: 收入模型桥

公开客户活动如何转化为 GigaAI 收入;硬件、软件和服务变现分开看。

这条流是商业化地图,不是收入确认排期。它把公开需求信号与仍未披露的现金转化机制分开。

[CI001, CI006, CI007, CI008, CI009, CI031]

4.2 公开牵引力 vs 收入可见度

公开牵引力存在,但主要是运营层面,而非财务层面。第一财经、QQ 和相关中文报道在三项头部商业化信号上收敛:大规模家庭交付前约 100 个 SeeLight S1 订单、与 Longsheng Technology 三年部署 1,000 台 Maker 系列机器人的框架,以及在 FAW Tooling Die 使用 GigaWorld-GigaBrain-Maker 栈执行拆垛、跨区域运输等任务的工业部署。更早的 Gasgoo 报道也显示,首批 Maker H01 发往湖北人形机器人创新中心,并把 2026 年描述为从模型叙事走向场景落地和交付的一年。软件侧,QQ 和 NCSTI 把 DriveDreamer 描述为已签约的基础设施业务,服务 30 多家国内外车企或自动驾驶客户;这一点重要,因为它暗示自动驾驶仿真分支不只是纸面资产。但这些牵引信号都没有闭合会计环路。公开来源没有披露已确认收入、递延收入、毛利、客户集中度,也没有说明 1,000 台框架中多少是有约束力的采购订单、多少是分阶段意向。它们也没有解释家庭订单是全额付款、部分预订、有补贴,还是免费试点投放。因此,公司比纯演示型创业公司有更多商业化证明,但距离财务上可承销的成长公司仍差得很远。[CI007, CI008, CI009, CI010, CI011, CI029]

公开财务缺口表
缺失的私有指标对分析的影响精确尽调路径
按业务线确认收入无法判断部署是否正在变成现金,也无法判断哪条业务线应享受软件式倍数要求提供月度收入桥,拆分 Maker 硬件、SeeLight 硬件、模型服务、DriveDreamer 和支持
按业务线毛利无法知道家庭铺开、工业安装或仿真器合同是在改善经济性,还是在稀释经济性要求提供产品级毛利瀑布,包含硬件 BOM、现场服务、货运和算力
现金、烧钱和资金可支撑期即便 2026 年募资规模很大,资本充足性仍只能凭印象判断要求提供董事会材料或贷款人材料,包含当前现金、月度烧钱和资金可支撑期情景
积压订单质量和收入确认规则如果里程碑宽松,1,000 台框架协议和家庭订单可能高估近期收入要求提供已签合同摘要,包括验收标准、取消权和收入确认政策
DriveDreamer 合同价值和续约数据驾驶基础设施可能是最干净的经常性收入流,但价值被隐藏要求提供头部客户 ACV、实施时间、毛利率,以及续约 / 扩张 cohort
家庭服务成本和事故画像规模化之前,家庭机器人可能因服务、安全和更换成本摧毁毛利要求提供试点家庭批次数据:安装时间、支持工单、事故率和每台活跃设备维护成本

这些是今天无法形成干净承销案例的具体盲点;每一行都点名了可以解决问题的具体文件或数据集。

[CI007, CI008, CI009, CI017, CI029, CI030]
FI003: 财务估算区间

对若干财务锚给出公开可支撑区间,并明确区分已报道资本与估算价格代理指标。

前三项数值单位为 RMB 百万元,最后一项为 USD 百万元;由于证据混合了币种和确定性层级,displayValue 承载可读单位。

[CI003, CI004, CI008, CI009, CI016, CI017]

4.3 单位经济性与资本强度

最有用的财务视角是比较式,而不是公司专属式,因为 GigaAI 几乎没有披露投资人通常用来判断机器人单位经济性的项目。外部锚点显示,这套栈资本强度很高。2026 年工业机器人指南把仅机械臂价格大致放在 USD25,000 到 USD180,000 区间;若把工装、集成和安全纳入,完整部署单元往往达到 USD80,000 到 USD400,000,多年总拥有成本远高于标价。Standard Bots 的 ABB 定价指南也指向同一方向:经销商报价因型号和地区而异,一套装备齐全的重载系统可能超过 USD120,000。对 GigaAI 本身,Humanoid.guide 的 USD160,000 Maker H01 锚点只能作为低置信度外部对照,不能当作经验证的公司定价。家庭端经济性更不确定。RoboActu 报道称 SeeLight S1 试点家庭免费拿到设备,硬件价格目标是在 2027 年中降至 RMB100,000 以下,这意味着家庭业务可能仍处在有补贴的数据采集和产品学习阶段,而非成熟毛利阶段。可比公开机器人公司财报也强化了谨慎结论。UBTECH 2025 年业绩显示人形业务强劲增长、毛利率 37.7%,但公司在扩产中仍录得显著亏损。iRobot 申报文件则显示,当需求、价格和成本结构无法匹配时,消费机器人业务会遭遇库存压力、促销压力和持续经营风险。对 GigaAI 来说,正确结论不是单位经济性很差,而是它仍未被证明,并且很可能按产品线高度异质。[CI016, CI017, CI018, CI019, CI020, CI021]

单位经济性表
指标数值 / 空值置信度为什么重要尽调要求
已确认收入运行率null没有收入规模,任何估值和资金可支撑期讨论都只是猜测按硬件、软件和服务披露月度及过去 12 个月确认收入
毛利率 %null毛利决定硬件放量和家庭支持是在创造价值,还是更快烧钱披露合并毛利率及各产品线贡献毛利
Maker 实际 ASPnullASP 决定硬件毛利和服务附加销售的上限提供试点交易与规模化交易的合同 ASP、安装收入和折扣政策
SeeLight 全口径家庭获客成本null家庭端经济性取决于安装、培训、维护和事故成本,不只是机器人 BOM按 cohort 提供每户部署、支持、维修和更换成本
DriveDreamer 年合同价值null如果合同可重复且粘性强,仿真基础设施可能是毛利最高的业务线披露平均合同价值、实施投入,以及续约 / 扩张率
可比毛利代理UBTECH 2025 毛利率 37.7%;iRobot 2024 毛利率 20.9%公开可比公司显示,机器人毛利结果会随产品组合和规模阶段大幅波动解释 GigaAI 预期相对工业和消费代理组合处在什么位置
可比已部署系统成本代理工业单元通常 USD80K-400K,七年 TCO 为 1.8-2.5x有助于解释企业买家为什么常靠报价、试点和 ROI 案例采购,而不是按标价下单提供客户销售流程中使用的内部 ROI 计算器假设

每个空值字段都代表缺少公开披露,不是零值。可比行已明确标为代理,而非 GigaAI 事实。

[CI020, CI021, CI024, CI026, CI027, CI030]
FI002: 单位经济性桥

基于公开代理指标,用定性桥展示 GigaAI 毛利可能在哪里生成,也可能在哪里被吃掉。

这座桥使用定性节点,因为公开证据能支撑成本驱动因素,但不足以为 GigaAI 搭出完整数字瀑布。

[CI016, CI017, CI020, CI021, CI024, CI026]

4.4 资本充足性与融资依赖

近期融资给了 GigaAI 空间,但披露不足以判断公司已经能自我供血。支撑最强的公开事实,是 2026 年 3 月至 6 月横跨 Pre-B、B1 和 B2 的融资潮,合计约 RMB3.5 billion,投后估值超过 RMB10 billion。第一财经和 NCSTI 关于资金用途的表述高度一致:管理层计划继续投入数据与算法「双金字塔」,迭代物理 AGI 基础模型,并扩大家庭和工业部署。换句话说,新资金投向进一步模型开发和商业化建设,而不是一台已经清晰盈利、正在吐现金的机器。这并不天然负面;前沿机器人业务常常需要这样。但资本充足性无法完整评估,因为已审阅公开来源没有披露账上现金、月度烧钱、跑道月数、债务负担、客户融资义务或项目融资结构。可比上市公司说明了这个遗漏为何重要。ABB 2025 年综合报告即便在成熟规模下,也记录了数十亿美元级年度资本开支和研发预算;UBTECH 则显示,人形机器人收入快速增长仍可能与扩产期的大额经营亏损并存。含义是,GigaAI 近期资本是战略缓冲,而不是证明公司无需下一轮融资就能支撑家庭推出、工业安装、支持义务和模型研发。下一轮融资讨论应从标题估值转向付费部署转化、重复软件收入和毛利耐久性的证据。[CI003, CI004, CI005, CI022, CI023, CI024]

资本充足性表
指标数值 / 状态置信度为什么重要尽调要求
近期募资2026 年 3-6 月募资 RMB3.5B当前文件中最强的公开偿付能力支撑确认总募资与净到账,以及是否包含老股或受限用途资金
融资后估值> RMB10B反映市场胃口,但不代表流动性提供股权结构表、清算顺位,以及任何棘轮 / 优先权条款
账面现金null没有当前现金,无法评估 runway披露非受限现金、受限现金和近期资金义务
月度烧钱null如果部署和研发烧钱极高,近期融资仍可能不够提供月度现金消耗,拆分研发、制造、销售和支持
资金可支撑月数null对一家同时扩硬件和模型的公司,资金可支撑期是资本充足性的核心测试提供当前计划下的基准资金可支撑期,以及转化延迟下的下行资金可支撑期
资金计划用途数据与算法系统、模型迭代,以及家庭 / 工业场景规模化部署说明资金投向持续建设,而不只是修补资产负债表按模型训练、制造、GTM、家庭运营和营运资本披露预算分配
下一轮触发条件应该是付费部署转化、重复软件合同和已披露毛利结构,而不是又一次叙事性估值上调未来融资质量取决于订单能否转化为现金效率高的收入证据定义董事会层面里程碑,用于支撑下一轮主融资或债务工具

本表关注未来资本充足性,不重复前文已经覆盖的完整历史融资时间线。空值表示在已审阅公开来源中未披露。

[CI003, CI004, CI005, CI034, CI035, CI036]
FI004: 资本强度 / 现金流地图

这张示意图展示,在完整自我造血可见之前,GigaAI 新募资本可能消耗在哪里。

只有期初融资额来自直接报道。所有分配桶均为作者估算,依据是公司披露的资金用途以及机器人资本密集度代理指标。

[CI003, CI005, CI022, CI024, CI025, CI034]

4.5 财务结论

可投资的正面在于广度。GigaAI 似乎不止一种变现选项,而且这些选项与真实公开牵引力对齐:工业机器人部署、早期家庭需求、基础模型服务叙事,以及看起来已有外部客户的自动驾驶仿真基础设施分支。可投资的负面在于不透明。公司仍不公布已确认收入、毛利率、烧钱或现金跑道,公开定价也缺席、以报价制为主,或只是第三方松散估算。这让故事更像「融资加现场证明」,而不是「可审计的财务扩张」。因此,我的财务结论是谨慎正面但尚不可承销:2026 年融资后,近端偿付风险看起来降低,但在 GigaAI 披露多少部署转化为收入、收入中硬件与软件各占多少、家庭推出附带哪些服务义务,以及放量后毛利率改善还是压缩之前,财务质量仍无法干净打分。投资人只有在公司能证明当前订单的付费转化、重复工业或仿真合同,并至少部分披露毛利、现金消耗和支持经济性时,才应认为下一轮融资有依据。[CI028, CI029, CI030, CI033, CI035, CI036]

4.6 图表

Chapter 05

05产品与技术

5.1 已交付产品、品牌与用户任务

GigaAI 已不再只是围绕单个机器人演示包装的研究品牌。其官网公开产品界面覆盖四条锚定路线——Maker H01、GigaBrain、GigaWorld 和 DriveDreamer——而 2026 年 6 月发布报道又加入家庭品牌 SeeLight,S1 和 S2 是同一栈面向家庭的一支。这意味着公司至少同时在卖三类东西:具身基础模型、自有机器人本体,以及面向自动驾驶等相邻垂直领域的世界模型工具。按工作流拆,Maker H01 是执行界面,GigaBrain 是决策层,GigaWorld 是具身数据与仿真层,DriveDreamer 是驾驶世界模型分支。[CE001] [CE003] [CE004] [CE006] [CE028] [CE033] [CE034] 用户任务也同样分裂。工业用户侧,已披露任务包括箱体拆垛、跨区运输、动态避障,以及真实工厂里的精细操作。家庭用户侧,SeeLight 系列围绕家务和更长期的家庭辅助展开。AV 客户侧,DriveDreamer 被销售为数据生成、长尾场景处理和闭环仿真的基础设施。最宽的解读是,GigaAI 想拥有从数据捕获到策略学习再到物理执行的整条栈,而不是只变现一个机器人 SKU 或一个模型 API。[CE010] [CE031] [CE033] [CE036] [CE041]

产品模块 / 资产矩阵
产品 / 资产主要用户状态 / 成熟度差异化尽调缺口
Maker H01工业、服务和开放场景运营方早期商业交付 / 试点铺开原生机身直接接入 GigaBrain 和 GigaWorld 闭环未公开 MTBF、机队可用率或服务网络
GigaBrain-0 / 0.5M*机器人开发者和机器人运营方已公开技术发布,并带有基准测试主张以世界模型为条件的 VLA 路径,包含结构化任务与运动规划公开证明多集中在基准和演示,缺少 SLA 证据
GigaWorld-0 / GigaWorld-Policy具身 AI 开发者开源框架加更新的策略层数据放大,加闭环仿真和 RL 训练营销主张对应的直接公开基准链接并不完整
SeeLight S1 / S2家庭和住宅运营方S1 已获订单;S2 宣布于 2026 年 Q3 推出面向消费者的品牌,绑定同一套基础模型栈安全认证、定价和维护模式未披露
Maker M01 / U-01 / E-01数据采集运营方作为数据金字塔一部分发布面向真机、手持和第一视角采集的专用硬件精确吞吐量、BOM 和部署规模未公开
DriveDreamer 系列AV OEM 和自动驾驶团队从研究走向商业的分支,并已有具名客户活动驾驶场景中的世界模型数据生成、重建和闭环仿真公开客户台账和基准反向链接仍不完整

各行拆分具身产品、数据硬件和 AV 世界模型产品,方便读者看清 GigaAI 在哪里卖机身、策略、数据引擎或相邻基础设施。

[CE001, CE003, CE006, CE010, CE028, CE033]
工作流 / 用例表
用户任务当前工作流GigaAI 方案可衡量收益限制
重复执行工厂搬运任务硬编码自动化,并为每个工位重新调参工厂工作流中的 Maker H01 + GigaBrain + GigaWorldFAW 案例称适配周期从数月缩短到数周没有按任务类别披露的长期故障率数据
推出家庭辅助机器人服务招募人员,或依赖狭窄的单一用途设备SeeLight S1 / 规划中的 S2 家庭平台100 户订单簿和 2026 年 Q3 扩张目标已公开没有公开家庭留存、事故或服务数据
采集高保真具身训练数据只能使用小规模且昂贵的机器人机队Maker M01 加 S1、U-01 和 E-01,覆盖多个数据层监督来源比纯机器人采集更宽每台设备的精确采集成本和数据量未披露
训练通用机器人策略在有限真机演示上微调GigaWorld 数据引擎 + GigaBrain 策略栈披露 10,931 小时预训练组合,并包含合成覆盖公司基准之外,合成到真实的贡献仍难审计
生成并测试 AV 边缘场景等待罕见道路事件,或为昂贵仿真闭环付费DriveDreamer / DriveDreamer4D / UniDriveDreamer声称降低角落案例采集负担和测试成本客户规模和基准链接尚未在公开信息中完全对齐

本表把公司的模型和机器人词汇翻译成买方任务,让数据基础设施、具身执行和 AV 仿真的区别更清楚。

[CE011, CE013, CE028, CE031, CE033, CE035]
FE002: 客户工作流 / 运营流

GigaAI 如何把场景需求转成数据、模型适配和实体部署。

[CE003, CE006, CE008, CE010, CE014, CE028]

5.2 双金字塔架构与世界模型栈

核心产品技术论点是 Dual Pyramid 系统:一个金字塔负责数据获取,一个负责算法学习。数据金字塔从广泛但弱监督的互联网视频,向上经过人类演示、世界模型仿真器、合成仿真,最终到高保真实体机器人数据。算法金字塔则从世界仿真推进到动作对齐,最终到经验强化。这个框架很关键,因为 GigaAI 主张具身扩展不会来自单一模型技巧,而会来自同时控制数据阶梯和学习阶梯。[CE008] [CE009] [CE010] [CE011] 在这套架构里,GigaWorld 是数据引擎,GigaBrain 是动作策略。官方材料和开放代码库把 GigaWorld-0 描述为一个开放框架,用来生成高泛化具身交互数据,并提供闭环仿真和 RL 训练支持。GigaBrain 则消费多模态输入——包括图像、点云、文本和具身状态——并输出结构化任务和运动规划。更高级的 GigaBrain-0.5M* 层加入 RAMP,用世界模型对未来状态和价值的预测来约束策略。因此,这套栈不只是「机器人加模型」;它是一套反馈架构,世界模型输出既成为训练数据,也成为动作上下文。[CE005] [CE006] [CE007] [CE013] [CE014] [CE016] [CE022]

技术 / 运营架构表
层级 / 组件角色关键依赖主要风险
五层数据金字塔从网络视频到真实机器人,提供由弱到强的监督信号低成本采集硬件加仿真栈覆盖范围和准确性很难同时放大
GigaWorld-0 / GigaWorld-Policy生成具身交互数据和世界-动作先验高保真仿真和世界模型训练基础设施基准证明和外部复现仍处早期
GigaBrain / RAMP把多模态上下文和未来状态 / 价值预测转成机器人动作稳定的数据格式、未来预测质量、人类在环铺开策略质量取决于世界模型预测是否有价值
GigaTrain / GigaDatasets / GigaModels提供分布式训练、整理、打包和可部署模型库开源维护和硬件可用性开源仓库展示能力,不保证生产支持
Maker / SeeLight 机身在工厂和家庭中承载策略,同时回流新数据机械可靠性、电池、机械臂和感知集成真实环境里,机身可靠性可能卡住模型优势
DriveDreamer 分支把世界模型经济性延伸到驾驶仿真和数据生成获取客户数据、传感器和评估闭环客户数量披露和基准反向链接并不均衡

架构行按价值创造位置分组:数据、世界模型、策略、基础设施、机身和相邻驾驶业务。

[CE006, CE008, CE009, CE014, CE018, CE019]
FE001: 产品架构图

从数据捕获到策略学习、再到具身执行的公开披露层级。

[CE006, CE007, CE008, CE009, CE014, CE022]
FE003: 关键依赖图

这些依赖决定 GigaAI 能否扩大产品交付、训练质量和可靠性。

[CE010, CE017, CE018, CE024, CE031, CE033]

5.3 部署、集成与可靠性证据

最强公开部署证明不是消费者应用指标,而是一串现场 rollout。Gasgoo 报道 Maker H01 首批交付湖北人形机器人创新中心;2026 年 6 月报道则称 GigaWorld + GigaBrain + Maker H01 三件套进入 FAW Tooling Die 执行真实工业流程,并计划与 Longsheng Technology 推进三年 1,000 台 rollout。家庭侧,SeeLight S1 绑定约 100 户订单簿,并计划 2026 年第三季度规模化运营。这些披露表明,GigaAI 把部署当成模型闭环输入,而不只是收入终点。[CE029] [CE030] [CE031] [CE032] [CE033] [CE034] 集成证明也强于许多硬件优先的机器人创业公司,因为公开代码库展示了明确接口。GigaBrain 包含推理服务器和客户端路径;GigaTrain 处理分布式训练和 checkpoint;GigaDatasets 打包并评估多模态机器人数据;GigaModels 覆盖训练、推理、部署和压缩。开源界面让这套栈更可读,但可靠性证据仍更多来自试点和基准,而不是运营:公开材料尚未披露正常运行时间、MTBF、现场故障率或客户支持 SLA。[CE017] [CE018] [CE019] [CE020] [CE021] [CE037] [CE039]

5.4 差异化、开放工具与相邻驾驶栈

GigaAI 的差异化不在单个基准奖杯,而在栈的广度。它已经开源了具身 VLA 代码库、世界模型代码库、训练框架、数据框架和共享模型库,公开 GitHub 组织在这些界面上也已有可观 star 牵引力。这给了公司强于典型封闭机器人 OEM 的开发者信号,也让产品故事更容易审计:这套栈可检查、可复现,并以基准为导向。[CE018] [CE023] [CE024] [CE037] [CE038] 驾驶分支同样重要,因为它拓宽了变现和数据飞轮论点。DriveDreamer、DriveDreamer4D 和 UniDriveDreamer 显示,GigaAI 并没有把世界模型当成机器人专用小众方向;它把同一套「数据机器」逻辑用于 AV 仿真、4D 重建和多模态 camera-LiDAR 生成。如果这些分支继续共享基础设施和人才,它们会强化公司关于护城河来自世界模型运营、而非单一本体形态的主张。风险在于,客户数量和基准规模主张在官方和媒体界面之间仍记录不均。[CE025] [CE026] [CE027] [CE028] [CE040] [CE041] [CE042]

FE004: 产品成熟度 / 能力图

基于文档深度、部署证明、信任披露和路线图风险,对各项资产做定性成熟度判断。

[CE018, CE023, CE024, CE029, CE031, CE033]

5.5 信任、质量、合规与路线图姿态

当前公开质量信号来自开放代码、基准参与和具名部署引用,而不是传统企业信任栈。代码库采用 Apache 2.0 许可,GigaBrain challenge 围绕 GigaWorld 和 RoboChallenge 正式化社区评测赛道,技术页面也展示训练、推理和评估路径,而不只是精修发布视频。这些都是公司希望外部使用和审视的有用证据。同时,公开官网仍未披露定价、正式安全认证、支持 SLA 或专门信任门户,这限制了买方仅凭公开材料判断生产就绪度的能力。[CE017] [CE023] [CE037] [CE038] [CE039] 路线图很激进。2026 年 6 月公开材料指向 Q3 里程碑:SeeLight S1 规模化运营、SeeLight S2 发布和 GigaBrain-1,随后在 12 个月物理 AGI 路线图中推出 GigaBrain-2 和 GigaBrain-3。该路线图与 Dual Pyramid 叙事在战略上自洽:更多家庭和工厂部署产生更多数据,更多数据喂给世界模型和动作模型闭环,下一代机器人理应变得更好用,而不只是舞台上更惊艳。但在 GigaAI 发布更清楚的安全、可靠性和基准链接证据之前,这一路线图应被理解为雄心很大,但尚未完全去风险。[CE011] [CE032] [CE033] [CE034] [CE035] [CE040]

信任 / 质量 / 合规表
控制 / 信号状态范围证明了什么缺口
RoboChallenge 排名主张公司公开声称,且存在外部具名基准具身真实机器人评估内部演示之外,至少有一个公开质量门槛官方页面没有从每个营销页面直接链接到结果台账
开源 Apache 2.0 许可核心仓库均可见研究和开发者采用技术栈可检查,也可合法用于实验开源许可不等于企业支持或赔偿
具名工业部署参考由独立报道提及工厂用例技术栈已经走出纯实验室演示没有公开可用率、MTBF 或铺开 KPI 仪表盘
具名家庭订单和 Q3 运营计划多篇 2026 年 6 月报道提及家庭机器人发布路径公司在尝试真实家庭部署,而不只是展厅演示安全流程、现场支持和保修模式未公开
正式安全 / 合规认证官方网站未公开呈现家庭和工业机器人信任姿态缺席本身就是尽调信号需要公布认证、测试制度和运行约束
面向客户的 SLA / 支持门户官方网站未公开呈现售后运营当前公开材料产品优先,不是支持优先需要披露响应时间、维护和事故处理

本表区分可见质量信号和缺失的信任信号;缺失项被列入,是因为家庭和工业机器人中披露缺席本身就很重要。

[CE002, CE012, CE017, CE024, CE037, CE038]
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
2025-11Maker H01 发布已宣布 / 已亮相说明本体策略早于当前交付推进官方资料包 + Gasgoo
2026-01 至 2026 年初Maker H01 首批交付据称已开始产品从亮相走向客户发货Gasgoo
2026-04FAW Tooling Die 工厂方案据称已落地工业集成被用作泛化能力验证阶段NCSTI + 36Kr
2026 Q3SeeLight S1 扩大家庭运营计划中家庭铺开将成为首个大规模公开家庭数据闭环NCSTI + 36Kr + Sohu
2026 Q3SeeLight S2 发布计划中第二代本体意在改善家庭可用性NCSTI + 36Kr + Sohu
2026 Q3GigaBrain-1 发布计划中下一代基础模型成为 Dual-Pyramid 时代第一个正式模型里程碑NCSTI + Sohu
未来 12 个月GigaBrain-2 / GigaBrain-3 扩展路径计划中模型路线图把未来能力绑定到 1000 万小时视频 + 100 万小时世界动作数据NCSTI + Sohu

日期混合了发布、交付和路线图节点;这是一条披露时间线,不是经审计的生产排期。

[CE011, CE029, CE031, CE032, CE033, CE034]

5.6 图表

Chapter 06

06客户

6.1 客户分层、买方结构与证据真正最强的地方

GigaAI 的公开客户故事比单个类人机器人 SKU 更宽,但证据在各分层之间并不均匀。今天最能支撑的分层是工业制造、家庭或养老试点,以及汽车仿真。工业证明由 FAW Tooling Die 的具名工厂部署、与湖北人形机器人创新中心的首批交付关系,以及无锡 Longsheng 项目锚定。汽车证据更窄但真实:ECARX 已正式把 DriveDreamer 集成进 AutoGPT,这给了 GigaAI 至少一个直接 Tier-1 风格的世界模型采用证明点。家庭和养老证据仍早,但可见度异常高,100 台家庭发布叙事在中英文媒体中反复出现。相比之下,3C 电子、仓储或物流更适合理解为公司宣称覆盖的分层,而不是具名、独立逐项列出的客户证据。这一区分很重要,因为它既让 GigaAI 的客户地图保持可信,又不会夸大公开证据。买方、用户、付款方在各分层中也不同:工厂和公共机构似乎赞助部署,运营团队是用户,国资背景试点或企业预算很可能是付款方;家庭则仍更接近试用参与者,而不是经常性订阅者。[CU001, CU002, CU013, CU017, CU022, CU023]

客户分层表
分层买方 / 用户 / 付款方具名证明规模信号商业化解读公开缺口
工业制造买方 / 付款方 = 工厂运营方或赞助方;用户 = 工厂运营团队FAW Tooling Die、Longsheng / Longsheng Weirui、湖北中心Longsheng 1000 台计划;据称首批交付已发生当前最强公开证明,因为场地和任务都有具名未公开合同金额、续约或正常运行时间
汽车 / 自动驾驶买方 = Tier-1 或 OEM 项目负责人;用户 = AI 与座舱团队ECARX AutoGPT 集成;Li Auto 只被间接提及ECARX 官方集成;公司相关报道反复提及 30+ 客户说法真实合作伙伴证明存在,但具名客户清单很薄30+ 客户说法背后没有公开逐项清单
家庭 / 养老照护买方不清晰;早期用户是试点家庭;付款方可能仍是 GigaAI 或赞助方武汉家庭试点、员工住房、人才公寓多个来源提到约 100 台;2026 Q3 / 2027 H1 时间因来源而异能见度高,但仍以试点为主订单与试用的混合口径、定价模型仍未定
湖北创新生态买方 / 付款方可能是公共机构或项目赞助方;用户 = 研究和场景团队湖北人形机器人创新中心首台 Maker H01 交付,加上数据工厂协作可作为部署和数据闭环证明很难把客户收入与生态合作价值拆开
仓储 / 物流可能是工业运营方和合作伙伴牵头的场景站点仅有 Longsheng 相关工业物流场景无锡和 Gasgoo 报道中的场景表述扩张路径合理,但尚未独立具名未披露独立客户或订单金额
3C 电子可能是工厂和电子组装商仅见行业报道中的分层级说法被称为标杆客户垂直领域应视为公司声称的触达,而非已证明的具名牵引未披露具名客户、场地或 KPI

各行把可支撑的具名证明与泛化的分层说法拆开;缺少合同金额或续约数据,本身就是客户质量评估的一部分。

[CU001, CU002, CU013, CU017, CU022, CU033]
FU001: 客户旅程图

GigaAI 当前客户路径先从生态发现进入真实场景测试,再走向规模化主张;续约证明仍待补齐。

[CU002, CU004, CU006, CU013, CU026, CU038]
FU004: 按分群划分的公开证明密度

工业制造和家庭试点的公开证明最密;3C 电子和物流在具名客户上仍落后。

数值是作者从已审阅来源中编码出的各分群不同公开证明点或具名证明持有方数量,不是公司披露客户数。

[CU001, CU033, CU034, CU035, CU036]

6.2 具名客户证明与从首批交付到规模主张的采用轨迹

采用轨迹最好不要只看单个标题,而要按阶梯读。第一,盖世汽车称,GigaAI 已开始交付 Maker H01,首台交给了湖北人形机器人创新中心。第二,2026 年 4 月,GigaAI 的工业验证进入具名汽车制造场景:一汽模具被披露为实际工厂点位,用于卸箱、跨区域转运、动态避障和精细作业。第三,2026 年 6 月又新增龙盛关系,公司与无锡伙伴宣布三年部署 1,000 台 Maker 系列机器人,并称首批已通过工业搬运验证且完成交付。第四,家庭路径从演示走到 100 台首发叙事。但采用线索在这里开始变得不够干净:有的媒体说订单,有的说员工住房或人才公寓测试,中国日报则称这是面向普通家庭的免费试用。最后,DriveDreamer 侧至少有 ECARX 这一直接公开客户证明,但更宽泛的 30 多个客户数仍是未逐项列明的说法。结果是:公司确实在从概念走向现场使用,但公开扩张语言仍更多由计划支撑,而不是由收入支撑。[CU003, CU004, CU006, CU007, CU008, CU009]

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
首次披露的 Maker H01 交付向湖北中心初始交付 1 台2026 年初Gasgoo产品从亮相进入客户发货未披露总订单簿
FAW 工厂铺开真实工厂任务已具名2026-04NCSTI + Yicai + 36Kr工业证明从概念走向具名生产场地没有装机规模或合同期限
Longsheng 规模计划3 年 1000 台机器人2026-06Gasgoo + Yicai + VOI公开记录中最大规模的已宣布扩张路径管线承诺,不是已交付机队
Longsheng 首批里程碑搬运验证通过并已交付2026-06Gasgoo说明至少部分设备通过了真实工业验收未披露首批数量
SeeLight 家庭发布叙事约 100 台设备 / 订单 / 试用2026 Q3 / 2027 H1来源:China Daily + Yicai + eWeek + Mike Kalil家庭铺开有公开动能,也有公开口径混乱付费订单与免费试用尚未对齐
家庭需求兴趣2000+ 条 WeChat 留言2026-06China Daily显示大规模发布前的消费者好奇心不是转化或留存指标
直接 DriveDreamer 合作伙伴证明ECARX AutoGPT 集成2025-03ECARX / GlobeNewswire显示至少一条在跑的汽车软件客户路径未披露收入或部署数量

这是一条披露阶梯,不是经审计的收入桥;越往后越偏计划,越少依赖已装机规模。

[CU003, CU006, CU009, CU010, CU013, CU018]
具名客户证明表
客户 / 证明持有方分层部署或用例生产 vs 试点结果 / 证明限制
湖北人形机器人创新中心工业 / 研发试验场首台 Maker H01 交付和数据工厂协作试点 / 验证场显示发货、数据采集和制造服务推广收入条款和持续采购量未披露
FAW Tooling Die汽车制造箱体卸载、跨区域运输、动态避障、精密作业试点到生产的桥接具名工厂任务和更快适配周期已公开未公开合同金额、正常运行时间或续约披露
Longsheng / Longsheng Weirui工业制造与物流三年 1000 台计划,加上首批搬运验证已宣布规模,早期已有交付公开记录中最强规模标题尚不等同于 1000 台已安装付费设备
ECARX汽车 AI / 仿真DriveDreamer 集成进 AutoGPT生产软件集成说法面向车企的世界模型采用有直接合作伙伴证明无法验证更广泛的 30+ 客户说法
武汉家庭试点 / 光谷住房家庭 / 养老照护首批 100 台 SeeLight 进入普通家庭、员工住房或人才公寓试点 / 试用确认真实家庭测试,而不只是展厅演示公开来源对设备是付费订单还是免费试用存在冲突

具名证明在具体场地、合作伙伴或家庭项目公开时最强;各行混合了试点、集成和规模宣布,因此成熟度需要谨慎解读。

[CU003, CU004, CU006, CU009, CU010, CU013]
FU002: 采用 / 部署漏斗

公开证明从宽泛分群主张很快收窄到少数具名场地,公开留存披露几乎没有。

作者基于已审阅公开证明集编码计数:分群 = 汽车制造、家庭或养老、汽车仿真、仓储或物流、3C 电子;具名证明持有方 = 湖北中心、FAW、龙昇、ECARX、武汉家庭试点。

[CU001, CU009, CU013, CU031, CU034, CU035]
FU003: 客户证明矩阵

具名证明之间的证据质量差异很大:工业任务更具体,但几乎所有案例的收入和留存能见度都很弱。

[CU003, CU006, CU009, CU013, CU018, CU020]

6.3 留存、满意度和服务可见性仍是公开记录中最弱的部分

相比许多只展示实体 AI 的初创公司,GigaAI 现在有更多可见部署证明,但公开记录在耐久性指标上仍严重滞后。已审阅来源没有披露 NRR、GRR、流失率、续约率、合同期限,也没有给出工业账户或家庭用户的 cohort 视图。因此,本章可以核验启动动作和具名点位,却无法判断客户是否续约、扩张,或安装后是否产生经常性利润。家庭试点来源在这一点上尤其说明问题。中国日报把首批 100 台家庭设备描述为免费试用,并把数据采集和故障识别纳入流程。eWeek 的现场报道补上了具体质量缺口:有些家务仍然慢、别扭或不整洁;这反而有价值,因为它说明机器人是在真实环境里测试,而不只是停留在精修演示中。中国日报提到 2,000 多条 WeChat 留言,说明有兴趣,但兴趣不等于转化。最诚实的读法是:当前满意度证明更多是轶事和运营层面的,而不是合同层面的。在 GigaAI 公布续约、服务和维护指标之前,留存应被视为尽调问题,而不是可推断的优势。[CU020, CU025, CU027, CU028, CU030, CU031]

留存 / 重复使用 / 满意度表
指标数值 / null分层证据质量尽调追问
NRRnull工业 + 软件无公开披露要求提供按具名客户拆分的账户级 NRR 和扩张情况
续约率null工业 + 软件无公开披露要求提供已签续约或重复采购历史
合同期限null工业 + 软件无公开披露要求提供平均期限和试点转付费时间
家庭延续率null家庭无公开披露跟踪免费测试后有多少试用家庭保留机器人
需求兴趣2000+ 条入站留言家庭China Daily 引用信号把好奇心与实际注册、付款转化拆开
任务质量提示部分家务仍慢或不够利落家庭eWeek 现场报道要求按家务类型提供任务成功率和人工介入率
售后模型仅在探索家庭China Daily / 湖北中心引述要求提供维护成本、响应 SLA 和服务人员配置计划

null 是有意保留:本章找到了发布证据,但没有公开留存、续约或服务质量 KPI。

[CU025, CU027, CU031, CU032, CU041]

6.4 扩张已经出现,但集中度和生态依赖仍是真实风险

公开材料里已经能看到最合理的扩张闭环:试点或首次部署产生数据,数据改进 GigaWorld 或 GigaBrain,升级后的技术栈覆盖更多场景,更丰富的证明再帮助公司拿下下一个点位或伙伴。湖北和武汉处在这个闭环中心,因为它们同时提供数据工厂、智能制造推广、政策资金和经过筛选的住宅测试床。龙盛则增加了第二条规模化路径:在无锡把制造支持与多年部署计划结合起来。这些都是优势,但也暴露了集中度风险。目前大部分具名证明落在少数政府关联或伙伴主导的生态里,公开来源没有披露收入或装机基数有多大比例依赖这些生态。龙盛 1,000 台计划也是管线承诺,不是已经落地的装机量。同样,家庭牵引在接触广泛中等收入需求之前,仍看起来集中在高端或政策支持的住房环境。因此,商业上行空间可信,但仅凭公开证据,市场还没有多元化到足以让投资人承销低集中度风险。[CU026, CU037, CU038, CU039, CU040, CU042]

扩张与集中度风险表
扩张驱动集中度风险影响尽调路径
湖北数据工厂与真实场景闭环严重依赖单一区域生态可能夸大证明向其他区域迁移的可行性要求提供非湖北客户管线和已赢单证据
Longsheng 1000 台计划单一合作伙伴规模依赖一个合作伙伴项目滑坡,就可能主导公开牵引表现要求提供分阶段交付排期、验收标准和取消条款
FAW 制造参考参考客户集中旗舰证明场地能帮助销售,但不能保证覆盖广度要求提供更多汽车工厂 logo 和重复场地扩张
ECARX DriveDreamer 集成具名汽车客户基础薄直接证明存在,但更广泛的汽车渗透可能仍浅要求提供 OEM 清单、量产 SOP 范围和软件收入模型
家庭人才公寓试点精选环境偏差试点成功未必能迁移到大众家庭要求按员工家庭、普通家庭和付费用户拆分试点组合
武汉和无锡政策支持场景资金政策辅助需求 vs 市场需求政府支持能加速证明,但会模糊真实付费意愿要求拆分补贴试点和市场定价部署

扩张向量真实存在,但多数具名证明仍集中在少数合作伙伴、区域和政策支持试验场。

[CU037, CU038, CU039, CU040, CU042, CU043]

6.5 图表

Chapter 07

07风险

7.1 按严重性排序的风险姿态

GigaAI 在野心和资金获取上异常强,但公开记录仍指向较高剩余风险,而不是已经降险的规模化故事。公司现在把自己定位为全栈实体 AGI 平台,覆盖世界模型、动作模型、数据引擎和原生机器人本体;公开融资报道称其三个月内融资 CNY3.5 billion,并同时瞄准工业和家庭部署。这些确实是优势,却会加重而不是降低执行压力。同一批来源显示,新资金正投向数据系统、模型打磨和机器人扩产,也就是说,在可持续单位经济性可见之前,业务仍依赖昂贵的迭代循环。外部怀疑者的方向也一致:人形机器人需求大体仍是假设,AI 稳健性还没达到市场级可靠性,安全和电池约束仍是门槛。实际来看,头部风险集中在五类:出口管制和算力风险、中国监管和备案风险、产品安全和缺陷责任、商业化和资本依赖,以及集中伙伴或关键人执行风险。因此,本章只在尽调能把公司声称的规模信号转化为合规、可靠性、客户广度和资本效率证据时,才把 GigaAI 视为可投资对象。[CR001, CR002, CR003, CR004, CR009, CR010]

监管 / 法律风险登记表
风险司法辖区 / 规则当前证据可能性严重性缓释成熟度剩余敞口尽调路径
AI 芯片出口收紧美国 BIS 先进计算管制BIS 现在要求向中国总部实体出口某些先进计算产品时,即便不在中国境内也要取得许可;H200 级销售带有条件,政策仍在变化。严重要求提供当前训练和推理硬件计划、获批供应商,以及美国 GPU 访问进一步收窄时的应急方案。
算法备案 / 安全评估中国 CAC 生成式 AI 和算法规则涉及舆论属性或社会动员能力的服务可能触发备案和安全评估义务;生成式 AI 规则还要求训练数据、透明度和标识义务。询问当前或计划中的服务是否已完成备案、安全评估或深度合成标识设计审查。
产品缺陷责任中国产品质量法如果机器人或配件存在缺陷,导致人身伤害或财产损失,生产者可能需要赔偿;警示说明和安全标准合规很关键。获取 SeeLight 和 Maker 线的产品警示、保险、事故响应 SOP,以及任何出货前安全验证结果。
人形机器人标准收紧中国国家人形机器人 / 具身 AI 标准体系中国已经发布覆盖人形机器人全生命周期的国家标准体系框架,包括安全和伦理。对照已公布的标准类别和预期认证路径,梳理 GigaAI 当前产品和软件实践。
劳动关系合规中国劳动 / 劳务派遣规则劳务派遣有比例上限,未签书面合同可能带来双倍工资和无固定期限合同风险;关联实体可能承担连带责任。审查工厂、演示和家庭运营团队中的劳动、劳务派遣、承包商和现场服务人员结构。
开源 / IP 与不正当竞争敞口中国 AI 和竞争义务CAC 规则要求尊重 IP,并禁止基于算法或平台的不正当竞争;与此同时,GigaAI 开源了技术栈中有实质意义的部分。要求提供 IP 策略、贡献者控制、模型许可合规,以及对仿冒商业化或侵权主张的监测。

严重性排序只考虑公开可见缓释之后的剩余敞口。这是一份部分基于公开来源的登记表,不是经律师完整审阅的法律清单。

[CR001, CR002, CR003, CR004, CR005, CR006]
FR001: 风险热力图

最高残余风险集中在外部依赖、未经验证的现场经济模型和上升中的监管义务交叉处。

评级为定性判断,依据已审阅公开证据,而非公司披露的内部 KPI 阈值。

[CR001, CR004, CR009, CR010, CR023, CR024]

7.2 地缘政治和监管风险

最结构性的外部风险在于,GigaAI 正好在监管密度上升的地方扩张。美国侧,BIS 指引现在表明,即便接收方位于中国境外,面向中国总部实体的先进计算出口也可能需要许可证;2025–2026 年的管制姿态扩大了对中国市场设计芯片以及经第三国转运的审查。这一点很重要,因为 GigaAI 公开技术栈明确围绕可扩展训练、多 GPU 执行、FP8 效率、世界模型数据引擎,以及甚至提到 NVIDIA Jetson 的部署路径。中国侧,负担也不再只是笼统的 AI 口号。中国已发布人形机器人与具身智能国家标准体系,CAC 规则也已经对训练数据合法处理、尊重知识产权、内容标识、透明度,以及具有舆论属性或社会动员能力的算法服务备案或安全评估义务提出要求。对 GigaAI 来说,实际问题不是每条规则今天都必然适用;而是家庭机器人、多模态数据采集和模型介导输出,会让公司在产品离开受控演示、进入更像消费者或更宽泛工业环境时,可能碰到其中多项规则。[CR001, CR002, CR003, CR004, CR005, CR006]

伙伴 / 依赖风险登记表
依赖交易对手 / 层级角色集中度失效情景严重性缓释剩余敞口
先进 AI 算力NVIDIA / 美国关联 GPU 栈训练、推理、ROS 加速许可证被拒或第三国执法收紧,会拖慢模型迭代,并推高成本或延迟。严重效率优化、小设备优化、替代硬件
工业验证点FAW Tooling、Longsheng、Hubei 创新中心已点名的部署锚点和信号客户进度延迟、范围降级或生产结果平淡,都会削弱收入叙事和技术可信度。扩大客户基础,并发布更多现场指标
家庭发布窗口SeeLight 早期采用者 / 分销商家庭采用信号订单无法转化为安全的规模化运营,或支持经济性恶化。分阶段推出、收窄用例、投入服务和安全运营
内部开源栈GigaTrain / GigaDatasets / GigaModels / GigaWorld核心数据、训练和部署工具框架 bug、维护拖累或 IP 泄露,会拖慢发布节奏或削弱护城河耐久度。Apache 许可治理、内部 QA、社区反馈
资本提供方近期财团和跟投投资人为模型 / 数据和落地扩张提供资金人形机器人情绪转弱或里程碑放慢,会在经营现金流被验证前推高未来融资成本。更早证明客户转化和毛利率路径

集中度只基于公开点名的交易对手评估。未披露的供应商、合同制造商和云安排仍不透明,因此本表仅作方向判断。

[CR001, CR002, CR003, CR011, CR012, CR013]
FR003: 依赖图

关键外部和内部依赖,可能阻断 GigaAI 从演示走向规模化部署。

具名节点仅反映已审阅公开材料中可见的依赖。

[CR001, CR004, CR011, CR012, CR013, CR016]

7.3 安全、责任和技术泛化风险

GigaAI 最难的产品风险在于,公司试图把三件难事压进同一个商业化周期:安全的具身硬件、由世界模型驱动的泛化,以及经济上可支撑的现场运营。公司自己的论文和代码库坦率承认,真实世界机器人数据采集既贵又耗时,世界模型生成数据本意是缓解这个瓶颈,而技术栈依赖多个内部框架和大量算力。这是一套连贯的技术策略,但不等于已经证明长周期 sim-to-real 迁移被解决。外部来源更谨慎:MIT Technology Review 强调,人形机器人仍缺少常识,采用节奏慢且高度行业化;IEEE Spectrum 则认为市场大多仍是假设,AI 还不够稳健,达不到市场要求,工业买家关心的是电池寿命、安全和可靠性,更接近 99.99% 的水准,而不是炫目的演示质量。如果 GigaAI 的 SeeLight 和 Maker 项目在故障率、服务和事故控制成熟之前进入家庭和工厂,中国产品质量法会带来真实敞口:可能危及人身或财产安全的产品必须符合安全标准、附带警示;缺陷造成伤害或损失时,生产者可能承担责任。[CR017, CR018, CR019, CR020, CR021, CR022]

运营 / 质量 / 安全风险登记表
失效模式证据可能性严重性缓释成熟度剩余敞口未解决缺口
仿真到现实泛化在真实部署中不及预期GigaBrain 和 GigaWorld 明确用合成数据和世界模型生成数据,减少昂贵的真机采集;但外部质疑者仍认为,能达到市场级稳健性这件事还没解决。家庭和工厂场景下,未披露公开的任务级失败率、漂移或重训练成本。
可靠性和停机时间达不到工业要求IEEE Spectrum 指出,工业客户看重 99.99% 可靠性,停机成本可能极高。Maker 机队未披露公开的正常运行时间、MTBF、备件或现场服务 SLA 数据。
电池 / 可维护性约束侵蚀可用劳动产出外部机器人评论认为,更强的人形机器人需要更多电力、更大重量,并承担更多安全取舍。GigaAI 主要机器人产品线未披露公开的续航、充电、维护或换电周期。
家庭或工厂安全事故损害采用轨迹家庭和工业落地会抬高直接接触人的风险;如果产品存在缺陷或警示不足,产品责任法也会带来敞口。严重本轮审阅未看到公开的事故、召回、认证或保险披露。
数据 / 隐私或算法治理失效家庭和工厂机器人会产生多模态数据流;CAC 规则要求数据安全、个人信息和标识管理。截至运行日审阅的官网内容,未看到公开隐私门户、DPA 套件或备案号披露。
仿真 / 工具链依赖老化或误导AirSim 已归档;GigaAI 自身技术栈也依赖多个内部框架和上游工具,而不是单一稳定的外部标准。需要工具路线图、供应商锁定地图,以及核心仿真或加速层变化时能够安全迁移的证据。

各行结合公司披露和外部机器人评论。公开可靠性和事故指标缺位时,剩余敞口仍然偏高。

[CR017, CR018, CR019, CR020, CR021, CR022]
FR002: 风险传导图

技术、法律和融资冲击如何传导到收入节奏、利润率和估值信心。

这张图是定性因果图,不是财务模型。

[CR001, CR003, CR016, CR019, CR021, CR023]

7.4 采用、资本和依赖风险

商业风险较少来自叙事不足,更多来自叙事跑在可重复采用之前。公开来源确实显示真实部署标志:Maker H01 交付给湖北人形机器人创新中心、一汽模具工厂用例、三年 1,000 台机器人龙盛计划,以及约 100 台 SeeLight S1 订单,目标是在 2026 年 Q3 扩量。但同一批披露留下了关键空白。它们没有公布现场可用率、MTBF、事故率、保修经济性、服务人员配置,也没有披露试点转化为经常性生产账户的收入情况。这使伙伴依赖变得重要:少数具名工业关系承担了过大的信号价值,因此任何锚定客户或部署伙伴的滑坡,都可能同时打击市场认知和学习闭环。融资风险同样不低。公司的融资速度令人印象深刻,但披露的募资用途显示,资本仍在为模型、数据和铺开扩张供血。换句话说,从公开经营经济性看,公司还不能自我证明;它更像一个融资充足、正在赛跑以证明经济性的竞争者。在更谨慎的人形机器人市场里,这一区分会变得关键。[CR011, CR012, CR013, CR014, CR015, CR016]

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尽调路径
创始人 / CEO公开的商业和技术叙事仍高度围绕 Huang Guan。扩大可见的运营班底和面向客户的领导层索取组织架构图、董事会构成、继任计划和下放后的 P&L 归属。
首席科学家 / 研究班底Zheng Zhu 是世界模型和具身 AI 叙事的主要可信度节点。留住更广的技术班底,并记录各子系统负责人索取高级技术人员名单、流失历史和研究到产品的交接流程。
合规 / 隐私 / 监管运营公开备案、隐私或信任界面缺位,说明外部很难看清合规能力。增设专职监管和隐私负责人审阅法务、隐私和算法治理职能的人名、汇报线和预算。
现场服务 / QA / 安全进入家庭和工厂规模化,需要支持、事故分诊、备件和安全治理;这些能力不能只靠研究人才。大规模推出前建立正式服务和质量体系要求提供服务人员配置计划、培训项目和事故升级指标。
HR / 劳动结构管理快速扩张和跨实体演示时,派遣、加班和承包商结构可能变成负债。收紧书面合同,并简化实体 / 问责结构按运营实体审计劳动合同、派遣使用、借调和外包。

人员风险来自已审阅公开来源中出现的人物,而不是完整内部花名册。运营职能在公开材料中沉默,只会提高不确定性,并不证明这些职能不存在。

[CR008, CR010, CR020, CR037, CR038, CR039]

7.5 缓释、监控和否决标准

风险图景中较积极的一面是,大多数头部风险都可以监控。出口管制压力可以通过披露的硬件选择、采购灵活性,以及公司是否展示小型设备部署进展——而不仅是前沿规模训练——来跟踪。监管风险可以通过公开备案编号、安全评估、隐私文件,以及家庭产品是否附带清晰警示、条款和事故处理路径来监控。产品风险应由硬现场证据判断:复购客户、场景适配时间下降且未以安全事件为代价、披露可靠性目标,以及家庭机器人能经受真实使用而不是精心管理的展示。资本风险应通过客户广度、服务毛利率,以及新融资是否仍主要用于补贴数据采集和算力来监控。最后,人员风险应通过梯队深度和可见运营成熟度来观察:更强的合规、QA、现场服务和企业支持领导力,会降低外界对信任过度集中在创始人和小型研究核心上的担忧。如果这些信号在下一个产品周期内没有改善,投资论题应从“跟踪”转向明确否决,而不是继续由世界模型热度托着。[CR008, CR010, CR016, CR025, CR033, CR035]

缓释和终止标准表
风险可监测触发项阈值 / 事件行动含义
算力 / 出口管制冲击硬件路线图和采购灵活性有证据显示,关键训练或推理里程碑依赖受控美国 GPU,且没有可行备选暂停承销激进规模假设,并下调路线图时间表。
中国备案 / 合规缺口公开备案号、隐私条款、安全评估更广家庭推出前,没有可信的备案、标识或隐私合规材料在法律准备度得到证据前,将消费者扩张视为受阻。
安全 / 责任事件事故、召回或重大警示缺陷任何严重用户或工人伤害、监管通知,或由缺陷驱动的召回升级到论点破裂审查;重新评估可保性和产品市场时点。
可靠性不足停机时间、MTBF、服务成本规模化试点后,仍没有稳定现场可靠性或经济上可支撑服务模式的证据将工业落地重新归类为试点展示,而不是可持续采用。
资本依赖收入证明相对融资节奏在公开证明客户广度或单位经济性改善前,又完成一轮大额融资假设稀释和时间表风险仍是核心问题,而非临时扰动。
客户集中已点名客户广度和重复使用下个周期后,FAW / Longsheng / Hubei 仍是唯一有意义的公开工业参考下调部署护城河,把订单信号视为集中验证,而不是广泛需求。

公共来源没有披露足够硬运营指标,无法设定完全量化的投资契约,因此这些阈值是定性的。

[CR001, CR002, CR003, CR011, CR012, CR014]
Chapter 08

08估值

8.1 投资论题与反论题

GigaAI 的正向逻辑不难看到。一个季度内,公司完成中国规模的资金储备:CNY1 billion Pre-B、近 CNY1.5 billion B1,以及 CNY1 billion B2 / 战略轮,同时把自己定位为全栈实体 AGI 玩家,覆盖世界模型、具身模型、工业机器人和家庭机器人。这个组合很重要,因为在具身 AI 里,资金获取本身就是护城河:它能支持硬件迭代、数据采集、仿真和制造学习曲线,而较弱同行负担不起。反论题在于,当前价格主要由品类动能和选择权支撑,而不是由已披露经济性支撑。已审阅公开来源仍没有公布 GigaAI 的收入、毛利率、客户集中度、积压订单转化或轮次条款。在这个行业里,这种缺失很关键。支撑最充分的高溢价同行,要么有官方软件平台叙事和全球投资人背书,例如 Figure 和 Physical Intelligence;要么正走向更强的财务披露,如 Unitree 的 IPO 路径。GigaAI 可能成为这些赢家之一,但今天的公开证据还不能证明它已经值得同等承销待遇。[CV001, CV002, CV003, CV004, CV005, CV006]

投资论点 / 反论点表
维度投资论点反论点什么会改变判断
资本获取三个月内 CNY3.5B 可为扩张提供资金资本动能可能跑在商业现实前面证明募资用途绑定已签收入和利润率改善
产品范围全栈 physical-AGI 雄心可能创造平台价值范围过宽可能掩盖薄弱的单位经济性用已披露定价和支持模式证明重复部署
商业证明家庭订单和 1,000 台目标显示需求订单和交付目标弱于已确认收入披露已签收入、积压订单和转化率
同业背景Figure 和 PI 显示投资人愿为具身 AI 支付溢价这些同业是全球叙事领头羊,平台定位更强说明为什么 GigaAI 相对中国同业也应获得类似溢价
披露类 IPO 透明度可以稍后到来今天估值 $1.5B,缺乏披露本身就是估值风险现在提供带日期的收入、毛利率和优先权条款
退出路径IPO 或战略出售仍有可能当前公开文件尚未达到退出准备状态证明规模、盈利路径和治理准备度

反论点主要挑战披露和估值质量,并不是否认 GigaAI 最终可能做成一家强公司。

[CV004, CV005, CV006, CV007, CV008, CV009]
FV001: 推荐逻辑

融资强度、同业背景、缺失披露和估值质量如何共同形成当前推荐。

定性决策树,而非加权模型;缺失披露节点是主导性闸门因素。

[CV004, CV005, CV006, CV032, CV038, CV041]

8.2 建议、信心和进入纪律

我们给予 GigaAI “继续研究”评级,信心为中,风险评级为高,估值姿态为偏紧。这不是看空公司的技术野心,而是基于今天公开可知信息的价格敏感判断。当前估值锚点超过 CNY10 billion,约 $1.5 billion;只有当 GigaAI 能把当前融资动能转化为已披露收入规模、可接受的硬件经济性,以及可获得溢价倍数的软件附着时,这一估值才站得住。支撑这一步跨越的证据仍然缺失。公开报道显示出有希望的商业化标志,包括据称约 100 台家庭机器人订单和 1,000 台年交付目标,但这些锚点仍弱于已报告收入、经常性软件支出或经审计的利润率披露。因此,进入纪律应明确:不要只为叙事支付当前价格;要求有日期的收入、毛利率、积压订单质量和优先权结构披露,或者以明显低于当前标记的价格进入。[CV005, CV006, CV007, CV008, CV033, CV034]

建议摘要表
维度评估当前判断依据
建议继续研究价格领先于已披露经济性
置信度融资事实清楚;收入和利润率不清楚
风险评级商业化和下轮降估值风险仍然重要
估值立场偏高当前估值需要情景成功支撑,不只是试点证明
当前融资锚点>CNY10B / ~$1.5B由 2026 年融资报道支撑,不是由公开收入披露支撑
决策含义先尽调,再谈价格承销前要求提供收入、利润率、积压订单和优先权细节

评估基于情景推演,因为已审阅公开文件未披露 GigaAI 收入、毛利率和融资条款。

[CV005, CV006, CV038, CV039, CV040, CV041]

8.3 融资背景、稀释悬顶和公开文件无法支持的内容

融资背景清楚,经济性却不清楚。第一财经和其他已审阅媒体基本一致认为,GigaAI 在 2026 年约三个月内融资 CNY3.5 billion,包括 3 月 Pre-B、4 月 B1 和 6 月 B2 / 战略轮,投后估值超过 CNY10 billion。这种资本形成规模是真实的,并显著提高了 GigaAI 在昂贵的训练加硬件竞赛中存活的概率。它也带来不同的尽调负担。到了这一阶段,正确的承销问题不再是公司能否融资,而是这些钱换来了多少所有权、清算优先权顺位、老股转让泄漏和 runway 延长。已审阅公开文件没有呈现任何这些条款。收入、积压订单和毛利率未披露,意味着投资人还无法测试最新一轮反映的是牵引、稀缺战略选择权,还是具身 AI 热钱市场。实际操作中,场景估值会占主导,因为常规轮次收入倍数还无法用已披露经营数据扎根。[CV001, CV002, CV003, CV004, CV005, CV006]

8.4 牛、基准和熊市情景

今天给 GigaAI 估值,正确方法是反问:未来需要出现什么披露,当前标记才会显得便宜、合理或明显过高。牛市情景中,GigaAI 把今天的融资爆发转化为真实部署规模:工业订单朝 1,000 台雄心扩张,家庭试点越过当前据称 100 台订单信号,软件或模型层附着在硬件部署上,并产生有意义的毛利率。沿着这条路径,公司可以证明明显高于当前标记的价值。基准情景更克制:商业化发生,但主要以硬件为重的试点和早期生产收入为主,因此已披露规模来得更慢,公司围绕今天轮次交易,或略低于今天轮次。熊市情景是,品类比投资人预期更长时间地保持高热度、低收入;在这种世界里,披露缺失叠加倍数压缩和下一轮偏重优先权,会造成真实 down-round 风险。由于收入未披露,场景框架比虚假的单点目标更诚实。[CV006, CV025, CV028, CV029, CV030, CV033]

牛市 / 基准 / 熊市情景表
情景概率信号核心假设隐含估值区间相对当前估值的 MOIC下行 / 上行触发项
牛市需要披露升级工业部署向 1,000 台目标扩张,家庭试点转化,软件附加提升经济性$2.2B-$3.6B~1.5x-2.4x由 2028 年前披露收入 >$250M 且证明利润率触发
基准按公开证据最可能试点转化,但业务仍偏硬件,披露改善缓慢$0.8B-$1.5B~0.5x-1.0x由收入规模扩大但缺乏高溢价软件经济性触发
熊市必须纳入承销商业化滞后,下一轮融资增加优先权包袱,品类倍数压缩$0.2B-$0.6B~0.1x-0.4x由收入未成规模或积压订单披露疲弱触发

估值区间是作者估算,锚定当前私人市场估值、Unitree 披露和公开机器人倍数;收入未披露,因此情景比精确值更重要。

[CV005, CV025, CV028, CV033, CV034, CV035]
FV002: 估值敏感性

不同未来收入和倍数假设下的示例估值结果,单位为十亿美元。

敏感性分析使用已披露公开可比公司区间作为参照;GigaAI 自身未披露收入,因此这是情景工具,不是预测。

[CV025, CV028, CV033, CV034, CV035, CV036]
FV003: 估值 / 回报区间

各情景下的估值区间,以及相对当前估值标记的隐含回报范围。

区间为作者估算,依据已披露的同行估值标记、上市可比公司倍数,以及 GigaAI 未披露收入这一事实。

[CV005, CV033, CV035, CV036, CV037]

8.5 可比集合:溢价上限、披露锚点和警示性上市可比

可比集合更支持谦逊,而不是确信。溢价侧,Figure 的官方融资路径从 2024 年 $2.6 billion 到 2025 年 $39 billion;Physical Intelligence 则从 $5.6 billion 走向下一轮超过 $11 billion 的报道。这些结果显示,市场有时愿意为具身 AI 平台选择权支付高价。但这两个可比本身都无法拯救 GigaAI 的价格:两者都是全球叙事领导者,且 Physical Intelligence 至少披露了软件订阅逻辑。更有用的锚点是 Unitree,因为它同时提供估值和经营披露。财新报道其 2025 年 6 月估值为 127 亿元、2026 年 IPO 目标为 420 亿元、2025 年收入为 17 亿元、利润为 2.8 亿元。上市机器人可比用另一种方式说明同一点。Symbotic 的市值和申报收入隐含高个位数年化销售倍数;Serve 则说明,收入极小的机器人公司也能在很小基数上打出吸睛倍数。放在这组可比中,GigaAI 当前标记在品类里并不荒谬,但相较已披露同行,证据不足。[CV009, CV010, CV011, CV012, CV013, CV014]

可比估值表
可比公司状态公开估值 / 市值运营指标锚点隐含倍数或信号相关性 / 局限
GigaAI2026 年私募轮>CNY10B / ~$1.5B收入未披露n/a当前标的;估值由融资支撑,不是由已披露经济性支撑
Figure AI私营2025 年投后 $39B收入未披露仅作为溢价上限显示私人具身 AI 热度;直接运营锚点弱
Physical Intelligence私营投后 $5.6B;洽谈中估值 >$11BSacra 称软件模式为每台机器人每月 $300平台溢价可用于模型层可选性,不用于判断硬件执行质量
Unitree私营 + IPO 路径私募估值 CNY12.7B;IPO 目标 CNY42B2025 年收入 CNY1.7B;利润 CNY280M按 IPO 目标约为 2025 年销售额 3.6x中国披露锚点最佳;与 GigaAI 的不透明度不能直接对比
AgiBot私营已审阅免费来源未清晰披露估值2025 年出货 5,100 台;已披露融资 $83M+商业化信号可证明出货规模,但没有清晰估值锚点
EngineAI私营估值未披露累计融资 CNY1B;T800 已量产融资 / 制造信号可用于判断中国同业强度,不用于倍数测算
Symbotic上市市值 $23.28B2026 财年 Q2 收入 $676M(年化约 $2.7B)年化销售额约 8.6x公开披露工业机器人锚点最佳
Serve Robotics上市市值 $0.54B2026 年展望约 $26M;2025 年收入 $2.7M约为 2026 年展望 20.8x显示早期上市公司倍数可在极小收入基数上走到极端

覆盖范围有限,因为多家私人同业未在免费公开来源披露收入或估值;列出倍数处,均为基于引用公开数据的近似计算。

[CV005, CV009, CV012, CV014, CV015, CV016]

8.6 退出准备度、论题破裂触发点和最终尽调问题

公开证据不支持今天把 GigaAI 视为已具备退出准备。公司显然拥有融资动能和品类位置,未来可能成为 IPO 或战略收购候选,但披露文件仍太薄,无法让这条路径被有信心地承销。最重要的论题破裂点很简单:如果带日期的收入、积压订单或毛利率披露出现,并显示业务规模远低于支撑 $1.5 billion 以上标记所需水平,估值逻辑会迅速塌陷。紧密相关的触发点包括:带有重度 senior preferences 的 down-round、订单信号主要是试点而非合同需求的证据,或硬件利用率提升慢于投资人假设的证明。因此,最终尽调应聚焦 2026 年确切已签收入、硬件与软件或服务分项毛利率、客户集中度、已签积压订单与取消权、2026 年融资后的 burn 和 runway,以及优先权结构。在这些问题得到回答之前,审慎的 IC 姿态是把 GigaAI 留在积极覆盖中,而不是把叙事兴奋转化为对价格不敏感的承诺。[CV006, CV038, CV039, CV040, CV041, CV042]

论点破裂和终止触发项表
触发项阈值 / 事件对投资论点的传导行动含义
收入披露未达到规模要求公开已签收入显著低于当前估值所需水平打破溢价倍数逻辑回避入场或重定价
毛利率只像硬件看不到软件挂载或利润率抬升证据公司估值会更接近低质量硬件倍数下调公允价值区间
在手订单质量弱试点、LOI 或可取消订单占主导削弱出货叙事可信度将增长说法视作宣传
下一轮带来沉重优先权包袱出现高级条款或大额老股退出即使账面估值守住,普通股上行也会受损重新测算投后经济性
商业化节奏滑坡1,000 台目标和家庭场景转化停滞把 GigaAI 拉向熊市情景价值带暂缓承诺,之后复盘

触发项聚焦估值质量,而不只看技术雄心;本章的核心决策变量就是价格敏感度。

[CV006, CV007, CV008, CV042, CV045]
最终尽调问题表
主题缺失证据重要性负责人 / 尽调路径
2026 年已确认收入按硬件、软件和服务拆分的季度及年化收入桥为任何倍数估值提供分母公司 CFO / 管理层资料室
毛利率按产品线和服务挂载拆分毛利率判断 GigaAI 应拿溢价,还是该按硬件折价财务尽调和客户访谈
在手订单质量已签订单、取消权、试点转量产转化区分宣传声量和合同需求销售运营尽调和合同审查
客户集中度头部客户组合和复购证据检验耐久性和议价权商业尽调和队列复盘
轮次经济性优先权堆栈、清偿顺位、期权池刷新,以及任何老股成分决定账面估值之后普通股还有多少上行法务尽调和股权表审查
烧钱和跑道2026 年各轮之后的现金消耗、资本开支需求和跑道检验在证据追上之前再次融资的概率财务模型复盘

要从情景权重很重的判断走向能落价格的投资决定,这些问题是最低信息包。

[CV006, CV033, CV042, CV045, CV046]
FV004: 投资 KPI

面向投委会的打分,聚焦对 GigaAI 价格敏感型决策最关键的变量。

打分是给委员会使用的定性辅助,不是机械模型;重点看估值质量和披露就绪度。

[CV004, CV005, CV006, CV029, CV032, CV038]

免责声明

本报告是基于公开来源的尽调简报,不构成投资建议。GigaAI 的许多关键承销输入——包括收入、利润率、员工人数、股权结构机制和客户集中度数据——仍属非公开信息,或仅有间接报道;因此,在完成一手尽调前,相关结论都应视为暂定。

证据索引

结论
编号陈述可信度来源
CO001 The active company website is gigaai.cc and the retrieved page title is “极佳科技.” SO001
CO002 The English-brand domain gigaai.com is a parked domain listed for sale rather than a live corporate site. SO002
CO003 Reviewed public coverage identifies the operating company as Beijing 极佳视界科技有限公司 / GigaAI and describes it as Beijing-based. SO003, SO011, SO029
CO004 The best-supported founding year for GigaAI is 2023. SO003, SO007, SO011
CO005 Huang Guan is publicly identified as GigaAI’s founder and chief executive officer. SO003, SO005, SO018
CO006 Public profiles say Huang Guan previously worked at Horizon Robotics and also held roles linked to Tsinghua, Microsoft Research Asia, and Samsung China Research. SO003, SO006, SO018
CO007 Zheng Zhu’s own homepage identifies him as GigaAI’s co-founder and chief scientist. SO028
CO008 Investor media identifies Sun Shaoyan as a co-founder with prior Alibaba Cloud and Horizon product experience. SO006, SO013
CO009 Investor media identifies Mao Jiming as an engineering leader with Baidu Apollo simulation experience. SO006, SO013
CO010 Reviewed public sources do not clearly disclose GigaAI’s board composition, voting control, or protective investor rights. SO003, SO014, SO015
CO011 GigaAI is repeatedly described as a four-part stack of embodied foundation model, world model, native body, and application scenarios. SO004, SO005, SO009
CO012 Public materials describe GigaWorld as GigaAI’s world-generation or world-model platform. SO003, SO020, SO026
CO013 Public materials describe GigaBrain as GigaAI’s embodied action or foundation-model layer. SO003, SO020, SO026
CO014 Maker H01 is presented as GigaAI’s industrial or general-purpose native robot body. SO003, SO008, SO022
CO015 SeeLight S1 is presented as GigaAI’s household robot product line. SO003, SO012, SO021
CO016 The public business model combines foundation-model capability delivered through software, API, or licensing with physical robot products and deployments. SO004, SO011
CO017 Reviewed public materials do not disclose GigaAI revenue or ARR. SO003, SO015, SO017
CO018 Reviewed public materials do not disclose current GigaAI headcount. SO003, SO015, SO017
CO019 GigaAI’s seed round was reportedly backed by Chentao Capital. SO013, SO029
CO020 Public coverage says GigaAI raised nearly RMB 50 million across angel and angel+ rounds in September 2024. SO014, SO029
CO021 Public coverage says GigaAI completed Pre-A and Pre-A+ rounds in 2025 for hundreds of millions of renminbi. SO014, SO015
CO022 Public coverage says Huawei Hubble and Huakong Fund invested in GigaAI’s 2025 A1 round. SO014, SO015
CO023 Public coverage says GigaAI’s 2025 A2 round raised RMB 200 million and was led by Fortune Capital with Huakong as a co-lead backer. SO014, SO015
CO024 Reviewed sources agree that GigaAI completed a nearly RMB 1 billion Pre-B round in March 2026. SO005, SO009, SO018
CO025 Reviewed sources agree that GigaAI completed a nearly RMB 1.5 billion B1 round in April 2026. SO003, SO007, SO014
CO026 Reviewed sources agree that GigaAI completed a RMB 1 billion B2 round in June 2026. SO003, SO004, SO016
CO027 Yicai reported that GigaAI raised RMB 3.5 billion, about USD 518 million, across three months in 2026. SO003, SO015
CO028 Multiple reviewed outlets reported that GigaAI’s post-financing valuation exceeded RMB 10 billion. SO003, SO007, SO014
CO029 Crunchbase recorded GigaAI as a Beijing-based company valued at USD 1.5 billion after an April 2026 Series B. SO010
CO030 Named B2 investors included Lion Partners Capital, the China-Belgium Direct Equity Investment Fund, and Wanxiang Qianchao. SO003, SO016
CO031 Some B1 coverage still described one key participant only as a “well-known tech giant,” leaving the full investor map partially undisclosed. SO007, SO014
CO032 Public sources say B2 proceeds are earmarked for data and algorithm systems, the physical AGI foundation model, and scaling robot products in household and industrial settings. SO003, SO016
CO033 Public 2026 materials target 1 million hours of vision-action data and 10 million hours of world-model pretraining data by year-end. SO004, SO011
CO034 Gasgoo reported that GigaAI began Maker H01 deliveries in early 2026 and sent the first unit to the Hubei Humanoid Robot Innovation Center. SO008
CO035 Yicai and other outlet coverage said SeeLight S1 had about 100 orders and planned broader delivery or operation in the third quarter of 2026. SO003, SO012, SO021
CO036 Reviewed sources said GigaAI announced a plan to deploy 1,000 Maker-series robots with Longsheng Technology over three years. SO003, SO004
CO037 Reviewed sources said GigaAI, FAW Tooling, and Alibaba Cloud announced a real manufacturing deployment in April 2026. SO003, SO004, SO011
CO038 DriveDreamer was publicly presented as a real-world-driven autonomous-driving world model released in September 2023. SO019, SO024
CO039 The GigaWorld-0 arXiv paper says generated world-model data improved physical-robot generalization without any real-world interaction during training. SO026, SO027
CO040 GigaAI maintains a public research footprint through a GitHub organization and open repositories for DriveDreamer, DriveDreamer4D, and GigaWorld-0. SO023, SO024, SO025, SO026
CO041 The public milestone path runs from world-model research into embodied models and then into robot deliveries and factory or household deployments. SO019, SO027, SO008, SO012
CO042 Tencent News commentary warned that valuation inflation, robot-brain capability, and commercial closed-loop execution remain open tests for the world-model wave. SO017
CO043 The split between the active .cc website and the parked .com domain is a real diligence note for English-language counterparties. SO001, SO002
CO044 Public materials are materially stronger on fundraising and roadmap claims than on board, revenue, ARR, customer-count, or headcount disclosure. SO003, SO015, SO017
CO045 The best-supported stage description is that GigaAI is a late-stage private embodied-AI unicorn after its 2026 B2 financing. SO003, SO010, SO016
CO046 Zheng Zhu’s homepage says he participated in financings above RMB 3 billion while GigaAI’s valuation exceeded RMB 10 billion. SO028
CM001 The relevant market for GigaAI should include deployable robotics software, perception, simulation, and adaptation budgets tied to physical workflows rather than all AI software. SM004, SM009, SM015
CM002 Status-quo substitutes for many embodied-AI jobs are fixed industrial robots, wheeled mobile robots, and human labor rather than general-purpose humanoids. SM002, SM018, SM020
CM003 Global robot density reached 162 industrial robots per 10,000 manufacturing employees in 2023. SM001
CM004 China’s robot density rose to 470 robots per 10,000 manufacturing employees in 2023 from 402 in 2022. SM001
CM005 South Korea reached 1,012 robots per 10,000 manufacturing employees in 2023, showing the global frontier of factory automation intensity. SM001
CM006 China already holds around 2 million industrial robots in operational stock and about 54% of global annual industrial robot installations. SM002, SM003
CM007 China’s 15th Five-Year Plan pushes AI toward physical applications and treats robotics as a strategic emerging industry. SM002, SM006
CM008 IFR argues universal humanoid factory helpers are unlikely to be mass-adopted in the near or medium term, with commercialization arriving closer to the end of the 2026-2030 plan period. SM002
CM009 Local Chinese suppliers increased their share of domestic industrial robot installations from 30% in 2020 to 57% in 2024. SM002
CM010 In electronics, 64% of global industrial robots are installed in China and Chinese suppliers provide 59% of that domestic industry. SM002
CM011 TBRC sizes the broad embodied-AI market at $3.22 billion in 2025, $3.8 billion in 2026, and $7.24 billion in 2030. SM009
CM012 TBRC’s embodied-AI category includes robots, exoskeletons, autonomous systems, and smart appliances, so it is broader than humanoids and narrower than generic AI software. SM009
CM013 MarketsandMarkets projects the U.S. humanoid robot market to rise from $857.9 million in 2025 to $4.601 billion in 2030. SM008
CM014 MarketsandMarkets projects South Korea’s humanoid market from $112.6 million in 2025 to $583.5 million in 2030. SM008
CM015 MarketsandMarkets identifies Asia Pacific as the fastest-growing humanoid region because of government-backed automation and component depth. SM008
CM016 Goldman Sachs framed a conservative humanoid market of roughly $6 billion over the next 10 to 15 years, but a blue-sky scenario of up to $154 billion by 2035. SM013
CM017 Morgan Stanley’s 2024 view projected a global humanoid population of 40,000 by 2030, 8 million by 2040, and 63 million by 2050. SM012
CM018 Morgan Stanley’s 2025 view projected $4.7 trillion in global humanoid revenue and roughly 1 billion humanoid units by 2050. SM011
CM019 Morgan Stanley says China’s manufacturing depth and government support make Chinese component suppliers major beneficiaries of the humanoid value chain. SM011, SM012
CM020 Bank of America projects annual humanoid shipments from 20,000 units in 2025 to 90,000 in 2026, 1.2 million in 2030, and 10 million in 2035. SM024
CM021 Bank of America projects a humanoid installed base of 300 million by 2040 and 3 billion by 2060, with household robots eventually representing 62% of units. SM024
CM022 BCG says current physical-AI value is concentrated in Levels 2 and 3—visual perception and dexterous manipulation—rather than in Level 5 reasoning. SM004
CM023 BCG estimates roughly 75% of traditional robotics total cost of ownership sits in initial setup and reengineering, and software-defined approaches can cut those setup and reengineering costs by up to 50%. SM004
CM024 BCG argues Level 5 reasoning with internal world models remains largely aspirational and is the gating constraint for truly general-purpose robotics. SM004
CM025 DeepMind says Gemini Robotics more than doubled performance on its generalization benchmark relative to prior vision-language-action models. SM005
CM026 DeepMind says Gemini Robotics-ER achieved a 2x-3x success rate improvement over Gemini 2.0 in end-to-end robot control settings. SM005
CM027 NVIDIA positions Isaac ROS as an open ROS 2 stack with GPU-accelerated perception, localization, mapping, and motion-planning modules for robotics developers. SM021
CM028 Microsoft says AirSim was built to generate large robotics training datasets and to debug systems in simulation because real-world experimentation is costly and unpredictable. SM022
CM029 WEF’s industrial-operations white paper says rule-based, training-based, and context-based robotics will coexist in real deployments. SM015
CM030 WEF says physical AI is easiest to apply where warehouses already have automated storage, retrieval, and robotics infrastructure. SM016
CM031 WEF argues physical-AI deployment should follow a trust ramp from analysis to supervised autonomy and often be installed in phases to spread capex and operational disruption. SM016
CM032 WEF’s Future of Jobs framing says robotics and autonomous systems are expected to transform operations at 58% of employers by 2030. SM014
CM033 WEF describes physical AI as a demographic necessity in aging economies, with Europe’s 65+ share rising from 21% today to 29% by 2050 and Japan already above 29%. SM014
CM034 CGTN reports Beijing launched a three-year plan to build an embodied-AI industrial ecosystem by 2027 and hosts more than 2,500 AI enterprises. SM007
CM035 CGTN reports a Beijing sorting robot can process up to 1,200 parcels per hour and its vendor has roughly 1,000 unit orders across more than 10 logistics centers, indicating logistics is an early commercialization path. SM007
CM036 Berkeley roboticist Ken Goldberg argues a 100,000-year data gap makes claims of humanoids becoming common in the next two to ten years unrealistic. SM023
CM037 Goldberg says simulation can help locomotion and acrobatics, but sim-to-real transfer for dexterous work like construction, kitchen tasks, or factory hand work remains elusive. SM023
CM038 IEEE says demand, battery life, reliability, and safety remain unresolved gating factors for scaling humanoids, and some industrial customers expect 99.99% reliability. SM018
CM039 IEEE quotes a robotics operator saying few customers have yet identified applications requiring several thousand humanoids per facility, making demand discovery a bottleneck. SM018
CM040 IEEE and Gill Pratt argue current AI advances improve robot brains more than bodies, and flat factories often suit wheeled systems better than legged humanoids. SM020
CM041 The HBS-BCG study found AI users finished some in-scope tasks 25% faster and 40% higher in quality, but were 19 percentage points less likely to get complex out-of-frontier tasks correct. SM010
CM042 Goldman said early humanoid applications are likeliest in factories, and economic viability depends on costs falling toward roughly a minimum-wage worker’s two-year salary. SM013
CM043 Bank of America identifies AI maturity, falling hardware costs, EV/autonomy supply-chain overlap, labor shortages, and humanoid compatibility with human environments as key adoption drivers. SM024
CM044 GigaAI’s near-term serviceable market is narrower than generic robotics TAM and is best approximated as budgets attached to industrial automation, logistics, and embodied-AI infrastructure where deployment loops already exist. SM004, SM016, SM021, SM024
CM045 The most credible direct buyers for world-model infrastructure are robotics OEMs, integrators, and large operators that own deployment tooling, changeover pain, and retraining workflows. SM004, SM016, SM021, SM022
CM046 Logistics is ahead of household humanoids because it already has data-rich infrastructure, measurable throughput KPIs, and tolerated phased rollouts. SM007, SM016, SM018
CM047 China is unusually important to embodied-AI commercialization because it combines policy sponsorship, component depth, domestic robot demand, and rising local supplier share. SM002, SM006, SM008
CM048 Contradictory market estimates reflect fundamentally different market boundaries—embodied AI, humanoids, industrial automation, and household robotics—rather than a single consensus TAM waiting to be averaged. SM008, SM009, SM011, SM013, SM024
CM049 IEEE survey evidence suggests people generally prefer special-purpose home robots over humanoids and treat safety, privacy, and price as major conditions for adoption. SM019
CM050 IEEE reported Agility’s Oregon factory is designed for over 10,000 Digit robots per year, showing supply-side manufacturing capacity can scale faster than proven end-market demand. SM018, SM025
CM051 WEF treats robots as part of AI’s next enterprise frontier, reinforcing that embodied AI should be evaluated as an applied compute and operations layer rather than as a standalone consumer gadget market. SM017, SM021
CP101 GigaAI is competing across both household and industrial embodied-robot workflows rather than only one humanoid niche. SP102, SP103, SP104
CP102 SeeLight S1 is positioned as a home robot and has publicly reported early demand of about 100 orders in Wuhan-linked deployment context. SP102, SP104
CP103 Maker H01 is a wheeled humanoid with dual 7-DOF arms and a maximum payload of about 5 kilograms per arm. SP103
CP104 Yicai reported that GigaAI raised CNY3.5 billion over three months in 2026 and said the company’s post-financing valuation exceeded CNY10 billion. SP102
CP105 GigaAI said the new capital would fund data and algorithm systems, its physical AGI foundation model, and scaling of household and industrial robot products. SP102
CP106 Publicly named GigaAI commercial proofs include Maker H01 deployments with FAW Tooling Die Technology and a three-year 1,000-unit framework with Longsheng Technology. SP102
CP107 EngineAI’s Shenzhen base is presented as a move into 10,000-unit delivery capability for T800 humanoids. SP106, SP107
CP108 EngineAI says each robot must pass 79 quality inspections and 46 simulation tests before delivery. SP106, SP107
CP109 Independent 2026 coverage says EngineAI raised about $200 million in Series B and pushed valuation past RMB10 billion with Luxshare joining the cap table. SP107, SP108
CP110 Public EngineAI coverage describes a broader line-up of T800, PM01, SA02, and JS01 products than GigaAI’s currently public two-product narrative. SP107
CP111 AgiBot’s official site emphasizes AGIBOT World and commercial mass production as core parts of its competitive story. SP109
CP112 AgiBot’s public GitHub organization shows active repositories for Omnihand, VLA work, simulation, and challenge baselines, signaling a broader developer surface than GigaAI’s fetched public materials show. SP110
CP113 Unitree’s official G1 page lists a public humanoid price of US$13.5K and describes a 23-to-43-DOF configuration range. SP112
CP114 CNBC reported that Nvidia selected Unitree for its first researcher robotics system, extending Unitree’s distribution signal beyond its own storefront. SP113
CP115 Unitree’s official site frames the company as an established global seller of high-performance robots rather than a pilot-only startup. SP111
CP116 Figure 03 is officially described as a general-purpose humanoid designed for Helix, the home, and world-scale deployment. SP114, SP116
CP117 Figure says it rebuilt supply chain and manufacturing processes for Figure 03 and that BotQ’s first-generation line can initially produce up to 12,000 robots per year. SP116
CP118 Figure’s news index publicly references both a BMW production milestone of 30,000 cars and a Series C funding announcement above $1 billion at a $39 billion post-money valuation. SP115
CP119 Physical Intelligence publicly positions itself as a general-purpose model layer intended to control any robot for any task. SP117
CP120 The Robot Report says Physical Intelligence raised $600 million in late 2025 to gather more data, expand partnerships, and grow its team. SP118
CP121 TechCrunch reported in March 2026 that Physical Intelligence was discussing another roughly $1 billion raise at a valuation above $11 billion while still having no commercialization timeline. SP119
CP122 FANUC, KUKA, and ABB each market broad automation portfolios spanning multiple industrial tasks that overlap with work humanoids aspire to automate. SP120, SP121, SP122
CP123 ABB explicitly highlights a broad service network while FANUC and KUKA emphasize catalog breadth and application coverage, reinforcing incumbents’ procurement advantage over novel humanoids. SP120, SP121, SP122
CP124 For many factory tasks, established fixed or collaborative automation remains a lower-risk substitute than buying a general-purpose humanoid robot. SP120, SP121, SP122
CP125 GigaAI’s wheeled architecture likely improves indoor stability and safety but is structurally less suited to stairs or uneven terrain than fully legged humanoids. SP103, SP104, SP116
CP126 GigaAI’s public strategy more closely resembles a hardware-plus-model hybrid than a pure model-layer business such as Physical Intelligence. SP102, SP117
CP127 Among the fetched primary peer sources, Unitree is the clearest public price anchor because its official G1 page shows a list price while GigaAI, EngineAI, AgiBot, Figure, and Physical Intelligence do not publish equivalent list pricing in the fetched set. SP103, SP105, SP109, SP112, SP114, SP117
CP128 GigaAI’s public distribution proof is concentrated in named pilots and orders rather than a broad public retailer, integrator, or developer channel. SP101, SP102, SP104
CP129 EngineAI’s Luxshare-linked financing and manufacturing-base disclosures indicate stronger public supply-chain leverage than GigaAI has documented so far. SP106, SP107, SP108, SP102
CP130 AgiBot’s dataset and GitHub footprint suggest a more visible public data and developer flywheel than GigaAI has shown in fetched sources. SP109, SP110, SP101
CP131 Figure’s combination of Helix, tactile hands, home-safe design, and BotQ manufacturing shows a deeper public full-stack integration story than GigaAI currently documents. SP114, SP115, SP116, SP102
CP132 If robot-agnostic model layers improve quickly, hardware vendors such as GigaAI could face faster commoditization of embodied-intelligence features. SP117, SP118, SP119
CP133 Switching costs in early humanoid deployments are more likely to come from workflow integration, safety validation, and collected task data than from current hardware uniqueness alone. SP102, SP117, SP122
CP134 Because commercialization remains early and timelines are unsettled, multi-homing across robot vendors is still feasible for many pilot customers. SP119, SP123, SP124, SP125
CP135 Berkeley researchers argue that useful dexterous humanoid work is unlikely within the next two, five, or even ten years because of a large data gap. SP123
CP136 MIT Technology Review reports that humanoids still lack common sense and that adoption is likely to be drawn out, industry specific, and slow. SP124
CP137 KrASIA reports that investors increasingly describe embodied AI as fuzzy, commercially unclear, and at risk of bubble dynamics despite rising valuations. SP125
CP138 GigaAI’s moat appears real but medium-durability at best because peers each own a stronger public advantage in pricing, openness, manufacturing scale, or incumbent channel depth. SP102, SP108, SP110, SP112, SP116, SP122, SP125
CP139 GigaAI’s current commercialization posture is pilot-first rather than retail-first because public materials emphasize orders, deployments, and next-quarter delivery rather than a storefront price. SP102, SP104, SP101
CP140 EngineAI’s public package is manufacturing- and scale-led rather than price-led because fetched sources emphasize factory throughput, QA, and funding rather than list pricing. SP106, SP107, SP108
CP141 Figure’s public package is capability- and manufacturing-led rather than list-price-led because official sources emphasize Helix, home use, and BotQ without public unit pricing. SP114, SP115, SP116
CP142 Physical Intelligence is competing on software runtime and robot foundation models rather than on a named robot unit sold with a public bill of materials. SP117, SP118, SP119
CP143 ABB’s claim of the broadest service network in the industry highlights why incumbent service capacity can outweigh embodied-AI novelty in procurement decisions. SP122, SP120, SP121
CI001 Aiqicha shows that GigaAI’s operating entity is scoped for AI software, industry system integration, robot sales and manufacturing, installation and maintenance, equipment leasing, and import-export activity. SI002
CI002 Aiqicha lists registered capital of RMB3.336289 million and paid-in capital of RMB2.132131 million for Beijing 极佳视界科技有限公司, but those registry figures are not evidence of current operating liquidity. SI002
CI003 Public reporting says GigaAI raised about RMB3.5 billion over roughly three months in March through June 2026. SI003, SI009
CI004 Multiple June 2026 reports place GigaAI’s post-financing valuation above RMB10 billion. SI003, SI009
CI005 Recent-round proceeds are publicly earmarked for continued data-and-algorithm investment, foundation-model iteration, and scaling household and industrial deployments. SI003, SI007
CI006 NCSTI describes GigaAI as having two commercialization lanes: foundation-model services delivered via software, API, and licensing, and native AGI products represented by the SeeLight and Maker lines. SI007
CI007 QQ reports that the DriveDreamer branch serves more than 30 automaker and autonomous-driving customers through signed or mass-production cooperation. SI009
CI008 Yicai reports that SeeLight S1 had about 100 orders before mass delivery began. SI003
CI009 Yicai says GigaAI signed a three-year agreement with Longsheng Technology to deploy 1,000 Maker-series robots in industrial settings. SI003
CI010 Gasgoo reports that the first Maker H01 shipment went to the Hubei Humanoid Robot Innovation Center. SI004
CI011 Gasgoo says GigaAI planned to deliver 1,000 native-body units during 2026 across multiple scenarios. SI005
CI012 GigaAI’s reviewed official surface does not publish public list pricing, self-serve checkout, or plan-tier pricing for Maker, SeeLight, or model services. SI001
CI013 Unitree publicly lists a US$13.5K pre-tax, pre-shipping price for the G1 on its official product page. SI010
CI014 Unitree’s storefront adds shipping, customs, and contact-sales logic for customized EDU versions even when a base humanoid price is public. SI011
CI015 The Unitree example shows that even transparent humanoid vendors still rely on quote-style sales for configured research versions, so public list pricing does not eliminate contract complexity. SI010, SI011
CI016 Humanoid.guide shows an external, unverified Maker H01 price anchor of roughly US$160,000. SI012
CI017 RoboActu reports that SeeLight S1 pilot homes get the robot free and that the hardware price target is below RMB100,000 by June 2027. SI014
CI018 IFR’s 2025 service-robot summary says robot-as-a-service fleets grew 31%, indicating subscription or rental models are increasingly used to reduce customer upfront capex. SI022
CI019 IFR’s 2025 service-robot summary says almost 20 million consumer service robots were sold in 2024, showing large demand in home-service categories even though price and business models differ by segment. SI022
CI020 An external 2026 industrial-robot cost guide places 6-axis arms at roughly USD25,000 to USD180,000, fully deployed cells at USD80,000 to USD400,000, and seven-year TCO at 1.8x to 2.5x initial capex. SI020
CI021 Standard Bots says ABB robots are generally sold through distributor quotes that vary by configuration and that a fully equipped heavy-duty setup can exceed USD120,000. SI021
CI022 ABB reported 2025 capital expenditures of about USD1.001 billion and R&D investment of about USD1.318 billion, underscoring how capital intensive industrial-automation scale can remain even for mature players. SI016, SI025
CI023 ABB says its Robotics division generated about USD2.3 billion of 2025 revenue and includes field services, spare parts, and digital services alongside robot systems. SI016
CI024 UBTECH’s 2025 annual results show RMB2.001 billion of revenue, RMB820.6 million from full-size embodied humanoid products and services, 37.7% gross margin, and a RMB789.8 million loss. SI017
CI025 UBTECH also says it reached annualized production capacity above 6,000 humanoids and completed 1,000-unit-level small-scale mass production and delivery, implying sizeable manufacturing investment even during commercialization. SI017
CI026 iRobot’s 2024 annual report says revenue fell 23.4% to USD681.8 million and that the company cut about 40% of its workforce as part of a restructuring to realign cost structure with near-term demand. SI019
CI027 iRobot’s filing says operating losses, negative operating cash flow, inventory pressure, and gross-margin deterioration created substantial going-concern risk, showing how home-robot economics can break under weak demand and promotions. SI019
CI028 GigaAI’s public commercialization story is currently hardware-led in both industrial and household scenarios, even though software and simulator monetization are explicitly described as future or parallel lanes. SI003, SI007, SI009
CI029 Public GigaAI traction disclosures emphasize orders, deliveries, and deployment frameworks rather than recognized revenue, ARR, or gross margin. SI003, SI004, SI009
CI030 No reviewed public source disclosed GigaAI revenue, ARR, gross margin, cash on hand, monthly burn, runway months, or debt obligations. SI001, SI003, SI006, SI007, SI009
CI031 Because Aiqicha includes equipment leasing and installation within the business scope, robot-rental or service-subscription monetization is legally plausible even though active public GigaAI RaaS contracts were not found. SI002, SI022
CI032 Industrial-robot realized pricing is usually project-specific because hardware cost, tooling, safety, and integration can materially exceed the robot sticker price. SI020, SI021
CI033 Foundation-model services and DriveDreamer infrastructure could become higher-margin software-like revenue streams, but public pricing, contract value, and renewal data are still undisclosed. SI007, SI009, SI023
CI034 The stated use of funds makes clear that GigaAI is still deploying capital into model development and scale-out rather than harvesting mature free cash flow. SI003, SI007
CI035 GigaAI’s near-term capital adequacy looks stronger after the 2026 fundraise, but adequacy cannot be underwritten without disclosed burn, cash, and runway. SI003, SI009
CI036 The next round should be justified by paid conversion from current orders and frameworks, repeat simulator or software contracts, and some disclosed margin evidence rather than by another headline valuation step-up alone. SI003, SI007, SI009
CI037 If the unverified Maker H01 price anchor is directionally right, a 1,000-unit industrial framework could imply large GMV potential, but recognized revenue timing, discounts, and service obligations remain unknown. SI009, SI012
CI038 GigaAI’s financial verdict is therefore positive on monetization breadth and traction, but negative on disclosure quality and still dependent on future financing and conversion proof. SI003, SI007, SI009, SI019
CI039 The public DriveDreamer GitHub repository supports the view that the driving stack is a standalone infrastructure asset, but open code is not evidence of paid contract value. SI023
CI040 Public industrial-robot incumbents such as ABB and FANUC foreground investor-reporting surfaces rather than public robot SKU checkout, which is consistent with enterprise quote-based selling being the norm in this category. SI021, SI024, SI025
CE001 The official site exposes Maker H01, GigaBrain, GigaWorld, and DriveDreamer as primary product routes. SE002
CE002 As of 2026-06-24, the official site navigation does not surface separate public trust, security, status, or support routes. SE001, SE002
CE003 Official product copy describes Maker H01 as a physical-AGI-native body with dual arms and a mobile base for home, commercial-service, and light-industrial scenes. SE002, SE008
CE004 GigaBrain is positioned as Maker H01’s brain and as the layer that provides end-to-end decision control. SE002, SE006
CE005 GigaBrain is described as taking image, point-cloud, text, and embodiment-state inputs and outputting structured task and motion plans. SE002, SE006, SE011
CE006 GigaWorld is framed as a world-model framework built specifically for VLA training and data-efficient embodied learning. SE002, SE012, SE019
CE007 GigaWorld-0 is presented as an open-source system for generating high-generalization interaction data and for powering closed-loop simulation and reinforcement-learning training. SE002, SE012, SE019
CE008 The disclosed Dual Pyramid data stack contains five layers: internet video data, human data, world-model simulators, synthetic simulation data, and real-robot data. SE003, SE007, SE009
CE009 The disclosed algorithm pyramid contains three layers: world simulation, action alignment, and experience reinforcement. SE003, SE007
CE010 GigaAI says the data stack spans SeeLight S1, Maker M01, U-01, E-01, and GigaWorld-0 rather than relying on a single robot platform. SE003, SE007, SE009
CE011 GigaAI says its data program targets 1 million cumulative training hours by the end of 2026. SE003, SE007
CE012 GigaBrain-0.1 is publicly claimed to rank first on RoboChallenge with a 51.67% average success rate as of 2026-02-09. SE023, SE011, SE007
CE013 GigaBrain-0.5 is described as being pretrained on 10,931 hours of robot data, 61% of which is synthesized by GigaWorld. SE023
CE014 RAMP is presented as a world-model-conditioned reinforcement-learning loop that alternates future prediction, conditioned policy updates, human-in-the-loop rollouts, and joint refinement. SE023
CE015 GigaBrain-0.5M* reports roughly 30-point gains over RECAP on difficult tasks such as box packing and espresso preparation. SE023
CE016 GigaBrain-0.5M* is reported to achieve near-perfect success on box packing, espresso preparation, and laundry folding across consecutive runs. SE023, SE002
CE017 The GigaBrain repository includes an inference server, a generic client, and an AgileX-specific client, implying a deployable client-server inference path rather than paper-only code. SE011
CE018 GigaBrain explicitly depends on GigaTrain, GigaDatasets, and GigaModels as its training, data, and model foundations. SE011, SE013, SE014, SE015
CE019 GigaTrain exposes distributed training modes including ZeRO, FSDP, and DDP together with mixed precision and resumable checkpointing. SE014
CE020 GigaDatasets supports curation, loading, evaluation, and visualization across images, videos, 2D and 3D boxes, 2D and 3D points, and LeRobot datasets. SE015
CE021 GigaModels positions itself as a shared infrastructure layer for training, inference, deployment, and compression across VLA, diffusion, and vision pipelines. SE013
CE022 GigaWorld-0 combines a video-generation layer with a 3D stack that includes 3D Gaussian Splatting, differentiable system identification, and executable motion planning. SE012, SE019
CE023 The GigaWorld repository exposes downloadable pretrained models plus training and inference scripts, indicating that external developers can reproduce parts of the data-engine workflow. SE012
CE024 The public GitHub organization page shows notable developer traction with roughly 2.5k stars for GigaBrain-0, 1.6k for GigaWorld-0, 1.3k for GigaWorld-Policy, 1.1k for GigaTrain, and 641 for GigaDatasets. SE010
CE025 DriveDreamer is positioned as a real-world-driven autonomous-driving world model trained with a two-stage pipeline on nuScenes. SE016, SE020
CE026 DriveDreamer4D uses world-model priors plus cousin-data training to improve 4D driving-scene reconstruction under complex maneuvers. SE017, SE021
CE027 DriveDreamer4D reports relative FID gains of 32.1%, 46.4%, and 16.3% over PVG, S3Gaussian, and Deformable-GS. SE017
CE028 Official product copy says DriveDreamer, DriveDreamer4D, and ReconDreamer address data shortage, corner-case collection difficulty, and testing cost for autonomous-driving customers. SE002
CE029 Gasgoo says Maker H01 shipments began in early 2026 and that the first delivered unit went to the Hubei Humanoid Robot Innovation Center. SE008
CE030 Gasgoo says GigaAI and the Hubei center were building a virtual-real embodied-intelligence data factory spanning control, collection, processing, training, and iteration. SE008
CE031 36Kr and NCSTI say the Maker H01 plus GigaBrain plus GigaWorld stack entered FAW Tooling Die’s factory in April 2026 for box depalletizing, cross-zone transport, dynamic obstacle avoidance, and precise operation. SE003, SE007
CE032 The same June 2026 reports say GigaAI plans to deploy 1,000 Maker-series robots with Longsheng Technology in Wuxi over three years. SE003, SE007
CE033 SeeLight is the household brand and SeeLight S1 is reported to have secured 100-home orders with scaled operations targeted for Q3 2026. SE003, SE007, SE009
CE034 SeeLight S2 is slated for Q3 2026 release and founder-edition reservations were said to open alongside the brand launch. SE003, SE007, SE009
CE035 Sohu reports the planned SeeLight S2 redesign reduces base volume by 60%, raises battery endurance by 70%, expands reachable task height by 40%, and supports hot-swappable batteries. SE009
CE036 NCSTI says GigaAI’s commercialization splits into foundation-model services delivered via software, API, or licensing and AGI-native products delivered via SeeLight and Maker robots. SE003
CE037 The public core open-source repos are Apache 2.0 licensed, which lowers legal friction for research adoption but does not by itself prove production support. SE011, SE013, SE014, SE015
CE038 GigaBrain Challenge 2026 at CVPR included dedicated GigaWorld and RoboChallenge tracks, showing GigaAI is investing in benchmark and community scaffolding around its technical agenda. SE024, SE025
CE039 The public official site and bundle do not disclose robot pricing, customer-facing SLA terms, or formal safety and compliance certifications. SE001, SE002
CE040 Several high-visibility performance claims, including WorldArena, RoboCasa365, and named customer-scale results, are repeated in media and marketing without a directly linked benchmark or customer ledger on the official site. SE002, SE003, SE007
CE041 Official product copy says GigaAI has signed and mass-production cooperation with more than 20 OEM and autonomous-driving solution companies for world-model-based data generation, closed-loop simulation, and reinforcement-learning solutions. SE002
CE042 June 2026 independent coverage says the DriveDreamer business already serves more than 30 domestic and overseas OEM and autonomous-driving customers. SE003, SE007
CU001 Supportable public customer evidence is concentrated in industrial manufacturing, home or eldercare pilots, and automotive simulation; 3C electronics and logistics are mostly claimed segments rather than named public accounts. SU007, SU009, SU011
CU002 In industrial manufacturing, the public record implies enterprise factories or public institutions are the buyer or sponsor, plant operators are the day-to-day users, and payment would come from enterprise or state-linked project budgets rather than consumers. SU003, SU004, SU021
CU003 GigaAI says the first Maker H01 delivery wave has already started and included a shipped unit to the Hubei Humanoid Robot Innovation Center. SU002
CU004 GigaAI and the Hubei Humanoid Robot Innovation Center partnered to build a world-model-driven virtual-real embodied-intelligence data factory. SU002, SU017
CU005 The Hubei Humanoid Robot Innovation Center is promoting GigaBrain applications in intelligent manufacturing and commercial services, indicating it serves as both a proving ground and an ecosystem distribution partner. SU018
CU006 FAW Tooling Die is publicly named as a Maker H01 deployment site for box unloading and cross-area transportation in real factories. SU004, SU009
CU007 The FAW Tooling Die deployment is also described as covering dynamic obstacle avoidance and precise operations in high-frequency factory tasks. SU008, SU009
CU008 NCSTI reports that the FAW deployment compressed scenario-adaptation cycles from months to weeks versus traditional automation. SU009
CU009 GigaAI and Longsheng Technology disclosed a three-year plan to deploy 1,000 Maker-series robots in industrial settings around Wuxi. SU003, SU004, SU005
CU010 Gasgoo says the first batch in the Longsheng program has already passed industrial-handling verification and been delivered. SU003
CU011 Longsheng-linked deployment language covers intelligent manufacturing, industrial logistics, precision machining, and lighthouse factories rather than a single narrow workstation. SU003, SU022
CU012 The Jiangsu Longsheng Weirui innovation center publicly highlights warehousing logistics and precision assembly as application targets, reinforcing logistics as a partner-led expansion path for embodied robots in Wuxi. SU022
CU013 ECARX officially integrated DriveDreamer into AutoGPT to accelerate autonomous-driving development cycles and reduce development costs for automakers. SU006
CU014 ECARX says DriveDreamer can also support hyper-personalized intelligent-cockpit features by analyzing multimodal passenger and vehicle inputs. SU006
CU015 Gasgoo reports that DriveDreamer has partners including Li Auto and ECARX, but the article does not provide contract details or deployment volumes for Li Auto. SU007
CU016 NCSTI and 36Kr both repeat that DriveDreamer serves more than 30 OEM or autonomous-driving customers, but neither reviewed public source itemizes the customer list. SU008, SU009
CU017 The strongest directly supportable public automotive customer proof is the ECARX integration; the wider 30-plus customer count remains a non-itemized company-claim cluster. SU006, SU008, SU009
CU018 China Daily says 100 SeeLight S1 units will enter ordinary households on a free-trial basis starting in the third quarter of 2026. SU011
CU019 Yicai, VOI, 36Kr, and NCSTI repeat company language that SeeLight S1 has about 100 orders and is slated for mass delivery in the next quarter. SU004, SU005, SU008, SU009
CU020 Public descriptions of the first 100 home units conflict on whether they are paid orders, employee pilots, or free household trials. SU011, SU013, SU015, SU024
CU021 Mike Kalil reports that the first 100 home units are entering Optics Valley talent apartments and are being provided free of charge to collect feedback from tenants. SU015
CU022 eWeek says later home trials are planned to prioritize households with elderly residents, children, or pets. SU013
CU023 Interesting Engineering frames eldercare and demographic support as a stated rationale for the household robot rollout. SU014, SU011
CU024 The earliest home deployments are described as employee housing, talent apartments, or high-end international community environments rather than broad mass-market households. SU015, SU020, SU024
CU025 China Daily says the household-trial announcement drew more than 2,000 messages on the company’s official WeChat account. SU011
CU026 China Daily says GigaAI will collect in-home task data during the trial, with consent, to verify functions, identify faults, and drive upgrades. SU011
CU027 eWeek reports that current household task execution can be slow and imperfect, including multi-minute book organization, double-digit-minute clothing folding, and spilled water during mug handling. SU012
CU028 Humanoids Daily argues that home robotics still faces major data, safety, and cost barriers and characterizes GigaAI as skipping an industrial-first learning curve. SU016
CU029 Inc/Fast Company says the first 100 pilot units are planned for employees’ homes before a wider Wuhan rollout. SU024
CU030 Public safety evidence currently rests on company-described compliant-control behavior that is supposed to freeze robot movement on contact with children or pets. SU013, SU024
CU031 No reviewed public source discloses NRR, GRR, churn, renewal rates, contract length, or cohort retention for either industrial or household customers. SU001, SU004, SU011
CU032 Public satisfaction proof is anecdotal rather than contractual: the reviewed corpus is dominated by demos, sign-up interest, and executive quotations instead of renewals or customer KPI dashboards. SU011, SU015, SU024
CU033 Gasgoo says benchmark clients already span automotive manufacturing, 3C electronics, warehousing and logistics, high-end guiding, and home terminals. SU007
CU034 Named public proof is strongest for FAW Tooling Die, Longsheng or Longsheng Weirui, ECARX, the Hubei Humanoid Robot Innovation Center, and Wuhan household pilots. SU002, SU003, SU004, SU006, SU011
CU035 No reviewed public source names a 3C electronics customer or discloses an order value for that segment. SU007, SU009
CU036 No reviewed public source names a warehousing or logistics customer beyond Longsheng-linked program language and scenario descriptions. SU003, SU007, SU022
CU037 The Hubei and Wuhan ecosystem is central to GigaAI’s proof set, combining the data factory, intelligent-manufacturing promotion, humanoid-industry policy, and curated residential pilot settings. SU018, SU019, SU020
CU038 Longsheng functions as both a manufacturing-supply partner and a scale customer channel, which can accelerate rollout but also raises partner-concentration risk. SU003, SU022
CU039 The 1,000-unit Longsheng plan is an announced three-year deployment commitment rather than a fully realized installed base or recognized revenue figure. SU003, SU004, SU005
CU040 The home deployment path is still pilot-heavy and subsidy-like because the first households are described as free-trial or feedback sites rather than disclosed paying subscribers. SU011, SU015, SU024
CU041 China Daily quotes the Hubei Humanoid Robot Innovation Center saying the household trial is also meant to explore a commercial model covering maintenance and after-sales services. SU011
CU042 TEDA’s automotive embodied-AI joint lab shows that FAW-linked deployments still require standards, system integration, and engineering validation before broad scaling. SU021
CU043 Wuhan and Wuxi local governments are funding humanoid-robot ecosystems and scenario demonstrations, so a meaningful share of GigaAI’s public traction is policy-assisted rather than purely market-proven. SU019, SU022
CU044 Early home pilots appear concentrated in premium or policy-backed housing environments before exposure to broad middle-income consumer demand. SU015, SU020, SU024
CU045 No reviewed public source discloses revenue share, installed-base share, or ARR concentration by top customer. SU003, SU004, SU011
CU046 Before treating the announced scaling path as durable revenue, diligence should request signed orderbooks versus pilots, pilot-to-paid conversion, top-customer concentration, renewal data, and service-unit economics. SU003, SU011, SU021
CR001 BIS guidance says a license is required to export advanced computing items to entities headquartered in Country Group D:5 or Macau even when those entities are located outside D:5 or Macau. SR002
CR002 BIS says export license applications for Nvidia H200, AMD MI325X, and similar chips to China will be reviewed case by case under security and testing conditions. SR001
CR003 Miller & Chevalier says the new 2025 AI-chip restrictions were triggered through BIS 'is informed' letters, cover H20-class chips, and reflect concern about diversion to Chinese supercomputers and third countries. SR003
CR004 China has released its first national standard system covering the full industrial chain and lifecycle of humanoid robots and embodied AI, with safety and ethics standards running through the lifecycle. SR004
CR005 China’s Interim Measures for Generative AI Services took effect on 2023-08-15 and were issued with concurrence from multiple ministries including MIIT and NDRC-related bodies. SR005
CR006 The CAC generative AI measures require providers to conduct training-data processing lawfully under cybersecurity, data-security, personal-information, and science-and-technology laws. SR005
CR007 The CAC generative AI measures require providers to respect intellectual property and commercial ethics, improve transparency, and improve the accuracy and reliability of generated content. SR005
CR008 Publicly reviewed sources identify Huang Guan as founder and chief executive while Zheng Zhu is presented as co-founder and chief scientist, implying a visibly concentrated leadership bench. SR015, SR020, SR021
CR009 MIT Technology Review argues that humanoid adoption is likely to be drawn out, industry specific, and slow rather than immediate at workforce scale. SR009
CR010 The active GigaAI website reviewed on the run date exposes only the title '极佳科技', providing little public trust, safety, or compliance detail compared with enterprise-grade vendors. SR013
CR011 Gasgoo says GigaAI began Maker H01 deliveries with a first wave of multi-customer orders and shipped an initial unit to the Hubei Humanoid Robot Innovation Center. SR014
CR012 Yicai says the SeeLight S1 has about 100 orders, GigaAI has a factory use case with FAW Tooling, and Longsheng Technology plans to deploy 1,000 Maker-series robots over three years. SR015
CR013 Pedaily and 36Kr say GigaAI targets Q3 2026 scaled operations for SeeLight S1/S2 and a three-year 1,000-robot Longsheng rollout in industry. SR018, SR019
CR014 Yicai says GigaAI raised CNY3.5 billion over a three-month period in 2026. SR015
CR015 Yicai says GigaAI raised CNY1 billion in March and CNY1.5 billion in April, taking post-financing valuation above CNY10 billion. SR015, SR017
CR016 Yicai says the latest financing proceeds are meant for data and algorithm systems, physical-AGI foundation-model refinement, and scaling household and industrial robot products. SR015
CR017 NVIDIA Isaac ROS is built on ROS 2 and optimized for NVIDIA GPUs, NVIDIA Jetson devices, and NVIDIA DGX-class hardware, showing how mainstream robotics acceleration still leans on NVIDIA infrastructure. SR030
CR018 GigaTrain advertises multi-GPU and multi-node training plus FP16, BF16, and FP8 support, indicating that GigaAI’s internal training stack is designed around significant compute scale rather than lightweight experimentation alone. SR026
CR019 The GigaBrain-0 paper says large-scale real-world robot data is expensive and time-consuming to collect, and that this bottleneck limits the scalability and generalization of current VLA systems. SR028
CR020 The GigaBrain repository says GigaBrain-0 trained on about 1,000 hours of real-world robot data and that GigaBrain-0.1 scaled the training data to 10,000 hours. SR023
CR021 The GigaBrain paper says world-model-generated data includes video generation, real2real transfer, human transfer, view transfer, and sim2real transfer to reduce dependence on real robot data while improving cross-task generalization. SR028
CR022 The GigaWorld paper claims that GigaBrain models trained on GigaWorld-generated data improved task success on physical robots without any real-world interaction during training. SR029
CR023 IEEE Spectrum says the humanoid market is still almost entirely hypothetical and that even leading companies have deployed only a small number of robots in carefully controlled pilot projects. SR010
CR024 IEEE Spectrum quotes an industry operator saying current AI is not robust enough to meet market requirements for multipurpose humanoids. SR010
CR025 IEEE Spectrum says industrial buyers care about battery life, reliability, and safety, and cites 99.99% reliability as the kind of expectation attached to production-line use cases. SR010
CR026 MIT Technology Review says stronger humanoids need more power, heavier batteries, more safety consideration, and complex manufacturing, while flashy demos may not map cleanly onto real jobs. SR009
CR027 A 2024 humanoid review paper says current humanoids remain distant from human-like to human-level intelligence. SR012
CR028 GigaModels is an Apache 2.0 open-source repository that includes GigaBrain and GigaWorld pipelines among a wider model toolkit. SR025
CR029 GigaTrain and GigaDatasets are also Apache 2.0 projects and are designed for straightforward installation and reuse. SR026, SR027
CR030 Because GigaAI has open-sourced core data, training, and model tooling, the stack is easier for developers and fast followers to inspect, reproduce, and benchmark than a fully closed robotics platform. SR023, SR024, SR025, SR026, SR027
CR031 The algorithmic recommendation rules require providers to establish algorithm security, technology ethics review, security incident response, data security, and personal-information protection controls. SR022
CR032 The same rules require marking synthetic information, providing users with opt-out and tag-deletion functions, and avoiding manipulative rankings or unfair differentiated treatment. SR022
CR033 Providers with public-opinion properties or social-mobilization capabilities must complete filing formalities within 10 working days and conduct a security assessment under the algorithmic recommendation rules. SR022, SR005
CR034 China’s Product Quality Law requires industrial products that may endanger human health or personal or property safety to meet national or industry safety standards and carry warnings where improper use can create danger. SR032
CR035 The Product Quality Law says sellers must repair, replace, refund, or compensate for qualifying quality failures and says producers are liable where product defects cause personal injury or property damage. SR032
CR036 The Product Quality Law provides a two-year limitation period from discovery and a ten-year long-stop for defect claims, except where a stated safe-use period has not yet expired. SR032
CR037 Chambers says labor dispatch in China is limited to temporary, auxiliary, or substitutable positions and the share of dispatched workers generally may not exceed 10% of the workforce. SR006
CR038 Chambers says employers should sign a written employment contract within one month, otherwise double salary can be owed, and overtime and statutory-holiday pay rules are prescriptive. SR006
CR039 DLA Piper says Chinese courts may recognize employment relationships without written contracts based on factual indicators and may impose joint liability on affiliated companies for unpaid wages or benefits. SR007
CR040 ADVANT Beiten says affiliated entities can jointly bear salary and insurance liability in group-employment structures and that foreign representative offices must use qualified dispatch agencies for Chinese staff. SR008
CR041 GigaBrain-0-Small is presented as efficient enough to run on NVIDIA Jetson AGX Orin, tying at least one public deployment path directly to NVIDIA edge hardware. SR028, SR030
CR042 Microsoft says AirSim has been archived and will receive no further updates, illustrating that important simulation platforms can age out even when they were once central to robotics development. SR031
CR043 Pedaily and 36Kr say GigaAI expects to accumulate about 1 million hours of world-action or high-quality visual-action data and 10 million hours of world-model pretraining video as part of its scaling roadmap. SR017, SR018, SR019
CR044 Reviewed public materials emphasize funding, orders, roadmap milestones, and named pilots far more than disclosed revenue, ARR, warranty cost, or support-economics metrics. SR014, SR015, SR018, SR019, SR021
CR045 The combination of external compute dependency, unresolved reliability thresholds, and capital-intensive scaling means GigaAI’s residual risk rating is high even after accounting for its fundraising momentum. SR001, SR010, SR015, SR026, SR028, SR030
CR046 The public technical story is concentrated in a small visible leadership set led by Huang Guan and Zheng Zhu rather than a broadly disclosed enterprise operating bench. SR015, SR020, SR021
CR047 Named industrial validation is concentrated in Hubei, FAW Tooling, and Longsheng, so slippage at a small number of anchor deployments could disproportionately hurt commercialization credibility. SR014, SR015, SR018, SR019
CR048 Because disclosed financing proceeds still fund data systems, model refinement, and rollout expansion, continued access to capital remains operationally important rather than purely opportunistic. SR015, SR018, SR019
CR049 A household robot business that captures multimodal data and uses model-mediated outputs creates elevated privacy, labeling, and algorithm-governance risk under China’s generative-AI and algorithm rules. SR005, SR022, SR029
CR050 The minimal official web surface weakens counterparties’ ability to verify compliance posture, support pathways, or safety documentation before deeper diligence begins. SR013
CR051 Current public evidence supports enthusiasm and milestone momentum, but it does not yet prove that GigaAI has broad customer diversity or self-sustaining operating economics. SR009, SR010, SR015, SR019, SR021
CV001 GigaAI raised about CNY1 billion in a Pre-B round in March 2026. SV001, SV002, SV003
CV002 GigaAI then closed a B1 round worth nearly CNY1.5 billion in April 2026. SV002, SV003
CV003 GigaAI completed a strategic/B2 financing round of CNY1 billion in June 2026. SV003, SV004
CV004 Reviewed sources converge that GigaAI raised CNY3.5 billion, or about $518 million, across those three 2026 rounds in roughly three months. SV002, SV003, SV004
CV005 The same 2026 financing coverage places GigaAI's post-financing valuation above CNY10 billion, or about $1.5 billion. SV003, SV004
CV006 Reviewed public sources do not disclose GigaAI revenue, ARR, gross margin, or customer concentration as of 2026-06-24. SV001, SV002, SV003, SV004
CV007 Yicai reported that GigaAI's SeeLight S1 had received about 100 orders and was set to begin mass delivery in the following quarter. SV003
CV008 Gasgoo reported that GigaAI was targeting 1,000 unit deliveries in 2026 across industrial and home-service scenarios. SV002
CV009 Figure announced that its Series C financing exceeded $1 billion at a $39 billion post-money valuation. SV005, SV007
CV010 Figure's earlier Series B financing raised $675 million at a $2.6 billion valuation. SV006, SV007
CV011 The move from Figure's $2.6 billion 2024 valuation to its $39 billion 2025 valuation represents roughly a 15x re-rating in about 18 months. SV005, SV006, SV007
CV012 Physical Intelligence closed a $600 million Series B in 2025 at a $5.6 billion post-money valuation. SV009, SV011
CV013 TechCrunch reported in March 2026 that Physical Intelligence was discussing another roughly $1 billion round at a valuation above $11 billion. SV010
CV014 Sacra describes Physical Intelligence as a software/model-layer business with pricing around $300 per connected robot per month rather than a hardware-only revenue model. SV011
CV015 Caixin reported that Unitree's latest market-based funding round in June 2025 valued the company at about CNY12.7 billion post-money. SV028
CV016 Caixin reported that Unitree's 2026 STAR Market IPO process targeted a valuation of roughly CNY42 billion, or about $6.2 billion. SV027, SV028
CV017 Caixin reported that Unitree generated about CNY1.7 billion of revenue and CNY280 million of net profit in 2025. SV028
CV018 Caixin also reported that Unitree's first-quarter 2026 net profit fell 47.7% even as revenue grew, showing commercialization does not eliminate volatility. SV028
CV019 Unitree's official G1 page lists a pre-tax, pre-shipping price of US$13.5K, giving the market a level of pricing transparency GigaAI does not yet provide publicly. SV014
CV020 EngineAI says cumulative Pre-A++ and Series A1 funding had already reached CNY1 billion before it completed A1+ and A2 rounds. SV015
CV021 EngineAI also says its T800 humanoid robot had entered mass production by the time of the latest financing announcement. SV015
CV022 Humanoid Index reports that AgiBot had disclosed more than $83 million of funding and 5,100 units shipped in 2025. SV025
CV023 CompaniesMarketCap shows Symbotic at about $23.28 billion of market capitalization in June 2026. SV020
CV024 Symbotic reported Q2 FY2026 revenue of $676 million and guided Q3 FY2026 revenue of $700 million to $720 million. SV017
CV025 Using Symbotic's June 2026 market capitalization and Q2 FY2026 annualized revenue implies a public-market multiple of roughly 8.6x sales. SV017, SV020
CV026 CompaniesMarketCap shows Serve Robotics at about $0.54 billion of market capitalization in June 2026. SV021
CV027 Serve Robotics reported 2025 revenue of $2.7 million and a 2026 revenue outlook of about $26 million. SV018
CV028 Using Serve's June 2026 market capitalization and its 2026 outlook implies a forward revenue multiple of roughly 20.8x, illustrating how extreme early-stage robotics multiples can be on small bases. SV018, SV021
CV029 Grand View Research estimates the global humanoid robot market at $1.55 billion in 2024, growing to about $4.04 billion by 2030. SV022
CV030 MIT Technology Review argues that humanoid deployment is running late and that sector revenue remains minimal despite high valuations. SV023
CV031 Across private embodied-AI peers, headline valuations frequently outrun disclosed revenue, making narrative leadership and future optionality major drivers of price. SV005, SV007, SV010, SV011, SV023, SV024
CV032 GigaAI's current mark sits below Figure's and Unitree's most aggressive 2026 reference points, but the quality of public support is weaker because GigaAI lacks Unitree-like revenue disclosure and a clearer software-model narrative than Physical Intelligence. SV003, SV005, SV011, SV027, SV028
CV033 Because GigaAI revenue is undisclosed, scenario-based underwriting is more defensible than any single-point multiple derived from public evidence. SV003, SV017, SV018, SV020, SV021, SV023
CV034 At a current mark around $1.5 billion, a fair base underwriting would likely require something like $150 million to $200 million of future revenue at roughly 8x to 10x quality-adjusted sales, or a comparably profitable hardware path. SV017, SV018, SV020, SV021, SV028
CV035 A plausible bull case would require disclosed revenue above roughly $250 million by 2028, meaningful software attach, and deployment scale that converts today's order signals into recurring commercial proof. SV002, SV003, SV005, SV011, SV028
CV036 A plausible base case is that GigaAI commercializes meaningfully but remains hardware-heavy, leaving fair value around $0.8 billion to $1.5 billion. SV003, SV025, SV028, SV020, SV021
CV037 A plausible bear case is that commercialization lags and the next round introduces meaningful preference overhang, pulling value toward roughly $0.2 billion to $0.6 billion. SV023, SV021, SV029
CV038 The evidence set supports a research-more recommendation rather than buy, because valuation diligence has not caught up with funding momentum. SV003, SV023, SV028
CV039 Confidence should be medium rather than high because the public file is good on rounds and peer marks but weak on GigaAI's own economics and round terms. SV003, SV004, SV017, SV018
CV040 A high risk rating is appropriate because commercialization timing, multiple compression, and preference-stack uncertainty can all hurt common-equity outcomes from the current price. SV019, SV023, SV029
CV041 The best-fitting valuation stance is stretched, since the current mark can be rationalized only under favorable future disclosure and execution rather than on disclosed present-day metrics. SV003, SV020, SV021, SV028
CV042 Entry discipline should require dated revenue, gross margin, backlog quality, customer concentration, and preference-stack disclosure before paying or averaging up at the current price. SV003, SV006, SV017, SV018, SV028
CV043 The strongest thesis for tracking GigaAI is that few Chinese embodied-AI startups can match its 2026 capital access while also pursuing a full-stack world-model-plus-robot strategy. SV002, SV003, SV005, SV008
CV044 The strongest anti-thesis is that peers with the clearest valuation support either disclose stronger operating evidence, like Unitree, or command global narrative premiums that GigaAI has not yet independently earned in public data. SV005, SV011, SV028
CV045 The most important thesis-break trigger is future disclosure showing revenue or backlog quality materially below what a $1.5 billion-plus valuation requires. SV003, SV023, SV028
CV046 Public evidence supports keeping IPO or strategic-sale optionality on the map, but it does not support treating GigaAI as exit-ready today. SV003, SV028, SV029
来源
编号出版方标题引文
SO001 GigaAI 极佳科技 极佳科技
SO002 Spaceship / parked page Domain for sale Listed with spaceship.com
SO003 Yicai Global Chinese Embodied AI, Robotics Developer GigaAI Raises USD518 Million Over Three Months GigaAI has raised CNY3.5 billion (USD518 million) from investors over a three-month period.
SO004 36Kr 持续领跑世界模型驱动物理AGI,极佳视界再获10亿元B2轮融资 极佳视界完成10亿元B2轮。
SO005 36Kr 极佳视界完成10亿元Pre-B轮融资,「世界模型」驱动构建全球头部的「具身基模」 极佳视界宣布完成近10亿元Pre-B轮融资。
SO006 36Kr Europe Tsinghua University Doctoral Graduate Raises Another 1 Billion Yuan Until June 2023, Huang Guan embarked on a new entrepreneurial journey and officially founded Giga Vision.
SO007 Gasgoo Seeds raised 2.5 billion yuan in a month, another embodied unicorn valuation surpasses 10 billion yuan That means GigaAI has raised 2.5 billion yuan in a single month, pushing its valuation past the 10 billion yuan mark.
SO008 Gasgoo GigaAI Begins Large-Scale Deliveries GigaAI has officially kicked off deliveries for its in-house general-purpose robot.
SO009 Gasgoo With 1 billion yuan invested, Gigaai aims to become the “OpenAI of the physical world” On March 5, Gigaai announced it has recently closed a nearly 1 billion yuan Pre-B financing round.
SO010 Crunchbase News New AI Unicorn Startups In April 2026: Frontier Labs, Ineffable Intelligence, Recursive Superintelligence Beijing-based GigaAI ... raised a $220 million Series B. The 3-year-old company was valued at $1.5 billion.
SO011 National Center for Science and Technology Information 北京极佳视界科技有限公司再获10亿元B2轮融资 此前2026年3月和4月,极佳视界已分别完成10亿元Pre-B轮和15亿元B1轮融资。
SO012 163 / NetEase Hao 极佳视界发布拾光S1,百台订单率先在武汉落地 “拾光S1”目前已获得真实家庭场景的百台订单。
SO013 PEDAILY / 投资界 极佳视界,一个月融资25亿 直到2023年6月,黄冠开启新的创业征程,正式创立极佳视界。
SO014 PEDAILY / 投资界 首发| 极佳视界一个月内再获数十亿融资,世界模型赛道还能火多久? 极佳视界正式完成近15亿元B1轮融资。
SO015 PEDAILY / 投资界 首发| 极佳视界三个月融资35亿,投资人开抢物理世界OpenAI 这已是极佳视界三个月内第三笔融资,累计金额高达35亿元。
SO016 PEDAILY / 投资界 极佳视界再获10亿元B2轮融资,持续领跑世界模型驱动的物理AGI 此前2026年3月和4月,极佳视界已分别完成10亿元Pre-B轮和15亿元B1轮融资。
SO017 Tencent News / iHeima 极佳视界一个月内再获数十亿融资,世界模型赛道还能火多久? 估值泡沫、机器人“大脑”能力与商业闭环的三重考验也摆在所有玩家面前。
SO018 Sina Finance 近10亿Pre-B轮融资落地,极佳视界凭什么成为具身智能“吸金兽”? 2023年,他离开鉴智机器人,创立了极佳视界。
SO019 Tencent Cloud Developer / 机器之心 这家世界模型公司发布中国版Sora级视频生成大模型,走向世界模型打造新一代数据引擎 2023 年 9 月,极佳科技推出了全球首个真实世界驱动的自动驾驶世界模型 DriveDreamer。
SO020 Tencent Cloud Developer 具身智能的“盗梦空间”!GigaAI最新重磅发布GigaBrain-0:世界模型驱动的VLA模型 以世界模型为核心的数据引擎。
SO021 HOKANEWS GigaAI Unveils SeeLight S1 Home Humanoid Robot for Real Household Tasks Wuhan has been chosen as the first deployment city for the initial batch of 100 units.
SO022 Humanoid Press Maker H01 | GigaAI’s Physical AGI Humanoid Robot Maker H01 is the first humanoid robot from GigaAI.
SO023 GitHub GigaAI-research [ECCV 2024] DriveDreamer.
SO024 GitHub DriveDreamer DriveDreamer is the first world model established from real-world driving scenarios.
SO025 GitHub DriveDreamer4D DriveDreamer4D is the first to utilize video generation models for improving 4D reconstruction in driving scenarios.
SO026 GitHub GigaWorld-0 GigaWorld-0 ... designed explicitly as a data engine for Vision-Language-Action learning.
SO027 arXiv GigaWorld-0: World Models as Data Engine to Empower Embodied AI VLA models trained on GigaWorld-0-generated data achieve strong real-world performance.
SO028 Zheng Zhu Zheng Zhu (朱政) Zheng Zhu is currently the Co-founder and Chief Scientist at GigaAI.
SO029 TMTPost / AsianFin Spatial Intelligence Company Giga AI Secures Nearly 100 Million Yuan in Financing Giga AI, a Beijing-based spatial intelligence company, announced ... nearly 50 million yuan in angel and angel+ financing.
SM001 International Federation of Robotics Global Robot Density in Factories Doubled in Seven Years The new global average robot density reaches a record 162 units per 10,000 employees in 2023.
SM002 International Federation of Robotics China Makes AI-powered Robots Core of National Strategy China’s manufacturing industry already has an operational stock of around 2 million units.
SM003 International Federation of Robotics Preliminary U.S. industrial robot installations results 2025 Annual installations in China reached 295,000 units in 2024. This represents a global market share of 54%.
SM004 Boston Consulting Group How Physical AI Is Reshaping Robotics Today—and What Comes Next Level 5 reasoning is the gating constraint for truly general-purpose robotics.
SM005 Google DeepMind Introducing Gemini Robotics and Gemini Robotics-ER, AI models designed for robots to understand, act and react to the physical world Gemini Robotics more than doubles performance on a comprehensive generalization benchmark.
SM006 The State Council of the People’s Republic of China China to nurture emerging, future industries Robotics is one of the strategic emerging sectors highlighted for accelerated development.
SM007 CGTN Robots at work: Beijing advances embodied AI development The M7 sorting robot can process up to 1,200 parcels per hour.
SM008 MarketsandMarkets Humanoid Robot Market Size, Share, Latest Trends & Growth Analysis, 2025-2030 The US humanoid robot market is projected to reach USD 4,601.0 million by 2030.
SM009 The Business Research Company Embodied Artificial Intelligence (AI) Market Insights To 2035 The embodied artificial intelligence (AI) market size has grown from $3.22 billion in 2025 to $3.8 billion in 2026.
SM010 Harvard Business School Working Knowledge Humans vs. Machines: Untangling the Tasks AI Can (and Can’t) Handle On more complex tasks, consultants using AI were 19 percentage points less likely to produce the right answer.
SM011 CNBC Morgan Stanley says humanoid robots will be a $5 trillion market by 2050. How to play it Morgan Stanley analysts forecast $4.7 trillion in global humanoid revenue by 2050.
SM012 CNBC Morgan Stanley expects 8 million humanoids to exist by 2040, names stocks to play the AI theme The bank forecasts a humanoid population of 40,000 by 2030, 8 million by 2040 and 63 million by 2050.
SM013 CNBC Goldman says humanoid robots will be a $6 billion market in 10 years – How to play the growing trend Goldman Sachs envisions a market of up to US$154bn by 2035E in a blue-sky scenario.
SM014 World Economic Forum Why the next decade of physical AI must be human-centric Robotics and autonomous systems are expected to transform business operations at 58% of employers by 2030.
SM015 World Economic Forum Physical AI in Industrial Operations Three complementary robotics systems are emerging that will coexist in the target state: rule-based, training-based and context-based robotics.
SM016 World Economic Forum Physical AI in the supply chain: How its promise can be realized With the growing adoption of automated storage and retrieval systems and robotics throughout the supply chain, the foundational infrastructure is now developed enough for businesses to apply physical AI.
SM017 World Economic Forum Spatial computing, wearables and robots: AI’s next frontier
SM018 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.
SM019 IEEE Spectrum Do People Really Want Humanoid Robots in Their Homes? Our survey showed that people generally prefer special-purpose robots over humanoids.
SM020 IEEE Spectrum Humanoid Robots and the AI Brain Shift We need world models. We need some way for the AI system to imagine, try things out, and truly reason.
SM021 NVIDIA NVIDIA Isaac ROS NVIDIA Isaac ROS is built on the open-source ROS 2 software framework.
SM022 Microsoft Research Aerial Informatics and Robotics Platform AirSim solves these two problems: the need for large data sets for training and the ability to debug in a simulator.
SM023 Berkeley News Are we truly on the verge of the humanoid robot revolution? I’m saying it’s not going to happen in the next two years, or five years or even 10 years.
SM024 Bank of America Institute Physical AI, part 2: Humanoid robots Annual humanoid robot shipments are projected to reach 1.2 million in 2030 and 10 million by 2035.
SM025 IEEE Spectrum Agility’s New Factory Can Build Thousands of Humanoids a Year At full scale, over 10,000 Digit robots will be built in Oregon annually.
SP101 GigaAI 极佳科技
SP102 Yicai Global Chinese Embodied AI, Robotics Developer GigaAI Raises USD518 Million Over Three Months The SeeLight S1 has already received about 100 orders and is set to begin mass delivery next quarter.
SP103 ChipSilicon Maker H01 | Humanoid Robot Maker H01 is a wheeled humanoid from GigaAI — pairing dual 7-DOF arms, rich sensing, and agile mobility.
SP104 RoboActu SeeLight S1 | RoboActu Cent SeeLight S1 sont déployés ce mois-ci dans des logements de fonction réservés aux salariés high-tech à Wuhan.
SP105 EngineAI ENGINEAI
SP106 RoboticsTomorrow EngineAI Launches Shenzhen Intelligent Manufacturing Base as First Batch of T800 Humanoid Robots Roll Off the Production Line to Begin Mass Delivery Each robot must pass 79 full-dimensional quality inspections and 46 working condition simulation tests before delivery.
SP107 Interesting Engineering One T800 humanoid robot every 15 mins, ENGINEAI’s new factory claims In April, the company finished its Series B financing round, raising $200 million.
SP108 Humanoids Daily EngineAI Secures $200 Million Series B as Manufacturing Giant Luxshare Joins the Cap Table The investment, which pushes the company’s valuation past 10 billion RMB ($1.4 billion), marks a critical milestone.
SP109 AgiBot AGIBOT Innovation (Shanghai) Technology Co., Ltd. Introducing AGIBOT World, the first Large Scale, Quality, Realistic Task Dataset and Ecosystem for Embodied.
SP110 GitHub AgibotTech AgibotTech/Omnihand-2025-SDK’s past year of commit activity.
SP111 Unitree Robotics 宇树科技—全球四足机器人行业开创者
SP112 Unitree Robotics Humanoid robot G1_Humanoid Robot Functions_Humanoid Robot Price Price(Tax and Shipping cost excluded) US $13.5K.
SP113 CNBC Nvidia picks Unitree for humanoid robot platform as Chinese startup eyes IPO Nvidia has selected Chinese humanoid robot maker Unitree for the first robotics system the chipmaker is selling to researchers.
SP114 Figure Figure Figure 03 is a general purpose humanoid robot for every day.
SP115 Figure News | Figure F.02 Contributed to the Production of 30,000 Cars at BMW.
SP116 Figure Introducing Figure 03 BotQ’s first-generation manufacturing line will initially be capable of producing up to 12,000 humanoid robots per year.
SP117 Physical Intelligence Physical Intelligence (π) Physical Intelligence is bringing general-purpose AI into the physical world.
SP118 The Robot Report Physical Intelligence raises $600M to advance robot foundation models The San Francisco-based company plans to use the financing to collect more data, make strategic partnerships, and grow its team.
SP119 TechCrunch Physical Intelligence is reportedly in talks to raise $1B, again Co-founder Lachy Groom told TechCrunch the company has no timeline for commercialization.
SP120 FANUC America Industrial Robots for Manufacturing The most extensive range of robots and cobots available in the market across every application and industry.
SP121 KUKA Industrial robots for every application and industry KUKA offers a comprehensive range of industrial robots, catering to diverse applications with precision, flexibility and efficiency.
SP122 ABB Robotics | ABB ABB’s collaborative robots are made for a wide range of tasks and serviced by the broadest service network in the industry.
SP123 Berkeley News Are we truly on the verge of the humanoid robot revolution? I’m not saying it’s not going to happen, but I’m saying it’s not going to happen in the next two years, or five years or even 10 years.
SP124 MIT Technology Review Why the humanoid workforce is running late The adoption of the technology will likely be drawn out, industry specific, and slow.
SP125 KrASIA Bubble or breakthrough? China’s humanoid robotics race faces reality check We placed multiple bets, but this is still a fuzzy space. There’s a real chance it’s a bubble.
SI001 GigaAI 极佳科技
SI002 爱企查 北京极佳视界科技有限公司 - 工商信息查询 - 爱企查 经营范围包括…智能机器人销售…工业机器人安装、维修…机械设备租赁…货物进出口。
SI003 Yicai Global Chinese Embodied AI, Robotics Developer GigaAI Raises USD518 Million Over Three Months Proceeds of the latest raise will mainly go toward ongoing investment in data and algorithm systems, the development and refinement of GigaAI’s physical AGI foundation model, and the scaling of its robot products in household and industrial settings.
SI004 Gasgoo GigaAI Begins Large-Scale Deliveries The initial unit — a "physical AGI native body" dubbed Maker H01 — has been shipped to the Hubei Humanoid Robot Innovation Center.
SI005 Gasgoo With 1 billion yuan invested, Gigaai aims to become the "OpenAI of the physical world" It has now started mass production and delivery for scenarios including data collection, industrial use, and services.
SI006 36Kr 极佳视界完成10亿元Pre-B轮融资,「世界模型」驱动通用机器人加速进入千行百业
SI007 国家科技信息中心 极佳视界再获10亿元融资!加速布局世界模型驱动的物理AGI 一类是面向行业伙伴输出基础模型能力的基础模型服务,通过软件、API、授权等方式赋能本体厂商、自主系统开发商与工业客户。
SI008 投资界 / PEDAILY 首发|华为押注的机器人,极佳视界刚刚又融10亿
SI009 腾讯新闻 首发| 极佳视界三个月融资35亿,投资人开抢物理世界OpenAI DriveDreamer系列…已与多家国内头部主机厂、海外及合资主机厂,以及AI芯片、Tier 1巨头达成签约定点与量产合作,服务海内外头部主机厂与自动驾驶公司超30家。
SI010 Unitree Robotics Humanoid robot G1_Humanoid Robot Functions_Humanoid Robot Price Price(Tax and Shipping cost excluded) US $13.5K.
SI011 UnitreeRobotics Unitree G1 $13,500.00 USD. Shipping costs between $300 and $1200.
SI012 Humanoid.guide Maker H01 by GigaAI - Wheeled Humanoid Robot - Humanoid.guide $ 160 000
SI013 ChipSilicon Maker H01 | Humanoid Robot Maker H01 is a wheeled humanoid from GigaAI — pairing dual 7-DOF arms, rich sensing, and agile mobility.
SI014 RoboActu SeeLight S1 | RoboActu Cible de prix matériel sous 100 000 yuans (14 700 dollars) à l’horizon juin 2027.
SI015 HOKANEWS GigaAI Unveils SeeLight S1 Home Humanoid Robot for Real Household Tasks
SI016 ABB INTEGRATED REPORT 2025 Purchases of property, plant, equipment, and intangible assets for continuing operations amounted to $1,001 million in 2025.
SI017 HKEX UBTECH ROBOTICS CORP LTD Annual Results Announcement for the Year Ended December 31, 2025 Our revenue increased by 53.3% to RMB2,001.0 million... Our gross profit margin increased by 9 percentage points to 37.7%... Our loss decreased to RMB789.8 million.
SI018 UBTECH Robotics Financial Reports | UBTECH Robotics
SI019 AnnualReports.com iRobot 2024 Annual Report & Form 10-K (PDF) Our total revenue for fiscal 2024 was $681.8 million, declining 23.4%... Our history of operating losses and negative cash flows from operations has raised substantial doubt about our ability to continue as a going concern.
SI020 EVS Robotics How Much Does an Industrial Robot Cost? Guide (2026) Robot arm only: USD 25,000–180,000... Total deployed cell: USD 80,000–400,000... 7-year TCO: typically 1.8–2.5x initial capex.
SI021 Standard Bots ABB robot prices in 2026: Full cost breakdown for arms, cobots, and RobotStudio ABB uses distributor-based pricing that varies by region and configuration. Contact authorized ABB distributors for current quotes.
SI022 Business Wire / IFR Service Robots See Global Growth Boom – IFR reports The robot-as-a-service fleet (RaaS) has grown impressively by 31%.
SI023 GitHub GitHub - GigaAI-research/DriveDreamer: [ECCV 2024] DriveDreamer: Towards Real-world-driven World Models for Autonomous Driving
SI024 FANUC Integrated Reports - Library - Investors
SI025 ABB Annual Reporting Suite 2025 | ABB 2025 in numbers: Revenues $33,220 mn; R&D investment $1,318 mn; Operational EBITA margin 19.0%.
SE001 GigaAI 极佳科技
SE002 GigaAI GigaAI official website application bundle
SE003 National Center for Science and Technology Information (China) 极佳视界再获10亿元融资!加速布局世界模型驱动的物理AGI
SE004 投资界 / PEDAILY 首发|华为押注的机器人,极佳视界刚刚又融10亿
SE005 Pandaily GigaAI Unveils Physical AGI "Dual Pyramid" System, Targeting the Embodied Intelligence Scaling Wall
SE006 36Kr 极佳视界完成10亿元Pre-B轮融资,「世界模型」驱动通用机器人加速进入千行百业
SE007 36Kr 持续领跑世界模型驱动物理AGI,极佳视界再获10亿元B2轮融资
SE008 Gasgoo GigaAI Begins Large-Scale Deliveries
SE009 Sohu 极佳视界发布物理AGI「双金字塔」体系:数据与算法如何撑起具身智能Scaling Law?
SE010 GitHub GigaAI organization page
SE011 GitHub open-gigaai/giga-brain-0
SE012 GitHub open-gigaai/giga-world-0
SE013 GitHub open-gigaai/giga-models
SE014 GitHub open-gigaai/giga-train
SE015 GitHub open-gigaai/giga-datasets
SE016 GitHub GigaAI-research/DriveDreamer
SE017 GitHub GigaAI-research/DriveDreamer4D
SE018 arXiv GigaBrain-0: A World Model-Powered Vision-Language-Action Model
SE019 arXiv GigaWorld-0: World Models as Data Engine to Empower Embodied AI
SE020 arXiv DriveDreamer: Towards Real-world-driven World Models for Autonomous Driving
SE021 arXiv DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation
SE022 arXiv UniDriveDreamer: A Single-Stage Multimodal World Model for Autonomous Driving
SE023 GigaBrain 0.5M* project page GigaBrain-0.5M*: a VLA That Learns From World Model-Based Reinforcement Learning
SE024 GigaBrain Challenge 2026 GigaBrain Challenge 2026 @ CVPR 2026
SE025 OpenReview CVPR 2026 Workshop GigaBrain Challenge
SE026 RoboChallenge RoboChallenge
SU001 GigaAI 极佳科技
SU002 Gasgoo GigaAI Begins Large-Scale Deliveries The initial unit — a physical AGI native body dubbed Maker H01 — has been shipped to the Hubei Humanoid Robot Innovation Center.
SU003 Gasgoo Deployment of 1,000 Units in 3 Years, GigaAI Signs Another Major Partnership The first batch has already passed verification for industrial handling functions and been delivered.
SU004 Yicai Global Chinese Embodied AI, Robotics Developer GigaAI Raises USD518 Million Over Three Months
SU005 VOI GigaAI Raises USD518 Million, Chinese AI Robots Increasingly Targeted by Investors
SU006 ECARX via GlobeNewswire / FinancialContent ECARX Integrates DriveDreamer World Model into AutoGPT This strategic integration will enhance AutoGPT’s understanding of the surrounding environment and empower automakers to accelerate autonomous driving development cycles, reduce development costs.
SU007 Gasgoo With 1 billion yuan invested, Gigaai aims to become the "OpenAI of the physical world" Public information shows Gigaai has landed benchmark clients in automotive manufacturing, 3C electronics, warehousing and logistics, high-end guiding, and home terminals.
SU008 36Kr 持续领跑世界模型驱动物理AGI,极佳视界再获10亿元B2轮融资
SU009 National Center for Science and Technology Information (China) 极佳视界再获10亿元融资!加速布局世界模型驱动的物理AGI DriveDreamer…服务海内外头部主机厂与自动驾驶公司超30家。
SU010 Sohu 极佳视界发布物理AGI「双金字塔」体系:数据与算法如何撑起具身智能Scaling Law?
SU011 China Daily / Invest in China Robot housekeepers set to enter homes in trial Hubei Giga World Robot Co has announced plans to send 100 humanoid robots to ordinary households on a free trial basis.
SU012 eWeek China’s GigaAI Sends 100 Humanoid Robots Into Homes for Real-World Chore Trial Organizing a few books can take more than five minutes, while folding a single piece of clothing may require over ten minutes.
SU013 eWeek China Is Testing Household Humanoid Robots That Can Cook, Clean, and Do Laundry The company then plans to roll out free household trials in Wuhan during the first half of 2027, focusing on homes with elderly residents, children, or pets.
SU014 Interesting Engineering Chinese firm launches a humanoid robot for daily house chores
SU015 Mike Kalil Huawei-Backed GigaAI’s SeeLight S1 Home AI Robot GigaAI is providing the robots free of charge. They’ll collect feedback from tenants, monitor how the robots handle real household layouts, and fix issues before wider seed-user home trials.
SU016 Humanoids Daily Home Truths: Why the Humanoid Butler Is Further Away Than It Looks Chinese challenger GigaAI is bypassing industrial steps entirely, aiming to deploy 100 pilot units of its wheeled SeeLight S1 domestic butler to employees before expanding to public trials in 2027.
SU017 World Robot Conference 湖北人形机器人创新中心有限公司-exhibitor-2026世界机器人大会-2026世界机器人大会
SU018 The Government of Wuhan The government of Wuhan The Hubei Humanoid Robot Innovation Center is actively promoting its application in intelligent manufacturing and commercial services.
SU019 The Government of Wuhan The government of Wuhan
SU020 The Government of Wuhan The government of Wuhan
SU021 TEDA Administrative Commission Tianjin’s First Automotive Manufacturing Embodied AI Lab Unveiled in TEDA
SU022 en.wuxi.gov.cn Wuxi launches provincial embodied intelligent robot innovation center
SU023 Humanoid Press Maker H01 | GigaAI’s Physical AGI Humanoid Robot
SU024 Inc. / Fast Company The Robot Butler Boom Has Begun—and China Is Way Ahead The company claims the first 100 pilot units will be deployed at the end of this month in employees’ homes.
SU025 Modern Mechanics 24 China Launches SeeLight S1 Household Humanoid Robot
SR001 Bureau of Industry and Security Department of Commerce Revises License Review Policy for Semiconductors Exported to China
SR002 Bureau of Industry and Security BIS guidance on advanced computing exports to China-headquartered entities
SR003 Miller & Chevalier Trade Compliance Flash: Key Takeaways from New BIS Restrictions on AI Chips to China
SR004 State Council Information Office China releases national standard system for humanoid robotics and embodied AI
SR005 Cyberspace Administration of China Interim Measures for the Administration of Generative Artificial Intelligence Services
SR006 Chambers and Partners Employment 2025 - China | Global Practice Guides
SR007 DLA Piper Supreme Court clarifies employment boundaries: What employers must do re contracts and labour relationships
SR008 ADVANT Beiten China Labour Laws – Changes from 1 September 2025 – New Interpretation (II)
SR009 MIT Technology Review Why the humanoid workforce is running late Humanoids are mostly not intelligent, and adoption will likely be drawn out, industry specific, and slow.
SR010 IEEE Spectrum Humanoid Robots: The Scaling Challenge The market for humanoid robots is almost entirely hypothetical and AI is not robust enough to meet market requirements.
SR011 IEEE Robotics and Automation Society Reality Is Ruining the Humanoid Robot Hype
SR012 arXiv Humanoid Robots and Humanoid AI: Review, Perspectives and Directions
SR013 GigaAI 极佳科技
SR014 Gasgoo GigaAI Begins Large-Scale Deliveries
SR015 Yicai Global Chinese Embodied AI, Robotics Developer GigaAI Raises USD518 Million Over Three Months
SR016 Pedaily 首发|华为押注的机器人,极佳视界刚刚又融10亿
SR017 Pedaily 首发| 极佳视界,一个月融资25亿
SR018 Pedaily 极佳视界再获10亿元B2轮融资,持续领跑世界模型驱动的物理AGI,加速生产力场景规模化落地
SR019 36Kr 持续领跑世界模型驱动物理AGI,极佳视界再获10亿元B2轮融资
SR020 Zheng Zhu Zheng Zhu (朱政)
SR021 TMTPost Spatial Intelligence Company Giga AI Secures Nearly 100 Million Yuan in Financing
SR022 DigiChina Translation: Internet Information Service Algorithmic Recommendation Management Provisions
SR023 open-gigaai giga-brain-0
SR024 open-gigaai giga-world-0
SR025 open-gigaai giga-models
SR026 open-gigaai giga-train
SR027 open-gigaai giga-datasets
SR028 arXiv GigaBrain-0: World Model Generated Data Empowered Vision-Language-Action Foundation Model for Generalist Robot Learning
SR029 arXiv GigaWorld-0: Scaling Embodied AI through World Model as Data Engine
SR030 NVIDIA NVIDIA Isaac ROS
SR031 Microsoft Research AirSim research project
SR032 WIPO Lex Law of the People's Republic of China on Product Quality
SV001 Pandaily Giga AI Completes Nearly USD 137 Million Pre-B Round - Pandaily
SV002 Gasgoo Seeds | Raised 2.5 Billion Yuan in a Month, Another Embodied "Unicorn" Valuation Surpasses 10 Billion Yuan GigaAI has recently closed a B1 financing round worth nearly 1.5 billion yuan.
SV003 Yicai Global Chinese Embodied AI, Robotics Developer GigaAI Raises USD518 Million Over Three Months GigaAI has raised CNY3.5 billion (USD518 million) from investors over a three-month period.
SV004 36Kr The embodied intelligence company "Xingdong Jiyuan" recently completed a strategic round of financing worth 1 billion yuan, with its valuation exceeding 10 billion yuan. The embodied intelligence company "Xingdong Jiyuan" recently completed a strategic financing round of 1 billion yuan, with its valuation exceeding 10 billion yuan.
SV005 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.
SV006 PR Newswire Figure Raises $675M at $2.6B Valuation and Signs Collaboration Agreement with OpenAI Figure ... has raised $675M in Series B funding at a $2.6B valuation.
SV007 Sacra Figure AI valuation, funding & news
SV008 Physical Intelligence Physical Intelligence (π)
SV009 The Robot Report Physical Intelligence raises $600M to advance robot foundation models
SV010 TechCrunch Physical Intelligence is reportedly in talks to raise $1B, again Physical Intelligence ... is in discussions to raise about $1 billion in new funding at a valuation exceeding $11 billion.
SV011 Sacra Physical Intelligence valuation, funding & news
SV012 AGIBOT AGIBOT Innovation (Shanghai) Technology Co., Ltd.
SV013 Unitree Robotics 宇树科技—全球四足机器人行业开创者
SV014 Unitree Robotics Humanoid robot G1_Humanoid Robot Functions_Humanoid Robot Price Price(Tax and Shipping cost excluded) US $13.5K.
SV015 EngineAI Capital Boosts Industrial Empowerment! EngineAI Completes A1+ and A2 Funding Rounds; T800 Makes Grand Debut to Accelerate Industrialization Having secured 1 billion yuan in cumulative Pre-A++ and Series A1 funding ... [EngineAI] has now completed both Series A1+ and Series A2 financing rounds.
SV016 U.S. Securities and Exchange Commission 10-K Symbotic FY'25 (1)
SV017 U.S. Securities and Exchange Commission Symbotic Reports Second Quarter Fiscal Year 2026 Results Symbotic reported revenue of $676 million, up 23% year-over-year, and net income of $9 million.
SV018 U.S. Securities and Exchange Commission Serve Robotics Announces Fourth Quarter and Full Year 2025 Results Full year revenue of $2.7 million ... [and] 2026 revenue outlook to approximately $26 million.
SV019 U.S. Securities and Exchange Commission patr-20251231
SV020 CompaniesMarketCap Symbotic (SYM) - Market capitalization
SV021 CompaniesMarketCap Serve Robotics (SERV) - Market capitalization
SV022 Grand View Research Humanoid Robot Market Size & Share | Industry Report, 2030 The global humanoid robot market size was estimated at USD 1.55 billion in 2024 and is projected to reach USD 4.04 billion by 2030.
SV023 MIT Technology Review Why the humanoid workforce is running late Despite high valuations, revenue across the humanoid robotics space remains minimal.
SV024 Sacra Figure vs Apptronik vs Agility Robotics
SV025 Humanoid Index AgiBot: Funding, Valuation, Robot Specs & More | Humanoid Index Total Funding $83M+ (disclosed) ... Units Shipped 5,100 (2025).
SV026 Humanoid Index EngineAI: Funding, Valuation, Robot Specs & More | Humanoid Index PM01 and SE01 compact humanoids. Sub-$10K price target.
SV027 Caixin Global Unitree Fast-Tracks Shanghai IPO With Target Valuation of $6.2 Billion - Caixin Global Unitree Robotics is fast-tracking its initial public offering ... seeking a valuation of roughly 42 billion yuan ($6.2 billion).
SV028 Caixin Global Humanoid Robot Maker Unitree Advances Toward $618 Million Shanghai IPO - Caixin Global Revenue grew to 1.7 billion yuan in 2025 with net profit of 280 million yuan.
SV029 CNBC Investors bet humanoid robots will transform industry and homes over the next decade
SV030 Figure Introducing Figure 03
SV031 Physical Intelligence π0: Our First Generalist Policy
SV032 Unitree Robotics Universal humanoid robot H1_Bipedal Robot_Humanoid Intelligent Robot Company | Unitree Robotics