VCs Target Enterprise Labor Spend as Big Tech Locks AI Compute

科技巨头锁定AI算力，风投资本瞄准企业人力支出 随着科技巨头垄断算力基础设施，风险投资正转向旨在替代企业劳动力的垂直AI智能体 风险投资人与企业创新领袖探讨了AI市场的演进趋势：行业正从浅层的软件套壳转向专业的垂直应用及自主智能体，直接角逐数万亿美元的人力预算，而非局限于传统的IT支出。讨论嘉宾强调，创始人的坚韧毅力、深厚的行业专长以及严密的企业流程梳理，是在供应商整合与在位企业竞争中脱颖而出的关键。尽管底层算力基础设施正向资金雄厚的巨头集中，但发言者总结认为，AI原生初创企业相比传统在位企业具备结构性优势，未来二十年最具决定性意义的垂直应用巨头目前尚未诞生。 AI 现已占据全球 53% 和美国 64% 的风投资金，市场正快速从缺乏防御壁垒的“套壳应用”（Thin Wrappers）转向由专有数据和垂直行业深耕驱动的垂直领域 AI（Vertical AI）。 (3:16) AI 正在将软件的潜在市场空间（TAM）从传统的企业 IT 预算（占营收的 5%–10%）大幅扩张至数万亿美元规模的企业人力与劳动力支出（占营收的 30%–40%）。 (8:11) 大型企业规模化落地自主 AI Agent 的关键瓶颈并非底层模型能力，而在于企业内部业务流程梳理（Process Mapping）与组织就绪度。 (14:29) 早期投资人对创始人的韧性、可指导性以及问题洞察力的重视程度已超过单纯的技术原型或 MVP。 (17:42) 底层基础设施层已被巨额资本与老牌巨头高度垄断，投资人预计未来二十年最具商业价值的 AI 巨头将诞生于垂直应用和自主智能体层。 (25:16)

Venture and enterprise investors argue that artificial intelligence is shifting from generic wrapper interfaces to domain-specific vertical applications and autonomous agents that target enterprise labor spend rather than traditional software budgets. While incumbents hold deep moats in foundational infrastructure, panelists suggest the next era's most valuable companies will be AI-native operators that successfully navigate corporate integration hurdles. TAM Expansion from Software Budgets to Labor (8:11) Artificial intelligence represents an unprecedented addressable market expansion because it targets human labor rather than legacy IT budgets. Ray Wu of Alumni Ventures calculates that traditional enterprise software and IT infrastructure historically consume 5% to 10% of corporate revenue, whereas employee compensation and operational workforce spend account for 30% to 40%. By serving as co-pilots and autonomous task execution systems, AI taps directly into this multi-trillion-dollar global labor pool. Mohammad Islam of Threshold Ventures and Ray Wu both note that while previous computing transitions focused on translating human logic into machine-readable code, modern AI utilizes natural language as the primary computing interface. This transition allows machines to assume computational and operational workloads, theoretically liberating human employees to focus on higher-order creative output, as also emphasized by Dr. Fei-Fei Li. Shift from Thin Wrappers to Vertical Moats (3:16) Amid an extraordinary concentration of venture capital—with Dan Primack's Axios data showing AI capturing 53% of global and 64% of US venture dollars—investors observe a swift maturation away from superficial software layers. Ashlyn Lackey of Prudential highlights that initial startups merely provided thin wrappers around models like OpenAI's ChatGPT or Anthropic's Claude. Defensibility now requires proprietary models integrated with specialized vertical knowledge and veteran enterprise go-to-market leadership. Serephina Ha of Founder Institute reports that the strongest early-stage founders leverage preexisting distribution and deep industry credibility. Examples include a former MarTech executive launching a platform with access to 44 million advertising users and a Chief Compliance Officer designing regulatory AI workflows. Both Ha and Lackey argue that founder resilience, coachability, and domain expertise outweigh generic technical prototypes in securing early-stage backing. Corporate Process Bottlenecks and Vendor Consolidation (14:29) Despite intense optimism around autonomous agent deployment, large organizations face substantial structural roadblocks. Ashlyn Lackey observes that enterprise workflows are rarely mapped well enough to allow agents to operate effectively. While she labels these hurdles as operational opportunities to reduce overhead and improve customer experiences, realization of AI agent ROI depends directly on corporate process engineering rather than model intelligence alone. Simultaneously, startups encounter severe enterprise sales headwinds. Lackey points out that corporate buyers are tightening procurement budgets, consolidating third-party software vendors, and extracting value from existing enterprise contracts rather than onboarding unproven point solutions. Consequently, Serephina Ha stresses that emerging startups cannot succeed in isolation; they must form strategic partnerships and collaborative ecosystem networks to bypass Big Tech's entrenchment in enterprise software distribution. Creative Destruction: AI-Native Versus Legacy Incumbents (21:43) Comparing current market dynamics to historical disruptions—such as online retail against traditional storefronts or streaming video against video rental chains—Ray Wu distinguishes between AI-native startups and legacy AI-enabled incumbents. While traditional corporations attempt to bolt artificial intelligence onto rigid legacy architectures, AI-first companies operate with structurally leaner cost foundations and autonomous workflows built from inception on foundational model loops. However, this cycle differs from previous platform transitions because mega-cap incumbents recognized the shift immediately, with the "Magnificent Seven" amassing roughly 15% of global equity market capitalization. Wu contends that while base infrastructure (compute, hardware, and foundation models) has largely consolidated around heavily capitalized players like OpenAI, Anthropic, and CoreWeave, the vertical application and autonomous agent layer remains entirely open, leaving room for unlaunched startups to become future market leaders. While panelists broadly agree that the single most valuable AI corporation of the next two decades has likely not yet been founded, early-stage contenders must overcome significant near-term friction. Entrenched cloud incumbents dominate capital-intensive infrastructure, and enterprises remain constrained by poorly mapped internal processes and vendor consolidation.

在Ai4研讨会上，来自Threshold Ventures、Alumni Ventures、Prudential与Founder Institute的投资人与行业高管指出，生成式AI正从缺乏壁垒的浅层包装转向具备深厚垂直行业知识与分发渠道的专属应用。各方认为，AI正在从争夺传统IT软件预算扩展至替代或辅助企业人力劳动池，尽管底层基础设施已被先行巨头高度垄断，但未来最具价值的企业将诞生于垂直行业自主智能体层。 资本演化：从浅层包装转向垂直深耕 (3:16) 讨论引用Axios记者Dan Primack的数据指出，AI投资已占全球风险投资资金的53%以及美国风投的64%，硅谷更有近三分之二的资本集中于此。Prudential新兴技术团队的Ashlyn Lackey认为市场正显著走向成熟：早期依托ChatGPT或Claude搭建的无防御性浅层界面（thin wrappers）正被淘汰，创始人转向利用垂直行业专门知识构建专属模型。与此同时，针对此前学术突破与企业实际销售脱节的问题，初创公司开始吸纳资深业务拓展主管以适应复杂的企业采购周期。 Founder Institute董事总经理Serephina Ha强调，具备垂直领域认知和既有客户分发能力的创始人具有明显优势。她举例称，有软银注资17亿美元的营销科技公司前CTO创业后可直接触达4400万广告受众，资深合规官创业也能依托既有行业信任快速部署软件。Ha预测垂直专家的壁垒在未来一至五年内依然坚固，但也指出量子计算等新型计算平台的商业化未来可能会再次重构技术格局。 市场空间扩张：从软件预算切入人力支出 (8:11) Alumni Ventures管理合伙人Ray Wu结合其横跨编译器开发、互联网与移动浪潮的投资经验分析，以往的技术范式旨在将人类语言翻译为机器代码，而现代AI直接将自然语言转化为操作界面。在市场规模上，传统企业IT与软件支出历来仅占总营收的5%至10%，而员工薪酬与人力运营支出高达30%至40%。AI副驾驶与自主智能体通过直接切入数万亿美元的人力劳动服务市场，使整体潜在市场空间（TAM）实现了数量级跃迁。 Threshold Ventures合伙人Mohammad Islam进一步阐述，AI的扩张本质上是释放人类从事高阶创造性活动，由机器承担计算与物理任务。Ray Wu借用亚马逊与巴诺书店、奈飞与百视达的历史对比，强调了“AI原生”（AI-native）初创企业与“AI使能”（AI-enabled）在位巨头的本质差异：传统巨头仅在旧系统上修补AI功能，而原生企业从底层基于大模型、自主代理与专属数据闭环构建，享有结构性成本优势与全新产品形态。 落地阻碍：企业流程映射与预算紧缩 (14:29) 在探讨自主智能体落地时，Serephina Ha指出大型科技公司的裁员潮推动了高质量人才向早期初创企业迁移，促使资深从业者在自身熟悉的垂直领域深耕。然而，Ashlyn Lackey从大型金融与保险企业的实践角度提出警示：大企业内部仍有大量业务流程尚未完成清晰的梳理与映射，若缺乏这一关键基础层，企业将难以高效部署AI智能体并验证投资回报率。她将这一瓶颈视作改善用户体验和缩减运营成本的“挑战性机遇”（ch-opportunities）。 商业层面的现实制约同样存在。Lackey指出，当前企业IT预算普遍趋紧，采购团队积极推进供应商整合，优先考虑从既有软件合同中挖掘衍生价值，极力避免引入未经检验的单点工具（point solutions）。大科技巨头对企业客户通道的深度把控，迫使年轻的初创企业必须依赖战略渠道合作与生态联盟以打破准入壁垒。 竞争格局：基础设施垄断与智能体机遇 (25:16) 关于未来20年最具价值的AI企业是否已经诞生，Ashlyn Lackey与Serephina Ha倾向于认为未来的行业巨擘目前尚未设立，并指出即便是OpenAI在数年前也仅被视作规模有限的研究项目。随着行业步入深水区，投资人对早期创始团队的考核越来越侧重于韧性、对指导意见的包容度（coachability），以及在寻找理想客户画像（ICP）过程中持续迭代的敏锐度。 Ray Wu从技术架构分层提出了更为审慎的预测：在基础设施层，由于底层算力搭建与模型预训练需要耗费海量资本，OpenAI、Anthropic及CoreWeave等先行者与巨头已构筑了极高的资金壁垒，初创企业机会渺茫；而在垂直行业应用与自主智能体层，能够根本性重构最终用户工作流的方案仍未成型，这片空白领域孕育着未来产生新超级巨头的最大潜力。 本场研讨会的核心共识在于，AI正在引发深远的市场创造性破坏，其商业价值重心正从高资本消耗的底层基建向纵深垂直的行业智能体转移。然而，大科技巨头对企业渠道的严格控制、企业内部业务流程映射的脱节以及采购预算的持续紧缩，构成了初创企业将技术优势转化为长期商业护城河的主要现实不确定性。

## Background 该内容由科技会议机构 Ai4 于 2025 年 8 月 25 日发布在 YouTube 平台上。根据页面简介，本次圆桌研讨会汇聚了来自顶级风投机构与大型金融企业的创新领袖，共同探讨快速演进的 AI 市场格局、前沿技术趋势、风投布局重心以及未来产业链中的赢家与输家。 ## TL;DR 研讨会深入分析了 AI 投资从底层算力基建向垂直应用与 AI Agent 演进的趋势，指出 AI 正在从传统的软件 IT 预算切入体量高达数万亿美元的人类劳动力重塑市场。嘉宾们强调，单纯依赖大模型的“套壳应用”缺乏防御性，拥有深厚行业专有知识、现有客户分发渠道与强交付能力的垂直领域老兵正成为最具突围潜力的主力军。尽管底层基础设施已被科技巨头高度盘踞，但面对企业业务流程重塑与人机混合劳动力的历史机遇，下一代颠覆性的 AI 原生超级巨头极可能诞生在尚未定型的垂直应用与智能体层级中。 ## Atmosphere & style 整场对话节奏明快、思辨深入，兼具宏观历史维度的长周期视野与企业一线落地的务实审视。主持人和来自孵化器、联合风投及金融机构的三位嘉宾配合默契，在碰撞出万亿市场兴奋点的同时，也冷静指出了企业流程映射缺失与巨头估值泡沫等残酷现实。交流风格真诚且干货密集，既有对创业者坚毅特质的温情勉励，又展现出成熟投资人在平台级技术变革中的理性与远见。 ## Key takeaways - [03:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=196s) 风投资金高度向 AI 领域集中，全美超六成早期资金涌入 AI，市场正从单纯依赖大模型的套壳应用（Thin Wrappers）转向具备垂直专业知识的专有模型与成熟 GTM 落地。 - [06:09](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=369s) 深耕特定产业的资深专家在 AI 落地中更具胜算，凭借现成的行业分发渠道和深厚的客户信任纽带，能够快速跨越验证期并在企业端完成部署。 - [08:11](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=491s) 传统 IT 支出仅占企业营收的 5%-10%，而人力成本占 30%-40%；AI 作为 Copilot 或替代劳动力直接切入企业人力预算池，开拓出数万亿美元级的全新潜在市场（TAM）。 - [10:45](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=645s) AI 正在演变为继存储、算力与网络之后的“第四计算基础要素”，驱动全栈 AI 服务商与“人机混合劳动力”（AI-first workforce）的兴起。 - [10:45](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=645s) 本轮浪潮不同于历史周期的关键在于科技巨头（Mag 7）第一时间全员重金抢跑，基础设施层门槛极高，迫使投资人必须在狂热中严守合理估值纪律。 - [14:29](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=869s) 大厂裁员潮促成了技术专家的良性重新分配，流出的顶级工程师与垂直行业专家结合，为初创生态注入了高素质的创业活力。 - [14:29](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=869s) 阻碍企业端充分发挥 AI Agent 价值的核心瓶颈在于：企业内部大量底层业务链路长期缺乏标准化梳理与端到端流程映射（Process Mapping）。 - [17:25](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1045s) 早期 AI 创始人最核心的特质在于：受教性（坚定信念与敏锐吸收反馈的平衡）、面对挫折寻找 PMF 的坚毅韧性（Grit），以及“行业专家 + 技术极客”的黄金搭档配置。 - [22:33](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1353s) AI 原生企业（AI-First）凭借全新的成本结构和无历史包袱的架构，将对仅做表面修补的传统“AI 赋能型”企业造成毁灭性的创造性破坏。 - [22:33](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1353s) 在企业普遍削减 IT 预算、倾向依赖现有供应商的环境下，初创企业难以单打独斗，必须依靠紧密的生态联盟与战略协作才能打破大厂获客壁垒。 - [25:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1516s) 底层基础设施层已被巨头高度盘踞并趋向整合，但未来 20 年重新定义问题解法、市值达数万亿美元的垂直应用与 AI Agent 超级巨头目前大概率尚未诞生。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=4s) 嘉宾背景介绍：跨越孵化器、早期基金与大型企业创投的多维视角 本次研讨会由 Threshold Ventures 合伙人 Mohammad（Mo）Islam 主持。Mo 首先介绍了自己的背景，Threshold Ventures 源自拥有 40 年硅谷投资历史的老牌知名风投机构 DFJ（Draper Fisher Jurvetson），曾早期投资了 SpaceX、Tesla、Box、Twilio 和 Skype 等行业巨头，他本人加入该团队已有 10 年，长期深耕早期科技投资，对当前 AI 浪潮充满热情。 各参会嘉宾随后逐一介绍了各自代表的机构生态： - **Ashlyn Lackey** 来自保德信金融（Prudential）的新兴技术团队。她的核心工作是为母公司筛选并对接潜在的企业级技术合作伙伴，同时主导运营 NJFAST 加速器项目——该加速器由保德信金融、Plug and Play、新泽西州政府以及斯蒂文斯理工学院（Stevens Institute of Technology）四方联合打造。 - **Serephina Ha** 是全球知名初创孵化器 Founder Institute 的董事总经理。该机构自 2009 年创立以来，专注于构想阶段（idea-stage）的科技初创企业孵化，累计孵化了超过 7,500 家初创公司，总估值达 350 亿美元。Serephina 本人曾于 2016 年作为 AI 初创公司创始人创业，如今致力于全方位赋能新一代创始人。 - **Ray Wu** 是 Alumni Ventures 的管理合伙人，常驻硅谷。作为全美最活跃的风投机构之一，Alumni Ventures 采用联合投资人（Co-investor）模式，每年投资 200 至 300 个项目，过去 10 年间累计募资 15 亿美元，在全美主要科技枢纽设立了 5 个办公室。该基金最大的特色在于其拥有由 85 万名校友组成的庞大网络社区，汇集了来自斯坦福、哈佛、麻省理工（MIT）、加州大学伯克利分校（UC Berkeley）及哥伦比亚大学等顶级学府的人才。基金通过这一网络帮助被投企业对接招聘、行业顾问及董事会成员；同时，依托已投资的约 1,600 家被投公司，建立了“CEO 服务网络”，促进被投企业 CEO 之间的业务互助与资源链接。此外，Alumni Ventures 致力于降低风投门槛，让更多合格个人投资者能够像配置公募基金股票组合一样，轻松构建多元化的非上市初创企业投资组合。 ```mermaid flowchart TD accTitle: 创投生态全景与嘉宾角色定位 subgraph S1["孵化与加速阶段"] A["Founder Institute<br/>(Serephina Ha)<br/>• 构想阶段孵化<br/>• 7500+ 初创项目"] B["NJFAST / Prudential<br/>(Ashlyn Lackey)<br/>• 垂直产业加速器<br/>• 链接大型金融企业场景"] end subgraph S2["早期与联合投资阶段"] C["Threshold Ventures<br/>(Mo Islam)<br/>• 专注硅谷早期投资<br/>• 捕获平台级/破坏性创新"] D["Alumni Ventures<br/>(Ray Wu)<br/>• 85万高校校友网络赋能<br/>• 每年 200-300 笔联合投资"] end S1 --> S2 ``` --- ### [03:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=196s) AI 投资狂潮下的理性转向：从套壳应用到专有模型与 GTM 落地 Mo Islam 抛出了关于当前市场狂热度的讨论，并引用了 Axios 资深科技记者 Dan Primack 的最新行业调研数据：当前在全球范围内，高达 53% 的风险投资资金正流向 AI 领域；而在美国市场，这一比例更是攀升至 64%，意味着全美近三分之二的风投资金都被 AI 吞噬。特别是在硅谷，“放眼望去除了 AI 广告还是 AI 广告”。Mo 借此向嘉宾提问：在如此汹涌的浪潮中，大家是选择激进做多全面下注，还是保持投资纪律、不盲目追逐那些光鲜亮丽的表面项目？ Ashlyn Lackey 率先分享了她的观察。她指出，AI 的第一波爆发期充斥着大量直接构建在 ChatGPT 之上的“套壳应用（thin wrappers）”，这类产品既缺乏独特的技术壁垒，也不具备长期的商业防御性（defensibility）。而眼下令她感到振奋的趋势是，越来越多的创业者开始跳出对头部大模型厂商的单纯依赖，转向构建专有模型（proprietary models）。这些创始人不仅对底层 AI 技术有着极高的敏锐度，更重要的是他们将自身的垂直领域专业知识（domain expertise）融入到模型与产品构建中，形成了真正持久的护城河。 此外，Ashlyn 提到了学术界技术成果转化的演变。最初的技术突破很大程度上由学术界学者点燃，但学者在向大型企业客户（B2B enterprise）销售产品时，往往在市场定位和商业化诉求上存在认知脱节。令人欣喜的是，如今这些由技术专家主导的公司正在积极引入成熟的 GTM（Go-to-market）市场化业务领袖和深谙企业长销售周期运作机制的销售人才，补齐了商业落地的关键拼图。 --- ### [06:09](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=369s) 垂直领域老兵入局：自带行业分发与信任的 AI 破局者 Serephina Ha 对 Ashlyn 的观点表示高度共鸣。从孵化器的实践经验来看，目前最具胜算和突围能力的创业者，大多是在特定垂直产业深耕多年的资深专家。 Serephina 列举了两个典型案例： - 一位曾担任软银（SoftBank）投资、估值达 17 亿美元的 MarTech（营销科技）独角兽 CTO 的创始人，近期选择离职重新创办了一家新的营销科技 AI 公司。该创始人此前已沉淀了触达 4,400 万广告用户的深厚资源，凭借在细分领域的深刻理解，能精准切入痛点进行行业颠覆。 - 另一位多次在知名企业担任首席合规官（CCO）的女性创始人，正利用自身数十年积累的合规经验与人脉网络，打造面向垂直合规场景的 AI 产品。 Serephina 尖锐地指出，许多年轻的 AI 创业者找上门来寻求资金支持去开发 MVP（最小可行产品），但他们甚至不知道客户究竟是谁，也根本没有进行过扎实的产品市场验证。相比之下，深耕垂直行业的资深老兵不仅拥有现成的行业分发渠道（distribution），更拥有天然的客户信任纽带，产品一旦就绪便能立刻在企业端完成无缝部署。她预测，这类具备行业壁垒和分发优势的垂直 AI 团队将在未来 1 至 5 年内快速推进并占领市场；至于 5 年之后，随着量子计算等底层新技术的商业化落地，市场或将迎来新一轮颠覆。 --- ### [08:11](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=491s) 跨越数十年的技术浪潮：从底层算力基建到人类劳动力替代的万亿市场 作为经历过多轮技术变革的资深投资人，Ray Wu 从宏观计算历史的维度进行了深度剖析。Ray 回忆道，自己早年是 C 和 C++ 编译器的底层程序员出身，亲历了计算机产业最初如何把人类语言一步步编译转译成机器可读的二进制代码；而到了大模型时代，“人类的自然语言本身直接变成了编程语言”，这种人机交互范式的逆转彻底重塑了软件产业的底层逻辑。在随后的职业生涯中，他曾在思科（Cisco）负责互联网浪潮下的战略投资与核心并购，也曾在移动互联网爆发期执掌惠普风投（HP Ventures）。 ```mermaid flowchart LR accTitle: IT 支出与潜在市场规模（TAM）维度跃迁 subgraph Old["传统企业软件模型"] A["企业总营收"] --> B["IT 与基础设施支出<br/>(仅占企业总预算 5% - 10%)"] end subgraph New["AI 时代劳动力重塑"] C["企业总营收"] --> D["人力劳动力成本支出<br/>(占企业总预算 30% - 40%)"] D --> E["全栈 AI / Copilot 方案<br/>开拓全新数倍级别 TAM"] end ``` 在 Ray 看来，当前的 AI 浪潮在市场规模（TAM）的潜在空间上与以往任何一次技术变革都截然不同： - **市场规模维度的质变**：传统企业软件（SaaS / IT 基础设施）通常仅占企业总营收预算的 5% 到 10%；然而，企业在人力劳动力（human layer，即全体员工薪酬与运营支出）上的开销通常高达 30% 到 40%。当 AI 作为 Copilot 甚至直接替代部分劳动力角色进入这一层面时，它直接切入了企业过去庞大的人力预算池，从而开拓出了传统 IT 视角下完全不存在的全新万亿级市场。 - **投资重心由基建向下游应用演进**：Alumni Ventures 在过去几年重度布局了底层基础设施领域，投资了包括 Lambda Labs、Cohere 以及 Groq 等在内的算力与基础模型公司，见证了英伟达（Nvidia）等底层硬件巨头捕获海量价值的过程。随着算力基础设施成本的持续下行，创新的重心正在加速向企业端应用、小型语言模型（SLM）以及行业垂直化 AI 扩散。 - **全栈 AI 服务商的兴起**：市场上正涌现出大量“全栈式 AI 服务公司（full-stack AI service companies）”，它们不仅提供纯软件交付，更融合了端到端的业务流程服务，彻底打破了传统软件外包与标准软件的边界。 就在 Mo 准备追问 Ray 关于历次技术平台跃迁的周期经验、并提及当天上午人工智能先驱李飞飞（Fei-Fei Li）的主题演讲时，讨论进一步引向了对过往历史周期教训的深入探讨。 ### [10:45](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=645s) 历史级技术浪潮与风险：四层计算架构与大厂齐聚的估值挑战 讨论从前序演讲中将人工智能比作“文明定义级技术”（类比 1870 年代电力的引入）切入。主持人 Mohammad Islam 向经历过多轮互联网技术周期的嘉宾提问：面对 Geoffrey Hinton 等学者对通用人工智能（AGI）及超级智能（ASI）潜在风险的预警，大家究竟是极度乐观，还是应当保持审慎？ Ray Wu 明确表达了高度兴奋。他指出每一次重大颠覆都孕育着风险投资的最佳窗口，而本轮周期带来了全新的商业形态，尤其是 AI Agent 与人类协同办公所构成的“人机混合劳动力”（AI-first workforce）。Ray 提出了一个系统性的四层模型框架：传统计算时代由存储（Storage）、算力（Computing）和网络（Networking）三大支柱组成，如今 AI 正在成为计算底层中不可或缺的“第四层核心要素”。在这之上，企业级应用催生出对安全、合规、隐私以及可观测性（Observability）的全新诉求；在更纵深的领域，通过数字孪生（Digital Twins）技术将物理现实映射入虚拟环境，结合 AI 模型更在生物工程和医疗保健等产业掀起了前所未有的加速发现浪潮。 然而，Ray 也表达了冷静与审慎：与以往技术周期的本质差异在于“巨头的全面同步入局”。回顾历史，初创企业之所以能有长跑道构建壁垒，是因为当时的巨头并不相信新趋势——微软起步做 PC 时 IBM 不相信个人电脑市场，Google 创立搜索引擎时 Yahoo 认为算法搜索无法取代人工分类。而在这一轮 AI 浪潮中，所有科技巨头都在第一时间调集巨额资金抢跑，直接拉高了全行业的投入与创业门槛。因此对于风险投资而言，核心挑战已经转向在狂热的市场中寻找具有合理估值（Right Valuation）的真实机会。Mohammad 对此深表赞同，并指出美股“七巨头”（Magnificent Seven）目前已占据全球股票总市值的近 15%，巨头以前所未有的资本集中度主导着基础模型与技术的输出。 ```mermaid flowchart TD accTitle: AI-First Architecture Layers subgraph Traditional["传统计算三要素"] A["存储 (Storage)"] B["算力 (Compute)"] C["网络 (Networking)"] end subgraph AILayer["AI 核心演进层"] D["AI 计算层 (第四基础要素)"] end subgraph Enterprise["企业级基础设施与支撑"] E["安全 / 隐私 / 合规 / 可观测性"] end subgraph Applications["应用落地与混合劳动力"] F["AI-first 混合劳动力 / 协同 Agent"] G["垂直领域革新 (医疗 / 生物工程 / 数字孪生)"] end Traditional --> D D --> E E --> F E --> G ``` - **历史周期的巨头滞后效应**：IBM 早期轻视 PC、Yahoo 轻视算法搜索，为初创企业留足了构建护城河的成长窗口期。 - **本轮周期的巨头垄断度**：科技七巨头（Magnificent Seven）凭借雄厚资本直接参与核心基础设施军备竞赛，占据全球约 15% 的股票市场资本总额。 - **投资平衡点**：在技术颠覆带来的海量机会与高估值泡沫之间寻求理性定价。 --- ### [14:29](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=869s) AI Agent 走向深水区：人才重流向与企业业务流程映射壁垒 随着讨论深入应用层，Mohammad 将焦点引向当下最炙手可热的概念——AI Agent（智能体），询问两位女性嘉宾如何看待 Agent 在工作流中的实际机会与切入点。 Serephina Ha 强调，智能体时代的突破依然取决于拥有垂直行业专业知识（Domain Expertise）的人。她提出，当前大型科技公司的裁员潮实际上是一场健康的人才重新分配，让极具天赋的技术专家流出大厂、创立或加入初创企业。她以自身支持的团队为例，强调很多优秀创业者在取得突破之前都经历过多轮公司失败；关键在于将失败视作寻找产品市场契合度（PMF）过程中的基石，并在信仰的垂直赛道中与优秀的伙伴同行。 Ashlyn Lackey 则从大型金融机构的实际业务视角浇了一盆冷水。经历过三家大型金融财团的创新实践，她指出阻碍企业端充分发挥 AI Agent 价值的最大障碍在于：企业内部有大量流程至今从未被清晰规范地梳理与映射（Process Mapping）。如果最底层的业务链路本身缺乏标准化沉淀，任何高级的 Agent 都无法有效自动化工作流。 面对主持人关于“企业落地面临严峻挑战”的总结，Ashlyn 风趣地纠正道，她并不喜欢“挑战”（Challenges）这个词，而倾向于将其称作“挑机”（Ch-opportunities，挑战中的机遇）。只要企业愿意投入时间和精力去梳理底层业务逻辑，AI Agent 在提升客户体验、获客、降本增效方面将带来不可估量的价值。 - **垂直人才回流**：大厂裁员潮释放出高素质工程师与领域专家，成为垂直 AI Agent 创新的源头活水。 - **企业采纳的底层短板**：企业内部缺乏标准化的端到端流程映射（Process Mapping），导致 Agent 落地面临断层。 - **商业价值落地视角**：跨越流程梳理门槛后，技术能够直接转化为提升客户体验、扩展客户群与优化收支结构的具体驱动力。 --- ### [17:25](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1045s) 早期创始人的核心特质：从受教性、韧性到万亿劳动力重构 针对台下的众多初创企业创始人与潜在创业者，主持人邀请全场嘉宾逐一分享早期投资与合作中最看重的创始人特质。 不同维度的答案勾勒出了 AI 时代优秀创始人的画像： 1. **受教性与反馈吸收力（Coachability & Ability to Take Feedback）**：Ashlyn 认为，创始人对自己的业务需要具备坚定的信念（Conviction），但这种信念绝不能演化成盲目执拗。他们必须清晰感知自身所处的位置，敏锐听取反馈并全力以赴推动产品达到 PMF。 2. **坚毅的韧性与解题能力（Grit & Problem-Solving）**：Serephina 认为韧性是第一位的。寻找 PMF 与理想客户画像（ICP）是一个极其痛苦、需要成百上千次迭代的过程；缺乏对所解决问题的绝对执着极易中途放弃。同时，创始人必须极具行动力与资源调动能力（Resourceful），善于向周围生态借力。 3. **好奇心、敏捷学习与跨界组合**：Ray 指出，如果面对的是颠覆带来的全新赛道，因为业内无人具备经验，此时“好奇心”以及快速适应与学习的能力胜过一切。而如果是将 AI 应用于特定垂直产业，则要求创始人本身拥有深厚的行业 Know-how，或者团队中必须完成“行业专家 + 技术极客”的黄金搭档配置。 当讨论反向抛给主持人 Mohammad 时，他从宏观投资逻辑阐述了当下的时代机遇。传统企业软件市场虽然体量庞大，也仅停留在数千亿美元规模；而生成式 AI 与 Agent 首次让技术触及了体量高达数万亿美元的“人类服务与劳动力市场”（Human/Labor Services Market）。借用李飞飞（Fei-Fei Li）当天早上的演讲观点，Mohammad 总结道，人类天生具备极强的创造力与原生才华，但在纯粹计算与复杂重复操作上不及机器。随着具身智能与物理 AI（Physical AI/Robotics）的发展，AI 不仅是在替代某些劳动环节，更是在全面释放人类走向高维创造力的潜能。 - **三大核心特质**： - **Coachability**：坚定信念与客观接受反馈、敏捷调优的平衡能力。 - **Grit**：面对反复寻找 PMF 与 ICP 挫折时的抗压迭代能力。 - **Curiosity & Domain Blend**：新赛道看重学习好奇心，垂直赛道看重行业认知与技术能力的互补。 - **潜在市场空间跃升**：从数百亿美元的 SaaS 软件替代，演化为跨越数万亿美元的劳动力与专业服务重塑，为人机协同与物理实体 AI（具身智能）铺平了道路。 ### [21:17](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1277s) 投资策略与市场淘汰：谁将在平台变革中出局？ 在探讨物理 AI（Physical AI）的庞大机会时，投资人强调在评估初创团队时，核心在于考察团队是否具备紧跟模型演进的敏捷适应能力，以及是否能构建起随时间推移而不断深化的长期“护城河”（Durable Advantage）。这种持久优势往往来源于专有数据或顶级人才的获取通道，但更重要的是团队能够清晰、精准地讲述其业务叙事。 随后，讨论的焦点从“谁是赢家”转向了“谁将成为输家”。每一次重大的底层平台技术变革都会伴随熊彼特式的“创造性破坏”（Creative Destruction）——在带来巨大技术红利的同时，必然会有旧势力被时代抛弃。正如当年 Google 崛起初期迅速颠覆并取代了传统报纸的分类广告业务一样，当前的 AI 浪潮也必将重塑既有格局，导致大批跟不上节奏的企业遭到洗牌或边缘化。 ```mermaid flowchart TD accTitle: 平台变革与市场淘汰演进图 subgraph Shift["底层平台变革"] A["颠覆性技术浪潮 (如 AI / 互联网)"] --> B["创造性破坏 (Creative Destruction)"] end subgraph Disruption["企业分化与淘汰"] B --> C["AI Native / 新原生代企业\n全新成本结构与原生架构"] B --> D["传统适配企业 (仅做修补)\n背负沉重历史流程包袱"] D --> E["业务被边缘化 / 遭颠覆淘汰"] C --> F["构筑数据与场景护城河"] end ``` --- ### [22:33](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1353s) AI 原生与 AI 赋能的分野：传统企业的转型困境与生态合作法则 在界定淘汰群体时，Ray Wu 提出了“新浪潮原生”（New Wave Native / AI First）与“新浪潮赋能”（New Wave Enabled）的本质区别。正如互联网时代初期，Amazon 作为互联网原生企业彻底击败了试图“触网”的传统书商 Barnes & Noble，Netflix 彻底颠覆了传统录像租赁巨头 Blockbuster 一样，传统企业在试图适配新技术时，往往受制于原有的组织架构与业务流程，极难实现彻底重构；相反，完全基于大语言模型、智能体（Agent）和专有数据从零构建的 AI 原生企业，具备全新的成本结构与能力底座。 与会嘉宾进一步探讨了初创企业的生存法则与大客户采购环境的残酷现状： * **生态协同与伙伴关系**：Serephina Ha 强调，未来的 AI 赢家必须是擅长协作与生态联盟的团队。正如 Y Combinator 投资组合内的企业相互合作赋能，以及历史上科技巨头与前沿初创公司深度绑定一样，在当前技术演进速度下，试图“单打独斗”的初创公司极难生存。 * **大厂壁垒与预算紧缩**：Ashlyn Lackey 指出，科技巨头对企业级市场（Enterprise）拥有极强的掌控力。当前大企业处于整体削减 IT 预算的周期，客户倾向于从现有供应商（Existing Vendors）中压榨更多价值，而非轻易采购全新的独立软件产品，这使得缺乏大厂生态支持的初创企业获客极为艰难。 --- ### [25:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1516s) 展望未来 20 年：万亿美元级的终极霸主是否已经诞生？ 针对“未来 20 年内全球市值最高的超级巨头是诞生于现有的万亿级巨头（如 Apple、Google、Meta、Tesla、OpenAI），还是尚未被创立”这一议题，嘉宾们展开了推演。Ashlyn Lackey 与 Serephina Ha 均认为那家终极伟大的公司“目前尚未诞生”。回顾历史，OpenAI 在数年前仅仅是一个非营利的科研项目，iPhone 的颠覆性爆发在最初也未被完全预料，技术的跃迁往往孕育于当前的盲区之中。 Ray Wu 则从产业链分层的视角进行了深度拆解： | 产业链层级 | 资本与竞争壁垒 | 市场格局研判 | |:--- |:--- |:--- | | **底层基础设施层 (Infrastructure)** | 资金门槛极高，已被 OpenAI、Anthropic 以及完成上市的 CoreWeave 等吸纳海量融资的玩家占据。 | 新初创公司极难后发赶超，该层级未来将走向巨头间的整合与并购（Consolidation）。 | | **垂直应用与智能体层 (Vertical Apps & Agents)** | 依赖对特定场景问题的全新解法与业务重塑，技术与模式仍在快速演化。 | **尚未定型**；颠覆性的超级企业极有可能在这一层级中诞生，目前仍处于爆发前夜。 | 小组最终达成共识：尽管底层算力与模型基础设施已被现有巨头深度盘踞，但在垂直场景与全新交互模式中，能够重新定义问题解法的下一代万亿级企业，仍有待创业者们从零打造。 ## Notable quotes > [03:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=196s) 放眼望去除了 AI 广告还是 AI 广告 — Mohammad Islam > [03:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=196s) AI 的第一波爆发期充斥着大量直接构建在 ChatGPT 之上的“套壳应用（thin wrappers）”，这类产品既缺乏独特的技术壁垒，也不具备长期的商业防御性。 — Ashlyn Lackey > [08:11](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=491s) 人类的自然语言本身直接变成了编程语言 — Ray Wu > [08:11](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=491s) 全栈式 AI 服务公司（full-stack AI service companies）不仅提供纯软件交付，更融合了端到端的业务流程服务，彻底打破了传统软件外包与标准软件的边界。 — Ray Wu > [10:45](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=645s) 传统计算时代由存储、算力和网络三大支柱组成，如今 AI 正在成为计算底层中不可或缺的“第四层核心要素”。 — Ray Wu > [14:29](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=869s) 阻碍企业端充分发挥 AI Agent 价值的最大障碍在于：企业内部有大量流程至今从未被清晰规范地梳理与映射（Process Mapping）。 — Ashlyn Lackey > [14:29](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=869s) 挑机（Ch-opportunities，挑战中的机遇） — Ashlyn Lackey > [17:25](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1045s) 创始人对自己的业务需要具备坚定的信念，但这种信念绝不能演化成盲目执拗。他们必须清晰感知自身所处的位置，敏锐听取反馈并全力以赴推动产品达到 PMF。 — Ashlyn Lackey > [22:33](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1353s) 新浪潮原生（New Wave Native / AI First）与新浪潮赋能（New Wave Enabled） — Ray Wu > [25:16](https://www.youtube.com/watch?v=Yt9GQ_1dU08&t=1516s) 目前尚未诞生 — Ashlyn Lackey ## Mentioned Mohammad Islam, Ashlyn Lackey, Serephina Ha, Ray Wu, Dan Primack, Geoffrey Hinton, 李飞飞 (Fei-Fei Li), Threshold Ventures, DFJ (Draper Fisher Jurvetson), Prudential (保德信金融), NJFAST, Plug and Play, Stevens Institute of Technology, Founder Institute, Alumni Ventures, Axios, SpaceX, Tesla, Box, Twilio, Skype, SoftBank, Cisco, HP Ventures, Lambda Labs, Cohere, Groq, Nvidia, IBM, Microsoft, Google, Yahoo, 七巨头 (Magnificent Seven), Barnes & Noble, Netflix, Blockbuster, Apple, Meta, OpenAI, Anthropic, CoreWeave, Y Combinator, ChatGPT, AI Agent, 数字孪生 (Digital Twins), 具身智能 / 物理 AI (Physical AI) ## Video description Ashlyn Lackey, Director, Innovation Strategy, Prudential Ray Wu, Managing Partner, Alumni Ventures Mohammad Islam, Partner, Threshold Ventures Serephina Ha, Managing Director, Founder Institute Abstracts In this session, leading venture capitalists will discuss the rapidly evolving AI market, highlighting key trends, emerging technologies, and predictions for the future. Panelists will share their perspectives on which sectors are poised to benefit the most from AI innovations and which areas might struggle to keep up. From advancements in generative AI to AI-powered automation, discover where VCs are focusing their investments and what factors will determine success or failure in the AI landscape. Gain insights into how the world of AI is shifting and who the big winners and losers might be. [Source](https://www.youtube.com/watch?v=Yt9GQ_1dU08)

[3:16] Mohammad Islam: 放眼望去除了 AI 广告还是 AI 广告 [3:16] Ashlyn Lackey: AI 的第一波爆发期充斥着大量直接构建在 ChatGPT 之上的“套壳应用（thin wrappers）”，这类产品既缺乏独特的技术壁垒，也不具备长期的商业防御性。 [8:11] Ray Wu: 人类的自然语言本身直接变成了编程语言 [8:11] Ray Wu: 全栈式 AI 服务公司（full-stack AI service companies）不仅提供纯软件交付，更融合了端到端的业务流程服务，彻底打破了传统软件外包与标准软件的边界。 [10:45] Ray Wu: 传统计算时代由存储、算力和网络三大支柱组成，如今 AI 正在成为计算底层中不可或缺的“第四层核心要素”。 [14:29] Ashlyn Lackey: 阻碍企业端充分发挥 AI Agent 价值的最大障碍在于：企业内部有大量流程至今从未被清晰规范地梳理与映射（Process Mapping）。 [14:29] Ashlyn Lackey: 挑机（Ch-opportunities，挑战中的机遇） [17:25] Ashlyn Lackey: 创始人对自己的业务需要具备坚定的信念，但这种信念绝不能演化成盲目执拗。他们必须清晰感知自身所处的位置，敏锐听取反馈并全力以赴推动产品达到 PMF。 [22:33] Ray Wu: 新浪潮原生（New Wave Native / AI First）与新浪潮赋能（New Wave Enabled） [25:16] Ashlyn Lackey: 目前尚未诞生

[0:04] Ray Wu: network-powered VC [3:16] Mohammad Islam: shiny objects [8:11] Dr. Fei-Fei Li: civilization-defining technology [11:23] Ray Wu: AI-first [14:29] Ashlyn Lackey: There are so many processes that we have not appropriately mapped, and unless we get that layer done, I don't know that we'll be able to efficiently and effectively leverage AI agents. [14:29] Ashlyn Lackey: ch-opportunities [21:43] Ray Wu: AI-native [21:43] Ray Wu: AI-enabled

[0:04] Hey everybody, thanks for having us and [0:05] thanks to AI for for organizing all [0:07] this. Uh I'm Moislam. I'll I'll be [0:10] moderating the panel today. Um I'll [0:12] quickly introduce myself and then ask [0:14] all the panelists to introduce [0:15] themselves. Uh I'm I'm a partner at [0:18] Threshold Ventures. We're part of a [0:20] larger firm called DFj. um early [0:22] investor in a number of companies, [0:24] SpaceX, Tesla, Box, Twilio, Skype, based [0:27] in Silicon Valley for the last uh 40 [0:29] years. I've been with the team for for [0:31] 10 and um early stage investor. Really [0:34] excited about AI. Glad to be here. [0:37] >> Hi, I'm Ashlin Lackey. I'm on the [0:39] emerging technology team at Credential. [0:41] So, in my role, I source and screen [0:43] companies for potential partnership with [0:46] the broader firm and I also work on our [0:48] accelerator partnership NJFAST. Uh, so [0:50] it's us, plugandplay, the state of New [0:53] Jersey, and the Stevens Institute of [0:54] Technology. Uh, like Mo said, excited to [0:57] be here, and thank you to the AI4 team. [1:01] >> Hi everyone, my name is Sarah Fina. I'm [1:03] a managing director at a founder [1:05] institute. So, we're startup um, [1:07] incubator, an ide startup incubator. So, [1:11] we've been around since 2009, incubated [1:14] over 7,500 tech startups. Um, valued at [1:18] $35 billion. Um, I used to be a startup [1:22] founder, AI startup founder in 2016. So, [1:25] now I currently support startup uh in [1:29] many different way. I'm very excited to [1:31] be here and thank you for coming to our [1:33] session. [1:36] >> Hi, uh, my name is Ray Wu, managing [1:38] partner with Alam My Ventures. We're one [1:40] of the most active VC fund in the US [1:42] doing 2003 300 deals on yearly basis. Uh [1:45] we're a co-investor. We raised 1.5 [1:48] billion in the last 10 years. Um the [1:51] five offices across the US in all the [1:54] tech hubs. Uh I'm based in Silicon [1:56] Valley. Um the the reason we call [1:59] Ventures is because we have a very large [2:01] community with about 850,000 committee [2:04] members. many of them come from the [2:06] leading universities networks uh [2:09] including Stanford, Harvard, MIT, [2:11] Berkeley, Columbia of the world. Um the [2:13] the value add we are to our our startup [2:17] is mainly the networks. So we um because [2:20] of such a networks it's great for the [2:22] founders to tap into essentially for [2:24] them to hire employees advisor fund [2:26] advisor board members but in parallel [2:29] because we're co-investor model we have [2:31] about 1,600 portfolio company now was we [2:34] have a CEO service network that kind of [2:36] CEO with our CEO so we actually help our [2:39] uh portfolio company build a network [2:41] amount our portfolio company as well so [2:44] from that we call ourel network powered [2:46] VC um the The reason we're created is [2:49] because we uh we want to democratize the [2:52] uh VC access for individual investors. [2:54] Just like when you're building up a [2:56] portfolio of mutual fund uh mutual fund [2:59] portfolios with all the public stock, uh [3:01] we want to enable anyone to uh expand [3:04] your credit investors to build a [3:07] portfolios of um uh private uh [3:10] investment in the startup space. Uh [3:12] great to be here and thanks for [3:14] participating. [3:16] >> Awesome. Thanks everybody. So we we have [3:18] a pretty diverse group of folks here, [3:20] uh, corporate investors, financial [3:22] investors, uh, former founders. So, uh, [3:25] you can get a pretty, uh, diverse [3:27] perspective of everything that's going [3:28] on. Um, and it's it's a pretty wild [3:30] time. Uh, maybe to kind of get into it a [3:32] little bit, uh, on the on the market and [3:36] how you guys kind of think about, uh, [3:38] what's happening and how you're feeling [3:40] about it. Um, you know, Dan Primac and [3:43] Axios had a good report. Uh, uh, [3:46] globally 53% of all venture dollars are [3:49] going into AI. Uh, in the US it's 64%. [3:52] So, it's, you know, almost like [3:53] twothirds of all venture dollars are are [3:56] are uh all about AI. And, you know, at [3:59] least in Silicon Valley where I'm based, [4:00] uh, that's that's all we hear. Um [4:03] curious to to get your individual [4:04] perspectives on how you're feel like uh [4:07] if if you're feeling bullish and leaning [4:09] in, if you're feeling bearish and you [4:11] know maybe staying more disciplined and [4:13] not chasing some of these shiny object [4:15] AI companies. Uh maybe touch on um a few [4:18] examples as well if you can. [4:21] >> Sure. I'm happy to kick us off. I think [4:24] all based in the [4:25] >> we're all very much in the [4:28] >> Yes. You know, you can't drive without [4:30] seeing another advertisement for [4:32] something. [4:34] >> But what I have personally seen is that [4:37] first wave of it was a lot layers on top [4:40] of chat GPT, not necessarily anything [4:43] unique and something necessarily [4:45] defensive. M [4:47] >> right now I'm excited that there are [4:49] founders that are moving away from [4:51] building on any of the large AI [4:54] companies and we're seeing more [4:55] proprietary models and that's what I'm [4:57] most excited about especially as we have [4:59] people who have developed this very [5:01] innate understanding of AI and they're [5:03] bringing their domain expertise to the [5:06] problem. [5:07] So it's it sounds like it's a it's um [5:09] sort of you know initially it was just [5:12] very like thin layers wrappers around [5:15] chat GPT or enthropics cloud models or [5:18] others but now you're actually seeing [5:20] people bring domain expertise bring data [5:22] and actually build something that's more [5:23] durable. [5:24] >> Exactly. And I think there's also a wave [5:26] of academics that we're building [5:29] >> that sparked things off. [5:32] This is not [5:34] in any way a commentary on academics, [5:37] but there was a little disconnect [5:39] between academics selling corporate [5:41] enterprises [5:43] and how they were positioning [5:45] themselves. In fact, a lot of those [5:47] companies are [5:49] hiring establish Thank you. Um, [5:54] I'm gonna [5:57] keep going. Um, establish go to market [6:00] leaders or people the the sales cycle of [6:03] course. [6:04] That's exciting to see as well. They [6:06] finally have nailed some. [6:09] >> Saraphina and Ray, what do you guys [6:11] think? So, you know, Ashlin sounds like [6:13] she's very excited about the application [6:15] layer companies that are building, you [6:17] know, things more robust now on top of [6:19] these models with proprietary data. [6:20] Where do where do you guys see [6:21] opportunities? [6:23] >> I absolutely resonate with Ashlin's [6:26] comment. um founders that I've seen in [6:29] our incubator, they're more of the [6:31] domain expertise. So I have a founders [6:34] um who built $ 1.7 billion dollar soft [6:37] pink um back company startup as a CTO in [6:41] marketing tech company. He's now quit [6:44] his job to build another marketing tech [6:46] company because he already has access to [6:49] $44 million advertisement users. So he [6:53] is utilizing his experience in marketing [6:56] tech field to build another company that [6:58] he can actually disrupt and another [7:00] founder is also multi-time chief [7:03] compliance officer. One of my u founder [7:06] that I really uh adore and love um she's [7:10] building something in compliance basis [7:12] because a lot of people are doing the [7:15] startup in their domain expertise [7:17] because they have a distribution within [7:20] their network. A lot of startups come to [7:23] me and asking for funding um to build [7:26] MVP and they don't even know which [7:28] customer they're trying to sell. They [7:30] don't even have a PMR. But these people [7:33] who are already in domain expertise and [7:35] vertical um domain, they already know [7:39] and earn the trust to deploy their [7:42] product as soon as those are ready. So I [7:44] think those people are gonna make [7:47] advancement at least one to five years [7:49] but after five years you never know what [7:52] kind of new technology will come in. [7:54] Maybe quantum computing will [7:55] commercialize and monetize. You never [7:57] know. [7:58] >> We're waiting for it. Yeah. [7:59] >> But so sounds like for you know it's [8:01] like vertical AI you know similar places [8:04] where there's domain expertise and you [8:06] think ultimately it's about like [8:07] creating business value for the [8:09] customer. [8:10] >> Ray. [8:11] >> Yeah. Um I've been in this IT industry [8:14] for a long time. I actually started as [8:17] CN C++ compilers uh programmers when I [8:20] started. [8:21] >> Wow. [8:21] >> So the yeah I've been in this industry [8:23] for a long time. So I watched the waves [8:25] when we first started as about translate [8:28] human language into binary code so [8:30] machine can understand and suddenly now [8:33] our language become the programming [8:35] language. So this is completely new way [8:37] in my mind. And second, I was in Cisco [8:41] doing leading investment major [8:42] acquisition during the internet waves. I [8:44] was in HP [8:47] uh in the in the mobile wave. I was [8:48] leading HP ventures at that moment. So I [8:50] watched multiple waves and this is [8:52] completely I I still think this wave is [8:55] a little bit new because this is the [8:56] first time we're seeing that no longer [8:59] we're just infrastructure software [9:02] layers where actually become more the [9:03] intelligent human layers on top of that. [9:06] So that actually bring a new dimension [9:07] almost in my mind in term of the overall [9:09] market size. So traditionally if you [9:12] think about where software companies [9:14] sell in the enterprise normally 5 to 10% [9:17] of the overall budget of the top line is [9:20] actually in the in this kind of IT [9:22] layers or infrastructure service layers [9:25] um but now if you think of how much you [9:27] spend on the human layers right [9:29] essentially all the employees are 30 [9:31] 40%. So you imagine that as a co-pilot [9:33] as a replacement that essentially open [9:36] up a complete new TM or the market that [9:38] doesn't exist from IT perspective and [9:41] then I I think that complete bring a new [9:44] possibilities into that system [9:46] >> and then uh we've been kind of investor [9:48] in infrastructure for the last a few [9:51] years ago. uh so for example lambda labs [9:54] uh coher rock of the world right it's [9:56] all infrastructure similar as you see [9:58] the waves in the nvidia of the [10:01] essentially a lot of value capture on [10:03] the bottom but as the cost [10:05] infrastructure comes down we start to [10:07] see more and more into the in the [10:09] enterprise into a smaller language [10:11] models into the u the new areas where [10:15] this kind of new verticalized AI got to [10:17] build out and there's a lot more [10:20] consolidation as well into the [10:21] application layers kind of building out [10:23] this kind of full stack AI service [10:25] companies that in addition to just being [10:28] a software company that actually provide [10:30] service along that dimension as well. So [10:32] there's many things that unpack because [10:33] of this new business model. Maybe I'll [10:35] follow up with you specifically because [10:37] you've lived through some of the prior [10:38] big platform shifts. Um, you know, Feay [10:41] Lee was here this morning at the keynote [10:42] and she one of the god the godmother of [10:45] AI and she was talking about AI as like [10:47] a civil civilization defining technology [10:50] similar to electricity when it was [10:53] introduced in you know 1870 I think she [10:55] said um how do you how do you guys kind [10:57] of think like you know for you [10:58] specifically since you've like lived [11:00] through the internet and uh you know [11:01] some of the prior waves um are you as do [11:04] you agree are you as bullish um are you [11:06] as excited about the opportunities ahead [11:08] or do you think we need to be more [11:10] cautious given some of what um Jeffrey [11:12] Hinton and others have talked about with [11:15] uh you know the risk with AGI ASI and um [11:19] you know uh with with what what might be [11:22] coming. [11:23] >> Yeah, we're definitely excited. I think [11:25] as I mentioned I I think this is a [11:27] completely new wave. So every wave come [11:29] with disruption where VC opportunities [11:32] is because disruption anytime there are [11:34] significant disruptions there's [11:35] opportunity within that cycle. So we're [11:38] definitely excited about this and we see [11:40] new opportunities especially not only [11:42] the infrastructure opportunities like I [11:44] mentioned we're in but also the [11:46] verticalization of that um into a human [11:50] co c co-pilot the the new business [11:52] possibility of thinking about the AI [11:55] agent working into uh with human [11:58] together so the new workforce almost [12:00] like a hybrid workforce we call AI first [12:02] work and the so we have four layers [12:04] right AI first first infrastructures [12:06] where We do think a lot of new computing [12:08] layers coming in. If you think about [12:10] traditional model there's storage uh [12:13] computing and networking right so AI [12:16] become almost in my mind become a new [12:18] layer that come into the overall [12:19] computing so that that become the almost [12:21] force component then that enables a lot [12:24] more this kind of uh additional [12:26] capabilities uh both beyond the [12:29] workforce but also bring when you're [12:31] coming into enterprise there's a lot of [12:33] the security compliance privacies uh [12:36] observability type of things you need to [12:38] build on on top of that and so that [12:40] brings new possibilities and we're also [12:43] seeing that um the new discovery we we [12:46] we we're looking at some of these new [12:49] optimization where you actually have [12:50] digital twins bring physical properties [12:52] into the virtual environment and you can [12:55] leverage the AI to create new uh [12:57] possibility much faster those in bio [13:00] engineering healthcare so there's a lot [13:01] of new possibilities uh built into this [13:05] um I'm a little bit cautious also only [13:08] because this cycle we see everyone [13:10] jumping at the same time. um [13:13] previously the reason their startup has [13:15] more runway is because actually the big [13:18] guys do not believe in that cycle right [13:20] so you think about when Microsoft [13:23] started with PC IBM actually don't [13:25] believe the PC market uh when Google [13:28] started search engine Yahoo actually [13:30] just don't believe the search actually [13:32] will replace human so there are certain [13:34] there's a delay you know certain factors [13:35] that allow startup to build this bar [13:38] intrigue give them long runways I think [13:40] this cycle is a little bit Everyone see [13:42] at the way we come in at the same time [13:44] every jump in there's a lot of spending [13:46] in this market so that actually brings [13:48] the the so for us it's more about what [13:52] type of opportunity versus what's the [13:54] valuation so a lot of these I think in [13:55] the VC is more about we're excited about [13:58] opportunity but the right valuation I [14:00] think it's a really great point on like [14:02] the big companies now do see it and [14:03] you're seeing a lot of you know great [14:05] models and technology still come out of [14:08] Google and Meta and others u you know if [14:10] you track the magnific magnific uh [14:12] magnificent seven excuse me it's I think [14:15] like 15% of all global uh um equity [14:20] capitalization in the world it's crazy [14:21] driven by uh seven literally seven [14:24] companies so um it's it's it's pretty [14:26] amazing to see the the scale now too um [14:29] I want to touch on something that you [14:31] said uh which is agents and that's been [14:33] like a big theme here so maybe from [14:35] Ashlin Saraphina how do you guys think [14:38] about agents that seems to be sort of [14:40] the new you know area where people are [14:42] looking for investment opportunities um [14:45] you know very much in line with the [14:47] application layer. Um where do you guys [14:49] find opportunities? How do you think [14:50] about workflows where agents might have [14:53] a play? [14:55] >> Um as I mentioned before um I truly [14:58] believe in people um with vertical [15:01] expertise. Um now with the AI [15:04] revolution, people are more empowered um [15:07] to do more things than we used to do. [15:11] And um with the layoff uh with big tech [15:13] layoff and I think um people are meant [15:16] to be assigned at where they should be [15:19] and I think all the smart people all the [15:21] capable and talented people should be [15:23] building startup or working with startup [15:27] or join them. So all the amazing talent [15:31] should flow into startup and in the [15:33] sector that you believe in and you have [15:35] a domain expertise to strive forward and [15:38] then make something impact to the new [15:40] world you know lovable um he even before [15:44] he built the lovable he failed multiple [15:47] companies before and although so it's a [15:51] finding a PMR is really critical in the [15:54] journey of finding a PMR you meet [15:57] different kind of people and work with [15:58] different kind of people. Don't think of [16:00] it as a failure. U think of it as a [16:02] stepping stone for you to move forward. [16:05] So don't give up and but work with [16:07] people that you believe in and then work [16:10] together. That's one of thing that I [16:12] really wanted to say. [16:16] I think jumping in on that agentic AI [16:18] front specifically you mentioned that [16:20] and I think about from the corporate [16:22] perspective so many corporates and I'm [16:24] on my third large corporation at this [16:26] point all in financial services there [16:28] are so many processes that we have not [16:31] appropriately mapped and unless we get [16:33] that layer done I don't know that we'll [16:35] be able to [16:39] efficiently and effectively leverage AI [16:42] agents [16:44] >> interesting so for you it's uh uh you [16:46] know kind of internal processes and [16:48] making that work with this new [16:51] technology you see challenges ahead and [16:53] work that needs to be done to make that [16:55] work. [16:56] I personally don't love the phrase [16:58] challenges. The the word I like is ch [17:01] opportunities. Challenge opportunities. [17:03] And I think there is so much that we can [17:06] do to either improve the customer [17:08] experience, get more customers, save [17:11] money, make money using this technology. [17:14] >> I think it's really just going to come [17:16] down to are we willing to put in the [17:18] time and effort to make sure we're [17:20] getting the most out of it. But this is [17:21] not a panel on ROI, so let's go [17:24] >> not go there. [17:25] uh you know we we definitely have you [17:27] know founders in the audience, folks [17:29] that are in industry that are you know [17:31] future founders. Um you know maybe just [17:34] going down the line uh talk about um [17:36] what you look for in in founders in [17:39] terms of backing them. Um you guys [17:41] talked a little bit about in terms of [17:42] domain expertise and data and you know [17:45] kind of vertical applications but u [17:46] maybe at the very earliest stages if uh [17:49] folks were to come to you what does it [17:51] takes to to get funded or partnered with [17:53] with your organizations? [17:56] >> I would say coachability and an ability [17:58] to take feedback. Um, I've [18:03] one of the best things that a founder [18:05] can have is conviction in their company, [18:07] but they also have to have an awareness [18:10] of what they need to hit product market [18:14] fit. [18:17] >> I would say grit because in order to [18:20] find PMR and your ICP, you need to [18:25] iterate, iterate, iterate in in that [18:27] process. If you don't believe in the [18:30] problem that you are to try to solve [18:33] people, a lot of people will give up. So [18:35] grit is number one quality that I would [18:38] love to see in a founder along with [18:40] problem solving skill. If you can't [18:42] solve the problem, if you had to ask so [18:46] many question and being resourceful um [18:50] then you'll be able to answer all the [18:51] questions because there are people [18:53] around you who are try to help you [18:57] >> right. Yeah. Um I agree to the grit for [19:01] sure. Um sustainability is important. Um [19:05] depends in my mind depends on the [19:06] problem you're solving is a new domain [19:10] because some new opportunity like I [19:11] mentioned on the because of disruption. [19:14] If it's a new opportunities, new [19:15] domains, new areas, I I say it's more [19:18] about uh curiosities, the ability to [19:22] learn very quickly because this is [19:24] something that nobody have experienced. [19:26] Then do you have the the curiosity are [19:29] you interested uh how quick you can [19:31] adapt to this environment uh how [19:33] flexible you are looking at this the the [19:35] the problems that need to be solved but [19:37] different ways approaching that so [19:39] that's the if if it's new problem then I [19:42] think by applying the existing the for [19:45] example AI in a vertical domain for [19:47] example then you you really need some [19:49] domain knowledge as well you can't just [19:51] come from somewhere and says I have no [19:53] idea where this thing is and I I And [19:56] even you don't have it, you probably [19:57] should build a team that has someone who [20:00] really know the domain has a domain know [20:02] how domain knowledge combined with the [20:04] the technical uh capability that you [20:07] have in the team. [20:08] >> Can we turn that question back on you [20:10] and you know what you look at? [20:12] >> Sure. Yeah. I mean you know I think it's [20:14] uh I I similar like at at a high level [20:17] when we make you know any investment [20:18] it's you know team technology and market [20:21] and I you know I think uh Ry Ry may have [20:24] made this point earlier this this may be [20:26] the biggest market market opportunity [20:29] ever you know it's uh like if you look [20:32] at the software market that a lot of our [20:34] companies have historically gone after [20:35] that's like hundreds of billions now you [20:38] know theoretically you could go after a [20:40] multi- trillion dollar human services [20:43] market, labor services, you know, for [20:45] the first time, you can actually um both [20:48] replace what someone is doing, but you [20:50] know, open up the ability to go do more [20:52] creative things. Um I really like the [20:54] way Fay was talking about it this [20:56] morning, which is like, you know, uh [20:58] humans human beings are naturally [20:59] creative and and we we do have a lot of [21:02] like innate talent when it comes to [21:03] that. We're not we're not as good as at [21:05] at other as machines at computing and at [21:08] um you know when you think about [21:09] physical AI like uh um in terms of [21:13] moving things around where like robotics [21:14] and some of the next generation of [21:17] physical AI that we're seeing is going [21:19] to be able to do that but I think the [21:21] the opportunity is going to be massive [21:23] and so we we do look for teams that can [21:25] adapt or are close to the models and [21:28] then understand what the you know [21:30] durable advantage is going to be for [21:32] them over time. often it's data, [21:33] sometimes it's access to talent, right? [21:36] It really depends, but I think having [21:38] that story and narrative be crisp is is [21:41] really important. Um, so we, you know, [21:43] we we've been talking a lot about like [21:45] the winners or who we think the winners [21:47] will be. Uh, maybe we shift to the other [21:50] side given, you know, the topic at hand [21:51] and who who who do you think are going [21:53] to be the losers in this market? [21:55] Because, you know, I I would say, um, [21:57] with any new big platform shift, you [22:00] know, there's creative destruction, [22:01] right? um a lot of positive, you know, [22:03] benefit from that. But, you know, [22:04] there's going to be some people that are [22:06] left behind. Like if you think about [22:08] like Google back in the day, you know, [22:09] they replaced um classified ads in [22:12] newspapers. Like that was I don't know [22:14] if people are, you know, I'm sure some [22:15] some people remember that. That was a [22:17] thing that when Google was first [22:19] introduced, that's some of what they [22:20] started to replace. Um and so uh you [22:23] know curious to get everybody's [22:24] perspective on who do you think is [22:26] either going to get displaced or left [22:29] behind or you know disrupted. [22:33] >> Yeah I I can start. Um thanks. Um I [22:37] think it's similar what you mentioned. I [22:39] do think there's in history there's I [22:42] call the new wave enabled or new wave [22:45] native. Right. So for us we're looking [22:47] at more what we call AI first. We're [22:49] launching our A first fund. Essentially, [22:51] we're looking at something that innate [22:53] come from the the the language models, [22:56] the agent, the uh the proprietary data [22:58] that attribute as a foundation build it [23:01] up, right? So, so you look at [23:02] historically like um I say Amazon is [23:06] internet native and then Barnes & Noble [23:08] kind of internet enabled and you see [23:10] what happens in that uh historical [23:12] context and you see Netflix and uh [23:15] whoever has been long enough, you know, [23:16] blockbuster. So, so that's kind of the [23:18] the traditional company that try to [23:21] adapt to the new technology. It's very [23:23] hard for them to completely change the [23:24] process where there's something new [23:26] completely new coming out from scratch [23:28] and building out of that the new cost [23:30] structure, the new enablement, the new [23:32] capability that completely build on top [23:34] of the AI. I think that's what could [23:36] potentially be the so I wouldn't mention [23:39] anyone by name, but I'm just thinking [23:40] that's the category we're looking at. [23:42] we're b we're funding the the AI kind of [23:45] first AI native type of companies [23:47] whereas more the AI enabled [23:50] >> um I'm a big believer of a collective [23:53] global impact. I think the winner of um [23:57] AI startup would be the people who can [24:00] collaborate and do great partnership [24:03] with each other. If you looked at uh Y [24:06] Combinator um portfolio startup, they do [24:09] partner each other to help each other [24:12] out and then do ICP and PMR um a lot of [24:16] stuff together and also in the past even [24:19] Netflix, Google, Microsoft, they do [24:22] collaborate with emerging technology [24:25] startups to make them frontier model and [24:29] then also um look into different types [24:33] of opportunities. [24:34] So therefore, if you're not a good [24:36] collaborator or not a good good at [24:38] partnership, if you want to do solely on [24:40] your own, it is it will be very hard for [24:43] you to survive um at this age. [24:46] >> I would agree with that and I think that [24:49] big tech has such a chokeold on [24:54] enterprises. So, if you are an [24:56] enterprise [24:58] focused company, if you don't build out [25:01] those partnerships, [25:02] >> you're not going to get it. Right now, [25:04] we're in a time where budgets are being [25:07] slashed. People aren't looking [25:08] necessarily for new vendors. They're [25:10] looking at their existing vendors and [25:11] how they can get more out of them. [25:16] um you know we we sort of talked about [25:18] this I think in uh the the prep session [25:20] for this uh because I you know I think [25:23] given um the scale of what we're talking [25:25] about the potential for AI you know um [25:29] you're seeing for the first time [25:30] companies that are worth like trillions [25:33] of dollars maybe multi- trillions if you [25:35] look at you know Apple Google Facebook [25:38] um Tesla potentially open AI is coming [25:40] up their last round was at 500 billion [25:43] um in 20 years, do do you guys think the [25:46] most successful uh most valuable [25:48] company, I should say, is going to come [25:50] out of that batch or do you think it's [25:52] going to be something completely new? [25:54] Has it been founded yet? And um tell me [25:56] who you think it's going to be. [25:59] >> I don't know that it's been created yet. [26:01] So, ask me again in 10 years. [26:02] >> Okay. So, you think it has it hasn't [26:04] been built. Okay. [26:06] >> I agree with Ashlin. I don't think it [26:08] has been built yet. And you know um open [26:13] AI was a research project just few years [26:17] ago and we didn't expect this at all and [26:19] same for iPhone and many other [26:22] technology. If I can predict the future [26:26] then I wouldn't be [26:28] >> and I also think if [26:29] >> I wouldn't be here I will be time [26:31] traveling or something. I think the [26:34] other piece is if you can predict the [26:35] future, odds are it's already here and [26:38] someone is looking to see where their [26:40] gaps are and how they can do better. [26:43] >> Ray, here's a lot. [26:44] >> Yeah. Um I think it depends on the [26:46] layers the way we look at it. Some of [26:48] these already because it takes so much [26:51] money to build up the infrastructure. So [26:53] all the infrastructure guys that's when [26:55] what you I think depends on the wave [26:58] always from infrastructure to middleware [26:59] to application. So the infrastructure [27:01] layers take so much money has already [27:04] been I mean doesn't matter it's open AI [27:06] is antropic um even core we've IPOed [27:09] right so we have a few of these kind of [27:11] company already I think they took a lot [27:13] of funding it I think it's much harder [27:15] for start to catch up in that layers [27:18] only because it's um it's already [27:20] whoever already been built I think it's [27:22] going to be more of a consolidation over [27:24] time within that layers but you think [27:26] about vertical applications the vertical [27:29] agent [27:30] new ways of solving problems. I think we [27:33] I don't think we have seen that yet. [27:35] It's essentially something to be built [27:36] just because there's new possibility. So [27:39] the way I we always compartmentalize is [27:41] think about different layers, different [27:42] components and where we are versus uh [27:45] generically talk about where things are. [27:48] >> So with that uh I think we're out of [27:50] time. So thank you for having us [27:51] everybody. Uh most valuable company yet [27:54] to be built. So hopefully we'll see you [27:56] soon. [27:56] >> Right. Let's give the panel a round of [27:58] applause.