{
 "about": "海外 VC / 投资人自己写的工业 AI 文章（投资备忘、赛道论点、市场地图、年度预测），每篇打开原文核对过标题、日期、作者。gist 是核心观点的中文转述，不是原文；seg 对应 layers.json 的 code，all = 赛道总论。",
 "articles": [
  {
   "t": "Protecting the Machines that Keep Things Running: Why We're Deepening Our Partnership with Tractian",
   "org": "Sapphire Ventures",
   "au": "Anders Ranum、Demi Obayomi、Michael Hsu",
   "date": "2026-10-22",
   "url": "https://sapphireventures.com/blog/sapphire-ventures-deepens-tractian-partnership-series-d/",
   "kind": "投资备忘",
   "seg": [
    "c5"
   ],
   "co": [
    "Tractian"
   ],
   "gist": "Sapphire 领投 Tractian D 轮，核心理由是「从告警到答案」：旧工具只报振动异常，Tractian 还告诉你哪里坏、为什么、下一步做什么；更重要的是传感器长期挂在机器上产生的专有物理数据，互联网上没有，能让每个模型越训越准。文中称全球 500 强每年因非计划停机损失约 1.4 万亿美元、占收入约 11%。",
   "cav": "数字为 Sapphire 文中所说；Sapphire 同时是 C、D 轮领投方，有利益关系。",
   "top": true,
   "vs": "工业 AI 的数据从哪来",
   "star": true
  },
  {
   "t": "Superhuman Hardware Design Is The Easy Part",
   "org": "a16z",
   "au": "Christian Keil",
   "date": "2026-10-20",
   "url": "https://www.a16z.news/p/superhuman-hardware-design-is-the",
   "kind": "赛道论点",
   "seg": [
    "b1",
    "b4"
   ],
   "co": [],
   "gist": "a16z 认为前沿模型已能生成 CAD、设计 PCB、选型采购现成零件，超人级硬件工程师很快会来；但数字世界和物理世界之间的接口是「有损」的——把设计写成实物、把实物测量读回电脑都很难，所以真正的瓶颈在设计之后的制造、供应链和测试。",
   "cav": "文中的 GPT-6 Astra、LuxoBench 等是作者自己做的测评，非行业公认基准。",
   "top": true,
   "vs": "设计与制造哪个才是瓶颈",
   "star": true
  },
  {
   "t": "Edition 67 | 2026 — The Autonomous Factory",
   "org": "Overlook VC",
   "au": "Amir Kabir",
   "date": "2026-10-15",
   "url": "https://overlookvc.substack.com/p/edition-67-2026-the-autonomous-factory",
   "kind": "赛道论点",
   "seg": [
    "e1"
   ],
   "co": [
    "Siemens Industrial Copilot / Eigen Engineering Agent",
    "NVIDIA Omniverse / Industrial AI Cloud"
   ],
   "gist": "Overlook 的创始合伙人说，自主工厂的失败方式变了：旧工厂是响亮地停线，AI 工厂则会不停线、持续产出次品，一个坏决策能让整批货发出去才被发现，召回成本远大于停机；而且同一供应商的软件一次更新能同时波及多家工厂，保险和 ISO 9001 之类以「固定流程」为前提的质量体系都没跟上。",
   "cav": "Substack 通讯，里面的案例（西门子 Erlangen 工厂、百事 Gatorade 吞吐量提升 20%、捷豹路虎网络攻击停产约五周）为文中引述，我没逐一核对出处。",
   "top": true,
   "vs": "自主工厂是红利还是新风险",
   "star": true
  },
  {
   "t": "The Burden of Proof: Evaluating Energy and Industrial Solutions in the AI Era",
   "org": "Energize Capital",
   "au": "",
   "date": "2026-07-24",
   "url": "https://energizecap.com/insights/the-burden-of-proof-evaluating-energy-and-industrial-solutions-in-the-ai-era",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Energize 提出一套判断工业软件「会不会被 AI 替代」的框架：数据获取与管理这类旧护城河正被 AI 冲击，而行业经验加 Agent 的组合会让护城河更深；工业和能源买家容错低，更愿意依赖有行业经验的老伙伴来引入 AI。",
   "cav": "框架的具体评分项在文末，我只读到前半。",
   "vs": "老牌工业软件会不会被取代"
  },
  {
   "t": "Thoughts and Highlights on Building Vertical AI by BVP",
   "org": "Exposition Ventures",
   "au": "Carlos Furlotti",
   "date": "2026-07-01",
   "url": "https://exposition.ventures/thoughts-and-highlights-on-building-vertical-ai-bvp/",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Exposition 说多数 AI 手册不用担心软件失灵，工业不同：工厂里一次幻觉就是停线、安全事故或违规，所以它把 Bessemer 的垂直 AI 打法按工业场景改成五条原则。",
   "cav": "页面只给出摘要，五条原则的具体内容我没读到；日期来自页面元数据（07-01）。",
   "vs": "老牌工业软件会不会被取代"
  },
  {
   "t": "The Industrialist's Case for Applied AI",
   "org": "Welkin Capital",
   "au": "",
   "date": "2026-06-15",
   "url": "https://welkincapital.com/insights/appliedAI",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Welkin 认为应用 AI 的价值更多流向拥有自有工作流、干净过程数据、能把 AI 嵌进决策回路的运营方，而不是模型厂商；AI 最值钱的地方是缩短「信号到决策、决策到执行」的距离，只提升「可见度」却不改变排班、库存和资本开支的，价值会被高估。",
   "cav": "Welkin 是投资机构而非典型 VC；未署作者；引用了 McKinsey 的研究。",
   "top": true,
   "vs": "AI 的价值与护城河归谁",
   "star": true
  },
  {
   "t": "MaintainX: From Seed to $3.6B Acquisition",
   "org": "Leonis Capital",
   "au": "",
   "date": "2026-06-08",
   "url": "https://www.leoniscap.com/research/maintainx-from-seed-to-3.6b-acquisition",
   "kind": "投资备忘",
   "seg": [
    "c1",
    "c5"
   ],
   "co": [
    "MaintainX"
   ],
   "gist": "Leonis 复盘 MaintainX 七年：制造业几乎什么都数字化了，唯独一线没有；做成这件事要从一线工人切入、一个班次一个班次赢得信任，而不是从仪表盘；工单、资产历史、故障规律形成的非结构化数据是复利护城河，也正是 Autodesk 出 36 亿美元收购的原因。",
   "cav": "Leonis 是早期投资人，文章带复盘叙事的立场；36 亿美元收购金额为文中所说。",
   "vs": "AI 先替哪一层的活"
  },
  {
   "t": "Industrial AI: The New Wave of Innovation Hitting Heavy Industry",
   "org": "Energize Capital",
   "au": "",
   "date": "2026-06-02",
   "url": "https://energizecap.com/insights/industrial-ai-deep-dive",
   "kind": "市场地图",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Energize 用缺工（2025 年 8 月美国约 40.9 万制造岗位空缺）、新产能投资、停机成本上涨 2 到 4 倍和联网成本下降解释工业 AI 为什么现在起量，并沿设计与仿真、工厂运营、自动化三段列出它看好的公司；文中附一张市场地图。",
   "cav": "数字为文中所说；公司清单在图里，正文未全部展开。",
   "top": true,
   "star": true
  },
  {
   "t": "Building the first intelligence layer for industrial assets: our investment in Scope",
   "org": "Index Ventures",
   "au": "Stephane Kurgan、Bastian Hasslinger",
   "date": "2026-05-19",
   "url": "https://www.indexventures.com/perspectives/building-the-first-intelligence-layer-for-industrial-assets-our-investment-in-scope/",
   "kind": "投资备忘",
   "seg": [
    "c1"
   ],
   "co": [
    "Scope"
   ],
   "gist": "Index 投 Scope，看中创始人扎进检验认证（TIC）行业亲自跟检的做法：产品先替检验员省掉纸笔和手工报告，因为这个行业每十年有四成检验员退休、补一个岗位平均要十个月；文中称前十大检验公司里已有六家的检验员在用，ARR 增长 9 倍、试点转化率 100%。",
   "cav": "数字为文中所说，由 Index 披露；检验认证属于资产检验，和制造业有交叉但不完全重合。"
  },
  {
   "t": "The Generational Opportunity for Industrial Software",
   "org": "Energize Capital",
   "au": "",
   "date": "2026-05-07",
   "url": "https://energizecap.com/insights/the-generational-opportunity-for-industrial-software",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Energize 认为美国制造投资自 2021 年以来增长约三倍、电网老化叠加 AI 算力需求，买方对工业软件的紧迫感前所未有；它把「工业 AI」看成工业软件里的新一类——把推理、预测和自动化直接嵌进工业工作流，但工业追求零故障，所以采用慢、试点长。",
   "cav": "Energize 重点是能源与工业，文中例子多来自电网。"
  },
  {
   "t": "From Bits to Atoms: Why We're Backing AI for the Physical World",
   "org": "Love Ventures",
   "au": "",
   "date": "2026-04-02",
   "url": "https://loveventures.co.uk/insights/from-bits-to-atoms-why-we-re-backing-ai-for-the-physical-world",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Love Ventures 认为模型一个周末就能复制 SaaS 的核心功能，应用层的护城河在被侵蚀；AI 让数字产出变便宜，稀缺的是物理工作，所以下一批有防御力的软件会长在工厂、基础设施等现实世界里。",
   "cav": "英国早期基金；引用了 Eclipse 合伙人的话，页面未署作者。",
   "vs": "AI 的价值与护城河归谁"
  },
  {
   "t": "The mindset shift to rebuild American factories",
   "org": "a16z American Dynamism",
   "au": "Erin Price-Wright",
   "date": "2026-02-17",
   "url": "https://www.a16z.news/p/the-mindset-shift-we-need-to-rebuild",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "a16z 合伙人在密歇根大学 Tauber 运营大会上的主旨演讲：美国制造业几十年「梦游」后，所有问题都绕回「怎么把制造业做好」；她把「工厂」定义成一套从第一性原理拆解、自动化、找瓶颈再重复的系统思维，号召创业者用 AI 去设计和运营工厂。",
   "cav": "这是演讲稿，我读到的是前半段，后半段关于具体 AI 用法的部分未逐字核对。",
   "top": true,
   "star": true
  },
  {
   "t": "The Intelligent Manufacturing Investment Opportunity: Why \"Atoms\" Beat \"Bits\" for Sophisticated Capital",
   "org": "Encore Industrial",
   "au": "",
   "date": "2025-12-30",
   "url": "https://www.encoreindustrial.co/post/the-intelligent-manufacturing-investment-opportunity-why-atoms-beat-bits-for-sophisticated-capital",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Encore 认为风险投资长期把钱投给软件而不是实体技术，智能制造被严重欠投：湾区 2024 年拿走美国风投 57%，文中称智能制造的风投交易数从 2019 年的 276 笔降到 2023 年的 192 笔；懂制造工艺又愿意做难题的投资人有结构性机会。",
   "cav": "Encore Industrial 是产业运营方，不是 VC；数字为文中引用，原始数据来源在文末脚注，我未核。"
  },
  {
   "t": "Big Ideas 2026: Part 2（American Dynamism 一节：Building the AI-native industrial base）",
   "org": "a16z American Dynamism",
   "au": "David Ulevitch",
   "date": "2025-12-10",
   "url": "https://a16z.com/newsletter/big-ideas-2026-part-2/",
   "kind": "年度预测",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "a16z 判断 2026 年能源、制造、物流、基建会被「AI 原生、软件优先」的新公司重建，这些公司从仿真、自动设计和 AI 驱动的运营起步，而不是给老流程打补丁；机会点在先进能源、重机器人制造、新一代采矿等。",
   "cav": "同一篇里工业部分只占一节，没有点具体软件公司名；Ulevitch 的署名按 a16z 该节标题，页面正文未单独标作者。",
   "top": true,
   "vs": "设计与制造哪个才是瓶颈",
   "star": true
  },
  {
   "t": "Vision-native AI opportunities: a precursor to intelligent robotics",
   "org": "Bessemer",
   "au": "Bhavik Nagda、Talia Goldberg",
   "date": "2025-12-02",
   "url": "https://www.bvp.com/atlas/vision-native-ai-opportunities-a-precursor-to-intelligent-robotics",
   "kind": "赛道论点",
   "seg": [
    "c4"
   ],
   "co": [],
   "gist": "Bessemer 认为视觉语言模型已能直接理解物理世界，在机器人动手之前，先有一大片「视觉原生软件」机会：用摄像头等采集现场画面、在边缘到云端做推理，从文档处理一直做到各类自动驾驶式的监控与自动化。",
   "cav": "主题是物理 AI 的前一站，工业质检只是其中一类应用，文章不是专讲工厂。"
  },
  {
   "t": "Litmus Secures Additional Investment Led by Insight Partners to Accelerate Industrial AI",
   "org": "Insight Partners",
   "au": "Josh Fredberg",
   "date": "2025-11-12",
   "url": "https://www.insightpartners.com/ideas/litmus-secures-additional-investment-led-by-insight-partners-to-accelerate-industrial-ai/",
   "kind": "投资备忘",
   "seg": [
    "a1"
   ],
   "co": [
    "Litmus"
   ],
   "gist": "Insight 的董事总经理 Josh Fredberg 称，让工业 AI 在碎片化数据上跑起来是制造业最难的事，Litmus 是少数真正打通 OT 与 IT 数据并转化成可衡量结果的平台；Litmus 称有 250 多个开箱即用连接器。",
   "cav": "这是新闻稿转载在 Insight 官网，观点只有 Josh Fredberg 的一句引语，算最弱的一篇。",
   "vs": "工业 AI 的数据从哪来"
  },
  {
   "t": "Putting People at the Center of Industrial Digitalization – Why We Invested in Augmented Industries",
   "org": "b2venture",
   "au": "",
   "date": "2025-09-30",
   "url": "https://www.b2venture.vc/stories/putting-people-at-the-center-of-industrial-digitalization---why-we-invested-in-augmented-industries",
   "kind": "投资备忘",
   "seg": [
    "c1"
   ],
   "co": [
    "Augmented Industries"
   ],
   "gist": "b2venture 认为工厂里技术员的经验不可替代，而传统课堂和 e-learning 培训太慢太泛，造成停机和风险；投 Augmented Industries 是押「以人为中心」的一线数字化——用技术持续给工人培训并保证合规和安全。",
   "cav": "页面日期是正文里的 2025-09-30。"
  },
  {
   "t": "Software for Hardware",
   "org": "General Catalyst",
   "au": "",
   "date": "2025-07-21",
   "url": "https://www.generalcatalyst.com/stories/software-for-hardware",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [
    "Senra Systems"
   ],
   "gist": "GC 把「Software for Hardware」当作工业复兴的核心类别：为物理世界从头设计的软件和系统，让硬件团队用上软件式的迭代；它看好既懂物理又懂 AI 的创始人，从工厂车间到海面都要有软件定义的平台，并以 Senra Systems（线束制造流程重做）为例。",
   "cav": "页面没署作者；一半篇幅讲国防和硬件平台（Anduril、Helsing 等），不全是工厂软件。",
   "top": true,
   "star": true
  },
  {
   "t": "Europe's Industrial Moment: Why AI Automation is the Key to Revitalizing Manufacturing Competitiveness",
   "org": "b2venture",
   "au": "Anna Bosch",
   "date": "2025-07-02",
   "url": "https://www.b2venture.vc/stories/europes-industrial-moment-why-ai-automation-is-the-key-to-revitalizing-manufacturing-competitiveness",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [
    "Soff",
    "Mandel AI",
    "Didero",
    "Augmented Industries",
    "ProGlove",
    "Celonis"
   ],
   "gist": "b2venture 说欧洲制造业占 GDP 约 15%，但利润被劳动力和能源成本挤压、出口份额从 2000 年的 21% 降到 2023 年的 17%，缺人也越来越严重；AI 自动化是提升竞争力的关键，并按投产前、生产、投产后画了创业公司地图。",
   "cav": "Anna Bosch 的署名来自我方底稿，页面本身未显示作者；数字为文中所说。",
   "top": true,
   "star": true
  },
  {
   "t": "Our Investment in Pelico",
   "org": "General Catalyst",
   "au": "",
   "date": "2025-06-18",
   "url": "https://www.generalcatalyst.com/stories/our-investment-in-pelico",
   "kind": "投资备忘",
   "seg": [
    "c2"
   ],
   "co": [
    "Pelico"
   ],
   "gist": "GC 领投 Pelico 4000 万美元战略融资，判断制造业缺的是一个实时决策层：把 ERP、MES、TMS 和车间传感器的碎片数据接到一起，让工厂团队发现中断、查根因、协同响应；强调它不是外挂分析工具，而是嵌进工厂工作流，并带仿真引擎让计划员试算方案。",
   "cav": "金额 4000 万美元为文中所说。",
   "vs": "工业 AI 的数据从哪来"
  },
  {
   "t": "AI is Rewriting the Rules of Supply Chain",
   "org": "Activant Capital",
   "au": "Rebecca Rodseth",
   "date": "2025-06-17",
   "url": "https://activantcapital.com/research/ai-is-rewriting-the-rules-of-supply-chain",
   "kind": "市场地图",
   "seg": [
    "c2"
   ],
   "co": [
    "Auger",
    "Raft",
    "Pallet",
    "Prewave",
    "Altana"
   ],
   "gist": "Activant 认为供应链碎片化、数据八成以上是非结构化的，AI 消除不了这种碎片，但能让企业更会管理它；把玩家分成 AI 编排者、AI 物流操作系统、AI 副驾三类。",
   "cav": "偏物流和供应链，对应图谱 c2 的外延。"
  },
  {
   "t": "Partnering with Nominal: Powering the Next Era of Hardware Engineering",
   "org": "Sequoia Capital",
   "au": "Anas Biad、Alfred Lin",
   "date": "2025-06-12",
   "url": "https://sequoiacap.com/article/partnering-with-nominal-powering-the-next-era-of-hardware-engineering",
   "kind": "投资备忘",
   "seg": [
    "b3"
   ],
   "co": [
    "Nominal"
   ],
   "gist": "Sequoia 认为再工业化、国防开支和 SpaceX 节奏逼硬件团队从「原型→测试→量产」的线性流程，变成像软件一样在设计和运营全程持续测试、持续改进，而硬件一直缺这套软件栈，Nominal 想补上。",
   "cav": "Nominal 做的是硬件测试数据和验证软件，对应「工程数据」多于「工厂 AI」。"
  },
  {
   "t": "BackOps: The Connective Intelligence Layer Powering the Modern Supply Chain",
   "org": "Construct Capital",
   "au": "",
   "date": "2025-06",
   "url": "https://www.constructcap.com/articles/backops-the-connective-intelligence-layer-powering-the-modern-supply-chain",
   "kind": "投资备忘",
   "seg": [
    "c2",
    "e1"
   ],
   "co": [
    "BackOps"
   ],
   "gist": "Construct 领投 BackOps 600 万美元种子轮，判断供应链后台的痛点是「异常处理」：一个托盘放错就要在 WMS、TMS、ERP、短信和电话间来回倒腾；BackOps 做的是连接这些系统和沟通渠道的智能层，覆盖第三方物流、制造和物流后台。",
   "cav": "页面没标日期，2025-06 按融资公告日期（2025-06-24）推断；金额为文中所说。",
   "vs": "AI 先替哪一层的活"
  },
  {
   "t": "Unlocking Value Through Industrial AI Agents",
   "org": "b2venture（与 Hitachi Ventures、AENU 联合）",
   "au": "",
   "date": "2025-05-22",
   "url": "https://b2venture.vc/stories/unlocking-value-through-industrial-ai-agents",
   "kind": "市场地图",
   "seg": [
    "e1"
   ],
   "co": [],
   "gist": "三家投资机构共同判断，SCADA 和传感器产生了海量数据，规则型控制系统缺乏推理和自适应，工业 AI Agent 是补这个「能动性缺口」的下一步；演进分三阶段（第一年垂直嵌入、第二年跨垂直即插即用、之后平台生态），能「从感知到决策再到行动」拥有全闭环的创业公司护城河最深，但现实中多数 Agent 仍是子系统。",
   "cav": "文中引用 Gartner 预测（2028 年三分之一企业应用内嵌 Agent）；附一张工业 AI Agent 市场地图，见 /maps。",
   "top": true,
   "vs": "自主工厂是红利还是新风险",
   "star": true
  },
  {
   "t": "Sense, Think, Act — How AI is Remaking Manufacturing: Part I",
   "org": "Construct Capital",
   "au": "Camila Key Saruhashi、Leland Chamlin",
   "date": "2025-04-08",
   "url": "https://www.constructcap.com/articles/sense-think-act-how-ai-is-remaking-manufacturing-part-i",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "Construct 用机器人学的「感知—思考—行动」框架给制造业 AI 分层：传感器和摄像头成本大降、64% 的工业企业上云，但数据仍然碎片化，AI 正在补上过去自动化里数据整合、决策、执行之间的缺口；它认为美国制造业欠投资多年，现在是一次「跳过几十年停滞」的机会。",
   "cav": "页面没标发布日期，2025-04 来自搜索索引；这是 Part I，我读了开头和框架部分。",
   "top": true,
   "star": true
  },
  {
   "t": "Why we Invested in Tractian",
   "org": "Upload",
   "au": "Alfredo Hahn",
   "date": "2025-02-10",
   "url": "https://alfredohahn.substack.com/p/why-we-invested-in-tractian",
   "kind": "投资备忘",
   "seg": [
    "c5"
   ],
   "co": [
    "Tractian"
   ],
   "gist": "Upload 跟投 Tractian 1.2 亿美元 C 轮（领投方 Sapphire、General Catalyst、Next47），称它用传感器加 AI 算法加自动化做软硬一体的资产管理，是工业的「Copilot」，目标零停机。",
   "cav": "发表在合伙人 Alfredo Hahn 的 Substack「LatAm Lens」，是跟投方的短文。"
  },
  {
   "t": "Rebuilding Western Manufacturing's Edge: Why We're Leading Trace.Space's $4M Seed Round",
   "org": "Cherry Ventures",
   "au": "",
   "date": "2025-02-06",
   "url": "https://cherry.vc/news/rebuilding-western-manufacturings-edge-why-were-leading-tracespaces-4m-seed-round",
   "kind": "投资备忘",
   "seg": [
    "b3"
   ],
   "co": [
    "Trace.Space"
   ],
   "gist": "Cherry 领投 Trace.Space 400 万美元种子轮：汽车等复杂产品动辄几十万条需求，要和多家供应商协同，旧的需求管理系统在 AI 时代已不够用；Trace.Space 用 AI 秒级处理需求并把供应商协作从几天缩到几分钟，帮西方制造商保住竞争力。",
   "cav": "金额与场景为文中所说。"
  },
  {
   "t": "Revolutionizing manufacturing – why we invested in assemblean",
   "org": "b2venture",
   "au": "",
   "date": "2025-02-04",
   "url": "https://b2venture.vc/stories/revolutionizing-manufacturing---why-we-invested-in-assemblean",
   "kind": "投资备忘",
   "seg": [
    "c3"
   ],
   "co": [
    "assemblean"
   ],
   "gist": "b2venture 看欧洲制造商受能源和成本压力，按需制造平台又撑不住多零件、大批量产品，找到可靠供应商很费时；assemblean（德国帕德博恩）用「生产即服务」平台承接工业级产品的大规模按需制造。",
   "cav": "assemblean 本质是制造服务平台，软件和产能撮合并重，和纯软件公司有区别。"
  },
  {
   "t": "Revolutionizing Manufacturing with AI: Why We're Thrilled to Lead Tractian's Series C",
   "org": "Sapphire Ventures",
   "au": "Anders Ranum、Demi Obayomi、Adam Liu",
   "date": "2024-12-05",
   "url": "https://sapphireventures.com/blog/revolutionizing-manufacturing-with-ai-why-were-thrilled-to-lead-tractians-series-c/",
   "kind": "投资备忘",
   "seg": [
    "c5"
   ],
   "co": [
    "Tractian"
   ],
   "gist": "Sapphire 相信「传感器加 AI」的预测性维护是改造工业运营的关键，工厂里一台关键设备意外停机每分钟损失数千美元，传统维护靠人工巡检，容易漏掉早期预警；因此领投 Tractian C 轮。",
   "cav": "领投方自己写的备忘。"
  },
  {
   "t": "Spearheading an AI-driven industrial revolution – why we invested in Tractian",
   "org": "NGP Capital",
   "au": "Debjit Mukerji",
   "date": "2024-12-05",
   "url": "https://www.ngpcap.com/insights/spearheading-an-ai-driven-industrial-revolution---why-we-invested-in-tractian",
   "kind": "投资备忘",
   "seg": [
    "c5"
   ],
   "co": [
    "Tractian"
   ],
   "gist": "NGP 认为传感器实时信息、更强的 AI 算法和更好的控制自动化能同时提升产量、质量和效率；非计划停机是最关键的效率指标，文中引西门子报告称大型企业停机成本可超年收入 10%，一小时停产近 100 万美元。",
   "cav": "数字引自文中转引的西门子报告。"
  },
  {
   "t": "Valdera Secures $15 Million to Help Manufacturers Make the Next Generation of Products Faster, Safer and More Sustainably",
   "org": "Index Ventures",
   "au": "Georgia Stevenson、Jan Hammer",
   "date": "2024-10-03",
   "url": "https://www.indexventures.com/perspectives/valdera-secures-15-million-to-help-manufacturers-make-the-next-generation-of-products-faster-safer-and-more-sustainably-than-ever-before/",
   "kind": "投资备忘",
   "seg": [
    "c3"
   ],
   "co": [
    "Valdera"
   ],
   "gist": "Index 领投 Valdera 1500 万美元 A 轮，看好用 AI 加实时市场数据做原材料和化学品的供应商发现、资质审核和关系维护，帮制造商更快上新品、少断货。",
   "cav": "Valdera 做的是原材料采购，偏化工和消费品制造商，不是离散制造的 RFQ。"
  },
  {
   "t": "Call for Startups",
   "org": "Greylock",
   "au": "",
   "date": "2024-07-30",
   "url": "https://greylock.com/greymatter/call-for-startups/",
   "kind": "赛道论点",
   "seg": [
    "b1"
   ],
   "co": [],
   "gist": "Greylock 公开列出各合伙人想投的方向，其中包括面向工业、制造、建筑的生成式 3D / CAD 设计，说明它把「物理世界的设计软件」当作 AI 机会之一。",
   "cav": "这是 Greylock 的全公司征集清单，工业只占其中一两条，页面按条列出各合伙人，不是专论。"
  },
  {
   "t": "Tailwinds of Change: Reengineering Hardware Design and Manufacturing",
   "org": "Menlo Ventures",
   "au": "",
   "date": "2024-04-04",
   "url": "https://menlovc.com/perspective/tailwinds-of-change-reengineering-hardware-design-and-manufacturing/",
   "kind": "市场地图",
   "seg": [
    "b1",
    "b2",
    "b3",
    "b4",
    "c1"
   ],
   "co": [
    "CoLab Software",
    "Duro",
    "Allspice",
    "Flux",
    "Quilter",
    "Zoo",
    "PhysicsX",
    "Siml.ai",
    "Dirac",
    "First Resonance",
    "Fulcrum",
    "Copia"
   ],
   "gist": "Menlo 认为硬件工程的软件栈四十年没变，被 80 年代的老厂垄断，设计评审还靠 PDF；云迁移、软件人才进入 SpaceX 式公司、回流政策叠加，加上 AI 当「配料」，新一代硬件工程软件有机会。盯三件事：打破封闭设计环境、用生成式 AI 简化设计测试制造、MES 上云。",
   "cav": "页面署名是 Menlo Ventures 而非个人；文章是 2024 年初，当时 AI 仿真公司还很早期。",
   "top": true,
   "vs": "老牌工业软件会不会被取代",
   "star": true
  },
  {
   "t": "Industrial Resilience",
   "org": "General Catalyst",
   "au": "",
   "date": "2024-02-02",
   "url": "https://www.generalcatalyst.com/stories/industrial-resilience",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [
    "Tractian"
   ],
   "gist": "GC 的「工业韧性」论点：西方把制造外包后，知识、数据和知识产权也跟着流失，现在需要重建工业基础；AI 的突破让软件有机会重塑自动化栈，文中用 Tractian 作为资产健康和维护领域的例子。",
   "cav": "文章同时覆盖国防、造船、能源等，工业软件只是其中一部分。"
  },
  {
   "t": "Our Investment in Tractian",
   "org": "General Catalyst",
   "au": "",
   "date": "2023-08-07",
   "url": "https://www.generalcatalyst.com/stories/our-investment-in-tractian",
   "kind": "投资备忘",
   "seg": [
    "c5"
   ],
   "co": [
    "Tractian"
   ],
   "gist": "GC 看好 Tractian 用智能传感器采集设备实时数据、预测机械和电气故障，把大多数工厂「坏了才修」的被动维护改成预防；认为这家巴西创始团队有机会改写维护人员的角色。",
   "cav": "这是 2023 年 Tractian B 轮时的文章，不是后来的 C/D 轮。"
  },
  {
   "t": "Laminar: Enabling Data Driven Liquid Manufacturing",
   "org": "Construct Capital",
   "au": "",
   "date": "2023-07",
   "url": "https://www.constructcap.com/articles/laminar-july-2023",
   "kind": "投资备忘",
   "seg": [
    "c5",
    "d1"
   ],
   "co": [
    "Laminar"
   ],
   "gist": "Construct 投 Laminar（原 H2Ok Innovations）种子轮，看好液体生产和监测还停留在间歇、过时的方式，用联网软件加定制液体传感器可降成本、提收入、保安全并省能。",
   "cav": "页面没标具体日期，月份来自网址；这是 Construct 较早期的文章，AI 含量不高。"
  }
 ],
 "debates": [
  {
   "q": "AI 时代，硬件产品的瓶颈在设计端还是在设计之后的制造端？",
   "a": "口径差在：口径不同在「从哪端算起」：a16z 的 Big Ideas 是 2025 年底对创业方向的展望，把设计仿真当起点；Superhuman 是 2026-10 对模型能力的实测，指出写回物理世界才是难处。同一家机构的两篇，一篇看起点、一篇看卡点；Menlo 则是 2024 年的盘点，当时 AI 仿真还很早期，没有区分设计与落地谁更难。",
   "refs": [
    "https://a16z.com/newsletter/big-ideas-2026-part-2/",
    "https://www.a16z.news/p/superhuman-hardware-design-is-the",
    "https://menlovc.com/perspective/tailwinds-of-change-reengineering-hardware-design-and-manufacturing/"
   ]
  },
  {
   "q": "工业 AI 做出来以后，价值和护城河留在模型、软件公司，还是留在拥有数据和现场的人手里？",
   "a": "口径差在：他们谈的「谁」不同：Welkin 站在投资运营方的角度（买 AI 的人赚钱），Love 和 Energize 站在投软件公司的角度（软件靠什么不被复制），Leonis 是复盘一家已被收购的公司。Welkin 并没说软件公司没价值，只说价值落点在运营方；Love 与 Energize 的分歧在于护城河到底是「物理世界本身」还是「行业经验加 Agent」。",
   "refs": [
    "https://welkincapital.com/insights/appliedAI",
    "https://loveventures.co.uk/insights/from-bits-to-atoms-why-we-re-backing-ai-for-the-physical-world",
    "https://energizecap.com/insights/the-burden-of-proof-evaluating-energy-and-industrial-solutions-in-the-ai-era",
    "https://www.leoniscap.com/research/maintainx-from-seed-to-3.6b-acquisition"
   ]
  },
  {
   "q": "让 AI Agent 闭环地感知、决策、执行，对工厂是效率红利，还是一种新的失败方式？",
   "a": "口径差在：b2venture 看的是能力和商业护城河（谁拥有完整闭环），并承认现实里多数 Agent 仍是子系统；Overlook 看的是失败模式（出错后谁发现、谁赔），两边没有直接反驳对方，是在回答不同问题。Overlook 还点出保险与 ISO 9001 这类以固定流程为前提的体系没有跟上。",
   "refs": [
    "https://b2venture.vc/stories/unlocking-value-through-industrial-ai-agents",
    "https://overlookvc.substack.com/p/edition-67-2026-the-autonomous-factory"
   ]
  },
  {
   "q": "工业 AI 的数据优势，是靠自己装传感器攒专有数据，还是靠接入现有 ERP / MES / OT 系统？",
   "a": "口径差在：口径不同在赛道：Sapphire 说的是预测性维护，这类场景现成数据稀缺，必须自己采；GC 和 Insight 说的是计划、数据平台，数据已躺在 ERP / MES / 历史库里，难点是打通。Insight 一篇只有一句引语，分量比另两篇轻。同一个 GC 另外还在 Tractian 上押过自带传感器的路线。",
   "refs": [
    "https://sapphireventures.com/blog/sapphire-ventures-deepens-tractian-partnership-series-d/",
    "https://www.generalcatalyst.com/stories/our-investment-in-pelico",
    "https://www.insightpartners.com/ideas/litmus-secures-additional-investment-led-by-insight-partners-to-accelerate-industrial-ai/"
   ]
  },
  {
   "q": "工业软件的老厂（CAD、MES、自动化软件）会被 AI 原生新公司取代，还是靠行业经验继续守住买家？",
   "a": "口径差在：Menlo 站在设计和工程软件，那里老厂的产品体验确实落后；Energize 和 Exposition 站在工厂运营和能源，那里可靠性压倒一切。分歧的根源是被替换环节的容错成本不同。Menlo 是 2024 年的文章，之后西门子买 Altair、Synopsys 买 Ansys 等大并购说明老厂在整合而非被撇开，这是我们的旁注，不是文章里的观点。",
   "refs": [
    "https://menlovc.com/perspective/tailwinds-of-change-reengineering-hardware-design-and-manufacturing/",
    "https://energizecap.com/insights/the-burden-of-proof-evaluating-energy-and-industrial-solutions-in-the-ai-era",
    "https://exposition.ventures/thoughts-and-highlights-on-building-vertical-ai-bvp/"
   ]
  },
  {
   "q": "AI 最先、最该替的是一线操作员，还是后台办公室和工程环节？",
   "a": "口径差在：Leonis 和 b2venture 把工人当入口，是「增强人」；Construct 的 BackOps 把后台人员的重复协调当对象，更接近「替人」。他们面对的岗位不同、信任成本不同：一线要先赢得信任，后台系统间搬运信息的工作更容易直接由 Agent 接手。这一条的两边是不同岗位，不算严格对立。",
   "refs": [
    "https://www.leoniscap.com/research/maintainx-from-seed-to-3.6b-acquisition",
    "https://www.b2venture.vc/stories/putting-people-at-the-center-of-industrial-digitalization---why-we-invested-in-augmented-industries",
    "https://www.constructcap.com/articles/backops-the-connective-intelligence-layer-powering-the-modern-supply-chain"
   ]
  }
 ]
}