OpenAI Enterprise Pivot Trails Anthropic Revenue Growth as Compute Scarcity Mounts

算力紧缺加剧之际 OpenAI企业级转型营收增速落后于Anthropic 在AI数据中心面临严峻电力与基础设施瓶颈的背景下，Anthropic在企业编程领域实现的年营收10倍增长，对OpenAI的市场估值构成威胁。 本期节目首先就纽约市拟议的豪宅税（pied-à-terre tax）展开辩论，随后深入分析了OpenAI与Anthropic之间的AI市场动态，重点探讨了企业编程Token、算力紧缺以及数据中心建设瓶颈。主持人还审视了围绕Eric Swalwell离开国会的政治动向，并分析了宏观市场趋势、通胀指标以及伊朗地缘政治事件对投资者情绪的影响。在穿插市场点评与AI生产力讨论的同时，节目最后以轻松的游戏环节回顾了那些著名的估值过高初创公司。 纽约市拟对曼哈顿中城 15 英里内超过 500 万美元的非主要住宅征收 3.9% 的豪宅税（pied-à-terre tax），引发市场对高端房地产开发停滞和资产贬值的担忧；相比之下，德州奥斯汀通过放开土地分区并大规模增加住房供应，在人口激增的同时实现了连续三年租金下降。 (0:42) OpenAI 泄露的内部备忘录披露其战略重心正全面转向企业级 Agent 应用与 API，以应对 Anthropic 的激烈竞争；后者凭借企业编程 token 计量计费模式，ARR 预计将从 10 亿美元爆发式增长至 800 亿至 1000 亿美元，增速远超依赖每月 20 美元固定订阅模式的 OpenAI。 (11:23) 云服务超大规模厂商（Hyperscalers）掌控了全球约 60% 的 AI 算力，使独立前沿模型实验室面临算力受限和运营瓶颈的风险；同时，电网容量短缺、土地审批延迟以及日益高涨的民意反对（NIMBYism），正倒逼数据中心运营商转向自建能源（BYOE）微电网模式。 (26:42) 全美目前约有 100 个、总价值约 1620 亿美元的数据中心项目面临争议或阻力，被否决项目的取消率翻倍至 40%；同时 AI 行业的公众好评度因持续聚焦存在性风险与失业问题而持续下滑，甚至低于部分政府机构。 (49:48)

The All-In Podcast hosts and guest Travis Kalanick argue that frontier artificial intelligence labs face an impending collision between exponential token-driven revenue growth and hard physical infrastructure limits. At the same time, peak macroeconomic valuation multiples and corporate change management barriers threaten to temper broader market optimism. Frontier AI Rivalry and Enterprise Monetization (11:23) OpenAI and Anthropic are locked in an escalating enterprise conflict, highlighted by a leaked memo from OpenAI Chief Revenue Officer Denise Dresser targeting Anthropic's reported $30 billion annual recurring revenue (ARR) run rate. While OpenAI expanded rapidly with ChatGPT's estimated one billion consumers, David Sacks noted that flat $20-per-month consumer tiers convert only 3% to 4% of users. Conversely, Anthropic has expanded at an estimated 10x annual clip—scaling from $1 billion to $30 billion ARR and targeting up to $100 billion by year-end—by prioritizing metered enterprise coding tokens, which businesses consume continuously as a utility. Valuation dynamics reflect these diverging growth trajectories. Travis Kalanick argued that scaling speed dictates investor sentiment, cautioning that venture subsidies cannot outstrip positive contribution margins over the long run. Despite OpenAI closing a record $122 billion funding round, secondary market buyers have reportedly stalled at its $850 billion valuation, requiring a projected $1.2 trillion public debut to justify existing math. Meanwhile, enterprise developers are rapidly switching providers; David Friedberg reported his organization shifted to 90% Anthropic reliance within six months due to its rapid model release cadence. Physical Grid Bottlenecks and Data Center Backlash (39:20) The hosts emphasized that physical infrastructure—specifically electrical grid access and data center construction—has superseded capital as the primary constraint on AI expansion. Cloud hyperscalers currently control 60% of global compute capacity, leaving independent labs vulnerable to server throttling that Chamath Palihapitiya likened to Friendster's historical capacity collapse. In response to severe compute constraints, Claude users reported a roughly two-thirds reduction in internal model thinking time prior to the Opus 4.7 update, while xAI is aggressively building out 555,000 GPUs across three facilities to ensure internal capacity. At the same time, regional pushback against server infrastructure is intensifying. Approximately 100 domestic data center projects valued at $162 billion are actively contested, with cancellation rates climbing to roughly 40%. The panel attributed this opposition to consumer fears of rising residential utility bills, state-level bans such as in Maine, and local political backlash against high-consumption tech facilities. Consequently, developers are turning toward 'bring your own energy' microgrids and on-site natural gas systems, such as those from Bloom Energy, to bypass utility grid queues entirely. Market Resilience Against Peak Valuation Metrics (70:34) Broader financial markets have reached new all-time highs despite the ongoing military intervention in Iran, which has reached its seventh week and accumulated over $100 billion in U.S. expenditures. Sacks and Kalanick observed that Wall Street has treated the conflict as a brief engagement nearing resolution, viewing public equity benchmarks as the governing metric for the Trump administration. However, Palihapitiya cautioned that standard historical valuation metrics, including the Shiller price-to-earnings ratio and the Buffett Indicator, sit near record highs while index gains remain narrow and concentrated in eight or nine mega-cap equities. Given the divergence between speculative valuations and defensive risk indicators, Palihapitiya advocated adopting a risk-off posture and warned that major private tech companies must tap the initial public offering window rapidly. He cited necessary liquidity events including SpaceX and a race between Anthropic and OpenAI to reach public markets. The hosts connected these elevated valuations to broader economic populism, noting that tax measures like New York City's proposed 3.9% annual pied-à-terre tax on secondary properties over $5 million risk depressing local development. Enterprise Implementation Realities and Agent Limits (77:37) The panel debated whether generative AI tools are driving genuine economic productivity or unoptimized waste. Jason Calacanis asserted that top-tier operators achieve 10x to 30x productivity multipliers, citing vertical enterprise applications like TaxGPT capturing 6% to 7% of practicing accountants. Conversely, Palihapitiya warned that enterprise chief financial officers will scale back growing token budgets if internal engineering audits reveal that models are merely generating unoptimized 'vibe-coded slop' without improving bottom-line profitability. Operational barriers within non-tech corporations remain the primary hurdle to enterprise integration. Kalanick identified human change management and complex, undocumented processes as the main bottlenecks slowing large-scale enterprise adoption. Furthermore, he noted that current autonomous agents remain fundamentally limited; personal experiments running automated trading agents on prediction markets like Polymarket showed that models lack autonomous judgment and will enter contradictory positions without continuous human supervision. While surging enterprise coding revenue provides tangible return on investment for frontier model developers, scaling remains caught between severe physical power bottlenecks and corporate implementation friction. The unresolved question is whether frontier labs can secure necessary energy and compute infrastructure before peak market valuations and capacity ceilings restrict further expansion.

All-In Podcast探讨了前沿人工智能领域的商业化分化与物理瓶颈，指出Anthropic在企业编程Token计费模式下的迅猛增速正对OpenAI的高估值构成直接挑战；与此同时，电力短缺、地方抵制以及宏观估值分歧正成为科技行业面临的核心制约。 模型竞争分化与商业模式之争 (11:23) OpenAI内部备忘录披露了其向企业代理平台转型的战略，并指责竞争对手Anthropic虚报营收。David Sacks指出，截至第二季度初，两家公司的年化经常性收入（ARR）均在300亿美元左右，但增速路径截然不同：OpenAI年增长3至4倍，主要依赖转化率仅3%至4%的20美元/月C端订阅；Anthropic则以10倍的年增速扩张，预计年底ARR可达800亿至100亿美元。这种差距主要由企业编程场景推动，企业将代码生成视为公用事业，按Token计量付费，形成了无限扩展的用量规模。 针对OpenAI完成的创纪录1220亿美元融资，Travis Kalanick指出资本固然能暂时换取规模，但效率终将战胜补贴，依靠高边际贡献率取得的有机收入才是长久立足的根本。投资者测算OpenAI需要达到1.2万亿美元的IPO估值才能支撑其最近一轮融资，但在8500亿美元的估值水平下，二级市场已难觅买家。David Friedberg亦补充称，得益于极快的产品迭代，其所在机构在六个月内将90%的底层使用迁移到了Anthropic。 算力瓶颈、电网制约与地方抵制 (39:20) 生成式AI的指数级扩张正遭遇现实物理条件的硬约束。Chamath Palihapitiya强调，云厂商巨头掌控着全球60%的AI算力，缺乏自有算力的模型开发商随时面临被云厂商限流的风险。与此同时，全美有约100个数据中心项目陷入争议，涉及总资本达1620亿美元，其中约40%的争议项目最终被取消，取消率同比翻倍。甚至缅因州等地区已出台立法全面禁止数据中心建设。 建设阻力源于电网过载担忧、AI安全组织的游说以及民众对科技巨头消耗公共资源的民粹主义反弹。传统公用事业公司受固定投资回报率（约10%）驱动，倾向于拉高资本开支而非降低成本，迫使运营商转向自带能源（BYOE）方案，借助天然气微电网和离网自发电绕过漫长的电网审批。在算力军备竞赛中，埃隆·马斯克正投资180亿美元将xAI的Colossus集群扩容至55.5万张GPU，Meta则规划了15万张GPU的Prometheus集群，基础设施建设正由单纯的资本战演变为电力与土地获取能力的博弈。 企业落地瓶颈与估值泡沫隐忧 (77:37) 对于AI驱动的生产力提升，嘉宾们观点存在分歧。Jason Calacanis认为AI带来了10至30倍的生产力爆发，并以税务工具TaxGPT在专业会计师群体中达成6%至7%的渗透率为例，证明前沿团队正拉开巨大竞争优势。然而Chamath提出质疑，认为边缘场景与小型初创公司的工具化应用无法支撑数万亿美元的市场估值，AI若不能在大企业核心业务中兑现巨额利润增长，便依然只是“玩具”。 Travis Kalanick从运营角度分析，传统大企业的变革阻力来自未文档化的复杂流程与庞大的科层管理体制；同时，当前的AI代理在自主决策与复杂逻辑上仍存在显著缺陷，缺乏人类监管极易陷入逻辑混乱。多位嘉宾还回顾了零利率政策（ZIRP）时期的泡沫案例，例如Allbirds因涉足AI数据中心概念导致股价单周暴涨400%以上，这与当年初创公司脱离毛利常识盲目扩张、最终走向破产或价值清零的教训如出一辙。 宏观指标背离、地缘风险与政治博弈 (0:42) 在宏观经济层面，尽管美军在伊朗军事行动已持续七周并耗资超1000亿美元，美股仍创下历史新高。Travis Kalanick指出投资者已将股市表现视作特朗普政策取向的风向标，认定其政策会在引发市场恐慌后迅速回调以维持美股高位。然而Chamath警告称，席勒市盈率与巴菲特指标（股市总市值与GDP之比）均触及历史高位，指数上涨仅靠八九家超大市值企业拉动，在极端市场分化下，投资者急需SpaceX或前沿AI企业的IPO来释放流动性并降低杠杆。 在地方政策与政治动态方面，纽约市拟对曼哈顿中城15英里内、价值超过500万美元的第二套房产征收3.9%的年度豪宅税，David Sacks警告这会摧毁价格敏感的高端需求并遏制城市开发；加州政坛方面，众议员Eric Swalwell因性丑闻突然退选加州州长并辞去国会席位，Sacks分析指出，这是加州开放式初选（Jungle Primary）选情告急下，民主党建制派为避免两名共和党候选人包揽决选资格而迅速采取的弃车保帅策略。 本期节目展现了硅谷核心圈对AI技术前景的深刻分歧：尽管前沿编程模型在企业端展现出指数级商业化变现潜力，但缺乏自主性能力的AI代理、传统企业艰难的流程重构，以及电力与土地审批的实体约束，正成为阻碍技术落地的实质阻力。在宏观估值指标逼近历史极值的背景下，市场能否通过关键龙头企业的IPO维持流动性，仍是悬而未决的挑战。

## Background 本期节目由知名播客频道 All-In Podcast 于 2026 年 4 月 17 日在 YouTube 平台发布。节目特邀 Uber 联合创始人特拉维斯·卡兰尼克（Travis Kalanick）做客，围绕纽约拟征收的高额豪宅临时居住税、OpenAI 泄露备忘录引发的企业级转型与估值竞争、AI 数据中心遭遇的电力与反建危机、加州众议员斯沃韦尔退选内幕，以及美股在消化伊朗和平预期下的破纪录行情等热点展开深入讨论。 ## TL;DR 四位核心主持人与卡兰尼克首先抨击了纽约拟开征的“次居所税”，指出惩罚性重税将摧毁高端地产需求并加速财富外逃。随后话题切入 AI 核心战局，深度探讨了 OpenAI 内部备忘录泄露后暴露的企业级战略摇摆、Anthropic 依托编程场景实现 10 倍年化增速及其二级市场估值反超，并警示全行业正迎头撞上地方环保抵制与电力短缺的物理天花板。在政治板块，大家剖析了民主党党内机器为保全选情而果断逼退斯沃韦尔的手腕与既视感；最后，嘉宾们辩论了美股在伊朗战事缓和预期下创下历史新高背后的估值割裂，并穿插了两轮竞猜泡沫初创公司的幽默游戏。 ## Atmosphere & style 全片充盈着典型的硅谷顶级投资圈内幕谈资与极度松弛的高密度思想碰撞，兼具极客深度与美式脱口秀式的犀利调侃。主持人们一边围绕私人飞机升级鄙视链、外卖骑手恶搞和综艺音效展开狂风暴雨般的互相拆台，一边在公用事业定价模型、地缘政治博弈及大模型架构等硬核议题上展现出冷酷而敏锐的商业直觉。整场对话节奏极快且情绪饱满，在尖锐揭露政治操盘与科技泡沫之余，散发着务实、辛辣而又充满精英洞察的智力狂欢气息。 ## Bulletin - [11:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=683s) OpenAI 首席营收官内部备忘录外泄，直指 Anthropic 虚报约 80 亿美元 ARR，与此同时二级市场上 Anthropic 的交易估值首次反超估值 8500 亿美元的 OpenAI。 - [49:48](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2988s) AI 算力遭遇实体硬瓶颈，全美约 100 个数据中心项目因居民抗议与电网过载受阻，涉及总额达 1620 亿美元，行业被迫加速转向自备能源（BYOE）离网模式。 - [59:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3563s) 加州众议员埃里克·斯沃韦尔因严重性丑闻退选加州州长并辞去国会职务，揭示出民主党当权派在全开放初选危机下冷酷引爆黑料的政治机器操盘战术。 - [01:12:08](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4328s) 美股在伊朗军事行动进入第 7 周之际接连创下历史新高，市场全面消化伊斯兰堡谈判并提前定价“和平溢价”，但席勒市盈率等指标与巴菲特持有的 3000 亿美元现金仍释放极强防御信号。 ## Key takeaways - [00:42](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=42s) 纽约市拟针对价值超 500 万美元的非自住物业征收年化 3.9% 的豪宅临时居住税，此举被指将大幅推高持有摩擦成本并摧毁高端住宅开发生态。 - [06:05](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=365s) 奥斯汀通过放宽土地开发管制并在人口流入下实现租金连续三年下降，与蓝州城市的邻避主义及高交易摩擦税负形成鲜明对比。 - [21:14](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1274s) OpenAI 战略重心面临企业级与 C 端的两线拉扯，而 Anthropic 凭借代码场景（Coding）的水电计量计费模式实现了每年 10 倍的爆炸性年化增速。 - [26:50](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1610s) 前沿大模型厂商极度依附云巨头现役算力配额，缺乏自有算力基础设施使初创公司在规模网络效应竞逐中极易被卡脖子。 - [42:54](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2574s) 全美社区抗议、水资源指控与地方公用事业垄断机制导致数据中心四处碰壁，迫使科技厂商转向自备天然气与光伏的离网 BYOE 供电模式。 - [42:54](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2574s) Anthropic 早期资助 AI 末日论与监管邻避主义以延缓对手自建数据中心，如今自身规模扩张被迫自建厂房时反遭反基建舆论反噬。 - [01:03:00](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3780s) 斯沃韦尔性丑闻黑料被政界私下雪藏数月，民主党当权派为避免加州全开放初选中出现两名共和党通吃决选的险情，决断“刺破脓包”逼其体面退选。 - [01:07:16](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4036s) 国会议员不受《公平披露规则》（Reg FD）直接约束，罗·康纳等人被披露的高额股票交易频次再次引发公众对国会内幕交易特权的声讨。 - [01:10:31](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4231s) 伯克希尔·哈撒韦坐拥 3000 亿美元巨额现金并保持防守，折射出顶级价值投资者对当前狭窄市场广度与极高席勒估值的警惕。 - [01:12:08](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4328s) 美伊在伊斯兰堡接触推动股市收复战时全部失地并破纪录，特朗普对股市形象（Optics）的敏感促使其将军事行动定调为短期涉险。 - [01:20:30](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4830s) 编码模型在模型层实现了真实的商业回报，但生成式 AI 在复杂传统大型企业的落地仍深陷官僚阻力与变革管理泥潭，尚未出现普遍的净利润海啸。 - [01:24:24](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=5064s) 创始人主导的科技公司研发敏捷度因 AI 成倍提升，但当前 Agent 仍缺乏独立推理与常识判断，脱离人机协同的“AGI 已至”属于过度包装炒作。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=0s) 特拉维斯·卡兰尼克做客，开场调侃与“人肉搜索”隐患 节目在一阵轻松的打趣中开场，主持群“核心四人组”迎来了重量级嘉宾——被誉为“原子之王”（The King of Atoms）的优步（Uber）联合创始人特拉维斯·卡兰尼克（Travis Kalanick）。卡兰尼克透露自己当时正坐在戴维·萨克斯（David Sacks）隔壁录制节目，萨克斯随即开玩笑警告大家千万别暴露具体地址、搞“人肉搜索”（doxxing）。杰森·卡拉卡尼斯（Jason Calacanis）立刻接梗称不用担心，纽约政客佐兰·曼达尼（Zohran Mamdani，节目中被戏称为“Mondummy”）已经在门外守着，正准备找上门去查封房产，因为他觉得“纽约已经没有有钱人可税了”，所以开始把目光转向其他城市搜刮税收。 ### [00:42](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=42s) 纽约拟征“豪宅临时居住税”引发争议 话题迅速切入纽约市拟推行的高额“豪宅临时居住税”（pied-à-terre tax，非主要居所房产税）。据坊间传闻与推测，针对价值超过 500 万美元的非主要居所（第二套及以上住宅），该税率可能高达每年 3.9%。杰森对此猛烈抨击，指出在曼哈顿中城方圆 15 英里范围内，几乎找不到 500 万美元以下的像样房产，这实际上是对任何在纽约置办次居所的人征收的惩罚性重税。萨克斯指出，这类买家是整个房地产市场中“需求弹性最高”的群体，他们根本不依赖纽约常住，完全可以把资金投到全世界任何地方；一旦开征如此沉重的持有税，将直接摧毁曼哈顿对第二套住房的需求，进而重创整个房地产市场。 面对关于“打压高端房价是否有利于普通人住房负担能力”的疑问，杰森反驳称，曼哈顿 57 街或格拉梅西公园（Gramercy）的顶层豪华公寓不可能被拆分成七套保障性廉租房，这种政策反而会彻底扼杀开发商盖楼的积极性。 讨论还涉及到曼达尼在宣传视频中的挑衅行为：他直接站在亿万富翁对冲基金巨头肯·格里芬（Ken Griffin）的豪宅楼下用手指向该房产拍摄视频。尽管查马斯·帕里哈皮蒂亚（Chamath Palihapitiya）认为格里芬买下该豪宅早已是公开的市场新闻、算不上严格意义的隐私泄露，但杰森极其严肃地强调，在数千万人观看的视频中指着富豪住宅煽动民怨是一种危险的“政治暗号”（dog whistle）。杰森联想到此前联合健康保险（UnitedHealthcare）CEO 遭枪杀，以及近期山姆·奥特曼（Sam Altman）住宅遭枪击和莫洛托夫燃烧瓶袭击的恶性事件，直言无论公众对某位公众人物评价如何，煽动暴力的做法都越过了红线。 - **拟议税率与门槛**：传闻每年约 3.9%，适用于非主要居所且估值超过 500 万美元的物业。 - **投资持有成本**：以一套 1000 万美元的住宅为例，叠加利息、通胀与该税率，持有人在 10 到 11 年内房产总成本几乎翻倍，需要涨到 2000 万美元才能保本，投资逻辑荡然无存。 - **安全担忧**：政客公开指认富豪物业拍摄宣传片，在社会对立加剧的当下极易诱发极端暴力事件。 ```mermaid flowchart TD accTitle: 纽约拟推豪宅次居所税的连锁反应 accDescr: 展示从开征每年3.9%的高额税率到最终重创城市财政与开发生态的逻辑演变 subgraph Policy ["政策出台"] A["拟征年化 3.9% Pied-à-terre 税<br/>（门槛 500 万美元以上次居所）"] end subgraph MarketResponse ["市场即时反应"] B["高净值外来买家需求断崖式下跌"] C["持有成本翻倍（10年翻倍）<br/>买家转向其他低税城市/酒店"] end subgraph DownstreamImpact ["下游经济与开发连锁影响"] D["顶层豪宅超高溢价消失<br/>新楼盘开发无法通过财务核算"] E["缺少富豪高消费与无成本税收<br/>城市财政收益受损、建设停滞"] end A --> B A --> C B --> D C --> D D --> E ``` ### [04:15](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=255s) “次居所文化”与伦敦空心化教训 查马斯表示自己并不推崇富豪在多地购置多套房产的“次居所文化”（pied-à-terre culture），直言自己度假只会选择住酒店。他以伦敦为例，说明大量海外资本将房地产作为单纯的资产储备手段（land banking），反而将部分高档社区“掏空”：在周四或周五的夜晚，切尔西（Chelsea）等老牌豪宅区宛如无人居住的“鬼城”。此外，这项税收如果涵盖租客非主要居所的各种长短期租赁物业，房东同样会被课税，最终可能只利好酒店业。 但萨克斯与弗里德伯格（David Friedberg）指出，英国引入印花税（Stamp Duty）重税改革后，伦敦的高端房地产市场遭遇沉重打击，不仅未能让普通人受益，反而损害了伦敦乃至整个英国的经济活力。全球财富之所以愿意停泊在纽约或伦敦，是建立在对当地法治环境、产权保护和世界级稀缺地位的信任之上。如果市政当局频频施加任意且不可预测的惩罚性重税，巨额资本便会毫不犹豫地撤出，流向其他保护资本的避风港。 ### [06:05](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=365s) 奥斯汀供需模式 vs 纽约干预主义 主持群进一步对比了不同执政理念下的住房政策。查马斯直言所谓“通过限制和惩罚富人来解决住房危机”纯属无稽之谈。以得克萨斯州奥斯汀以及内华达州、佛罗里达州为例，这些地方奉行自由放权和尊重土地开发权的原则——“地是你的，你有权开发它”。 数据显示，奥斯汀在过去十年中城市规模几乎翻倍，并在人口持续净流入的同时，实现了连续三年租金与住房价格双双下降。萨克斯与杰森犀利地总结了这一鲜明对比：民主党主政的蓝州城市口口声声高喊解决“住房负担能力”，却纵容邻避主义（NIMBY）阻碍新建任何住房；而共和党主政的地区通过鼓励充分开发、增加市场供应，真正实现了租金与房价的平抑。 ### [08:18](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=498s) 高净值业主对城市财政与城市建设的隐形补贴 深入到市政财政与商业开发的微观逻辑，嘉宾们指出了政客政策中严重的经济盲区。像肯·格里芬这样的外来豪宅业主，不仅早已缴纳了极其高昂的常规房产税，而且由于他们一年中大部分时间并不住在纽约，几乎不消耗任何城市公共资源、治安警力或公共设施，对纽约市财政来说纯粹是“高利润客户”。 更关键的是，在大规模综合开发项目中，正是因为有这种对价格不敏感的超级富豪（“大鲸鱼”）愿意以惊人的单价为顶层复式等溢价单元买单，才使得整个大楼建设项目的账目能够“算得过来”（make the whole project pencil out）。高层豪华单元的高利润在事实上交叉补贴了整体建筑工程与周边配套的落成，构成了纽约地平线上吊车林立、建设繁荣的基础。一旦消灭了这部分需求，不仅外来高净值人群在纽约看球赛、光顾高档餐厅的衍生消费会随之流失，整个城市的房地产投资与开发周期也将陷入深度停滞。 ### [09:09](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=549s) 财富外流与蓝州地产税危机：从伦敦非本地居民税到加州豪宅税 全美与欧洲主要城市对高净值人群加税的政策正在引发资本的逆向流动。Travis 指出，很多富豪每年缴纳三四百万美元的税款，却几乎没有享受任何市政服务，对城市来说这原本是“纯利润”。伦敦废除“非本地居民”（non-dom）税收优惠地位后，大批富豪迅速转向苏黎世、卢加诺和米兰等税收更为友好的城市。与会嘉宾提到，英国政府官员私下表示目前尚未看到足够剧烈、足以拉响“五级火警”的冲击，但这种政策带来的伤害往往是渐进式的“温水煮青蛙”，等到城市真正被掏空，局面将极难逆转，因为不会有某个突发的灾难性时刻迫使政客立刻改弦易辙。 与此同时，美国加州各大城市也在推行类似的激进税负。洛杉矶出台了 5% 的豪宅税（Mansion Tax），而旧金山的房产交易过户税更为严苛：起征点为 500 万美元，超过 2500 万美元的豪宅税率更是高达 6%。加上约 5% 到 6% 的房产中介佣金，在旧金山出售一套高净值房产的交易摩擦成本已经飙升至 13% 左右。Sacks 直言这种税收带有某种倒算性质，就像政府平白无故抢走了屋主的几间房，直接导致洛杉矶等地的豪宅交易量彻底枯竭，房屋翻修套利（house flipping）几近停滞。他尖锐地警告：“在蓝州，你的财产根本不安全，那里的政治阶层总觉得自己有权分走一杯羹。”拥有资金选择权的富豪们正加速将资产撤离蓝州。 - **伦敦资本外流目的地**：瑞士苏黎世、卢加诺，以及意大利米兰。 - **旧金山房产过户税**：500 万美元起征，超过 2500 万美元税率达 6%；叠加中介佣金后，综合出售摩擦成本高达约 13%。 - **核心市场后果**：高额交易税导致洛杉矶与旧金山的高端房地产交易量断崖式萎缩。 --- ### [11:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=683s) OpenAI 内部备忘录泄露：怒怼 Anthropic 虚高营收与企业级转型之争 OpenAI 首席营收官（CRO）Denise Dresser 向全体员工发送了一份长达四页的内部备忘录，并在极短时间内被“战术性外泄”给媒体。备忘录火药味极浓，直接开火抨击竞争对手 Anthropic，声称对方对外宣称的 300 亿美元年化运行率（run rate）纯属虚夸（cap），其中至少有 80 亿美元是通过与 AI 模型分发平台的营收分成和特定会计手法注水得来的。Dresser 甚至在备忘录中对 Anthropic 的安全叙事发起价值观层面的攻击，指责其品牌故事完全建立在“恐惧、限制，以及少数精英阶层应当垄断控制 AI”的说辞之上。 备忘录同时披露了 OpenAI 的重大战略转向：公司正全力争夺企业客户，并决心夺下智能体平台层（agent platform layer）。此前 OpenAI 招募了开源项目 OpenClaw 的架构师 Peter Steinberger，业内普遍猜测其意图是将后续创新锁在 OpenAI 内部，而非回馈开源社区。与此同时，竞品阵营全面开花：Perplexity 业绩暴增、营收翻了四倍，xAI 的马斯克正在酝酿全新杀手锏，而 Polymarket 预测 OpenAI 新模型 Spud 有 75% 的概率于下周发布。然而，《金融时报》引述匿名投资人的抨击，直指 OpenAI 已经“严重丧失专注度”：“你手握 ChatGPT 这样一个拥有 10 亿用户、年增速高达 50% 到 100% 的业务，为什么还要跑去死磕企业级和代码？这简直是一家毫无重心的公司。”目前随着 Claude 与 Gemini 的攻城略地，ChatGPT 的市场份额正在被逐步侵蚀。 ```mermaid flowchart TD accTitle: OpenAI 战略分歧与竞争格局 accDescr: 展示 OpenAI 面对 Anthropic 竞争与自身战略重心的分化选择 subgraph External["外部竞争压力"] ANT["Anthropic Claude<br/>发布节奏极快 / 增速强劲"] GOOG["Google Gemini / Meta 开源<br/>蚕食 C 端流量份额"] end subgraph Controversy["内部备忘录交锋"] MEMO["CRO 泄露备忘录<br/>指责 Anthropic 营收虚高 80 亿<br/>批判精英恐惧叙事"] end subgraph Split["业务重心抉择"] CONS["消费级市场 ChatGPT<br/>10 亿级用户 / 估值潜力 3-4 万亿美元"] ENT["企业级与开发者 Codex<br/>抢占智能体平台 / 估值潜力 2-3 万亿美元"] end External --> MEMO MEMO --> Split CONS -. 投资人担忧分散精力.-> ENT ``` --- ### [13:50](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=830s) 双轮驱动还是战略分散？Codex 对阵 Claude 的企业价值推演 针对 OpenAI 是否应当全面转向 B 端开发者的争议，Chamath 结合其投资机构 8090 的实测经验提出了不同维度的观察。在日常、常规的编程任务中，Anthropic 的 Claude 模型集成往往更为顺手且稳定；但一旦面对逻辑极其复杂、执行周期极长的长程编程任务（long horizon coding tasks），OpenAI 的 Codex 则展现出明显更强的性能与功能上限。因此，在技术底座上 OpenAI 绝非无的放矢。 从资本商业化的角度来看，Chamath 认为两线作战虽然会引发上下文切换的损耗，但背后的商业价值诱惑巨大。若全力巩固消费级霸主地位，ChatGPT 未来有望支撑起 3 万亿至 4 万亿美元的企业价值；若能单独划定资源、组建独立团队死磕以 Codex 为核心的企业级开发市场，又能在 B 端捕捉 2 万亿到 3 万亿美元的市场空间。二者叠加，长远来看足以构筑一个 7 万亿到 9 万亿美元市值的超级帝国。关键在于管理层必须将消费端团队与企业端团队进行彻底的组织物理隔离，让 C 端团队全力冲刺，同时赋予 B 端团队自主决策权，避免相互牵制。 - **8090 编程模型测评**：日常代码工作使用 Claude 组合更为稳定；复杂、棘手的长周期任务 Codex 优势明显。 - **潜在估值路径推演**：消费端垄断对应约 3 至 4 万亿美元，企业级生态对应约 2 至 3 万亿美元，远期整体体量上看 7 至 9 万亿美元。 - **组织结构要求**：必须设立边界清晰的独立业务单元，杜绝 C 端与 B 端之间的上下文切换内耗。 --- ### [15:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=923s) 次级市场反转与规模飞轮：Anthropic 估值反超 OpenAI 的隐忧 在非公开股权二级市场上，格局正在发生被称为“大反转”（The Flippening）的历史性剧变：Anthropic 的二级市场交易估值有史以来首次反超了 OpenAI。此前 OpenAI 刚刚完成了一轮高达 8500 亿美元估值的融资（最新一轮吸纳了 1220 亿美元资金，堪称全球一二级市场有史以来最大单轮融资）。然而彭博社等信源显示，目前二级市场根本没有买家愿意在这个估值水平接盘，甚至有投资人指出，OpenAI 的 IPO 市值必须达到 1.2 万亿美元以上才能让这轮入场的投资机构保本。 曾在 Uber 经历过残酷死斗的 Travis 深刻指出，在当前的 AI 军备竞赛中，“增长就是一切”（Growth is the whole damn thing）。无论你当下的绝对体量多大，只要竞争对手以数倍于你的速度在狂飙，投资人就会在模型中折现这一未来趋势。规模决定了计算资源（compute）、生成 token 吞吐量以及强化学习（RL）数据飞轮的正向循环，这与当年 Uber 依靠资本为武器、通过网络效应彻底甩开 Lyft 和 DoorDash 的逻辑如出一辙。如果 OpenAI 的营收与用户增速落后于 Anthropic，即便账面依然庞大，其护城河也会被加速掏空。Friedberg 同样指出，Anthropic 近期的产品迭代与模型发布速度堪称史无前例，不仅最新发布的 Opus 惊艳市场，其功能更在短短数天内全面压制并取代了 OpenClaw，展现出令人恐惧的技术执行力。 - **二级市场估值逆转**：Anthropic 场外定价首次超越估值 8500 亿美元的 OpenAI。 - **单轮融资金额**：OpenAI 完成了创纪录的 1220 亿美元巨额融资，但后续退出面临极高估值溢价压力（预期至少需 1.2 万亿美元 IPO）。 - **平台规模网络效应**：更大的规模带来更多客户 token 吞吐与强化学习数据，资金优势正成为算力壁垒的核心加速器。 ### [18:53](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1133s) 效率胜过补贴：Anthropic 的造血飞轮与 OpenAI 的融资神话 讨论切入到大模型格局的最新剧变。Travis Kalanick 指出，Anthropic 的势头不仅体现在用户规模的膨胀上，更体现在其极其恐怖的业务迭代节奏上，这种产品升级的密度已经明显凌驾于其他所有对手之上。以他自己的团队为例，仅仅在六个月前，整个组织还在重度依赖 Cursor 和 Google Gemini，而如今短短半年时间，内部已有大约 90% 的工作流全面倒向了 Anthropic。这种强大的飞轮效应正在底层迅速成型。 谈及竞争格局，Travis 直言必须对 OpenAI 能够完成 1200 亿甚至 1300 亿美元的天价融资表达“极大的敬意”，但这并不意味着胜负已定。资本固然能够砸出早期的网络效应和规模，但如果对手是靠真实收入和边际贡献利润（Contribution Profit）来推动规模扩张，那么“运营效率最终必然碾压资本补贴”。没有人能无休止地一轮接一轮去融千亿美元级别的资金，这种融资狂欢总有撞墙的一天。如果 Anthropic 正在依靠源源不断的真实商业营收完成自我造血，而其他玩家依然深陷外部资本输血的泥潭，从长期来看，这种能够自给自足吞吐研发算力的造血机器，对任何竞争对手而言都极其致命。 面对主持人关于老牌科技七巨头（Mag 7）是否还有机会争夺前二的提问，大家梳理了巨头们的弹药库：Meta 和 Google 依然手握庞大核心业务产生的充沛净利润，并将其源源不断地重新配置到底层基础设施中；特斯拉现金流充裕，SpaceX 坐拥巨额现金，马斯克不仅在全力扩建 Colossus 超级计算集群，还在推进自建晶圆厂（Fab）的计划。市场究竟会沦为 OpenAI 与 Anthropic 的二人转，还是巨头们依然能凭借底层基建反扑，悬念依然巨大。 --- ### [21:14](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1274s) 企业级代码场景的指数级爆发与消费级订阅的局限 David Sacks 赞同 Google DeepMind、xAI 和 Meta 依然具备顶级的资源与团队，绝未出局，但他将核心焦点拉回到对 OpenAI 的批判上。Sacks 指出，外界批评 OpenAI 缺乏专注力是完全合理的，比如完全让人摸不着头脑地斥资数亿美元去收购播客网络——“如果你们真打算花几亿买个科技播客，我们这帮人一直都在这儿，Dario（Anthropic CEO Dario Amodei）赶紧联系我们，直接给我们把票验了！”虽然现场随即调侃巨头们根本买不起 All-In 团队，但 Sacks 强调，外界指责 OpenAI 转向企业级市场（Enterprise）则是彻头彻尾的误判。OpenAI 当前最致命的问题不是进入企业级，而是进入得不够坚决、做得不够精准。 Sacks 搬出了一组令人震撼的对比数据：进入第二季度时，OpenAI 和 Anthropic 的年化运行率（ARR）大约都在 300 亿美元上下。即便剔除渠道分销伙伴的收入，Anthropic 在纯同口径下略低 20%，两者的增速也完全不在一个量级： - **OpenAI 的同比增速**：维持在每年 3 至 4 倍。 - **Anthropic 的同比增速**：达到了惊人的每年 10 倍（从去年同期的 10 亿直接飙升至 100 亿，到今年第一季度末已冲到 300 亿，按照 Brad Gerstner 等投资人的测算，年底极有可能直奔 800 亿至 1000 亿美元）。 在对数坐标图上，Anthropic 的收入曲线几乎是一条以“10 倍”为阶梯笔直向上的倾斜线，这在商业史上前所未见。如果 Anthropic 翻 10 倍只需 1 年，而 OpenAI 需要 2 年，终局的胜负显而易见。而拉开这一差距的核心本质在于**商业模式与场景**： - **企业端（代码场景）**：Anthropic 极度专注企业级市场，尤其是编程代码（Coding）。企业对代码 Token 的付费模式类似于自来水和电力计量计费（Metered Basis），业务跑得越多、效率越高，企业就越乐意付出无上限的预算，需求呈井喷式扩张。 - **消费端（C端订阅制）**：OpenAI 押注的消费者市场付费意愿极低，全网仅有 3% 到 4% 的转化率，且普遍习惯 20 美元/月的“不限量自助餐（All-you-can-eat）”包月模式。这种商业模式在算力成本膨胀的当下，收入天花板极低且无法线性规模化。 ```mermaid flowchart TD accTitle: 商业化模式路径对比 accDescr: 展现企业级计量付费与消费级包月订阅在增长飞轮上的本质差异 subgraph Enterprise ["企业级代码市场 (Anthropic 路径)"] E1["极度聚焦 Coding 场景"] --> E2["按 Token 计量计费 (类似水电模式)"] E2 --> E3["企业用量无限扩容 / 极高付费意愿"] E3 --> E4["年化收入 10X 指数级垂直增长"] end subgraph Consumer ["消费级订阅市场 (OpenAI 路径)"] C1["泛 C 端日常交互与对话"] --> C2["$20/月 All-you-can-eat 订阅制"] C2 --> C3["仅 3%-4% 付费转化率 / 算力边际成本高"] C3 --> C4["年化增速放缓至 3-4X 撞上天花板"] end ``` --- ### [25:02](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1502s) 物理现实的铁壁：算力瓶颈、缩减思考与“末日论”的反噬 然而，Sacks 同样发出了清醒的警告：即便 Anthropic 的对数曲线再完美，也不可能打破物理定律。今年如果冲到 1000 亿美元，明年难道真能凭空跃迁到 1 万亿美元？物理世界的硬约束正在迎头撞来——全球算力储备、电力供应、数据中心建设周期都是无法逾越的实体壁垒。 事实上，Anthropic 已经显露出了撞墙的端倪。前不久大量核心开发者在社区抱怨 Claude 出现异常，在执行复杂提示词时“思考时长直接被砍掉了三分之二”，疑似官方为了应对服务器过载和算力枯竭而暗中降配。尽管当天刚刚发布的 Opus 4.7（替换原有的 Opus 4.6）似乎恢复了深度思考能力，但这背后很可能伴随着更高的调用单价与更严苛的限流。 更具讽刺意味的是意识形态的反噬。Sacks 指出，Anthropic 早期为了商业利益一直大力资助和倡导“AI 灭绝论”与监管邻避主义（Doomer Nimbyism）。在自己只需要向云巨头租用算力时，用安全主义卡死对手自建数据中心的审批是绝佳的商业防御策略；但如今，当他们自身的爆炸性体量倒逼其必须亲自下场大建特建超大型数据中心时，Anthropic 极有可能被自己亲手培育的反基建思潮作茧自缚（"hoisted on their own petard"）。 - 此外，打磨顶级代码模型的附带红利在于**自举（Bootstrapping）**：内部工程团队使用自家 AI 编程平台打造更强大的下一代 Anthropic 模型，形成了“产品变得更好”与“企业为此买单”的双重收益闭环。 --- ### [26:55](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1615s) 算力博弈与云巨头的掐脖子危机 针对代码模型的热潮，讨论迅速迎来了冷静的反思。当前社交网络上已有大量工程师公开炮轰 Claude Desktop 充斥着“氛围编程烂代码（Vibe coded slop）”。AI Agent 必须在受到极严密引导和规则约束的前提下才能成为真正的生产力倍增器。残酷的现实在于： 1. 行业目前根本没有任何一家公司真正搞懂了如何大规模凭借 Agent 交付工业级复杂软件； 2. 没有任何管理层知道如何重构与 Agent 协同的组织架构； 3. 财务上更是无人通晓如何做合理的 Token 预算管控——优步（Uber）CTO 近期甚至公开表态直接放弃，因为工程团队在无节制的调试中彻底刷爆了公司的 Token 额度。 更深层次的地缘商业危机正在前沿实验室与云巨头之间爆发： - **前沿实验室的致命依附**：无论是 OpenAI 还是 Anthropic，在早期阶段租用超大规模云服务商（AWS Bedrock、Google Cloud、Azure）的算力是省时省力的最优解，但伴随业务极速膨胀，这种租赁关系已演变为致命的战略依附。一旦缺乏底层的自建算力支撑，实验室在云巨头面前就犹如手捧铁钵乞讨算力配额的乞丐。 - **超大规模云服务商的卡脖子博弈**：最新数据显示，几家核心超级云巨头牢牢锁定了全球超过 60% 的现役算力。从博弈论角度出发，当自研模型（如 Google Gemini、微软自研生态等）全面发力时，云巨头完全有动机通过算力配额管控来卡死前沿实验室的扩张咽喉，迫使整个 AI 战争的技术制高点重新向拥有发电厂、电网和服务器机房的物理巨头倾斜。 ### [28:50](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1730s) 算力封锁与巨头算力军备竞赛 通过限制或节流算力供给，巨头完全有能力遏制后起之秀的势头。在社交网络时代，Friendster 曾经风靡一时，但由于系统极其缓慢、服务器响应跟不上爆发的需求，导致 MySpace 趁虚而入抢夺了全部市场，随后 Facebook 又再次将其颠覆。如今的大模型企业正面临相似的境遇：如果前沿实验室拿不到足够的底层算力支撑，哪怕市场需求再旺盛也会被活活卡死。OpenAI 试图摆脱对微软及外部算力支出的依赖，但除了亲自下场重度投资算力设施外几乎别无选择，Anthropic 亦是如此。与此同时，企业端订阅模式正在经历巨变——各大模型厂商已无法继续贴钱补贴 Token 消耗。一旦企业发现每月的 Token 账单暴涨 30%、40% 甚至 50%，而仔细审查生成的代码后却发现充斥着一堆无用的低质垃圾（slop），这种缺乏商业正向回报的粗放开销必将遭遇严厉的预算审查。 与此同时，超大规模算力集群的竞争正在急剧升级： - **Elon Musk 与 xAI 的 Colossus**：正在横跨三栋大楼扩张至 55.5 万张 GPU，累计投资额高达 180 亿美元；Elon 还与 Cursor 达成算力租赁协议，将多余算力对外出租用于模型训练，正式切入超大规模数据中心服务业务。 - **Meta 的 Prometheus**：计划于 2026 年上线，预计部署 15 万张 GPU。 对于平台来说，超前过度建设算力具有天然的战略优势：既能确保自有模型优先获得充足资源，又能将剩余算力高价倒手给竞争对手。对于 Anthropic 此前未公开全面发布 Mythos 模型的举动，马克·安德森（Marc Andreessen）等人指出，其真实原因很可能是因为缺乏足够的算力去支撑其日常调用。Mythos 模型的规模极其庞大，单个 Token 的推理成本据传高达 Opus 的 10 到 20 倍。由于 Anthropic 必须将宝贵的算力留给即将发布的 Opus 4.7，因此借由“发现潜在代码重大安全漏洞、留出修补时间”的名义推迟发布，不仅掩盖了算力匮乏的现实，反而在政府和公众面前塑造出极具前瞻性与利他主义的技术标杆形象，打了一场漂亮的公关营销战。 ```mermaid flowchart TD accTitle: 大模型算力瓶颈与商业模式传导 accDescr: 算力短缺导致模型受限并传导至企业端成本核算的流程图 subgraph Infra["底层算力供给 (Compute Bottle-neck)"] A["巨头控制顶级算力<br/>(Colossus / Azure / AWS)"] --> B["初创模型厂商算力吃紧<br/>(OpenAI / Anthropic)"] B --> C["高成本大模型被迫雪藏/限流<br/>(如 Mythos 延后商用)"] end subgraph Enterprise["企业端商业传导 (Cost Pass-through)"] B --> D["取消 Token 资本补贴<br/>(转嫁真实推理成本)"] D --> E["企业级 Token 账单暴增 30%-50%"] E --> F["管理层严查低效代码/垃圾产出<br/>(反思 ROI，削减预算)"] end ``` ### [32:33](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1953s) Polymarket 恶搞与“涓滴航空学” 在转向下一个严肃商业议题前，主持群注意到预测市场 Polymarket 上出现了一个离谱的交易标的：Anthropic 在年底前收购《All-In》播客的概率竟然被炒到了 37%，成交量显示甚至高达数千万美元。几位主持人立即起哄拆台，质疑这根本是虚假造势或自导自演的恶搞（"Jay Cal slop"），认为背后可能连 92 美元的真金白银都没有，纯属任何人都能随意创建预测盘口的产物。 调侃之余，几位亿万富豪主持人还借机开启了私人飞机机型鄙视链的互怼与自嘲。众人打趣称，万一真被天价收购，就该全面升级座驾了：“不用买皮拉图斯 PC-24 了，谁手上有湾流 G650 的赶紧换成 G800，把二手的顺位留下来，我也终于能放弃美联航的白金卡了。”这种打趣被众人戏谑地冠以“涓滴航空学”（Trickle-down avionics）的名头。 ### [33:53](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2033s) Allbirds 转型 AI 暴涨与零利率狂欢的反思 网红跑鞋品牌 Allbirds 宣布转型人工智能及数据中心方向，这一看似荒诞的跨界竟推动其股价单日暴涨超过 400%。回顾这家曾经以“环保羊毛鞋”闻名硅谷、被戏称为“最丑运动鞋”的 DTC 品牌，其兴衰史正是 2021 年零利率政策（ZIRP）狂欢顶峰的典型注脚。Allbirds 在 2021 年上市时估值曾突破 40 亿美元，IPO 募资高达 3.5 亿美元；然而随着热潮褪去，其核心品牌资产最终仅以 3900 万美元被贱卖（仅约为 IPO 融资额的 10%）。这种狂热宛如 90 年代末互联网泡沫重演——当时只要把公司名字加上“.com”就能坐享估值暴涨。 这番讨论引发了关于当年实体业务被当成纯软件公司盲目估值的深刻反思： - **忽略毛利率与销货成本（COGS）**：在纯软件领域，服务增量客户的边际成本趋近于零；但在物理世界制造实体消费品，每双鞋都受制于供应链、物流与原材料成本。然而在狂热期，硅谷投资人完全无视这一基本财务常识。 - **两到三年远期增速贴现**：特拉维斯·卡兰尼克（Travis Kalanick）指出，2021 年的估值扭曲不仅是实体与数字业务的差异，更是大量充裕流动性在通胀未显现前的盲目投机。投资人不再看次年业绩，而是直接按照每年两三倍的增速提前透支企业未来两到三年的远景，导致估值体系彻底失真。 - **Bird 共享单车的兴衰教训**：大卫·萨克斯（David Sacks）针对其早年领投 Series A 的 Bird 共享滑板车进行了反思。当年估值曾达二三十亿美元的 Bird 虽敏锐捕捉到了短途出行的微出行（micro-mobility）趋势，却最终被城市监管的紧箍咒扼杀。许多城市不仅没有为小型电动出行工具规划专属车道，反而采取一刀切的禁令与配额限制；甚至在 Bird 已拿下 80% 市场份额的城市，强行引入四家竞品按人头均分微不足道的牌照份额，最终将原本极具潜力的创新业态挤压致死。 ### [38:04](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2284s) 监管俘获的恶果：从共享滑板车到自动驾驶汽车 当地方政府开始插手市场并人为指定赢家、平分份额时，整个市场的经济模型和网络效应便被彻底摧毁。嘉宾们指出，这种做法消除了竞争，剥夺了企业创造价值的空间，是典型的“最恶劣的监管俘获（Regulatory capture at its worst）”。大家警告称，自动驾驶汽车（AV）领域也必须警惕同样的陷阱：纽约和波士顿等城市已经在动类似的心思，试图模仿当年出租车牌照（Medallion）的做法，人为给上路的自动驾驶车辆设定几千辆的配额上限，而不是任由市场自由竞争。这种政策的最终后果是价格飙升、技术创新无法落地，消费者得不到任何实惠，整个市场的生机被直接抹杀。 与此同时，话题转回了运动鞋品牌 Allbirds（被戏称为转型为“New Bird AI”）的荒谬投机闹剧。市场上充斥着这家公司通过 5000 万美元可转债购入算力、全力押注 AI 的传闻，甚至有人开玩笑说他们是不是就买了“8 张 H100”。然而，散户的狂热直接把股价推高到了每股 14 美元，单周暴涨超过 450%。嘉宾们调侃这是向 WallStreetBets 致敬，“市场的非理性时间往往比做空者的资金承受力还要持久”。 - 纽约与波士顿正尝试对自动驾驶汽车实行总量配额限制，重蹈出租车牌照导致成本高企、扼杀创新的覆辙。 - Allbirds 宣布转型 AI 后一周内股价暴涨 450%，达到每股 14 美元。 --- ### [39:25](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2365s) 严重的算力与电力瓶颈：为什么 AI 数据中心四处碰壁 在看似荒诞的市场现象背后，资本市场其实押对了一个残酷的底层现实：全行业正面临极其严重的“算力枯竭（massively compute constrained）”。近期类似 Jane Street 向一家新型 Neoscaler 投资 10 亿美元并签订 60 亿美元算力协议等令人咋舌的交易，均源于此。算力瓶颈主要体现为两大致命难题： ```mermaid flowchart TD accTitle: AI 数据中心建设面临的双重瓶颈 accDescr: 展现算力受限下电力供应与土地审批两大阻碍及其引发的链条反应。 subgraph Bottlenecks["底层资源约束"] P["电力缺口<br/>电网排队耗时数年"] L["土地与审批壁垒<br/>地方民意反弹与禁令"] end subgraph Solutions_Frictions["应对与现实冲突"] B["现场天然气供电方案<br/>（如 Bloom Energy 迅速获批）"] V["社区公投罢免官员<br/>（如推翻 60 亿美元审批）"] end subgraph Crisis["前沿实验室危机"] F["Friendster 效应崩溃<br/>（因算力不足导致服务停摆）"] end P --> B L --> V B --> F V --> F ``` 首先是电力。像 Bloom Energy 这样的公司股价之所以如火箭般垂直飙升，是因为它们提供了能够利用现场天然气发电且排放极低的解决方案，企业无需耗费数年苦等电网接入批准，就能快速拿到清洁空气许可并直接驱动数据中心。其次是土地与厂房审批。美国普通民众对 AI 的态度正逐渐转向负面，背后的原因交织着末日论的渲染、对自身工作被替代的恐惧，以及对技术红利仅被极少数万亿富豪掠夺的不满。 这种民意反弹最直接的体现就是选民在地方层面全力狙击数据中心。例如，某小镇在批准了一处 60 亿美元的数据中心建设项目后，当地居民一夜之间将半数理事会成员公投罢免，随即换上新人推翻原决定；缅因州甚至直接立法全面禁止兴建数据中心。把希望寄托在得克萨斯州也是不现实的，当地的电网容量和天然气供应同样捉襟见肘。对于 OpenAI 和 Anthropic 而言，这无异于“五级火警”——如果拿不到土地、电力和厂房，它们的营收增长将直接撞墙，而原因根本不是产品不好或缺乏用户需求，纯粹是因为遭遇了当年的“Friendster 效应”：算力撑不住激增的流量，网站根本无法保持在线。 - Jane Street 向新型 Neoscaler 注资 10 亿美元，并签订了价值 60 亿美元的算力采购协议。 - 某地方城镇批准 60 亿美元数据中心项目后，选民迅速通过选举罢免了半数理事会成员以撤回批准。 - 缅因州通过法案全面禁止兴建 AI 数据中心。 --- ### [42:54](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2574s) 环保反弹、伪草根运动与 Anthropic 的战略自食其果 大卫·萨克斯（David Sacks）进一步剖析了数据中心在全美陷入空前孤立的三大驱动因素，并预测未来甚至可能有多达 30 个州明令禁止数据中心建设： 1. **掠夺公用电网的恐慌**：大量无资质的地产投机商（Wildcatters）在未解决自有供电方案的情况下盲目申报项目，导致社区强烈抗议数据中心挤占公共电网并推高居民电价。为此，白宫推动了“用电户保护承诺（Ratepayer Protection Pledge）”，要求所有超大规模云厂商（Hyperscalers）承诺建设新数据中心必须自备电源，实现对电网的“电力中和”，甚至在用电低谷时向电网反向输送电力。 2. **AI 末日派的人为操纵（Astroturfing）**：诸如“生命未来研究所（Future of Life Institute）”等末日论组织意识到无法靠“终结者毁灭人类”的说辞说服大众，便转而在地方环保议题上大做文章，编造并散播“数据中心将抽干居民生活用水”的虚假指控，通过科技富豪的巨额捐款在地方基层煽动邻避主义（NIMBYism）。 3. **Anthropic 的政治同盟反噬**：最具讽刺意味的是 Anthropic 自身。早期它们为了给 OpenAI 和 xAI 等对手使绊子，选择在政治上与这些末日论及邻避团体结盟，寄希望于只从第三方云厂商采购算力而不亲自下场建厂。如今第三方算力触顶，Anthropic 被迫也要自建数据中心，当年的战略盟友反而成了自己头顶的紧箍咒，未来有效利他主义者（EA）该如何调转口风将极具戏剧性。 --- ### [45:44](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2744s) 民粹风暴的实体靶子：数据中心沦为“富人的神庙” 弗里德伯格（David Friedberg）提出了更为深刻的社会心理学视角：全美乃至整个西方世界正在积聚一股强烈的仇富情绪，而数据中心正是这种贫富撕裂在物理世界上最醒目的图腾。 在大众眼中，数据中心就是“超级富豪的神庙（the temple of the wealthy）”，是科技精英与建制派亿万富豪从底层敛财、把自己送上太空、却将普通人远远抛在身后的核心运作机器。普通消费者至今并没有从 AI 中感受到生活被切实改善的红利——绝大多数价值目前都滞留在企业级市场，普通人能体验到的无非也就是在 ChatGPT 上查个医疗建议而已。这种巨大的疏离感让民粹主义情绪不断膨胀，庞大冰冷的数据中心因而成了民众发泄怒火与破坏欲的最佳物理靶点，其性质就像曼哈顿针对富人征收的豪宅税（pied-à-terre tax）一样，成为了对抗富人阶级的攻击载体。针对这一困局，嘉宾们最后补充指出，由于数据中心的数据传输依托于光速，理论上并不需要非得建在大城市周边，只要有电，它们完全可以建在更偏远的地区。 ### [47:33](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2853s) 数据中心的迁徙、公用事业的盈利死结与民粹反噬 在讨论数据中心的选址困境时，嘉宾们指出数据中心选址的核心诉求本质上是廉价电力与低延迟；如果部分州开始禁止或抵制数据中心建设，微秒级的延迟差异根本无法阻挡资本的脚步，数据中心将直接转移到太空、冰岛、得克萨斯州甚至任何接纳它们的地方。Travis 认为，这种针对数据中心的抗议与此前对非自住住宅税（pied-à-terre tax）的讨伐如出一辙，不过是民粹主义寻找的下一个情绪宣泄靶子，未来甚至可能轮到私人飞机燃油税。民粹情绪抬头的根本原因在于政府债务失控、效率低下并过度承诺，最终引发美元贬值，同时未能给民众带来实质回报。 针对公众对电费上涨的担忧，David Sacks 此前参与推动的“用电户保障承诺”（Ratepayer Pledge）被视为一项关键进展。但 Chamath 对此提出了公用事业垄断机制层面的尖锐质疑。他剖析了公用事业公司的底层商业模式：公用事业本质上是由州和县级政府颁发特许经营权的地方垄断企业，其每年制定预算时的定价逻辑取决于投资规模，因为法规允许它们在基础设施投资额的基础上获得大约 10% 的固定资本回报率（ROE）。这种激励机制促使电力公司不断寻找扩建和翻新电网的借口——例如以应对山火风险为由将输电线路埋入地下，耗资上百亿美元并顺理成章地获取 10% 的稳定收益。因此，公用事业公司天然具备如同保险公司一般逐步推高终端价格的内在冲动。即便数据中心在现场自建电力设施，也无法扭转公共电网这一制度性痼疾，这也是“表后供电”（behind-the-meter）模式愈发关键的原因所在。 ```mermaid flowchart TD accTitle: 公用事业资本回报机制 accDescr: 展示地方公用事业公司通过基础设施投资获得固定回报的运作逻辑 subgraph Regulation ["监管与特许权"] A["政府特许经营垄断权"] --> B["公用事业委员会 (PUC) 审批"] end subgraph UtilityModel ["公用事业商业逻辑"] B --> C["提出高额资本开支项目<br/>(如投资 100 亿美元线缆入地)"] C --> D["按投资额赚取约 10% 固定回报"] D --> E["推高终端用户电价"] end subgraph Solution ["数据中心应对方案"] F["表后供电 (Behind-the-Meter)<br/>与自备能源 (BYOE)"] -.->|绕过电网垄断机制| E end ``` - **公用事业资本回报率**：地方垄断公用事业公司获准在其批准的资本支出投资上赚取约 10% 的固定收益，缺乏主动压低成本的动力。 - **项目支出规模**：以山火防护线缆地下化为例，单一公用事业升级项目的预算申报即可达 100 亿美元。 --- ### [49:48](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2988s) 蓝领就业繁荣、百座数据中心阻击战与自备能源（BYOE） Sacks 反驳了“数据中心不创造就业机会”的流行观点，强调数据中心建设掀起了蓝领行业前所未有的就业狂潮。由于生成式 AI 带来的巨额资本开支（CapEx）绝非一两年的短期风口，正如英伟达首席执行官黄仁勋所预测的那样，这一建设浪潮可能会持续十到二十年。在施工现场，电工、木工、泥水工、混凝土浇筑工、铺路工和设备安装工的需求爆发，不仅创造了数以万计的高薪蓝领工作，还将电工等核心工种的工资推高了 25% 至 30%。 然而，现实阻力依然严峻。全美目前有约 100 个数据中心项目正遭到地方层面的激烈抵制，历史数据显示，每 100 个受争议的项目中约有 40 个最终被迫取消；今年此类受阻项目的取消率已较去年翻了一倍以上，这批搁浅项目背后所对应的经济总价值高达约 1620 亿美元。面对地方电网受限与抗议，数据中心行业正在快速转向“自备能源”（BYOE, Bring Your Own Energy）模式。例如 Crusoe 的 Chase 与 CoreWeave 的 Michael 在访谈中均提及，他们不再完全依赖公共电网，而是将天然气发电、柴油机组与太阳能设施直接带到现场，构建离网自给系统。 - **薪资涨幅**：数据中心密集建设推动电工、木工等蓝领建筑工种薪资上涨 25% 至 30%。 - **受阻项目规模**：全美目前约 100 个数据中心面临诉讼与抵制，涉及总投资额高达 1620 亿美元，面临取消的项目占比约为 40%。 - **核心模式**：自备能源模式（BYOE）通过现场接入天然气、柴油和光伏，成为化解电网容量瓶颈的主流方案。 --- ### [51:31](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3091s) 海湾国家算力地缘战、虚假指控与 AI 行业的公关危机 针对那些以本土能源紧缺为由反对数据中心的言论，Sacks 指出逻辑上的矛盾：如果反对者并非单纯反技术进步，理应支持美国在能源富集盟友国建设数据中心。Sacks 回忆起自己一年前曾极力推动海湾阿拉伯国家合作委员会（GCC）盟国采用美国顶尖技术建设数据中心，当时却遭到了铺天盖地的批评与阻挠，反对者构陷该项目“服务于中国利益”。然而事实是，伊朗伊斯兰革命卫队（IRGC）已正式将这些美方及海湾盟友合建的数据中心列为威胁摧毁的重点资产清单，甚至发动了导弹袭击；如果真如外界诬蔑的那样有利于中国，绝不可能沦为伊朗的袭击目标。Sacks 直言，诸如“数据中心消耗过多水资源”以及阻挠 GCC 算力合作等论调完全是由特定势力精心策划的虚假信息战（Psyop），并点名指出 Anthropic 等机构曾强硬游说、全力阻挠中东数据中心合作。 对话随即引申到美国 AI 行业在公众沟通上的巨大溃败。与中国全社会对 AI 的普遍正面预期形成鲜明对比的是，目前 AI 在美国公众眼中的支持率已经跌到与 ICE（美国移民及海关执法局）、民主党建制派甚至伊朗政府同等低迷的水平。其根源在于西方 AI 行业树立的两位核心发言人形象极具负面色彩：一位是 Anthropic 的 Dario Amodei，整天向公众宣扬“末日论”，声称未来所有人都会失业、所有账户与数据都会被黑客攻破；另一位则是 OpenAI 的 Sam Altman，刚被调查记者 Ronan Farrow 写了一篇长达 7 万字的深度特稿曝光，文中多名知情人士直言 Altman 具有“反社会人格”（sociopath）。嘉宾们呼吁，AI 行业必须将聚光灯转移到黄仁勋（Jensen Huang）、埃隆·马斯克（Elon Musk）或迈克尔·戴尔（Michael Dell）身上，真正专注于利用 AI 解决民众关切的“三座大山”——医疗、住房与教育（例如 Joe Lonsdale 支持的 Alpha School 的成功教育实践）。 --- ### [54:37](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3277s) 互动竞猜秀：“估值太离谱”第一轮，130亿美元的猴子头像 节目随后切入轻松的恶搞游戏环节——“估值太离谱”（The Price is Wrong），主题是竞猜那些估值严重虚高后破裂的灾难级初创企业。Jason 担任主持人，伴随着标志性的综艺音效，首先邀请来自得克萨斯州奥斯汀、号称“PayPal 黑手党执法官”的 David Sacks 登台答题。 第一道抢答题的线索极其直白：**“这家初创公司估值一度超过 130 亿美元，其核心商业模式是用虚假的互联网货币贩卖 JPEG 图片，当时全网失心疯般地斥巨资狂买各种猴子头像。”** Sacks 思考良久，嘴里念叨着“无聊猿”（Bored Apes），但迟迟想不起平台的具体名字，卡在关键节点无法作答。一旁的 Chamath 跃跃欲试、不停高喊抢答。主持人读秒倒计时结束后将机会给到 Chamath，Chamath 瞬间报出正确答案——**OpenSea**，成功斩获 100 分。 随后 Jason 调侃 Chamath 出场：“这位让‘独裁者’（Dictator）一词名副其实、来自加州帕洛阿尔托的选手，也是大家最喜欢的 DoorDash 外卖骑手。”在长达九年的兼职送外卖玩笑中，Chamath 打趣自曝外卖潜规则：“我送餐通常偷吃四根薯条，有时还要舔一下卷饼。”荒唐的发言引得全场哄笑，甚至借机感慨：“这就是为什么我们急需配送机器人上岗，没人想吃被外卖员舔过的卷饼。” - **涉案标的**：NFT 交易平台 OpenSea，估值峰值曾突破 130 亿美元。 - **竞赛战况**：Sacks 误答为无聊猿未得全名，Chamath 抢答 OpenSea 成功先拔头筹。 ### [56:48](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3408s) 《价格猜错啦》第一轮收官：Clubhouse、Juicero 与失败独角兽的宿命 演播室里的竞赛气氛愈发热烈，主持人 Jason Calacanis 继续抛出下一个创业公司谜题：“这家初创公司估值曾高达 40 亿美元，在 COVID 疫情期间让人们可以通过移动设备语音畅聊，还能收听各种中庸平庸的 VC 高谈阔论。” 还没等提示完全念完，Chamath Palihapitiya 便抢答高呼：“天哪，是 Clubhouse！” Chamath 凭借敏锐的记忆力再斩获 200 分，状态火热。众人感叹互联网的遗忘速度如此之快，昔日红极一时的明星独角兽一旦失败，就迅速从公众记忆中彻底蒸发进虚无之中。 随后，镜头切给了远在爱达荷州土豆田秘密基地的 David Friedberg。Jason 自封为“幽默的杰克”（Jocular Jake / Jocular Calcaneus），打趣调侃起 Friedberg 的土豆科研日常，追问他最喜欢吃哪种土豆，是扇贝焗土豆、浓郁芝士泥还是普通土豆泥。调侃过后，Jason 为这位纯素食主义者出题：“这家初创公司曾估值 2.7 亿美元，专门售卖放入榨汁机里的果汁包装包，被称为‘果汁界的 iPhone’。” 听到提示后，Friedberg 激动得接连大喊了十多声“我知道！选我！”，最终脱口而出准确答案：“Juicero！” 尽管被众人起哄怀疑偷偷低头查了手机，Friedberg 还是为自己赢回了 100 分。 最终，Chamath 夺得本轮冠军。Jason 为他颁发了一连串恶搞大奖：前往墨西哥卡波的成人度假村“Temptations 2”三人同行游（调侃其“三人同居/thrumble”），或者一套附带纽约市新出台巨额税单与 DoorDash 外卖兑换券的第二居所（pied-à-terre）。在一阵爆笑中，大家一致认为这个回顾破产独角兽的游戏必须成为播客的固定常设环节。 - **Clubhouse**：巅峰估值 40 亿美元，凭借疫情期间的纯语音社交和创投圈大佬开房布道风靡一时，如今热度已烟消云散。 - **Juicero**：融资数亿美元、估值达 2.7 亿美元的冷压榨汁机项目，后被彭博社曝出其昂贵的专属果汁包用双手直接挤压就能出汁，迅速沦为硅谷科技泡沫的经典笑柄。 --- ### [59:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3563s) 埃里克·斯沃韦尔爆雷退选：华盛顿丑闻与被精心雪藏的黑料 欢声笑语过后，话题急转直下，切入政坛重磅丑闻。素来沉闷的华盛顿特区近期桃色新闻频发，就连 TMZ 都特意在 DC 开设了记者站。加利福尼亚州联邦众议员埃里克·斯沃韦尔（Eric Swalwell）突然深陷严重的性丑闻泥潭，不仅退出了加州州长竞选，还被迫辞去国会众议员职务。虽然在法律层面未定罪前仍推定无罪，但其政治生涯显然已遭受毁灭性打击。 作为播客里的“调查记者”，David Friedberg 分享了一段耐人寻味的内幕爆料。早在去年 12 月斯沃韦尔刚传出有意角逐加州州长时，出于对加州未来命运的关切，Friedberg 曾致电加州政界的各方知情人士打探情况。多位互不关联的独立消息源当时就明确告诉他：“别看好这家伙，他私底下给多名女员工发送不雅私密照，身上有一大堆见不得光的黑料迟早要爆出来。” 当时 Friedberg 并没有全信，甚至一度将其当成竞争对手为阻击其参选而编造的政治谣言，因为他理所当然地认为，如果这些恶劣行径确凿属实，早就该被受害者或媒体公之于众了。 ```mermaid flowchart TD accTitle: 斯沃韦尔丑闻曝光与民主党机器运作逻辑 accDescr: 展示丑闻从私下流传到集中爆发，再到民主党因选情考量进行干预的因果链路 subgraph Timeline ["丑闻时间线"] A["2025年12月：斯沃韦尔谋划参选加州州长"] --> B["政界私下广泛流传不雅照与性骚扰黑料"] B --> C["黑料被刻意压制并雪藏长达数月"] C --> D["2026年4月：全网集中爆发协同爆料"] end subgraph PartyMachinery ["民主党选情考量与干预"] E["加州全开放初选（Jungle Primary）选票极度分散"] F["两名共和党候选人（如 Steve Hilton）民调均达14%-15%领跑"] G["民主党面临决选被两名共和党通吃的灭顶危机"] E & F --> G end Timeline --> H{"高层政治决断"} PartyMachinery --> H H --> I["提前引爆脓包（Lance the Boil）逼退斯沃韦尔"] I --> J["辞去国会职务并彻底退出州长竞选"] ``` 令 Friedberg 深感震惊的是，在整整四五个月的时间里，这个圈子里的许多人对这些劣迹心知肚明，受害者与知情者却默契地集体保持沉默，直到某一个精确的时间节点，所有猛料才如同经过严密策划一般排山倒海般协同释放。这引发了极其残酷的灵魂拷问：那些掌握黑料的政客为何不在第一时间为受害者主持公道？为何要特意将受害者的痛苦作为政治筹码扣在手里，非要等到最能产生致命杀伤力的选举关头才抛出？ --- ### [01:03:00](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3780s) 民主党机器的“刺破脓包”战术：从加州初选危机到拜登退选既视感 面对是谁在幕后操盘的疑问，众人半开玩笑地提到了南希·佩洛西（Nancy Pelosi），但随后讨论深入到了民主党党内机器的生存法则。嘉宾们指出，民主党体制内存在极深的利益绑定与圈内人机制，该党本质上是一个庞大的利益分配机器，首要目标是维护源源不断流向党内利益集团的利益链条（"gravy train"），绝不允许任何局外人或不可控因素砸了基本盘。党内当权派真正青睐的或许是像凯蒂·波特（Katie Porter）这样根正苗红的全国及加州圈内人。 加州特殊的选举制度是引发这场政治处决的直接诱因。加州实行独特的“全开放丛林初选”（Jungle Primary），得票排名前两名的候选人无论党籍直接进入决选。当时民主党阵营候选人过多，导致党内选票被严重稀释分流；反观共和党阵营，史蒂夫·希尔顿（Steve Hilton）等人各自拥有约 14% 到 15% 的稳定支持率，牢牢占据民调前列。如果照此发展下去，决选甚至可能出现“两名共和党人对决”的滑稽灾难。民主党当权派不仅需要迅速精简候选人队伍，更深知一旦斯沃韦尔侥幸闯入决选，共和党在正赛阶段砸出这些致命丑闻，民主党将毫无招架之力而丢掉加州帅位。 为了防患于未然，民主党高层做出了“刺破脓包”（"lance the boil"）的决断，在私下劝退未果后，迅速祭出黑料将其斩落马下。这种冷酷而高效的党内清洗，让嘉宾们瞬间联想到了 2024 年拜登被迫退选的经典一幕：在灾难性的总统辩论前，主流媒体与建制派统一口径盛赞拜登“思维极其敏锐”（"sharp as a tack"）；而辩论翻车当晚，民主党内部短信漫天飞舞，立场在数小时内瞬间逆转，随后佩洛西等党内巨头直接上门逼宫，逼迫拜登为了政党利益体面出局。政治机器在面对存亡危机时展现出的无情与决绝，在斯沃韦尔身上再次得到了教科书般的重演。 ### [01:05:37](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3937s) 民主党机器的铁腕操盘：从逼退拜登到斯瓦尔韦尔出局 萨克斯（David Sacks）回顾了民主党高层在幕后运作政局时那套冷酷而高效的手腕。他提到当初民主党高层劝退拜登时开出的条件极其直白：“我们可以用体面的方式解决，也可以用强硬的方式解决（We can do things the hard way or the easy way）。”结果几天前还在模仿《华尔街之狼》经典台词宣称“他们休想让我离开”的拜登，在神秘消失了一周后，突然就通过推特发布了退选声明，甚至签名都疑似是用自动签名机（auto pen）签署的。整个过程充满戏剧性，明眼人都看得出拜登是被党内大佬硬生生架出局的。而在这两起重大退选事件的暴风眼中心，始终伫立着同一个核心操盘手——南希·佩洛西（Nancy Pelosi）。 佩洛西不仅是党内操盘者，更是埃里克·斯瓦尔韦尔（Eric Swalwell）长达二十年的政治领路人。早在约二十年前，正是佩洛西发掘了斯瓦尔韦尔并支持他进军国会；即便后来斯瓦尔韦尔卷入与涉嫌中国间谍方芳（Fang Fang）的丑闻、共和党人全力试图将他逐出众议院情报委员会时，也是佩洛西力排众议院阻力保全了他的政治前途。因此，萨克斯直言，党内绝对不可能在没有获得佩洛西首肯的情况下擅自按下彻底毁灭斯瓦尔韦尔的“核按钮”。斯瓦尔韦尔的结局与拜登如出一辙，党内大佬同样递给了他那道选择题，但他“蠢到听不进劝告”，最终招致了最残酷、身败名裂的“强硬解决手段”。 - 斯瓦尔韦尔因涉嫌性丑闻等负面风波被迫退出加州州长竞选，并宣布辞去国会众议员职务。 - 佩洛西在民主党内部扮演着终极政治“教父”角色，无论涉及总统换将还是议员断送仕途，均需其点头首肯。 --- ### [01:07:16](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4036s) 国会山“新股神”横空出世：从佩洛西到罗·康纳的交易神话 话题随后转向国会议员们惊人的炒股表现，现场气氛立刻变得充满调侃与讽刺。主持人们先是赞叹佩洛西长期以来作为“传奇日内交易员”的无敌战绩，接着又爆出播客老友、加州国会议员罗·康纳（Ro Khanna）的惊人近况：罗·康纳在过去一年的股票交易额竟然高达 6 亿美元，其交易频率甚至超过了全球顶级量化做市商 Citadel Securities，今年以来的投资收益率甚至压过了佩洛西一头。查马斯（Chamath Palihapitiya）打趣道，如果罗·康纳哪天被选民赶出国会，自己随时准备在 Alt Capital 给他留一个董事总经理（MD）的职位。众人更是哄笑称佩洛西很快就会在暗中给罗·康纳“捅上一刀”，因为他发财的内幕交易消息居然没有提前向佩洛西报备。 萨克斯借此剖析了自己曾经支持罗·康纳的原因以及如今政坛环境的巨变。当初萨克斯欣赏罗·康纳，是因为他是少数在推特言论自由事件中公开发声支持言论自由的民主党人，并且是进步派党团中唯一主张通过外交途径解决俄乌冲突的清醒之声。尽管罗·康纳曾主张征收富人财富税，但萨克斯当时认为这只是一种无关痛痒、不切实际的政治姿态。然而随着近年来政治“奥弗顿之窗（Overton window）”的大幅剧烈漂移，原本荒谬的财富税如今竟成了可能落地的真实提案，两人的政治距离也随之拉开。 谈到议员炒股与投资巨头巴菲特（Warren Buffett）的对比，查马斯为自己此前对巴菲特战绩的分析进行了辩护。他指出巴菲特的投资回报在 2000 年美国证券交易委员会颁布《公平披露规则》（Regulation FD）之前和之后呈现出极端的“双峰分布”：在没有严厉信披法规限制的蛮荒时代，巴菲特的收益率高到破表；而一旦市场必须遵循公平信披规则，包括巴菲特在内的所有正规投资人收益率全部被大幅压制。相比之下，国会议员却享有着不受 Reg FD 约束的法外特权，他们可以在委员会听证会甚至是机密信息保密室（SCIF）里获取第一手机密内幕后，第一时间跑出来实时下单交易，这才是所谓“战胜巴菲特”背后的残酷真相。 - 罗·康纳被爆出在公开披露中有高达 6 亿美元的频繁股票交易纪录，引发舆论对国会议员内幕交易的广泛声讨。 - 《公平披露规则》（Reg FD）严禁上市公司向特定投资机构提前泄露非公开信息，但美国国会议员在履职过程中获取的情报并不受该法规的直接约束。 --- ### [01:10:31](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4231s) 极度分裂的市场指标：巴菲特的 3000 亿现金与离散行情 查马斯随后将视角转向宏观资本市场的异常混乱态势。从传统经典的估值指标来看，多项历史关键数据均已拉响警报：希勒席勒市盈率（Shiller P/E）目前处于历史极值高位，衡量股市总市值与 GDP 比例的“巴菲特指标（Buffett Indicator）”同样触顶。如果单看这些指标，市场俨然已经来到了见顶回调的前夜。然而，让华尔街感到无比棘手的是，当前美股出现了极度的“离散化（dispersion）”行情：实际上只有八到九家科技巨头在不知疲倦地狂拉指数刷新历史新高，而市场中剩下的大多数普通企业股价却表现平平，并未享受到牛市红利。 这种两极分化的割裂行情让许多价值投资者感到迷茫。查马斯指出，伯克希尔·哈撒韦账面上如今趴着超过 3000 亿美元的惊人巨额现金资产。尽管如今的伯克希尔在运营层面已不再全由巴菲特一人事必躬亲，但面对席勒指标和自身指标的双双爆表，巴菲特团队选择坚守自己熟悉的投资铁律，任凭数千亿美元现金闲置吃利息也绝不盲目追高入场。这一庞大的现金储备释放出一个强烈而清晰的信号：在巴菲特的眼中，当前市场上根本找不到具有性价比的真正投资机会，他们正在耐心等待一场全市场级别的估值均值回归与暴跌回调。 - 伯克希尔·哈撒韦现金储备达到创纪录的约 3000 亿美元，凸显顶级价值投资机构在泡沫估值下的防守姿态。 - 市场广度极其狭窄，指数的上涨几乎完全由少数几家龙头科技巨头（约 8-9 家）驱动，多数股票处于落后状态。 --- ### [01:12:08](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4328s) 伊朗战事第 7 周股市创新高：伊斯兰堡谈判与“和平溢价” 杰森抛出了一个令许多投资者大惑不解的矛盾现象：美国对伊朗展开军事行动已经进入第 7 周，国会和政府在这次中东军事行动中已经砸进去了高达 1000 亿美元的真金白银，然而美股大盘不仅没有发生预想中的恐慌性崩盘，反而在近期将所有战争跌幅彻底抹平，并在周三和周四接连创下历史新高。杰森询问萨克斯和特拉维斯·卡兰尼克（Travis Kalanick），资本市场为何能如此轻松地“对战争耸耸肩”并一路狂飙。 萨克斯给出了极其清晰透彻的地缘政治与金融逻辑。他指出，随着美伊双方代表在伊斯兰堡（Islamabad）举行会谈，市场已经开始全面定价战争即将迎来和平解决的前景。特朗普总统近期亦公开明确表示军事行动的目标已基本达成，战争即将接近尾声，并将其定性为一次短期的“涉险行动（excursion）”。萨克斯分析称，虽然伊斯兰堡会谈并未当场签署和平条约，但要求两个交火数周、且在过去近半个世纪中彼此敌对的国家在 24 小时内彻底化解所有恩怨本就是不切实际的幻想。股市作为“终极预测市场”，敏锐地捕捉到了局势缓和的确定性动向： - 美股市场在周二就已经收复了自对伊开战以来的全部失地。 - 周三股指强势刷新历史高位，周四更进一步创下盘中崭新纪录。 萨克斯严谨地做出免责声明，强调自己虽然与政府关系密切，但绝非代表官方表态，也未掌握任何公众所不知道的机密情报，这纯粹是作为一个观察者对市场交易行为与总统公开言论的理性解读。紧接着，杰森打趣地引出了特拉维斯在西洋双陆棋（backgammon）比赛中狂胜自己时提出的全新政治假说：这并非特朗普惯常被嘲弄的“临阵退缩（Trump always chickens out）”，而是特拉维斯独创的“特朗普永远在乎公关形象（Trump always cares about optics）”理论。 ```mermaid flowchart TD accTitle: 伊朗冲突下的美股市场定价传导机制 accDescr: 展示从伊斯兰堡外交接触到股市创历史新高的逻辑链条 subgraph 地缘与外交信号 A["美伊双方在伊斯兰堡接触"] --> B["特朗普公开表示军事目标接近完成"] B --> C["官方定调冲突为短期涉险行动 (Excursion)"] end subgraph 市场预期与行情反应 C --> D["市场将股市视为终极预测市场"] D --> E["全面定价战争和平解决预期"] E --> F["周二抹平开战以来的全部跌幅"] F --> G["周三及周四接连刷新历史新高"] end ``` ### [01:14:29](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4469s) 特朗普的风向标与股市博弈：高波动的“扑克局” 特拉维斯·卡兰尼克（Travis Kalanick）提出了一个俯瞰全局的宏观视角，他将股市比作唐纳德·特朗普（Donald Trump）的“风向标”。尽管当前市场波动率上升、VIX 指数高居不下，但标普 500 指数实际上在一个相对狭窄的区间内震荡运行，展现出了在各种极端政策噪音下异乎寻常的韧性。在特拉维斯看来，特朗普的政策操盘非常在意股市表现，绝不会容忍标普 500 指数跌破底线；他的惯用套路是先通过激进的言论和动作制造恐慌、打破僵局，随后回归务实并推动谈判落地，而市场交易员们如今已经逐渐摸清并消化了这种规律。 杰森·卡拉卡尼斯（Jason Calacanis）随即打趣，这种风格就像在牌桌上与查马斯·帕利哈皮蒂亚（Chamath Palihapitiya）或者知名牌手艾伦·基廷（Alan Keating）打德州扑克：你看着他几乎每一手牌都在玩，疯狂输钱，让人困惑“这家伙到底怎么还没破产”，结果到了深夜牌局行将结束时，他突然在两个超大底池连续撞出绝杀顶牌（the nuts），一把彻底翻盘摆脱困境。 ```mermaid flowchart TD accTitle: 特朗普政策与股市互动的心理博弈循环 accDescr: 展示特拉维斯所描述的特朗普制造市场震荡与拉回实操的循环过程。 subgraph Cycle ["特朗普的风向标循环"] A["激进政策表态 / 制造紧张局势"] --> B["市场波动率飙升 / 恐慌蔓延"] B --> C["股市逼近承受底线"] C --> D["回归务实谈判 / 政策修正与利好出台"] D --> E["市场回暖反弹 / 确立新估值区间"] E --> A end ``` ### [01:16:03](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4563s) 矛盾的市场信号：估值见顶还是牛市暴冲？ 查马斯随后展示了一组相互矛盾的市场宏观数据。一方面，席勒市盈率（Shiller PE）目前正徘徊在历史高位附近；衡量美股总市值与美国 GDP 之比的“巴菲特指标”同样打破了历史极值。从历史估值周期来看，这两大指标都在明确提示投资者应该采取偏防御的“避险”（risk-off）姿态。 但令分析师极为困惑的是，另类技术与动量信号却在指向截然相反的方向：根据历史离散度数据显示，一旦标普 500 在 4 月上半月实现 5% 以上的涨幅，全年剩余时间的平均涨幅往往高达近 32%，而今年前几个月大盘本身就已经累计上涨了约 7.5%。查马斯直言，现在的市场处于一个极度割裂的状态，投资者可以在海量数据中轻松找到任何一条来支撑自己的既有偏见。出于对极高估值的谨慎，查马斯本人倾向于全面收缩风险敞口，并急切等待大体量独角兽上市以降低杠杆套现离场： - **SpaceX IPO**：查马斯认为其尽早公开上市至关重要，是帮助市场锁住流动性与减仓套现的核心窗口。 - **Anthropic 与 OpenAI 的抢跑之争**：两家顶尖大模型实验室中必须有一家率先抢跑完成上市，抢在估值泡沫破裂前锁定资金。 ### [01:17:37](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4657s) AI 效率神话 vs. 商业变现实质：初创爆发还是玩具狂欢？ 杰森对市场的乐观情绪提出了完全不同的依据。他认为聪明的长线资金看好的是企业全面应用 AI 后的超级盈利潜力：过去的微软 Office 办公软件提升了约 30% 的效率，互联网革命带来了约 50% 的生产力提升，而当今顶尖公司的优秀员工在无限算力加持下，效率正在提升 10 倍、20 倍乃至 30 倍。杰森在自己投资的 600 家初创公司中观察到，虽然目前只有 10% 到 20% 的团队真正精通如何驾驭模型，但那些正确落地的企业正全面碾压竞争对手。他以两家公司为例： - **MicroOne**：专注为大语言模型构建暗池数据与语料采集技术，业务爆发式增长。 - **TaxGPT**：专注于财税会计垂直场景，已经渗透并占据全美约 6% 至 7% 的注册会计师工作流。 查马斯对这一观点进行了极其锋利的驳斥。他坦言自己掌管的业务至今还没有看到 AI 带来哪怕一丝“利润或收入的海啸式增长”。查马斯强调，小公司在边缘细分赛道的小打小闹根本无法撑起数万亿美元的超级市值；大企业之所以迟迟无法大规模部署 AI，并非因为决策者愚蠢，而是因为大型跨国实体的业务流程远比初创企业复杂精密千百倍。查马斯尖锐地总结道：“如果你不能证明这项技术可以在主流的、顶级联赛级别的关键业务场景中跑通，它就依然只是个玩具。” ### [01:20:30](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4830s) 模型层 ROI 显现与企业级落地困局 大卫·萨克斯（David Sacks）在双方争论中充当了裁判。他坦言自己在落地现状上更偏向杰森，承认企业级组织变革管理难度极大，麦肯锡等机构的调研也印证了目前大量大型企业的生成式 AI 转型项目均以失败告终。然而，自下而上的技术渗透却极为迅猛，尤其在代码编写领域，过去几个月编程模型的迭代带来了前所未有的指数级爆发增长。这直接扭转了此前的市场担忧： - **模型层 ROI 突破**：过去市场普遍质疑巨头在数据中心与算力上的天量资本支出（CapEx）是无底洞泡沫，因为模型层缺乏商业回报；但目前编码模型等垂直能力的爆发，已经让模型层实现了切实的商业回报与指数级收入。 - **应用层 ROI 拷问**：当前的疑问被顺延到了应用层，即大型软件与应用生态能否将这些底座能力转化为真实的组织净利润。 查马斯随后重申了自己的警惕：回顾移动互联网浪潮，消费端应用的爆发极为迅速，谷歌和 Meta 在成立早期就展现出极其恐怖的造血能力与巨额净利润；但反观生成式 AI，如果全行业一致认定“万亿美元估值的底座必须由企业级付费来支撑”，那么至今为止市场上竟然找不出哪怕一家产生可观利润规模的标杆企业。萨克斯最终也赞同了查马斯对高估值的担忧，表示即便不考虑突发地缘事件，仅凭席勒市盈率等估值指标，投资者在当前点位采取更为保守防御的姿态也是非常理性的选择。 ### [01:23:10](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4990s) 传统企业的变革阻力与“自主企业”的现实鸿沟 关于 AI 能否拯救传统企业并维持估值增长，讨论聚焦到了极其骨感的落地细节上。传统企业的估值往往被市场压得很低，如果 AI 转型能够成真，上行空间固然诱人，但正如席间强调的，“细节才是关键”。要把一个盘根错节、流程繁琐的传统实体业务在一夕之间改造成所谓的“自主企业（autonomous enterprise）”，远比动动嘴皮子困难得多。 ```mermaid flowchart TD accTitle: AI 企业落地现实与两极分化 accDescr: 展现传统企业因流程与管理层阻力导致 AI 落地困难，与创始人领导的科技公司快速敏捷落地的对比 subgraph Legacy["传统企业困境"] A["传统复杂企业"] --> B["官僚机构与中间管理层阻力"] B --> C["未文档化的隐性业务流程"] C --> D["变革管理推进缓慢"] end subgraph Tech["硬核科技公司"] E["创始人领导的科技公司"] --> F["全员转向拥抱 AI 开发"] F --> G["研发与部署节奏翻倍提升"] end ``` Travis Kalanick 对此给出了总结：在大型机构中，推行自动化最大的障碍其实是“变革管理（change management）”。这本质上不是技术问题，而是人事问题——那些坐在位置上的中间管理层、技术官僚和各级雇员，以及企业内部大量甚至从未形成书面文档的复杂工作流。理论上所有人都在谈论 AI 颠覆一切能有多快，但在实际操作中，推动这群人改变既有习惯是一项极其艰巨的工程。 - 传统企业当前往往面临严峻的估值折价，虽然理论上引入 AI 的改善空间极大，但实际改造难度被外界严重低估。 - 阻碍传统企业 AI 落地的核心痛点在于“官僚体系（bureaucrats）”与人际协同，海量关键业务细节依赖未经文档记录的人工作业。 --- ### [01:24:24](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=5064s) 科技公司的敏捷红利与大言不惭的 AGI 泡沫 与传统企业的举步维艰相反，在真正由创始人领导的硬核科技公司和上市新贵中，AI 已经带来了立竿见影的质变。Travis 透露，他与多位一线科技公司的创始人兼 CEO 交流时，大家的反馈惊人一致：通过将公司文化全面转向“以 AI 为核心的开发模式”，产品开发节奏、迭代部署以及新功能的推出速度都得到了成倍的飞跃，这种生产力红利是毋庸置疑的真实突破。 然而，Travis 与众人也直言不讳地戳破了某些人宣扬的“AGI 已经降临”的虚妄叙事，认为这不过是某些利益相关方为了自身利益在“王婆卖瓜（selling their book）”。只要真正亲手做过 Agent 开发的人都心知肚明，目前的 AI Agent 根本称不上聪明。Travis 分享了自己的“副线投资任务”：他开发了一批自主 Agent 尝试在 Kalshi、Polymarket 等预测市场上进行自动化下注与投资，结果发现这些 Agent 表现极其滑稽——团队甚至不得不花大量时间去教导它一个极其基本的常识：如果你想投资赚钱，绝不能在同一个赌局的正反两面同时下注。目前的 Agent 毫无审美与品味（taste），一旦无人看管就会迷失在任务细节的森林里，现阶段绝对离不开“人机协同（Human-in-the-loop）”。 - 创始人主导的科技公司通过 AI 重构研发流程，显著缩短了版本发布和功能迭代周期。 - 预测市场实测印证当前的 Agent 仅能应付重复性任务，完全无法执行复杂的全新策略，甚至连基本的套利与对冲逻辑都会发生逻辑自相矛盾。 - 全场一致认为高呼 AGI 已至纯属言过其实的炒作。 --- ### [01:26:31](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=5191s) 趣味加时赛：竞猜估值爆雷企业与剧终谢幕 在播客接近尾声之际，Jason 专门为 Travis 加设了估值竞猜游戏“The Price is Wrong”的加时赛（Bonus Round），并顺带对 Travis 展开了一番无厘头的恶搞介绍，戏称他是来自佛罗里达州塔拉哈西的一名“专业水上运动选手”，平日打打零工补贴家用，周末则在湖面教滑水。在一片欢笑声中，比赛正式展开。 加时赛题目均瞄准了科技圈历史上估值被严重扭曲的灾难性泡沫： 1. **第一题**：某初创公司曾以 90 亿美元估值吸纳了 9 亿美元融资，随后在 2018 年彻底清盘解散，其以低沉嗓音闻名的女性创始人兼 CEO 目前正在联邦监狱服 11 年有期徒刑。Travis 迅速抢答出著名的血液检测骗局“Theranos”，轻松斩获 100 分。 2. **第二题（全员抢答决胜局）**：某主打移动端优先的短视频流媒体平台，从全硅谷顶级投资机构手中卷走 17 亿美元巨额融资，却在上线短短 6 个月内就宣告倒闭。David Sacks 和 Chamath 展开激烈争夺，Chamath 先是脱口喊出创始人 Jeffrey Katzenberg 的名字，随后精准抢答出“Quibi”，成功锁定胜局。 赢下比赛的 Chamath 获得了主持人调侃的“终极大奖”——一套位于奥斯汀东区靠近机场的单身公寓（pied-à-terre），被全场戏谑该地段鱼龙混杂，适合寻欢作乐。节目的最后，几位主持人顺势公布了线下活动的进展：Chamath 主办的 Liquidity 闭门会议 500 个名额早已被抢购一空，现场商讨是否追加 50 到 100 个转播席位；同时提醒听众尽早前往 allin.com 抢购即将售罄的 2026 年 All-In 峰会门票。节目在全场招牌式的片尾混剪金句与互损欢笑声中正式收尾。 - **Theranos**：累计融资 9 亿美元，峰值估值 90 亿美元，2018 年破产清算，创始人伊丽莎白·霍姆斯（Elizabeth Holmes）获刑 11 年。 - **Quibi**：由 Jeffrey Katzenberg 牵头、募集 17 亿美元的短视频流媒体先驱，问世仅 6 个月便轰然倒闭。 - **线下活动动态**：Liquidity 闭门名额达到 500 人上限，团队考虑增设分会场；All-In Summit 门票持续热销，提醒听众避免因延误而提前数月售罄。 ## Notable quotes > [09:09](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=549s) 在蓝州，你的财产根本不安全，那里的政治阶层总觉得自己有权分走一杯羹。 — David Sacks > [11:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=683s) 你手握 ChatGPT 这样一个拥有 10 亿用户、年增速高达 50% 到 100% 的业务，为什么还要跑去死磕企业级和代码？这简直是一家毫无重心的公司。 — 投资人 (引述) > [15:23](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=923s) 增长就是一切。运营效率最终必然碾压资本补贴。 — Travis Kalanick > [21:14](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=1274s) 如果你们真打算花几亿买个科技播客，我们这帮人一直都在这儿，Dario 赶紧联系我们，直接给我们把票验了！ — David Sacks > [38:04](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2284s) 最恶劣的监管俘获。 — David Sacks > [45:44](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=2744s) 超级富豪的神庙，是科技精英与建制派亿万富豪从底层敛财、把自己送上太空、却将普通人远远抛在身后的核心运作机器。 — David Friedberg > [01:05:37](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=3937s) 我们可以用体面的方式解决，也可以用强硬的方式解决。 — David Sacks > [01:17:37](https://www.youtube.com/watch?v=SFdqX7IY7RY&t=4657s) 如果你不能证明这项技术可以在主流的、顶级联赛级别的关键业务场景中跑通，它就依然只是个玩具。 — Chamath Palihapitiya ## Mentioned Travis Kalanick, David Sacks, Jason Calacanis, Chamath Palihapitiya, David Friedberg, Zohran Mamdani, Ken Griffin, Sam Altman, Denise Dresser, Peter Steinberger, Dario Amodei, Elon Musk, Marc Andreessen, Nancy Pelosi, Eric Swalwell, Steve Hilton, Katie Porter, Ro Khanna, Warren Buffett, Donald Trump, Jensen Huang, Michael Dell, Joe Lonsdale, Jeffrey Katzenberg, Elizabeth Holmes, Uber, OpenAI, Anthropic, ChatGPT, Codex, Claude, Gemini, Perplexity, xAI, Colossus, Meta, Google, DeepMind, Amazon, Microsoft, Tesla, SpaceX, Cursor, Allbirds, Bird, Bloom Energy, Jane Street, Crusoe, CoreWeave, OpenSea, Clubhouse, Juicero, Theranos, Quibi, MicroOne, TaxGPT, Berkshire Hathaway, Polymarket, Kalshi ## Video description (0:00) Bestie intros: Travis Kalanick joins the show! (0:42) Mamdani taxes the rich: new pied-à-terre tax coming to NYC (11:23) OpenAI's identity crisis: Leaked memo, enterprise pivot, Anthropic valuation flip (27:28) Big Tech's compute dominance: How will this impact frontier labs? (33:53) Allbirds stock up over 400% on AI pivot, and what's behind the datacenter disaster? (54:37) The Price is Wrong: Name That Overvalued Startup ROUND 1 (59:23) Eric Swalwell drops out of CA governor race and resigns from Congress: Who knew what and when? Impact of the Democratic Machine (1:10:31) State of the market: Confusing indicators, pricing in Iran peace, AI's mixed results, inflated valuations (1:26:31) The Price is Wrong: Name That Overvalued Startup ROUND 2 Apply for Summit 2026: https://allin.com/events Follow Travis: https://x.com/travisk Follow the besties: https://x.com/chamath https://x.com/Jason https://x.com/DavidSacks https://x.com/friedberg Follow on X: https://x.com/theallinpod Follow on Instagram: https://www.instagram.com/theallinpod Follow on TikTok: https://www.tiktok.com/@theallinpod Follow on LinkedIn: https://www.linkedin.com/company/allinpod Intro Music Credit: https://rb.gy/tppkzl https://x.com/yung_spielburg Intro Video Credit: https://x.com/TheZachEffect Referenced in the show: https://www.texastribune.org/2025/01/22/austin-texas-rents-falling https://www.cnbc.com/2026/04/13/openai-touts-amazon-alliance-in-memo-microsoft-limited-our-ability.html https://polymarket.com/event/gpt-5pt5-released-on https://www.ft.com/content/04ac7917-940b-4606-be5f-9eb895a7d982 https://x.com/RihardJarc/status/2044725910566220095 https://x.com/RoundtableSpace/status/2041897604708233650 https://x.com/chamath/status/2043811625967530056 https://x.com/pmarca/status/2042742413098450998 https://www.reuters.com/business/allbirds-shares-jump-over-400-plans-pivot-ai-sneakers-2026-04-15 https://www.wsj.com/business/data-centers-are-a-gold-rush-for-construction-workers-6e3c5ce0 https://x.com/zerohedge/status/2043520934930239860 https://x.com/APompliano/status/2041946174677217733 https://x.com/IAPolls2022/status/2041589297489305717 https://www.tmz.com/2026/04/13/tmzdc-staff-starts-today https://www.nbcnews.com/politics/elections/eric-swalwell-drops-bid-california-governor-sexual-misconduct-allegati-rcna277009 https://x.com/great_martis/status/2044454944808571015 https://x.com/GlobalMktObserv/status/2044059352466690501 https://x.com/mattcerminaro/status/2044546097230626819 https://www.politico.com/news/2026/04/13/influencers-allegations-eric-swalwell-00869517 https://x.com/JoeBiden/status/1815080881981190320 https://x.com/TimRunsHisMouth/status/1815088017847333273 https://www.npr.org/2024/07/08/nx-s1-5032737/biden-tells-democrats-stop-speculation #allin #tech #news [Source](https://www.youtube.com/watch?v=SFdqX7IY7RY)

[9:09] David Sacks: 在蓝州，你的财产根本不安全，那里的政治阶层总觉得自己有权分走一杯羹。 [11:23] 投资人 (引述): 你手握 ChatGPT 这样一个拥有 10 亿用户、年增速高达 50% 到 100% 的业务，为什么还要跑去死磕企业级和代码？这简直是一家毫无重心的公司。 [15:23] Travis Kalanick: 增长就是一切。运营效率最终必然碾压资本补贴。 [21:14] David Sacks: 如果你们真打算花几亿买个科技播客，我们这帮人一直都在这儿，Dario 赶紧联系我们，直接给我们把票验了！ [38:04] David Sacks: 最恶劣的监管俘获。 [45:44] David Friedberg: 超级富豪的神庙，是科技精英与建制派亿万富豪从底层敛财、把自己送上太空、却将普通人远远抛在身后的核心运作机器。 [65:37] David Sacks: 我们可以用体面的方式解决，也可以用强硬的方式解决。 [77:37] Chamath Palihapitiya: 如果你不能证明这项技术可以在主流的、顶级联赛级别的关键业务场景中跑通，它就依然只是个玩具。

[0:42] David Sacks: You take out that part of the market that in effect was price insensitive and was subsidizing all these projects, and I think development's going to dry up to a large degree. [17:37] Travis Kalanick: efficiency will outstrip subsidy. [25:02] David Sacks: hoisted on their own petard of doomer nimbyism [26:42] Chamath Palihapitiya: Remember what the biggest problem with Friendster was? Friendster was slow as a dog... then MySpace came in and took all the share, then Facebook came in and took all their share. [31:50] Unknown: Oh, wow, they're holding it back cuz it's so amazing. [31:50] Unknown: Looks like the All-In podcast... 37% chance of buying the All-In podcast. [31:50] Unknown: There's no way this is real, dude... It's Jay Cal slop. [31:50] Unknown: Time to upgrade the planes... Trickle-down avionics! [33:53] Unknown: Allbirds just pivoted from ugly sneakers... to AI and the stock has ripped. [33:53] Unknown: This reminds me of the late '90s where all you had to do was change your name to whatever.com. [33:53] Unknown: Software, you never really had to worry about COGS or gross margin... people in Silicon Valley weren't really trained to look at gross margin. [33:53] Unknown: If the cities had leaned into it... created a small EV lane, it could have transformed cities. Instead, they banned it... regulatory capture at its worst. [39:20] Chamath Palihapitiya: We are massively compute constrained. Massively. [39:20] Unknown: Maine just passed the bill that bans all data center buildings. [39:20] Unknown: For Anthropic and OpenAI, it is a five-alarm fire... It will entirely be because of the Friendster effect. You just couldn't keep the site up. [39:20] David Sacks: The data center... is the temple of the wealthy. It is the machinery of the rich, elite tech billionaires. [62:49] David Sacks: The Democratic establishment and all of the mainstream media were saying that Biden was sharp as a tack... And then he had that disaster debate performance... text messages were flying between the Democratic Party insiders... Nancy Pelosi was reported as having gone to the president said, 'We can do things the hard way or the easy way.' [67:16] Chamath Palihapitiya: When you have to follow the rules of disclosure, everybody's returns got knee-capped... Now, what's interesting to note is the reason why Nancy Pelosi's returns are so consistently good is Reg FD does not apply to people in Congress... They can learn things in their committee meetings... get out of the SCIF quick and get that trade in. [70:34] Travis Kalanick: The stock market is Trump's weather vane... He does not let the S&P go down too low. People get the sort of panicking thing now... he makes moves, and then he comes back to reality and gets things done... and the traders are getting used to it. [76:03] Chamath Palihapitiya: I think that it is crucial that the SpaceX IPO get done ASAP. And then I think it's even more crucial that one of Anthropic and OpenAI front-run the other one and get out first. And the first two charts are the reasons why.

[0:00] All right, everybody. Welcome back to [0:01] the number one podcast in the world. [0:04] We've got the core four here and dare I [0:08] say The king of Adams. The king of [0:11] Adams, yes. [0:13] Captain Travis Kalanick is here. How you [0:15] doing, brother? I'm pretty good. Pretty [0:17] good. I'm uh [0:19] sitting here doing the podcast just next [0:21] door to David. Ah, yes, there you go. [0:24] Don't reveal our locations. Please don't [0:26] dox. [0:26] >> put my address out there, dude. No, [0:28] that's true. [0:29] >> Don't worry, Mondummy did. [0:31] He's outside your house as right now [0:32] asking them to foreclose on you. [0:34] >> 3.9% or something? [0:36] Yeah, he's he he decided there's no rich [0:38] people left in New York, so he's looking [0:40] for other cities to tax. [0:42] >> Is it 3.9% a year? Is it per year or is [0:45] it [0:46] >> know if the percentage has been released [0:47] yet, but the speculation I've seen is [0:48] 3.9%, but I don't think that's final. [0:51] But yeah, it's a pied-à-terre tax, so if [0:53] you have a second home, yeah, every [0:55] year. Wow. And by the way, it's for any [0:58] home over 5 million. [1:00] There's no homes under 5 million in [1:02] Manhattan. This is not a rich person [1:04] tax. This is within [1:07] 15 miles of Midtown Manhattan, you're [1:10] paying an extra tax. Uh I don't know [1:12] >> But only but J Cal, only if it's a [1:13] pied-à-terre. So, what it means is that [1:15] the most second homes Well, yeah, but [1:17] the most elastic part of the market is [1:19] what they're targeting for this tax. So, [1:21] in other words, [1:22] people who don't live in New York, who [1:23] just have it as a second or third home, [1:25] who could buy that property anywhere [1:29] Yeah. are now being taxed the most. So, [1:31] what do you think that's going to do? [1:32] It's going to have a massive impact on [1:34] demand [1:36] for second homes in New York, which will [1:37] crash the whole market. Yes. [1:39] Congratulations, Mondummy. But in a [1:41] weird way, that'll be good for housing [1:43] affordability in New York. Well, that [1:45] that's sort of the the claim, but I [1:46] don't think it'll be good for it because [1:48] there'll be no incentive to build more. [1:50] Yeah. [1:52] All units matter. Every time you add [1:53] units, people upgrade, and it's not like [1:55] these are going to be low-income [1:57] housing. Like penthouse on 57th Street [2:00] or you know, in Gramercy, that's not [2:03] low-income housing. You'd have to break [2:05] it into seven units. Makes no sense. But [2:07] by the way, I don't know if you guys saw [2:08] the video, not to get too serious, [2:10] but he's doxxing [2:12] a certain billionaire who owns a certain [2:14] place and he's literally pointing at his [2:15] home. No, I said that to you, Jason. [2:17] He's not doxxing because everybody's [2:18] known for years that Ken Griffin bought [2:20] that place. Everybody knows that [2:22] address. Everybody knows that unit. We [2:24] all knew it. It was marketed widely. I [2:26] don't think that's really doxxing. He [2:28] doesn't live there and everybody knew he [2:30] owned it. It would be very different if [2:32] it was a place where somebody was [2:34] keeping their primary residence and you [2:35] didn't know and they stood in front of [2:37] the house. That I would agree with you. [2:38] I think this one is a little bit more [2:39] tenuous. [2:40] >> Okay, fair enough. But what I will tell [2:42] you is it's a dog whistle. [2:44] Crazy people, and this thing's been seen [2:46] by 30, 40, 50 million people now. [2:48] It's a dog whistle to say [2:49] >> That's true. That's true. [2:50] >> That's the next United Healthcare CEO [2:52] and in the week that a fire a Molotov [2:56] cocktail and a bullet get shot into Sam [2:57] Altman's house, it's deadly serious. And [2:59] if you reversed it, I always reverse it. [3:01] What if a Republican sat outside of [3:04] Bernie Sanders' second or third home and [3:06] said we should this is his pied-à-terre, [3:08] this is his summer home, we should add [3:12] you know, taxes to it and I'm sure you [3:13] could look up Bernie Sanders' home [3:14] pretty easily. So, just [3:17] before you point at people's homes and [3:18] say this is the villain, [3:21] be careful, folks, because then if [3:22] something does happen to that person, [3:24] like you know, what happened to Sam [3:26] Altman this week. You can feel however [3:27] you want about him, but nobody deserves [3:29] to have their house firebombed or shot [3:32] at. Period, full stop. [3:34] I just think that it's going to kill the [3:37] demand for, you know, maybe people who [3:39] already have a home in New York like Ken [3:41] Griffin, they probably are just going to [3:44] suck it up and pay the tax and keep [3:46] whatever they have, but if you were a [3:48] person who is thinking [3:50] about buying a pied-à-terre in New York, [3:52] you there's no way you would do it now. [3:54] Yeah. [3:54] >> Because you don't know what the tax [3:55] rate's going to be and it's going to [3:57] keep going up. After a decade with [3:59] interest and inflation, you've [4:01] effectively almost double the price of [4:02] your unit. So if you're going to buy a [4:03] $10 million unit, [4:06] you're probably then looking at a [4:08] $20 million [4:10] purchase price just to break even after [4:11] about 10 or 11 years. That's crazy. The [4:13] math doesn't work anymore. What is the [4:15] downstream effect of these individuals [4:17] not coming to New York, going to a [4:19] Knicks game, going to a restaurant? No, [4:21] they spend money. They spend money in [4:23] New York. They have their birthday [4:24] parties in New York. They do all kinds [4:26] of things in I'm not a big fan of the [4:27] pied a terre culture. I I own one house. [4:29] I don't like this whole multiple houses [4:31] you flitter-flattering everywhere. It [4:33] ruined London. [4:35] Because when when you look at London, [4:36] >> you vacation? Chamath, where do you [4:38] vacation? I go to a hotel. [4:40] Okay. [4:42] So, my point is this that I think the [4:44] way that it works is it's not just pied [4:46] a terre. If you [4:48] if you own a home that you fully rent [4:50] out, but that is not the primary [4:52] residence of the person who is renting [4:55] it, [4:57] it's also a problem. So, this is any [5:00] even like like a you know, you had a [5:02] two-week rental or a one-week rental or [5:04] a 30-day rental or even a five-year [5:07] rental and it was somebody else's second [5:09] home, the owner still gets gets hit. [5:12] So, the rental the sort of rentals of [5:15] like folks who want to come to that city [5:17] or who don't want to own in that city, [5:20] uh [5:21] but it's a second place. The The second [5:23] place thing goes away. So, maybe this is [5:25] just good for hotels. [5:27] I think it's good for hotels, but if you [5:28] look at London, it's probably the best [5:30] example where there was a lot of people [5:32] who used London as a place to store [5:34] assets. Real estate became the primary [5:37] way in which they would do that. [5:39] It hollowed out parts of London where [5:42] it's not as if it was unlivable, it's [5:43] just that it was unlived. Nobody was [5:46] there. You drive around Chelsea, you [5:48] drive around [5:50] certain parts of London and it was like [5:51] a ghost town on like a Thursday night or [5:54] a Friday night. That is an issue. It's [5:56] the land banking. It's like your Bitcoin [5:58] is hollowing out a neighborhood and and [6:00] there is something there. But you know [6:01] what? The whole uh housing thing is [6:05] complete cap and utter because [6:08] if you move to a place like Austin or [6:10] you are in Nevada or Florida, you see [6:12] what happens when you allow people to [6:14] build units. [6:15] In Austin, 3 years in a row, rents [6:18] and housing prices have gone down while [6:21] net migration has gone up. But that's [6:23] been good, right? For Austin, you guys [6:24] would say? Net good? [6:26] >> Incredible. Yeah, amazing. [6:28] Austin has like roughly doubled as a [6:30] city over the past decade and yet the [6:32] rent for, you know, whatever one or [6:33] two-bedroom apartments gone down. [6:36] That's incredible. So, in other words, [6:37] if you let people build to satisfy the [6:39] demand, you won't have this problem. [6:42] And then who's stopping the building? [6:45] It's Democratic cities. It's NIMBY [6:47] people. And then in a Republican town, [6:50] they're actually building units [6:53] for [6:54] and affordability. So, you have one [6:55] group saying they care about [6:56] affordability and they're doing nothing [6:58] about it and they're stopping it. And in [6:59] another place, they're like, "We're just [7:01] going to let you build because it's your [7:02] right to build because it's your land." [7:04] That's the approach in Texas. It's your [7:06] land, you have the right to develop it, [7:08] go. By the way, the high end of the [7:10] market in London has basically turned [7:12] over and collapsed. Sax, to your point, [7:13] like they introduced the stamp tax, [7:15] which I think is equivalent to this tax [7:17] that the New York City mayor is [7:19] proposing. And if you look at London as [7:21] a guide, the real estate market just bit [7:23] it at the high end. And that's I don't [7:26] think that's going to be good for that [7:27] city or even the UK as a whole. There's [7:29] a handful of cities where people do what [7:32] you say, which is kind of park money [7:33] there. The reason why they do that is [7:35] because they believe that that city A [7:38] has the rule of law and B is a unique [7:40] world-class city that's going to keep [7:42] appreciating. [7:43] And if you have new management that [7:45] doesn't really believe in the rule of [7:46] law, that basically keeps imposing all [7:47] these arbitrary taxes, that money is [7:50] going to flee and find other kinds of [7:51] investments. People aren't going to park [7:53] their money there. That has to be a bad [7:54] thing for the city. It's like, you know, [7:56] who cares if the top floor of that [7:58] building where Ken Griffin lives is [8:00] owned by one guy who who isn't there [8:02] that much. It's not going to affect the [8:03] city that much, but having all these [8:05] billionaires from all over the world [8:06] decide to park money [8:08] in that American city has to be good for [8:10] the US, [8:11] just like it's good for the UK. [8:13] And if you give that up, [8:16] then again, the money will just go [8:18] somewhere else. Well, to to your point, [8:20] the other thing is they're already [8:21] paying taxes on the property. [8:24] And because they're not there very [8:25] often, they're not using city services. [8:29] So, they're paying taxes on the [8:30] property, they're not using city [8:32] services. They're profitable to the [8:34] city. Yeah. Think about a developer [8:36] who's underwriting some new project. You [8:39] know, the fact that a whale like Ken [8:41] Griffin is willing to overpay in the [8:44] sense of price per foot for that top [8:47] floor [8:48] might make that whole project pencil, [8:50] and this is what contributes to the [8:51] vitality of New York. There's constantly [8:53] development going on. There's constantly [8:55] cranes. And so, you take out that part [8:58] of the market that in effect [9:01] was price insensitive and was [9:03] subsidizing all these projects, and I [9:05] think development's going to dry up to a [9:07] large degree. Mhm. He's paying he must [9:09] be paying three or four million in taxes [9:11] every year and getting no services to [9:12] your point, Travis. Good point. Like all [9:14] profit for the city. [9:15] >> London also did something else, which is [9:16] that they essentially crippled what's [9:18] called non-dom status, which is the big [9:20] tax arb if you're moving or parking [9:23] assets in London. [9:24] And to your point, Sachs, what did all [9:26] the rich people do? They just redirected [9:27] themselves to Zurich, to Lugano, to [9:30] Milan. [9:31] And they they took advantage of more [9:33] hospitable tax policy in other places. [9:37] Now, what the people in Britain would [9:38] tell you, I actually met with the UK [9:39] government yesterday, [9:42] they don't see a meaningfully enough [9:44] measurable impact yet where they think [9:46] it's a five-alarm fire. So, the real [9:47] question is is it more of a slow bleed [9:49] and a slow melt [9:51] and at some point it's just hollowed out [9:53] and it's very hard to reverse cuz [9:54] there's not going to be a cataclysmic [9:56] acute moment where people say, "Oh my [9:58] god, we need to reverse these policies." [10:00] It doesn't seem like that's in the [10:01] offing. You know, LA did something a [10:04] little different but similar as they [10:05] introduced that 5% mansion tax. It's on [10:08] like all the areas except for like [10:10] Beverly Hills and San Francisco has [10:12] that, too. Yeah, San Francisco got that, [10:14] too. I think it's actually like 6% in [10:16] San Francisco. [10:17] >> it's 6% over 25 million and it starts at [10:20] 5 million. So, there's a transfer tax, [10:23] an excess transfer tax above $5 million [10:25] properties that scales up above 25 [10:28] million to an extra 6% on top of your [10:30] brokerage commission. So, when you sell [10:32] a house in San Francisco, you're paying [10:35] 13% now? Which is why you look at the [10:38] transaction volume in the LA real estate [10:40] market and it's just completely dried [10:42] up. You know, people aren't doing the [10:43] house flipping anymore like that kind of [10:45] stuff because the transaction costs are [10:47] too high. [10:48] But, it just shows, I mean again, this [10:50] tax was imposed retroactively. For [10:52] example, I had my house in SF and then [10:53] they just take [10:55] you know, a couple of rooms of it. [10:58] And we talked about this on the show [10:59] when it happened. [11:01] But, my my point is just your property [11:03] is not safe in blue states. And wealthy [11:06] people who have a choice of where to [11:07] park their money are going to [11:09] increasingly realize that and they're [11:12] not going to buy I think real estate in [11:14] blue states is dangerous because [11:16] the political class thinks that they can [11:18] take a chunk of it. And you know, the [11:20] wealthy people are going to react to [11:21] that and they're going to move their [11:22] money elsewhere. All right, let's go to [11:24] topic number one. OpenAI is apparently [11:27] suffering from a bit of an identity [11:30] crisis. On Sunday, OpenAI's Chief [11:32] Revenue Officer, Denise Dresser, sent a [11:34] four-page memo to employees. Obviously, [11:36] it leaked immediately. Probably the [11:38] point of it. And she called out [11:40] Anthropic. She said their $30 billion [11:42] run rate is cap. Inflated by $8 billion [11:46] due to a revenue share and some [11:47] accounting with AI model providers. [11:50] Chamath you pointed that out in the last [11:51] couple weeks. Also, she said Anthropic's [11:54] story is built on, quote, fear, [11:56] restriction, and the idea that a small [11:58] group of elites should control AI. [12:01] Obviously, she's a fan of the pod. She [12:02] also laid out Open AI's pivoting, and [12:06] she said they are going hard after [12:08] business customers, and they want to win [12:11] the agent platform layer. If you [12:13] remember, they hired the architect of [12:15] the open-source project Open Claw. They [12:17] didn't acquire Open Claw. So, Peter [12:20] Steinberger is working at Open AI. [12:23] Cynical people said, "Hey, maybe they [12:24] want his next set of innovations to go [12:26] inside of Open AI's products as opposed [12:28] to the open-source one." I kind of agree [12:29] with that directionally. [12:31] There's obviously Perplexity computers [12:33] doing really well. They quadrupled their [12:34] revenue. And I was over at XAI earlier [12:38] this week with Elon, and I can tell you [12:40] he's got some very cool stuff coming. [12:43] And this new model, Spud, is coming from [12:46] Open AI. Here's your Polymarket 75% [12:49] chance Spud is released [12:52] next week. [12:54] Additionally, and we'll go to the panel [12:56] in just a second, on Tuesday, 2 days [12:59] after this memo came out. Obviously, [13:01] these memos to go directly to the press, [13:03] so they're obviously trying to undercut [13:04] Anthropic's valuation, and they feel [13:06] that's a threat. That's my take. The FT [13:08] cited anonymous Open AI investors [13:11] who are frustrated with the company's [13:13] lack of focus. Here's the anonymous [13:15] quote. Quote, you have ChatGPT, a 1 [13:18] billion user business growing 50 to 100% [13:21] a year. What are you doing talking about [13:23] enterprise and code? It's a deeply [13:25] unfocused company. [13:27] And we talked about this, ChatGPT's [13:29] market share is going down as the number [13:31] of users is going up because Gemini and [13:34] Claude gaining significantly. And Meta [13:37] just last week released their first [13:39] proprietary model. And Apple doesn't [13:42] have a product in market yet, but they [13:43] do have a lot of users. So, here's your [13:45] gen AI website traffic. [13:49] Let's start uh Chamath with you. Your [13:52] thoughts on [13:54] OpenAI, should they be pivoting straight [13:57] into business developers and getting [13:59] focused on that, or should they stay [14:00] focused on the consumer where they are [14:02] the verb? As it relates to complex long [14:05] horizon coding tasks, what I can tell [14:08] you from my team at 80 90 is Codex is [14:11] better [14:13] generally than Anthropic. And so, what [14:15] happens is for the more day-to-day work, [14:19] I think it's more reliable to use the [14:22] Claude ensemble of models. [14:25] But when you're dealing with something [14:26] that's very tricky and very complicated [14:28] and a little bit more long horizon, [14:31] Codex is [14:32] really functional and really good. So, [14:35] I think if you're looking at it through [14:37] the lens of OpenAI, what they're [14:39] probably saying is, "Hey, hold on a [14:40] second. [14:41] If we allocate our resources and just [14:43] double down and crush consumer, [14:46] that's probably three or four trillion [14:49] of enterprise value. [14:51] And then if we slowly refocus the [14:53] company with the rest of the resources [14:54] and double down on Codex and do [14:56] something meaningful in enterprise, we [14:58] can probably capture [14:59] two or three trillion there. And now all [15:01] of a sudden, you can paint a picture for [15:03] a [15:04] seven, eight, nine trillion-dollar [15:05] market cap in the fullness of time, not [15:07] tomorrow, obviously. But Codex is really [15:09] good, and there's a business to be had [15:12] in both. You have to separate the two [15:13] businesses. You can't have a lot of [15:15] overlap because there's too much context [15:17] switching. You got to let the consumer [15:18] team run, [15:20] and then you got to isolate the [15:21] enterprise team and let them do what [15:22] they think is right. Travis, uh there's [15:24] a big debate going on amongst the [15:26] investors. The FT piece also questioned [15:28] the $850 billion valuation of Open AI. [15:31] Secondary markets have now priced [15:34] Anthropic higher than Open AI for the [15:36] first time. This is the flipening that [15:38] people predicted. One investor said Open [15:40] AI would need to IPO at a valuation of [15:42] 1.2 trillion for the last round to make [15:44] any sense. [15:46] But there's no buyers currently at the [15:49] $850 billion valuation that Open AI just [15:52] closed, according to Bloomberg. [15:55] Travis, how do you uh handicap this race [15:58] between these two leading frontier [16:00] models? [16:01] Growth is king right now in this in this [16:04] world uh in this segment. Growth is the [16:06] whole damn thing. [16:08] And if Anthropic is growing faster than [16:11] Open AI by a significant clip, [16:14] the investors right now are going to [16:15] play it forward. [16:17] And there it [16:18] you know, you start to get network [16:20] effects around compute, network effects [16:22] around the number of tokens you're [16:24] pushing out for various customers, [16:26] enterprise or consumer. [16:28] And ultimately, you know, it's not great [16:30] today, but how that plays into [16:33] reinforcement learning, and the things [16:34] getting smarter over time. [16:36] There's so many [16:39] there's so much upside to volume and [16:41] scale that if they are growing faster at [16:44] the same size, [16:47] even if Open AI is still growing, but if [16:49] they are half the growth rate, a third [16:51] the growth rate, a fifth the growth [16:53] rate, [16:55] I would be I'd be worried. [16:57] And you saw this at Uber specifically [16:59] when you accelerated away from the [17:02] competitors like Lyft and DoorDash, [17:04] yeah? Yeah, you had to. Yeah, network [17:06] effects was the whole thing at the end [17:07] of the day. And network effects was [17:09] based on scale. So yes, I'm I'm sort of [17:12] like coming from my very specific [17:14] experience. But if you believe there's [17:16] network effects with the scale of data [17:18] that you have and the scale of customers [17:20] and the revenue that is cash that's [17:23] coming in that you redeploy into compute [17:26] and so say there's network effects of [17:28] large compute, [17:30] I'd be very worried [17:32] if I'm OpenAI and I'm seeing somebody [17:34] growing faster at the same size. So, [17:37] Freeburg, when you look at this race [17:39] between these two giants, maybe your [17:41] thoughts on the flywheel as it relates, [17:43] as Travis is pointing out, to advantages [17:47] in compute, reinforcement learning, and [17:50] then also the ability to fundraise. One [17:51] of the great things that Travis and the [17:53] team did was as they were pulling away, [17:57] they just sucked all the oxygen out of [17:59] the room by using capital as a weapon. [18:02] So, your thoughts, Dave, on this [18:03] high-stakes game because there's also a [18:05] point at which, and I'll just end on [18:06] this, there's a point at which you could [18:09] run off the cliff. You raise so much [18:10] money [18:11] and you deploy it so fast and the [18:13] revenue doesn't catch up to it and then [18:14] you go public and the markets don't [18:15] believe the story. So, your thoughts, [18:17] Freeburg. I don't know the financials of [18:19] the two companies well enough. [18:20] Obviously, Sam has no problem raising [18:22] money. The guy didn't he just close like [18:23] a $150 billion round or something? How [18:25] much was it? [18:27] >> billion. [18:27] >> $122 billion. Largest round ever raised [18:30] in any market, I think, private or [18:32] public. [18:32] >> Yeah, probably. Yeah, I mean, that's [18:33] great. [18:34] >> public, yeah. That doesn't seem to be an [18:35] issue. What I've noticed is just the the [18:38] pace of innovation at Anthropic is uh [18:42] for my experience, unprecedented. I [18:43] mean, they're [18:45] release cadence is extraordinary. [18:47] They've basically supplanted Open Claw [18:49] already with this release they did a few [18:51] days ago and then today the new Opus [18:53] model got dropped. So, there's something [18:55] about the momentum not in necessarily [18:59] just in user growth, but in how they're [19:01] operating this business that just seems [19:03] to be head and shoulders above everyone [19:05] else in the cadence of of upgrades. If I [19:08] look back 6 months ago, I think we We [19:10] pretty heavy on Cursor [19:12] and Gemini and now I think we're [19:14] probably 90% Anthropic in just the last [19:17] 6 months in my my organization. There's [19:19] something very powerful about the [19:20] flywheel they have going on. [19:22] Yeah, but here's where I go just real [19:24] quick Dave is [19:27] mad respect on a [19:29] 120 or 130 billion dollar raise. I mean, [19:32] Yeah. this is obviously next level, but [19:35] you can use capital and investment to [19:39] acquire scale and network effects [19:41] associated with it. But if somebody is [19:43] getting that scale [19:46] with revenue and let's call it [19:48] contribution margin, contribution [19:50] profit, [19:52] efficiency will outstrip subsidy. [19:55] And it will you can't just keep raising [19:57] 100 billion dollar things forever. [19:59] That's where the train will stop. And if [20:02] Anthropic is funding theirs through [20:04] revenue [20:05] and other folks are funding it through [20:08] investment, there's like a short-term [20:11] that's a short-term solve, but it the [20:13] long run is whoever is scaling their [20:16] actual usage and system and ultimately [20:20] with contribution profit that then soaks [20:23] up the need for investment, that's a [20:25] that's a that's a very scary machine if [20:28] you're competing against it. Which is [20:30] exactly Sacks what the [20:32] um [20:33] legacy Mac 7 are doing. You have massive [20:36] profits [20:37] from uh Meta's core business, Google's [20:40] core business that are being redeployed [20:42] into infrastructure and even uh Tesla [20:45] which has a lot of profits in SpaceX [20:47] which has a lot of cash on hand and [20:49] they're building out Colossus and other [20:52] assets including uh Elon is working on [20:54] building a fab as folks have been [20:56] talking about. So, Sacks, is there a [20:58] chance for the legacy companies, the Mac [21:01] 7s to compete in this or are we looking [21:04] at Open AI and Anthropic are one and two [21:07] and then everybody else can uh fight for [21:09] third place and the and the and the and [21:11] the bronze as it were. [21:14] Well, I think Google's clearly in the [21:15] mix. I mean DeepMind has an outstanding [21:17] team and I think Elon's still in the mix [21:19] with xAI and then you've got Meta also [21:23] has the resources. And they seem to be [21:25] further behind, but they're going to [21:27] compete. Look, let me just go back to [21:28] the central premise here. [21:30] I agree that there's some valid [21:33] criticism that OpenAI has been unfocused [21:38] and should be, you know, moving forward [21:41] more focused in what they do. I have no [21:44] idea, for example, what they're doing [21:46] buying up podcast. That's not us. So, I [21:49] don't know what what that was about. [21:51] Um if you were going to buy a tech [21:53] podcast for a few hundred million [21:54] dollars, I mean, um we were here. We're [21:57] here. We're here. [21:59] Dario, reach out. [22:00] Small potatoes for you guys though. [22:02] Small potatoes. Yeah, because they can't [22:04] afford you. They can't afford you guys. [22:06] Yeah, they thought we were too [22:06] expensive. [22:08] Little did they know, we would have sold [22:09] out. But um [22:11] Punch the ticket, Dario. [22:14] But look, this other part of the [22:16] criticism of OpenAI that they shouldn't [22:19] do enterprise is totally misguided. One [22:22] of the reasons why they should have been [22:23] more focused is to do more enterprise [22:25] and get enterprise more correct. Now, [22:28] why do I say that? To Travis's point [22:30] about growth rates, [22:33] it's true that OpenAI and Anthropic as [22:36] of the beginning of Q2, so let's say 2 [22:38] weeks ago, they were both around 30 [22:40] billion of revenue and that memo from [22:42] that OpenAI employee was right that if [22:45] you compare them on an apples-to-apples [22:47] basis, then Anthropic is about 20% less [22:50] because they are including [22:52] revenue made by their channel partners. [22:54] But that doesn't matter, what matters is [22:55] the growth rate, [22:57] you know, again to Travis's point. [22:59] And let me just put some numbers around [23:01] this. The Open AI growth rate's been [23:03] around three to four X a year. The [23:05] Anthropic growth rate has been around 10 [23:07] X a year. So, they went from let's call [23:11] it one to 10 billion of ARR last year, [23:14] and by the end of Q1 this year, they're [23:16] already at 30 billion of again, let's [23:19] call it their revenue. [23:20] And they're on their way, you know, like [23:22] Brad Gerstner was saying on our podcast, [23:24] I think in the last couple of weeks, [23:25] they're going to end this year at 80 to [23:28] 100 billion, at least on the current [23:30] trajectory. [23:31] And so, you can plot their revenue on a [23:33] logarithmic graph. I mean, again, no [23:35] one's ever seen anything like before, [23:37] where every unit on the Y axis is [23:39] another X is 10 X, and it's a straight [23:41] line. Yeah, that's crazy. It's crazy, [23:44] right? So, if it's taking Anthropic, [23:46] let's say, one year to 10 X, and it's [23:48] taking Open AI two years [23:51] to achieve a 10 X, then it's obvious [23:53] which one's going to win. [23:55] Now, what is the reason for this? It's [23:57] because Anthropic was very focused on [24:00] enterprise, specifically coding. [24:02] And what you're seeing is that [24:04] businesses are willing to pay for coding [24:06] code tokens on a meter basis, let's call [24:09] it like electricity. The more they use, [24:11] the more they're willing to pay. [24:14] And their usage just continues to scale [24:16] and scale. Consumer is completely [24:18] different. I mean, consumer is the thing [24:19] that Open AI prioritized. Consumers have [24:21] a lower willingness to pay. Maybe only [24:23] three or four percent of them are [24:24] willing to convert to premium in the [24:26] first place. And what they want is a $20 [24:28] a month all you can eat subscription. [24:31] So, the revenue simply doesn't scale the [24:32] same way that enterprise does. And so, [24:35] if you want to tap into the scalable [24:37] revenue source in the market right now, [24:39] you have to go after enterprise. So, you [24:41] know, again, where I would agree with [24:43] the criticism of [24:45] Open AI is maybe they should have been [24:47] more focused, but they need to be more [24:48] focused specifically [24:50] to pursue coding and enterprise. And if [24:54] they don't catch up soon your point, [24:56] then you could see Anthropic taking a [24:58] lead here that, let's say over the next [25:00] 1 or 2 years, could be insurmountable. [25:02] Just by the way, let me just say say one [25:04] thing is [25:05] even though Anthropic's revenue has [25:08] followed this graph, this exponential [25:09] graph very predictably, [25:11] it can't do that forever, right? Like [25:13] let's say it does get to maybe 100 [25:15] billion this year. Can it really get to [25:18] a trillion dollars in revenue the year [25:19] after that? Seems hard to believe, [25:21] right? And and and the reason is because [25:23] as you hit new levels of scale, you [25:26] encounter new problems. I mean, you're [25:27] simply going to run out of compute or [25:30] electricity, data centers, you know, [25:32] infrastructure. There are physical [25:34] limits or there's limits in the physical [25:36] world that you're going to hit. And [25:37] there's already some evidence that [25:39] Anthropic is hitting some of those [25:40] limits. Users were complaining, for [25:42] example, that Claude was thinking less. [25:45] Did you guys see this? That that, you [25:47] know, a typical Claude prompt, they they [25:49] seem to have cut down on the thinking [25:51] time by about 2/3. Now, I saw someone [25:53] tweeting today that they just released [25:55] Opus 4.7, replacing Opus 4.6, and the [25:58] thinking is back, but, you know, maybe [26:00] they're charging more for that, hard to [26:02] say. But they're going to hit some sort [26:04] of physical limits, and I do wonder if [26:06] over the next year [26:08] Anthropic will reconsider whether its [26:10] support for all this like doomer [26:12] nimbyism [26:13] was the right call because it kind of [26:16] made sense for them [26:18] from a business standpoint when their [26:20] competitors were building data centers [26:21] and they were just getting compute from [26:23] the hyperscalers, but now that they're I [26:26] think going to have to move into the [26:27] game of building their own data centers, [26:28] they might regret salting the earth [26:31] for data centers all over the country. [26:33] And I wonder if that'll be the natural [26:36] limit [26:37] of their growth is they'll be hoisted on [26:39] their own petard of doomer nimbyism. And [26:42] they're also using the [26:44] coding platform to build Anthropic [26:47] itself better. So that in and of itself [26:49] is a reason to nail coding. You you get [26:51] the double whammy. You can make a better [26:53] product and you can get paid for it. [26:55] >> Yes and no. Hold on because if you look [26:56] at Twitter, people are panning Claude [26:58] Desktop and what they said is this is [27:00] all a bunch of vibe coded slop. I think [27:02] we have to remember if you keep these [27:03] agents on task and they're guard wheeled [27:06] properly, [27:07] this stuff is a force multiplier. The [27:09] problem is nobody knows how to do this [27:11] really well yet. Nobody has built real [27:14] products of scale largely using agents [27:17] yet. Nobody knows how to give an example [27:20] of how an org structure should be [27:21] redefined. Nobody knows how to budget [27:24] properly. We had the CTO of Uber say I [27:26] give up. I've hit my token budget. The [27:28] problems I see it are twofold. [27:32] The first is just to build on Sachs's [27:35] point. All of these frontier labs have a [27:37] very serious issue which is both OpenAI [27:39] and Anthropic are growing so fast [27:42] that they're at a point now where they [27:44] need their own infrastructure. It's kind [27:46] of like when you first start building [27:47] any kind of company, you're just much [27:49] easier renting capacity from the [27:52] hyperscalers. And it's a dependency. [27:54] Yeah. [27:55] >> But then it becomes a dependency. [27:56] Exactly. And now when you're so big, [27:58] it's actually strategically a huge [28:01] mistake to not have your own compute [28:03] supply. Why? [28:05] Because if you look at who's leading, [28:06] the frontier labs are leading. [28:08] And Sachs, to your point, you mentioned [28:09] this on X. You're like, there was all [28:11] this doomerism but maybe it was tied to [28:13] compute capacity because when Bedrock [28:15] opened up more capacity for Anthropic, [28:17] all the doomerism went away. You're left [28:19] wondering like is it really tied to just [28:21] the fact that they were just trying to [28:22] throttle usage? So if you're a frontier [28:24] lab, you don't want to have to go [28:26] through Amazon and GCP and Azure and tin [28:28] cup for access and capacity. What you'd [28:31] much rather have is go straight to your [28:33] customer. On the other side, [28:35] if you're Gemini or [28:38] Microsoft or Meta and you have all this [28:41] compute because I saw a stat this week, [28:43] the hyperscalers control 60% of all the [28:46] compute. [28:47] So the game theory there is if you [28:49] kneecap the frontier labs, it'll give [28:50] you some chance to catch up. And it [28:52] gives you time to catch up because no [28:54] matter what the demand is on the upside, [28:56] you remember you guys remember like in [28:57] social networking when Friendster was [28:59] the cat's meow? Yeah. Cat's meow. [29:03] Remember what the biggest problem with [29:04] Friendster was? [29:05] >> Friendster was slow as a dog. Yes. Yes. [29:09] And what happened? Then MySpace came in [29:11] and took all the share. Then we came in [29:12] Facebook and we took all their share. [29:14] So, there is a way where you can [29:16] handicap and kneecap these companies by [29:18] throttling compute access to them. [29:21] So, A, they are forced to now go and get [29:24] in the game, which is weird because [29:25] look, OpenAI has tried to displace some [29:29] of the Stargate spend. [29:31] I don't see any path except they're [29:32] going to have to do it themselves and [29:33] Anthropic will have to do it themselves. [29:35] But then separately, the other problem [29:36] is [29:37] when you change the subscription model [29:39] in enterprise [29:40] and you say, "Hey, we're not going to [29:42] subsidize any more tokens." [29:44] What's going to happen is all these [29:46] token budgets going to go crazy and what [29:47] Freeburg said is going to happen where [29:48] he's like, "Hey guys, why are you [29:49] spending all this money? What are you [29:50] making?" [29:52] And you inspect the code and you're [29:53] like, "What is this slop?" [29:55] And you're not going to add 30, 40, 50% [29:58] opex [29:59] to produce nothing. [30:01] So, I think that that's an open question [30:03] and that question will become more [30:04] amplified over the next year as [30:07] they push the costs off of them. So, as [30:09] Travis said, no more subsidy from the [30:11] capital. [30:13] You have to grow into it, but you're not [30:14] going to support negative gross margins. [30:16] So, you're going to pass through the [30:17] token costs. [30:18] I So, I think it's a very dynamic moment [30:20] right now for these [30:21] >> And if you look at the ranking [30:23] of these clusters and who has the most, [30:26] right now, people have forgotten about [30:28] Colossus, which Elon's been building. [30:30] He's expanding to 555,000 [30:33] GPUs across three buildings, 18 billion [30:36] in investment. He If then if you look at [30:38] Prometheus, Meta's planned [30:41] 2026, that's 150,000 GPUs. So, this is [30:45] uh what you [30:45] >> Elon just announced a deal. He announced [30:47] a deal this morning with Cursor. [30:49] So, Elon's Elon's renting a bunch of [30:51] capacity. So, he's now getting [30:52] effectively into the data center [30:53] business. He's going to be a hyper [30:54] scale. [30:55] So, he's he's going to use as much as he [30:57] can for xAI and whatever is left over [30:59] he'll give to Well, in this case he's [31:01] giving to Cursor to train their model. [31:03] He could give it to others. [31:04] >> you might as well overbuild capacity [31:06] because that way your own models will be [31:08] in a privileged position and you can [31:09] sell the rest of to your competitors. [31:12] But, Jamath, to your point about the [31:14] thing I was saying on X, I actually I [31:16] think I was retweeting Mark Andreessen [31:18] who pointed out that one of the reasons [31:21] why Anthropic might have wanted to hold [31:22] back Mythos is they simply didn't have [31:24] the compute to serve it. The model was [31:26] huge and very expensive to serve, [31:29] something like maybe even 10 or 20 times [31:31] the token cost of of say Opus. They knew [31:34] Opus 4.7 was coming out, right? So, they [31:37] hold it back knowing that they don't [31:39] have the compute to serve it anyway, [31:42] and they save their compute for the next [31:43] iteration of Opus, and then by holding [31:46] it back they create this impression of [31:47] scarcity and altruism, and it turns into [31:50] this gigantic marketing event for their [31:53] product because everyone in the [31:54] government's like, "Oh, wow, they're [31:55] holding it back cuz it's so amazing." [31:58] Now, look, I think it may have been [32:00] genuinely altruistic as well in the [32:02] sense that Mythos does reveal coding [32:05] vulnerabilities that people didn't know [32:08] about before. And, you know, it does [32:10] make sense to give [32:11] time to companies with large code bases [32:13] to patch these dormant bugs and [32:16] vulnerabilities, but [32:17] it's looking more and more like [32:19] Anthropic could not have offered that [32:21] model commercially anyway [32:23] because it was just too big and [32:24] expensive, and they need to create space [32:27] for Opus 4.7. [32:28] So, [32:29] it's an interesting theory on what [32:31] actually happened there. All right, [32:33] guys. Uh before we go to our next story, [32:34] some breaking news here. Uh here's your [32:36] Polymarket, gentlemen. I don't This is I [32:39] don't know if you're placing some [32:40] insider bets here Freeberg, but looks [32:42] like the All-In podcast and Tropic now [32:45] 37% chance of buying the All-In podcast. [32:48] This is live [32:49] >> time? from Polymarket. Yeah, this is by [32:51] the end of the year. By end of year. [32:53] Okay, good. By end of the year, so it's [32:55] This isn't real, is it? Yeah, [32:56] absolutely. They've been People have [32:58] been trading on this. This is heavily [32:59] traded [33:00] >> be. It says not 92 million dollars of [33:02] volume. They can't be. [33:03] >> the number three I'm just calling it [33:06] out. There's no way this is real, dude. [33:08] There's no way this is It's breaking [33:09] news, guys. I don't I don't control the [33:11] news flow. It's Jay Cal slop. This is no [33:14] way this is real. That's what I mean. [33:15] Anybody can create a prediction market, [33:17] I guess. Hey, what's the what's the [33:19] volume of that thing? [33:20] >> 92 million. That's not real. [33:22] There's 92 dollars behind it. There's no [33:24] way that's real, guys. Come on. [33:26] >> Guys, it's all good. Nice job, man. [33:28] Guys, time to upgrade the planes. I just [33:31] Hey, guys, just I don't need the PC-24. [33:33] I'll take somebody's G650. Whoever's got [33:35] a G650, go up to the 800. It's going to [33:37] be trickle-down to Jay Cal. Let's do it. [33:40] I'm giving up my United Platinum status. [33:42] Trickle-down economics. [33:43] >> Trickle-down avionics. All right, [33:45] listen. Story number two. Trickle-down [33:48] avionics. [33:49] That's good. [33:52] This is pretty good. Okay, number two. [33:53] Allbirds. Speaking of data centers, [33:57] Allbirds just pivoted from ugly [33:59] sneakers, the squaziad, to AI and the [34:02] stock has ripped. Talking about people [34:04] Global Behavior podcast getting bought [34:07] by Frontier Models and sneaker companies [34:10] pivoting to [34:12] data centers. Allbirds, as you know, is [34:14] the ugliest sneakers on the planet. [34:16] And [34:17] this became a massive dilution in our [34:20] industry that this company was worth [34:22] billions of dollars. They went public in [34:25] 2021. This might be one of the peak zerk [34:27] moments. Raised 350 million dollars in [34:30] their IPO. Sacks, were you an investor [34:33] in this thing? [34:34] All birds. All birds back in No. No, no, [34:37] no, no. You're thinking about the [34:38] scooter company. So, that was the [34:39] scooter company Bird. That's right. [34:40] Okay. Yeah, don't don't remind me of the [34:43] all the investors that didn't work. But, [34:44] no. [34:45] Bird was crazy. This reminds me of the [34:50] late '90s where all you had to do was [34:51] change your name to whatever.com. Yeah. [34:54] Yes. And you get a huge pop in your [34:56] valuation. And you could spin it out. [34:58] Barnesandnoble.com became [35:00] And then Barnes and Noble's [35:02] And so, the stock crashed and never [35:05] looked back. There's your stock chart. [35:07] Um they sold off all their brand assets [35:08] for 39 million, about 10% of what they [35:10] raised in the IPO. [35:12] Congratulations to whoever owns those [35:14] ugly ass sneakers. Why were they worth 4 [35:16] billion? They sold sneakers. [35:17] >> Collective delusion. Collective [35:19] delusion. People in Silicon Valley liked [35:21] them and they just thought it was the [35:22] next Nike. Tulips. [35:24] It was tulips. Well, I think I think [35:26] that was an era in Silicon Valley where [35:29] people rewarded rapid growth without [35:32] really looking at gross margins or cost [35:34] of goods sold. People didn't really make [35:36] the distinction between software and [35:37] everything else, right? [35:38] >> Right. Software, you never really had to [35:41] worry about COGS or gross margin because [35:44] the incremental cost of serving a [35:47] customer with software is like almost [35:48] zero. Yeah. So, people in Silicon Valley [35:50] weren't really trained to look at gross [35:52] margin and there was this rash of [35:54] physical world companies that all of a [35:57] sudden started getting crazy valuations. [35:58] Travis, quite frankly, you might have [36:00] the success of Uber might have ushered [36:02] in this era of you know, these physical [36:05] world companies that started getting [36:06] valued like software companies, even [36:08] though obviously they didn't deserve it. [36:10] David, thank you so much for that shout. [36:11] I really appreciate it. You and Airbnb [36:15] Anyway, let's get back to [36:17] >> no, no, no, no. Hold on. Hold on. The [36:18] era of 21 is super interesting though, [36:20] guys. I don't believe it was a physical [36:22] versus digital thing. I believe it was [36:25] a moment in time in in the COVID ZIRP, [36:29] massive money going in, we hadn't seen [36:31] inflation yet. It was all of that [36:34] happening, and the investor class was [36:36] basically deciding, "We're going to look [36:38] two and three years forward on your [36:41] current growth." [36:43] And it wasn't one year forward, it was [36:45] like two or three years forward, and [36:47] that's where these crazy valuations got [36:48] weird. So, if you went from zero to a [36:50] hundred, you know, if you went from a [36:52] hundred to three hundred million dollars [36:54] in the last year, they'll play that two, [36:56] three years forward and go, "Oh, yeah, [36:58] you're totally twenty times bigger. [37:00] We'll pay you for that now." Yeah. [37:02] >> That's where it got weird. [37:04] It got weird. Hey, Bird Scooters was [37:06] valued at two or three billion. They [37:07] were doing micro-mobility and You're [37:09] just trolling them, Jay. I'm no, I we're [37:11] giving everybody their flowers for [37:13] incredible investments and say even even [37:16] the mighty Sachs could trip up. Did you [37:18] sell the shares for IPO? Yeah. No, no, [37:21] we don't do that. [37:22] Uh it was a Series A we did. I mean, [37:24] look, we we got it right in the sense [37:26] that it was a total phenomenon. Yeah. [37:28] >> But then the cities just cracked down on [37:29] it and killed it. Look, if the cities [37:31] had leaned into it, yeah, look, if the [37:33] cities had reacted differently, if [37:34] they'd leaned into it, if they had [37:36] created, let's say, a scooter or like a [37:39] small EV lane, it could have transformed [37:42] cities. I mean, it would have been a lot [37:43] easier to get around, but instead, they [37:44] banned it, they limited it, they didn't [37:46] create designated areas for it. And then [37:50] the the coup de grace was basically when [37:52] they would take a city where Bird [37:54] already had like 80% dominant market [37:56] share, and then they would say that, [37:58] "Well, we're going to choose four [37:59] operators and give them each 25% of, [38:02] say, a thousand scooter allocation." [38:04] That just killed the economics for [38:05] everybody. Yeah, there's no market [38:07] anymore. There's no market, yeah. Yeah. [38:09] There was no marketplace network effect, [38:11] right? When when they're basically just [38:12] picking the winners and deciding the [38:13] market. [38:14] >> can't com- you can't compete and [38:16] splitting it even, so then [38:17] >> Yeah, there's no there's no competition. [38:19] You can't drive value. [38:20] >> Regulatory capture at its worst. [38:22] >> Guys, like, we should watch for this in [38:24] the in the autonomous car space, too. [38:27] Cities may get cute and start doing [38:28] things like that. And And what they did [38:30] on scooters, they could do on cars. [38:33] Ooh. That's a really good point. Yeah, [38:34] you're right. Instead of just letting [38:35] the market play out, they say, "Well, [38:38] we're only going to have X number of [38:40] thousand autonomous cars." New York is [38:42] literally doing this right now. Like the [38:45] medallion. [38:46] >> and Boston. [38:46] >> the prices prices go up, the innovation [38:49] doesn't get realized, consumers don't [38:51] benefit, no one benefits. [38:53] Basically, you've deleted the market. [38:55] Yes. [38:56] And anyway, just to wrap up this story [38:58] on [39:00] the the shoe company, they're now New [39:01] Bird AI. They bought eight H100s, I [39:04] think, with $50 million in a convertible [39:06] note. And [39:08] the stock has gone up. [39:10] >> Not eight. Stop. Eight really No, it's [39:12] not eight. Oh, no, it's a joke. [39:14] The stock's now at $14 a share. It's up [39:16] 450% in the last week. Shout out to Wall [39:18] Street Bets for [39:20] >> staying Can I say something serious? [39:21] >> more than the shorts can stay solvent. [39:25] Can I say something serious? Yes, [39:27] please. There are a handful of [39:28] transactions that have happened in the [39:30] last few days [39:31] that if you [39:33] look far away, are head-scratchers. [39:35] So, this is one. [39:38] I don't know if you guys saw recently, [39:39] but Jane Street did a [39:41] a billion-dollar investment in uh [39:43] essentially a new Neoscaler. [39:46] And then also did a $6 billion compute [39:48] deal with them. [39:50] That was like a little interesting. So, [39:53] what would I like to say about this? I [39:54] think the thing that the capital markets [39:56] are getting right [39:58] is that we are massively compute [39:59] constrained. [40:01] Massively. [40:02] And those are two problems. One is the [40:05] power. So, if you look at companies like [40:07] Bloom Energy, it has gone absolutely [40:09] straight up vertical nuclear. [40:12] And the reason is because Bloom [40:14] has a solution that allows you to use [40:16] nat gas [40:18] that allows you to do something on site [40:20] and critically allows you to get your [40:22] clean air permits [40:24] very quickly because it has very very [40:26] little emissions and it's been proven as [40:28] such. And so instead of waiting for [40:30] years to get on the grid, if you wanted [40:32] to build a data center [40:34] you can now use their services. [40:36] The other part that's going absolutely [40:38] nuclear is the actual land and the shell [40:41] because it's turning out it's impossible [40:43] to get these approvals. Now, why is [40:45] that? And I sent Nick an image. [40:47] The reason is because underneath at the [40:49] core of it all is there's a tide [40:52] that is shifting on AI. [40:55] The American population is incrementally [40:58] getting more and more negative [41:00] on the whole subject matter writ large. [41:03] And it's not clear exactly why [41:06] they're doing that. Maybe it's the [41:08] doomerism which we talked about last [41:09] week. Maybe it's the fear of the job [41:10] loss that Jason has been talking about. [41:13] Maybe it's just this idea of yet another [41:15] wave of innovation that's only going to [41:17] benefit a few [41:19] in an extreme way minting trillionaires [41:21] all over the place while everybody else [41:23] stands still. I don't exactly know [41:26] what's causing it, but the sentiment is [41:27] shifting. [41:29] And as the sentiment shifts, the most [41:31] scaled action that they can take, [41:33] they're taking which is then they are [41:34] going and voting down data centers. [41:36] Here's an example which was insane. [41:38] A town approves a $6 billion data center [41:40] build [41:41] and half the the the board gets ousted. [41:46] Voted out overnight so that they can put [41:49] in new people to undo the decision. [41:52] So if you look at this all around the [41:54] country [41:55] the answer is not, oh we're only going [41:57] to build in Texas. That doesn't work. [41:59] There's not enough power, there's not [42:01] enough grid capacity, there's not enough [42:02] nat gas [42:04] that allows that to happen. [42:05] Maine just passed the bill [42:08] that bans all data center buildings. So [42:10] I think the reason why all birds went [42:12] crazy [42:14] is a very, very small canary in a very [42:17] important coal mine, which is we are [42:19] absolutely compute constrained. [42:22] I think if you play this out, the real [42:23] problem again goes back to Anthropic and [42:25] OpenAI. If I were them, it is a [42:27] five-alarm fire for them. [42:29] They, more than anybody else, needs to [42:31] get their hands on compute. They need to [42:33] have land power shell, but otherwise [42:35] that revenue could either slow down or [42:38] hit a wall, and it will not be because [42:41] of product quality and adoption. [42:43] It will entirely be because of the [42:45] Friendster effect. You just couldn't [42:46] keep the site up. And I think that that [42:48] that would be a huge, huge problem for [42:50] everybody. Sacks, final thoughts on data [42:52] centers [42:53] and the build-out. [42:54] >> Well, I think I think Chamath is right [42:55] that the the data center has become very [42:57] unpopular. Probably in 30 states there's [42:59] going to ban them outright. And then [43:02] it's very hard to get projects approved. [43:03] Look, I think there's a few reasons for [43:05] this. One is that there are a lot of, [43:08] let's call them real estate developers [43:09] who are kind of wildcatters who are out [43:11] there trying to get entitlements, and [43:13] they did bring a lot of projects up for [43:15] local permits where they didn't have a a [43:17] power solution. And there's no question [43:20] that local communities do not want the [43:22] data centers drawing off the grid, [43:25] thereby increasing residential prices if [43:28] that data center is not bringing its own [43:30] power generation. And the administration [43:33] agrees with this. This is why the [43:34] president did the ratepayer protection [43:35] pledge, where we got all the major users [43:38] of the data centers, you know, all all [43:40] the hyperscalers to agree that they [43:43] would not build new data centers without [43:45] bringing their own power. So, again, it [43:47] was designed to be power neutral to the [43:50] grid, or in fact, it would increase the [43:52] amount of energy available to the grid [43:53] because these data centers would give [43:55] back when they're not at peak usage. So, [43:58] that's sort of like category number one [44:00] is there's this fear of electrical rates [44:03] going up. But there's a couple other [44:04] categories I think of groups. So, the [44:06] second one was future of life and a lot [44:09] of these like doomer groups saw that [44:11] data centers were a way to stop AI [44:13] progress. And there are interviews with [44:16] some of these doomer folks who say [44:18] things like we have to meet people where [44:20] they are, meaning that they've been [44:22] unable to convince people that AI is [44:24] going to lead to the Terminator, but [44:25] they can convince them that AI data [44:28] centers are going to use up their water, [44:29] for example, which isn't true. And so [44:32] the lot of the NIMBYism has been kind of [44:34] astroturfed by a lot of the doomer [44:36] groups which have a lot of money thanks [44:39] to contributions from a few tech [44:41] billionaires like we've talked about in [44:43] the past. So that's category number two. [44:45] Category number three, [44:47] ironically again, is Anthropic itself. [44:49] It has allied itself politically with a [44:52] lot of the doomer groups, a lot of the [44:54] NIMBY groups. It didn't seem to matter [44:57] in the first couple of years because [44:58] Anthropic had made the strategic [45:00] decision not to build its own data [45:02] centers. So they probably thought that [45:04] they were just throwing sand in the [45:05] gears of OpenAI or xAI, their [45:09] competitors, and they would just rely on [45:11] hyperscalers to get their compute. And I [45:14] think that that strategy has now [45:16] backfired in the sense that they [45:18] apparently have reached the limits of [45:20] the compute that's available to them by [45:23] buying it from a third party, and they [45:24] need to build their own data centers. [45:25] It's going to be very interesting to see [45:27] how they adapt to that and how [45:29] the message around data centers changes [45:32] over the next year as let's call it the [45:34] effective altruists decide that all of a [45:36] sudden data centers or certain kinds of [45:38] data centers might be a good thing [45:40] because they serve their mission. So I [45:42] think that's going to be a very [45:43] interesting thing to watch. I'll tell [45:44] you the one thing I think you're [45:45] missing, which is that most people in [45:48] America really are starting to really [45:51] hate rich people. [45:53] And there's no physical space that [45:55] better represents the wealth in America, [45:57] the wealth creation that's happened that [45:59] a lot of people feel left behind from [46:01] than the data center. What other [46:03] physical space is there to go to? It is [46:05] the temple of the wealthy. It is the [46:09] mechanism, the tool, the machinery of [46:11] the wealthy. It is the way that the [46:14] rich, elite, tech, kind of political [46:17] connected billionaires [46:19] that we're obviously all attached to are [46:22] taking from the poor, getting themselves [46:26] ahead, shooting themselves to space, [46:28] leaving everyone else behind. And the [46:31] data center, I think, is the [46:32] representation of their progress. And it [46:35] is a representation of the progress that [46:37] others don't feel. So, that's why I [46:39] think it is physically like the [46:41] manifestation that people want to attack [46:42] and destroy. Uh there's very little [46:45] rationale about, "Oh, let's stop AI [46:47] robots from killing us." I think people [46:48] just don't see the value in AI. The [46:50] average person doesn't see the value in [46:52] AI today yet. Like we just talked about, [46:54] so much of the value of AI is showing up [46:56] in the enterprise and in the rebuilding [46:57] of enterprises. [46:59] But for a consumer's life to actually be [47:01] altered in a meaningfully positive way, [47:02] most people don't feel that yet. The [47:04] best thing they see is some medical [47:06] advice they're getting on ChatGPT or [47:07] something. And that's kind of the end of [47:09] it for them. So, I think there's a lot [47:11] of this populism that's swollen and [47:14] that's taken over not just the US, but [47:16] probably a good chunk of the West. And [47:17] and the data center is the target. [47:20] It is the pied de terre of this space, [47:22] in a way. This is their attack vector. [47:24] >> To David's point, they'll probably ban [47:26] data centers in 30 states. [47:28] Data centers can output and input at the [47:30] speed of light, so they can be anywhere. [47:32] I mean, you know, the reason you put [47:33] data centers in different states is [47:34] cheap power, low latency. So, maybe you [47:37] get a couple of milliseconds of latency. [47:38] So, the data centers can go anywhere. [47:40] So, as soon as all the states start [47:42] banning the data centers, the data [47:43] centers will just go to space, to [47:46] Iceland, to Texas, to wherever. And then [47:50] and then the data center and then [47:51] everyone will be forced to move on, [47:52] Travis. They'll basically find the next [47:54] uh a for populism, which will be [47:57] something else. The pied-a-terre, the [47:58] data center, what's next? You know, this [48:00] is this is part of what's going on right [48:02] now. Private jet fuel. [48:03] >> Where we've got this kind of level of [48:05] debt from the I always bring it back to [48:07] this because I do think this is at the [48:08] root cause of populism is the fact that [48:11] the government promised so much, it's so [48:12] inefficient, and as a result, it's [48:14] destroyed the value of the dollar and [48:15] not given anyone anything that they [48:17] thought they were getting. And we do [48:18] have the ratepayer pledge, Sax, to [48:20] address the energy issue. That was a big [48:22] win, I think, and you worked on that, I [48:24] believe. Yeah. But hold on, can I just [48:27] say something? I think that ratepayer [48:28] pledge is important. It doesn't change [48:31] the business model of the utility. [48:33] Meaning, Mhm. [48:35] what the utility is allowed to do is [48:37] every year build a budget. [48:39] And the way that they decide how to [48:41] charge you [48:42] is that they are allowed to make [48:44] investments, right? [48:46] And then they're allowed to earn 10% [48:49] roughly [48:51] on the amount of investment that they [48:52] make. What do you think their incentive [48:55] is? [48:55] Their incentive is to find ways to keep [48:57] investing and upgrading the [48:58] infrastructure. [49:00] So, I suspect what you'll see is, [49:01] independent of what the data centers do [49:05] on site, which I think is smart and [49:06] good, and I'm glad you guys did that, [49:09] it doesn't slow down the actual [49:11] fundamental business model of the [49:12] utility because those are local monopoly [49:15] licenses that are granted at the state [49:18] and county level. [49:19] And so, when you go inside and you look [49:21] at those utilities, what they're allowed [49:22] to do is say, "Well, I'm going to bury [49:24] the lines underground because there's a [49:26] wildfire threat. That's going to cost 10 [49:28] billion." And they present that, the PUC [49:30] has to say yes, and they're allowed to [49:32] make 10% on 10 billion. [49:34] So, the business model of the utility [49:35] has to be looked at because it is [49:37] independent of all of these other [49:38] things. And they have an incentive, like [49:40] insurance companies, to just walk prices [49:43] up over time. It's true. That's why I [49:44] think behind the meter so so important [49:46] there, for sure. [49:48] Yeah. But look, I mean, this idea that [49:49] that data centers don't create jobs is [49:51] just wrong. I mean, it's been a huge [49:52] boon for blue collar jobs and the [49:54] construction industry. While they're [49:57] building the data center. Yeah, but that [49:58] that construction could be a wave that [50:00] goes on for a decade or two decades. I [50:02] mean, the CapEx is not going down, it's [50:04] increasing. Listen to, for example, what [50:06] Jensen says about his projections. I [50:08] mean, it's not like the CapEx is a [50:09] one-year thing. We're seeing tens of [50:11] thousands of new construction jobs being [50:13] created and 25 to 30% higher wages for [50:17] electricians, carpenters, the guys who [50:20] hang drywall, pour concrete, create [50:22] roads, install equipment. I mean, these [50:24] are blue collar jobs and it's creating, [50:27] again, not just new jobs, but also wage [50:29] increases. It's a good thing. It's not a [50:32] bad thing. I think the issue is the the [50:33] issue is with [50:34] >> but it's not the same as a fab where the [50:36] jobs are permanent and they're there. Uh [50:38] we can debate it all we want. You can [50:41] There's about 100 data centers right now [50:43] that are being contested. The data is [50:45] for every 100 that are contested, about [50:47] 40 get canceled. That number is [50:50] increasing for this year. It's more than [50:51] doubled already from what the number was [50:53] last year. [50:54] And the total economic value of those [50:56] 100 data centers right now is about 162 [50:58] billion dollars. Jason, the the thing is [51:00] is that there are permanent jobs because [51:03] what Where's the energy that's powering [51:05] that? Where's the semiconductors that's [51:07] powering that? You know, there's a lot [51:10] of [51:11] We we actually had um Chase from Crusoe [51:14] and Michael from Coreweave on the All-in [51:17] interview show and they said they're [51:18] bringing energy with them. That's their [51:20] big thing. Bring BYOE, bring your own [51:22] energy to the space. They're bringing in [51:25] natural gas, they're bringing in diesel [51:27] fuel, they're bringing in solar. BYOE is [51:29] the model. Well, can can I also say, [51:31] listen, if if you were of the opinion [51:34] that you didn't want more data centers [51:36] in America because it was using up, say, [51:38] scarce energy resources or something [51:40] like that, but you weren't just [51:42] anti-progress altogether, then what you [51:44] would want to do is at least see more [51:47] data center construction among US [51:48] allies, right? Energy-rich US allies. [51:52] Well, a year ago, that's one of the [51:53] things I worked on was allowing the Gulf [51:55] states to build data centers with [51:58] American technology for American [52:00] companies. And remember the controversy [52:02] that caused? Everyone accused us that [52:04] somehow this was like serving China. [52:06] Well, those data centers are getting [52:07] bombed, and I don't think they'd be [52:08] getting bombed right now by Iran if they [52:11] were serving China. These were [52:13] >> They're very strategic, yeah. The IRGC [52:15] has put these data centers on a list of [52:17] assets that they've threatened to [52:19] destroy because they're American assets [52:22] or [52:23] >> Yeah. they're assets of our Gulf state [52:24] partners in partnership with America and [52:28] American companies. This whole idea that [52:30] somehow this was a threat to American [52:32] national security was a total hoax, just [52:35] like the data center uses too much [52:37] water. And what I'm saying is that a lot [52:39] of the same forces were behind this hoax [52:41] as are behind these data center hoaxes. [52:43] Now, Freeberg is right about the [52:44] resentments, but those resentments get [52:47] whipped up and marshaled by people who [52:50] have an agenda, and that agenda is [52:52] well-funded and strategic. And I can [52:55] tell you that in the case of the GCC [52:57] data centers, Anthropic was adamantly [53:00] opposed to that, and they were lobbying [53:02] against it, and they were again salting [53:04] the earth against those projects. And I [53:06] think the whole issue's kind of moot now [53:07] anyway because I think the data centers [53:09] have been blown up, but uh or they've [53:11] been threatened to, but it just shows [53:13] how ridiculous [53:14] and and what a psyop some of the [53:16] opposition to these ideas are. Mhm. And [53:19] it's a cell phone in many ways. If you [53:21] think about who are the top two [53:23] spokespersons for our industry and what [53:26] they're communicating to Americans [53:29] versus say how the Chinese view AI, [53:31] which is incredibly positive, we've got [53:33] Dario who says it's the end of days, [53:36] everything's going to be hacked, all [53:38] your files are going to be hacked, all [53:40] your accounts are going to be hacked. [53:41] And then on the other And everybody's [53:43] losing their job. That's our top [53:45] spokesperson and Dario and Anthropic. [53:47] The other top spokesperson just had a [53:49] 70,000 word piece written about him [53:51] where Ronan Farrow said, "I had a dozen [53:54] people tell me unprompted Sam Altman's a [53:57] sociopath." Those are the top two [53:59] spokespeople. Those people cannot be the [54:01] spokespeople for this industry, and this [54:03] industry's got to get focused on fixing [54:05] the big three. Healthcare is going to be [54:07] dramatically improved by AI. Housing [54:10] could be dramatically improved. I'm not [54:11] sure who's working on that. And [54:14] obviously education and the cost of [54:15] education. We need somebody out there, a [54:18] Michael Dell, you know, Jensen, Elon, [54:21] just explaining how we're all positive [54:23] this could be because right now AI is as [54:26] popular as is less popular than ICE, the [54:29] Democrats, and the government of Iran. [54:32] Joe Lonsdale at Alpha School. On the [54:34] education side, I think you just have to [54:35] look at Alpha School. It's It's working. [54:37] All right. So, that's your story, uh [54:40] Chamath. That's the story of All Birds, [54:43] the most overvalued startup. But uh [54:45] we're going to play a little uh cue the [54:47] music here. We're going to play a little [54:48] game show for everybody. This is The [54:50] Price uh David Sacks is wrong. Yes, this [54:53] is The Price is wrong. The price of the [54:56] startup, Travis, is wrong. [55:00] And this is the game show where we guess [55:03] which overvalued disaster are we talking [55:05] about. First up, hailing from Austin, [55:07] Texas, an enforcer of the PayPal mafia, [55:10] it's Mr. David Sacks. Welcome to the [55:11] program, David Sacks. Are you ready to [55:13] play The Price is wrong? [55:15] >> Let's do it. Let's play along. All [55:16] right. Here we go. The Price is wrong. [55:19] This startup was once valued at over $13 [55:21] billion. [55:22] Their main business model, selling JPEGs [55:24] for fake internet money. People lost [55:26] their minds and started spending [55:28] millions of dollars on monkey images. [55:30] Play the thinking music, please. [55:33] David Sacks, [55:35] can you name that startup once billion? [55:39] 13 billion for Is it Is that like the [55:41] board apes thing? [55:43] Okay, you're closing in. Remember format [55:45] in the name of the question. [55:47] >> I know I know They're selling JPEGs for [55:49] fake internet money. Me me me pick me [55:51] pick me pick me. [55:52] >> you'll get your chance Chamath. [55:55] Oh, I think I know what you're talking [55:56] about. All right, let's move on to the [55:58] next I can't remember. [55:59] They're spending millions of dollars. [56:01] Okay, Can I say it? Can I say it? Can I [56:02] say it? [56:02] >> time Chamath. It's your turn to steal. [56:04] OpenSea OpenSea Correct. 100 points for [56:08] Chamath Palihapitiya. Okay, an amazing [56:11] steal an amazing steal. And next up that [56:13] puts [56:14] >> Okay, here we go. He puts the dick in [56:16] dictator hailing from Palo Alto [56:18] California. It's your favorite [56:20] DoorDasher Chamath Palihapitiya. How [56:22] long you been DoorDashing there Chamath? [56:24] Nine years. Nine years. You like that? [56:27] What do you like about that? Is it the [56:28] interaction with the people? Is it the [56:29] tips? Is it stealing a couple of french [56:31] fries? What do you like about being a [56:32] DoorDasher there? I typically take four [56:34] fries. I lick the burrito. [56:36] Okay, licking burritos. Did he just say [56:38] licking burritos? [56:39] >> why we need robot delivery robot [56:41] delivery drivers. I mean it's just [56:44] >> It's coming. Okay. We don't want our [56:46] french fries being licked. Let's get the [56:48] music going here. This startup once [56:50] valued at 4 billion Oh. allowed you to [56:53] talk to other people on your mobile [56:55] device during COVID. And you could [56:57] listen to mid VC Yes yes yes I got it I [56:59] got it. It's called generic startup [57:01] device. Can you name that overvalued [57:03] startup? [57:03] >> god, it's called Clubhouse. Oh my god, [57:05] he's on a heater. That's 200 points for [57:08] Chamath Palihapitiya. Hey, that's a lot [57:10] of DoorDashes. He's going to do very [57:12] well here in the final round. Okay, [57:14] >> memory. How How soon we forget. I can't [57:16] I couldn't place any of these names. [57:18] Okay, finally streaming in from a potato [57:21] field at an undisclosed location in [57:23] Idaho. It's It's Dr. David Friedberg. [57:26] You're You're in the potato sciences, [57:28] correct? Mr. Friedberg. [57:30] >> That's right. Potato science. You like [57:31] the potatoes. You have a favorite potato [57:33] dish. Is it the scallops? What do you [57:35] like to do with your potatoes? What is [57:36] your name, Mr. What is the host's name? [57:38] What is your name, host? It is me, uh [57:40] Jocular Jake Hal. Let's go. Here we go. [57:42] Jocular Jake Hal here at the helm. [57:45] You you you like a certain type of [57:46] potatoes there? What do you like? [57:48] Jacular Jacular It's Jacular Calcaneus. [57:50] And what do you like there on the [57:51] potatoes? What's your favorite there? [57:52] You like a creamy mash? Little little [57:55] little little cheesy You're liking [57:57] you're liking the cheese. Okay, very [57:58] good. Okay, David, this startup was once [58:00] valued at $270 [58:02] million. This is your favorite. I [58:03] understand you're a vegan. They sold [58:05] juice. These juice packages went to a [58:08] juice machine. I know I know I know I [58:09] know I know I know I know I know I know [58:10] I know I know I know I know I know I [58:11] know I know I know I know I know I know [58:11] I know I know I know I know I next [58:12] iPhone, the iPhone of juice. Can you [58:14] name Oh, I know it. [58:15] >> I know it. I know it. Pick me. Juicero I [58:18] know it. I know it. I know it. Pick me. [58:19] Juicero Juicero Yes, that's right. [58:21] That's 100 points on the board for you. [58:23] Wait, are you Wait, are you searching? [58:24] It looks like he's searching. Did he [58:26] search? Oh my god, are you on your [58:27] device? He looked like he was on his [58:29] device. Internet free. Internet free [58:31] here. [58:32] He's on his device. All right, folks, [58:33] there you have it, a clear winner, [58:34] Chamath Palihapitiya. Tell him what he's [58:37] won. He's won a trip to [58:40] Temptations 2 in Cabo. You're going [58:42] directly to Temptations 2, an adult [58:44] resort. Uh that's for you and two of [58:46] your friends, okay? Enjoy Temptations 2, [58:48] Chamath Palihapitiya. Two of your [58:50] friends. Uh What a fun game this is. You [58:53] can bring your thrumble. I think this is [58:55] going to need to be a come a regular [58:57] feature of the pod. Okay, landing. We [58:59] find out Everyone forgets the unicorns [59:02] that don't work. They just like [59:03] disappear into the ether. [59:05] All right. All right. There you go. You [59:07] won a pied-à-terre on a uh [59:11] on a You can fish with a tax bill in New [59:13] York. You got a pied-à-terre available [59:15] to you as a your first-place winner, [59:17] Chamath Palihapitiya, DoorDash or food. [59:20] Hello help a pied-à-terre. All right, [59:22] listen, uh there's been a lot of [59:24] shenanigans going down in DC, the most [59:26] boring city in the world, where [59:27] apparently everybody's uh uh getting a [59:30] little frisky after hours. TMZ launched [59:33] a news bureau. [59:35] Eric Swalwell is out of the governor's [59:37] race. There's a lot of dark stuff that's [59:41] been released. He is innocent until [59:43] proven guilty, but it's not looking [59:46] good. He also resigned from Congress, [59:48] Jake Hal. [59:49] And he resigned from Congress, as well. [59:51] And Friedberg, [59:52] our investigative journalist, now [59:56] working as a stringer for TMZ. [59:58] Friedberg, what have you learned? What's [60:00] in Friedberg's day? [60:02] >> learned anything. I [60:03] My um Okay. You have an analysis, then, [60:05] maybe. My anecdote on this is [60:09] back in December, when [60:13] it was first rumored that Swalwell was [60:15] going to run for governor, [60:17] I started making some calls to various [60:21] folks to be like, "Hey, you know, [60:24] what do we think of this guy? Is he [60:26] going to be a good candidate? You know, [60:28] obviously I and a lot of other [60:30] people, before they evacuate the state, [60:32] care a lot about the future of [60:33] California. [60:35] So, I started checking around. I spoke [60:36] to several people who independently told [60:40] me that there's knowledge about this guy [60:43] sending, you know, pics to employees, [60:46] and that this guy has a bunch of stuff [60:49] that's going to come out about him." [60:51] So, I heard all of this not from one [60:53] person, but from several different [60:55] sources. This was back in December, [60:57] going into January. And I largely kind [61:00] of dismissed it, cuz I was like, "If [61:01] this is true, this would have all come [61:03] out already. I'm like, there's no way [61:05] this is true. If it was true, like, [61:06] people would have talked about it, they [61:07] would have made a thing about it. If [61:09] multiple people are telling me about [61:10] this, [61:11] then I've got to assume [61:13] that it's a rumor that's being used to [61:15] block him from running for governor, [61:17] versus it being a real thing. Because [61:19] multiple people had this knowledge and [61:21] this information. [61:22] So, this was December, January where I [61:24] had these conversations and at that [61:26] point nothing had come out. So, I was [61:27] like, "Okay, it doesn't seem like this [61:28] is real." And then everything that I had [61:31] been told started to come out in the [61:33] last week. So, the striking aspect of [61:35] all of this for me was how much [61:37] knowledge there was about these various [61:40] incidents with the guy. [61:42] How so many people had this knowledge [61:44] and how no one had actually brought the [61:46] knowledge to bear, which begs the [61:49] question, [61:52] why did they not do what was right by [61:54] the victims? Or why did the victims sit [61:58] on the sidelines waiting for the right [62:00] moment to all come out together? [62:03] Because this was broad knowledge within [62:05] a community of people and they made the [62:07] choice not to bring it forward with that [62:09] knowledge. And that's what was so [62:11] striking to me about this whole thing. I [62:12] had honestly dismissed the whole thing [62:14] as just being rumor-mongering to try and [62:16] besmirch the guy. And it turned out that [62:18] these were all being held back [62:20] purposefully and deliberately for a very [62:22] particular moment in time when they were [62:24] all brought forward to be used. And let [62:26] me just be clear, these are all [62:28] allegations. Nothing's been proven in [62:29] court. He does get his day in court. [62:31] Absolutely. And and and again, we just [62:34] want to make sure there's a bunch of [62:35] allegations here. Yeah, [62:37] and all that I'm And all that I'm saying [62:39] is that people had told me about these [62:42] supposed claims five months ago, four [62:45] months ago, multiple people and had [62:46] chosen not to bring it forward and that [62:48] the victims had not come forward [62:49] publicly with these And then there was [62:51] this coordinated effort to bring [62:53] everything forward at the same moment. [62:55] And that's what was so striking to me is [62:56] just how coordinated all of this was. [62:58] Who's controlling it? Who do you think [63:00] like Is there like a meeting? Is there [63:01] like a [63:03] a council? [63:04] >> Nancy Pelosi. [63:06] Jesus Christ, it's so [63:07] >> know if that's true. Well, I don't not [63:09] just her. I'll tell you my sense of it. [63:11] My sense of it is that there are certain [63:13] insiders and he's not an insider. [63:15] And those insiders are like Katie [63:17] Porter's of the world. Katie Porter, I [63:19] think, is who the Democratic [63:21] establishment wants to be governor. [63:23] She's an insider to [63:25] national Democrat. She's an insider to [63:28] California Democrats. [63:30] And I think that she's the preferred [63:32] candidate. Tom Steyer She's a spousal [63:34] abuser. I don't know that Katie Porter [63:36] is like this ultimate insider, but I [63:38] think there are Look, there are clearly [63:40] insiders. The Democratic Party is a [63:42] machine [63:43] that exists to siphon off as much money [63:45] as possible from the public till to the [63:47] interests that support the party. And [63:50] it's their gravy train. And they're not [63:51] going to let anyone stop that gravy [63:53] train. [63:53] >> Zach. I don't know if it's just the [63:54] Democratic [63:55] >> They're not going to let that gravy [63:56] train stop for 1 second. Now, the [63:58] Democratic Party had a huge problem in [64:00] this California governor's race, which [64:02] is that the Democratic field was very [64:04] fragmented. And so, the two Republican [64:06] candidates actually were polling the [64:08] highest. And so, that just so the [64:11] viewers have context, California has [64:12] this weird jungle primary system where [64:14] the top two go to a runoff. They don't [64:16] have a Democrat lane and a Republican [64:18] lane. They just take the top two jungle [64:20] primary and then they go to the runoff. [64:22] Even today still, Hilton and was it [64:25] Bianca or whatever, they are polling at [64:27] both around like 14 or 15% and if the [64:29] election were held today, you'd have two [64:31] Republicans in the runoff. So, the [64:33] Democrats needed to winnow the field [64:36] down and have fewer candidates. In [64:39] addition to that, they must have been [64:40] concerned that all of this oppo [64:43] onslaught would come out once it was him [64:46] versus say Steve Hilton. And they didn't [64:48] want it to come out later when they [64:51] could lose the election. So, the powers [64:53] that be made the decision to lance the [64:55] boil. Probably [64:57] there was a conversation with him to [64:58] tell him to get out of the race. He [65:00] didn't listen. And by the way, remember [65:01] this feels a lot like what happened with [65:03] Joe Biden when he had to drop out of the [65:06] presidential race. Remember [65:07] >> surprisingly like that. [65:09] >> Democratic establishment and all of the [65:11] mainstream media were saying that Biden [65:13] was sharp as a tack, okay? And then he [65:16] had that disaster debate performance [65:18] with Trump and it became clear that [65:21] >> Yeah, and you could just see the text [65:23] messages were flying during the debate [65:25] between the Democratic Party insiders. [65:27] And by the time that debate was over, [65:29] they had congealed on a new position, [65:31] which is that Biden had to step aside. [65:33] And then Nancy Pelosi was reported as [65:35] having gone to the president said, "We [65:37] can do things the hard way or the easy [65:40] way." Imagine that. Like telling the [65:42] president of the United States, "We can [65:44] do things the hard way or the easy way." [65:46] And then Biden disappeared for a week [65:48] and magically he stepped aside by tweet. [65:52] Remember that? He published a statement [65:54] that appeared to be done by an auto pen. [65:57] People were speculating whether he was [65:59] even behind this or whether the staff [66:00] pushed him to do it. The whole thing was [66:02] extremely weird. But he, you know, was [66:05] clearly muscled out of it. Just a few [66:07] days before he had said, "I'm not [66:08] leaving the race no matter what." I [66:10] mean, it was like straight out of that [66:11] Wolf of Wall Street meme. "I'm They're [66:14] not getting me out of here." And then a [66:15] few days later he's resigning by tweet. [66:19] And again, Nancy Pelosi appears to be [66:21] the figure at the center of both these [66:22] things. [66:23] She was Hold on. Hold on. Pelosi Hold on [66:25] a second. Pelosi is reported as having [66:28] been Swalwell's mentor. I guess she [66:30] found him roughly 20 years ago to run [66:33] for Congress in the first place. When [66:35] the Republicans wanted to kick Swalwell [66:38] off of the Intelligence Committee, the [66:39] House Intelligence Committee, for [66:41] allegedly being involved with that [66:43] Chinese spy Fang Fang, [66:45] it was Pelosi who protected him. So, [66:49] she's sort of been a central figure in [66:51] his career. I'm not saying she approved [66:54] of anything he did, but [66:55] >> Listen, politics is dirty. [66:56] >> But there's no there's no way [66:58] that that button gets pushed [67:00] without going to Pelosi for the [67:02] sign-off, right? I mean, she's like the [67:04] boss of this operation. [67:06] And I think that the same thing happened [67:08] to Swallow that happened to Biden is [67:10] they went to him and said, "We can do [67:11] things the hard way or the easy way." He [67:13] was too dumb to listen and they did [67:15] things the hard way. [67:16] Yeah, shout out Nancy Pelosi, incredible [67:19] day trader and shout out to friend of [67:22] the pod Ro [67:23] Bro who's now beaten Nancy Pelosi. I [67:25] think his trades this year he's he's [67:27] actually even beaten [67:29] Nancy Pelosi. That was the that was the [67:31] difference. [67:32] >> Ro Khanna has traded $600 million of [67:34] stock. He trades more frequently than [67:36] Citadel Securities. What? I mean it's [67:38] incredible. Yeah, where's his PR [67:40] director? [67:41] I want to get in on it. The congressman [67:43] you've been supporting for years tax is [67:45] great stock trader. He's Yeah, [67:47] apparently we should have been talking [67:48] to him not about taxes. We should have [67:50] been talking to him about trades. There [67:52] were things I liked about Ro Khanna. I [67:53] mean he did support freedom of speech [67:55] with the whole Twitter thing. You know [67:58] what it was is that the Overton window [68:00] shifted so much. [68:01] >> You supported him? Yeah, when I [68:02] supported Ro Khanna freedom of speech [68:05] was a big issue and he was one of the [68:06] only Democrats to support that. He was [68:08] also I think the only member of the [68:10] progressive caucus to support a [68:11] diplomatic track for Ukraine which I [68:13] supported and I gave him credit for [68:15] that. And yeah, I knew that he was in [68:17] favor of wealth tax but I thought that [68:18] was just a very unserious proposal that [68:21] you know, it wasn't in play in any way. [68:22] Well, the Overton window shifted so much [68:26] that now the wealth tax really is a [68:27] possibility. So, you know, things have [68:29] changed. [68:30] Here it is. Shout out to our boy Ro [68:32] Khanna. Look at this. Right about now [68:35] Chamath's thinking about making a [68:36] managing director offer here is I mean [68:38] Ro, if it doesn't if you get booted out [68:40] of office you could become a managing [68:42] director at Alt Capital. [68:44] >> statistically like even if you had [68:46] insider information making that much [68:47] money is like very hard. Holy cow, Nancy [68:51] Pelosi is going to shiv Ro Khanna next [68:54] for not giving her the uh [68:56] giving her the uh inside track on [68:58] whatever he's betting on. What did he [68:59] bet [69:00] >> Well, look any any as we all know any [69:02] investor can have a good quarter or a [69:04] year, but to put up the kind of returns [69:06] that Nancy Pelosi has done over decades [69:08] is nothing short of miraculous. It's [69:10] generational. It's a generational run. [69:12] Give her her flowers. She's [69:14] substantially better than Warren [69:15] Buffett. [69:19] I mean, it's Warren Buffett, Stan [69:21] Druckenmiller, and Nancy Pelosi are [69:24] three of the most accomplished investors [69:25] of all time. [69:27] Whatever you did to the poor Buffett [69:29] fans, Chamath, they are incredibly angry [69:33] at you for desecrating his legacy. [69:37] I didn't desecrate his legacy. I pointed [69:38] out one [69:39] unavoidable fact, which is his returns [69:42] were bimodally distributed pre- and [69:44] post-Reg FD. [69:46] When you have to follow the rules of [69:47] disclosure, everybody's returns got [69:50] knee-capped. [69:51] When there was no disclosure rules, [69:54] his returns were off the charts. That's [69:55] just a mathematical truism. Now, what's [69:58] interesting to note is the reason why [70:00] Nancy Pelosi's returns are so [70:01] consistently good is Reg FD does not [70:03] apply to people in Congress. [70:06] That should be the takeaway. They can [70:08] learn things [70:09] >> to stop them. They can learn things in [70:11] their committee meetings. [70:13] In fact, there are situations where [70:15] things are disclosed and then they are [70:17] trading in real time. Shout out to the [70:19] SCIF. Yeah, get out of the SCIF quick [70:21] and get that trade in. [70:22] >> So, I have enormous respect for Warren [70:24] Buffett and what he's done. It's It's [70:25] He's He's the goat of goats. But, the [70:27] returns post-Reg FD are They are just [70:30] materially worse than they were pre-Reg [70:31] FD and that's just What's he going to do [70:32] with that cash position, Chamath? [70:34] Speaking of Buffett, they The market [70:36] What? 300 billion sitting there? The [70:39] market is in a very complicated moment [70:40] right now. If you look at [70:42] historical indicators of value, [70:44] so if you look at Schiller as an [70:46] indication of value, it's peaking. [70:50] If you look at the Buffett index, [70:53] it's peaking. [70:54] So, there are things that when you look [70:56] at it, look like all-time highs. [70:59] And the problem with that is you would [71:01] say, "Oh, man, but [71:03] there's this weird dispersion happening [71:05] in the market." Dispersion means [71:07] literally a few companies are hitting [71:09] all-time highs. I think it's like eight [71:10] or nine. [71:12] And everybody else is not. So, it's a [71:14] really complicated moment. It's hard to [71:16] understand what's going on, but he's got [71:18] a lot of cash. If he's sticking to his [71:20] knitting, [71:21] he's looking at the Schiller index, and [71:23] he's looking at his own indicator, which [71:24] shows all-time highs, and he's waiting [71:26] for a correction. He's not really in [71:28] charge anymore. Like Berkshire Hathaway [71:29] is not Warren Buffett anymore, even [71:31] though he's a big owner, he's not really [71:33] doing it. But, I mean, the fact that [71:35] they're sitting on that massive pile of [71:36] cash says something. That they're not [71:38] putting it to work in the market says [71:40] they don't see an opportunity yet. [71:42] Sachs, Travis, I'm curious your takes. [71:45] Uh either one of you can go in whichever [71:46] order you want. On how is the market [71:50] crushing it while we're in week seven of [71:52] a war, and we put a hundred billion [71:55] dollars or something into this [71:57] military activity in Iran, and [72:01] the market is pricing it in, shrugging [72:03] it off, and we're hitting all-time [72:04] highs. Sachs, I don't know if you're [72:06] feel comfortable. [72:07] >> Yeah, yeah. Yeah. [72:08] >> Look, I think it's pretty [72:09] straightforward, which is that in the [72:12] wake of the meeting in Islamabad, that [72:14] the market is feeling confident and [72:17] pricing in that the war is going to get [72:21] resolved. The president also recently [72:23] said that it's very close to being [72:25] wrapped up. The military [72:27] objectives are close to being achieved, [72:30] and he's made it sound like it's going [72:32] to be resolved. Yes, a deal was not [72:35] signed in Islamabad. I always thought [72:37] that was an unrealistic expectation that [72:39] these two countries, which are at war [72:40] with each other, and in fact have had [72:42] hostile relations for almost 50 years, [72:45] are going to resolve all their [72:46] differences in 24 hours. It's not [72:47] realistic, but the impression that the [72:50] market I think has, and clearly is [72:51] trading on this is that that war is [72:54] going to be what Donald Trump said, [72:55] which is an excursion and something that [72:58] is on its way to being resolved that [73:01] they've made progress. And J Kyle, [73:03] you're right. All week has been spending [73:05] incredibly strong. I think by Tuesday [73:08] the market had recovered [73:10] all of its losses since the start of the [73:12] war. I think it made a new high [73:14] yesterday on Wednesday and is making new [73:16] highs, fresh highs today on Thursday. [73:19] So, you just have to say that at the [73:21] present time [73:23] the market thinks and I I would consider [73:25] the stock market to be the ultimate [73:26] prediction market that this war is on [73:29] its way to being resolved. Just to [73:31] qualify, I'm not speaking as a member of [73:33] the administration. Of course not. I'm [73:35] not saying that I know something, okay? [73:38] >> know anything. You don't know anything. [73:39] I got it. There's no no clipping. [73:42] I'm not representing anyone and I don't [73:45] know anything different than what any of [73:47] you know. I'm just saying that I think [73:50] this is what the market is clearly [73:52] pricing in and I'm paying attention to [73:54] what the statements are of the president [73:57] and vice president. And this Travis, you [73:59] and I were talking the other day while [74:00] you were [74:01] Yeah. slaughtering me in backgammon when [74:04] you won our eight-point match, you're [74:06] almost caught up. [74:08] About the taco trade. Not Trump always [74:11] chickens out. You have a theory about [74:12] Trump always cares about optics. Unpack [74:15] it for us, Travis. Yeah, so while I was [74:18] beating you in backgammon Ooh. [74:21] I sort of had a [74:22] You know, there's like the in cuz J Kyle [74:24] asked me and the inside is is like and I [74:27] I think this maybe is just the more [74:29] simple view of things, which is [74:32] Trump's weather vane is the stock [74:34] market. [74:35] Like we see Vola's up, we see the Vix is [74:38] really high, but you know what? The S&P [74:40] is trading in this band that's actually [74:42] fairly tight given the crazy that's [74:44] going down. [74:46] And everybody's nervous, but actually he [74:49] he moves in the policy space. He's he he [74:53] does not let the S&P go down too low. [74:56] People also get the sort of the [74:57] panicking thing now, which is like he [75:00] gets people nervous, he makes moves, and [75:03] then he comes back to sort of reality [75:05] and gets things done is it practical and [75:08] and probably the the better parts of [75:10] what he he might be, you know, what he [75:12] does well, and people are pricing that [75:14] in. So, there's there's all the things [75:16] Sachs is saying about the nuances and [75:18] the details, but the sort of super high [75:20] level like I'm in lower earth orbit view [75:24] is like [75:25] the stock market is Trump's weather [75:28] vane. And he's he will go to the place [75:31] that makes the stock market come up. And [75:33] if he's go and and if it's up and it's [75:35] too high, he almost feels like it's too [75:37] easy, so he makes it hard on himself [75:39] again, and then he brings it back up, [75:41] and the traders are getting used to it, [75:43] I think. Yeah, it's like playing poker [75:45] with Chamath with Alan Keating. It's [75:46] just like, how is this guy solvent? He's [75:49] literally playing every hand of poker. [75:51] He's losing tons of money, and then all [75:53] of a sudden by the end of the night, he [75:54] hits two big pots, and he's got the nuts [75:56] both times, and he pulls out of the [75:58] stall. What's your take on the market [76:00] today, Chamath? [76:03] Yeah, I mean, I'll just do it again, but [76:04] I think the the Shiller PE, Nick, I sent [76:07] it to you. [76:08] Shows near all-time highs. Then the [76:10] second is the Buffett indicator, which [76:11] is the sum of all [76:13] US equities divided by GDP. [76:15] It's also at all-time highs. [76:18] So, this would generally mean [76:21] that you need to be increasingly a [76:22] little bit more risk off. But then the [76:24] opposite side of that, and I sent you a [76:26] third one, and this is why it's so [76:27] confounding, is you have signals showing [76:29] everything, which is which typically [76:31] doesn't happen. [76:32] And this is this dispersion point where [76:35] when you see this performance and it's [76:37] up 5% [76:40] in the first, you know, half of April, [76:41] typically the market is up almost 32% on [76:44] average for the rest of the year. And we [76:46] were already up, as Sac said, you know, [76:48] 7 and 1/2% already in the first 10 [76:50] >> read this graph. This is crazy. I don't [76:52] even know what's going on here. [76:53] >> Hey, this is just a this is just a [76:54] dispersion. But the [76:56] the the idea of all of this is I think [76:59] we're in a moment where you can find a [77:01] piece of data [77:03] to underwrite your bias. [77:05] And I think that's where there is a lot [77:07] of danger. I don't know, for me [77:08] personally, I'm generally more risk off [77:10] right now. [77:12] And more importantly, I'm waiting for [77:14] these IPOs so that I can, to be very [77:16] honest with you, delever and get some [77:18] chips off the table. I think that [77:21] it is crucial that the SpaceX IPO get [77:23] done ASAP. And then I think it's even [77:27] more crucial [77:29] that one of Anthropic and OpenAI [77:30] front-run the other one [77:32] and get out first. [77:34] And the first two charts are the reasons [77:36] why. [77:37] I it's fairly obvious to me what's going [77:39] on here. It you have [77:41] uh these multiple trends going on at the [77:43] same time. The reason people are traders [77:45] are valuing this, and I think there is [77:46] some smart money in the market right [77:48] now, is what you're seeing in terms of [77:50] the earnings potential of these [77:52] companies as they deploy AI, as they [77:55] have unlimited intelligence, and the top [77:57] employees at the top companies become 10 [78:00] or 20 or 30 times more productive. [78:02] That's never been seen before. Microsoft [78:04] Office may have made you 30% more [78:07] efficient. The internet may have made [78:08] you 50% more efficient. But none of [78:10] these things made you 10 times. And you [78:11] had a really interesting point that you [78:13] slipped in earlier, Chamath, which was [78:15] nobody knows how to harness these things [78:17] yet, right? And it's producing slop. The [78:19] truth is, 10% of people do know how to [78:21] harness it, 20% maybe. And I watch this [78:23] in my own organizations, and I watch it [78:25] in 600 portfolio companies. The ones [78:27] that do deploy it correctly, they are [78:30] running the table on the ones that are [78:32] not. So, the efficiency boom that we're [78:33] going to see at these companies, whether [78:36] it's Meta or Uber or Airbnb [78:38] >> maybe I'm just an idiot. Whoever [78:39] executes it properly is going to be [78:40] phenomenal. [78:42] The earnings will be insane. Maybe I'm [78:44] an idiot, but [78:45] it has not translated into a tsunami of [78:48] more revenue and more profit for me yet. [78:51] Maybe I'm the only one and maybe it's [78:52] everybody else but me. [78:54] But I haven't seen it. [78:57] I see it in a lot of companies. So I [78:58] think in private companies, we have two [79:00] companies as just two examples, MicroOne [79:03] which is doing data, you know, dark [79:05] pools of data for these language models [79:07] and they have built technology that [79:09] allows them to build data and allows [79:11] them to collect data to help the large [79:12] language model companies grow and they [79:14] are on a tear by using this to identify [79:17] and find great people who can then [79:19] contribute to these corpuses and then we [79:21] have one tax GPT that is making the [79:23] accountants I think they have six or [79:24] seven percent of all accountants using [79:26] their platform. [79:27] >> you agree with me? They're just ripping [79:28] I hope you can agree with me. [79:30] I hope you can agree with me with the [79:31] following statement. [79:33] Small companies enabling at the edges is [79:36] not where these guys [79:38] will build a multi-trillion dollar [79:39] market cap. I think we're we're both in [79:41] alignment. Big companies, dumb [79:43] companies, slow to implement the [79:44] technology. Startups No, they are dumb [79:47] when it comes to implementing new [79:48] technology. They're always the laggards [79:50] when it comes to implementing new [79:51] technology. The reason that they are [79:52] slow is not because they are dumb. The [79:54] reason that they're slow is their [79:55] business is much more sophisticated and [79:57] much more complicated than many other [79:59] businesses. And what I'm saying is if [80:01] you can't prove [80:03] that this works in the big time, prime [80:06] time, big league use cases, [80:09] it's a toy. [80:11] All right, Sachs, where where where do [80:12] you stand? Are you in the J Cal position [80:14] the startups are showing the way, [80:15] they're being massively efficient, [80:16] they're growing revenues like we've [80:18] never seen in the startup community [80:19] before with less people [80:21] or you're in the Chamath camp, hey, big [80:23] companies are not having a drop to the [80:25] bottom line and they're smart or both [80:26] things are true. Where do you stand, [80:27] Sachs? Be the adjudicator of this case. [80:30] >> I mean, honestly, I'm probably closer to [80:31] you, J. Cal. I hate to say that. Um [80:34] Yeah, and I mean, listen. Because, look, [80:37] I think people are still figuring out [80:39] how to drive business value out of AI. [80:41] And change management is hard, and the [80:43] bigger the company, the harder it is. [80:46] So, what's happening right now is there [80:48] are a lot of transformation projects at [80:50] large enterprises that are failing. [80:51] There's like a big McKinsey study on [80:53] that. But, if you look at activity from [80:56] the bottom up, I think it's very [80:58] interesting. It's becoming more [80:59] interesting. [81:00] And over the last several months, [81:02] obviously with coding reaching a new [81:04] level, [81:06] we're starting to see very interesting [81:07] things happening there. And obviously [81:09] the revenue that's now being generated [81:10] from these coding models is, you know, [81:12] again, it's exponential like we've never [81:13] seen before. So, the ROI [81:17] is finally there at the model layer. [81:20] Meaning, before, you know, people were [81:22] were saying that it's a bubble because [81:25] you had all this massive CapEx at the [81:26] data center level, and there was no ROI [81:28] coming [81:30] at the model layer. Now we have the ROI [81:32] at the model layer, and I guess we're [81:34] sort of questioning whether [81:37] the ROI will be there at the application [81:38] level. [81:39] But in any event, I think things are [81:40] progressing. I mean, look, I [81:41] fundamentally I'm bullish on this whole [81:42] thing. Okay. No, no, no, let me be [81:44] clear. Hold on, hold on. [81:46] Obviously I'm bullish. I'm in this I'm [81:48] in the space. I'm doing it. You're in [81:49] the arena. [81:51] We have seen in every single wave, [81:54] in the mobile wave, we needed consumers [81:56] to show up at scale. So, we needed [81:57] consumer experiences, and it was very [82:00] obvious that there were these consumer [82:02] businesses that were going to be, if not [82:03] already, incredibly, incredibly [82:05] profitable. Google and Facebook [82:08] were profitable within the first few [82:10] years. They never looked back. [82:12] And all I'm trying to point out to you [82:14] guys is there is not one great example [82:17] yet. If we believe enterprises where all [82:19] the money is, [82:20] and if we believe that's what's going to [82:22] underpin these trillion-dollar [82:23] valuations, just please somebody show me [82:26] a couple of good examples of scaled [82:27] profits. [82:29] Okay, fair enough. Travis, uh final word [82:32] here on But by the way, can I say one [82:34] other thing? I think I think Chamath has [82:35] a good point about being risk off right [82:37] now because if you just look at [82:38] valuation metrics, [82:40] they do seem to be quite high. And I've [82:42] seen other uh versions of those metrics [82:44] as well. So, [82:45] look, it's very very hard to time the [82:47] market. I don't try. [82:49] But I think you could make an argument [82:50] just based purely on [82:52] valuation levels, not events, that you [82:55] want to adopt a more conservative [82:56] posture right now. [82:58] I I think the bet you're making then, [82:59] Travis, is are these valuations [83:03] so [83:05] high right now on the standards that the [83:09] efficiency [83:10] that AI could bring to these companies [83:13] is [83:14] uh not real enough to continue the [83:16] growth from here. It's not a catalytic [83:18] enough technology. And you know, when [83:21] you look at [83:22] >> the opposite. You know, the all all [83:23] these traditional companies have [83:24] horrible valuations. They've been [83:25] crushed. So, my point is if AI is real, [83:28] the upside is also real. All I'm saying [83:31] is [83:31] the details, guys, matter. It's very [83:34] hard to take a very complicated business [83:37] and all of a sudden, quote-unquote, [83:38] transform it. [83:40] It's not as easy as it sounds. That word [83:42] is easy to say. For sure. Travis, go [83:45] ahead. Final word. Final word from [83:46] Travis, please. [83:47] >> First is when it comes to big companies, [83:49] I think the big thing about, let's call [83:50] it the [83:52] the autonomous enterprise, [83:54] is change management [83:56] is the big boy. And in change management [83:59] actually is about all the people that [84:00] already work there, the middle managers, [84:01] the technocrats, the bureaucrats, the [84:04] the whatever, the crats. The crats. [84:07] >> And getting the change management going, [84:10] the change management going there is is [84:11] a human thing, and it's very tricky with [84:15] very complex processes, many of which [84:17] are not even documented. [84:19] And in theory, it's just all going to [84:20] happen real fast, but in practice like [84:22] that's hard. So, that that's part one. [84:24] Part two, what I'm seeing with true tech [84:27] companies, [84:28] real companies, public, I would say [84:31] public like [84:33] hardcore public companies, I mean like, [84:35] you know, founder-led, you know, folks [84:38] that are really cranking [84:40] public companies and tech companies that [84:42] are sort of up-and-comers, [84:45] I talked to CEOs across the board and [84:47] they're like, they're fired up about the [84:50] development, the sort of the [84:53] the productivity and deployment schedule [84:56] and like the new features they're able [84:58] to roll out much, much, much faster [85:01] because they've pivoted their culture [85:03] sort of very pro AI development. [85:06] Um, and I'm getting like almost at this [85:09] point a consistent feedback from real [85:11] founder CEOs that like [85:14] this stuff's real and it's it's it's not [85:17] just hype. Now, there are folks pushing, [85:19] you know, selling their book that are [85:21] like, "Oh, we're at AGI and all this." [85:23] Anybody who's worked with these agents [85:26] and done the AI dev stuff, [85:31] yeah, there's a lot of good stuff, but [85:32] they're not they're not that smart yet. [85:35] They're just not that smart. Anybody [85:37] who's done like I've got a side quest [85:39] where I'm just investing, I've agents [85:41] investing and betting on Cal She and [85:44] Poly and, you know, these other places [85:46] and [85:48] it's silly how dumb the agents are, even [85:50] their best agents, to be honest. [85:53] Yeah. The agents are what you you have [85:56] to be human in the loop with the agent. [85:58] The agent has no taste. The agent can do [86:00] repetitive tasks. The agent is not going [86:02] to do something novel and they quickly [86:04] can get lost in the forest. We had to [86:06] spend a lot of time, we had to spend a [86:07] lot of time with our investing agents [86:11] getting them on board with the idea that [86:14] if you want to make money investing, you [86:16] can't be on both sides of the same bet. [86:20] Yeah. [86:21] You know what I mean? There it is. It's [86:22] just like that. That's where we're at. [86:25] It's the AGI is not here and it's kind [86:27] of silly for folks, I think, to sort of [86:28] suggest that it is. Absolutely. Hey [86:31] guys, we got to wrap. [86:33] Travis, you didn't get to play uh The [86:35] Price Is Wrong. I have two more. You [86:37] guys want to do a bonus round of The [86:38] Price Is Wrong? Yeah, do it. Do it. Do [86:40] it. See if See if Travis gets it. Okay, [86:42] two more. But you guys can steal. You [86:44] guys can steal here. Let me get my [86:45] background music. I got to give me the [86:47] music. Travis, you're going to get to do [86:48] the bonus round here. Let's go. [86:51] >> we got Travis Kalanick. He's here. He is [86:54] a professional water sports player from [86:57] Tallahassee, Florida. How is it down [87:00] there in Tallahassee, Florida? I [87:01] understand you do a little handy work to [87:03] pay the bills, but you spend your most [87:05] of your time out there on the lakes and [87:06] the oceans doing water sports. Yeah, you [87:08] like the water sports. I love the water [87:10] sports. I I [87:12] doing water ski lessons on the weekends. [87:14] Anybody's interested. [87:15] All right, I'll be out there at at [87:17] Tallahassee, Florida. Okay, now, here we [87:19] go. It's your job [87:22] to figure out The Price Is Wrong. The [87:24] Price Is Wrong. [87:26] Here we go. It's so funny. Okay. [87:28] >> Is Wrong here. Here we go. This is your [87:30] bonus round. [87:32] Yeah. [87:33] All right. [87:34] This startup raised 900 million at a 9 [87:38] billion dollar valuation. This is a big [87:40] one. This is a big one before dissolving [87:43] in 2018. [87:45] And their famously deep-voiced founder [87:48] CEO [87:49] is currently serving an 11-year sentence [87:51] in a federal prison. Yes. [87:54] Can you [87:55] name that mispriced startup? Can you [87:56] name the mispriced startup? [87:59] It rhymes with Schmarinos. [88:02] Schmarinos, okay, you're you're closing [88:03] in here. Be careful. You have to [88:06] formulate a question. [88:07] I'll answer. Theranos. Theranos, okay, [88:10] very good. 100 points, you're on the [88:12] board. And here is your next one. This [88:14] is your next one. [88:16] Okay, this is the overtime. All three [88:18] players get to play this one. Whoever [88:21] zooms in first and says it, this is your [88:23] final. [88:24] It's the mispriced startup here. The [88:27] price is wrong on this startup here. [88:29] >> Chamath. You can do this. Come on, [88:30] Chamath. You got this now. Here we go. [88:32] You can do this, Chamath. This [88:33] short-form [88:34] mobile-first streaming platform raised [88:37] $1.7 billion from top investors Come on, [88:40] Chamath. Jeffrey Katzenberg Snap, the [88:42] the the the the uh what is it called? [88:43] >> it out. TOP INVESTORS ACROSS QUIBI. ALL [88:46] of Silicon Valley and shut down in just [88:48] 6 months. Quibi. Quibi. [88:49] >> Quibi. [88:50] Wrong. No. David Sacks, wrong. [88:52] No, no, I know. This was the uh [88:54] >> company. Jeffrey This Katzenberg Quibi. [88:56] QUIBI. QUIBI. QUIBI. [89:00] I WAS SO [89:01] CLOSE. YES. THERE'S YOUR CHAMPION, [89:03] EVERYBODY. Let's tell him what he won. [89:05] He apparently he's won a pied-à-terre [89:09] in downtown uh East Austin. A [89:11] pied-à-terre in East Austin by the [89:13] airport. God knows what shenanigans are [89:15] going down there. I think if you need [89:17] methamphetamine or a date, you're going [89:19] to have an easy time there, Chamath. [89:21] Okay. Oh, hey Chamath, I'm getting [89:22] hammered with people who want to come to [89:24] the sold-out Liquidity. 500 was the cap [89:26] you put on it. You've been [89:28] around as dictator. Can we add 50 seats? [89:31] I think we can add 50 or 100. I don't [89:33] know. Alyssa, can we add 100? Get a [89:34] simulcast room going. [89:36] All right. We'll see you all. Oh, and uh [89:38] the All-In Summit tickets are on sale. [89:40] Go to allin.com. [89:42] Uh and yeah, don't get shut out of the [89:43] summit, folks, cuz this happens every [89:45] time. You you guys email me 2 months [89:47] out, the tickets have been sold out for [89:48] 4 months. So, get your tickets now [89:50] uh at allin.com. Another amazing [89:53] episode. And we'll see you next time. [89:55] Bye-bye. Bye-bye. [89:58] We'll let your winners ride. [90:01] Rain Man David Sacks [90:05] And I said, "We open-sourced it to the [90:07] fans, and they've just gone crazy with [90:09] it." [90:09] >> Love you as I am. Queen of quinoa. I'm [90:12] going all in. [90:18] Besties are back. I'm going all in. [90:21] That is my dog taking a dump in your [90:23] driveway. [90:24] Wait a minute. [90:26] Oh, man. [90:29] We should all just get a room and just [90:30] have a one big huge orgy cuz they're all [90:32] just useless. It's like this like sexual [90:34] tension that they just need to release [90:35] and have. [90:37] Wet your [90:38] Wet [90:39] Wet your feet. [90:41] Wet. [90:42] We need to get Mercy's arm back. I'm [90:44] going all in. [90:51] I'm going [90:53] all in.