A16z, Accolade Say Direct Compute Scaling Rewrites VC Portfolio Math

A16z与Accolade称算力直接扩展重构风投组合逻辑 资本直接转化为算力集群构建了自我强化的技术护城河，使超额回报向顶尖平台集中，同时让传统软件杠杆收购陷入困境。 研讨小组成员认为，AI直面全球30万亿美元的经济劳动力，其影响范围远超传统IT软件，并通过允许资本经由原始算力直接转化为技术优势复合增长，颠覆了风险投资的历史规律。由于品类赢家占据了绝大部分市场价值，资产配置方必须将AI从边缘对冲转向核心配置，重注前10%的多阶段平台。与此同时，ChatGPT问世前的传统SaaS资产在私募股权债务重压下面临严峻的估值压缩与终局风险，表面化的AI功能嫁接十分危险。展望未来，推动数万亿美元级产出的终极驱动力将从数字聊天机器人过渡到实体机器人、自主系统，并攻克关键的电力与数据中心基础设施瓶颈。 风险投资的幂律法则达到了前所未有的集中度：在过去二十年间的3,000家美国风投机构中，仅有约20家（占比不足1%）能持续产生3倍以上的净回报，而整个行业的平均水平仅为1至2倍净回报。 (0:00) 与以往周期中过剩资本导致人员臃肿并带来组织拖累不同，前沿AI领域的资本可以直接转化为原始算力，从而加速模型能力提升并拓宽竞争护城河。 (1:34) AI的普及速度是企业SaaS的近4倍，仅用约4年时间便达到了1,000亿美元的年化运行率（run-rate），而SaaS则耗时15年。 (3:43) 从IT软件视角评估AI会极大地低估其潜在市场总额（TAM），因为AI瞄准的是认知劳动——该市场规模约为企业软件总支出的40倍。 (5:35) 风投组合的逻辑要求早期阶段能容忍约60%的亏损率以捕获极端的品类赢家，而成长阶段基金的目标则是承担10%至20%的亏损率，并在集中仓位上博取10倍以上的上行回报。 (8:33)

## Background 该视频由知名风险投资机构 a16z 于 2026 年 9 月 10 日发布在 YouTube 频道。内容由 a16z 合伙人 Jen Kha 与 David George 携手母基金 Accolade Partners 的 Aram Verdiyan 共同探讨，主要聚焦于 AI 如何重构科技投资的极端幂律法则、前沿大厂如何借助资本直接转化算力形成全新复利壁垒，以及机构投资者在资产配置、传统 SaaS 处置与物理基础设施瓶颈面前应采取的策略转变。 ## TL;DR 对话深入探讨了 AI 时代风险投资的底层变迁：不同于过去过度注资引发的组织内耗，如今资本可直接转化为算力并转化为内生复利壁垒，驱动前沿超级巨头的退出天花板迈向千亿美元乃至万亿美元级别。面对全美三千家 VC 中仅有 20 家能稳定交付 3 倍净回报的残酷现实，资产配置者正将 AI 从边缘的“卫星配置”调整为核心乃至超核心敞口，同时警惕真假难辨的短期虚假增长。传统 SaaS 资产因缺乏 AI 原生架构正遭遇估值腰斩与退出通道关闭的严峻考验，而兼顾长周期耐心与阶段性流动性创造的 GP 才能穿越周期。未来真正的战略机遇不仅在企业级与消费级智能体，更扎根于电网通电速度、超高密度数据中心与先进硬件等供给侧物理咽喉。 ## Atmosphere & style 整场对谈兼具业内顶尖投资人的老友重聚温情与直击行业痛点的冷峻洞察。开场因 Aram 曾任职 a16z 十周年的纪念道具而笑声不断、松弛风趣，但随后迅速切入全行业最敏感的尖锐议题——估值泡沫、LP 接飞刀焦虑与传统 SaaS 终局价值归零风险。三位嘉宾言辞直率且极富颗粒度，既有对极端幂律数学模型的严谨推演，又有对一线真实使用数据穿透虚假指标的务实笃定，最终在畅想具身智能与迎接机器时代的豪迈展望中收束，展现出一种极具感染力的高密度前瞻气场。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=0s) 生成式 AI 重构风险投资幂律法则，资本可直接转化为核心算力并转化为产品复利壁垒，打破了过去传统初创公司过度注资招人导致管理内耗的死局。 - [03:43](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=223s) AI 仅用时 4 年便达到 1000 亿美元行业营收，而传统 SaaS 耗时 15 年；全美经济劳动力支出是软件支出的 40 倍，AI 任务替代的潜在市场空间远超传统软件许可证模式。 - [10:40](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=640s) 全美 3000 家风险投资机构中仅有约 20 家能在过去 20 年间稳定实现 3 倍净回报，行业正加速走向极度两极分化，缺乏早期生态和平台能力的纯中后期基金举步维艰。 - [39:30](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2370s) 传统高杠杆软件资产面临终局价值被 AI 侵蚀的风险，公开市场市销率腰斩，而寄希望于通过裁减客服等外部手段进行“AI 私募股权”改造往往会引发 NPS 崩塌并加速业务违约。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=0s) 风险投资的幂律法则从手工作坊式的局部特征演变为整个科技行业的系统性格局，头部领跑者承载了数万亿美元级别的潜在价值。 - [00:44](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=44s) 资本可以直接购买算力并转化为模型与产品的核心护城河，资本运作彻底告别了依靠注资堆砌人员规模的传统逻辑。 - [03:43](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=223s) AI 在资产配置中必须从组合边缘的“卫星配置”升级为不可或缺的“核心甚至是超核心”地位。 - [03:43](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=223s) 顶尖科技公司的退出天花板正在从过去周期的 100 亿美元和当前的 400 亿美元，向 1000 亿美元以上量级跃升。 - [05:56](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=356s) AI 正在直接承接报销、账单与行政等庞大人力任务，其通过重构劳动力释放出的潜在市场规模是传统垂直 SaaS 的 10 倍以上。 - [05:56](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=356s) 科技行业各层级呈现非零和的同步扩张，例如法律大模型 Harvey 的普及反而促成了更多专业交互与计费工时增长。 - [08:33](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=513s) 顶级早期基金的高额回报依赖于高达 60% 的项目失败容忍度，而成长期基金则必须依赖单笔 10 倍以上回报的重仓标的驱动整体业绩。 - [10:40](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=640s) 全美 3000 家风投机构中不足 1%（仅约 20 家）能够持续交付跨周期的 3 倍净回报，平庸风投基金的表现甚至显著逊于 PE 和二级公开市场。 - [13:41](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=821s) 风险投资加速呈现“中间层之死”，市场仅留下极其敏锐垂直的前沿微型种子基金与具备 700 人投后赋能平台的全周期超级大牌机构。 - [17:04](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1024s) 纯中后期基金难以立足，晚期投资的核心入场券和追投权完全依赖于从早期种子阶段与创始人建立的深厚信任网络。 - [22:18](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1338s) AI 初创公司充斥着单月 ARR 年化与生态内互买的虚假增长信号，判断真实牵引力的关键在于客户是否从试用防范逆转为强制倒逼使用。 - [25:23](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1523s) LP 与 GP 存在激发动机错配：GP 最大恐惧在于错失下一个超级平台（错失之错），而 LP 往往为了保住饭碗倾向于贴近基准线并规避下行暴雷（作为之错）。 - [28:09](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1689s) LP 在投资组合构建上容易陷入过度分散陷阱；若仅在黑马基金上下注 1% 仓位，即使获得 10 倍回报也无法对整体母基金净值带来实质推动。 - [29:36](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1776s) 公募与私募软件估值体系经历大洗牌，缺乏 AI 原生动力的传统软件资产面临终端价值归零风险以及无买家接盘的流动性困局。 - [34:26](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2066s) 代码生成任务具备完备数据、确定验证与可模拟环境，向复杂商业通用知识工作的扩散速度会慢于悲观预期，为存量软件调整留出了时间窗口。 - [38:05](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2285s) 2021 至 2022 年间发生的数千亿美元软件杠杆收购随着公允价值腰斩导致底层杠杆率激增，已成为当前私募信贷市场的严重隐患。 - [39:30](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2370s) 简单地将 AI 嫁接在传统企业外部无法挽救老旧业务，反而易因损坏用户体验导致 NPS 下滑并加速财务危机，唯有从地基底层重构才有一线生机。 - [40:57](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2457s) AI 独角兽的成长周期被大幅压缩，顶尖 GP 需在周期高位通过并购与二级交易主动制造流动性，同时护送核心赢家冲刺公开市场。 - [44:01](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2641s) 未来的超级物种将脱离单纯的聊天界面，在具身机器人、全自动驾驶、占 GDP 18% 的医疗健康以及物理现实重塑中诞生。 - [46:36](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2796s) AI 发展正面临物理供给侧瓶颈，能源并网审批延误与下一代 10 倍超高密度数据中心基础设施的重构成为决定机器时代能否降临的关键卡点。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=0s) 极端幂律的到来与资本复利逻辑的重构 风险投资领域的底层价值创造逻辑正在经历一场深刻的巨变。长期以来，幂律法则（Power Law）一直被视为早期风投这一“手工作坊式行业”（cottage industry）的独有特征，但如今它正在演变成整个科技投资生态乃至宏观经济中的系统性格局。当前全美约 3000 家风险投资机构中，在过去二十年间能够持续交付 3 倍净回报（3x net returns）的仅有区区 20 家。而随着生成式 AI 的崛起，这种极端马太效应愈发剧烈：像 SpaceX、OpenAI 和 Anthropic 这样的前沿巨头，合计潜在企业价值已经达到了 3.5 万亿至 5 万亿美元之巨，但令人震惊的是，在这些超级独角兽走向成熟或上市之前，大量机构出资人（LP）和主流资产配置者在这一领域的敞口却严重不足。 ```mermaid flowchart TD accTitle: 传统初创公司与前沿AI实验室资本运作对比 accDescr: 展示了传统软件初创公司过度注资导致的组织臃肿，与AI前沿实验室注资直接转化为算力复利优势的路径对比。 subgraph Traditional["传统软件初创公司模式"] A["注入巨量资本"] --> B["大规模招募人员 (1000+人)"] B --> C["协调内耗与管理摩擦增加"] C --> D["产品迭代受阻、效率稀释"] end subgraph FrontierAI["前沿 AI 实验室模式"] E["注入巨量资本"] --> F["直接采购大规模算力 (Compute)"] F --> G["模型能力跃升与产品力增强"] G --> H["资本与算力驱动的内生复利壁垒"] end ``` a16z 合伙人 David George（DG）指出，当前的幂律分布比过去 10 到 20 年间的任何时候都更加极致，甚至超过了此前消费互联网网络效应爆发的鼎盛期。过去科技投资中虽有规模报酬递增（increasing returns to scale），但传统初创公司最常见的死法之一就是“过度注资”：盲目招募上千名员工，导致组织陷入协调内耗、管理混乱以及优先级冲突；但前沿 AI 实验室彻底打破了这层组织瓶颈。在 DG 的投资生涯中，"这是第一次你把巨额资本砸向一家公司，资本能直接为它带来滚雪球般的竞争壁垒"。资本被直接转化为核心算力（Compute），而算力直接提升模型与产品性能，形成了纯粹的规模经济效应。 - 全美 3000 家 VC 机构中，仅有 20 家在过去 20 年里稳定取得 3 倍净回报。 - SpaceX、OpenAI 与 Anthropic 三家领军企业合计承载了约 3.5 万亿至 5 万亿美元的潜在企业价值。 - 资本运作模式转变：从过去“注资雇人导致管理摩擦”转变为“资本直接购买算力，推动产品和商业护城河持续复利”。 --- ### [03:43](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=223s) 资产配置的颠覆：从边缘配置到超核心地位 对话中出现了一段温馨而有趣的叙旧：来自母基金 Accolade Partners 的 Aram Verdiyan 曾于整整十年前在 a16z 供职。为了纪念这十周年，主持人 Jen Kha 专门从机构“地下秘库”（catacombs）中翻出了一件珍贵的超大号纪念物，引得现场欢笑连连。回望十年前，整个市场还在为出现 10 亿美元规模的单支风投基金而叫苦不迭、忧心忡忡。然而短短十年间，一级市场的体量已彻底跃升为 5 万亿至 6 万亿美元的庞大资产类别，企业选择在更长时间内保持私有化的趋势不可逆转。 针对当下的资产配置策略，Aram Verdiyan 提出了极为鲜明的观点：AI 仅用了 4 年时间就实现了 1000 亿美元的行业总营收，而当年的传统 SaaS 达到同等收入体量足足花费了 15 年。更重要的是，这一增长背后的推理需求（inference demand）几乎是无限的。AI 正在同时冲击交通运输、劳动力、专业服务、资本运作与组织协同等各个关键板块，历史上从未有过任何一项技术范式能够同时触及高达 30 万亿美元的 GDP 领域。面对如此体量的增长飞轮，机构投资者如果仍把 AI 当作投资组合边缘的“卫星配置”（satellite position），将是严重的战略失误；AI 必须成为资产组合中的“核心（core）甚至是超核心（super core）”。 从退出回报的角度来看，顶层投资标的的天花板已被无限拉高。David George 回顾自己 18 年前从事成长股权投资的经历时感慨，彼时前 10% 顶尖退出的门槛大约在 100 亿美元，如今已飙升至 400 亿美元左右，而随着 Anthropic、OpenAI 未来的潜在退出，这一数字很快会迈向 1000 亿美元量级。上一轮科技浪潮创造了 25 万亿美元的新增市值，而正在展开的这一轮 AI 浪潮在体量上无疑将全面超越以往。 - SaaS 达到 1000 亿美元营收耗时 15 年，而 AI 仅用时 4 年。 - AI 正在同时影响涉及 30 万亿美元规模的 GDP 核心板块。 - 风险投资从 18 年前的“边缘手工作坊”成长为 5 万亿至 6 万亿美元规模的独立资产大类。 - 头部企业的退出规模正从过去周期的 100 亿美元、当下的 400 亿美元，向 1000 亿美元级别跃迁。 --- ### [05:56](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=356s) 远超传统 SaaS 的 TAM：重塑劳动力与非零和生态 关于 AI 的潜在市场空间（TAM），Aram Verdiyan 坦言业界一直存在严重误判。以医疗健康领域为例，全行业每年花在医疗 IT 软件上的预算仅约 600 亿至 100 亿美元；然而，医疗体系中用于报销索赔（claims）、账单核算（billing）和日常行政等纯劳动力任务的支出，却是一个规模达万亿美元的庞大市场。AI 的核心目标并非替代传统的 IT 软件许可证，而是直接承接并执行原本由人类劳动力完成的实际工作任务。如果按照任务所承载的经济价值和分成率来重新定义 TAM，AI 的市场空间将是传统企业软件或垂直 SaaS 的 10 倍以上。 ```mermaid flowchart LR accTitle: 医疗健康IT与医疗劳动力市场空间对比 accDescr: 显示传统医疗IT软件百亿美元市场与医疗行政劳动力万亿美元市场的量级差异。 subgraph LegacySaaS["传统 SaaS 视角"] A["医疗 IT 软件预算"] --> B["600 亿 - 1000 亿美元 TAM"] end subgraph AIEra["AI 任务替代视角"] C["医疗报销、账单与行政等劳动力"] --> D["1 万亿美元以上潜在 TAM"] end ``` David George 补充指出，美国宏观经济中花在劳动力上的资金是软件支出的约 40 倍。但这绝不意味着人类劳动力将直接归零，而是任务形态的深度重塑与再构。Jen Kha 分享了一个生动的法律行业实例：一家合作律所的高级法律顾问曾感慨自己极度喜爱法律 AI 工具 Harvey，但原因却出人意料——原本很多客户由于专业门槛限制，对复杂条款往往直接弃权听从律师意见；如今借助 Harvey，客户自认为也掌握了专业知识，反而在各种细节议题上与律师反复深入辩论，最终导致律所的计费工时（billable hours）不仅没有减少，反而显著上升。这表明 AI 带来的经济效应具有极强的“扩张性”（expansionary）。 针对市场上持续争论的“模型层与应用层究竟谁吃掉谁”、“开源模型是否会扼杀专有前沿实验室”等零和思维（zero-sum thinking），DG 明确表达了反对态度。他引用业内同行的观点指出，由于整个市场空间正被指数级撑大，甚至可能出现“一切层级都将蓬勃发展”（everything is going to work）的繁荣景象。单纯用传统软件周期的挤压竞争视角来推演 AI 生态，往往会陷入“过于狭隘的局限”。 - 宏观层面，美国整体经济在人力成本上的开支约为软件支出的 40 倍。 - 医疗 IT 软件市场年支出为 600 亿至 1000 亿美元，但医疗行政与账单处理等劳动力市场规模超 1 万亿美元。 - 法律垂直大模型 Harvey 的实际反馈表明，AI 普及推动了更深入的专业交互，反向刺激了账单计费工时的增长。 - 产业链各层（前沿大模型、开源生态、上层行业应用）更可能呈现非零和的同步扩张，而非单一维度的赢者通吃。 ### [08:33](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=513s) 赢家通吃与品类扩张：风险投资的容错率与损失率 围绕 AI 是否会让科技行业走向更加彻底的“赢家通吃”，讨论展开了更深层的推演。Jen Kha 提出，上一代科技在许多细分赛道上确实呈现出赢家通吃的局面，但当前这一代 AI 似乎在底层逻辑上更为不同，因为投资人正在对诸多基本面进行重新定价。David George（DG）对此给出了更为精细的界定：如果拉长时间维度观察所有领军科技公司的市值增长，就会发现成功的公司数量极其庞大，整个科技行业本身并非单一维度的赢家通吃；然而，在**特定品类内部**，幂次法则（Power Law）体现得淋漓尽致——头号赢家吞噬了绝大部分的市场份额和市值，而第二名往往只能在边缘“争夺残羹冷炙”（playing for scraps）。 与此同时，技术的演进正在带来品类数量的爆发式扩张。就像二十年前，CRM 根本算不上一个成熟的独立品类，市场上仅有 Siebel Systems 等少数几家公司在摸索，如今它已膨胀为一个万亿美元级的庞大市场。David 指出，风险投资的底层机制在于对损失的极高容忍度，如果在某一期基金中没有遭受足够多的本金亏损，反而说明投资团队冒的风险远远不够。数据表明，历史上回报最顶尖的早期风险投资基金，其投资失败率（loss rate）通常高达 60% 左右；而在成长期投资阶段，虽然失败率会下降至 10% 到 20% 的合理区间，但依然需要依靠少数能够实现 10 倍以上回报的标的来驱动整期业绩。投资者的职责在于判断哪些赛道具备信服力，并在每一个靠谱的品类中押注头号选手；如果整个品类最终未能爆发，也是投资组合必须承担的系统性风险。 - **早期基金损失率**：顶尖早期基金的投资失败率往往在 60% 左右；不出现高亏损率意味着没有承担足够的风险。 - **成长期基金回报指标**：成长阶段的损失率通常控制在 10% 至 20%，但组合必须依赖 10 倍以上（10x+）回报的项目拉动整体表现。 - **品类扩张效应**：以 CRM 历史为例（从早期的 Siebel Systems 到后来的巨头生态），AI 将催生出远比过去更为广阔的全新应用品类。 --- ### [10:40](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=640s) 风投中的“极度一致性”：3000家机构中仅有20家实现持续3倍回报 市场择时往往充斥着认知偏差。投资人在当下常常感叹资产估值过热，但事后复盘却往往发现当时的领军企业其实非常便宜；唯有少数真正的市场泡沫期例外。正因如此，Aram Verdiyan 经常告诫 LP（有限合伙人）：在风险投资这一资产类别中，“持续一致的配置”（consistency）比其他任何领域都更加致命，因为颠覆性技术的爆发节点往往难以精准预测。面对当前诸如大模型等前沿 AI 技术，许多没有直接入场渠道的机构投资者正陷入落后的焦虑中，甚至因急于追赶而产生动作变形和非理性的追高行为。 Aram 分享了一组令人震撼的行业内部研究数据：Accolade 统计了美国过去二十年间的 3,000 家风险投资机构，最终发现**仅有 20 家公司**能够做到持续创造 3 倍以上的净回报（3x net TVPI）。这里的标准甚至不是要求在这 20 年间连续募集的七八期基金全部达标，而仅指在二十年跨度内至少有 3 到 4 期基金实现 3 倍净回报——在全行业占比不足 1%。这种极度稀缺的一致性背后，核心原因在于这些赢家在每一个投资周期（vintage）中，都无一例外地拥有锁定品类定义型巨头（category-defining companies）的通道与份额。光是“在投资组合里挂上明星 Logo”毫无意义：对于早期基金而言，必须持有足够高的股权占比；而对于成长期投资，集中下注的仓位规模更是决胜关键。当单一重仓项目能够占据整期基金 5% 到 10% 以上的规模时，单家巨头回报整期基金（fund-returning）的数学模型才可能在晚期阶段成立。相比之下，Cambridge Associates 过去十年的全行业平均数据显示，平庸风投基金的净回报仅在 1 到 2 倍之间，流动性却被锁死十年，其收益表现甚至显著逊于成熟的 PE（私募股权）和公开市场二级投资。 - **顶级机构稀缺度**：全美 3,000 家 VC 中，仅有 20 家（占比低于 1%）在过去二十年实现跨周期的持续 3 倍净回报（3x net TVPI）。 - **行业均值困境**：据 Cambridge Associates 数据，过去十年普通风投基金的回报普遍徘徊在 1x 到 2x，远落后于公开市场和 PE，无法匹配其十年流动性锁定期。 - **晚期重仓数学逻辑**：晚期投资的顶级公司必须占到基金总规模的 5% 至 10% 以上，才能依靠单一超额回报项目覆盖整期基金本金。 --- ### [13:41](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=821s) “中间层之死”与全生命周期平台的飞轮效应 风险投资市场正加速分化为两极，导致所谓“中间层之死”（death of the middle）的现象愈演愈烈。三人甚至打趣称，每次提到“中间层之死”都应该罚酒一杯。Jen Kha 援引了投资人 Eddie 当天发布的一条推特：“对超大型 VC 基金的需求是由创始人驱动的，而不是 LP。创始人普遍想要的是具备真正规模化能力、能做全生命周期投资、并在招聘和获客上提供强效支持的超级品牌。LPs 只是在后知后觉地跟进，因为这与传统分散投资的常识背道而驰。” ```mermaid flowchart TD accTitle: a16z 平台飞轮模型 accDescr: 展示从前沿认知与行业专家洞察，到赢得创始人心智、投后赋能，再到沉淀顶级声誉与回报的风险投资闭环飞轮。 subgraph Expertise["认知与获取"] A["深耕细分领域与前沿认知<br/>(Domain Expertise & POV)"] B["赢得优质创始人与热门项目<br/>(Win Category Winners)"] end subgraph ValueAdd["投后与赋能"] C["700人运营平台赋能<br/>(700-Employee Operating Platform)"] D["改变发展曲线与降低系统风险<br/>(De-risk & Bend the Curve)"] end subgraph Moat["声誉与资本回流"] E["跨周期业绩持续性与品牌壁垒<br/>(Persistence of Returns & Brand)"] F["顶级创始人圈层推荐效应<br/>(Killer References & Network)"] end A --> B B --> C C --> D D --> E E --> F F --> A ``` 谈及究竟何为“中间层”，David George 分析了当前存活最好的两种模式：一类是极其垂直、规模小巧但专注前沿深科技的微型早期基金，凭借极深的学术或技术圈层认知迅速锁定天使轮；另一类则是像 a16z 这样的巨型全产品线平台，能够陪伴创业者从 Seed 轮一直走到 IPO。在这两极之间缺乏鲜明特色的中型基金正陷入残酷的竞争夹缝。创始人出让股权的核心诉求在于“谁能帮我降低失败风险”（derisk the outcome）。a16z 拥有多达 700 名员工，并将基金收取的管理费绝大部分重新投入到企业服务、招聘、政府事务等赋能团队中，其目的就是为了在关键时刻“扭转公司的成长曲线”并在激烈争夺中拿下项目。 David 将这一模式总结为一个不断自我强化的闭环飞轮：顶尖研究产生前瞻认知，前瞻认知帮助团队向创始人证明合作价值以赢得 Deals；赢下投资后，庞大的内部平台协助公司降低失败风险并扩大战果；一旦成功，不仅为基金沉淀持久的超额收益，更在头部创始人圈层中树立无可替代的口碑推荐链条，进而吸引下一代顶尖创业者慕名而来。讨论最后，Aram 抛出了一个新的假说：对于估值在两三千万美元以内、基金规模低于一亿美元的 Pre-seed / Seed 轮小型基金，它们是否能够与大型基金共存？因为当初期同时涌现出六七家做同类 AI 方向的团队时，大型平台往往倾向于等待一至两轮，直到品类赢家的胜势相对明朗之后再重仓下注。 ### [17:04](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1024s) 早期品牌引力与全周期投资的飞轮效应 在探讨小型种子基金与头部大机构的关系时，Aram Verdiyan 指出，专注最前端的小型基金完全可以在大机构出手前的一两个阶段（如 Pre-Seed 轮）建立自己的根据地和胜率，与大平台形成互补。Jen Mehl Kha 与 David George（DG）对此表示赞同，指出 a16z 与整个生态中的天使及种子基金都保持着非常健康的合作，但 a16z 自身也在向更前端延展，例如运营 Speedrun 孵化计划。当前市场正在发生深刻演变，创始人对顶级风投品牌展现出了极强的“优先依附性”（Preferential Attachment）。对创业者而言，他们往往并不在乎资金究竟来自 a16z 内部的哪一个专注领域的子基金，而是一心渴望“进入这一引力轨道”（in that orbit），后续平台自然能调配相应基金进行承接。因此，风投机构本质上必须保持极强的早期属性与灵活性，即便不投某一笔交易，也必须率先站在最前沿摸清市场图景，以便在后续阶段做出明智的决策。 ```mermaid flowchart TD accTitle: 早期与成长型双轮驱动的投资飞轮 accDescr: 展示早期投资建立的生态与关系如何反哺成长型阶段，并强化整体品牌吸引力 subgraph EarlyStage["早期投资引擎"] A["Speedrun / 早期种子投资"] --> B["获取前沿技术洞察与创始人深度信任"] end subgraph Orbit["品牌轨道效应"] B --> C["维持领跑地位与强力 Pro Rata 权"] C --> D["成长基金提供规模化后续资本"] end subgraph GrowthStage["后期规模放大"] D --> E["在关键头部企业单笔下注5-10%基金额度"] E --> A end ``` 这种模式彻底重塑了风险投资的竞争壁垒。在 10 年前，早期机构向后期拓展或跨阶段布局尚处雏形，而如今缺乏早期捕获能力的“纯中后期基金”已难以立足。David George 坦言，自己的成长基金之所以能取得成功，根本原因就在于 early stage 的根基支撑：“我们的业务始于早期，也终于早期。”早期布局不仅为后期投资带来了无可比拟的直接渠道（Access）、信息优势与创始人关系，成长基金充沛的后续资金也反过来让早期合伙人能向创始人承诺长周期支持，从而在种子轮竞争中胜出。Aram Verdiyan 从 LP 视角对此高度认可：最强大的后期投资机构必然附着在一个强大的早期品牌之上；若要在某个品类定义级公司中一掷千金（单笔开出 5 亿美元支票或占据基金 5% 到 10% 的仓位），必须仰赖从创立初期就与团队建立的深厚信任，毫无早期根基的纯后期（De Novo Late Stage）机构根本无法凭空拿下这种核心份额。 - 创始人品牌依附：顶尖创始人首要诉求是进入顶级机构的生态轨道，而非纠结投资工具的来源。 - 业务本质统一：风投核心始终是早期捕获能力，唯有提早入局才能掌握投资主动权并保留后续追投（Pro Rata）的权利。 - 阶段互锁效应：成长阶段资本提供规模化追投背书，早期网络提供信息与入场券，纯中后期单点打法被挤压。 --- ### [20:32](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1232s) LP 视角下的风投四分天下与 AI 虚实之困 从资产配置与 LP（出资人）的结构性视角来看，Aram Verdiyan 梳理了当前风投市场的四大细分阵营： 1. **Pre-seed 与 Seed 基金**：规模通常在 1.5 亿美元以下，仅在美国就有接近 2000 家； 2. **“混乱的中部”机构（Messy Middle）**：规模介于种子与大牌之间，数量有数百家之多，处境最为尴尬； 3. **大型综合平台型机构（Big Firms）**：如 a16z 这类跨阶段、多武器库的头部品牌； 4. **纯后期专项基金（Dedicated Late Stage）**：因缺乏早期支撑，Accolade 等 LP 近年来已基本不再配置此类资产。 然而，AI 的爆发正在让投资人的工作变得前所未有的艰难。融资轮次规模急剧膨胀，交易推进节奏被无限压缩，而最令人困惑的是市场上充斥着大量真假难辨的“虚假业务增长”（Confusing Traction）。Aram 形象地描述了当下孵化器里屡见不鲜的怪象：一家刚出加速器的初创公司宣称“我们只用了一个月就从零做到了 500 万美元 ARR”，但背后不仅毫无客户留存周期的检验，甚至很多仅仅是同批次学员互买互刷、将单月收入乘以 12 换算出来的伪 ARR。然而，它们却能以此在极高估值乘数下迅速完成融资。更加剧挑战的是，在每九家这种虚火泡沫型的公司中，偏偏就混着一家真正做到数百万扎实 ARR、估值高企并最终蜕变为类似 Cursor 级别的下一个时代奇迹。 - 资产分类版图：美国本土 sub-150M 的种子基金接近 2000 家，而大机构与精选微型基金正分流大部分超额回报。 - AI 估值失真：缺乏完整续约周期的单月爆发被简单年化，同行业相互采买造成数据虚高。 - 甄别难度激增：大机构虽有资本实力等待相对明确的信号再重锤领投，但在极早阶段准确筛出极少数真金的容错空间被严重压缩。 --- ### [22:18](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1338s) 穿透虚假 Traction：从 Cursor 的争议到 Harvey 的真正爆发 面对纷乱的市场噪音，如何去伪存真？在讨论 Cursor 早期那笔“300 万美元 ARR 对应约 4 亿美元估值”的投资时，外界曾普遍认为估值极其疯狂，甚至就在收购传闻前后仍有大量声音唱衰 Cursor“必死无疑”。David George 强调，在早期阶段，财务指标往往无能为力，核心完全在于对创始人认知的判断以及对底层真实需求的把握。他笑谈自己每天贴在电脑屏幕上的便利贴只有一句话：“市场真的在向你的产品索要更多吗？”（Is the market demanding more of your product?） 在产品上线仅数月的初期阶段，传统意义上的财务分析、队列分析（Cohort Analysis）和续订率数据完全无法提供有效信号，甚至会误导判断，投资人必须深入到一线客户的真实使用肌理中去。David George 以法律 AI 独角兽 Harvey 的发展历程给出了经典印证： - **第一阶段（初期势头与真实脱节）**：Harvey 由顶尖研究员与专业律师联合创立，商业拓展极具进攻性，早期迅速签下了数家知名律所。但如果仔细剖析其背后的软件部署与实际使用强度，表现相当平庸，并未展现出真正的杀手级黏性。 - **第二阶段（推理模型催生质变）**：随着具备高级推理能力的大模型（Reasoning Models）上线，局面发生了根本性颠覆。律师从产品中获得的实际价值实现跃迁，用户活跃与使用频率瞬间爆发。 - **第三阶段（被动防御转向刚性需求）**：客户心态彻底完成了从“担忧模型幻觉”到“所有委托客户主动强制要求律所必须使用 Harvey”的逆转。 David George 总结道，任何人都可以摆弄财务模型和留存表格，但唯有深潜至技术演进与客户需求交汇的细微纹理中，才能在缺乏财务确证的迷雾里，比市场更早识别出从虚假泡沫中破土而出的真正胜者。 - 质性分析重于量化数据：对于极早期的 AI 公司，常规财务指标往往失效，理解客户替代方案与痛点深度是唯一解法。 - 创始人判断维度：长周期的高频互动，审视其思考深度与快速迭代能力，是应对估值波动的锚。 - 临界点信号（Infection Signal）：从客户“试用防范”转变为“客户倒逼供应商必须使用”，是 AI 应用跨越鸿沟的核心标志。 ### [25:23](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1523s) LP 的顾虑与激发动机错配：“接飞刀”还是保住饭碗？ 对于投资机构而言，早期投资业务至关重要，因为早期团队往往扎根于底层技术与产品的最前沿，能够敏锐捕捉到像 Cursor 这类颠覆性产品的崛起。然而转向募资端时，作为业内极为高产的募资推手，Jen Kha 面对 LP（有限合伙人）时遇到的最大阻力正是对市场时机与泡沫的焦虑——大家都在心惊肉胆地追问：“我们现在是不是在接一把下坠的飞刀（catching a falling knife）？”市场估值是否过热、AI 的实际潜力能否兑现，始终笼罩在 LP 的心头。 更深层的阻力则植根于 LP 与 GP 之间截然相反的激励机制与风险容忍度。对于风险投资家（GP）来说，最大的罪过是“错失的错误”（error of omission）——如果漏掉了下一个 Facebook 或 Uber，GP 就可能被合伙人或市场淘汰。然而，LP 的世界遵循着完全不同的生存法则：“没人会因为投资了 IBM 而被解雇”，LP 只有在投出彻底爆雷的灾难性项目（invest into a mess）时才会面临丢饭碗的风险。如果一名 LP 错过了最前沿的基础大模型，但整体回报刚好贴着基准线甚至略低一点，其职位依然稳如泰山。这种“上行空间收益不大、下行暴雷万劫不复”的结构，使得传统 LP 缺乏拥抱剧烈变革的内生动力。 面对这种错配，头部 VC 的职责不仅是兜售“别错过下一代超级赢家”的 FOMO 情绪，更在于引导机构投资者看清技术浪潮的方向，帮助他们理解 AI 将如何不仅重塑风险投资本身，更将全面渗透并重写他们手头所有大类资产的底层逻辑。 ```mermaid flowchart LR accTitle: GP与LP激励机制差异 accDescr: 对比GP因错失超级项目而被解雇与LP因犯错才会被解雇的激励不对称 subgraph GP["GP（风险投资机构）"] GP_Fear["最大恐惧：错失机会（Error of Omission）"] GP_Out["错过下一个 Facebook / Uber 会被市场淘汰"] GP_Action["激进承担风险，追逐极端幂律回报"] GP_Fear --> GP_Out --> GP_Action end subgraph LP["LP（出资人 / 机构分配者）"] LP_Fear["最大恐惧：合规与暴雷（Error of Commission）"] LP_Out["只要贴近基准线，哪怕错过前沿模型也能保住饭碗"] LP_Action["厌恶变革风险，缺乏追逐上行极致的动力"] LP_Fear --> LP_Out --> LP_Action end ``` --- ### [28:09](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1689s) LP 的核心功课：准入、筛选与真正能带来改变的“仓位规模” LP 资产配置的核心框架无外乎三件事：准入（Access）、筛选（Selection）与仓位规模构建（Portfolio Construction & Sizing）。数据显示，全市场约 3,000 家风投机构中，能够持续创造卓越回报的顶尖机构仅有 20 家左右，即便在这 20 家中，能跨周期保持业绩一致性的也只有半数左右。因此，LP 既要保证进入头部老牌基金的门额，又要挖掘下一代新锐机构。 在投资组合的构建上，许多机构往往犯下分散投资的大忌。在风险投资这种幂律特征极强的资产类别中，如果资产包里塞进了 50、60 甚至 70 家风投基金，整体表现必然被平庸吞没，甚至很难超越行业平均水平。而风投的行业平均回报，根本无法补偿其长期流动性锁定的折价和高达 60% 的项目亏损率；相比之下，私募股权（PE）往往能以更低的风控成本和极低的亏损率提供 1.5 到 2 倍的净回报。 更严重的通病在于仓位下注过轻。现实中屡见不鲜的情形是：一位配置人敏锐地发现了一支极具潜力的优质黑马基金，判断完全正确，但最终只把自身资产池的 1% 分配给它。结果即便这支基金获得了惊人的 10 倍回报，折算到整个母基金层面也仅仅贡献了 10% 的收益，对大盘整体净值几乎“根本没有推动力”（does not move the needle at all）。 - **全市场分化悬殊**：3,000 家 VC 机构中真正能跑赢大市的仅约 20 家，且其中仅约一半能维持历史业绩的一致性。 - **过度分散陷阱**：配置 50 至 70 家 VC 基金等同于被动买入风投大盘指数，在扣除流动性溢价后性价比极低。 - **下注必须有厚度**：缺乏仓位集中度的成功挑选毫无意义，1% 权重的 10 倍回报无法改变母基金整体回报命运。 --- ### [29:36](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1776s) 跨资产类别的 AI 渗透：公募 SaaS 估值分化与私募股权（PE）的困境 AI 已经成为贯穿所有资产类别的“结缔组织”（connective tissue）。在公募软件市场，估值体系已然彻底崩塌并重建：过去曾有数十上百家传统 SaaS 公司以 10 倍以上的市销率（P/S）交易，如今这一梯队被残酷压缩至仅剩 15 到 20 家。而这寥寥无几能够站稳 10 倍 PS 以上的企业，无一例外都在借助 AI 实现业务增速的重新跃升，涉及监控、安全、AI Agent 部署等关键节点。资本市场的态度非常决绝——如果一家公司无法与 AI 这一全社会 GDP 增长最快的驱动力紧密绑定，其估值乘数就会被无情打折。 这种冲击对私募股权（PE）的打击尤为致命。过去，PE 往往倾向于以 15 到 20 倍的 EBITDA 估值收购那些年增长 10% 到 20% 的稳健软件公司，靠按席位收费的 SaaS 模式和财务杠杆获取稳健收益。但时至今日，这类老旧软件资产在公开市场的市销率已经惨跌到 2 倍左右。更可怕的不是估值腰斩，而是**这类公司很可能找不到下一个接盘者**。 当下的并购与买方市场在评估任何一家软件公司时，首要考量的就是其“终局价值”（terminal value）——它是否能抵御 AI 的侵蚀与替代？在公开市场，数据的反馈极为严苛：“每增加 1 个百分点的内生业务增速，在估值贡献上相当于提升 3 个百分点的 EBITDA”。2021 年市场疯狂推崇自由现金流与盈利能力，而如今市场重新为承担风险的高增长买单；缺乏高速内生增长且管理团队缺乏 AI 原生思维的传统 PE 软件资产，正陷入进退维谷的流动性泥潭。 | 资产类别 / 时代阶段 | 传统估值与投资逻辑 | AI 时代全新估值与现实冲击 | |:--- |:--- |:--- | | **公募 SaaS 市场** | 数十家企业轻松享有 10x+ P/S；席位制软件稳步提价 | 仅剩 15-20 家能维持 10x+ P/S；只有展现出 AI 驱动增速逆势上扬的企业才配享溢价 | | **私募股权（PE）旧软件持仓** | 支付 15-20x EBITDA 收购增长 10-20% 的成熟软件，追求杠杆现金流 | 公开市场同类资产仅值 2x P/S；面临终端价值归零风险，恐面临“无买家接盘”的流动性死局 | | **估值敏感度偏好** | 极度追求静态盈利与 EBITDA 率（尤其在利率抬升周期） | 增长重回王座：“1 个百分点的内生增长等同于 3 个百分点的 EBITDA 估值贡献” | --- ### [32:25](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1945s) 终局价值焦虑与 Intercom 式的“自我革命” 与 PE 市场的疲软形成鲜明对比的是，风险投资领域的大型并购退出规模甚至开始反超传统 PE。今年 PE 规模最大的两笔交易（如 EA 游戏约 500 亿美元的收购案、Medline 约 500 亿美元的退出）虽然体量庞大，但风投领域的超级整合——例如 Cursor 出售给 SpaceX 这一级别的并购——在估值规模和技术溢价上毫不逊色。 传统软件公司摆脱泥潭唯一的出路在于根本性的“自我重构”，而 Intercom 的绝地反击则被奉为当下的孤本典范（an N of 1）。Intercom 的做法极具魄力与毁灭性：让创始人重掌帅印，大刀阔斧推翻旧有业务体系，将其彻底打造为一个 AI 原生产品，完成规模化后再重新推向市场。这种做法无异于“为了重生而主动自杀现有业务”（almost suiciding your existing business）。这种脱胎换骨式的涅槃在 VC 支持的创业公司中尚有一线生机，但在依靠稳定现金流偿还杠杆债务的私募股权（PE）运作模式下，几乎是无法承受且不可能落地的禁区。 然而，手握大量传统 SaaS 资产的并不仅是 PE，深耕行业数十年的传统 VC 同样面临着庞大旧存量资产的处置难题。面对那些前 AI 时代的 SaaS 投资组合，投资机构必须在护航老旧资产与果断转向 AI 原生前沿之间做出艰难抉择。 ### [33:56](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2036s) 传统 SaaS 的困境：停滞的估值与消失的退出通道 对话从一个非常尖锐的问题切入：如果投资组合里有一家 2016 至 2021 年间（即 ChatGPT 问世前）成立的传统软件公司，它表现尚可但完全缺乏 AI 原生功能，增速放缓至 30% 左右，而在风险投资账面上仍按 10 到 20 倍的市销率（P/S）计价，那么这类公司的出路究竟在哪里？如今公开市场对这种标的毫无兴趣，甚至连一年前可能接盘的顶级私募股权巨头（如 Silver Lake）也已经彻底对其失去了收购胃口。 面对这类资产，投资人和创业者都在探寻底线真相。一位投资组合公司的创始人兼 CEO 在周末私下交流时直言不讳地要求听到真话：“别给我讲播客里的那套场面话，告诉我到底在发生什么。”现实是复杂的，两种看似矛盾的判断正在同时发生：一方面，AI 确实是前所未有的划时代变革，将彻底重塑所有行业；另一方面，那些具备自我革新和适应能力的软件公司，其长期价值依然存在，绝非瞬间归零。 - 传统 SaaS 困境特征：年增速放缓至约 30%，账面估值维持在 10–20 倍收入的高位，缺乏 AI 原生架构。 - 退出渠道收窄：公开市场 IPO 窗口对传统软件关闭，以 Silver Lake 为代表的大型 PE 机构也不再愿意接盘此类资产。 --- ### [34:26](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2066s) AI 扩散的现实：代码生成只是“假动作”与万亿级蓝海 从技术扩散到实体经济的角度来看，传统软件并非立刻失去一切阵地。一个经常被讨论的观点是：AI 在代码编写领域的爆发，在某种程度上是一次“假动作”（head fake）。因为代码拥有几乎完美的数据记录、完全可验证的结果以及高度可模拟的环境；然而，商业社会中的绝大多数知识工作根本不具备这三大特征。因此，AI 向更广阔知识工作领域的渗透速度，可能会远慢于许多人的悲观预期。传统软件公司如果能及时融入 AI 解决方案并主动调整商业模式，依然有演进的时间窗口。 然而，这丝毫不削弱投资机构对 AI 的绝对看好。在当前的头部投资组合中，约 95% 的净资产价值（NAV）并不在这些承压的传统软件公司上，而是高度集中于那些增速持续暴击历史纪录的纯正 AI 领跑者。目前 AI 在企业内部的消耗呈现出极其悬殊的阶梯分布：全美企业的月度 AI 支出中位数仅为每人每月 12 美元，但排名前 1% 的企业每人每月的 AI 支出已经高达 7,000 美元。即使是最激进的前沿银行，其在 AI 工具上的投入也仅占其人力成本的 1% 左右。整个 AI 产业仅仅依托两三千万核心用户的采用，就创造出了超越传统巨头单月增量的惊人营收；而在全美甚至全球庞大的知识工作者体量面前，这场生产力革命才刚刚拉开序幕。 ```mermaid flowchart TD accTitle: AI 渗透与任务特征对比 accDescr: 对比代码生成任务与通用商业知识工作任务的特征差异 subgraph Coding["代码编写领域（已爆发）"] C1["数据记录极其完备"] C2["输出结果确定且可验证"] C3["环境与运行完全可模拟"] end subgraph Business["通用知识工作（扩散缓慢）"] B1["上下文碎片化且非结构化"] B2["质量主观且难以量化验证"] B3["依赖真实商业环境无法简单模拟"] end Coding -->|被视为| A1["'假动作' Head Fake"] Business -->|决定了| A2["传统企业改造与演进的缓冲窗口"] ``` - 渗透率极端分化：全美企业员工月均 AI 支出中位数为 12 美元，顶尖前 1% 的企业高达 7,000 美元/人/月。 - 人力成本占比：即便在推进最激进的金融机构中，AI 工具的预算也仅占员工人力成本的约 1%。 - 增长势头对比：最头部的 AI 原生公司以创纪录的速度扩张，月增营收体量已超越传统科技巨头，而底层用户渗透仅覆盖了 1,000 万至 3,000 万人。 --- ### [37:24](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2244s) 资产配置洗牌：告别零和思维与 CalPERS 的逆向调仓 在机构投资端，关于传统与前沿的再平衡正在剧烈上演。加州公务员退休基金（CalPERS）曾因未抓住特定技术周期而错失数十亿美元的潜在收益，如今他们正大举纠偏：将其私募股权配置从 91% 削减至 58%，同时将风险投资与成长期投资的占比从 9% 大幅拉升至 43%。这种极端的资产调仓表明，大型机构正极力向技术创新前沿倾斜，但市场并不应该滑向绝对的“零和思维”。 许多传统软件企业依然能找到自我价值释放的路径。例如 Airtable 剥离其“HyperAgent”业务板块，在保留核心基本盘的同时将前沿智能体产品独立运营，这证明存量公司在 AI 浪潮下依然拥有多样的生存和变现途径，并非只有全面颠覆一种结局。 - CalPERS 资产配比变化：传统私募股权（PE）配比从 91% 下降到 58%，风投及成长期投资（Venture & Growth）从 9% 激增至 43%。 - 应对策略：摒弃非此即彼的零和博弈思维，传统软件公司可以通过分拆或重组高潜力 AI 业务（如 Airtable 剥离 Agent 业务）实现价值重塑。 --- ### [38:05](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2285s) 私募信贷与杠杆收购的隐秘雷区 随着过去十年私募信贷（Private Credit）的野蛮生长，软件资产不仅关乎股权投资，更深度卷入了数千亿美元的信贷市场。在 2021 至 2022 年的高峰期，全市场发生了 2,000 亿至 3,000 亿美元的软件杠杆收购（LBO），其中伴随着超过 2,000 亿美元的债务融资，当时的收购估值普遍高达息税折旧摊销前利润（EBITDA）的 25 到 32 倍。 然而，在经历了公开市场软件板块的大幅回调与暴跌（所谓“SaaS 启示录”式大清算）后，这些被收购公司的实际公允价值普遍已经腰斩。资产估值的急剧萎缩直接导致底层债务的杠杆比率（Leverage Ratio）被动暴增，这正是近期私募信贷市场遭遇大额赎回的根本原因。如果一家高杠杆在身的软件公司无法证明自身对 AI 冲击的防御能力，它将在股权清零和债务违约的双重夹击下陷入泥潭。 | 指标 / 维度 | 2021–2022 高峰期 LBO 软件交易 | 当前市场现状 | |:--- |:--- |:--- | | **交易与债务体量** | 2,000–3,000 亿美元交易总额，负债超 2,000 亿美元 | 面临严重的流动性枯竭与信贷赎回潮 | | **估值倍数 (EV/EBITDA)** | 25 倍 – 32 倍 EBITDA | 估值普遍腰斩，公允价值急剧收缩 | | **资产负债表健康度** | 基于高增长预期的安全杠杆设计 | 资产缩水导致实际杠杆率失控飙升 | --- ### [39:30](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2370s) 为什么“AI 私募股权”绝非万灵药 部分机构寄希望于通过“AI 私募股权”模式来拯救陷入困境的存量资产，试图买下传统公司后通过注入 AI 降本增效，但这种想法往往只是天真的幻想。正如当年“把西尔斯百货（Sears）搬上网页并不能把它变成亚马逊”，亚马逊的成功源于底层物流、仓储与系统从零开始的原生重构，而非仅仅套上一层前端外衣。 很多 PE 背景的软件企业急于采摘所谓的“低垂果实”，第一步就是裁减人工客服、换上 AI 客服智能体。但在业务流程根本没有被底层重构的情况下，强行切断客户习惯的人机交互体验，往往会立刻引发客户净推荐值（NPS）的大幅崩塌。而在企业级软件领域，NPS 的每一个百分点下滑都与客户流失和营收缩水直接挂钩；对于背负巨额并购贷款的公司而言，收入下滑会迅速引爆债务违约的恶性螺旋。靠派遣一名所谓的投后“运营合伙人”（Operating Partner）进驻企业、喊着“全面拥抱 AI”的口号根本行不通。除非企业管理层拥有如 Intercom 那样决绝的创始人意志，且董事会与股东全员高度协同，敢于承受断腕转型的短期剧痛，否则外部强加的 AI 改造只会加速毁灭。 > "就因为你把西尔斯百货搬到了网站上，它也不会变成亚马逊。你必须从地基和物流层面彻底重构，才能造就亚马逊。" --- ### [40:57](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2457s) 退出周期之辩：复利与耐心的终极回报 对于风险投资与成长期投资，市场上最强烈的质疑主要集中在极长的流动性回收周期（Timeline to Liquidity）。如今一家独角兽企业保持私有化状态的平均时间已长达十年以上，中途往往伴随着高密度的后续融资轮次，同一家公司的名字会反复出现在五六家不同投资机构的年报中。不仅如此，IPO 也不再等于即时分红变现，在持有 10% 到 15% 股权的情况下，机构往往需要 12 到 24 个月甚至更久的时间才能通过公开市场完成有序减持。 对此，核心的应对逻辑在于对顶尖赢家复利效应的敬畏。全行业三千多家风险投资机构中，真正能够持续创造超额回报的只有前二十家左右（前 1%）。对于这类机构捕获的划时代平台型企业，过早寻求流动性反而是最大的战略失误：三四年前任何投资人都不可能愿意提前卖掉手中的 Stripe 或 Databricks 股份。 与此同时，AI 原生企业的生命周期正在被剧烈压缩：Anthropic 从 2021 年完成初始融资到准备迈入公开市场仅用了约五年时间；而像 Cursor 这样的新生代工具，从首次融资到展现极高并购或商业价值的速度更是空前迅猛。顶级风投通过捕获这类企业收回基金本金的速度，甚至可能超过传统私募股权基金。为了直观说明这一现象，a16z 讲述了一个真实的案例：大约一年半以前，他们针对已经存续 16 年的 a16z 一期基金（Fund I）去征询 LP 的意向，询问他们是否希望对种子轮就入局并持有至今的 Stripe 仓位进行变现分红，因为作为风投基金，跨越 16 年的陪跑在形式上早已“功德圆满”。 - 退出周期的现实压力：独角兽平均私有化时间超过 10 年，IPO 后的减持锁定期通常需要 12–24 个月。 - 幂律分布的集中度：在约 3,000 家风投机构中，仅有约 20 家头部机构能够稳定产生超额回报。 - AI 时代的周期压缩：Anthropic 从首次获得投资到筹备 IPO 仅用约 5 年时间，Cursor 等企业成长节奏同样显著快于传统 SaaS。 - a16z 一期基金的 Stripe 案例：基金存续第 16 年时，管理团队主动向 LP 发问，探讨是否对种子轮持有的 Stripe 股权进行清盘分红。 ### [43:00](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2580s) LP 的退出偏好分化与 GP 的流动性构建艺术 在基金运作后期，流动性退出往往伴随着复杂的利益平衡。即使面对成立长达 17 年、必须清算退出以终结周期的老基金，当 GP 去询问投资人是否需要拿回流动性时，所有 LP 竟然异口同声地表示：“不，我们宁愿让它继续复利增长（No, we'd rather let this continue to compound）。”这种现象揭示了不同类型机构投资者（LP）截然不同的资金诉求：大学捐赠基金（Endowments）拥有极长的投资视野，倾向于让资产长期在市场中滚雪球；家族办公室（Family Offices）则极度排斥在短期内拿回现金，因为这会触发沉重的资本利得税负担，他们同样更希望资产在免于纳税的环境下持续复利。因此，投资界很难用一刀切的刻板眼光来概括所有 LP 的诉求。 然而，这也极其考验普通合伙人（GP）主动“制造流动性”的能力。回顾 2021 年的高点，许多机构盲目惜售而错失了套现良机，而真正顶尖的 GP 则懂得在周期高位主动落袋为安。在当前的新周期中，GP 更加需要通过精准的节奏把控来平衡流动性：一方面通过早期的并购交易（M&A）实现部分阶段性套现，向 LP 返还资金；另一方面则保留核心头寸，耐心等待真正的超级赢家最终通过 IPO 走向公开市场，充分享受企业价值全面复利释放带来的红利。 - **捐赠基金（Endowments）**：超长投资周期，追求长坡厚雪的终极复利，不急于要求返还本金。 - **家族办公室（Family Offices）**：极度受税收驱动，倾向于延迟资本利得确认，更愿资产长期沉淀在优质标的中。 - **优秀 GP 的流动性策略**：打破坐等 IPO 的被动模式，在周期波动中通过并购、二次转让（Secondary）主动创造流动性，同时护送核心赢家冲刺公开市场。 --- ### [44:01](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2641s) 展望 10 万亿乃至 100 万亿美元市值的超级新物种 当对谈抛出“谁将成为下一家 10 万亿乃至 100 万亿美元市值的公司”这一震撼命题时，现场气氛瞬间变得热烈而带有思考的重量。尽管 100 万亿美元听起来极其夸张，甚至需要经历跨越两代的技术周期（Tech Cycles），但能够支撑起 10 万亿以上体量的终局赢家，绝不可能仅仅依靠现有存量业务的修补，而必然诞生于那些尚未被数字科技彻底改造的全新产品领域： 1. **企业级 AI 的全面扩散**：目前企业端对 AI 技术的实际渗透与扩散才刚刚起步，远未进入深水区。 2. **重塑消费者交互形态（Consumer AI）**：当前以对话框为主的聊天机器人界面只是一种“拟物化（Skeuomorphic）”的初级过渡形态。未来真正的原生 AI 应用绝不会是一个聊天窗口，而是能够全自动、主动（Proactive）替用户完成现实复杂工作、创造巨大增量价值的智能代理。虽然当前全球已有数亿人在使用聊天工具，但这仅仅触碰到了消费级 AI 潜力的皮毛。 3. **具身智能与机器人（Robotics）**：机器人产业展现出的体量潜能未来将远远超越单纯的语言大模型，且极有可能在未来 10 年内发生爆发性突破。 4. **全自动驾驶与出行（Autonomy）**：自动驾驶目前几乎处于原初阶段，全美活跃运行的 Waymo 机器人出租车（Robotaxi）尚不足一万辆，未来全球范围内留给造车、出行与运营网络重构的战略空白极其宽广。 5. **占 GDP 18% 的医疗健康**：无论是医疗服务的交付流程，还是基于 AI 的新药研发（Drug Discovery），目前全行业充其量只是触及了冰山一角。接下来的 10 年将见证生命健康领域的巨额价值重构。 6. **物理世界的全面重塑**：从国防科技、高端制造到新型数据中心基础设施，AI 正在向整个物理现实全面渗透。未来 10 年诞生媲美 OpenAI 或下一个 SpaceX 级别的万亿级巨头，大概率就会扎根在这些实体与智能交织的领域。 --- ### [46:36](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2796s) 供给侧瓶颈与“机器时代”的万亿级物理底座 当前 AI 产业发展的最大制约绝非需求不足，而是一场彻底的“供给侧危机”。传统的软件创投生态习惯于聚焦上层应用与模型，但真正的咽喉却卡在底层的物理现实中。美国在芯片研发、前沿大模型及应用软件等技术栈右侧拥有极强优势，且有成熟的 VC 体系支撑；然而在底层能源与电网等技术栈左侧，却面临着极其严峻的体制与工程瓶颈。 ```mermaid flowchart LR accTitle: AI供应链价值链瓶颈图 accDescr: 从左侧物理基础设施瓶颈流向右侧模型与应用层 subgraph LeftSide["技术栈左侧：物理基础设施瓶颈"] A["能源生成 & 电网传输<br/>(速度与审批瓶颈)"] --> B["下一代超高密度数据中心<br/>(10倍功率密度重构)"] end subgraph RightSide["技术栈右侧：成熟创投生态优势"] C["先进制程芯片 & 存储"] --> D["前沿基座模型 (Frontier Models)"] D --> E["企业级/消费级应用"] end B --> C ``` 美国当前面临的真正痛点并非电力发电总量不够，而是“通电速度（Speed to power）”受阻——环评审批、输电线网铺设以及繁杂的监管流程极大地拖慢了步伐，而其他国家每年新增的可再生能源并网装机容量往往是美国的 10 倍以上。同时，下一代 AI 算力集群对电力密度的需求直接飙升了 10 倍以上，传统数据中心根本无法通过修修补补来改造适配，必须从物理架构层面彻底重构。 很多持怀疑态度的 LP 容易将本轮 AI 浪潮类比为当年的互联网泡沫（Dot-com）或新冠疫情期间的短期红利暴增。但本质区别在于，AI 当前的商业牵引力（Traction）是极其真实的硬核需求，绝非短暂虚幻的泡沫收入。谁能有效解决能源接引、超高密度数据中心建设以及下一代芯片与存储架构的物理瓶颈，谁就将捕获价值 500 亿甚至 1000 亿美元以上的时代机遇。正如对话最后充满激情的总结：“是时候迎来机器时代了，让我们迎接机器的大举降临（It's time for machine age. Let's bring the machines）。” ## Notable quotes > [00:00](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=0s) 这是第一次你把巨额资本砸向一家公司，资本能直接为它带来滚雪球般的竞争壁垒 — David George > [13:41](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=821s) 对超大型 VC 基金的需求是由创始人驱动的，而不是 LP。创始人普遍想要的是具备真正规模化能力、能做全生命周期投资、并在招聘和获客上提供强效支持的超级品牌。LPs 只是在后知后觉地跟进，因为这与传统分散投资的常识背道而驰。 — Eddie > [17:04](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1024s) 我们的业务始于早期，也终于早期。 — David George > [22:18](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1338s) 市场真的在向你的产品索要更多吗？ — David George > [25:23](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1523s) 没人会因为投资了 IBM 而被解雇 — Aram Verdiyan > [29:36](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=1776s) 每增加 1 个百分点的内生业务增速，在估值贡献上相当于提升 3 个百分点的 EBITDA — David George > [33:56](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2036s) 别给我讲播客里的那套场面话，告诉我到底在发生什么。 — 投资组合公司创始人 > [39:30](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2370s) 就因为你把西尔斯百货搬到了网站上，它也不会变成亚马逊。你必须从地基和物流层面彻底重构，才能造就亚马逊。 — David George > [43:00](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2580s) 不，我们宁愿让它继续复利增长 — LP 投资者 > [46:36](https://www.youtube.com/watch?v=bsdJd2VeLvg&t=2796s) 是时候迎来机器时代了，让我们迎接机器的大举降临 — David George ## Mentioned Jen Kha, David George, Aram Verdiyan, Erik Torenberg, Eddie, a16z, Accolade Partners, SpaceX, OpenAI, Anthropic, Harvey, Siebel Systems, Cambridge Associates, Speedrun, Cursor, IBM, Silver Lake, CalPERS, Airtable, Intercom, Sears, Amazon, Stripe, Databricks, Waymo, Electronic Arts, Medline ## Video description a16z’s Jen Kha and David George sit down with Accolade Partners’ Aram Verdiyan to discuss how AI is changing the power law of technology investing, why the largest companies can compound advantages in ways that weren’t possible before, and what that means for how investors construct portfolios. They explore why AI may be much bigger than traditional software, with applications reaching into labor, healthcare, transportation, services, and other major parts of the economy. David explains why capital itself can now reinforce an AI company’s advantage by buying more compute, while Aram makes the case that AI should increasingly be treated as a core allocation rather than a satellite position. The conversation also gets into the changing economics of venture and growth investing, how to distinguish real AI traction from early hype, what AI means for legacy software and private equity, and why some of the largest opportunities may still be ahead in robotics, autonomy, healthcare, energy, and physical infrastructure. Timestamps: 00:00 - Intro 00:44 - Why Power Law Is No Longer Just a Venture Thing 01:34 - Every Venture-Backed IPO Combined: Where Does It Go From Here? 03:43 - Rethinking Portfolio Construction from a Blank Sheet 08:27 - Why This Era of AI Is Categorically Winner-Take-All 10:40 - Why Consistency Matters More Than Ever in Venture 20:32 - How Venture Has Structurally Changed Since the 2000s 25:49 - Are We Catching a Falling Knife? LP Sentiment Today 33:39 - The Legacy SaaS Problem: What to Do with the Old Book 39:30 - Why "AI Private Equity" Isn't a Panacea 47:17 - The Real Bottleneck: Data Centers, Chips & the Machine Age Resources: Follow Aram Verdiyan on X: https://x.com/aramverdi Follow Jen Kha on X: https://x.com/jkhamehl Follow David George on X: https://x.com/DavidGeorge83 Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Find a16z on X: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Show on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Show on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see http://a16z.com/disclosures. [Source](https://www.youtube.com/watch?v=bsdJd2VeLvg)

[0:00] David George: 这是第一次你把巨额资本砸向一家公司，资本能直接为它带来滚雪球般的竞争壁垒 [13:41] Eddie: 对超大型 VC 基金的需求是由创始人驱动的，而不是 LP。创始人普遍想要的是具备真正规模化能力、能做全生命周期投资、并在招聘和获客上提供强效支持的超级品牌。LPs 只是在后知后觉地跟进，因为这与传统分散投资的常识背道而驰。 [17:04] David George: 我们的业务始于早期，也终于早期。 [22:18] David George: 市场真的在向你的产品索要更多吗？ [25:23] Aram Verdiyan: 没人会因为投资了 IBM 而被解雇 [29:36] David George: 每增加 1 个百分点的内生业务增速，在估值贡献上相当于提升 3 个百分点的 EBITDA [33:56] 投资组合公司创始人: 别给我讲播客里的那套场面话，告诉我到底在发生什么。 [39:30] David George: 就因为你把西尔斯百货搬到了网站上，它也不会变成亚马逊。你必须从地基和物流层面彻底重构，才能造就亚马逊。 [43:00] LP 投资者: 不，我们宁愿让它继续复利增长 [46:36] David George: 是时候迎来机器时代了，让我们迎接机器的大举降临

[31:22] David George: Our data shows 1 percentage of growth in the public markets is equivalent to three percentages of EBITDA. [25:49] Aram Verdiyan: you don't get fired for buying IBM [31:22] David George: suiciding your existing business to rebuild it [43:00] Aram Verdiyan: No, we'd rather let this continue to compound. [22:56] David George: Is the market demanding more of your product? [12:51] Jen Kha: death of the middle [44:01] David George: two tech cycles away [44:01] David George: skeuomorphic version [47:17] Aram Verdiyan: speed to power

[0:00] We've looked at the data of 3,000 [0:01] venture capital firms in the US. Only 20 [0:03] have achieved consistent 3x net returns [0:05] over the last two decades. [0:06] >> Right now, clearly the power law is more [0:08] extreme than it has been in the last 10 [0:11] to 20 years of technology investing. For [0:13] the first time, you can take capital and [0:15] throw it at a company and it compounds [0:17] their advantage. [0:17] >> AI is attacking every facet of the GDP, [0:20] transportation, labor, services, [0:22] capital, coordination. There hasn't been [0:24] a technology paradigm that hits on 30 [0:26] trillion in GDP at the same Elon has [0:28] talked publicly about Grockbot on Sam's [0:30] side. He's talked about Astra and some [0:32] of the longrunning capabilities that are [0:33] going to come out soon. [0:34] >> What do you think is going to be the [0:35] next hundred trillion dollar market cap [0:38] company? [0:38] >> It is possible that [0:44] Welcome back to the A16Z podcast. [0:47] Something fundamental has changed in how [0:49] value gets created. Power law used to be [0:51] just a feature of a cottage industry in [0:54] venture capital and now it's systemic [0:55] throughout and particularly the three [0:57] frontier model companies SpaceX, OpenAI, [1:01] Anthropic represent somewhere between [1:03] three and a half to 5 trillion dollars [1:05] of potential enterprise value. And [1:06] shockingly before SpaceX went public, a [1:09] lot of our LPs and also the broader [1:12] institutional allocator community didn't [1:13] have a lot of exposure to it. And so [1:15] we'll talk about today why uh [1:17] potentially portfolio construction and [1:19] asset allocation may have changed um why [1:21] power law is not just only in the [1:23] venture capital industry and then [1:25] particularly where and how value [1:27] actually compounds today. David George [1:29] around Verdian thank you for joining me. [1:31] >> Great to be here. Thanks for hanging [1:33] out. [1:33] >> Thank you for having us here. [1:34] >> Awesome. Awesome. Awesome. Okay. So DG. [1:36] So if you add up every venturebacked IPO [1:40] for the last six years, all of them [1:41] together, where does it go from here? [1:44] Right now clearly the power law is more [1:46] extreme than it has been in the last you [1:49] know 10 to 20 years of technology [1:50] investing probably going back to you [1:52] know the emergence of the network effect [1:54] driven consumer companies. Um there are [1:56] many reasons why that's the case. [1:58] Increasing returns to scale have always [2:00] been a dynamic in our business. [2:02] Obviously it's well covered how a [2:03] network effect business can have [2:05] increasing returns to scale but so can [2:07] software businesses right and and they [2:09] can take different forms but you know [2:10] brand reputation in the market the [2:12] accumulation of resources all provide [2:14] competitive advantages that all still is [2:17] the case but right now especially with [2:20] the labs for the first time you know in [2:23] my career you can take capital and throw [2:26] it at a company and it compounds their [2:28] advantage and this is a thing like how [2:31] do you screw up a a startup, um well, [2:34] you know, throw too much money at it and [2:36] have them hire a thousand people and [2:37] then, you know, you create all these [2:38] coordination issues and overhead issues [2:40] and and dueling priorities and and it [2:42] sort of gets messed up because you can't [2:43] hire enough people to do enough things [2:45] fast enough. Um, now that's not the [2:47] case. You can throw dollars at compute [2:50] and compute can make products and the [2:52] businesses better. And so to me, it's [2:54] not terribly surprising that the power [2:55] law is more extreme. Right now, [2:57] economies of scale are a very real thing [3:00] uh in the AI market. and I think will [3:01] continue to be the case. [3:03] >> So, Ron, so first of all, uh you're not [3:04] just one of our longtime LPs at [3:06] accolade, uh but incidentally, it's been [3:08] exactly 10 years since you were actually [3:11] an employee of A16Z. And so, for your 10 [3:15] year anniversary, since you were last [3:16] year, I've brought this gem back. [3:18] >> Oh my god. Oh my gosh. This is amazing. [3:21] Why do you still have this? This is [3:22] amazing. [3:22] >> So, we dug in the catacombs uh and we [3:24] made this extra large version uh just [3:26] for for posterity here. I can't believe [3:28] that I got this from the from the [3:30] catacombs. But incidentally, during the [3:33] last 10 years, a lot has changed in the [3:35] world. And and if you remember, at that [3:36] point in time, people were belly aching [3:39] about fund sizes being too large back [3:41] then. [3:42] >> And you had one at a billion. I remember [3:43] >> the the first one out of billion. First [3:45] venture fund. Exactly. Exactly. And so, [3:48] you know, a lot has happened since then. [3:49] How do you think about your venture [3:50] portfolio juxtaposed against your [3:52] private equity one? And then just also [3:54] generally asset allocation. We were [3:56] talking about this on the way in. If you [3:57] were to start from a blank sheet of [3:58] paper again knowing what you know now, [4:01] how would you have constructed [4:02] differently? [4:02] >> Yeah, let's take venture today. We've [4:04] reached 100 billion in revenue in AI. It [4:06] took SAS 15 years to get to the same [4:08] point. AI did that in four years and [4:11] we're not even close to anywhere in [4:13] terms of the penetration of demand. The [4:15] reason DG is saying you can throw [4:16] capital at it the fun that's a function [4:18] of unlimited demand for inference and we [4:22] are at a point where AI is attacking [4:25] every facet of the GDP transportation [4:27] labor services capital coordination [4:30] there hasn't been a technology paradigm [4:32] that hits on 30 trillion in GDP at the [4:34] same time and so you have the fastest [4:37] growing technology it's hitting on all [4:39] parts of the GDP so as an allocator it's [4:43] hard not to make the case, you should [4:45] it's not a satellite position. You [4:47] should be core or a super core in some [4:48] shape or form. [4:50] >> I'm very biased, but if you just think [4:52] about the the shape of the markets and [4:54] how they've changed since I started my [4:57] career, um you know, in private equity [4:58] and growth equity, you know, that was I [5:01] guess 18 years ago. This was a cottage [5:03] industry and now it's not. And I talk [5:05] about this all the time, but our asset [5:06] class is 5 to6 trillion dollars of [5:08] value. and the dynamics around companies [5:11] staying private longer, they're not [5:13] going to reverse. [5:14] >> Yeah. I mean, you had that insight in [5:15] 2019 when you left G. It's venture- like [5:18] outcomes in late stage, which is now [5:20] happening. Like, [5:21] >> so it's no longer just early stage and [5:24] you IPO when you have 100 million in [5:25] revenues. [5:26] >> Yeah. The top desk outcomes, I think, [5:28] used to be 10 billion and now they're [5:29] like 40 billion. [5:30] >> Yeah. Or soon to be probably 100 billion [5:32] by the time anthropic and then open AI [5:33] came come out. Yeah. [5:34] >> Yeah. Yeah. Yeah. And look, this makes [5:36] sense, right? Like the last cycle [5:37] created 25 trillion of market cap, new [5:40] market cap and a bunch of that went to [5:41] the incumbents. [5:41] >> Y [5:42] >> but a lot of it went to to startups and [5:44] the new startups and you know each one [5:45] of these subsequent waves gets bigger [5:47] than the prior one. And so [5:49] >> you know our expectation is you know [5:50] that take the 25 trillion and like it's [5:53] going to be a larger number. [5:54] >> Yep. [5:54] >> Yeah. And I'm [5:56] constantly confused about the TAM of AI. [5:58] We'll love your thoughts on this. Um [6:00] like take healthcare. Healthcare spends [6:02] like 60 to 100 billion on healthcare it [6:04] per year. But AI is hitting on actual [6:08] labor and the value of tasks that are [6:10] being performed in healthcare. That's [6:11] claims, billing, administration. That's [6:13] a trillion dollar industry. So the TAM [6:16] of AI can be 10x plus bigger than [6:18] traditional SAS or healthcare IT. And [6:20] what is that value? Well, what's the [6:22] economic value of a task that's being [6:23] performed like that's the TAM you're [6:25] looking at. And then there's some [6:26] capture rate that the AI company will [6:28] take. But we have no idea what how big [6:31] the TAM can get. to your point, it's you [6:34] look at every wave, the incumbents are [6:36] 10x smaller uh over time. So, it's hard [6:39] to estimate it, but I can tell this for [6:42] myself. I've been chronically wrong [6:44] about how big these outcomes can get. [6:46] >> Yeah, same here. Yeah, labor. I mean, [6:47] look, if you just look at like how much [6:49] of, you know, dollars are spent in the [6:51] US economy on labor versus software, [6:53] it's like something like 40 times more. [6:55] Now, that doesn't mean importantly that [6:58] um labor is going to go away. Like I [7:01] think labor is just going to get [7:01] reinvented and so we'll we'll end up [7:04] with um you know sort of reimagination [7:06] of the task that humans do. Um but I [7:09] think that's actually the the the whole [7:11] point of AI is like you're going after [7:12] this different thing. And so to equate [7:14] it to software and say oh it's the next [7:16] evolution of software is far too [7:17] limiting. [7:18] >> Our legal council uh says to us often [7:20] it's like I love Harvey. All my clients [7:22] think they're lawyers now and they could [7:24] actually spar with me on topics where [7:26] they would have probably been like I [7:27] don't really understand this. I'm just [7:28] going to defer to you. So my billable [7:29] hours have only gone up with the advent [7:31] of the usage of AI. And so all those use [7:33] cases are are massively massively um [7:36] probably underappreciated and we don't [7:37] still even know [7:38] >> they're expansionary. Yeah. Is the whole [7:39] point. Yeah. [7:40] >> And then there was there was this whole [7:41] thesis where well Frontier Labs are [7:43] going to cannibise the apps which layer [7:44] is going to win. Turns out like everyone [7:47] is sort of growing. [7:48] >> Yeah. Yeah. Yeah. We just um you know a [7:50] friend of mine did a podcast where he [7:52] described like everything is going to [7:53] work kind of thing. Mhm. [7:55] >> Um, and um, you know, I I describe it [7:58] slightly differently, but it's like, you [7:59] know, we get questions all the time from [8:01] LPs when they ask us like, [8:02] >> you know, layer in the stack, [8:04] >> which layer in the stack is going to [8:05] work. And I'm kind of like, I don't [8:06] know, the the market is going to be so [8:08] big. Like, I think everything might [8:10] work. Now, there's going to be a lot of [8:11] companies that don't work and there may [8:13] be idiosyncratic categories that don't [8:15] work. Um, but I think by and large, it's [8:17] far too limiting to think like, oh, if [8:19] open source does a good job, it's bad [8:21] for the labs and, you know, vice versa. [8:23] Uh so we try and like remove ourselves [8:25] from thinking in a zero sum way like [8:26] that. [8:27] >> Yeah. [8:27] >> Extract it out because why why do people [8:29] think it's going to be a winner or take [8:31] all and and you know if we can [8:33] hypothesize but you know the last era of [8:35] technology it was probably winner take [8:36] all in a lot of categories but this [8:38] feels categorically different because [8:39] we're reunderwiting a lot of the [8:41] fundamentals. So maybe extract it out. [8:43] Yeah, look, winner take all is an [8:45] interesting way to describe it because [8:46] like if you just look at the market cap [8:49] growth of all the leading technology [8:51] companies like there are many many many [8:54] that were successful like there wasn't [8:56] winner take all right now there's an [8:58] important distinction like we very much [9:00] are believers in the power law within a [9:02] given category so like the winners will [9:04] capture the vast majority of the market [9:06] share um and market cap and second place [9:09] is like playing for scraps you know um [9:11] >> uh but I think there will see a massive [9:13] expansion of the amount of categories [9:15] that we have, right? And so, you know, [9:17] if you go back 20 years like CRM was not [9:20] really a category I mean it was small. [9:22] It was like seable systems and and [9:23] things like that. Um but you know now [9:26] it's a massive category. And so I think [9:28] the same thing will happen. We've seen [9:29] it in every technology market that we [9:31] invest in. Again, our approach is um you [9:34] know in our business we can tolerate [9:36] loss, right? And like if we're not [9:39] losing money in a given fund on a given [9:41] amount of investments, we're not taking [9:43] enough risk, right? And so, you know, if [9:46] you look at our best performing venture [9:47] funds over time, I think the loss rate [9:50] is 60%. Yeah. [9:51] >> Or so [9:52] >> on early stage. Yeah. [9:52] >> On early stage. Um, now at the growth [9:54] stage, like the loss rate will be lower, [9:56] but it's probably going to be in the 10 [9:57] to 20% range. And that's appropriate [9:58] because [10:00] >> with that, you will get investments that [10:02] we make that, you know, 10x or more in [10:04] returns. And so, um, you know, if we're [10:06] doing a good job, we're backing the [10:08] leading company in every category that [10:10] is a credible category. And if the [10:12] category works out well, then we do a [10:14] great job. And if the category doesn't [10:16] work out well, that's okay. That's kind [10:18] of like the risk that we live with. [10:19] >> Yeah. Yep. Yep. And and embedded in that [10:22] is also kind of timing because I I know [10:25] around you and I lament on this in that [10:27] I think a lot of folks often times think [10:30] that things are overheated in that [10:32] moment in time and then you look back in [10:34] retrospect and it turns out everything [10:36] was actually quite cheap but then [10:37] there's these aberrations in the market [10:38] where it's probably actually true and so [10:40] I know you advise a lot of your LPs on [10:42] on the cons importance of consistency in [10:44] venture capital probably even more than [10:46] any other asset class because you just [10:47] never know when these technologies can [10:49] come out. maybe walk through that [10:51] because there's a lot of institutional [10:52] allocators out there who actually don't [10:54] have access to a lot of the frontier [10:57] certainly models now. They're trying to [10:58] play catch-up and and in some instances [11:00] probably maybe introducing some adverse [11:03] behavior that is a little bit too [11:05] reflective of things being a little bit [11:06] too frothy. So maybe unpack that for us. [11:08] >> Yeah, I mean the extreness of the power [11:10] law that DG talked about. If you as an [11:13] allocator have not had access to the top [11:16] 5 to 10 companies over the last 5 to 10 [11:18] years, you're significantly behind in [11:20] terms of returns. And let's take a step [11:23] back. Like we've looked at the data of [11:24] 3,000 venture capital firms in the US. [11:26] Only 20 have achieved consistent 3x net [11:28] returns over the last two decades. [11:30] >> Sorry, say that one more. 20% have 20 [11:34] >> less than 1%. [11:35] >> Wow. [11:36] >> Consistent 3x net returns. [11:37] >> That's incredible. And it's actually you [11:39] don't have you don't need seven eight [11:41] funds in those 20 years. Are it was do [11:43] you have three to four 3x net TVPI funds [11:46] over a 20-year period. We found only 20 [11:49] >> firms that have done that. [11:50] >> Wow. [11:50] >> Consistency in venture is really really [11:52] hard. [11:53] >> But what's interesting is the consistent [11:55] ones [11:56] >> consistently had access to the category [11:57] defining companies every vintage. [11:59] >> Yeah. [12:00] >> Um now there are exceptions and by the [12:02] way just having the logo is not [12:05] sufficient enough. If you're early stage [12:06] and you have a large fund, you need to [12:08] own enough of it. [12:09] >> If you're late stage, DJ, I'm curious if [12:11] you agree, sizing is really critical. [12:13] >> Yeah. [12:14] >> Uh so venture- like returns are possible [12:16] in late stage, but your best company [12:18] should be 5 10% plus of your fund. That [12:20] way you can actually return the fund on [12:22] a single company. Fund returning math in [12:24] late stage didn't exist before. It now [12:26] does. [12:26] >> Yeah. [12:27] >> Um so we have found the right portfolio [12:30] sizing and the ones that consistently [12:31] have gotten access are in that top 20 [12:34] out of 3,000. So if you don't have them, [12:35] there's a huge dispersion of returns. [12:37] And if you don't have those, you're [12:39] getting the average venture return. If [12:42] you look at Cambridge data, the average [12:43] venture return over the last 10 years is [12:45] 1 to 2x. [12:46] >> You'll do better in private equity. [12:47] You'll definitely do better in the [12:48] public markets. You don't need to lock [12:50] up your money for 10 years. [12:51] >> Yeah, for sure. Yeah. I I just pulled up [12:53] a tweet uh from our friend Eddie who [12:55] posted actually this this morning. He [12:57] said, "Internest in big VC funds has [12:59] been driven by founders, not LPs. [13:01] Founders more often than not want the [13:03] brand that can scale, be a life cycle [13:05] investor, and help land customers/hires. [13:08] LPs have slowly followed along, but most [13:10] are still dragging their heels because [13:12] it actually it's counter to conventional [13:14] wisdom. [13:14] >> Yeah. And the outcomes are larger, so [13:16] funds can be larger. There are some [13:17] exceptions in those 20. There are some [13:19] small firms that are focused on niche [13:20] vertical markets [13:22] >> or they're playing at a stage that's so [13:25] much earlier than the bigger firms where [13:27] like there is not a lot of competition [13:29] with the bigger firms. Now the problem [13:31] with that strategy is you have to stay [13:33] consistent in terms of fund size and [13:34] their strategy. If you start getting [13:36] bigger over time then you bump into the [13:37] big firms and I think it it becomes [13:38] really really hard to stay consistent. [13:40] >> Yeah. Yep. Death of the middle. [13:42] >> Death of the middle of the middle. [13:43] >> We said we were going to drink every [13:44] time we said death of the middle. So [13:46] >> I'm very complimentary of of many of our [13:47] peers in the in the venture ecosystem. [13:49] Um you know like but this this death of [13:52] the middle thing [13:53] >> how do you define it? [13:54] >> Like what's the middle? Uh so I think [13:57] like sort of what you described like the [13:59] highly specialized funds you know some [14:01] of the ones that were very early to AI [14:04] with like deep deep deep domain experts [14:07] >> um have done a pretty good job right [14:09] like they've done a good job and and you [14:11] know sometimes they can move fast get [14:13] into things or take shares of deals that [14:15] we want to do and you know that that is [14:17] that is a reality um then I think [14:20] there's like the you know whatever we're [14:22] you know large you know large scale [14:24] venture right in theense is that we have [14:26] many product lines and we can scale all [14:28] the way from, you know, a seed all the [14:30] way through to when you go public. Um, [14:32] and I think we have some peers who [14:34] employ a similar strategy, right? And, [14:36] you know, there's I'd like to think that [14:38] we're the best, but there's there's a [14:39] few other folks who do that. Everything [14:41] else in between that, I think, struggles [14:44] to compete a little bit for the reasons [14:45] that Eddie said, right? The what does [14:48] the founder care about? the founder [14:49] cares about um you know sort of taking [14:52] capital from a partner that they think [14:54] can derisk the outcome for themselves [14:57] like that if you were just to simplify [14:59] it that is the simplest way to describe [15:01] what the founder really cares about. Now [15:03] they care about the partner, right? Like [15:04] the government person. So you have to [15:05] you have to be a good actor and all [15:07] those things. But there's a reason why [15:09] we build up a tremendous amount of [15:11] resources, right? Like that's why we [15:12] have 700 employees. Uh that's why we [15:14] take, you know, the management fees that [15:16] we make on our funds and we invest them [15:18] in operating resources because we think [15:20] that it will one bend the curve on the [15:23] outcome and two help us to win deals. [15:25] Um, and so, you know, when founders [15:27] select their partners often, you know, [15:29] if it's a hot deal, like they'll have [15:30] many alternatives. Like, that's the [15:32] revealed preference. And, and you know, [15:34] as Eddie said, we're doing an okay job [15:36] with that. What I agree with him about [15:38] then is our LPs coming to invest in us [15:41] is a byproduct of that, right? Um, and [15:44] so, you know, our business, our business [15:46] is a flywheel. Um, the flywheel starts [15:49] with um, you know, are we are we deep [15:51] domain experts? uh are we going to have [15:53] a point of view uh that is the right [15:55] point of view? Um can we demonstrate to [15:57] the founder that we you know are the [16:00] right partner for for her or him? Um if [16:02] so, we win the deal. If we can help to [16:05] make the outcome better, that's great. [16:07] Um and then if we do make the outcome [16:09] better, there's two things that happen. [16:11] One, um our business has persistence of [16:14] returns partially because um you know [16:16] the the new founder wants to be around [16:18] the winners, right? they want they care [16:20] about because there's important brand [16:22] and that has knock on effects for them. [16:24] Um and then secondly, you know, by being [16:26] a part of the winners and helping them, [16:28] you know, in small ways, uh we we create [16:30] sort of killer references and the [16:32] founders then tell the other founders, [16:33] hey, you should work with these folks. [16:35] And so that's the way the flywheel works [16:36] in our business. [16:37] >> One theory curious to get your both of [16:39] your takes is [16:41] >> take preed seats. So sub 20, 30, $40 [16:43] million valuation, sub$und00 million [16:45] funds, they can coexist with the big [16:47] firms because at inception stage, say [16:50] there's seven AI companies kind of doing [16:51] the same thing. I would think a large [16:53] firm like Andre would want to wait for a [16:56] round or two until there's more relative [16:58] certainty because one thing you don't [17:00] want to do is be in the number two or [17:01] number three like you said, you have to [17:02] be in the category winner. So you rather [17:04] wait for that round and actually double [17:06] down and lead the A or the B. Mhm. [17:08] >> So the small firms, they can carve out a [17:11] niche for themselves a clip or two [17:13] earlier than the big firms and actually [17:15] have a right to win. They can do really [17:16] well and be complimentary to the big [17:18] firms. [17:18] >> Yeah. [17:19] >> Do you agree with that or like Yeah. [17:20] Yeah. Yeah. Yeah. And look, we have like [17:22] very healthy relationships with with [17:23] seed funds across the ecosystem. Um we [17:26] also do do seed ourselves, right? But [17:28] preede for sure, you know, sort of [17:30] earlier than we typically. [17:30] >> The seed you would do I think is like [17:32] chunkier bigger seed, right? Like a [17:34] serial entrepreneur. [17:35] >> They used to be a Yeah. [17:37] >> Yeah. that that definitely is our that [17:39] said, you know, we have our speedrun [17:40] program which we just came from actually [17:42] earlier this uh this morning and I don't [17:44] know I think [17:45] >> I think the market is evolving where [17:47] we're where um because founders have [17:50] such a preferential attachment as as DG [17:53] was saying to the brands we get to look [17:55] at everything and sometimes it does make [17:57] sense for us to do the preede and seed [17:59] and so you know you kind of want that [18:01] flex of capability and for the founder [18:03] to your point they kind of don't really [18:04] care where your focus is they just want [18:07] to be in that orbit and like you'll find [18:10] the funds to kind of match to it and so [18:12] I think we can coexist in this world but [18:14] I I think it's also very um important [18:17] like our business is principally an [18:19] early stage business we have to be first [18:20] to the poll and we may not actually do [18:23] the investment but we have to at least [18:24] understand the landscape and market to [18:26] be able to actually make the informed [18:27] decisions later on and so [18:28] >> I spoke to a few founders at speedrun [18:30] today and they very much are hoping to [18:32] stay in the orbit [18:32] >> yeah long term right and so this is you [18:35] know a new phenomena that were 10 years [18:37] ago again this is starting to really [18:40] take shape as more of the early stage [18:42] folks started doing later stage and then [18:44] extending process stack but not quite in [18:46] the way that it is today and DJ I don't [18:48] know if you you would agree with that [18:49] like it's just virtually impossible to [18:51] have this sort of mid-stage business [18:53] effectively without having the early [18:55] stage and then also the late stage and [18:56] come behind it as well. [18:58] >> Yeah, I mean look I'm very biased uh but [19:00] I think the reason that we've been [19:01] successful as a growth fund is because [19:02] of our early stage business. I said that [19:04] all the time like our business starts [19:05] and ends with early stage and so um you [19:08] know that provides us a tremendous [19:09] amount of advantages at the growth stage [19:12] um in terms of access information, [19:14] knowledge um relationships etc. Um and I [19:18] would think that you know our early [19:20] stage partners would probably say that [19:21] the growth business provides them [19:22] benefits too because it allows you know [19:24] us to scale up and deepen partnerships [19:26] with founders over time and that helps [19:27] to win deals at the at the early stage. [19:29] >> Yeah. Totally. Totally. And you can't [19:31] there's both sides. You can't only do [19:33] the early, you can't also only wait till [19:35] the late too. And so your point earlier [19:36] around it looks like increasingly that [19:39] firms are converging to a handful of [19:41] names. Like that's probably true because [19:43] you know again this power law dynamic [19:45] but also at the same time almost [19:47] majority of those logos so to speak we [19:49] have to get at the early because that's [19:50] the only way we maintain the ball [19:52] control and also participate in the [19:53] prata and then some. And so that's [19:55] obviously the the business that [19:56] >> we're big believers that the strongest [19:59] latestage franchises have a huge early [20:02] stage franchise attached to them. Like [20:03] the ability to win is multiplied when [20:05] you have an early stage franchise. Yeah. [20:07] Like it's [20:08] >> um and we talked about sizing in late [20:10] stage where you can whatever the size of [20:12] your late stage fund is, if you can at [20:14] scale put 5 to 10% of your fund in one [20:16] of the category defining companies. The [20:18] way you could do that is because you had [20:19] an early stage franchise that developed [20:21] that relation with the entrepreneur and [20:22] the management team early on. It's [20:24] really hard to come in as a denovo late [20:26] stage firm and write a $500 million [20:27] check. [20:28] >> Yeah. Yeah. Yeah. No, it's very hard. [20:29] Yeah. Uh Ive I've I've lived that world. [20:32] >> How do you think about from the LPC how [20:35] venture is fundamentally [20:38] perhaps a structurally different job [20:40] than maybe when you started your career [20:42] also as well? And you know how do you [20:43] think about also asset allocation within [20:45] venture because there's actually [20:46] subasses within venture as you think [20:48] about portfolio construction as well. [20:50] Yeah, I mean there's I mean in a very [20:51] simplistic way there's in our mind [20:53] there's four ways to do venture preceded [20:56] seed so think sub 150 funds there's [20:59] thousand thousand close to 2,000 today [21:01] in the US alone [21:02] >> messy middle we talked about and then [21:05] there's a lot of firms there by the way [21:07] thousands of not thousands hundreds of [21:08] firms and then the big firms [21:09] >> and then dedicated late stage so there's [21:11] four ways to play it [21:13] >> um we have done the larger firms [21:16] >> for decades now we've done the seed [21:18] firms we've select effectively done a [21:20] few in the messy middle [21:22] >> and we haven't done dedicated late stage [21:24] for the reasons we talked about. [21:26] >> Um how is it changing? AI is actually [21:30] making our jobs harder than ever before. [21:32] >> Um it's making it harder because rounds [21:35] are larger [21:36] >> in general. They're faster. [21:40] The traction that's happening in the [21:41] industry is confusing. And here's why [21:44] it's confusing. You can have a company [21:45] I'm actually really curious to hear this [21:47] from you because we hear this a lot. [21:49] company comes out of Pick Your [21:50] Accelerator. I went from zero to five [21:51] million ARR in a month. [21:53] >> Yeah. [21:53] >> There's no renewal cycle yet on that [21:55] company. [21:55] >> Yeah. [21:56] >> And they're raising off of that traction [21:58] at huge multiples. And a lot of times [22:00] they're selling to each other in a [22:01] cohort potentially. And it's not even [22:04] ARR, but they're multiplying by 12. So, [22:06] but for every C for nine companies like [22:09] that, there's one really special one [22:11] that's doing a couple of million in ARR [22:13] actual ARR that has a huge valuation [22:15] that will go on to be the next cursor. [22:17] >> Yeah. So it is really really tough [22:19] actually today to parse out like what's [22:22] real attraction what's not. Valuations [22:24] are really high. This is why the big [22:26] firms do well. I actually think they can [22:28] wait or they have enough relative [22:29] certainty in the next round then lead [22:31] that round. But even then there isn't a [22:32] lot of certainty. When you guys did [22:33] cursor I don't think there was a lot of [22:36] certainty. And how many for how many [22:38] months were people saying cursor is [22:39] dead? [22:39] >> Oh even the morning of the acquisition [22:41] announcement people were still saying [22:43] that cursor is dead. I'm like, they just [22:45] announced that they were going to be [22:46] acquired by SpaceX for 60. [22:48] >> Tell me if I'm wrong. 3 million AR, $400 [22:49] million round or somewhere maybe around [22:51] there. [22:51] >> Yeah. Something like that. Yeah. [22:52] >> Like a lot of people would say like [22:54] that's crazy. Why did they do that deal? [22:56] >> Yeah. Yeah. Yeah. Yeah. Well, look. [22:57] Okay. So, like founder judgment is a [23:00] very important thing, right? And so, you [23:02] know, getting to know founders over [23:04] time, spending a lot of time with them, [23:06] seeing how they think, like I'd like to [23:08] think that, you know, especially my [23:09] early stage partners are pretty good at [23:10] that. Um, secondly, like it is hard to [23:14] parse out real verse, you know, kind of [23:16] misleading attraction. And misleading in [23:18] the sense that like you can't take [23:19] market signal from it. Not that [23:20] anybody's doing any misleading. Um, [23:23] >> there's probably some of that too. [23:24] >> Yeah, there's probably some of that too, [23:25] but like I go back to [23:26] >> I love every now and then people are [23:28] like helping to redefine what AR [23:30] actually means. I'm like this is a very [23:31] helpful PSA for the [23:33] >> This is always helpful. Yes. Uh, yeah. [23:35] But like I don't know, come back to, you [23:37] know, is the market demanding more of [23:38] your product? Like that is always the [23:40] question. That's post a note on my on my [23:42] computer screen. Like [23:43] >> how do you know that when the company's [23:44] been only operating or selling for a [23:46] couple of months? [23:46] >> You're not going to be able to do it [23:47] with financial analysis. You'll have to [23:49] do it by really understanding the [23:51] customers, talking to the customers. And [23:52] then, you know, it's like one of the [23:54] things that I said about Harvey overtime [23:55] as an example, right? Um they did a [23:58] really good job commercially early days [24:00] like cuz they were they were smart. They [24:02] were re there was like a research plus [24:03] lawyer combo. um you know they got some [24:06] momentum and some and some you know [24:08] high-profile law firms to sign up early [24:10] days um but you know the usage was not [24:13] very good right and so like if you [24:15] looked at the actual deployment of it [24:18] just it looked like mediocre compared to [24:20] some other software firms um you know [24:22] and AI [24:22] >> from a retention standpoint [24:24] >> not retention no [24:26] usage of it um [24:28] >> now fast forward post reasoning models [24:31] like that totally flipped [24:32] >> and uh you could see you absolute [24:35] takeoff of adoption, right? And so a [24:37] bunch of different things happened at [24:38] the same time. Lawyers got way more [24:40] value out of the product. You could see [24:42] it in usage and engagement. Um, and then [24:47] it became because it was like high [24:49] utility usage and engagement. It almost [24:52] became a flip from what was previously [24:54] like, oh, we're scared of things like [24:56] hallucinations to, no, no, no, every [24:58] client is actually demanding the law [25:00] firms use the product. Um, and so, you [25:03] know, I think we look for markets like [25:04] that. We try to catch them early, like [25:06] we try to catch them earlier than when [25:07] than we did at Harvey. Um, but, you [25:09] know, that's the kind of signal that we [25:10] look for. You got to go you got to go [25:12] lay it down. Like, everyone can do [25:14] cohort analysis. Everyone can look at [25:15] renewal data. Um, but like understanding [25:18] the texture of the market and what the [25:19] customers actually want and need in [25:20] their alternatives. I think, you know, [25:23] that's how you make the decision. That's [25:24] why it's so important to have the early [25:26] stage business. Um, because they're the [25:28] deepest in the in the technology, in the [25:29] products. Um, and they, you know, they [25:31] obviously saw it in cursor and they've [25:32] seen it in many other things. [25:34] >> Jen, what's the biggest push back? I [25:36] mean, you're the most prolific [25:36] fundraiser I know. So, what's the push [25:38] back you get from [25:39] >> You could be very gainfully employed at [25:40] this firm. [25:41] >> I I agree that you are a prolific [25:42] fundraiser. [25:43] >> Um, what's the biggest push back you're [25:45] getting from LPs on like the state of [25:47] AI, the state of venture? [25:49] >> So, a lot of it is is worries around um, [25:52] you know, are we catching a falling [25:54] knife here? Just like the timing of the [25:56] market, where we are, like are things [25:57] overheated, etc. And so we talked a lot [25:59] about this at the the outset, you know, [26:01] around valuations and what the potential [26:03] of the market is, but I do get a lot of [26:07] sentiment from the LPs that their job is [26:09] also about to fundamentally change as [26:11] well. And and you mentioned earlier one [26:12] aspect of it around AI making it more [26:14] challenging to evaluate opportunities [26:16] and funds, but the other aspect of it as [26:19] well is the the LP historically has not [26:23] been incentivized to actually embrace [26:26] change in some respects, right? you [26:28] know, this is very much a job where um [26:32] the end goal is actually somewhat [26:35] diametrically opposed with the risk [26:37] tolerance of the GP and and this is just [26:39] the the mechanics of the industry, but I [26:41] often times say, you know, a GP can get [26:44] fired for missing out, you know, the [26:47] next, you know, Facebook, the next Uber, [26:49] right? Like that is the error of [26:50] omission and like that is fireable. [26:52] >> But LPs on the flip side only get fired [26:54] if you invest into Amanda. So in some [26:56] respects like the incentive outcomes are [26:59] actually completely opposite of the [27:01] >> you don't get fired for investing in IBM [27:02] if you're an LP. [27:03] >> Exactly. Yeah. And in fact like you [27:05] don't potentially even get fired for not [27:07] investing at all. [27:08] >> Yeah. [27:09] >> And so [27:09] >> if you miss if you miss the frontier [27:10] models back to your first question as an [27:12] LP but you kind of are along the [27:14] benchmark. Right. [27:15] >> Maybe slightly below the benchmark [27:16] you're keeping your job. [27:17] >> Right. Right. And that's fascinating. [27:19] >> And and also for most folks and I'll [27:22] leave fund of funds out of the the [27:23] equation because it's a different [27:24] different piece, but but you know for a [27:26] lot of folks the upside actually is not [27:29] that interesting for them. So the pitch [27:31] of like hey you're going to miss out on [27:32] the next potential generation incentive [27:34] misalignment. [27:34] >> It's it's actually quite um direct. So [27:37] so where do we go from there in terms of [27:39] the LP kind of role and see like it I [27:42] think we also have an important role and [27:44] function. And we oftent times talk about [27:46] in the context of our job as a leader of [27:48] the venture capital industry is we have [27:50] to help folks understand where the [27:51] future is going. And part of that is [27:54] understanding how to infiltrate not just [27:56] within their venture capital allocation [27:58] but across their entire portfolio. And [27:59] that I think is way more interesting [28:01] than just saying hey like you might miss [28:02] out on this next generation returns or [28:04] the optimization of like the next [28:06] frontier model or you know one or two [28:08] power law companies etc. [28:09] >> Access selection sizing is what LPs do. [28:12] So access [28:14] >> you could argue you have the data to [28:15] figure out who has done well [28:16] historically [28:19] >> out of those 20 firms out of 3,000 like [28:22] you are not going to see consistency [28:24] right maybe half of them are consistent [28:26] >> but the LP's job is also to find the [28:28] nextg firms as well as continue [28:31] accessing that so one is you access two [28:33] is selection three is portfolio [28:34] construction sizing and it's critical [28:36] from an LP standpoint [28:38] >> because if you have an asset class where [28:39] 20 firms out of 3,000 do well you should [28:42] concentrate in those 15 20 firms pretty [28:44] consistently. So when I see a portfolio [28:46] with 50 60 70 venture capital firms, [28:50] it's very hard for me to imagine that [28:52] the overall portfolio can generate [28:54] better than the average. [28:55] >> And again, I'm going back to the average [28:57] in venture. That's just not what one [28:59] >> not compelling enough for the liquidity [29:00] >> relative to any of the other asset [29:02] classes, public markets, private equity. [29:03] I mean, private equity can probably get [29:04] you like one and a half to 2x net [29:06] without the lock up, without the risk [29:07] you're taking in. You talked about 60% [29:09] loss ratio. P doesn't have that, [29:10] >> right? Uh P has other problems we can [29:12] talk about today when it comes to AI [29:14] software, but um so portfolio [29:17] construction sizing for an LP is [29:19] critical. Like I've seen too many times [29:22] >> an LP or an allocator find an [29:25] interesting fund, actually get it right [29:27] >> and put 1% of their fund into it. [29:30] >> Great. You 10xed it. It returns 10% of [29:32] your fund. It does not move the needle [29:34] at all. [29:34] >> Yeah. Yeah. Yeah. Yeah. Where do you all [29:36] think we are today in that evolution [29:39] that Jen was talking about of how you [29:42] would I know you're biased, but the [29:43] appropriate [29:46] percentage of overall capital allocated [29:49] to venture and growth compared to [29:51] private equity or public markets or real [29:53] assets, credit, whatever it [29:55] >> Yeah, this is the this is hard for me to [29:56] answer because all I do is venture [29:59] growth. So, I would be biased to say [30:00] it's should be superersized. Um, [30:02] >> we like that answer though. [30:05] Look, look at the public markets today. [30:08] We vibe coded like something that [30:10] created a great way for us to assess AI [30:12] resiliency of public companies and now [30:14] we're doing that on the private side and [30:15] it's helping us hugely in in our growth [30:17] equity portfolio. But the in the SAS [30:21] public markets there only like 15 to 20 [30:24] companies max trading above 10 times [30:26] revenues which is by an insane number [30:27] because it used to be dozens and dozens [30:29] a few years ago. And every one of those [30:30] companies for the most part is showing [30:32] acceleration of growth from AI. you're [30:34] either in monitoring, security, [30:35] deployment of agents, etc. So, it goes [30:38] back to the same principle where if you [30:40] are in some shape or form tied to AI, [30:42] which is the fastest growing facet of [30:44] all the elements of the GDP, then you [30:46] should superersize it in your portfolio. [30:48] That will have impacts in the public [30:50] markets. Even private equity today, when [30:52] they're doing a new investment, they're [30:54] looking for something that's AI native. [30:55] They're not looking to buy a workflow [30:57] software company growing 10% that's [30:59] seedbased. Like, that's just not [31:01] happening. They're looking for the [31:02] system of record. that can show [31:04] acceleration with an AI native [31:06] management team. [31:06] >> Yeah. [31:07] >> So that connective tissue of AI is [31:10] actually across every asset class today. [31:13] >> Yeah. [31:13] >> And the one of the strongest way to play [31:15] it is I mean it's it's probably through [31:17] venture. [31:18] >> Yeah. [31:18] >> But that's where selection access and [31:20] portfolio construction is really [31:21] critical. [31:22] >> Yeah. Well, even the exits though we [31:24] were talking about um you know Silverl [31:26] potentially you know buying workday for [31:27] example just like doing more provocative [31:29] things for example in private equity [31:31] land when you have the capabilities to [31:34] potentially infuse and and bring them [31:36] into into the future as a part of that. [31:37] The other version of it is also exits in [31:40] venture now way exceed private equity. [31:43] Like I you know like I was looking this [31:45] up last night. Private equity this year. [31:47] The biggest exits are um the buyout of [31:50] EA which was like around 50 billion and [31:52] Medline which is around 50 billion. [31:53] Curser let's not let's exclude the IPOs. [31:56] Like cursor was the M&A sale to SpaceX [31:59] was way bigger than that. And so the pro [32:01] the problem with that is like [32:03] >> you look at the pre-hat GPT vintages and [32:05] private equity [32:06] >> you would have paid I don't know 15 to [32:09] 20 times EBIDA for an software asset [32:11] that's growing 10 20% max [32:14] >> y [32:15] >> if you look at the public markets today [32:16] that asset is trading at two times [32:18] revenue [32:18] >> y [32:19] >> and the problem is not that just the [32:21] valuation there might not be a buyer for [32:23] that company because [32:25] >> if you're looking at a software company [32:26] today the first thing you think about [32:27] what is the terminal value is it [32:29] resilient from AI. The best way to show [32:32] that is organic growth acceleration. Our [32:34] data shows 1 percentage of growth in the [32:36] public markets is equivalent to three [32:38] percentages of IBIDA. [32:39] >> Yep. [32:40] >> So, by the way, it's funny because in co [32:41] everyone was like we need to be [32:42] profitable. It was the inverse and now [32:44] it's the opposite. [32:44] >> No, no. 21 it was the inverse. Post it's [32:47] basically like fully uh aligned with [32:49] risk, right? Correlated with like risk [32:51] in the public. [32:52] >> Exactly. Um so, unfortunately, a lot of [32:54] those private equity deals, they're not [32:55] growing fast enough. they're not showing [32:57] that acceleration and they may not have [32:59] the management teams to revamp the [33:01] business like what intercom did is a [33:03] great example. Bring the founder bank uh [33:06] back, revamp the whole business, create [33:08] an AI native product, scale it, and then [33:10] sell. Like it's almost like you're [33:12] suiciding your existing business, which [33:14] in private equity is really hard to do. [33:16] >> It's really hard to do. I just spent a [33:17] little time with the founder uh and uh I [33:19] I went up to him at an event and I was [33:21] like I just gave him a big high five and [33:22] I'm like, "You did it, man. [33:24] >> You did. This is the this is the thing [33:25] that like is really really hard to do. [33:27] You know, it's an end of one right now. [33:29] >> Um but you know, there's a bunch of [33:31] really good founders who are capable [33:33] with those businesses, public markets, [33:34] private markets, [33:35] >> uh who I think are are going to take a [33:37] crack at it. Um so we'll see. [33:39] >> Yeah. May maybe on that thread though, [33:41] DJ, because we we also sometimes get the [33:43] push back as well like for folks who [33:44] have been in venture and allocated to [33:46] venture, they might also have a similar [33:48] problem where they do have the legacy [33:50] SAS businesses also as well. like what's [33:52] the balance between how you think about [33:54] the historical stuff? [33:55] >> Let me ask you that question. So, I'm [33:56] going to piggyback off of Jim's question [33:58] to you. You got a pick a 2016 through [34:02] 2021 pre-Cat GPT vintage software [34:04] company that was fine but doesn't have [34:07] that AI native features anymore. It's [34:09] not accelerating. It's growing like 30%. [34:11] On the venture books, it's at like 10 20 [34:13] times revenue. [34:16] It's not can go public anymore. Like no [34:17] one cares to take that public. [34:19] >> Yeah. Silver Lake has no interest in [34:21] that company anymore. They would have a [34:22] year ago. They don't. [34:23] >> Yeah. [34:23] >> What happens to that company? [34:26] >> Um, you know, look, it's like [34:28] >> we have a lot of exposure to those [34:29] companies, too. So, [34:30] >> I would say it's like it's it's very [34:32] TVD, right? Like I, you know, I was with [34:34] um one of our CEO founders this weekend [34:36] and, you know, he was like, [34:38] >> you know, give me the straight scoop. [34:39] Like, what do you actually think is [34:40] happening? And, um, you know, not don't [34:42] give me a he actually said to me, don't [34:43] give me a podcast answer. Um, which is [34:46] ironic. Um and you know I think both [34:49] things can be true that AI is like the [34:52] biggest generational change that we've [34:53] ever seen and it's going to transform [34:55] industries and also there will be some [34:58] enduring value of software companies [35:00] that are able to adapt right um part of [35:03] the thing that we're monitoring which [35:05] makes us extremely bullish about AI is [35:07] just actual diffusion into the real [35:09] economy right so coding I think if if [35:13] you were to paint the the bullish [35:14] scenario for regular software and for [35:16] slower pace of change, you would say um [35:20] you know coding hit but that's kind of a [35:22] head fake, right? Like coding coding is [35:25] perfectly documented, right? So it has [35:26] like perfect data uh it's verifiable and [35:29] it's simulatable, right? And so um like [35:32] most tasks in business do not share [35:34] those three attributes and so you know [35:37] maybe the diffusion into other knowledge [35:39] work beyond you know coding will take a [35:41] lot longer. That would be the case to [35:43] make for the software companies. Um, and [35:45] then some of them will evolve and and [35:47] have AI, you know, solutions and they'll [35:50] change their business models. Um, and I [35:52] think that's a must. Um, but that would [35:54] be the case for why maybe, you know, [35:56] it's a little bit overblown. I think if [35:58] you look at the way that a lot of the [35:59] public SAS companies have reacted over [36:01] the last few months, like that's I think [36:02] there's a little bit of a growing [36:03] realization in that. Um, all that makes [36:06] me super super super bullish on AI [36:08] though, right? So like if you look at [36:10] our portfolio um you know we have some [36:12] of those companies but about 95% of our [36:15] NAV is not in those companies right like [36:17] right there it's it's in the it's in the [36:18] companies that are you know growing very [36:20] fast accelerating etc. Um you know the [36:23] average I think it was the median [36:26] company in the US is spending $12 per [36:30] employee on AI per month. The top 1% of [36:35] the data set that we've seen is spending [36:37] $7,000 [36:38] >> Wow. per employee on AI per month. So um [36:42] not only have we had like limited [36:44] diffusion beyond coding, but if you just [36:46] look at diffusion of, you know, the [36:49] shape of who is consuming tokens and [36:51] actually getting real value out of AI [36:52] today, we're super early, right? Like [36:54] banks, you know, the most cutting edge [36:56] banks are probably doing 1% of headcount [36:58] cost on on AI tools. Um and so the [37:02] reason this makes me very bullish is [37:04] these are the fastest growing companies [37:05] we've ever seen, like of all time. [37:07] Again, they're adding more revenue per [37:08] month than the than the mega cap tech [37:10] companies. Um, and yet it's probably on [37:12] the back of adoption of like 10 million [37:15] users, maybe 20, maybe 30 max. And you [37:19] know, there's one and a half billion [37:20] dollar workers in the US. And I think [37:21] it's going to transform the way we do a [37:23] lot of work. [37:24] >> Yeah, Kalpers's famously uh lost out on [37:28] billions of gains by not investing in [37:29] their back. They're making up for loss [37:31] time now. they they've uh converted [37:32] their portfolio from 91% to 58 and [37:35] venture and growth from 9 to 43. There's [37:38] probably some balance in between those [37:39] things. Uh but you know, they're they're [37:41] leaning hard into um but there's going [37:43] to be a lot of value still that's going [37:44] to be accreted in some of these [37:45] historical companies. And you I was sort [37:47] of joking around about bending. That was [37:48] probably a great outcome for Air Table. [37:50] You know, outside of the fact that, you [37:51] know, they're going to actually [37:53] >> spin off the hyper agent piece of the [37:54] business and actually um I think do [37:56] really interesting things with that. But [37:57] in the scheme of things, I think there's [37:58] going to be a lot of homes for a lot of [38:00] things. And this zero sum thinking I [38:01] think is is probably the pitfall of [38:03] which we would would advise against. [38:05] >> So we talked about um we've talked about [38:07] you know venture growth, we've talked [38:08] about private equity, you know we've [38:10] talked about um public markets. Um and [38:12] by the way the composition of all of [38:13] those have like radically changed over [38:15] the last 10 years. Um we also have had [38:18] the emergence of entirely new categories [38:21] that are available in the private [38:23] markets like private credit. Um do you [38:26] have a view on sort of outlook of those [38:28] on a relative basis [38:30] >> in software in particular? [38:32] >> Yeah, I'd say in technology [38:35] >> uh the vantage point we have is looking [38:37] at private credit which resides in a lot [38:40] of private equity software portfolios [38:41] which is hundreds of billions. [38:43] >> Mhm. [38:44] >> I'll give you one statistic. So you look [38:46] at 2122 [38:48] about 2 to 300 billion in LBO software [38:51] transactions happened with over 200 [38:53] billion in debt taken out. The average [38:56] valuation for those software deals over [38:57] 25 to 32 times IBIDA. Those companies [39:00] today are worth probably half that. The [39:03] reason you're seeing redemptions in the [39:04] credit markets in private credit is [39:05] exactly that. They're looking at the [39:07] public markets. You've had CS [39:08] apocalypse. It's been a massive [39:10] correction in software and the val and [39:12] you can see a contraction in valuations [39:14] which means the leverage ratios have [39:16] gone up dramatically. So [39:19] if you are a software company that is in [39:22] somewhat not resilient to AI I think [39:24] you're challenged both in terms of your [39:27] equity position also credit as well. [39:30] >> By the way even the um AI version of [39:35] private equity is not completely [39:37] insulated. We oftentimes talk about like [39:40] you know just because you put Sears on a [39:42] website didn't make it Amazon right you [39:44] have to have the benefit of building [39:45] Amazon from the studs logistically to [39:48] make it Amazon. It's not just the [39:50] website and in a lot of instances with [39:51] the private equity backed companies that [39:54] are now just infusing AI. We've seen it [39:56] actually in some of our companies as [39:58] well where the peer competitor is like [40:00] oh the first thing I'll do is of course [40:02] hire AI customer service agents because [40:04] that's like an easy lowhanging fruit. [40:06] Like turns out if you don't actually [40:08] build in the workflow, you start to turn [40:10] customers very quickly if they're used [40:11] to talking to a human and for every [40:13] dollar uh every drop in NPS is like a [40:16] direct correlation with drop in revenue [40:18] and then you start to spiral especially [40:20] if you have debt ladened on top of it. [40:21] So oftentimes uh sometimes we hear from [40:24] folks like well I'll just do the AI you [40:26] know kind of version of private equity. [40:28] It's not a pan panacea for for [40:31] generating returns especially when it's [40:33] just so categorically different from a [40:34] technological perspective to actually [40:36] infuse that throughout the company as [40:38] well. [40:38] >> Yeah, you can't just throw an operating [40:40] partner at the company and say let's put [40:41] AI on it. It just doesn't work. [40:43] >> You need to completely and if you have [40:45] by the way if you do have a founder [40:46] mentality at the management team like it [40:48] is possible but the board has to be [40:50] aligned the all the investors have to be [40:52] aligned [40:53] >> and you do have to make some really hard [40:54] decisions the way intercom did. [40:56] >> Yeah. Yep. [40:57] >> Yeah. So obviously this is a group that [40:59] is very pro you know venture and growth [41:00] this category. Um let's talk about the [41:03] legitimate opposition to it and what is [41:05] the case you know for why maybe the you [41:08] know the risk that you're taking or [41:09] whatever it may be with venture and [41:10] growth you know doesn't justify it. [41:14] >> Uh I mean the push back we get a lot is [41:16] timeline to liquidity. So it takes the [41:19] average unicorn is private for 10 plus [41:21] years typically and then you got all [41:24] these follow on rounds that are [41:25] happening pretty quickly one after the [41:27] other. You see maybe the same logo in [41:30] five six different firms [41:32] >> and the question is how do you get out [41:33] of it [41:34] >> um and so [41:35] >> and an IPO isn't actually a [41:37] distribution. [41:38] >> It takes it could take 12 24 plus months [41:40] before you actually get liquidity out of [41:42] an IPO. like especially if you own 10 [41:44] 15% at IPO like it's going to take a [41:46] long time if you're in the generational [41:47] company. So we get that push back a lot [41:50] in terms of timeline to liquidity. [41:51] >> Mhm. And then um what is the what is the [41:55] sort of counter to that push back? [41:58] >> Well counter to that push back is going [42:00] back to 3,000 firms 20 do well [42:04] consistently. If you're in the top 1% of [42:05] those firms and you have a category [42:07] winner you want to make sure that [42:08] compounds actually. [42:10] >> Yeah. Um would you have wanted to sell [42:12] stripe data bricks or any of those other [42:14] companies three four years ago? Like the [42:16] answer is unanimously no. Uh now could [42:19] some of these companies go public [42:20] earlier than not? Sure. Entropic was [42:22] really first funded in 2021. It's about [42:24] to go public 5 years later. [42:25] >> Yeah. [42:26] >> Cursor from first acquisition from first [42:29] financing to acquisition is short. Like [42:32] >> so the best venture firms actually have [42:34] fund returning liquidity pretty quickly [42:37] maybe even quicker than private equity. [42:38] But that subset of firms is tiny. [42:40] >> Yeah. [42:40] >> Yeah. Yep. Actually, very famously uh a [42:43] year and a half ago or so, we went to [42:45] our fund one LPs. At that point in time, [42:47] the fund one was 16 years old and we had [42:49] this position in Stripe that we invested [42:51] at the seed stage and we asked all of [42:52] our LPs like, "Hey, do you want [42:54] liquidity out on this?" We recognized, [42:56] you know, the job that we came to do is [42:58] now done 16 years in. Like, do you want [43:00] liquidity back on this? And every single [43:02] one of those LPs said, "No, we'd rather [43:04] let this continue to compound." And then [43:06] ultimately a year later, you know, we [43:07] decided to to make that exit because [43:09] it's 17 years in. We've got to get this [43:10] this liquidity out. We got to wrap up [43:12] the fund, etc. But, you know, so many [43:15] LPS, I think, is very specific to [43:18] certain categories, right? Endowments [43:20] would prefer to let it run. Family [43:21] offices, quite frankly, don't want the [43:23] money back because they don't want to [43:23] pay taxes on it. They'd rather have it [43:25] continue to compound. And so there is [43:27] specific nuance with each LP group where [43:29] it's very hard to paint a broad brush [43:31] stroke stroke on like across the board [43:34] on everyone wanting the same thing. But [43:36] I also think to your point [43:37] >> the very best LPS, excuse me, the very [43:40] best GPS have manufactured along the way [43:43] liquidity and particularly in 2021 when [43:45] a lot of folks didn't take money off the [43:46] table. You know, I think that was a good [43:48] sign of the first indicator and now in [43:50] this next cycle, it's can you actually [43:52] get some early liquidity out through M&A [43:55] and then let you know, potentially the [43:56] winners IPO over time with the fullness [43:58] of of compounding as well. [44:00] >> Yeah, exactly. [44:01] >> I'm going to end on this note because I [44:02] I thought this is an interesting [44:03] question that you and Gavin were were uh [44:05] tossing back and forth DG on. Um he he [44:08] uh didn't want to answer the question on [44:09] what will be the next $10 trillion [44:11] company, but he had certainty around [44:13] what will be the next 20 trillion market [44:16] cap company. So, I'm going to ask both [44:17] of you, uh, what do you think is going [44:19] to be the next, uh, hundred trillion [44:22] dollar market cap? [44:24] >> Oh my gosh, here we go. I can't even [44:26] think in those things. [44:27] >> Yeah, we're, uh, that's probably two two [44:29] tech cycles away, not just one. Uh, it [44:32] is possible that we have entirely new [44:36] companies that get created. And I think [44:38] a lot of the market cap creation that [44:39] you would talk about that would drive a [44:41] 10 trillion outcome or more um, is in [44:44] new product areas that haven't yet been [44:45] touched, right? So [44:46] >> yeah, [44:47] >> like what I talked about the sort of [44:50] diffusion of the technology into the [44:52] enterprise like we're nowhere, right? [44:54] >> Um, [44:55] >> you know, a year ago everyone talked [44:56] about consumer all the time. No one even [44:57] talks about consumer AI anymore. But [44:59] that is going to like the end use case [45:01] for consumers is not going to be a [45:04] chatbot interface. Like that's the [45:05] skumorphic version. We're going to have [45:07] a native version. It's going to be [45:08] proactive. It's going to do work on our [45:09] behalf. It's going to it's going to [45:11] create a ton of value for for consumers. [45:13] And we're kind of nowhere on that. I [45:15] mean, yeah, there's, you know, a billion [45:16] chat users, but, you know, like we're [45:19] that's like scratching the surface. Um, [45:22] we are, um, nowhere on robotics, but I [45:25] think robotics is going to be bigger [45:26] than the language stuff. Um, and I think [45:28] it's going to happen in the next 10 [45:29] years. [45:30] >> Um, we are almost nowhere on autonomy, [45:33] right? Like there's fewer than 10,000 [45:34] Whimos uh, live in the US, way fewer [45:38] robo taxis. Um, and then, you know, a [45:40] lot of open space for others to build in [45:42] that area, too. Um, we, you know, [45:45] healthcare is 18% of GDP. Yeah. [45:47] >> Like we, we've, we've done nothing to [45:50] scratch the surface either on care [45:51] delivery or on drug discovery yet. I [45:54] mean, there's some companies that are [45:55] working on it, showing some early signs [45:56] of progress, but I think the progress [45:58] that we make there in the next 10 years [45:59] is going to be massive. Um, and then, [46:02] you know, we're in this interesting era [46:04] of reimagining, um, all things physical [46:07] world from defense to manufacturing to [46:10] data centers. Um, and so I look at the [46:13] confluence of all these trends and I'm [46:16] like, yeah, it may feel like we've, you [46:18] know, we've we've done a lot with AI [46:20] already. Um, but 10 years from now, [46:22] we're going to look back and say, oh my [46:24] gosh, like those other major areas [46:26] created a ton of value. [46:28] >> Um, and so I'm excited that I think the [46:30] next SpaceX AI or OpenAI are probably [46:33] going to get created. Um, and they'll [46:35] probably be in those kinds of domains. [46:36] >> Yeah. [46:36] >> Yeah. I'll add one category that to me [46:39] is both a concern but a huge [46:40] opportunity. It's a plug for your new [46:41] fund on the on the opportunities fund [46:43] that you did. I think very [46:46] simplistically about like the bottleneck [46:48] in AI today. It's not demand, it's on [46:49] the supply side. So you got energy, the [46:51] grid data center, then you got chips, [46:53] then you got frontier models and apps. [46:55] The US is amazing at the right side of [46:57] that. So like chips and onwards. The VC [46:59] ecosystem supports that well. Um I think [47:03] the new fund you have is really going to [47:04] help on the left side as well because [47:06] the US doesn't have a problem with [47:07] energy generation. It has a problem with [47:09] speed to power. [47:10] >> Yeah. [47:10] >> That's permissioning transmission. [47:12] That's regulatory. Other countries are [47:14] putting out 10x more renewable capacity [47:16] a year. [47:17] >> So that is a real bottleneck and that [47:20] means reimagining the data center. You [47:22] talked about the density being 10x plus. [47:23] Well, you can't just repurpose an old [47:25] data center for the for new AI facility. [47:27] So this is where the new fund you have [47:29] can create not 10 50 but 100 billion [47:32] plus opportunities as well that can [47:34] really solve the bottleneck and I think [47:35] that is a real concern because demand is [47:37] not the concern. Um, a lot of I've heard [47:40] LP say this is like the dotcom or this [47:42] is co. It's not because the traction is [47:44] real and it's not ephemeral revenue like [47:46] >> co [47:47] >> the bottleneck could be supply but if [47:48] you have the right inputs like the fund [47:50] that's you're now backing those [47:52] companies next generation chip companies [47:53] memory etc. That's a huge opportunity. [47:57] >> It's time for machine age. Let's bring [47:58] the machines. [47:59] >> I love it. [48:00] >> All right let's close on that. Thank you [48:01] both so much. It was super fun. Thank [48:02] you for having us. Awesome. See you.