War With Iran Drives Oil Market Swings and US Political Fallout

对伊战争引发石油市场震荡与美国政治余波 中东军事冲突升级与能源价格飙升威胁经济稳定，并可能使共和党在即将到来的中期选举中面临国内强烈的选情反弹。 在本期节目中，主持人们分析了与伊朗的军事冲突引发的严峻宏观经济震荡，评估了2026年美国中期选举前的油价波动、全球基础设施风险及政治风险。他们还探讨了AI领军企业OpenAI与Anthropic在算力限制下的空前超高速增长，以及公众和监管机构对AI数据中心和模型安全日益强烈的抵制反弹。讨论最后详细审视了地方层面的财富税提案（如华盛顿州的“百万富翁税”和加利福尼亚州的《亿万富翁税法案》），并阐明了资本外逃如何促使高净值人群迁往佛罗里达州和德克萨斯州等免征个人所得税的州。 Brad Gerstner 介绍了“Invest America”（特朗普账户）计划的迅速推开，在国情咨文后每日有超过 10 万名儿童完成注册，该计划预计将于 7 月 4 日正式上线。 (0:00) 伊朗军事冲突升级与霍尔木兹海峡航运受阻导致国际原油价格剧烈震荡，高盛将全年 PCE 通胀预期上调至 2.9%，市场对通胀溢出效应及标普 500 估值压缩的担忧持续升温。 (3:48) 海湾关键基础设施（如海水淡化厂与储油库）面临严重报复风险，同时中国 20% 的石油消费依赖伊朗和委内瑞拉，促使各方有强烈动力在即将举行的高峰会前寻找冲突缓和通道。 (19:18) Anthropic 与 OpenAI 的年化营收分别飙升至 140 亿美元和 200 亿美元，AI 正加速切入企业人力预算；但 1 吉瓦算力中心高达 500 亿美元的建设成本，意味着实现盈亏平衡需要 5 至 6 年的回本周期。 (27:05)

The All-In Podcast panel argues that escalating military conflict with Iran threatens global macro stability, while the artificial intelligence sector faces severe headwinds as record enterprise revenue growth collides with massive data center capital costs, fractured public messaging, and grassroots infrastructure blockades. Iran Escalation Risks and Geopolitical Off-Ramps (3:48) The military confrontation with Iran has driven wild swings in energy markets, with Brent crude spiking from $84 to $119 before hovering near $100 following attacks on commercial cargo ships in the Strait of Hormuz. Brad Gerstner notes that Goldman Sachs revised its PCE inflation forecast up to 2.9%, raised core PCE to 2.4%, and cut GDP growth estimates by 30 basis points. David Sacks warns of a catastrophic regional "dead man's switch" if military action expands, highlighting vulnerabilities in Gulf desalination plants—which supply drinking water to over 100 million people, including roughly 70% of Riyadh's supply—alongside regional oil processing depots. While prediction markets price a 45% chance of a Democratic midterm sweep amid domestic voter frustration over inflation and foreign conflict, the hosts identify an economic off-ramp centered on China. Unlike the United States, which produces and consumes approximately 20 million barrels per day, China imports 20% of its oil from Iran and Venezuela to fuel essential transport and manufacturing. With Beijing pressured by high youth unemployment and an acute energy squeeze, the panel projects the U.S. will wind down active operations within 30 days, leaving China and regional powers to pressure Tehran into reopening maritime transit. AI Revenue Hyper-Growth Versus Infrastructure Capex Leading AI model developers are generating historic top-line revenue, with Anthropic hitting a $14 billion annualized run rate in February 2026 ($6 billion in a single 28-day window at a $380 billion valuation) and OpenAI reaching a $20 billion run rate. However, Chamath Palihapitiya challenges the durability of enterprise software spending, contending that most commercial traction reflects corporate experimentation rather than profitable production workflows. He points to Amazon Web Services mandating strict human verification for AI-generated software after autonomous agents triggered multiple severity-1 outages. Conversely, David Sacks contends that code-generation tools address a genuine enterprise shortage of software engineers by delivering functional code on a per-token basis. At the same time, computing bottlenecks and infrastructure costs are testing the limits of private capital. Chamath reveals that the projected capital expenditure for a 1-gigawatt data center build in Arizona escalated from an initial estimate of $4–5 billion to over $50 billion. Applying former OpenAI CFO Sarah Friar’s estimate that 1 gigawatt yields roughly $10 billion in annual revenue, facilities require a 5-to-6-year payback timeline before reaching positive net income, highlighting the steep J-curve confronting AI hardware investments. Messaging Fragmentation and Data Center Cancellations The panel diagnoses an escalating public relations crisis across the AI industry, citing conflicting communications from leadership: Dario Amodei warning of an existential "supergod," Sam Altman framing compute as a metered commodity, and Palantir's Alex Karp warning against disrupting the white-collar labor base without national alignment. David Sacks highlights Stanford's AI optimism index, which shows roughly 80% positive sentiment in China and Asia compared to the low 30s in the United States and Europe. Sacks argues that incumbent lab executives deliberately fuel apocalyptic doomerism to secure regulatory capture and exclude open-source competition, citing proposed New York legislation restricting AI medical and legal tools. This negative narrative has translated into local resistance against physical infrastructure. Boosted by Effective Altruism groups like the Future of Life Institute, local protests resulted in a 40% cancellation rate for targeted U.S. data centers. Chamath calculates that 25 canceled facilities in 2025 represented 5 gigawatts of lost capacity ($50 billion in lost annual revenue), with 2026 on track for 40 cancellations totaling 7 gigawatts ($70 billion lost). While cancellations remain heavily concentrated in Virginia and Indiana, Texas has recorded zero project cancellations across 150 gigawatts of requested capacity. Wealth Taxes, Capital Flight, and Deregulation (69:52) Addressing domestic economic policy, the hosts critique state-level millionaire and wealth taxes as self-defeating measures that accelerate capital flight. Following Washington state's enactment of a 9.9% tax on income over $1 million, former Starbucks CEO Howard Schultz relocated to Florida, mirroring Jeff Bezos's earlier departure. Chamath cites a Hoover Institution Monte Carlo simulation analyzing California's proposed wealth tax, which found a 71% probability of a negative Net Present Value and an estimated $25 billion budget deficit, shifting an estimated $2,500 annual burden onto middle-class households. Jason Calacanis argues that policymakers should abandon asset redistribution in favor of breaking structural monopolies. He advocates targeting local housing zoning restrictions, dismantling higher-education accreditation cartels, and using AI to reduce personal healthcare expenditures, pointing to nuclear energy deregulation as a model for spurring economic abundance. The hosts conclude that geopolitical entanglements, regulatory overreach, and uncoordinated technology PR threaten U.S. economic momentum, maintaining that sustainable progress depends on disciplined military off-ramps, rapid local data center approvals, and structural market deregulation.

本期节目聚焦于美国面临的地缘冲突、人工智能商业化瓶颈与各州财政政策危机。主持团队指出，伊朗战事引发的能源震荡与政治反噬正迫使白宫寻求外交撤离通道，而前沿AI实验室在营收爆发式增长的同时，正遭遇企业落地真实性质疑、数据中心建设阻力与严峻的舆论公关危机。 伊朗战事升级、能源冲击与地缘撤离机制 (3:48) 伊朗冲突爆发后，布伦特原油在数个交易日内于每桶84美元至119美元之间剧烈波动，霍尔木兹海峡多艘商船遭袭更推高了航运风险。Brad Gerstner引用高盛最新分析指出，油价飙升已导致PCE通胀预期从2.1%上调至2.9%，核心PCE调至2.4%，GDP增速预测下调30个基点，标普500估值倍数从24倍压缩至21倍。为抑制油价，美国能源部协调国际能源署（IEA）实施了4亿桶全球战略石油储备释放，其中美方出资1.72亿桶。预测市场显示，美军在3月底前进驻伊朗的概率为27%，年内概率达57%；同时，选民对海外干预的厌恶与通胀回升导致共和党在中期选举丢掉国会两院的概率升至45%。 David Sacks与Chamath Palihapitiya强调，美方应在摧毁伊朗常规军力后迅速宣布胜利并退出。Sacks警告，伊朗掌握着针对海湾基础设施的“死手开关”：一旦波斯湾沿岸海水淡化厂遭袭，将危及阿拉伯半岛1亿人口的供水（包括利雅得70%的淡水），且针对炼油设施的打击将使海峡通航变得毫无意义。此外，针对即将举行的中美首脑会谈，嘉宾指出中国20%的石油消费依赖伊朗与委内瑞拉，直接关乎其全部重工业物流与生产原料，而美国本土石油日产量与消费量基本持平（均为2000万桶），因此封锁海峡对中国经济构成严重不对称冲击，中方与海湾国家更有动力施压德黑兰以恢复通航。 AI营收爆发、企业落地争议与基建J曲线 前沿模型实验室正在刷新科技公司的扩张速度。Anthropic的年化运行营收（run rate）在2026年2月达到140亿美元，仅用14个月便实现12倍增长，估值升至3800亿美元，单月收入达60亿美元；OpenAI年化营收亦达到200亿美元，估值达8400亿美元。Nvidia首席执行官黄仁勋近期向两家机构注资400亿美元，并暗示这将是其公开上市前的最后私募轮次。然而，Chamath对该营收的实际质量提出质疑，认为当前财富500强企业的采购多属于防守型或实验性预算，缺乏能带来持续利润改善的核心生产级用例。他列举AWS因AI编写未经审核的代码导致多次一级严重故障（sev-1），迫使亚马逊下达禁止未经人工审核直接部署AI代码的禁令，说明医疗诊断与高频金融等高责任场景短期内难以全面接受模型幻觉。 David Sacks认为，软件代码生成是目前唯一得到验证的爆发级企业用例，正在通过按Token计费的方式直接补充长期短缺的工程人力产能。针对资本密集度，Chamath以其在亚利桑那州筹建的1吉瓦（GW）数据中心为例，说明项目总资本支出已从最初预估的40亿至50亿美元膨胀至超过500亿美元。引用OpenAI首席财务官Sarah Friar“每吉瓦算力对应约100亿美元年营收”的测算模型，500亿美元的重资产投入意味着AI数据中心面临长达5至6年的投资回收期，行业在技术上高度依赖定制芯片、开源算法优化来熨平这一严峻的盈利J曲线。 舆论叙事分裂、数据中心邻避抗议与开源竞争 AI行业面临急剧恶化的公共形象危机。民调显示公众对AI的正面支持率跌至极低区间，甚至低于移民及海关执法局（ICE）。嘉宾分析其原因在于领军人物公关信息的极度混乱：Dario Amodei宣扬“全能超级神”的技术末日论，Alex Karp强调AI打破白领政治经济特权并呼吁服务国家力量，而Sam Altman则将其包装为如水电般按量计费的基础设施。Sacks指出，头部闭源公司高管渲染生存危机实质是为了推动牌照制度，构筑监管壁垒以阻扼开源竞争；斯坦福AI乐观指数显示，中国及亚洲国家公众对AI持正面预期的比例高达80%，而美欧仅停留在30%左右，恐慌情绪更引发纽约州提案禁止AI提供基础法律与医疗建议等过度监管。 由有效利他主义（EA）基金主导的游说体系已阻碍物理层基建。参议员Bernie Sanders公开呼吁暂停全美数据中心建设，全美被抗议的数据中心项目取消率已达40%。Chamath统计，2025年全美共有25个数据中心项目被叫停（损失5吉瓦容量），直接造成约500亿美元年化潜在收入流失；2026年前两个月已有约100个项目遭遇抗议，预计将取消40个（损失7吉瓦与700亿美元年营收），两年累计对美国经济造成约1200亿美元的年化损失，这些阻力主要集中在弗吉尼亚州和印第安纳州，而得克萨斯州在150吉瓦申请量下维持零抗议取消。Brad Gerstner则强调，即使初创企业日常85%的Token负荷转移至本地开源模型，前沿模型单月营收依然激增，证明AI的整体潜在市场空间（TAM）远超市场保守预期，闭源与开源将长期共存。 富人税引发资本外流、财政缺口与结构性解法 (67:09) 地方财税政策正引发高净值人群加速迁徙。华盛顿州立法机构通过了将于2029年实施的“富豪税”，对超过100万美元的年收入征收9.9%的附加税，预计波及3万户家庭以筹集40亿美元税款。法案通过当日，星巴克前首席执行官Howard Schultz宣布结束其在西雅图长达44年的居住，搬迁至佛罗里达州，此前Jeff Bezos也因华盛顿州开征7%资本利得税于2023年移居迈阿密。Chamath援引胡佛研究所针对加州拟议《亿万富翁税法案》的10万次蒙特卡洛模拟指出，该税收在71%的情景下产生负净现值（NPV），将引发富人出逃并导致约250亿美元的预算净缺口，最终迫使1000万中产家庭每户承担约2500美元的补足税款。 针对Bernie Sanders与Ro Khanna在联邦层面提出的5%年度财富税提案，Sacks与Gerstner批评其实质是在20年周期内对私人财产实施变相征收，即使加州教师工会与州长Gavin Newsom也对地方版方案持保留意见，但Newsom的表态亦被视为在为联邦统征财富税铺路。Jason Calacanis主张，缓解生活成本危机应通过拆除教育资质垄断、放宽住宅分区建设限制以及利用AI实现自主医疗管理等供给端改革，效仿轻监管州模式以降低生活刚性支出，而非诉诸激进的财富剥夺手段；同时，Brad Gerstner提及全美推行的“投资美国”（Invest America）儿童账户计划，正在以日增10万账户的速度为3000万儿童提供本金支持，意在让普通家庭直接分享资本市场增长红利。 本期讨论表明，海外地缘冲突的外溢效应正严重制约美国国内的宏观经济与选举预期，而前沿AI与基础设施的扩张速度正在遭遇地方政治抗议、重资产回报周期与社会叙事分裂的现实约束。如何在地缘对抗中锁定战略撤离点，并在鼓励核心技术研发的同时避免惩罚性财税政策造成资本出逃，是决定未来美国经济竞争力的核心考验。

## Background 本期节目由知名商业科技播客 All-In Podcast 于 2026 年 3 月 13 日在 YouTube 平台发布。节目由 Jason Calacanis、Chamath Palihapitiya、David Sacks 以及替班嘉宾 Brad Gerstner 共同主讲，视频简介聚焦于伊朗战争的经济冲击与撤离策略、Anthropic 和 OpenAI 创纪录的营收狂飙与公关灾难，以及华盛顿州通过“富豪税”导致星巴克创始人出走迈阿密等热点议题。 ## TL;DR 四位主持人深入剖析了中东局势升级对国际原油市场与美国大选的连锁冲击，呼吁政府克制并寻找外交脱身通道，同时指出大国博弈的核心仍在于中美关系。在科技前沿板块，节目激烈辩论了 AI 龙头企业营收暴涨背后究竟是实验性开支还是真实生产力落地，并直面末日论炒作导致的公众信任危机与算力基建受阻困境。最后，嘉宾们结合胡佛研究所模型痛陈西海岸各州激进富豪税对税基和中产阶级的反噬，呼吁通过打破监管垄断与科技赋能来解决民生核心痛点。 ## Atmosphere & style 整场节目节奏紧凑、交锋激烈，既有硅谷顶级投资人之间刀刀见血的商业冷思考，又充斥着老友互怼、插科打诨的轻松笑料。主持人们在严肃研判地缘危机与宏观经济风险时神情凝重、直言不讳，而在拆解科技巨头公关翻车与富豪避税搬家时又极尽讽刺与幽默。听众仿佛置身于一场私密而高信息密度的精英智囊闭门夜话，既能感受到前沿产业剧变带来的紧迫感，又能在嬉笑怒骂中看清复杂博弈的底层逻辑。 ## Bulletin - [03:48](https://www.youtube.com/watch?v=HiVej9_fvl8&t=228s) 中东战事推高布伦特原油至接近每桶120美元后回落，高盛将美国全年PCE通胀预期上调至2.9%，各方主张白宫应迅速寻求“宣布胜利并抽身”的退出通道以防战事演变为持久泥潭。 - [27:05](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1625s) 大模型营收迎来爆炸式增长，Anthropic单月营收达60亿美元且ARR跃升至140亿美元，OpenAI达200亿美元；但其背后的万亿级资本开支与算力基建面临5至6年的漫长回报周期考验。 - [01:02:26](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3746s) 受有效利他主义（EA）智库渲染与过度末日公关反噬，全美遭抗议的数据中心项目有40%被迫取消，仅2025至2026年累计就将从美国经济盘中抹除约1200亿美元的潜在年化营收。 - [01:07:48](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4068s) 华盛顿州强行通过针对年收入超百万美元征收9.9%的“富豪税”，星巴克创始人霍华德·舒尔茨随即斥资4400万美元迁居迈阿密，加州与联邦层面的财富税构想亦面临资本外流与宪政争议。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=HiVej9_fvl8&t=0s) Invest America（“特朗普账户”）申领迅速推进，日新增注册超10万名儿童，旨在通过直接持有优质美股股权化解长期贫富分化与社会福利难题。 - [08:11](https://www.youtube.com/watch?v=HiVej9_fvl8&t=491s) 原油市场对特朗普“战争很快结束”的表态在数纳秒内反应暴跌，专业资本坚信中东战局缺乏持久化的真实逻辑，短期恐慌已随多国释放4亿桶战略储备而降温。 - [10:19](https://www.youtube.com/watch?v=HiVej9_fvl8&t=619s) 伊朗战争若失控升级将引发不可承受的次生灾害，包括对海湾脆弱海水淡化厂的打击危及1亿人生存，以及以色列因本土防空过载可能动用极端武器。 - [15:16](https://www.youtube.com/watch?v=HiVej9_fvl8&t=916s) 若无法在数月内脱离对伊冲突，通胀抬头与保守派选民的基本盘撕裂恐导致共和党在2026年中期选举惨败，终结特朗普第二任期的政治遗产。 - [19:09](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1149s) 对伊博弈的隐形杠杆在于中国对波斯湾重油的高度依赖，月底的中美首脑峰会促使北京有极强动力推动达成地缘大交易以稳固自身能源命脉。 - [28:59](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1739s) Anthropic 与 OpenAI 的营收扩张创下人类软件史新纪录，其增长动力来自于顶层智能体由争夺传统企业IT预算升级为直接替代并侵吞人力劳动力预算。 - [32:12](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1932s) 企业级AI落地依然面临“实验性测试流水”与“真实生产级ARR”的断层，幻觉失误及缺乏SLA保障限制了随机模型在核心生产工作流中的全面部署。 - [38:30](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2310s) 代码辅助与生成是目前企业级AI唯一经过充分验证的超级杀手级应用，有效打破了全球长期受限的软件工程人力供给瓶颈。 - [44:03](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2643s) AI基础设施呈现出恐怖的资本支出黑洞，亚利桑那州单个1吉瓦数据中心造价超500亿美元，对应每年100亿美元营收，企业需耗费5至6年才能实现收支平衡。 - [46:10](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2770s) 科技巨头在公关沟通上陷入严重自我矛盾，前期利用末日论与自主意识造神以高估值吸金，后期面临商业兑现压力又对军工及各行业无底线抛售并四处道歉。 - [56:15](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3375s) 民意调查显示美国大众对AI的抵触与失业焦虑已使其公众好感度跌至底谷，这与中国及亚洲地区高达80%的技术乐观度形成极其鲜明的跨文化反差。 - [58:35](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3515s) 纽约州拟取缔AI提供医疗与法律咨询被批为最荒谬的行业工会保护主义立法，实质性剥夺了底层弱势群体获取廉价专业援助的平权权利。 - [01:02:26](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3746s) 弗吉尼亚和印第安纳等州因极端环保和反AI叙事大量扼杀数据中心审批，而德克萨斯州凭借超150吉瓦的友好电网与政策接纳成为算力基建最大赢家。 - [01:06:04](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3964s) 企业实际架构正快速演进为“前沿闭源实验室做高层推理规划+开源端侧模型负责高并发执行”的混合模型分工，AI总体潜在市场（TAM）远超行业预期。 - [01:10:30](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4230s) 蒙特卡洛模拟表明加州富豪税有71%概率陷入250亿美元财政赤字陷阱，高流动性富豪离境出逃最终只会将财政窟窿转嫁给普通中产家庭。 - [01:17:23](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4643s) 遏制阶级对立与极左财富没收浪潮的治本之道在于以第一性原理去监管，运用AI与民营资本大幅压低教育、住房与医疗的核心民生供给成本。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=HiVej9_fvl8&t=0s) Brad Gerstner 替班登场与国情咨文幕后轶事 节目开场，Jason Calacanis 宣布 David Friedberg 因忙于“拯救世界”、培育新型土豆或藜麦、抱子甘蓝而缺席，由其钦定的替班嘉宾 Brad Gerstner 坐镇。Jason 调侃道，自 Brad 在美国总统国情咨文（State of the Union）演讲中被点名表彰后，大家就一直没在节目中碰过面，随即追问 Brad 那场全美瞩目的点名是否属于提前彩排。Brad 透露，自己当时完全不知道会被点名，事后才得知讲稿原文中根本没有这一段，纯粹是总统临时即兴加进去的。受邀出席国情咨文对他而言意义非凡，因为这项宪政传统已延续了 250 年；当晚他与家人坐在狭小的国会众议院大厅内，右手边不远处就是第一家庭、Jared Kushner 与 Ivanka Trump。 Jason 随即打趣 Brad 当时在镜头前向全美观众“比心致意”的动作：“你把心掏出来托得恰到好处，要是手臂再抬高 5 度向外伸直，那就成行纳粹礼了，那些‘特朗普账户’可就要变成超级种族主义账户了！”全场哄堂大笑。Brad 接着分享了“特朗普账户”（Trump Accounts / Invest America）的惊人进展：目前每天有超过 10 万名儿童注册申领，已有数百万儿童激活了账户；全美近 3000 万符合条件的儿童只要申领就能获得至少 250 美元的初始投资。该计划将于 7 月 4 日正式上线，总统在国情咨文中的大力背书加速了全民参与，其核心愿景就是让普通工薪阶层的孩子也能直接持有美国优质企业的股权，共享资本主义的红利。 - **特朗普账户（Invest America）申领进展**：每日新增注册超 10 万名儿童，已有数百万儿童认领账户；全美近 3000 万儿童符合申领至少 250 美元起步资金的资格。 - **项目上线时间**：预计于 7 月 4 日美国独立日正式全面上线。 --- ### [02:50](https://www.youtube.com/watch?v=HiVej9_fvl8&t=170s) 股权版“捐赠誓言”与全民基本收入的讨论 针对日益加剧的贫富差距和关于全民基本收入（UBI）的社会讨论，Jason 提出了一个构想：既然有人把 Invest America 当作 UBI 的雏形，那能否发起针对企业股权的“捐赠誓言”（Giving Pledge）？比如让 Alphabet 的 Larry Page、Sergey Brin 或 Meta 的 Mark Zuckerberg 等科技巨头自愿承诺，在未来 20 年内将其 5% 的股份注入全美儿童的账户中。哪怕每个孩子最终只分得某家公司十分之一、百分之一甚至千分之一股，长期积累下来也会是一笔改变命运的财富。Brad 证实业内确实已经讨论过类似方案，并神秘地透露“敬请期待，在临近 7 月 4 日时会有爆炸性重磅消息公布”。 --- ### [03:48](https://www.youtube.com/watch?v=HiVej9_fvl8&t=228s) 伊朗战事爆发与原油市场的剧烈震荡 话题随后转向中东局势。Jason 直言，尽管战争关乎生死与伊朗人民的自由，但本期播客更聚焦于其引发的经济冲击与二阶效应。在过去 5 个交易日中，原油市场经历了极端波动：布伦特原油在战争第 7 天（周五）飙升至 84 美元，第 10 天（周一）暴涨至 119 美元，随后回落至 84 美元；紧接着，由于 3 艘悬挂泰国、日本和马绍尔群岛国旗的商船在霍尔木兹海峡遇袭，油价再度被推高至 100 美元，录制播客时油价在 99 美元上下徘徊。Jason 回忆起 1978 年的原油危机，当时加油站甚至需要按车牌尾号单双号排队一两个小时；2008 年由于中国需求爆发与“石油峰值”担忧，原油按现值计算曾冲上 216 美元；2022 年俄乌冲突爆发时油价曾达 115 美元（相当于现值 133 美元）。 与此同时，地缘风险急剧升级：伊朗新任最高领袖穆杰塔巴（Mojtaba Khamenei）公开宣布将继续封锁霍尔木兹海峡，将其作为施压敌手的杠杆；《华尔街日报》援引中东研究所资深学者分析称，重新开放海峡可能需要动用地面部队；预测市场 Polymarket 上的交易员给出 27% 的概率认为美军将在 3 月底前进驻伊朗，而年底前进军的概率则高达 57%。 ```mermaid flowchart LR accTitle: Brent Crude Volatility and Strait of Hormuz Crisis accDescr: 过去交易日中布伦特原油价格随战事与商船遇袭事件大幅波动的轨迹 subgraph PriceMoves ["布伦特原油价格剧烈异动"] D7["第 7 天：飙升至 $84"] --> D10["第 10 天：暴涨至 $119"] D10 --> Drop["迅速回落至 $84"] Drop --> SpikeBack["3 艘商船遇袭：反弹回 $100"] end subgraph Escalation ["局势升级信号"] MLeader["最高领袖穆杰塔巴：维持霍尔木兹海峡封锁"] PolyOdds["Polymarket 预测：美军年底前进入伊朗概率 57%"] end PriceMoves -. 催生通胀隐忧.-> Escalation ``` - **布伦特原油震荡区间**：5 个交易日内于 84 美元至 119 美元之间大幅宽幅震荡，录制时约为 99 美元。 - **遭袭货轮背景**：在霍尔木兹海峡遇袭的 3 艘船只并非油轮，而是悬挂泰国、日本、马绍尔群岛国旗的普通货船。 - **Polymarket 地面部队概率**：押注美军在 3 月底前进入伊朗的概率为 27%，在 2026 年底前进入的概率高达 57%。 --- ### [06:13](https://www.youtube.com/watch?v=HiVej9_fvl8&t=373s) 滞胀阴影下的高盛预测与“特朗普主义”定力 面对高企的油价与不确定性，Brad 深入剖析了宏观经济的连带反应。高盛当日发布的最新评估报告大幅上调了通胀预测，将全年 PCE 通胀预期从 2.1% 提高至 2.9%；即便剔除能源价格，核心 PCE 也从 2.2% 上调到了 2.4%，这意味着高油价的连锁传导效应正在渗透整个经济体。高盛同时将美国全年 GDP 增长预测下调了 30 个基点，并预期失业率将有所攀升。标普 500 指数的估值倍数也从数月前的 24 倍回落至 21 倍，反映出市场浓厚的避险情绪。 然而，Brad 认为华尔街可能出现了误判。市场当前陷入了某种类似于阿富汗和伊拉克战争的创伤后应激障碍（PTSD），担忧美国再次卷入持久战的泥潭。但 Brad 指出，“特朗普主义”（Trump Doctrine）在本质上是极为务实的，与新保守主义（Neocon）有着根本区别：特朗普的目标非常有限且明确，即精准摧毁并削弱对美国国家安全的实质威胁，绝无兴趣在全球推行所谓的麦迪逊式民主或充当世界警察去搞政权更迭（Regime Change）。虽然特朗普乐见伊朗民众自发推翻政权，但他绝不会让美军重蹈切尼时代劳民伤财的覆辙，因此这场危机的持续时间大概率会比市场恐慌预期的要短得多。 - **高盛宏观数据修正**： - 全年 PCE 通胀预期：从 2.1% 上调至 2.9%。 - 核心 PCE 通胀预期：从 2.2% 上调至 2.4%。 - 全年 GDP 增速预期：下调 30 个基点（-0.3%）。 - 股市估值：标普 500 远期市盈率由高点的 24 倍压缩至 21 倍。 --- ### [08:11](https://www.youtube.com/watch?v=HiVej9_fvl8&t=491s) 市场的反射性验证：一场不会失控的短期冲突 Chamath Palihapitiya 接过话茬，从市场微观结构的反射性印证了 Brad 的判断。他指出，本周最值得关注的关键瞬间，是当总统特朗普被问及战事走向并回应称“战争很快就会结束”时，金融市场在“几纳秒之内”瞬间将原油价格从每桶 120 美元直接砸到 90 美元。 Chamath 强调，这种强烈的瞬时反射性动作说明了一切：如果市场真的坚信这是一场旷日持久的全面战争，交易员们根本不会理睬总统的言论，甚至会反向做多将油价推得更高。油价的暴跌证明了最聪明的专业资金（Sharps）心里非常清楚，伊朗冲突根本不存在长期延宕的路径。伊朗军方当下激进的“虚张声势”与挺胸示威，主要是为了在国内保全面子，并为下一任执政者争取尽量有利的统治资本，冲突不会演变成无休止的持久消耗战。 ### [09:21](https://www.youtube.com/watch?v=HiVej9_fvl8&t=561s) 战略石油储备释放与市场降温预期 Chamath 认为此前的市场异动更像是一个“投掷探路气球”（trial balloon），印证了大众最初的判断——这场冲突在短期内就会告一段落。他赞同 Brad 的观点，即短期内物价与油价确实可能出现剧烈脉冲，但政策层面的应对动作非常迅速。在 3 月 11 日，能源部长 Chris Wright 在总统指示下迅速启动了国际能源署（IEA）成员国的协调机制：美国方面释放了 1.72 亿桶战略原油，全球各成员国更是联合释放了约 4 亿桶石油。 这笔庞大的抛储规模能够有效平抑短期油价暴涨的冲击。更关键的是，据估算全球战略储备中还有大约 10 亿桶石油可供调配。Chamath 指出，结合这多重因素来看，最糟糕的恐慌阶段大概率已经过去，当前的核心议题已经转向如何寻找停火与降级的“退出通道”（off-ramp）。 ### [10:19](https://www.youtube.com/watch?v=HiVej9_fvl8&t=619s) 萨克斯拆解升级风险：霍尔木兹海峡、海水淡化厂与核阴影 David Sacks 完全赞同寻找退出通道的主张。他强调美军已经重创了伊朗的军事实力，其海陆空三军遭到毁灭性打击，此时此刻正是“宣布胜利并抽身退出”（declare victory and get out）的最佳时机，这也正是金融市场极度渴望看到的局面。然而，以共和党内新保守派（Neocons）为主的一股激进势力却在极力鼓吹战争升级，呼吁派遣地面部队、推进政权更迭，甚至在《华尔街日报》刊登社论主张不要寻找下台阶，而是无休止地轰炸下去。Sacks 随即深入剖析了战争一旦升级可能引发的三层灾难性连锁反应： ```mermaid flowchart TD accTitle: 战争升级风险传导链 accDescr: 展示伊朗战争升级对海湾国家基础设施、民生及以色列局势的连锁反应 subgraph GulfCrisis["第一层与第二层：海湾国家遭受打击"] A["对伊朗油气设施展开打击"] --> B["伊朗对海湾国家油气设施发起报复"] B --> C["如阿曼巨型油库遭袭，产油能力瘫痪"] B --> D["针对海水淡化厂的软目标打击"] D --> E["阿拉伯半岛1亿人面临断水生存危机"] end subgraph IsraelRisk["第三层：以色列防空承压与极端反制"] F["伊朗饱和式远程导弹反击"] --> G["以色列防空拦截弹药耗尽"] G --> H["以色列本土遭受毁灭性打击"] H --> I["以色列或动用核武器进行极端报复"] end GulfCrisis --> J["全球经济重创与不可逆的人道灾难"] IsraelRisk --> J ``` - **霍尔木兹海峡与海湾油气设施报复**：目前霍尔木兹海峡已被封锁，而更可怕的后果在于，一旦伊朗自身的油气基础设施遇袭，德黑兰必将对海湾盟国展开“对等报复”（tit-for-tat）。此前阿曼的巨型油库就已遭无人机或导弹摧毁。一旦海湾多国的油气生产设备全面受损，即便未来海峡重新开放，中东也无法恢复原油生产与出口。 - **海水淡化厂与生存绝境**：阿拉伯半岛本质上是一片沙漠，高度依赖海水淡化工程，利雅得约 70% 的淡水依靠淡化技术，整个阿拉伯半岛有近 1 亿人口以此为生。海水淡化设施属于极脆弱的民用软目标；此前已有伊朗淡化设施遭袭、随后科威特设施疑似遭到报复的报道。若关键水厂被毁，整片地区在人道主义层面将直接变为不毛之地，“没有水人类根本生存不了多久”。伊朗实际上对海湾国家的命脉握有一枚“自杀式触发开关”（dead man's switch）。 - **以色列的安全极限与核风险**：尽管以色列在社交媒体上实施了信息管制，但流出的零星消息表明，以色列本土在短短两周内遭受了建国以来最沉重的打击。海湾国家主要面临无人机与短程导弹威胁，而以色列则持续受到远程弹道导弹袭击。一旦铁穹等防空系统弹药告罄，以色列重要城镇与设施将面临灭顶之灾，届时极可能迫使以色列动用核武器进行报复，将局势彻底推入深渊。因此，唯有通过外交谈判达成停火与和解，才是规避全面浩劫的唯一出路。 ### [15:16](https://www.youtube.com/watch?v=HiVej9_fvl8&t=916s) 政治反噬：中期选举民调滑铁卢与 MAGA 阵营分裂 Jason Calacanis（JCal）对于局势的前景表达了强烈担忧。他认为在没有摩萨德、中情局情报支撑的情况下，虽然特朗普通常只在胜率极高且具备撤离方案时才会动手，但目前的战况完全看不到体面的退出路径。Polymarket 上的职业投注者给出美军派地面部队入场（boots on the ground）的概率甚至高达 57%。如果真演变成地面战，特朗普的第二任期就等于提前终结，因为这彻底背叛了他当初竞选的核心承诺——选民选他是为了阻止第三次世界大战爆发，而不是为了主动引爆对伊战争。 这种外交冒进正在引发全方位的政治动荡与反弹： - **保守派意见领袖倒戈**：Tucker Carlson、Marjorie Taylor Greene（MTG）、Joe Rogan、Matt Walsh 以及 Megyn Kelly 等 MAGA 阵线旗手群情激愤，痛斥此举背叛了“美国优先”的基本盘，甚至称其为“MAGA 运动的终结”。 - **民调断崖式下跌与国会失守风险**：Polymarket 数据显示，民主党在 2026 年中期选举中实现参众两院“全面横扫”（clean sweep）的概率飙升到了 45%。如果战争再拖延六个月，共和党将失去两院控制权，并在 2028 年大选彻底惨败，特朗普 2.0 的所有施政遗产都会化为泡影。 - **民生痛点叠加**：普通美国工薪阶层并不持有股票，对中东战事毫不关心；相反，他们直面的是通胀重回 3% 以上的威胁、略有抬头（微升 10 个基点）的失业率，以及对特朗普家族商业交易的质疑。 当其他主持人和嘉宾打趣 JCal 是把对爱泼斯坦案卷宗未完全公开、ICE（移民与海关执法局）执法严苛（此前免除 Kristi Noem 职务）等所有旧账一股脑“塞进下水道水槽”（kitchen sinking）时，JCal 极力澄清这绝非个人情绪发泄，而是条理清晰的形势研判：如果白宫不能在短时间内找到下台阶停火，中期选举的惨败将不可避免。Brad Gerstner 随后将讨论拉回主线，指出不论分析路径如何不同，所有人的共识都在于：必须尽快找到从这场战争脱身的方案。 ### [19:09](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1149s) 寻找伊朗战争撤离通道：霍尔木兹海峡与大国博弈 节目的讨论从伊朗战争的撤离路线（Off-Ramp）切入。布拉德·格斯特纳（Brad Gerstner）与查马斯·帕利哈皮蒂亚（Chamath Palihapitiya）均认为，特朗普的路线绝非传统的新保守主义（Neocon）路线；外界虽然目光全盯在伊朗身上，但底层的真实棋局是“关于中国、中国、还是中国”。查马斯强调，月底特朗普即将与中国最高领导人展开为期三天的关键会晤，这是当今主导世界的两大超级大国——守成大国美国与寻求复兴的中国——之间极具历史意义的峰会。查马斯敢拿“一美元赌一个甜甜圈（dollars to donuts）”打赌，中方有着极其强烈的动力在这次为期三天的峰会中达成一项历史性的“大交易”（Grand Bargain），而特朗普总统一定会充分利用中东局势作为地缘博弈的谈判筹码。 面对杰森·卡拉卡尼斯（Jason Calacanis）关于“如果以色列持续打击伊朗，被封锁的霍尔木兹海峡究竟如何重新开放”的追问，布拉德给出了清晰的脱身构想：美国应当直接宣布胜利并果断退出，正如萨克斯（David Sacks）所建议的那样，对外宣称“美军的任务只是削弱并摧毁其军事威胁，绝非搞什么输出民主的社会实验；伊朗人民该怎么走由他们自己决定”。布拉德指出，美国目前每天生产2000万桶石油、消费2000万桶石油，国内供需基本打平，海峡危机对美国只是个“温和的小麻烦”，但对严重依赖中东能源的中国、亚洲其他国家以及天天疲于躲避伊朗导弹的海湾盟友来说，则是灭顶之灾。一旦美军抽身，利益休戚相关的海湾国家、印度和中国自然会形成巨大合力，对伊朗施加前所未有的压力以确保航道安全。布拉德为军事行动进行了“坚壁清野式（steel-man）”的辩护：虽然军事干预伴随巨大风险，但如果袖手旁观任由誓言消灭美国的伊朗获得核导原料、或者任由南美的门罗主义防线崩溃，其长期风险同样不可承受；只要坚持“进场重创其能力，然后迅速撤离（degrade and get out）”，就符合美国的根本安全利益。 ```mermaid flowchart TD accTitle: 伊朗战争对各方的能源冲击与地缘博弈 accDescr: 展示霍尔木兹海峡危机对美、中及海湾国家的不同影响机制 subgraph US["美国本土态势"] A["能源自给: 日产销约2000万桶"] --> B["定点清除威胁后迅速撤离"] end subgraph ME["中东与海湾局势"] C["霍尔木兹海峡航运受阻"] --> D["海湾邻国与盟友承受防御压力"] end subgraph CN["中国面临的硬约束"] E["20%石油供给受阻 (伊朗/委内瑞拉)"] --> F["战略储备仅能支撑5-6个月"] G["青年男性失业率约25%"] --> F F --> H["亟需在首脑峰会达成大交易"] end C --> CN B -.->|"转移区域护航与施压责任"| ME B -.->|"形成战略谈判筹码"| H ``` - **美国原油供需现状**：日产量约2000万桶，日消费量约2000万桶，处于近乎自给自足的战略平衡状态。 - **美军行动策略**：反对陷入长期驻军与“民主改造”，主张在重创敌方进攻能力后快速宣布达成目标并撤出。 --- ### [22:58](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1378s) 真正的杠杆在中国：“大交易”预期与新保守主义之争 查马斯进一步拆解指出，委内瑞拉与伊朗这两场风波的核心指针最终都指向北京。开战以来，最值得投资人和战略家玩味的一个核心事实是：中国既没有替伊朗出头动用武力，也没有在国际上为伊朗公开背书，甚至根本没有取消月底与特朗普的峰会。查马斯解释，这是因为中国极度依赖这些被封锁的外部能源——其国内原油消费总量的20%来自委内瑞拉和伊朗，而这20%更是其运输命脉（汽车、公交、飞机）和化工基础原料不可或缺的重油来源。尽管中国建立了规模可观的战略石油储备（SPR），但绝对经不起这场危机持续五六个月的长期消耗。若局势僵持，叠加中国当下高达25%的年轻男性失业率，其经济将面临难以承受的剧痛。因此，峰会不仅不会吹，反而成了中国急需寻求化解困局的救生索。 在几位主持人互开玩笑、吐槽杰森又把“整个厨房水槽（kitchen sink，意指把所有杂七杂八的论据一股脑堆砌上来）”端出来之后，大卫·萨克斯给出了总结性研判。萨克斯赞同长期战争在政治上极其不得人心，若演变成持久战，势必在2026年乃至2028年的选举中反噬共和党；但所幸特朗普拥有极度敏锐的政治嗅觉，向来青睐短促、果断、雷霆般的特种突击与定点打击（如当年的“午夜重锤”行动或针对委内瑞拉马杜罗政权的突袭），而非深陷泥潭。萨克斯痛批以《华尔街日报》评论版为代表的新保守主义（Neocon）建制派是“在所有事情上从来没有对过一次的一群人”，这些人若得逞，在伊拉克和阿富汗驻军20年都不会撤退。因此，白宫当前最关键的抉择就是无视这批企图扩大战争目标的鹰派聒噪，遵从政治本能尽快体面收场。 - **中国能源承压数据**：国内约20%的石油消费依赖委内瑞拉与伊朗供应；战略石油储备难以维系5至6个月以上的断供冲击。 - **社会指标警讯**：查马斯指出当前中国年轻男性失业率已达25%左右，能源通胀对就业底盘冲击巨大。 - **政治与战术定调**：特朗普军事偏好倾向于“短平快”的精准震慑，而非新保守主义派系主张的大规模驻留与政权更迭。 --- ### [27:05](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1625s) AI史诗级创收狂飙：OpenAI与Anthropic的千亿军备竞赛 地缘政治探讨告一段落后，四人将目光转回播客的核心强项——人工智能与科技前沿。杰森直言，OpenAI 与 Anthropic 在营收规模和成本扩张的狂飙速度，刷新了人类商业史的极限记录。根据披露的最新行业数据，达里奥·阿莫代伊（Dario Amodei）执掌的 Anthropic 在2026年2月的年化营收运行率（ARR）一举跃升至140亿美元，仅用14个月就实现了从10亿美元到140亿美元的惊人飞跃，同比实现惊人的12倍增长（12x YoY），上个月估值攀升至3800亿美元。与此同时，OpenAI 在2025年底的年化营收运行率也达到了200亿美元，用24个月时间完成了从20亿美元到200亿美元的跃升，上月估值更是飙至8400亿美元。 ``` 【营收飞跃对比】 Anthropic: $1B ARR ──(14个月暴增14倍)──> $14B ARR (最新估值: $3,800亿) OpenAI: $2B ARR ──(24个月暴增10倍)──> $20B ARR (最新估值: $8,400亿) ``` 谈到这里，杰森在屏幕上调出了一张自己制作的恶搞梗图——截取自经典电影《侏罗纪公园》的后视镜视角，镜中赫然是一只狂奔追赶的霸王龙（象征达里奥·阿莫代伊），配字调侃山姆·奥特曼（Sam Altman）一回头发现达里奥已经咬紧牙关逼到了跟前，随时可能在赛道上实现反超。几位好友顿时爆发出笑声，惊叹 Anthropic 在企业级 B2B 市场的推进势头远超所有人预料。杰森随后把问题抛给了作为两家顶级 AI 独角兽共同投资人的布拉德。杰森打趣询问布拉德当年在《BG2》播客上被山姆·奥特曼放话“要原价买回股份”后究竟有没有抛售；布拉德当场爽朗回应：“从那之后，我不仅没卖，反而大幅加仓买入了更多。”面对眼下两家公司为追求 AGI 在实现盈利前可能需要吞噬2500亿乃至5000亿美元的“疯狂 J 曲线（J-curve）”，两家巨头究竟谁是更具性价比的投资标的，成为了全场最为关切的核心谜题。 - **Anthropic 业绩指标**：截至2026年2月年化运行率达140亿美元；14个月内从10亿美元飙升至140亿美元；年复合增长达12倍；最新估值3800亿美元。 - **OpenAI 业绩指标**：2025年底年化运行率达200亿美元；24个月内从20亿美元扩充至200亿美元；最新估值8400亿美元。 - **资本动态**：投资人布拉德·格斯特纳确认其不仅长期坚定持有 OpenAI，近期更逆势大幅追加了两大巨头的投资筹码。 ### [28:59](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1739s) Anthropic 与 OpenAI 的惊人收入与上市时机 面对关于 Anthropic 与 OpenAI 是否应当上市以及上市概率的提问，Brad Gerstner 幽默地打趣称“作为投资者必须像对待自己的孩子一样一视同仁”，因为两家都是极其出色的企业。他指出，就在两三个月前，市场与科技界同行对 AI 基础设施的巨额资本开支仍普遍持强烈怀疑态度，认为这些投资看不到增量收入回报。然而，在 2026 年 1 月和 2 月，行业迎来了堪比“原子裂变”的历史性时刻：Anthropic 在短短 28 天的 2 月份斩获了高达 60 亿美元的单月营收。这一单月业绩甚至超过了 Snowflake 和 Databricks 两大软件巨头历经 12 年拼搏所积累的年收入总和；两家 AI 龙头仅凭今年前四到五个月的业绩，就能比肩 SpaceX 全年的总营收。 当被问及是什么在支撑如此庞大的营收爆发时，Brad Gerstner 解释称，行业在 Claude Opus 4.6 和 ChatGPT 5.4 面世后跨越了一个关键临界点。无论是 Claude Code、Codex 还是 ChatGPT，顶层智能体（agents）的运算能力已不再局限于争夺传统的“企业 IT 预算”，而是开始大规模补充并替代人类工作，直接侵吞体量更为庞大的“人力预算”（labor budgets）。全美数以百万计的企业正在调动这些智能体替自己干活，并心甘情愿为此买单，因为产出价值远大于调用成本。正如 Kevin Weil 所言，“今天的模型与智能体将是它们未来生命周期中最愚蠢的一天”，而算力与算法能力的爆发才刚刚进入早期阶段。 关于上市问题，Brad 坚定认为两家公司理应在今年走向公开资本市场。首先，机构投资者的配置需求极度旺盛；其次，两家企业当前面临的算力瓶颈比过去三年任何时刻都要严峻，极需从公开市场获得更低成本的巨额资本来扩建算力；最后，将普通散户排斥在资本主义史上最重要的一波财富浪潮之外具有不稳定性。他特别提及，英伟达首席执行官黄仁勋近期向两家公司联合注资 400 亿美元时曾明确表示，这笔钱将是他最后的私募投资，因为他预计两家公司今年内都将启动 IPO。 - **单月 60 亿美元**：Anthropic 仅在 28 天的 2 月份就录得 60 亿美元收入，增速改写了全球企业软件史。 - **预算维度的范式转移**：AI 采购从每年几万美元级别的软件/IT 预算，转变为直接吞噬数百万至数千万美元级别的人力与外包开支。 - **黄仁勋的判断**：英伟达注资 400 亿美元后明确预期两家公司将在年内上市，以此作为其私募阶段投资的收官之作。 --- ### [32:12](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1932s) Chamath 的警示：实验性预算还是真实生产力？ Chamath Palihapitiya 随后对这种狂热的收入增长提出了尖锐的冷思考。他直言，目前在真正的传统大型企业内部，根本找不到哪怕一个凭借 AI 实现持续性利润率扩张（margin expansion）的可靠范例，绝大多数部署本质上依然处于小型试水阶段。之所以能凭空冒出几十亿美元的营收，是因为每一家大公司的管理层都必须在董事会汇报中勾选“我们已部署 AI”的考核项。全美成千上万家企业在管理层考核压力下，为员工采购每月 200 美元以上的企业订阅，这种由“防御性心态”堆砌出的数字并不能证明收入的健康与持久。 Chamath 将其与 Snowflake 和 Databricks 进行了深度对比：使用这两家软件的大型客户都在借此创造巨额利润与真实收入，因为其产品深度嵌入在支撑核心营收的关键业务生产工作流（critical production workflows）之中。而今天的 AI 远未达到这一地位。在医疗领域，一旦模型出现误诊或程序编码错误，企业可能面临巨额罚款乃至牢狱之灾；在金融领域，若因 AI 模型的幻觉造成投资组合错配，合规处罚与法律诉讼将接踵而至。因此，没有一家受监管行业敢将核心操作流程完全托付给随机概率模型。 为了佐证这一观点，Chamath 援引了亚马逊近期引发行业震荡的内部禁令：由于 AI 智能体编写的代码直接导致 AWS 爆发了三到四起一级系统故障（Sev-1 faults），亚马逊随即颁布严格指令，严禁在未经人类工程师逐行复核批准的情况下将此类代码部署进生产环境。Chamath 感叹，AWS 能够拥有令人惊叹的“12个9”（99.9999999999%）的可靠性，全赖人类工程师编写的、永不出错的确定性代码。他警示大家不要盲目夸大当下的市场狂欢：“这是一个仍处于摸索早期的行业，大量涌入的都只是测试预算，绝不能把它过度包装成已经完成的企业级革命。” ```mermaid flowchart TD accTitle: 核心生产工作流与实验性AI部署对比 accDescr: 展示以Snowflake为代表的传统核心系统与当前面临Sev-1故障风险的生成式AI在可靠性上的本质差异 subgraph 核心确定性生产体系 ["核心确定性生产体系 (Snowflake / AWS)"] A["确定性代码与人工架构"] --> B["12个9的高可靠性 (SLA保证)"] B --> C["承载核心关键业务工作流"] C --> D["实现可持续的利润率扩张"] end subgraph 实验性生成式AI部署 ["实验性生成式AI部署 (当前现状)"] E["管理层‘AI Checkbox’考核"] --> F["海量每月200美元测试账号"] F --> G{"面临业务风险 (如 AWS Sev-1故障)"} G -->|严禁黑盒上线| H["退回‘人工复核+辅助工具’模式"] G -->|核心责任缺失| I["实验性年化流水 (易遭预算削减)"] end ``` --- ### [35:12](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2112s) 实验性流水 vs 生产级 ARR 的激烈辩论 主持人 Jason Calacanis 将讨论推向核心争议点：在剔除掉不要求企业级服务协议（SLA）、类似于 Netflix 订阅性质的 20 美元个人消费者开支之后，两家巨头数以百亿计的企业级营收中，究竟有多少是真正的生产级系统收入，又有多少只是随时可能蒸发的实验性支出？ Brad Gerstner 回应称，自己早在业内率先提出了“实验性年化流水”（experimental run rate revenue）与“真正年度经常性收入”（ARR）的区别，承认作为投资人必须时刻甄别哪些收入会重复发生。但他反驳认为，将全部大企业部署都归为实验是站不住脚的：像 Palantir、英伟达、美国政府以及美军各部门，已经将 AI 大模型推向了全面生产阶段，AI 甚至在当前伊朗的军事行动中成为了决定胜负的关键基础设施，绝非可有可无的玩具。Brad 认为，只要 3 月份的营收环比继续狂飙，就足以证明其实用性正在不可逆地沉淀为长期需求。 然而 Chamath 立即打断并给出了强硬反击。他指出，“实验性开支完全可以不受约束地持续很长时间”，不能把单纯的付费延续等同于业务生产化。Chamath 结合自身投资与实际运营一线的数据直言：“我自己就在这些基础模型之上搭建应用，我每年付给模型厂商数百万美元，每天消耗的 token 数量都在急剧增加。但真相是，我的经济产出和公司收入根本没有以同样的速度增长！”他犀利地指出，连站在技术最前沿的极客团队每季度开支翻三倍都换不来收入的三倍增长，更遑论那些反应迟钝数步的财富 1000 强传统企业。这种模型调用成本与实际产出效益脱节的现状，表明当前的繁荣背后隐藏着巨大的投机泡沫与认知落差。 - **消费者与开发者开支的区别**：数百万工程师虽有公司报销每人每月 200 美元的工具费用，但内部并未建立严密的产出验收机制，缺乏针对幻觉引发事故的 SLA 约束。 - **国防与政府刚需**：Brad 强调五角大楼与 Palantir 在实战与军事情报层面的全量部署证明了技术已具备最高等级生产力属性。 - **前沿团队的真实 ROI**：Chamath 披露其实际运营痛点——Token 账单每季度翻倍增长，但自身业务收入并未获得对等提振，警告大企业客户终将面临算力投入回报率（ROI）的现实考量。 ### [38:30](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2310s) 企业级 AI 的真实回报大辩论：代码生成成为首个破局点 关于企业级 AI 的投资回报率（ROI），现场展开了激烈的交锋。对话由迈克尔·戴尔（Michael Dell）近期的表态切入——戴尔指出，一年前各家公司还在苦苦寻找 ROI，而如今已经在 AI 投资中看到了实打实的高额回报。萨克斯（David Sacks）随即切入这场辩论的核心：当前数十亿美元的企业端收入中，究竟有多少属于生产环境中的真实落地，又有多少仍停留在概念验证（PoC）阶段？萨克斯指出，若审视目前的企业端收入，真正的超级杀手级应用（breakout use case）压倒性地集中在**代码辅助工具（Coding Assistance）**；相较于消费端普遍偏向于问答、搜索与写作聊天的聊天机器人，企业级市场的真金白银几乎全在为代码买单。 萨克斯分析认为，对代码的需求具有极高的可扩展性，因为软件工程长期以来都是整个人类经济中最受人力供应约束的领域。即便是在全球吸纳顶尖工程师能力最强的硅谷，人才荒也从未缓解，每一家初创公司的产品路线图瓶颈永远是“招不到足够的工程师”。对于财富 500 强（Fortune 500）等传统巨头而言，与硅谷争夺工程师更是难上加难。AI 极大地打破了这种长期存在的供应瓶颈，释放了海量的潜在代码需求。随着 Token 成本持续下跌、辅助工具日渐成熟，企业能够按量计费（metered basis）购买代码，彻底摆脱了过去繁琐的招聘筛选、入职维持以及提供 KIND 能量棒等各种硅谷福利的重负。 在萨克斯看来，企业通过 Agent 自动化进一步利用这些代码将成为下一个重大爆发点。针对查马斯（Chamath Palihapitiya）对企业落地的怀疑，萨克斯表示赞同麦肯锡（McKinsey）的一项研究结论：在财富 500 强企业中，除代码之外的大量试验性试点项目进展并不顺利，由于面临巨大的变革管理（change management）阻力，企业在全面业务重塑上依然充满不确定性。但仅仅是代码生成和随之而来的 Agent 生态，其市场空间就庞大到足以支撑行业持续扩张。 - **核心应用分化**：消费端偏向研究辅助与文本写作，企业端首个被完全验证的刚需是代码辅助与生成。 - **劳动力供需瓶颈**：初创企业和财富 500 强的共同瓶颈始终是软件工程师短缺，代码生成能力将研发资源的获取变成了像用水用电一样的按需服务。 --- ### [41:59](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2519s) 劳动力替代之争：大企业内部阻力与初创公司的全生产流改造 萨克斯对布拉德·格斯特纳（Brad Gerstner）使用的“劳动力替代（labor displacement）”一词提出了异议。他认为外界往往将百亿美元级的大模型收入与大规模裁员画上等号，但实际上，由于以往软件工程师供给被严重人为受限，AI 带来的更应该被称为“赋能与生产力增益（augmentation）”，如同“通上自来水与电力”一样激活原本无法执行的增量业务，而非单纯削减现有岗位。 杰森·卡拉卡尼斯（Jason Calacanis）则从落地场景给出了更犀利的实操洞察。他直言，观察 AI 从实验走向真实生产环境，绝不能看大企业。大公司内部天然存在抗拒新技术落地的官僚机制——中层管理人员深知，一旦成功推行 AI 自动化来缩减人力规模，其最终结果往往是“把自己的岗位也给优化没了”。因此，大型组织在自我保护机制下推进极其缓慢。 相反，杰森指出真正的生产级（production-ready）变革发生在初创公司前线。在当前的创业生态中，LLM 已经深度嵌入日常业务流：法律合同审查、市场营销文案、销售开发代表（SDR）获客线索挖掘、基础记账与人力资源（HR）流程等传统上需要雇佣外包、咨询顾问或全职员工的脏活累活（blocking and tackling chores），全部被大模型在生产环境下规模化接管。 ```mermaid flowchart TD accTitle: AI Adoption Barrier vs Startup Execution accDescr: Diagram illustrating big tech bureaucratic resistance compared to startup rapid production deployment. subgraph BigTech["大型传统企业 (Fortune 500)"] A["高层引入 AI 试点项目"] --> B["中层管理抵抗 (恐慌裁员波及自身)"] B --> C["变革管理受阻 / 试点失败"] end subgraph Startups["初创企业生态 (Startups)"] D["生存与效率导向"] --> E["直接接入大模型生产管线"] E --> F["法务 / 营销 / 记账 / SDR / HR 全面替代外包"] end ``` --- ### [44:03](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2643s) LLM 行业的 J 曲线困局：500 亿美元/吉瓦的数据中心算力经济学 讨论随后转向商业模式的残酷现实：这门昂贵的生意何时才能实现日历年盈利？杰森展示了一张由 Claude 生成的“大模型行业 J 曲线（J-curve）”图表，指出即便按保守的 5000 亿美元全行业累计投入计算，AI 领跑企业要想收回投资，也将复制特斯拉（Tesla）、优步（Uber）和亚马逊（Amazon）耗时十数年的漫长投资回收周期。 查马斯以其正在亚利桑那州建设的 **1 吉瓦（1 Gigawatt, 1GW）数据中心**项目为例，现场拆解了令人触目惊心的基础设施资本支出（CapEx）： 1. **预算失控膨胀**：该项目立项获批时最初预估总投资仅为 40 亿至 50 亿美元，随后一路攀升至 100 亿、150 亿、200 亿美元，如今仅包含供电外壳（powered shell）、土地购置、审批牌照、电网接入以及基础设施与团队，整体造价已狂飙至 **500 亿美元以上**。 2. **单位经济效益公式**：结合 OpenAI 首席财务官萨拉·弗莱尔（Sarah Friar）此前的披露——每部署 1 吉瓦算力大约对应每年 **100 亿美元的年化收入**（Annual Revenue）。 3. **回收周期测算**：算力本质上是“能源即智能（Energy equals Intelligence）”。以单吉瓦 500 亿美元资本支出对应每年 100 亿美元营收计算，企业仅实现名义上的投资打平（break-even）就需要 **5 到 6 年的投资回收期**，真正进入净利润兑现阶段已是第 6 至第 8 年。 查马斯强调，全行业正在拼命尝试压缩这条 J 曲线的深度与面积：寄望于更高效的新一代硅基芯片（如英伟达黄仁勋即将发布的结合 Groq 架构特性的新方案）、底层开源生态的演进等，来削减资本黑洞。然而目前的物理定律与工程现实依然冷峻，这笔巨额账单亟需科技行业在硬件效率层面实现结构性突破。 - **亚利桑那项目成本**：从最初预算 40-50 亿美元膨胀 10 倍以上，突破 500 亿美元。 - **算力创收指标**：1 GW 容量对应约 100 亿美元年营收。 - **资本回收周期**：打平需 5 至 6 年，利润窗口期推迟至 6 至 8 年后。 --- ### [46:10](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2770s) AI 叙事的两步走闹剧：从末日论融资到疯狂求生的公关反转 查马斯跳出单纯的财务数字，从长远的技术周期视角给出了一套宏观分析框架。他回顾指出，从 PC 与第一代互联网（Web 1.0）的电子商务浪潮，到以移动化（Mobile）和社交媒体（Social）为核心的第二代互联网（Web 2.0），再到如今的 AI，所有重大科技周期在进入实质性商业化阶段时都会上演一模一样的戏码。 查马斯将 AI 行业的演进归纳为两个充满讽刺意味的阶段： > **第一阶段：以“末日恐慌（Scary Doomerism）”进行 A/B 测试来疯狂吸金。** 创业者们在融资阶段不断测试投资人的心理底线，最终一部分 AI 阵营发现，制造末日恐慌是拿到天量资金最有效的利器。Anthropic 的达里奥·阿莫代伊（Dario Amodei）等人经常如同先知般公开发表言论，宣称“所有的工作都将被摧毁”、“模型已经开始具备自主意识（Sentient）”。在被末日论与造神论震撼的投资人面前，数百亿、甚至上千亿美元的融资支票轻松落袋。 > **第二阶段：资金到手后的现实反噬与全向变现。** 当融得巨资、模型训练完成并推向市场后，投资人开始严厉逼问：“收入究竟在哪里？”巨大的营收兑现压力迫使这些原本标榜“AI 安全与伦理”的先知们撕下伪装，开始不加节制地向任何潜在客户兜售大模型。查马斯特别提到了近期 Anthropic 面对五角大楼（美国国防部）态度的惊人 180 度大转弯（flip-flop）——当面对真正的变现考核与生存焦虑时，曾经高调鼓吹的宏大叙事瞬间让位于对订单的狂热追逐。 ### [47:40](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2860s) 硅谷巨头的公关混乱与自我矛盾 Chamath 继续抨击行业领导者在公共政策与国家安全事务上的轻浮态度。他直言，Anthropic 之前表现得就像个“轻浮不严肃的半吊子合作伙伴”（un-serious dilettante），结果美国政府直接放话要将其踢出局——这意味着数十亿美元的收入瞬间灰飞烟灭。这立刻引得投入数十亿美元的幕后投资人急眼并施压：“稍等一下，这绝对不行，你必须服从规矩、重回正轨。”于是 Dario Amodei 迅速反水改口（flip-flops），声称“非常抱歉，我不是故意的，我们来修复关系”。Chamath 认为，这一切闹剧恰恰证明 AI 行业仍处于极其早期的野蛮生长阶段，根本还没想清楚自己在人类社会中究竟该处于什么生态位。 为了剖析该问题的核心，现场播放了两段对比鲜明的行业领袖视频片段。Alex Karp 在讲话中警示，全行业包括硅谷都严重低估了新技术的颠覆性破坏力：如果这项技术严重摧毁了某一特定政党基本盘（受过高等教育、大部分投民主党的女性选民）的经济与政治权力，同时又让得不到扶持的军人及蓝领阶层感到被孤立，却妄想在政治上全身而退，那简直是“住进了疯人院”。AI 正在削弱受人文教育的民主党选民的经济地位，同时抬升接受职业培训的蓝领男性群体的经济权重。面对这种剧烈重构，社会必须就技术的用途达成共识，并向那些丢掉体面工作的人给出解释。更关键的是，如果在军事国防上将 AI 脱钩，人们将无法向美国大众交代为什么要承担撕裂社会肌理的巨大风险——发展 AI 唯一的正当理由就是维护国家安全，“如果我们不做，对手就会做，我们就会受制于他们的法治”。反观 Sam Altman 的回应则显得极为功利而扁平，他将大模型公司的商业模式直接归结为“售卖 Token”，并将其描绘为水和电一样的公用事业（Utility），任由用户付费按量取用。 ```mermaid flowchart TD accTitle: 硅谷三大 AI 叙事模式对比 accDescr: 展示 Dario Amodei、Alex Karp 与 Sam Altman 三人截然不同且相互冲突的公众沟通定位 subgraph Narratives["三大分化话语体系"] Dario["Dario Amodei (Anthropic)<br/>'神明救世论'<br/>宣称打造有感知超神，预言普通人来日无多"] Alex["Alex Karp (Palantir)<br/>'地缘对抗与社会契约论'<br/>警告阶层颠覆风险，必须绑定军工与国家安全"] Sam["Sam Altman (OpenAI)<br/>'公用事业论'<br/>按表计费卖 Token，如同自来水与电力"] end subgraph Impact["现实公关与民意后果"] Chaos["公开进行 A/B 测试式混乱发声"] Backlash["社会恐慌蔓延，民调支持率跌至谷底"] end Narratives --> Chaos Chaos --> Backlash ``` Chamath 总结道，行业如今充斥着三种互不兼容的说辞：Dario 在鼓吹“拥有感知力的超级神明，只有我们能保护你，但普通人来日无多”；Alex Karp 严厉警告不可既颠覆社会阶级又回避国家责任；Sam Altman 则只想把 Token 当成自来水按表计费。疯狂的融资泡沫和过热的营收狂欢，彻底掩盖了企业建立可靠、严谨公关体系的责任。各家甚至直接在公众面前对核心叙事做“A/B 测试”。Chamath 亮出民意调查图表痛批，由于这些不负责任的发声，AI 的公众好感度已跌落至仅略高于民主党和独裁国家的水平，甚至连执法机构 ICE（美国移民及海关执法局）的支持率都高于 AI。他呼吁行业必须保持清醒克制：大方承认眼前有海量实验与不确定性，不要蔑视行业法律与执照规范，以稳健负责的姿态推进。 - 关键观点：Dario Amodei 因面临失去数十亿美元政府订单的风险，在投资人压力下迅速向五角大楼低头道歉。 - 选民结构重塑：技术重创人文教育背景的白领女性群体，反而强化了职业蓝领男性的相对地位，埋下政治动荡隐患。 - 民调惨状：AI 在美国的受喜爱程度已沦落到接近独裁国家的边缘，甚至低于 ICE。 ### [52:35](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3155s) 产业恐慌与历史周期律之辩 面对 Chamath 的抨击，Jason 将话题引向 Brad Gerstner，询问 AI 产业是否正面临严重的公关危机。与中国大众将 AI 视作生产力丰裕与发展机遇截然相反，美国民众普遍陷入了焦虑与恐慌，担忧技术失控导致大规模失业、贫富悬殊加剧以及“富者愈富、穷者愈穷”。各大科技巨头各自为政、缺乏协调，甚至不断发表吓坏公众的激进言论。 Brad 坦承这确实是一场合情合理的批判。从历史维度来看，19 世纪末的工业革命初期同样爆发了强烈的社会排斥，不仅伴随着街头示威、暴力冲突，还催生了针对“强盗男爵”（Robber Barons）的激烈阶级斗争。因此，面对这种体量的产业变革，普通大众产生失业恐惧是完全符合历史规律的，部分业内人士对未来的末日式预测更是起到了反作用。Brad 强调，今年将是 AI 产业证明自身对人类切实有益的关键决战年，尤其需要在医疗健康、药物研发和普惠教育等领域拿出铁证。 尽管民意普遍消极，Brad 依然持乐观态度。他援引数据指出，当今美国有 70% 的工作岗位在 40 年前根本不存在，数字化浪潮虽然淘汰了一批旧工种，但最终通过做大蛋糕创造了更多繁荣。虽然 Dario 警告本轮变革的迭代速率极快，人类社会可能难以及时完成就业替代，但历史大势终将走向丰饶。当 Brad 试图将讨论拉回“这是否为好投资”时，Chamath 当场打断辩驳，直指核心根本不是投资赚钱的问题——大模型公司是在淘金热中稳赚不赔的“卖铲人”，而下游企业则是买下铲子去淘金的大冤种，大部分应用层最终都会被淘汰。整个行业充斥着未解难题，必须摒弃虚妄宣传，老老实实面对现实。 - 历史镜鉴：工业革命初期同样经历了严重的阶级撕裂与社会抗议，颠覆性技术必然伴随大众阵痛与防卫心理。 - 就业重组数据：美国目前 70% 的现有工作岗位在 40 年前均不存在，技术进步长期来看具备再造就业的扩容效应。 - 淘金隐喻：Anthropic 和 OpenAI 扮演着高利润“卖铲人”角色，买铲淘金的下游应用生态绝大多数将面临破产清洗。 ### [56:15](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3375s) 中美 AI 乐观度断层与地缘民调落差 David Sacks 加入讨论，明确指出 AI 行业目前的公关答卷无疑是灾难性的。巨头们不仅没有建立公众信任，反而把大众吓得魂飞魄散，导致 AI 的公众支持率一路俯冲至近乎与伊朗等敌对地缘实体看齐的尴尬境地。Sacks 调侃称，伊朗花了 47 年才把自己的公共声誉搞砸，而硅谷的这帮天才只用了短短两年就达成了同等“成就”。 ```mermaid flowchart LR accTitle: 斯坦福 HAI 全球 AI 态度民意调查对比 accDescr: 对比中国与美国受访者在面对 AI 利弊时的态度分歧 subgraph Survey["核心问卷: AI 带来的益处是否大于危害？"] China["中国受访群体<br/><b>约 80% 赞同</b><br/>普遍秉持技术乐观与丰饶预期"] US["美国受访群体<br/><b>仅 30% 左右甚至更低</b><br/>普遍陷入失业焦虑与技术敌意"] end China -.-> Pros["积极融入与拥抱革新"] US -.-> Cons["技术抵触与政治反噬"] ``` Sacks 进一步指出，这种对技术的极度恐慌在很大程度上属于“美国特有症候群”。全球各大主要经济体的情绪追踪数据显示，外部世界对 AI 的未来远比美国本土要乐观得多。斯坦福大学 HAI（人类中心人工智能研究所）曾发起过一项全球 AI 乐观度调研，当被问及“你认为 AI 对人类而言是利大于弊还是弊大于利”时，约 80% 的中国受访者坚定给出了利大于弊的正面回答；而在美国，这一比例仅徘徊在 30% 左右的极低水平，近期甚至可能进一步走低。不仅是中国，整个亚洲国家的大众普遍对 AI 技术展现出了积极接纳与拥抱的开放态度，与美国国内蔓延的悲观与敌视形成了极度鲜明的跨文化反差。 - 民意两极化：斯坦福调研显示，中国约 80% 的受访者对 AI 持正面乐观态度，而美国受访者的认可度仅有 30% 左右甚至更低。 - 地域分化：恐慌与反技术浪潮高度集中于美国本土，亚洲国家普遍将该技术视为赋能社会的增量红利。 ### [57:17](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3437s) AI 恐慌与监管俘获：为何科技巨头热衷于宣扬末日论 大卫·萨克斯（David Sacks）指出，与亚洲等地区的乐观态度相比，美国和西欧对 AI 普遍弥漫着悲观情绪。这在很大程度上不是 AI 本身的技术现实决定的，而是由西方独特的媒体环境以及背后的推手塑造的。好莱坞数十年来产出了大量反乌托邦电影，而如今各大 AI 公司的 CEO 们又在公关上表现得极其糟糕——他们不断在公开场合宣扬 AI 将颠覆一切、让所有人失业。萨克斯直言，这种言论绝非偶然：一部分 CEO 或许只是缺乏基本的沟通技巧，但另一些人则是深思熟虑地在推行“监管俘获（Regulatory Capture）”战略。 查马斯·帕利哈皮蒂亚（Chamath Palihapitiya）表示赞同，指出这不仅源于科技领袖的盲目自大，也是一种商业卡位。他们一方面为了以极高估值进行巨额融资而在市场上大吹大擂，另一方面则企图人为制造监管反弹，从而促使政府出台严格的牌照准入与许可机制（Licensing and Permissioning Schemes），顺理成章地将后来者阻挡在门外。而在西方媒体“流血才能上头条（If it bleeds, it leads）”的流量机制助推下，恐慌叙事被无限放大，正向温暖的故事反而无人问津，使得 AI 在民意调查中几乎沦为除民主党和伊朗之外最不受欢迎的议题。 ```mermaid flowchart TD accTitle: 监管俘获与舆论恶性循环 accDescr: 展示科技巨头末日叙事如何通过游说转化为扼杀创新的立法闭环 subgraph Narrative["舆论引爆"] A["CEO 宣扬'AI抢走所有人工作'<br/>(融资宣传 + 巨头卡位)"] --> B["媒体渲染末日叙事<br/>(越猎奇越有流量)"] end subgraph InterestGroups["行业游说"] B --> C["行业协会借题发挥<br/>(医生、律师协会自保防颠覆)"] C --> D["游说州议会出台禁令与准入限制"] end subgraph BadOutcomes["现实恶果"] D --> E["出台荒谬法案<br/>(禁止 AI 提供医疗与法律建议)"] E --> F["底层弱势群体丧失平权工具<br/>(请不起律师与家庭医生)"] end ``` ### [58:35](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3515s) 荒谬的保护主义立法：以纽约州拟禁 AI 法律与医疗咨询为例 纽约州即将出台的一项新法案引发了全场的猛烈抨击：该法案计划直接取缔由 AI 聊天机器人提供的医疗和法律咨询。萨克斯与杰森·卡拉卡尼斯（Jason Calacanis）痛斥这是“有史以来最愚蠢的立法”。从投资回报率和普惠价值来看，医疗与法律咨询恰恰是 AI 能给普通消费者带来的最大红利，而这项禁令伤害最深的正是最底层的穷人。 杰森描绘了现实中触目惊心的鸿沟：贫困阶层根本请不起按小时计费的昂贵律师，当他们面临被房东驱逐等生死攸关的危机时，AI 是他们唯一能依靠并做法律研究来为自己辩护的武器；同样，许多没有医保、看不起全科医生的低收入人群，也是依靠 AI 来排查病症、寻找救济方案。AI 本是抹平资源鸿沟、实现阶层平权的伟大工具，但如今各行业的既得利益集团——如各州的法律与医疗行业协会——为了保护自己的地盘免受技术颠覆，正顺水推舟地利用 AI 恐慌向州议员施压游说，最终促成了这类荒唐的立法封锁。 - **阶层打击**：禁令直接剥夺了低收入人群在面对房屋驱逐诉讼或突发疾病时获取免费专业级建议的权利。 - **游说闭环**：AI 公司高管向外界宣扬技术即将摧毁世界经济；传统行会借机恐吓议员“必须踩刹车”；纽约州立法者顺水推舟出台扼杀创新的保护主义法案。 ### [01:00:20](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3620s) EA 末日论智库的巨额金权与伯尼·桑德斯的算力停建令 讨论随后聚焦到伯尼·桑德斯（Bernie Sanders）本周抛出的重磅主张——他公开发表言论，呼吁全美应当全面暂停（Moratorium）所有新建数据中心的审批与建设，理由是“AI 太危险了”。萨克斯敏锐地指出，桑德斯的演讲稿中有一句论调几乎是逐字逐句抄袭自“生命未来研究所（Future of Life Institute, FLI）”的宣传口径，即所谓“AI 面临的监管居然比一家三明治店还要少”，这完全是脱离现实的谎言。 萨克斯揭露，FLI 背后是由有效利他主义（Effective Altruism, EA）信徒资助的“末日论智库（Doomer Think Tanks）”，这些机构坐拥数十亿美元的雄厚资金。例如以太坊创始人 Vitalik Buterin 曾向其捐赠了价值 6 亿美元的柴犬币（Shiba Inu / Dog coins），达斯汀·莫斯科维茨（Dustin Moskovitz）等富豪也在背后长期注资。这些富甲一方的末日派智库四处资助新闻机构的记者研究金与编辑席位，在主流媒体中大量安插撰写负面报道的“笔杆子”，并不遗余力地在全美煽动数据中心的邻避效应（NIMBYism）。他们散布所谓“数据中心会推高居民电价”（事实上科技公司已承诺承担增量成本并自建独立发电厂）以及“耗尽水资源”（现代数据中心早已全面采用循环水冷技术）等虚假信息，试图全方位阻遏 AI 的发展进程。 - **巨额政治献金**：生命未来研究所等 EA 智库手握数以十亿计的资金储备，通过设立新闻基金和学术教席深度操纵舆论叙事。 - **虚假环保叙事**：针对数据中心的缺电、耗水攻击大多站不住脚，现代超大规模数据中心不仅使用封闭式内循环水冷，而且科技巨头正在大规模部署自备能源。 ### [01:02:26](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3746s) 算力被废！每年 1200 亿美元收入化为乌有，德州成为大赢家 查马斯拿出了一组详实的追踪图表，痛陈失控的公关叙事对美国算力基础设施造成的灾难性破坏。数据显示，在全美所有遭遇地方民众抗议的数据中心项目中，最终有大约 40% 会直接被取消。在 2023 年，这还只是个别案例（仅有 1 到 2 座被取消）；但随着 2024 年科技行业混乱且极端的末日公关，地方层面的对抗情绪被彻底点燃。 ``` 【数据中心被取消造成的年化收入损失测算】 2025 年损失： 5 GW 容量取消 × $100 亿/GW = $500 亿/年 2026 年预期： 7 GW 容量取消 (约40座项目) = $700 亿/年 ───────────────────────────────────────────── 两年累计从经济盘中抹除的年化营收： $1,200 亿/年 ``` 到 2025 年，局势彻底失控，全美共有约 25 个数据中心项目被迫取消，损失了整整 5 吉瓦（GW）的电力容量。按照 OpenAI 首席财务官萨拉·弗里亚尔（Sarah Friar）的收入测算模型（每吉瓦约对应 100 亿美元年营收），这直接导致每年 500 亿美元的年化收入化为泡影，连带使联邦和地方政府痛失数以百亿计的税收。而在 2026 年，仅截至 2 月底，全美就有约 100 个数据中心正遭遇抗议；若按 40% 的取消率推算，将有约 40 个项目流产，意味着又有 7 吉瓦容量折损、每年再损失 700 亿美元收入。两年之内，原本可为美国带来的 1200 亿美元年化营收被硬生生抹除。 查马斯指出，目前被取消的项目几乎全部集中在弗吉尼亚州和印第安纳州。相比之下，德克萨斯州因民众反对而导致的数据中心取消数量至今为零。德州目前拥有超过 150 吉瓦的数据中心容量接入申请，州长阿博特（Greg Abbott）和参议员特德·克鲁兹（Ted Cruz）对算力投资敞开怀抱，甚至被戏称“只要在推特上艾特他们，他们就会请你打扑克并光速放行”。 - **40% 死亡率**：凡是遭到地方组织抗议阻挠的数据中心，最终有近四成胎死腹中。 - **弗吉尼亚与印第安纳**：全美数据中心建设受阻最严重的重灾区。 - **德州对比**：凭借充裕的电网与友好的政治环境，德州手握 150+ GW 的庞大管线，零项目因地方抗议搁浅。 ### [01:04:59](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3899s) 开源模型与端侧算力的奇袭：挑战闭源巨头的千亿美元估值 在节目的最后，杰森将话题抛给了重仓投资了头部闭源模型的 Altimeter 创始人布拉德·格斯特纳（Brad Gerstner），抛出了一个直击商业模式核心的尖锐问题：开源模型的爆发式成长和苹果芯片（Apple Silicon）端侧算力的飞跃，是否正在成为闭源龙头巨额估值的巨大逆风？ 杰森透露，在他自己的公司体系内，目前已有高达 85% 的 token 消耗被开源模型接管。他接触到的几乎所有初创公司都在采取同样的架构策略：在本地或低成本集群部署开源模型处理日常绝大部分负载，只有在遇到极其复杂、自身无法解决的高难度任务时，才会调用最顶级的收费商业大模型。不仅如此，安德烈·卡帕西（Andrej Karpathy）刚在上个周末发布了划时代的 AutoResearch 自主研究工具，全球各地的极客开发者们正通过 OpenClaw 等工具组装并自主训练、微调大型语言模型。一条完全不依赖闭源垄断的平行开源生态正在飞速壮大。布拉德随即给出了回应，明确表态自己对开源生态的蓬勃发展不仅高度关注，而且持有极大的热情。 ### [01:06:04](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3964s) 混合模型架构下的 AI 市场爆发：远超预期的巨额 TAM 与活动公告 深入观察头部前沿企业的实际落地方式可以发现，当前最先进的科技公司正在普遍采用“模型组合”（Ensemble of models）策略：通常利用前沿闭源实验室（Frontier labs）的模型来进行高维度的规划与推理（Planning），随后交由开源模型（Open source models）去负责低成本、高并发的具体执行（Execution）。然而，真正震撼业界的并非仅仅是开源模型的快速逼近，而是在存在如此强大开源替代品的环境下，前沿闭源实验室依然展现出人类商业史上前所未见的光速吸金能力——例如 Anthropic 竟能在单月内暴增五六十亿美元的营收。这种态势表明，AI 的潜在市场总额（TAM）要远远超出所有人的想象，就像当年 Uber、Google 和 Amazon 所经历的跨越式拐点一样，开源模型固然会分得一杯羹，但前沿模型的市场空间依然庞大到不可思议。 在切入下一话题前，节目组穿插了几项重要公告：即将在洛杉矶举办的 All-In Summit（9月13日至15日）以及 Liquidity 活动门票几近售罄，听友可前往官网抢票；同时，节目组正式上线了听众调查问卷（allin.com/survey），呼吁早期铁粉与支持者积极参与填写反馈。 - **前沿模型商业化奇迹**：尽管开源模型几乎紧咬前沿水平，Anthropic 仍创下单月增长 50 亿至 60 亿美元营收的技术史纪录。 - **产业共识**：AI 跨越核心临界点，其实际 TAM 远比行业最初预估更宏大，闭源实验室与开源生态正在走向分工协作而非单纯的零和博弈。 --- ### [01:07:48](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4068s) 华盛顿州通过“富翁税”，星巴克创始人霍华德·舒尔茨出走迈阿密 全网热议的新闻聚焦在华盛顿州刚刚通过的“富豪税”（Millionaire Tax）。该法案规定，从 2029 年起，对年收入超过 100 万美元的群体额外征收 9.9% 的税款。预算中心预估该税种将波及约 30,000 个高收入家庭，为州普通基金每年带来约 40 亿美元进账，名义上用于公共教育、高等教育和医疗保健。极其讽刺的是，法案通过当天，在西雅图扎根 44 年的星巴克传奇创始人霍华德·舒尔茨（Howard Schultz）便火速宣布迁居佛罗里达州迈阿密，豪掷 4400 万美元在瑟夫赛德（Surfside）买下临海公寓。 主持人团队随即开启了对舒尔茨的冷嘲热讽与幽默调侃。Jason 戏谑舒尔茨也玩了一出“搬家避税大法”，而大家则戏称他离开西雅图纯粹是因为“爱上了迈阿密的古巴三明治”。更尴尬的是，舒尔茨此前参与总统竞选时曾公开声称“富人应当承担更多税负”，如今却在政策落地当天用脚投票。这也是继 2023 年 11 月杰夫·贝索斯（Jeff Bezos）因华盛顿州出台 7% 资本利得税而迁居迈阿密之后的又一重磅出逃案例。 ```mermaid flowchart TD accTitle: 州级富豪税引发的高净值人群迁移链条 accDescr: 展示华盛顿州出台富豪税如何促使富豪离开并产生溢出效应的流程逻辑 subgraph Policy ["政策出台"] A["华盛顿州通过富豪税法案<br/>(年入 >$100万 额外加征 9.9%)"] B["预计波及 30,000 家庭<br/>预期增收 40 亿美元"] end subgraph Exodus ["资本与人才出逃"] C["星巴克创始人舒尔茨离开西雅图<br/>(斥资 $4400万 入住迈阿密 Surfside)"] D["此前贝索斯出走佛州<br/>(规避 7% 资本利得税)"] end subgraph Feedback ["治理后果"] E["高流动性高净值人群加速流失"] F["中产阶级被迫填补财政预算漏洞"] end A --> B A --> C A --> D C --> E D --> E E --> F ``` --- ### [01:10:30](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4230s) 查马斯痛斥西海岸财政昏招：胡佛研究所蒙特卡洛模拟揭示 250 亿赤字陷阱 面对 Jason 提出“为何地方政客明知高净值人群具备极强流动性、能随时迁往佛州、德州甚至迪拜与新加坡，却依然执意推行高达 60%-70% 综合税率”的质问，Chamath 直言不讳地批评西海岸各州的政客“不仅施政极其低效，而且在算术上愚蠢至极”。华盛顿州不过是在步加利福尼亚州的后尘，而这类激进税制在州层面上从来就无法奏效。 Chamath 引用了胡佛研究所（Hoover Institution）最新发布的评估报告。该研究针对加州拟议的富豪税进行了 100,000 次蒙特卡洛模拟测算： - **净现值灾难**：在高达 71% 的模拟情景中，该税收方案的净现值（NPV）均为负数，预期将给州财政带来约 250 亿美元的巨大资金黑洞。 - **荒谬的估算错误**：立法者在数据模型中严重高估了加州亿万富翁的真实存量、低估了这批人目前缴纳的税收贡献，同时虚高了未来能收上来的税额——原本预计增收 1000 亿美元，实际恐怕只有 400 亿美元。 - **连带受害的中产阶级**：由于出逃富豪原本每年贡献 30 亿至 50 亿美元税收，富人跑路后造成的税基流失与财政亏空，最终只能转嫁给加州现存的 1000 万个中产阶级家庭，相当于每户每年需额外背负约 2500 美元税负。 华盛顿州在长达 23 小时的马拉松辩论后强行通过类似法案，其未来两年的结局注定比加州更加惨烈。大家指出，政府部门的资源浪费极其惊人，在消除体制内欺诈（Fraud）与行政挥霍之前，政客根本不具备任何提高税收的“道德制高点”。 --- ### [01:13:25](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4405s) 联邦财富税狂想与“镀金时代”阶级斗争历史回潮 讨论进一步延伸到联邦层面的极端税改提议。参议员伯尼·桑德斯（Bernie Sanders）与众议员罗·康纳（Ro Khanna）近期抛出了全美财富税（National Wealth Tax）提案。与加州仅征收一次性 5% 的方案不同，该联邦提案主张每年征收 5% 的资产净值税。主持人严厉抨击这一逻辑等于在 20 年内由联邦政府将公民合法积累的财富全部“合法充公”，本质上就是赤裸裸的社会主义财富掠夺。 大家指出，税收制度必须遵守契约精神——如果纳税人已经为自己辛勤赚取的收入缴过税，剩余的合法资产就应当神圣不可侵犯，绝不能因为政府 10 年后缺钱就跑回来强行没收。这种公然抢夺资产的政治口号令人联想到 1870 至 1920 年的“镀金时代”（Gilded Age）。当时针对卡内基、洛克菲勒等产业巨头的财富仇视，曾直接激化为大铁路罢工、普尔曼大罢工等导致数百人丧生的流血社会冲突。政客为了选票刻意煽动阶级对立（Class warfare）极其危险，各方必须为狂热情绪降温，将精力放回财政资源配置效率与治理欺诈等务实议题上。 - **联邦财富税方案**：主张每年对资产征收 5%，等同于在数学上用 20 年时间实质性剥夺公民的私有产权。 - **历史镜像与社会警示**：回顾镀金时代的阶级冲突与流血罢工，痛批当代政客以再分配为名煽动民粹与仇富情绪，无视社会生产力的根基。 ### [01:15:04](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4504s) 加州富豪税博弈：工会分赃不均与2028全美财富税暗流 围绕加州酝酿的“富豪资产没收税”（Billionaire Tax Act, BTA），布拉德·格斯特纳（Brad Gerstner）指出当前州内政治生态出现了一个耐人寻味的现象：就连加州教师工会也表态反对这项针对富豪的税案，因为他们清醒地意识到，资产没收政策必将引发资本大逃亡，最终导致流入加州公共教育体系的财政税收反而缩水。不仅正在竞选下届加州州长的圣何塞市长马特·马汉（Matt Mahan）坚决反对，就连现任民主党籍州长加文·纽森（Gavin Newsom）也公开予以抵制。格斯特纳强调，所有政治领袖都必须站出来表明立场——这不仅是反对一项税收，更是一场根本性的路线抉择：你究竟是站在商业、创业精神、创新以及推动全州经济增长的一边，还是站在其对立面。如同美国在镀金时代没有像欧洲那样放弃资本主义和创业精神，美国如今必须捍卫这一根基。 大卫·萨克斯（David Sacks）随即切入更残酷的政治现实：加州部分工会目前之所以反对这项资产剥夺税，纯粹是因为这次提议由联合医疗工会西部工会（SEIU-UHW）单方面主导发起，其他工会没能“分到一杯羹”（wet their beak）。坊间已有传言，加州教师协会（California Teachers Association, CTA）正在紧锣密鼓地起草属于自己的富豪资产剥夺方案，专攻下一届选举。萨克斯讽刺道，到了2028年，各大工会一定会统一口径，“让所有的猪都围到同一个食槽前”，届时所有公会结盟协同作战，法案通过的几率将大幅激增。 更具长远威胁的是，萨克斯预测到2028年，征收全美范围的全国性财富税将彻底成为民主党竞选纲领的“基本盘”（table stakes），伯尼·桑德斯（Bernie Sanders）激进的财富没收立场已逐步演变为主流党意。萨克斯特别点破了加文·纽森看似反对该法案背后的政治玄机：纽森在声明中声称“单个州无法单独推行这项政策，因为还有其他49个州在展开资本竞争”。纽森故意隐去了后半句潜台词——这必须由联邦政府在全国层面强制统一执行，让资本无处可逃。纽森正是通过这种巧妙的话术为自己预留回旋空间，以便在2028年向全国最高权力发起冲刺时顺理成章地全面拥抱全国财富税。 ```mermaid flowchart TD accTitle: 财富税政策推演路径 accDescr: 展示加州工会博弈如何演进为2028年全国性财富税倡议的推演流程。 subgraph CA_Current ["加州当前僵局 (2026)"] SEIU["SEIU-UHW 单方推行 BTA"] --> Clash["其他工会未分润 / 纽森表态反对"] Clash --> Argument["理由: 单州无法竞争，将致资本外逃"] end subgraph Projection_2028 ["2028 政治演变预测"] Unite["全工会联盟重新整合方案<br/>(CTA + SEIU 等利益均沾)"] FedTax["纽森等温和派转向:<br/>推动联邦层面全国统一财富税"] Unite --> TableStakes["财富税成为民主党全国竞选纲领标配"] FedTax --> TableStakes end CA_Current --> Projection_2028 ``` - **工会分歧本质**：加州教师工会反对当前的富豪税提案，核心原因在于 SEIU-UHW 独占了利益分配，未给其他公共工会预留配额。 - **政治人物对立**：圣何塞市长马特·马汉（竞选州长）与现任州长加文·纽森均对当前版本的 BTA 表达了反对。 - **纽森的潜台词**：纽森提出加州不能单打独斗对抗其他49个免税或低税州的竞争，实际上是为未来支持联邦统一征收财富税做战略铺垫。 --- ### [01:17:23](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4643s) 抵御社会主义浪潮的第一性原理：用AI与去监管解决民生之痛 杰森·卡拉卡尼斯（Jason Calacanis）提出了终结这种财富没收和社会主义狂潮的根本解法。他主张从第一性原理出发，剖析普通美国家庭与父母的真正诉求：他们想要的无非是让孩子接受优质教育、能够负担得起一套体面的住房、享有健全实惠的医疗保健，以及吃上健康的食物。杰森认为，AI 与技术创业者完全有能力直击这几个核心痛点，前提是政府必须准许民营力量进军这些被过度监管的深水区。 杰森逐一列举了改革路径：在教育领域，必须彻底打破僵化的高等教育认证垄断体制；在住房领域，必须废除严苛的土地与建筑审批限制，借鉴德克萨斯州、佛罗里达州和内华达州的发展经验迅速释放供给；在医疗保健领域，这正是 AI 能够发挥降本增效巨大威力的地方，通过赋能以患者为主导的“自主医疗模式”，可以实现医疗支出的指数级下降。杰森大声疾呼，只要政府与企业家合力解决住房、教育与医疗的供给问题，人民自然丰衣足食，根本不需要倒退向没收民财的社会主义泥潭。 在这场变革中，杰森认为唐纳德·特朗普（Donald Trump）扮演着关键的“监管粉碎机”角色。特朗普打破传统阻力，促使核电重返美国国家战略议程，这是此前历任总统都未能实现的壮举；如今他完全有能力在住房去监管领域复制这一突破。杰森最后呼应了此前对中东战争的批评，郑重呼吁华盛顿必须认清主次：“我们必须砸碎这些阻碍民生的监管枷锁，不要再去中东挑起毫无意义的对外战争，而是集中精力为美国人多建房子。” - **美国家庭核心诉求**：优质子女教育、平价自住房屋、负担得起的优质医疗、健康平价的食品。 - **核心打破方向**：破除教育资格认证垄断（Education Accreditation Cartel）、放宽各州建筑分区管控（参考德州与佛州）、运用 AI 重构自主医疗生态。 - **政策与领导力评价**：特朗普被视为具备打破官僚监管能力的催化剂，正如其成功重启全美核能议题一样，应将这种势头转化为国内住房建设，而非耗费资源于海外战事。 --- ### [01:18:55](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4735s) 尾声调侃与播客收尾：抢占最后发言权与全剧终曲 在杰森完成慷慨激昂的总结陈词后，查马斯·帕里哈皮蒂亚（Chamath Palihapitiya）立刻上线打破煽情气氛，笑着挖苦道：“J-Cal，你永远都得给自己抢最后一句发言权。”杰森毫不示弱地笑称：“因为我是主持人，我自然排在最后！你要是想当主持人，下次你来主持我第一个讲。大家要不然给我鼓鼓掌，要不然就朝我扔烂番茄吧！”萨克斯随之笑着打圆场表示认可，几位好友在欢声笑语中完成了最后的互怼。 杰森正式为本期节目拉下帷幕，向特邀做客的嘉宾布拉德·格斯特纳表达了诚挚感谢：“感谢好兄弟布拉德（Bestie Brad）的加盟！”随后，标志性的片尾洗脑电音与历年名场面语音集锦同步响起——萨克斯的“让赢家继续狂奔”（Let your winners ride）、“雨人萨克斯”（Rain Man）、弗里德伯格的“藜麦女王”（Queen of Quinoa）、粉丝开源恶搞热梗以及关于宠物狗在车道大便的经典爆笑切片依次闪回，为这场长达八十分钟、横跨中东战争危机、AI超级营收周期、开源生存死局与全美财富税风暴的重磅对话画上了句号。 ## Notable quotes > [00:00](https://www.youtube.com/watch?v=HiVej9_fvl8&t=0s) 你把心掏出来托得恰到好处，要是手臂再抬高 5 度向外伸直，那就成行纳粹礼了，那些‘特朗普账户’可就要变成超级种族主义账户了！ — Jason Calacanis > [10:19](https://www.youtube.com/watch?v=HiVej9_fvl8&t=619s) 宣布胜利并抽身退出 — David Sacks > [19:09](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1149s) 关于中国、中国、还是中国 — Chamath Palihapitiya > [28:59](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1739s) 今天的模型与智能体将是它们未来生命周期中最愚蠢的一天 — Kevin Weil (引述) > [32:12](https://www.youtube.com/watch?v=HiVej9_fvl8&t=1932s) 这是一个仍处于摸索早期的行业，大量涌入的都只是测试预算，绝不能把它过度包装成已经完成的企业级革命。 — Chamath Palihapitiya > [35:12](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2112s) 我自己就在这些基础模型之上搭建应用，我每年付给模型厂商数百万美元，每天消耗的 token 数量都在急剧增加。但真相是，我的经济产出和公司收入根本没有以同样的速度增长！ — Chamath Palihapitiya > [47:40](https://www.youtube.com/watch?v=HiVej9_fvl8&t=2860s) 如果我们不做，对手就会做，我们就会受制于他们的法治 — Alex Karp (引述) > [56:15](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3375s) 伊朗花了 47 年才把自己的公共声誉搞砸，而硅谷的这帮天才只用了短短两年就达成了同等‘成就’。 — David Sacks > [01:00:20](https://www.youtube.com/watch?v=HiVej9_fvl8&t=3620s) AI 面临的监管居然比一家三明治店还要少 — Bernie Sanders (引述) > [01:15:04](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4504s) 单个州无法单独推行这项政策，因为还有其他49个州在展开资本竞争 — Gavin Newsom (引述) > [01:17:23](https://www.youtube.com/watch?v=HiVej9_fvl8&t=4643s) 我们必须砸碎这些阻碍民生的监管枷锁，不要再去中东挑起毫无意义的对外战争，而是集中精力为美国人多建房子。 — Jason Calacanis ## Mentioned Brad Gerstner, Jason Calacanis, Chamath Palihapitiya, David Sacks, David Friedberg, Donald Trump, Jared Kushner, Ivanka Trump, Larry Page, Sergey Brin, Mark Zuckerberg, Mojtaba Khamenei, Chris Wright, Tucker Carlson, Marjorie Taylor Greene, Joe Rogan, Matt Walsh, Megyn Kelly, Kristi Noem, Dario Amodei, Sam Altman, Kevin Weil, Jensen Huang, Michael Dell, Sarah Friar, Alex Karp, Vitalik Buterin, Dustin Moskovitz, Bernie Sanders, Greg Abbott, Ted Cruz, Andrej Karpathy, Howard Schultz, Jeff Bezos, Ro Khanna, Matt Mahan, Gavin Newsom, Invest America (Trump Accounts), Polymarket, Goldman Sachs, International Energy Agency (IEA), Wall Street Journal, OpenAI, Anthropic, ChatGPT, Claude, Claude Code, Codex, Snowflake, Databricks, SpaceX, Nvidia, Palantir, AWS, McKinsey, Future of Life Institute (FLI), Stanford HAI, Altimeter, Apple Silicon, AutoResearch, OpenClaw, Starbucks, Hoover Institution, SEIU-UHW, California Teachers Association (CTA) ## Video description (0:00) The Besties welcome Brad Gerstner! (3:48) Economic fallout of the Iran War, escalation scenarios, impact on midterms (19:18) Off ramp strategies, Gulf state involvement, the China angle (27:05) Anthropic and OpenAI scaling revenue faster than any company ever (46:11) AI's PR disaster, open source's future (1:07:51) Washington passes "Millionaire Tax," Howard Schultz bails for Miami Follow Brad: https://x.com/altcap Take the survey: https://allin.com/survey Apply for Liquidity: https://allinliquidity.com Apply for Summit: https://theallinsummit.com 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.google.com/finance/quote/BZW00:NYMEX https://www.cnbc.com/2026/03/11/cargo-ship-struck-strait-of-hormuz-uk-iran-war.html https://www.cnn.com/world/live-news/iran-war-us-israel-trump-03-12-26?post-id=cmmnhwyod000l3b6wdinc1dw5 https://www.wsj.com/world/middle-east/ending-iran-war-quickly-carries-big-risks-for-the-u-s-and-allies-60c003de https://polymarket.com/event/us-forces-enter-iran-by https://x.com/altcap/status/2029223717356879931 https://www.wsj.com/business/energy-oil/iea-proposes-largest-ever-oil-release-from-strategic-reserves-275f4e5c https://www.wsj.com/opinion/iran-war-oil-operation-epic-fury-mojtaba-khamenei-0d2edb9c https://www.cnn.com/world/live-news/iran-war-us-israel-trump-03-12-26 https://x.com/sentdefender/status/2031827082934665293 https://polymarket.com/event/balance-of-power-2026-midterms https://polymarket.com/event/march-inflation-us-annual https://www.ft.com/content/7cab4ec7-4712-4137-b602-119a44f771de https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic https://www.axios.com/2026/03/06/pentagon-anthropic-amodei-apology https://www.ft.com/content/97bda2ef-fc06-40b3-a867-f61a711b148b https://x.com/WallStreetMav/status/2032115119879045512 https://x.com/TheChiefNerd/status/2032012809433723158 https://www.nbcnews.com/politics/politics-news/poll-majority-voters-say-risks-ai-outweigh-benefits-rcna262196 https://hai.stanford.edu/news/most-read-the-stanford-hai-stories-that-defined-ai-in-2025 https://x.com/DrTechlash/status/2030734402339365220 https://www.semafor.com/article/12/07/2025/ai-critics-funded-ai-coverage-at-top-newsrooms https://www.cbsnews.com/news/howard-schultz-starbucks-ceo-leaving-seattle-washington-millionaire-tax/ https://www.politico.com/story/2019/02/13/howard-schultz-2020-taxes-1167363 https://x.com/chamath/status/2032135944284094910 https://www.hoover.org/research/net-present-value-billionaire-tax-act-assessment-fiscal-effects-californias-proposed https://www.visualcapitalist.com/mapped-which-u-s-states-gained-the-most-residents-in-2025 https://www.sanders.senate.gov/press-releases/news-sanders-and-khanna-introduce-legislation-to-tax-billionaire-wealth-and-invest-in-working-families #allin #tech #news [Source](https://www.youtube.com/watch?v=HiVej9_fvl8)

[0:00] Jason Calacanis: 你把心掏出来托得恰到好处，要是手臂再抬高 5 度向外伸直，那就成行纳粹礼了，那些‘特朗普账户’可就要变成超级种族主义账户了！ [10:19] David Sacks: 宣布胜利并抽身退出 [19:09] Chamath Palihapitiya: 关于中国、中国、还是中国 [28:59] Kevin Weil (引述): 今天的模型与智能体将是它们未来生命周期中最愚蠢的一天 [32:12] Chamath Palihapitiya: 这是一个仍处于摸索早期的行业，大量涌入的都只是测试预算，绝不能把它过度包装成已经完成的企业级革命。 [35:12] Chamath Palihapitiya: 我自己就在这些基础模型之上搭建应用，我每年付给模型厂商数百万美元，每天消耗的 token 数量都在急剧增加。但真相是，我的经济产出和公司收入根本没有以同样的速度增长！ [47:40] Alex Karp (引述): 如果我们不做，对手就会做，我们就会受制于他们的法治 [56:15] David Sacks: 伊朗花了 47 年才把自己的公共声誉搞砸，而硅谷的这帮天才只用了短短两年就达成了同等‘成就’。 [60:20] Bernie Sanders (引述): AI 面临的监管居然比一家三明治店还要少 [75:04] Gavin Newsom (引述): 单个州无法单独推行这项政策，因为还有其他49个州在展开资本竞争 [77:23] Jason Calacanis: 我们必须砸碎这些阻碍民生的监管枷锁，不要再去中东挑起毫无意义的对外战争，而是集中精力为美国人多建房子。

[44:03] Chamath Palihapitiya: I am building a 1 gigawatt data center in Arizona... When I greenlit that project, I thought it was going to be a four or five billion dollar investment... Then it went to 10. Then it went to 15. Then it went to 20. And now it's upwards of 50 billion dollars for the powered shell, for all the land, for all the permits, then for all the infrastructure, all the people, all of it. [60:20] David Sacks: Future of Life is one of these EA-funded doomer think tanks... Vitalik Buterin donated 600 million dollars of dog coins to them... you can influence a lot of public discourse with that.

[0:00] All right, everybody. Welcome back to [0:01] the number one podcast in the world. [0:03] Free Birds out saving the world, [0:04] creating new potatoes, or I don't know, [0:06] quinoa, maybe some Brussels sprouts. I'm [0:08] not sure what he's working on at this [0:10] point. In his place, his personal [0:12] favorite bestie, he always says that [0:13] when I'm not here, I want Brad Gerstner [0:16] in the seat. Welcome back. We haven't [0:18] seen you on the pod since [0:21] your shout-out at the State of the [0:23] Union. Take us behind the scenes for a [0:25] brief moment here, Brad, of what it's [0:27] like to get a shout-out from POTUS at [0:30] the State Did you know it was coming? [0:32] Did you choreograph this thing? Did you [0:33] Did you choreograph that, or was that [0:35] more spontaneous? [0:36] >> had no idea it was coming. [0:38] And in fact, I I found out after the [0:40] fact that it wasn't in the speech, and [0:42] the president added it to the speech. [0:44] So, I don't even think it was, you know, [0:46] a few days few days before going to [0:48] happen. But, we got an invite to the [0:50] State of the Union, and you know, [0:51] listen, it's an institution. This has [0:53] happened [0:54] every year for 250 years [0:57] in the country. I've never been. I [0:59] thought I did know he was going to talk [1:00] about Trump accounts, so I figured if [1:01] I'm ever going to go, that's the time to [1:04] go. And I have to say, you know, I'm [1:05] just a sucker for democratic [1:07] institutions and democratic traditions. [1:09] It It was an extraordinary night. Set [1:11] aside, you know, the headlines about [1:13] what Democrats did or Republicans did, [1:15] just the whether it's a Democrat [1:16] president or Republican president, that [1:17] this happens every year. You have to go [1:19] report on the State of the Union. So, it [1:21] was a It was a special night. Did dinner [1:23] ahead of time. We're in the chamber. The [1:24] chamber, as you all know, is very small. [1:27] And so, you know, just to your right was [1:30] the first family, and and Jared, and [1:32] Ivanka, and [1:34] And so, you know, we were there as to [1:36] observe, like everybody else. And wow, [1:39] it was It was quite a moment, and [1:42] >> I want to just say, you did a great job, [1:44] because when you sent your heart out [1:46] to all of America, I took it. I took it. [1:49] I took it. I was like, [1:50] >> [laughter] [1:50] >> it out and you cupped it at the right [1:53] angle. [1:53] >> Right. Right. You had just got up a [1:55] little bit extra, and out and out 5 [1:58] degrees. You're being a Nazi. But it [2:00] would have been would have been no [2:00] bueno. Those would be some super racist [2:03] Trump accounts. [2:04] >> [laughter] [2:04] >> Keep your protractor and your ruler out [2:07] when you send your heart out. Okay. One [2:08] final thing on a Jason. [2:10] You know, we're signing up over 100,000 [2:12] kids a day [2:14] to these Trump accounts. We have [2:15] millions of kids who've already claimed [2:17] their account. We have nearly 30 million [2:19] kids in America who are eligible for at [2:21] least $250 if they just go claim their [2:23] account. These things are going to be [2:25] live on July 4th. And what it really [2:26] showed, I think, the country, it [2:28] accelerated after the State of the Union [2:30] because the president, you know, really [2:32] believes this is a way to get everybody [2:34] in Main Street America into the game of [2:36] capitalism and get them all directly [2:38] owning, you know, the great companies in [2:39] America. So, it meant a lot to me in [2:42] that regard that it highlights the [2:44] importance of the program. So, I was [2:45] deeply grateful to the president [2:47] for not only making sure this happens, [2:48] but the shout-out is pretty cool. Good [2:50] for you, bro. I have um an interesting [2:53] idea for you. I'm sure it's come up [2:55] already, but with this whole discussion [2:57] of UBI, somebody said to me, "Oh, you [2:58] know, I really like these Trump accounts [3:00] your friends did, the the Invest [3:02] America, because it's like the start of [3:03] UBI." And I was like, "Well, that's not [3:05] exactly the intention, but I get it." [3:07] And with wealth disparity going on in [3:09] the country that has a lot of people [3:11] concerned, [3:12] what if there was a giving pledge around [3:15] equities? And people could opt into it, [3:17] they don't have to, but if somebody [3:19] like, [3:20] I don't know, Larry and Sergey or [3:21] Zuckerberg said, "I want to pledge 5% of [3:24] my shares to go into kids accounts over [3:27] the next 20 years." [3:29] What an amazing, beautiful thing that [3:31] could be, and it would be incredibly [3:32] material to get whatever it is, a tenth [3:35] of a share, a hundredth of a share, a [3:36] thousandth of a share of whatever [3:38] company. Has that come up yet as an [3:39] idea? I'm sure it's obvious, right? [3:41] It's come up. Stay tuned. But yes, we're [3:44] going to have [3:45] some banger announcements as we head [3:47] toward July 4th. [3:49] All right, let's talk about the war in [3:50] Iran. Obviously, there are much more [3:52] important issues than financial ones, [3:54] life, death, the freedom of the people [3:57] of Iran, but we're uniquely qualified, I [3:59] think, to talk about the economic [4:01] fallout, second-order effects, [4:02] first-order effects, and there has been [4:05] massive volatility over the last five [4:07] trading days. [4:09] Just talking about Brent crude oil, and [4:11] we'll we'll key the discussion off of [4:13] that [4:14] type of oil. It spiked $84 on Friday. [4:17] That was day seven of the war, 119 on [4:19] Monday, day 10. [4:21] Dropped back down to 84. Jumped back up [4:23] to 100 after three commercial ships were [4:25] hit in the strait on Wednesday. [4:28] Those ships, by the way, were not oil [4:29] tankers. They were carrying cargo. They [4:31] were flagged as Thai, Japanese, and [4:33] Marshall Islands. Brent crude currently [4:36] at 99 when we're taping this. It'll be [4:38] at something different by the time you [4:39] listen to the pod, I'm sure. [4:41] But it's quite a spike. And here's a [4:43] second chart that shows you the spikes [4:45] over time. I was old enough to remember [4:47] the oil shock of 1978. We had to like [4:50] get in line at the gas station based on [4:52] your license plate number, and you had [4:54] to wait an hour or two to get gas. Gulf [4:56] War, obviously, it hit $100 in 20 $26. [5:00] 2008, we hit kind of a peak moment, $216 [5:04] in today's dollars. That was the peak [5:06] oil discussion. Demand from China went [5:08] off the charts. When Russia invaded [5:10] Ukraine, we hit 115, which would be 133 [5:13] in today's dollars. So, [5:15] this is not new, but it is significant [5:19] and breaking news today, Iran's new [5:22] supreme leader, Mojtaba, says he's [5:25] keeping the strait closed as a tool to [5:27] pressure the enemy. [5:29] Wall Street Journal on Thursday quoted a [5:31] senior fellow at the Middle East [5:32] Institute saying that reopening the [5:35] strait will require ground troops. [5:37] Polymarket, 27% chance that US forces [5:41] enter Iran by the end of March, and 57% [5:43] by the end of the year. So, So sharps [5:45] over at Polymarket believe we will have [5:47] boots on the ground. Let me stop there. [5:49] Brad, your thoughts on what happens when [5:52] oil hits this kind of number and we have [5:54] this uncertainty of hey, this could [5:56] last, [5:58] you know, two more weeks [5:59] or it could last 6 months, it could last [6:01] a year. [6:02] Nobody seems to know and how it [6:05] resolves, [6:06] we just had a really interesting talk [6:08] with Graham Allison. [6:10] Um, how it resolves is also a major [6:12] unknown. Your thoughts? [6:13] >> Right. [6:14] So first is [6:16] obviously there huge direct costs as oil [6:19] prices go up, right? Oil is a component [6:21] of a lot of consumer, [6:22] you know, and enterprise products. And [6:24] it also hurts consumer confidence, [6:25] enterprise confidence. Goldman Sachs is [6:27] out today with some analysis where they [6:29] updated kind of the economic knock-on [6:31] effects, right? So they raised their PCE [6:34] inflation forecast from 2.1 to 2.9, [6:37] right? For the year. So that's a huge [6:39] jump, right? In terms of their expected, [6:43] you know, PCE inflation. Core PCE, which [6:45] excludes oil, okay, is they forecast it [6:48] up from 2.2 to 2.4. So they're saying [6:51] even if you excluded the direct price of [6:53] oil, the knock-on effects is going to [6:55] cause a little more inflation. They [6:57] lowered their GDP forecast by 30 basis [6:59] points for the year. [7:01] And they also expect higher unemployment [7:03] as a result of this for the year. All of [7:05] that is weighing on the sentiment in the [7:07] market. Remember just a few months ago [7:08] the S&P peaked at 24 times, now we're at [7:11] 21 times. But I think the market may be [7:13] getting it a little bit wrong, right? [7:15] The Trump doctrine, you know, I tweeted [7:17] about this last week. I think the Trump [7:18] doctrine is far more pragmatic than the [7:21] neocon doctrine, right? I think we I [7:24] think Trump has a very limited set of [7:26] goals. He wants to destroy and degrade [7:27] threats to America's national security [7:29] interest. He doesn't want to spread [7:31] democracy. So my suspicion is that these [7:33] impacts are shorter duration, but right [7:36] now the market's having a little bit of [7:38] post-traumatic stress, flashbacks to [7:39] Afghanistan and Iraq and wondering if we [7:42] might be wandering into a quagmire. [7:45] All right. And just he may not [7:48] um in terms of the doctrine, he has said [7:50] he wants to see uh the people rise up [7:52] there. So, [7:53] it might be splitting hairs, but I think [7:55] uh he might not actually be for regime [7:57] change. He says he wants the regime to [7:59] change. [7:59] >> being equal, I don't think he minds if [8:01] the people bring it to themselves. The [8:02] question is whether the US is going to, [8:04] you know, put boots on the ground and [8:05] try to spread Madisonian democracy like [8:08] like the Cheney doctrine was. And I [8:10] think this is very different. [8:11] >> Chamath, uh your take on the economic [8:13] impact and [8:15] you know, and any other uh things you'd [8:17] like to add about the war in Iran. [8:20] I think the most important thing that I [8:22] saw this week was [8:25] I think President Trump was asked about [8:27] the war, and he said the war would be [8:29] over very soon. [8:32] What did the market do? [8:34] The market literally took oil from 120 a [8:36] barrel to 90 a barrel. [8:40] Almost in, you know, in a nanosecond. [8:43] I think that that sort of tells you what [8:46] everybody thinks. [8:47] To the extent that [8:49] the market really didn't believe it, oil [8:52] would not have budged. And if anything, [8:54] it would have faded those comments. [8:56] And you probably would have seen oil [8:58] stay at around 120 or even go slightly [9:01] higher. [9:02] So, the fact that there was this [9:03] reflexive move, I think is a belief by a [9:06] lot of the sharps [9:07] that there is no path to a sustained [9:10] conflict. There's going to be a lot of [9:11] chest bumping from the Iranians, [9:13] obviously, because they need to save [9:15] face, and they will want to set up [9:17] whoever comes next to have the most [9:19] successful chance of governing. So, my [9:21] perspective is that was a trial balloon. [9:24] I think it validated what everybody [9:25] thought, which is that this is going to [9:27] be a short-run thing. I agree with that. [9:31] The downstream impact is, I think, [9:33] correct what Brad said, which is it [9:35] could show up [9:36] in some short-term price spikes. [9:39] But then on March 11th, you saw what [9:41] Chris Wright did, [9:43] which is the president activated a whole [9:45] bunch of member countries in the IEA. [9:48] And I think Chris released 172 million [9:50] barrels. I think there's a coordinated [9:52] release of about 400 million barrels of [9:54] petroleum. [9:56] That's going to dampen the effect of any [9:57] price spike. [9:58] On top of that, I think the estimate is [10:00] there's probably another billion or so [10:02] more barrels that one could [10:05] release out of strategic stockpiles. So, [10:08] I think that both of these two things [10:10] together kind of paint a picture that [10:12] probably the worst is behind us. And I [10:14] think now it's about finding the [10:16] off-ramp. [10:19] Sachs, your thoughts? [10:21] Well, I agree that we should try to find [10:23] the off-ramp. I mean, I agree with what [10:25] Brad and Jamar said about that. [10:28] Look, we've degraded Iranian [10:29] capabilities massively. Their army, [10:32] navy, air force all been destroyed. [10:35] This is a good time to declare victory [10:36] and get out. And that is clearly what [10:38] the markets would like to see. [10:40] You are seeing, however, a faction of of [10:43] people, [10:44] I'd say largely but not exclusively in [10:46] the Republican Party, who want to [10:47] escalate the war. And who are calling [10:50] for things like ground troops or regime [10:54] change or they simply want the pounding [10:57] of Iran to just keep going on and on. I [10:59] saw an op-ed in the Wall Street Journal [11:01] to that effect that we shouldn't try and [11:02] find an off-ramp. We should just keep [11:04] going with this. [11:05] And I just want to lay out, I think, [11:07] some of the risks of what [11:10] an escalatory approach [11:12] could entail. So, first of all, we're [11:13] all seeing that the the Straits of [11:15] Hormuz are closed right now. [11:18] We don't want that to persist longer [11:20] than it has to. But there are actually [11:23] worse outcomes than that. So, if the [11:25] Iranians get hit, if their oil and gas [11:28] infrastructure gets hit, they've already [11:31] said they're going to engage in [11:33] tit-for-tat retaliation against the Gulf [11:34] states. And we saw there was recently um [11:37] the Iranians blew up this giant oil [11:39] depot in Oman. [11:41] You saw some of those images. They could [11:43] continue to target the oil and gas [11:45] infrastructure across the Gulf states. [11:48] And if that happens, it won't really [11:50] matter if the straits get reopened [11:52] because you won't be able to restart oil [11:53] and gas production in the Middle East. [11:55] So that would be I think a much worse [11:57] outcome that could result from [12:00] escalation. [12:01] Furthermore, there's an even worse, I [12:03] think, scenario [12:04] from there, which is the region is very [12:06] dependent on desalination plants. [12:09] I think something like 70% of Riyadh [12:11] gets their water from desalination. I [12:13] think it's something like 100 million [12:14] people on the Arabian Peninsula that get [12:17] their water from desal. I mean, it's [12:19] basically a desert, right? [12:21] And those desal plants are soft targets. [12:23] You already saw there was I think there [12:25] was one desal plant in Iran that got [12:27] hit. And then it caused Iran again [12:28] tit-for-tat to hit a desal plant. I [12:31] think it was in Kuwait. I could be off [12:33] about that. [12:34] But in any event, if you see that type [12:36] of destruction continue, you could [12:38] literally render the Gulf almost [12:41] uninhabitable. I mean, you're just not [12:42] going to have enough water for 100 [12:44] million people. And human beings just [12:46] cannot survive very long without water. [12:48] So that would be a truly catastrophic [12:51] scenario. And we're talking about [12:54] destroying the Gulf states economically [12:56] and then also from a humanitarian [12:57] perspective. So I think we have to take [12:59] things like this into account when you [13:01] hear people preaching for or advocating [13:03] for escalation. [13:05] You also have to I think consider the [13:06] impacts on on Israel. I mean, it's hard [13:09] to know exactly how much damage Israel [13:11] is taking right now. There's [13:13] a social media blackout. But what you're [13:15] starting to hear trickle out is that [13:17] Israel's getting hit harder than they've [13:19] ever been hit before in their history. [13:20] And we're only 2 weeks into this. If [13:22] this war continues for weeks or months, [13:26] >> then Israel could just be destroyed or [13:29] very large parts of it. Now, I think [13:31] Israel is a harder target than the Gulf [13:33] states. Their infrastructure is more [13:35] hardened. Also, they're further away. [13:37] The Gulf states are vulnerable to drones [13:39] and short-range missiles, whereas Israel [13:42] is mainly vulnerable to long-range [13:43] missiles. Nonetheless, at some point [13:46] their air defenses could become [13:48] exhausted if it hasn't happened already, [13:51] and Israel could get seriously [13:52] destroyed. Then you have to worry about [13:53] Israel [13:55] escalating the war by contemplating [13:56] using a nuclear weapon, [13:58] which would truly be [14:00] catastrophic. So, there's a lot of [14:02] scenarios here, a lot of really [14:05] frightening scenarios about where [14:07] escalation could lead. And even though [14:10] the United States is a much more [14:11] powerful country than Iran, they [14:13] essentially have a dead man's switch [14:16] over the economic fate of the Gulf [14:18] states and even potentially beyond that, [14:21] you know, the habitability of some of [14:23] these these countries. [14:24] So, I do tend to think that this is a [14:26] good time to declare victory. I think [14:29] Brad, you're right that the president [14:31] has never said that democracy promotion [14:34] is one of his objectives. Yes, Jake Hal, [14:35] obviously everyone would welcome if the [14:38] people rose up and chose a new regime, [14:40] but that's not something that we've said [14:42] we have to accomplish. [14:45] And this would be a really good time to [14:46] take stock of where we are and try, I [14:49] think, to seek an off-ramp. And look, if [14:51] escalation doesn't lead anywhere good, [14:54] then you have to think about well, how [14:55] do you de-escalate? And [14:57] de-escalation, I think, involves [14:59] reaching some sort of ceasefire [15:01] agreement or some sort of negotiated [15:03] settlement with Iran. And we can get [15:05] into more of what that looks like, but I [15:07] think that's the big picture is that if [15:09] escalation could lead in all these [15:11] horrifying directions, then I think [15:13] that's not the right approach. You have [15:14] to look at de-escalation. [15:16] Jake Hal, what what where are you on [15:17] this? Complicated. I have my personal [15:19] feelings on regime change, and since we [15:21] don't have the information uh the Mossad [15:24] and the CIA and Trump has, I do think [15:27] Trump would only do this if he had a [15:29] very high probability of success and an [15:32] off-ramp. However, it's not looking good [15:35] with the off-ramp right now [15:37] and it could be quite chaotic. I think [15:39] if the neocons get their way and the [15:41] people on Polymarket are correct, the [15:44] sharps who say 57% chance we'll have [15:46] boots on the ground, I think this is [15:48] kind of the end of Trump's second term [15:50] and if you were to put together [15:53] the series of mistakes that he's made um [15:57] and the administration's made, they're [15:59] really at the heart of why people voted [16:02] for him. You take starting a war like [16:04] this specifically with Iran, that's what [16:07] we were told was the reason to vote for [16:10] President Trump. He was not going to [16:12] take us down this path. He was not going [16:14] to [16:15] risk World War III. He was not going to [16:17] risk a nuclear possibility as as [16:20] Sachs correctly points out. And now you [16:24] have all the MAGA supporters from, you [16:26] know, Tucker to MTG to Rogan, Matt [16:28] Walsh, Megyn Kelly, they are all up in [16:30] arms about this is the end of MAGA and [16:33] this is, you know, a massive betrayal. [16:35] There's the one B betrayal where Trump [16:39] wouldn't release the Epstein files. [16:40] We'll put that aside because I don't [16:41] think that's as important as starting [16:43] World War III. [16:44] Um and then obviously the insane [16:47] unnecessary cruelty we talked about many [16:49] times on this podcast of ICE agents [16:50] which he has [16:52] corrected by getting rid of Kristi Noem. [16:54] So, you you start putting these things [16:55] together, [16:56] if this continues for another 6 months, [17:00] it's basically going to result in [17:03] the [17:04] Democrats [17:05] doing a clean sweep in the midterms. [17:08] Here's the chart that I think, you know, [17:10] the Republicans really need to look at [17:12] at how misguided, [17:15] you know, [17:16] uh this all is. This is the chart that [17:18] should be absolutely terrifying. Nobody [17:21] wanted this war or very few people [17:23] wanted it besides the neocons and the [17:25] people of Iran probably and the [17:27] Israelis. [17:28] But the chances of the Democrats [17:29] sweeping now is up to 45%. This just [17:32] happened. [17:33] The Democrats are going to sweep. [17:36] They're then they're going to win in [17:37] 2028 and the entire agenda of Maga and [17:41] Trump's 2.0 will be gone. And then you [17:43] look at just absolutely ignoring the [17:46] working man. [17:47] Inflation going up above 3% as you [17:50] pointed out is a likelihood. Brad, [17:52] unemployment ticking up. Still very low, [17:54] but it's ticked up 10% worth keeping an [17:56] eye on. These foreign affairs things are [17:58] the least [18:00] important to the American people. It's [18:02] very very low on the list of priorities. [18:05] And people are looking at Trump and what [18:08] they believe is the enriching of his [18:09] family and all these business deals and [18:12] then they [18:12] >> kitchen sinking it. [18:13] >> It's literally what I was thinking. [18:15] >> You're bringing everything. I I 1 2 3 4. [18:18] Number one, doing the war that everybody [18:20] said he should not do include and that [18:23] that was why we should vote for him. [18:24] Number two, the Epstein files. Number [18:25] three, the ICE cruelty. And number four, [18:27] not working for the American working man [18:31] who doesn't own equities. Those are [18:32] four. 1 2 3 4. It's not a kitchen sink. [18:35] This is not my personal feelings on [18:36] this. This is my assessment of the [18:38] situation. If he doesn't find an [18:39] off-ramp quickly, they're going to lose [18:43] both houses in the midterms. That's I [18:44] think [18:45] the thing Trump needs to really consider [18:47] and I think he will consider it. I think [18:49] he's going to find an off-ramp. [18:50] >> Right. That is the topic. The topic was [18:52] are we going to find an off-ramp or not [18:53] find an off-ramp and in Iran and I think [18:56] Sachs made the argument that there's [18:58] danger that neocons and others are [19:00] arguing that we expand, put boots on the [19:02] ground. You're saying if he doesn't, [19:04] it'll be a disaster. Chamath and I both [19:06] say he will, right? And so [19:09] No, wait. Wait. Will what? He will find [19:12] an off-ramp over, you know, in the [19:13] nearer term because the Trump doctrine [19:16] is not the neocon doctrine. As much as [19:19] people want to talk about Iran, Iran, [19:21] Iran, I think as I explained last week, [19:23] I think this is about China, China, [19:25] China. [19:26] And you have to remember at the end of [19:28] this month he has a pivotal 3-days [19:31] with Xi Jinping in China. This is going [19:34] to be an absolutely historic [19:37] convening [19:38] of the two superpowers [19:42] that run the world. One, [19:44] which is us, we are the established, [19:46] and one, which is China, who wants to be [19:49] re-ascendant. [19:50] And I would bet dollars to donuts [19:54] that there is going to be an enormous [19:55] incentive [19:57] for Xi to negotiate a grand bargain in [20:00] those 3-days [20:02] and do something historic for himself. [20:06] And I think that the president will use [20:08] that if he thinks that it creates [20:09] leverage. I think is a great insight. [20:11] How does the Strait of Hormuz open? If [20:13] this war is dragging on and Israel, [20:16] which seems to be the driving force in [20:19] this, if Israel keeps it up with Iran, [20:23] how do we ever get the strait open [20:24] again? I think the off-ramp is that the [20:26] United States, [20:28] you know, declares victory, does what [20:29] Sax says, and says, "Listen, we degraded [20:32] and we destroyed. That's what we came [20:33] here to do. We did not come here for [20:35] some experiment in in democracy. We wish [20:38] the best to the Iranian people to do the [20:40] things they need to do." And if Iran [20:42] does not back down, if after that [20:44] declaration Iran continues to destroy [20:46] cargo containers moving through the [20:48] narrow straits, [20:50] I think you're going to see Iran's [20:52] neighbors and Israel and others get very [20:55] involved as it pertains to Iran because [20:57] it's in their interest. Listen, the [20:59] United States produces 20 million [21:01] barrels of oil a day and we consume 20 [21:03] million barrels a day. This is a This is [21:05] a modest problem for the United States. [21:07] This is a massive problem for China. [21:09] This is a massive problem for Asia. This [21:11] is a massive problem for all of our [21:13] friends in the Gulf who are trying to [21:14] dodge Iranian missiles right now. So, [21:16] there are a lot of people in the world [21:18] who will take up arms to deal with the [21:19] Iranians if the United States isn't [21:22] there because we can take care of [21:23] ourselves. Your position, Brad, just to [21:25] confirm it is [21:27] we are going to leave the war in the [21:28] next 30 days and then if the straits are [21:30] not open, [21:32] then China, India, and all the Gulf [21:35] uh countries that are impacted by it, [21:37] they will protect it. They will fight [21:39] Iran. I think they'll put a lot of [21:40] pressure on Iran not to continue firing [21:43] missiles at their ships, right? At the [21:45] end of the day, this is not just an [21:47] American problem. [21:48] Right? And let's be clear, we're always [21:51] involved in this part of the world. The [21:53] only question is are we going to have an [21:54] active armada that's engaged in active [21:57] military activities against Iran? And [21:59] what I'm suggesting again, and listen, [22:01] anytime you try to clean up a mess like [22:03] this, there is risk. [22:05] This is not a risk-free, you know, [22:08] initiative by the United States, nor was [22:09] Venezuela. But let me [22:12] steel-man the alternative. Doing nothing [22:15] and allowing Iran to procure, you know, [22:17] the ingredients for a nuclear missile [22:19] when they are set on the destruction of [22:21] the United States and US interests, [22:22] doing nothing in Venezuela while the [22:25] Monroe Doctrine is totally wrecked and [22:26] we let our adversaries take up, you [22:28] know, positions in in South America, [22:31] those also have risks, right? Those [22:33] carry a lot of risks. And so, we're [22:36] we're weighing these two risks. Again, [22:38] for me, I don't like the fact that we're [22:40] engaged in military activities here, but [22:42] I will tell you I am very much on the [22:44] side that if we're going to go protect [22:46] American national security interests, [22:48] you go in, you do the the degrading of [22:51] their capability, and you get out. And I [22:53] think, you know, that's what I hear out [22:55] of the president. [22:56] Chamath, you had a follow-up. [22:58] All roads lead to China. I think that [23:00] there you're going to see [23:02] Xi offer up a grand bargain. And I think [23:04] it's up to the president to decide [23:05] whether he wants to take it and see what [23:06] he wants to add to it to get something [23:08] done. But I just don't see them meeting [23:10] and coming out with nothing. I think I [23:12] see them going in and coming out with [23:14] something that's historic. [23:17] And I think that all of this, Venezuela [23:19] and Iran together, is all about China. [23:22] >> Let me just say one thing as to that, [23:23] Chamath, I think because I think the [23:24] point is absolutely spot on. Right? [23:28] Probably the single greatest takeaway [23:30] for us from an investment perspective at [23:31] the start of this war [23:33] was that the Chinese, [23:35] right, didn't take up arms on behalf of [23:38] Iran, [23:39] aren't defending Iran, [23:41] and they didn't cancel the summit with [23:43] the president. [23:45] Right? The very fact that the [23:46] >> Because they need him. They need the [23:49] oil. 20% of their entire domestic [23:52] consumption is oil from Venezuela and [23:55] Iran. 20%, but it's not 20% because it's [23:58] literally 100% of anything that's [24:01] feedstock, anything that's transport, [24:03] cars, buses, planes. [24:06] They are in an enormous world of hurt. [24:08] Now, they have a strategic petroleum [24:09] reserve as well, [24:11] and it's quite robust, but it's not [24:14] robust enough to sustain five or six [24:16] months of this. It's not that robust. [24:19] So, at the end of the day, who is going [24:21] to be hurting the most? It is China. [24:24] And so, if you play this game theory [24:25] out, the reason he kept it is because [24:27] now he needs the summit even more. Could [24:29] you imagine if the president canceled? [24:32] That would be a disaster for the [24:33] Chinese. [24:35] So, the fact that it's still on the [24:36] books, [24:38] if I was Xi, I'd be like, "How do I [24:40] negotiate and help find the off-ramp? [24:42] How do I end up fixing this faster?" All [24:46] right. Remember, you have 25% [24:47] unemployment of young men inside of [24:50] China. [24:51] 25% today. What do you think it goes to [24:54] in five months with no oil? [24:57] It's in the It's [24:58] >> That's the unemployment rate you should [24:59] be focused on, Jason. [25:02] Oh, the China issue is a separate one. [25:04] Just to [25:05] >> separate. No, no. That was separate from [25:07] my point is my point. I I was bringing [25:09] up a different point. [25:10] >> kitchen sink didn't include the Chinese. [25:12] I get that. I'm just adding to your [25:14] kitchen sink. [25:14] >> I didn't have a kitchen sink. I have [25:15] four very salient points. All right, [25:17] Sachs, I'll give you the final word [25:18] here. [25:20] Well, look, that was a bit of a a [25:21] broadside, Jay Cow, where you kind of [25:23] did kitchen sink it. But look, here [25:25] here's the part I'll agree with you [25:26] about, which is it doesn't take a [25:28] political genius to understand that long [25:31] wars are unpopular. [25:33] It will hurt the Republicans in the [25:36] midterms of '28 if this does turn into a [25:38] long war. Fortunately, I think the [25:40] president understands that. His [25:41] political instincts are impeccable, and [25:43] he's always favored [25:46] short, decisive, swift actions, military [25:49] actions, whether it was midnight hammer, [25:50] whether it was the Maduro raid. I think [25:53] that is his inclination and preference. [25:55] And I think we are pretty much at or [25:58] close to a point where the president's [26:00] going to have to decide on next steps. I [26:02] think he's indicated that we have [26:05] completed our objectives, and I think [26:07] it's just important that we don't let [26:09] this neocon wing of the party try to [26:12] expand the objectives or aims of the war [26:16] because frankly they've always been [26:17] wrong about everything. I mean, these [26:19] are people who never wanted to get out [26:21] of Iraq and Afghanistan, would still be [26:23] there after 20 years if they had their [26:25] choice. [26:26] So, I think it's just important to not [26:28] listen to those people. And look, it's [26:29] not just one op-ed in The Wall Street [26:32] Journal. The Wall Street Journal is kind [26:34] of the tip of the spear representing [26:35] that whole neocon establishment. And I [26:38] think it's just important that this is [26:40] the time to frankly ignore those voices [26:42] and let the president do what I think [26:44] his political instincts are telling him [26:45] to do, which is to wrap this thing up. [26:48] I'm in strong agreement, and it is my [26:50] hope, too, that he wraps it up quickly, [26:52] and that we don't have any more loss of [26:54] life. All right, we'll keep discussing [26:57] this ongoing breaking news story in the [27:00] coming weeks, but back to our zone of [27:02] excellence, AI and tech. OpenAI and [27:06] Anthropic are scaling revenue and costs [27:08] faster than we've ever seen in the [27:10] history of, well, business, the world. [27:13] Revenue at these two companies growing [27:15] gosh, like unprecedented levels. Here [27:18] are the reports, and I believe your [27:20] investors in both these companies, Brad. [27:22] Anthropic hit a $14 billion run rate [27:25] last month, February. [27:27] That means they have grown revenue from [27:29] $1 billion [27:30] uh to $14 billion in 14 months. Yeah, [27:33] 12x year over year. [27:36] They're valued at a meager $380 billion [27:39] uh last month. This feels like a bargain [27:41] given the growth. OpenAI ended 2025 at [27:44] $20 billion annualized run rate. And [27:47] they've grown revenue from $2 billion to [27:48] $20 billion in 24 months. They're valued [27:51] at $840 billion last month. Uh and man, [27:54] it looks like [27:56] Sam Altman has Dario in the rearview [27:59] mirror. [28:00] He could get lapped any moment. Lots of [28:03] debate [laughter] Where'd you find this? [28:04] What the hell is this? [28:05] >> That one, I made that. This is Dario [28:07] closing in. [28:09] What is that? A velociraptor? What is [28:11] it? A T-Rex? [28:12] >> famous scene from Jurassic Park. Oh my [28:15] god. But I mean, I don't think anybody [28:16] expected Dario to be coming around the [28:18] bend this fast, but he's right behind [28:20] apparently, and they're winning [28:22] obviously the business-to-business side [28:24] of the business. [28:25] The J curve on these companies is [28:27] insane. 250, 500 billion, who knows [28:29] what's get gets invested [28:32] before these companies reach [28:33] profitability, Brad, but you're invested [28:36] in these two companies. Yeah. Unless you [28:38] sold when Sam Altman told you he would [28:40] buy his shares back on the famous BG2 [28:42] episode. I don't think you sold it. I [28:44] bought [28:44] >> a lot more since then, Jason. I bought a [28:46] lot more since then. That's important [28:48] information for us to have. Quick [28:50] question for you. Number one, what's a [28:52] better buy here? Yeah. Uh Anthropic at [28:54] 380, OpenAI at 840. [28:56] Uh and then I think people want to know [28:58] if these companies are going to go [28:59] public, what you if you think they [29:01] should go public, what is the chance of [29:03] that happening? Take these questions [29:04] however you like. [29:06] Well, I mean listen, you know, love your [29:08] children equally. [29:09] They're both incredible companies. [29:11] Anthropic unquestionably has a lot of [29:13] financial momentum, you know, and OpenAI [29:16] is seeing a lot of momentum themselves, [29:17] right? But the single most important [29:18] question this year, right? Was would AI [29:21] revenue show up? [29:23] And just 60 days ago, 90 days ago, there [29:26] was tremendous skepticism. No way all of [29:28] these infrastructure investments were [29:30] going to pay off. There's no incremental [29:32] revenue coming out of AI, including many [29:35] of our friends. But in February we had [29:38] in January and February we really had [29:40] kind of a nuclear moment, right? The [29:42] splitting of the atom moment. I mean, we [29:44] had a six billion-dollar month out of [29:46] Anthropic in February, right? What [29:49] widely reported, okay? Let that set in [29:52] for a second, right? Six billion dollars [29:55] in a month and it was only a 28-day [29:57] month, okay? That's more revenue than [29:59] the annual revenue of Databricks and [30:01] Snowflake that are two of the greatest [30:03] software companies of all time after 12 [30:06] years, right? They could do in the first [30:08] four or five months of this year the [30:10] total revenue of SpaceX this year. What [30:12] is driving that? Just explain to the [30:14] audience what's driving it. Is it token [30:16] use? Is it cloud subscriptions? We [30:18] crossed a threshold [30:20] with Opus 4.6, [30:22] right? And we saw it again with ChatGPT [30:24] 5.4, where the models and the agents on [30:27] top of them, whether it's Claude Code, [30:29] Codex, ChatGPT, they're no longer [30:32] competing with IT budgets. They're now [30:34] augmenting labor. They're competing with [30:36] labor budgets. You could not possibly [30:38] have a six billion-dollar month. It is [30:40] impossible to do that by displacing IT [30:43] budgets. Millions of other companies [30:45] across America say, "Oh my god, let's [30:47] spin up these agents and have them do [30:49] things for us and we're willing to pay [30:50] for it because the product of that [30:53] effort is worth the money to us." And [30:55] the revenue and the usage momentum, I [30:57] will tell you in the month of March [30:59] continues and it only accelerates from [31:01] here. As Kevin Weil has said, "The [31:03] models and the agents are the dumbest [31:05] today they will ever be." Right? We're [31:08] in the early innings of compute compute [31:10] and algorithmic uh capability. And so, [31:14] you know, like that to me is the [31:15] observation of this moment. Should they [31:16] go public? I said yes, they should go [31:18] public for several reasons. There's tons [31:20] of institutional demand. They need cheap [31:23] access to money to continue to build out [31:25] uh the compute they need to support. [31:27] They They There is more compute [31:28] constraint in these businesses this very [31:30] day than they've had any time in the [31:32] last 3 years. So, they need access to [31:34] the capital. And then finally, I think [31:36] you have to have the retail investor in [31:38] the game. These are two of the most [31:40] important companies in the history of [31:41] capitalism, in the history of America. [31:43] It's destabilizing not to have them [31:45] public. You know, Jensen said last week [31:48] that he expected the 40 billion he [31:50] recently invested in these two companies [31:52] would be his last money in because they [31:54] were both going to go go public. He [31:56] thought they He said they would both go [31:57] public this year. I think that, you [31:59] know, they're they're preparing and [32:01] heading down that path. You know, and [32:03] and listen, I want to get some of these [32:05] shares in the accounts of all these kids [32:06] that we're opening up because they're [32:08] really, really important companies to [32:10] the future of the American economy. [32:12] Chamath, you had some [32:14] insight into the quality, durability of [32:17] this revenue. There's not a single [32:19] good example that we can find of [32:22] sustained [32:23] positive [32:25] margin expansion and impact of [32:28] AI inside of a true corporate enterprise [32:31] that is not right now a small test. [32:34] There's not. [32:35] So, where does 6 billion come from? [32:37] Because everybody has to show up to [32:40] their board and have an AI checkbox. And [32:43] everybody is thousands and thousands of [32:45] companies. And when you have tens of [32:48] thousands of companies as customers [32:49] paying $200 plus a month, it's not that [32:52] hard to show up with that kind of [32:54] revenue. The real question is the [32:55] following. [32:57] If you take you use the Databricks and [32:59] Snowflake example. [33:01] If you look at the companies that use [33:04] that software, [33:06] those companies [33:08] generate enormous revenues and enormous [33:11] margins. And these products are in [33:14] critical [33:15] production workflows that underlie those [33:18] revenues and profits. [33:20] That is just not true with AI today. We [33:22] have all kinds of claims, [33:24] but we are still experimenting. [33:27] Why are we experimenting? [33:29] Because we know it's important, [33:31] but we don't yet really know what to do. [33:33] You can't just slot this in to a [33:35] critical workflow in healthcare, and all [33:37] of a sudden show up where if you make a [33:39] misdiagnosis or if you make a a [33:42] mischaracterization of a procedure, [33:45] you can get fined and go to jail. The [33:46] companies that are in healthcare don't [33:48] do that. [33:49] If you're in financial services [33:51] and you make a mistake about somebody's [33:52] portfolio or you make a misallocation [33:55] and you point to a model, you will get [33:57] sued and you will be in trouble. [33:59] None of these things have transitioned [34:01] from it's interesting, it's experimental [34:04] to it's the core critical operational [34:07] workflow. [34:07] >> Mhm. [34:08] Just the interesting. There will be a [34:10] transition in revenue quality [34:12] when that happens. A great example of [34:14] this is Amazon. [34:16] Why does Amazon issue an edict that says [34:19] you cannot use this stuff [34:21] inside of AWS unless a human now reviews [34:24] and approves it? Because what happened? [34:26] They had three or four sev one faults [34:29] from a bunch of code that was written by [34:31] agents that brought down AWS. Now, look, [34:33] I've told you, I love AWS for one reason [34:36] because it's hyper-reliable. I hate AWS [34:39] for the same reason that [34:40] hyper-reliability comes at enormous [34:41] cost. I pay it. But, I pay it to never [34:44] have a sev-1. [34:46] The reason they have 12 nines of [34:47] accuracy is because it's humans and [34:50] deterministic code that never fails. It [34:52] doesn't mean that two companies can't [34:54] get to 20, 30, 40 billion of revenue. [34:56] What it means is we have to be honest. [34:58] This is an industry that's early. [35:01] We are all figuring it out. There's a [35:03] lot of test budgets that are going at [35:05] it. It will slowly and methodically [35:08] emerge into production, but let's not [35:11] oversell what this moment is. Okay, [35:13] Brad, I want to give you a [35:14] well-constructed question here to [35:15] respond, and then Sachs, we'll go to you [35:17] if you have some input. Of the 20 [35:19] billion, how much of it do you think is [35:22] experimental? What percentage is [35:24] experimental versus production based on [35:27] >> out strip out the [35:29] the consumer spending for cuz like you [35:31] know, that's half of it. Okay, so let's [35:34] I'll I'll put aside the consumer Sure, [35:35] great idea. We'll put the consumer [35:36] subscriptions aside. They're obviously [35:38] getting value, or they wouldn't be [35:39] subscribing, but it's like Netflix or [35:41] >> way, for that, where it can be extremely [35:44] faulty, and there's no SLA that you're [35:46] giving. It's a phenome- These are [35:48] phenomenal products. [35:50] >> Yes, for 20 bucks a month, well worth [35:51] it, and consumers have decided it's [35:53] worth it. So, I think we're in agreement [35:54] there. [35:55] >> And also, for the individual engineer, [35:57] of which I suspect there's a few [35:59] million, [36:00] who get to pay $200 a month and have [36:02] their company subsidize it. The company [36:04] knows that these costs are being [36:06] incurred, but there is no tick and tying [36:09] at the end of it where then you review [36:10] the code in a different way because you [36:12] you you're worried that there's [36:13] hallucinations, as Amazon just [36:14] demonstrated. Yeah, there's a story in [36:16] the There's a story in the FT about [36:18] Amazon having some blast radius from [36:20] some computer-generated AI, and they're [36:23] putting controls in place. We'll put [36:24] that in the show notes. Brad, let's get [36:26] to the specific question I asked. Of the [36:28] tens of billions of dollars in revenue [36:30] between the two companies that's not [36:31] consumer, [36:33] what percentage do you think is [36:35] production quality versus experimental [36:37] to Chamath's point? Yeah, I mean, [36:39] listen, and I've been, you know, I [36:40] coined the phrase experimental run rate [36:42] revenue, right? Versus annual recurring [36:45] revenue, right? Like I think it's I [36:47] think Chamath's point is a really [36:48] important one. As an investor, I have to [36:50] discern what's repeating, right? What's [36:52] recurring and what's not. What I would [36:54] suggest is, of course, there's a lot of [36:56] experimentation because these things [36:58] haven't been around that long. But I [37:00] suspect, right, that Palantir, [37:04] the US government, [37:05] the US military, Nvidia, and a lot of [37:09] other major enterprises would argue [37:12] they've gone full production. In fact, [37:14] it's existential to the wartime effort [37:16] going on in Iran right now. That doesn't [37:18] sound to me like experimental as much as [37:21] it sounds like production capability. [37:23] And I will tell you what will prove this [37:25] one way or the other is in the month of [37:27] March, do revenues continue and go up? [37:31] Well, that's not true. The [37:32] experimentation can go on forever. [37:34] That's not true. That's not true. Okay, [37:35] well, if the experimentation goes on [37:37] forever, that sounds like recurring to [37:38] me. We scratch the surface of of the [37:40] number of companies that even know how [37:42] to adopt AI. So these numbers will go to [37:44] the stratosphere. I'm not debating that. [37:46] >> Yeah. Okay. Look, I'm on the same side [37:48] of the bet as you are. I want these [37:49] numbers to keep going to the moon. I'm [37:51] just being much more circumspect and [37:53] honest with myself to say, I see it on [37:55] the ground. I sit on top of these [37:57] models. I am paying these models [37:59] millions of dollars a year. I am. Yep. [38:02] And what I'm telling you is my revenues [38:03] don't go up faster than their revenues. [38:05] I'm consuming more tokens every single [38:08] day. [38:09] Do I get more economic output? I am not. [38:11] And I would say that my team is at the [38:13] leading edge, and so I suspect a Fortune [38:16] 1000 company is steps behind my team. [38:19] And if I am spending triple every 3 [38:21] months and not seeing my revenues [38:24] tripling. [38:25] I suspect these other companies are in a [38:27] similar situation. I'd finalize it, but [38:29] ask it on Friday when you're with [38:30] Michael Dell because I've had this [38:32] conversation recently with Michael Dell. [38:34] And Michael said a year ago, companies [38:35] weren't seeing ROI. Today, they're [38:37] seeing very big ROI in their AI [38:39] investments, but I think that's the [38:40] question on the table. But What [38:42] companies? Which companies? Yeah, let's [38:44] let's talk about Goldman Sachs Sachs. Of [38:46] course I'm probably seeing an ROI. [38:48] Sachs, we're agreeing. The company This [38:50] is experimental in large part, and this [38:53] is a new tool. So, by definition, you [38:56] have to experiment before you put it in [38:58] production. What's your take on this [39:00] grand debate? How much of this revenue [39:03] is experimental versus real? Well, look, [39:05] when you're talking about enterprise [39:06] revenue, what you're really talking [39:08] about is coding assistance. That's been [39:11] the breakout use case. It's really the [39:13] first big breakout use case on the [39:15] enterprise side. The consumer side's [39:16] been more of like, you know, research [39:18] and writing, that kind of stuff. The [39:20] chatbots. [39:21] But enterprise is all about about coding [39:23] assistance. My sense is that the demand [39:26] for code is very scalable. Software [39:27] engineers has always been an area of the [39:29] economy where companies have never been [39:31] able to hire enough. Even Silicon [39:33] Valley, which is the most attractive [39:34] place for software engineers to work, [39:36] we've never been able to recruit and [39:39] attract enough of them. The rate [39:41] limiting factor on the progress of every [39:43] startup I've ever invested in is not [39:46] having enough engineers to code up the [39:47] product roadmap. [39:49] And then you look at the rest of the [39:50] economy, Fortune 500, and so forth and [39:52] so on. They have hardly been able to [39:54] recruit software engineers at all [39:56] because they've all gone to Silicon [39:58] Valley. So, I think you're dealing with [39:59] a a part of the economy where there's [40:01] always been a massive supply shortage. [40:04] And I don't know what the natural limit [40:05] on that is, but my sense is that there's [40:08] a tremendous latent demand for the [40:11] ability to generate code [40:14] in large quantities, create new [40:16] products, you know, as the cost of code [40:19] goes down, as the coding assistance get [40:21] better, you can code up new types of [40:22] products. And then of course it's going [40:24] to lead to agents, which is another [40:27] way of basically using the code that [40:29] gets generated. So my sense is that this [40:31] could be very scalable. I don't know [40:33] where it it taps out exactly. [40:35] Where I think Chamath is right is that I [40:38] think there is a change management [40:40] aspect to this [40:41] in Fortune 500 companies for example, [40:44] and they haven't really wrapped their [40:45] heads around how exactly they're going [40:47] to use this. There was a McKinsey study [40:49] that showed that a lot of these pilot [40:51] projects in Fortune 500 companies were [40:53] experimental, and a lot of them were [40:54] proving not to be successful. So I do [40:56] think like as you go beyond coding [41:00] into, you know, company transformation [41:02] things like that, it becomes a little [41:04] bit more speculative. That's not to say [41:06] it won't happen. I think it will happen. [41:08] I'm actually I'm bullish, but I do think [41:11] that we're still waiting to see what the [41:13] breakout use cases beyond coding will [41:15] be. Probably agents will be the next big [41:16] one. But I think Brad's right that [41:18] that's big enough to see [41:21] you know, this scale for a while. [41:22] Because, you know, there's a thing about [41:24] the thing about code is you're paying [41:25] for code on a metered basis right now. [41:27] You're paying per token, which is kind [41:29] of an amazing deal for companies, right? [41:32] Because before they had to go through [41:33] this recruiting process to find [41:36] engineers, source them, [41:38] vet them, you know, keep them happy, [41:40] give them all the perks, the KIND bars, [41:43] all you know, uh and so to be able to [41:45] buy code on a metered basis as the cost [41:47] per token keeps going down, [41:50] it's kind of an amazing deal. [41:52] >> It was just metered on people. Now it's [41:53] Now to your point it's it's metered it's [41:55] metered in a different way, but it's [41:56] still metered. Yeah. And let me just, [41:59] you know, you used this term [42:01] labor displacement, Brad. That's like [42:02] the one part where I might disagree with [42:04] you is because there was such a shortage [42:08] >> Yeah. of software engineers. I think [42:10] when people hear the word labor term [42:12] labor displacement they might start to [42:14] think that 6 billion of incremental [42:17] revenue means 6 billion of layoffs and I [42:20] don't think it does and and the way to [42:22] to thread that needle is the fact that [42:24] we were artificially constrained in the [42:27] number of software engineers and how [42:28] they could be used and how rapidly they [42:31] could be acquired and all that kind of [42:32] stuff. So to be able to now turn that on [42:35] like electricity and that's kind of what [42:36] we're talking about exactly is just such [42:38] a huge [42:40] game changer and unlock for the whole [42:42] economy. Yes. [42:44] >> And that's what I think is really [42:44] exciting about it. [42:45] >> It's augmenting it's augmenting human [42:47] labor, right? We're not at a place yet [42:50] where it's displacing and this is the [42:51] definition of productivity gains. I'm [42:53] going to make just uh two quick points [42:55] here. [42:56] The place to look for this actually [42:58] moving from experimental into production [43:00] is not not at big companies. Big [43:03] companies are actively resisting this [43:04] management and big companies will resist [43:06] it because it means lowering head count [43:07] and it means the person who implements [43:09] it might actually implement themselves [43:11] out of a job. So that is the natural [43:13] resistance you'll see in big companies. [43:14] That's not where to look for adoption of [43:16] new tech. [43:16] >> That's not why it's happening. That's [43:18] not why it's basically happening. [43:19] >> my point and then you can you can [43:20] counter it. [43:21] Startups are the place to look at this [43:24] and that's where I am on the ground and [43:26] what I'm seeing there is that startups [43:28] are using this in production for their [43:31] legal work, for their marketing, for [43:33] SDRs, [43:35] uh for their accounting, reviewing legal [43:37] documents. This is all [43:40] uh work that would normally they would [43:41] hire consultants for, outsource, uh or [43:45] make hires for. And what I'm seeing on [43:46] the ground is it's production ready in [43:48] startups who are using it in those [43:51] categories. HR as well, accounting, [43:53] marketing, all of that. All that [43:55] blocking and tackling, all those chores [43:57] are being done currently with these LLMs [44:00] and they're doing it in production and [44:01] they're doing it at scale. Just a quick [44:03] second point here. Here's the J curve [44:05] and this is the question I think we'll [44:07] get to in our next segment. When does [44:08] this become a profitable business? If [44:11] And And you asked this of Sam in that [44:13] famous clip on the BG2 podcast, [44:15] RIP BG2. [44:17] Here you go, the LLM industry J curve. I [44:19] just asked Claude to make this for [44:21] [laughter] me. [44:22] If you have 500 billion, I think you [44:24] would agree it's probably going to be [44:25] around that number Brad invested in all [44:27] of this. And then [44:28] >> More. A lot more. Okay, [44:30] so 5 billion is an underestimate here. [44:32] And then when do we actually see these [44:34] large language model companies hit [44:37] profitability in a in a calendar year? [44:39] It took Tesla, Uber, [44:41] Amazon, you know, decade plus in each of [44:43] those cases to win back their [44:45] investment. If you put it at 10 billion [44:47] >> This is a really good chart. Here's the [44:48] precise math on this. So I am building a [44:51] 1 gigawatt data center in Arizona. Okay. [44:54] When I greenlit that project, I thought [44:56] it was going to be a four or five [44:57] billion dollar investment. I was like, [44:58] okay, whatever. [45:00] Then it went to 10. Then it went to 15. [45:01] Then it went to 20. And now it's upwards [45:04] of 50 billion dollars for the powered [45:06] shell, for all the land, for all the [45:08] permits, then for all the [45:09] infrastructure, all the people, all of [45:11] it. Okay. [45:13] Sarah Friar said, I think it was about a [45:15] year ago, maybe less than a year ago, [45:17] that for them every gigawatt is about 10 [45:19] billion of annual revenue. So if you [45:22] think about that J curve Jason, really [45:23] the way to think about it is look, [45:25] energy equals intelligence. [45:27] For every gigawatt that they're trying [45:28] to spend, they have a five-year payback. [45:30] It's roughly what it means just to get [45:31] to break even. And then year six, seven, [45:33] and eight will be where the profit is. [45:35] Now how do you shrink the J curve? [45:37] You have better silicon. We're going to [45:39] see something from Jensen in a in a week [45:41] or two that uses [45:42] a bunch of the stuff that we partnered [45:44] with him at Groq on. [45:46] There'll be other people. There'll be [45:48] open source. So all those things can [45:50] shrink the depth and the surface area of [45:52] that J curve so that you can get out of [45:54] it faster. But right now that that thing [45:56] is roughly accurate, which is it's about [45:58] 50 billion per gigawatt. And it's about [46:01] a [46:01] five to six-year payback just to get [46:04] into the money, and then it's about 10 [46:05] billion a year. [46:07] And the technology industry has to do [46:09] something to make this better. [46:10] Could I, though, [46:13] take a step back and give you just a [46:14] different framing of all of this? [46:17] Please. [46:18] I think the big thing that we're [46:19] debating [46:21] is actually something we've seen in [46:22] every other technology trend [46:26] when it started to get some really [46:27] meaningful traction. So, in the first [46:30] generation of the internet when you [46:31] started to see e-commerce and all these [46:32] other business models. [46:34] Then in the second big wave of the [46:35] internet around the move to mobile and [46:37] the move to social. And then now we're [46:39] seeing this big wave around AI. [46:42] And I think what happens [46:44] is in step one [46:47] entrepreneurs [46:48] are AB testing [46:51] what it takes to raise money. [46:53] Okay, that's step one. [46:56] And I think what has happened [46:58] is that at least some parts of the AI [47:01] ecosystem [47:02] have decided that this crazy [47:06] scary doomerism is the best way to raise [47:09] money. [47:10] Where every now and then they come out [47:12] and they say, "All the jobs will be [47:14] destroyed." And Prophetic, you know, [47:15] Dario says that. "This thing is [47:16] sentient." [47:18] And investors are like, "Okay, here's 10 [47:20] billion. Here's 50 billion. Here's 100 [47:22] billion." [47:23] But then the second step happens. They [47:25] get the money. They start to do the [47:26] training. They start selling. [47:29] And then the investors are like, "Hey, [47:30] where's the revenue?" And so then they [47:32] start selling everywhere. [47:34] And then, if you see in the Department [47:36] of War example [47:38] all of a sudden you flip-flop. You [47:40] become sort of an un-serious dilettante [47:42] like partner to the American government. [47:44] They're like, "We're going to boot you [47:44] out." That's billions of revenue gone. [47:46] And what happens? Well, the same [47:47] investors that gave billions of dollars [47:49] are like, "Hey, hold on a second. [47:51] That's absolutely not allowed. You need [47:52] to conform and get back on track." And [47:53] so what does Dario do? He flip-flops. [47:55] And he's like, "Oh, I'm really sorry. I [47:57] didn't mean to. Let's sort of make [47:58] good." All of that, to me, is an [48:00] industry that's still in its very early [48:03] phases and still figuring out what its [48:05] place in society is. [48:07] So, what is the problem? [48:09] The problem is the following two clips, [48:11] and I'll just have Nick play these, and [48:13] I'd love your guys' reaction. The one [48:15] thing though that I think, even now, is [48:18] underestimated by all actors in [48:21] industry, [48:22] and including in Silicon Valley, is how [48:25] disruptive these technologies are. If [48:27] you are going to disrupt the economic, [48:29] and therefore political, power [48:31] significantly of one party space, highly [48:34] educated, often female voters who vote [48:38] mostly Democrat, and military and [48:41] working-class people who do not feel [48:43] supported, and you feel like that's you [48:45] believe that that's going to work out [48:47] politically, you're in an insane asylum. [48:49] Like, that you cannot have it This [48:51] technology disrupts humanity's trained, [48:55] largely Democratic voters, uh and and [48:57] makes their economic power less, and [49:00] increases the power economic power of [49:03] vocationally trained, working-class, [49:04] often male [49:06] uh voters, and and and so, these [49:09] disruptions are going to disrupt every [49:11] aspect of our society. And it to make [49:15] this work, we have to come to an [49:17] agreement of what it is we're going to [49:20] do with the technology, how are we going [49:22] to explain to people who are likely [49:24] going to have less good and less [49:26] interesting jobs, from their [49:28] perspective, and how is it that we are [49:31] going And by the way, on the military [49:32] thing, these technologies are dangerous [49:35] societally. The only justification you [49:38] could possibly have would be that if we [49:41] don't do it, our adversaries and uh will [49:44] do it, and we will be subject to their [49:45] rule of law. So, you If you decouple [49:48] this from the support of the military, [49:50] you're going to have an enormous problem [49:52] explaining to the American people why is [49:54] it that we're absorbing the risk of [49:56] disrupting the very fabric of our [49:58] society including the most powerful [50:02] parts of our society uh if it's not [50:05] because it's about maintaining our [50:07] ability to be American in the near term [50:09] and and and long term. Now watch Sam's [50:11] reaction. Fundamentally, our business, [50:14] and I think the business of every other [50:15] model provider is going to look like [50:18] selling tokens. But we see a future [50:23] where [50:24] intelligence is a utility like [50:26] electricity or water and people buy it [50:30] from us [50:32] on a meter and use it for whatever they [50:34] want to use it for. So, if you take [50:36] those three messaging veins on a [50:38] spectrum one is we have a sentient [50:40] supergod, we're the only ones that can [50:43] protect you from it, but you know, your [50:45] days are numbered. That's Dario. Alex, [50:47] which is hey, hold on a second, you [50:48] can't have it both ways. You can't both [50:50] say it on the one hand and then try to [50:52] run the fabric of society and flip it. [50:54] You need to be much more circumspect. [50:56] And then Sam's, which is we want to sell [50:57] tokens as a service. [50:59] I think the point is that this industry [51:01] right now, that revenue traction if [51:03] anything else [51:05] has distracted people from actually [51:07] getting on the same page and being much [51:09] more methodical [51:10] and much more reliable and trustworthy [51:13] in explaining all of this and managing [51:15] the expansion of this. And so, what I [51:18] would say is all of this fundraising [51:20] gobbledygook has actually created this [51:22] breathlessness that is not useful and [51:25] isn't helping. [51:27] And I would say there needs to be a lot [51:28] more seriousness by these folks to [51:30] actually run this business thoughtfully. [51:32] You can't be a dilettante, you can't [51:34] flip-flop, you can't pressure test AB [51:36] test this kind of messaging in public. [51:40] But I understand why you're doing it [51:41] because the stakes are so high. You're [51:43] playing this enormous poker game. [51:45] But I think we need to do a better job [51:46] of explaining all this to people because [51:48] right now my end of this is look at this [51:50] chart. [51:51] This is now the result of those three [51:53] messages. Here is where AI is. It is [51:55] slightly above the Democratic Party and [51:58] an autocratic state, that's where AI is. [52:02] ICE is more popular than AI. [52:05] So [52:06] >> very popular. [52:07] So [52:08] to me, this is really the crux of this. [52:11] Where we are not really being honest. It [52:14] would be much better if we said soberly, [52:16] there's a lot of experimenting, this [52:18] revenue is great, but we don't really [52:20] know what's real, we're going to try to [52:21] figure it out, we're going to work [52:22] methodically, there's a lot of regulated [52:24] industries, we're going to work within [52:26] those, we're not going to flaunt the law [52:27] and the rules, licensure will still mean [52:30] something. That's a way better, [52:32] thoughtful, mature message. [52:35] And Brad, what do you think? And rant. [52:38] And rant. Great rant. Uh, [52:40] Brad, does the industry have a PR [52:42] problem? Obviously these recent surveys, [52:45] and especially comparing them to China, [52:48] where people see AI as abundance and [52:50] like this incredible new technology they [52:52] want to embrace. Here, people are [52:53] scared. People are scared they're going [52:55] to lose their job, people are scared [52:57] about wealth [52:58] disparity, the rich get richer, the poor [53:00] get poorer. There's a lot of fear here [53:02] in the United States. What can our [53:04] industry do [53:06] to turn this around in terms of [53:08] communication from the big companies? [53:11] They don't seem to be communicating in [53:13] any coordinated fashion, and and they [53:15] obviously are scaring the out of [53:17] the public. Yeah, no, listen, I think [53:20] it's I think it's a fair rant and a fair [53:22] point. At the start of the Industrial [53:23] Revolution, at the start of, you know, [53:26] in the in the late 1800s, we had similar [53:28] social responses to innovations that [53:31] were occurring. We in fact had some [53:33] violent clashes, we had demonstrations [53:36] in the street, we had the entire robber [53:38] baron movement, you know, so class [53:40] warfare and worse is, you know, has come [53:43] with other rebel you know, kind of [53:45] industrial changes of this magnitude. [53:47] So, it doesn't surprise me that we have [53:49] a lot of anxiety by people that they may [53:51] lose their job. And I think there are [53:53] people out there who are kind of [53:55] forecasting into the future [53:58] in ways that are scary to, you know, the [54:00] average person who's who's listening to [54:02] this. And I don't think that's [54:03] particularly helpful. So, could we do a [54:05] better job messaging? No doubt about it. [54:08] But if I just rewind to kind of where we [54:10] started, I actually think the industry [54:12] is um you know, this is going to be a [54:15] pivotal year for the industry to [54:17] demonstrate, right, how this is really [54:19] beneficial for humanity. I think we're [54:22] going to be able to demonstrate that [54:24] it's very beneficial from a healthcare [54:26] perspective, from a drug discovery [54:27] perspective, from an education [54:29] perspective, etc. But we need to have a [54:31] coordinated effort because Chamath [54:33] right, it's deeply unpopular in the [54:35] country. I happen to be on the [54:37] optimistic side of this. 70% of the jobs [54:40] that exist in the United States today [54:42] did not exist 40 years ago. [54:44] Right? Right? We've gone through the [54:46] digital disruption that put a lot of [54:48] people out of work, but the abundance [54:51] and the and and the recreation of new [54:53] jobs, right, expanded the pie for [54:56] for largely everyone. I think that will [54:58] be the case here. If you listen to [54:59] Dario, he says the concern is that the [55:01] disruption occurs at a faster and and [55:04] and and bigger rate. And so that we [55:06] can't keep up with kind of that [55:07] replacement. I think that's another fine [55:09] point. But if we just go back to where [55:10] we started the conversation, which was [55:13] are these good investments, right? Do [55:16] >> That's not the conversation. No, of [55:17] course they're good investments. Of [55:19] course you're going to make money. [55:20] That's not what it's about. That's not [55:23] He asked the question. You made the [55:24] argument. Jason asked the question, [55:27] right? Are these companies simply [55:29] selling tokens at a loss, right? And we [55:32] moved into the [55:33] >> No, no, no. They're they're selling at a [55:34] profit. I'm buying them and losing [55:36] money. Right? [55:38] In the 1849 Gold Rush Yeah. Anthropic [55:41] and OpenAI and all of these model makers [55:43] are selling the pick and shovel in the [55:45] Gold Rush. I am buying it and I'm trying [55:47] to pan for gold. But as with the Gold [55:50] Rush, most of these companies will go [55:51] out of business. [55:53] And all I'm saying is if we are really [55:55] circumspect and honest [55:57] there is still way more to figure out [55:59] than has been figured out. This is not a [56:02] solved problem and I think we would [56:03] would behoove everybody to just tell the [56:05] truth about this. It would be way better [56:07] to be honest. This is not figured out. [56:10] >> I would say I think the data the cards [56:11] that are being turned on the table move [56:13] me in the exact opposite direction. [56:15] Okay, let me get Sachs involved and then [56:17] I'll give my take. Sachs, you have any [56:18] thoughts here? [56:19] Well, I have some thoughts on the [56:21] question you asked about is the industry [56:23] doing a good job with PR? I think the [56:26] answer is clearly no. I think they are [56:27] scaring the bejesus out of the public [56:30] and that's why the popularity is right [56:31] down there with [56:33] I don't know, what was it? Iran? [56:35] It's pretty [56:36] >> [laughter] [56:36] >> I think we're a little more popular than [56:38] than Iran, but um But look at Iran [56:40] Iran's had 100 years to it up. So [56:42] we've only had two. [56:43] Yeah, or 47 anyway. [56:45] >> Yeah, Iran's had 47 years to it up. [56:47] We've only had two. But it is very much [56:49] a US specific problem. If you look at [56:52] data sentiment data across countries [56:55] what you'll see is that other countries [56:56] are much more optimistic about AI than [56:59] the US. I think the Stanford did a study [57:02] on AI optimism. They simply asked the [57:04] question do you think AI is going to be [57:06] more beneficial than harmful? Something [57:08] like 80% of people in China said yes, in [57:10] the US it was in the 30s and it might be [57:12] even lower now. And it's not just China [57:14] and the US. You see across Asian [57:16] countries they tend to be pretty [57:17] optimistic. And then the US and Western [57:20] Europe tend to be pretty pessimistic [57:21] about it. I think that's less about the [57:24] reality of AI and more about our media [57:26] environment and who influences it. [57:28] Obviously you have the influence of [57:30] Hollywood has created a lot of dystopian [57:32] films about AI. You've got the fact that [57:35] like we talked about these CEOs are [57:37] doing a horrible job and they keep [57:39] talking about putting everyone out of [57:40] business. I mean, this is I think been [57:42] not accidental. I would say some of [57:44] these CEOs [57:46] are speaking this way because they're [57:48] not very good at coms. I think others [57:50] are actually doing it because they see a [57:52] strategy there. You know, they're going [57:54] for a regulatory capture agenda. [57:56] >> a good point, Sachs. It's delusions of [57:58] grandeur plus they're positioning their [58:00] companies. Yeah. It could be for [58:01] financing reasons like you've mentioned [58:03] Chamath, it's like they want to tout [58:04] this stuff for fundraising. But also, I [58:07] think that some of it is to [58:10] create [58:11] a regulatory backlash that they can then [58:13] control, you know, create a licensing [58:15] scheme or permissioning scheme. [58:17] And that's a big part of it, too. [58:20] And then I think you just have the fact [58:21] that in our media environment, the scare [58:24] stories are the ones that get a lot more [58:26] attention than the heartwarming stories, [58:29] you know, if [58:30] it bleeds, it leads type thing. So, you [58:32] can just see how unpopular it is for all [58:34] of these reasons. You know, New York is [58:35] about to outlaw medical and legal advice [58:38] from AI chatbots, which by the way, [58:40] that's probably the the most obviously [58:43] valuable and highest and highest ROI [58:45] thing for for a consumer. And it hurts [58:48] poorest people the worst. Hurts poorest [58:50] people the worst. But you understand [58:52] like if you're say a professional [58:54] association that sees it as your job to [58:56] protect your industry from disruption, [58:59] you might actually want to spread FUD [59:02] about AI in order to then seek those [59:05] protections through your state [59:06] legislature. Well, because if you have [59:08] companies that are fanning those flames [59:10] and those companies are the ones in the [59:12] industry, it's making your job even [59:13] easier. But just think about the poorest [59:15] person. They they can't afford a lawyer [59:17] and they want to do their own research [59:19] and they research the legal stuff to in [59:22] order to fight an eviction. Or there are [59:24] poor people who don't have a primary [59:25] care doctor, they're not insured and [59:27] they find a way to deal with, you know, [59:29] some medical issue they're having, and [59:31] this is the greatest thing. It's the [59:32] level It's a It's It levels the playing [59:34] field for [59:36] for people or people without resources. [59:38] So, this is the craziest stupidest [59:40] legislation ever. [59:41] >> I give them [59:43] the sky. [59:43] >> sky of the week. No, the problem is very [59:44] the problem is very specifically that [59:47] these people that rely on these models [59:50] to make a healthcare diagnosis or get a [59:52] legal opinion to help improve their [59:54] lives, [59:55] the makers of those tools are telling [59:58] everybody that they're about to bring [59:59] death and destruction upon the economy [60:01] in the world. So, then the lawyers and [60:03] the doctors are like, well, then maybe [60:05] we should slow this down, and they tell [60:07] their lobbyists, who then go to New York [60:10] and then tell the New York legislators, [60:11] "Hey, uh [60:12] these guys are like trying to wreak [60:14] havoc." And then they're like, "Oh, [60:16] yeah, well, then maybe we should shut it [60:17] down." That is the loop that's [60:19] happening. [60:20] Incredible. [60:20] >> That was the That was the Bernie Sanders [60:21] moment this week where he said, "We [60:23] ought to have a moratorium on all data [60:25] centers being built in the United States [60:27] because AI is dangerous." That was his [60:29] message, and that is what he's pushing. [60:31] >> And actually, that brings me to another [60:32] point, which is [60:34] if you look closely at Bernie Sanders' [60:35] messaging, one of the talking points he [60:37] used was literally verbatim from Future [60:39] of Life Institute. Uh I think it was [60:41] something about how AI is less regulated [60:43] than a sandwich shop, which is just not [60:45] true. But Future of Life is one of these [60:47] EA-funded doomer think tanks, and [60:50] they've got something like a billion [60:51] dollars. Vitalik Buterin donated 600 [60:53] million dollars of dog coins to to [60:56] Anyway, you've got I mean, this is one [60:58] really weird sort of quirk about our [61:01] media environment is we have these [61:03] EA-funded think tanks with literally [61:06] billions of dollars. You know, it's guys [61:08] like Dustin Moskovitz and They're the [61:10] D-cells. They're the Democratic [61:12] Socialists kind of warm. It's so weird [61:13] that like New York is now taking the [61:15] crown of most state from [61:16] California. I don't know how this [61:18] happened, but they just seem to be [61:19] making every mistake possible. [61:21] >> But, [61:21] >> on. Let me just finish my point because [61:23] you've got these let's call them doomer [61:24] think tanks funded by these EA [61:26] billionaires. They have literally [61:27] billions of dollars. You can influence a [61:29] lot of public discourse with that. A [61:31] lot. Short. And they are behind a lot of [61:34] the NIMBY stuff around data centers. [61:36] They've been spreading a lot of the FUD [61:38] around data centers increasing your [61:39] electricity prices, which again, I think [61:41] is a solved problem now because the [61:44] users the AI companies have agreed to [61:46] pay for the incremental costs and stand [61:48] up their own power generation. They've [61:50] been spreading a lot of stories about [61:51] water usage, which is just totally made [61:53] up. I mean, the modern data centers [61:55] recycle their water. [61:57] So, they don't use up water. But again, [61:59] you've kind of got these doomer groups [62:01] who are just trying to stop AI however [62:02] they can. [62:04] And they're extremely well funded and [62:05] they're having a big impact. And I think [62:06] actually this is one of the reasons why [62:09] you're seeing in the US again that AI I [62:11] think is [62:12] it's basically the least popular thing [62:13] they can poll for except for the [62:14] Democratic Party and Iran. And by the [62:16] way, FLI, the Future Life Institute, [62:18] they also fund journalism [62:20] fellowships and endowments at [62:22] publications. So, Who probably writes [62:24] negatively about AI? [62:26] Yeah, exactly. Here's Sax, look at this [62:28] chart. It goes back now a little bit [62:30] earlier than '23, but I had the [62:32] data accurately from '23, so we're going [62:34] into the fourth year. [62:35] About 40% of all protested data centers [62:39] in America get canceled. [62:42] And so in 2023, this was a non-issue. [62:44] There were a few data centers that were [62:46] protested. And a few of those, 40% of [62:49] them, I think it was literally like one [62:50] or two got canceled. [62:52] But then starting in '24, when you have [62:54] this divergence of messaging or this [62:56] chaotic slipshod messaging, [62:58] and it was just a fever pitch to raise [63:00] money, [63:01] what you started to see was this [63:03] fomenting of negative perspective by [63:06] individual people on the ground. [63:09] And so in 2024, about 40% of all [63:11] protested data centers were canceled. [63:13] Still a small number, you could ignore [63:15] it. But last year was when the bottom [63:17] fell out. [63:18] We had about 25 data centers canceled, [63:21] about 5 gigawatts that got canceled. If [63:24] you use Sarah Friar's number, that's $50 [63:29] a year of revenue, [63:31] which is off the table because of what [63:33] happened in 2025. Now, that has [63:35] implications to everybody. Look at the [63:37] amount of taxes that that would actually [63:39] raise for federal and local and state [63:41] governments. All gone, vanished. [63:43] In 2026, [63:45] just at the end of February, there were [63:47] about a hundred data centers being [63:49] protested, [63:50] which if you flow that through, will [63:52] mean about 40 [63:54] will get canceled. [63:56] And that number right now is about 7 [63:58] gigawatts, so another 70 billion dollars [64:01] a year revenue. [64:02] So, just last year and this year, we've [64:05] taken off the table a hundred and twenty [64:06] billion dollars of revenue per year. [64:10] This is a wake-up call that this [64:11] messaging is wrong. These people are not [64:14] doing what is right on behalf of a very [64:16] nascent and critical industry for [64:18] America. [64:19] There's only so much that Sachs can do, [64:21] the White House can do. [64:22] All these other people are kind of at [64:24] the periphery, but if the people that [64:26] are on the ground don't get their [64:28] together, this is a national disaster. [64:30] Just to give people a sense of where [64:33] this is happening, almost all of these [64:35] cancellations are Virginia and Indiana, [64:37] according to some of some cursory [64:38] research here. [64:40] And there have been zero cancellations [64:42] due to local opposition here in the [64:43] great state of Texas, where we have over [64:46] 150 gigawatts of data capacity requests. [64:49] So, if you want to do this, come to [64:51] Texas, talk to Abbott, talk to Ted Cruz, [64:53] you just CC them on your tweet, and [64:55] they'll have you to the poker game, and [64:57] they will greenlight it. Brad, before we [64:59] move on, want to get your opinion on [65:01] open source and how powerful it is and [65:04] how powerful Apple silicon is getting. [65:07] I'm not sure how you factor this in at [65:09] Altimeter [65:10] into your thinking, but this seems to me [65:12] to be a massive headwind against the two [65:16] big bets you have. [65:18] All of these open source models, we [65:19] started running them [65:21] picking up about 85% of our tokens right [65:23] now, and every startup I know [65:25] is saying, "We are standing up our local [65:27] models, and we only use the top models, [65:31] the paid ones, uh when we have jobs we [65:33] can't do." So, I I'm just curious your [65:34] thoughts on that, and you can add to [65:37] that the auto research project from [65:39] Karpathy that came out this weekend. For [65:41] people who don't know, we now have [65:44] a group of tinkerers who are setting up [65:45] their open claws, and now setting up [65:47] large language models, and now trying to [65:48] train them with this auto research tool. [65:51] This seems like a parallel track that [65:52] could be material. I'm just curious if [65:55] you're monitoring it at all. [65:56] >> I mean, first I would say that uh [65:58] I am [66:00] very enthusiastic for open source. Okay. [66:02] We see it in widespread use [66:04] everywhere. But, here's the interesting [66:06] thing. You know, for the advanced [66:07] companies, they're doing some like [66:09] planning, [66:10] you know, with the frontier labs, and [66:12] then they're kind of doing the [66:13] execution, if you will, with the open [66:15] source models. So, they're running an [66:16] ensemble of model strategy. But, here's [66:19] what I think's more impressive. [66:21] We have incredible open source models [66:24] nearly on the frontier, and [66:27] notwithstanding that, we're seeing [66:29] companies like Anthropic add five or six [66:31] billion dollars of revenue in a single [66:34] month, which is extraordinary. We've [66:36] never seen anything like it in [66:38] technology. And that's in the face, [66:40] Jason, of all this open source usage. [66:42] So, what does it tell me? It tells me [66:44] that the TAM is dramatically bigger than [66:48] any of us think that it is. And that, [66:50] you know, when we when we look back on [66:52] this period, [66:54] you know, that will be the big takeaway. [66:55] It's a takeaway with Uber, the takeaway [66:57] with Google, the takeaway with Amazon. [66:59] The TAM was way bigger. We've crossed an [67:01] important threshold. Open source will be [67:03] a part of it, but clearly the front [67:05] frontier labs can do well even in the [67:07] face of it. All right, a little [67:09] housekeeping. [67:10] The All-In Summit is coming and [67:12] liquidity is uh I think it's sold out or [67:15] about to sell out. We might have a [67:16] couple tickets left. You can find both [67:18] events at allin.com/events. [67:21] And uh All-In Summit tickets if you want [67:23] to get there quickly before they sell [67:25] out. September 13th, 14th, and 15th in [67:27] Los Angeles again. [67:29] And [67:30] uh for all the All-In listeners, we're [67:31] launching a survey today. This is super [67:33] important to take the All-In survey if [67:36] you made it to this point in the [67:37] podcast. If you are [67:39] an early true believer in the pod, if [67:41] you're one of the uh All-In stands, we [67:43] need you to fill out this survey. [67:44] allin.com/survey. [67:46] allin.com/survey. [67:48] All right, let's uh wrap up with this [67:49] final story that went viral. The [67:51] millionaire tax has hit Washington [67:54] state. Howard Schultz, [67:56] CEO uh long-time CEO of Starbucks, [68:00] has bowed and he's gone to Miami. [68:02] Surfside. He's in Surfside. He bought a [68:05] He bought a condo in Surfside. He uh He [68:07] pulled a Jay Cal. He pulled a Jay Cal. I [68:09] think you mean a sexy boo? [laughter] [68:12] Well, yeah, but I was never I was never [68:14] a Starbucks liberal before I left the [68:16] state of California. [68:16] >> Listen, I I don't know how many times I [68:18] have to make this correction. I am a [68:19] moderate. I literally voted four [68:21] elections in a row for Republicans. [68:23] People have asked me for the receipts. [68:24] Pataki, Giuliani, um [68:27] and uh Bloomberg. I literally for almost [68:30] a [68:30] >> Okay. Okay. voted exclusively [68:32] Republican. Washington's millionaire tax [68:35] passed this week. Here's what the tax [68:37] is. People making more than $1 million a [68:39] year will pay an extra 9.9% [68:42] in tax starting in 2029. [68:45] The Budget Center estimates the tax will [68:47] impact 30,000 households bringing [68:48] another four billy uh for the state's [68:51] general fund. Um [68:53] the funds are supposed to go towards [68:54] public schools, higher education, and [68:55] health care. [68:56] In a huge coincidence, on the same day [68:59] the new tax was passed, [69:01] Howard Schultz, the billionaire [69:02] Starbucks founder [69:04] Huge coincidence, did you say? Yeah. [69:07] >> [laughter] [69:08] >> Just unrelated stories. [69:10] >> Unrelated in unrelated story, yeah. [69:11] Unrelated news. [69:12] >> He will be leaving Seattle after a [69:13] 44-year run. [69:15] Because he found out about these [69:17] incredible Cuban sandwiches [69:19] >> an opportunity There was an opportunity [69:21] to buy a $44 million condo in Surfside. [69:23] He couldn't pass it up. It just happened [69:25] to be on the same day that they passed a [69:26] millionaire tax. He He had the Cuban [69:29] sandwich at [69:30] Le Sandwich and he He fell in love. [69:34] Schultz has been getting crushed [69:36] after saying when he ran for president [69:37] that he would be willing to pay more [69:39] taxes. Bezos obviously left back in [69:41] November of 2023. [69:44] And people speculated maybe the 7% [69:46] capital gains tax [69:48] would [69:49] have influenced that, who knows. [69:52] So, I guess, Chamath, what is the end [69:55] game here because [69:57] for these local politicians [70:00] they must have learned the lesson [70:03] that [70:04] people of means can move. They have the [70:07] ability to buy new homes, put their old [70:11] homes on the market. They're very [70:13] mobile. And they could even leave the [70:15] United States and go to Singapore or [70:17] Dubai or other locations in the world. [70:20] Why are they still enacting these and [70:22] will they continue to still enact these [70:24] until we get to 60, 70% tax rates and we [70:27] just lose all of the creators and this [70:29] becomes An Randian? I think that state [70:31] politicians on the West Coast are very [70:35] ineffective and not not very smart. [70:38] Nick, there's a There's a tweet that was [70:40] published. I think maybe it was an [70:41] infographic [70:43] that showed net migration rates of every [70:45] single state [70:47] for 2025. Washington is a few months [70:50] behind California [70:53] in trying to enact these stupid [70:56] taxes. [70:57] And the reason they're stupid is these [70:59] kinds of things don't work at the state [71:00] level. [71:01] And we know what it's already done in [71:03] California because the Hoover [71:04] Institution just published something [71:06] this morning. [71:07] And it's a complete indictment of of [71:10] what the billionaire tax was trying to [71:12] do. And by the way, this billionaire tax [71:13] is only polling [71:15] right now at 25% of the votes it needs. [71:18] So, maybe it'll find a way to get on the [71:20] ballot. And then even then it'll have an [71:22] uphill climb to get voted in. [71:24] But look at the destruction that it has [71:26] done in California by just announcing [71:28] it. [71:29] The Hoover Institution basically ran [71:31] this Monte Carlo simulation. They ran [71:33] 100,000 runs. [71:35] And in 71% of those runs, it comes out [71:37] with a negative NPV. [71:40] And if you expected value it out, it's [71:42] about a $25 billion hole. [71:45] They also found that they over counted [71:48] the number of billionaires in [71:49] California. So, that number was wrong. [71:51] They under counted the amount of revenue [71:53] that they pay. So, that was wrong. And [71:55] they over counted the estimate of how [71:57] much money that they would make. So, [71:58] they're not good at math. They're not [72:00] good at math. So, when you add it all [72:02] up, they thought they were going to make [72:03] 100, they're actually going to make 40. [72:05] The people that left pay, you know, [72:08] three to five billion dollars [72:10] a year of taxes. [72:12] It's going to create a $25 billion hole. [72:14] You're going to have the middle class [72:15] that's now going to have to foot this [72:17] because this is net revenue that's not [72:19] going to come into the budget. [72:21] That's about 2,500 per [72:23] middle class household. There's about 10 [72:25] million in California. [72:27] So, that's what's happened just by [72:29] making the threat. Right. [72:31] >> Washington had a 23-hour debate and [72:33] passed the law. [72:35] So, I suspect when you look back on this [72:37] in 18 or 24 months, it'll be as bad or [72:40] worse than California. [72:42] These things don't make sense. [72:45] The reason they don't make sense is that [72:46] you are putting good money after bad. We [72:49] all know that money that goes to the [72:51] state governments [72:53] are wasted. We just don't know how much. [72:55] And so when you keep asking more, [72:57] eventually the smart people say enough's [72:59] enough. I'm out of here. We might find [73:00] out how much. I think Barry Weiss is on [73:02] the case. I don't know if you saw her do [73:03] her CBS report this week. She's She's [73:06] going hard for fraud and until you get [73:09] fraud out of the system, I don't think [73:11] you have the moral high ground to raise [73:13] taxes. I think that should be the [73:14] message [73:15] >> Well, [73:16] that's that [73:17] all That should be your campaign promise [73:19] when you run. Okay. Hi, [73:21] I'm Jason Calacanis and I will get rid [73:22] of fraud and lower your taxes. [73:25] Look, you may have seen an even more [73:27] severe tax was proposed at the federal [73:29] level where Bernie Sanders and and I [73:31] think Ro Khanna came out with their [73:33] version of a national wealth tax where [73:35] it wasn't just 5% once like in [73:38] California. It was 5% per year. Per [73:40] year. So in other words, in roughly 20 [73:42] years the federal government's is going [73:43] to take all of your money. I mean, [73:45] that's it Look, this is socialism. This [73:47] is another way to get to the same end [73:48] point which is the government owns [73:50] everything. They seize. Well, I mean the [73:52] seizure part of it I think is [73:55] the nuance point we have to get across [73:57] which is if you earned it and paid your [73:58] taxes already, It's yours. [74:00] >> can just decide, you know what? We [74:02] didn't take enough 10 years ago. We need [74:05] to go seize that. Hey, when you sold [74:06] your hammer, [74:06] >> Right. it we didn't take enough. We we [74:08] need to take it now. [74:09] >> Raise your hand if you believe the [74:11] things you own are better off being [74:13] owned by Bernie Sanders and Ro Khanna. [74:16] Raise your hand if that's what you [74:17] believe. I mean, you'd have to be an [74:18] idiot to believe that. Ro Khanna I'll [74:21] tell you the last time we saw these [74:22] proposals of asset seizures was during [74:25] the Gilded Age, you know, 1870 to 1920. [74:29] You know, then it was Carnegie and [74:31] Rockefeller. [74:32] You know, what's interesting, I went [74:33] back and looked at it. There were [74:35] actually like it was actual warfare. We [74:37] had hundreds of people killed in [74:39] clashes, [74:40] um, you know, during the Great Railroad [74:41] Strike, the Pullman Strike, etc. Um, and [74:44] it was all over this. And so I think [74:46] shame on the politicians that are [74:48] fanning the flames of class class [74:49] warfare. [74:51] Um we all need to bring the temperature [74:53] down. There are fair debates on whether [74:56] states have enough resources to fulfill [74:58] their obligations to the citizens. There [75:00] are fair debates about fraud. All has to [75:02] be taken on. But, I think it's [75:04] interesting in the state of California, [75:06] right? I think the teachers union is [75:07] against the the billionaires tax because [75:10] they know it's going to lead to less [75:11] dollars for education for the state of [75:13] California. Matt Mahan, who's running [75:14] for governor, is against the tax. The [75:16] current sitting governor, Democrat, [75:18] against the tax. Right? They all need to [75:20] step up and explain, not just that [75:22] they're against the tax, but we you're [75:25] either on the side of business and [75:27] entrepreneurs and creativity and moving [75:30] the state forward and growing the [75:31] economy, or you're against it. And [75:33] that's what's at stake here. And [75:34] fortunately, the outcome of the battle, [75:38] uh you know, during the Gilded Age was [75:39] that America didn't abandon [75:41] entrepreneurialism. We didn't abandon [75:43] capitalism like Europe did. And now we [75:46] leaned [75:46] >> Now we played it [75:48] One point on that. So, you mentioned [75:50] that some of the unions in California [75:52] are opposed to this this asset seizure [75:54] tax. That's only because they weren't [75:56] cut in on it. [75:57] >> Right. Because So, there's already [75:58] rumors that the California Teachers [76:01] Association is working on their own [76:03] version [76:03] >> Oh, boy. of a billionaire asset seizure [76:07] for not this election cycle, but for the [76:08] next one. [76:09] >> Yeah. And next time they're going to [76:11] have their ducks in a row, and they're [76:12] going to have all the pigs at the [76:13] trough, and all the unions are going to [76:14] get together. [laughter] [76:16] Because the SEIU UHW, they did this on [76:20] their own. So, again, they didn't allow [76:23] all the other groups to wet their beak. [76:24] So, I think that unfortunately, I think [76:26] that's going to be corrected. If this [76:27] one doesn't pass, they'll correct it for [76:29] '28, and it's more likely to pass [76:32] because they're all going to do it. And [76:34] the other thing is that I think that by [76:36] '28, this national wealth tax will just [76:39] be [76:40] a standard part of the Democratic [76:42] platform. [76:42] >> It could be table stakes, yeah. Yeah, [76:44] it's table stakes. I think that, you [76:45] know, the Bernie Sanders real kind of [76:46] position will be the position of the [76:48] Democratic Party. And you even see Gavin [76:50] Newsom creating wiggle room for himself [76:52] to embrace this position. If you look [76:54] closely at his statements formally [76:57] opposing the the BTA in California, what [77:01] he says is that a state can't do this by [77:02] themselves because they got 49 other [77:04] states to compete with. So, what he's [77:07] saying is that, you know, look, [77:09] California's operating in a competitive [77:10] environment. One state can't do it. And [77:13] then he's leaving [77:15] the other part elliptical, which is, [77:16] well, the the federal government needs [77:18] to do this. And I think that you can [77:19] expect him to embrace that position by [77:22] 2028. [77:23] >> There is a [77:24] way out here from this socialist [77:26] movement. It's a It's very simple. If [77:28] you think from first principles, what [77:30] does an American want? What does an [77:31] American family want? What What do [77:32] mothers and fathers in this country [77:34] want? They want to educate their kids. [77:36] They want to be able to own a nice home. [77:38] They want to have decent health care. [77:39] They want to have healthy food. It's a [77:41] very small subset of issues. And AI is [77:45] uniquely positioned to solve a lot of [77:47] these problems. And entrepreneurs can [77:49] come in and take these highly regulated [77:51] industries, if we're allowed [77:53] to participate in them. Education, we [77:56] have to break this accreditation, you [77:59] know, cartel. And then housing, we have [78:01] to break these regulations like the [78:03] great state of Texas, Nevada, and [78:05] Florida have. And then, you know, when [78:07] it comes to health care, this is where [78:09] AI could have a tremendous impact. And [78:11] entrepreneurs can have a tremendous [78:12] impact in lowering the cost of health [78:13] care and letting people solve for that [78:16] with their own, [78:17] you know, health care led, you know, [78:19] self-led health care. These are the [78:20] problems. If you solve for these [78:22] problems, [78:23] people's homes, people's health, [78:25] people's education of their kids, we're [78:27] going to solve these problems and we [78:28] don't need to go to socialism and seize [78:31] people's assets. That's what [78:32] entrepreneurs should be doing. That's [78:33] what entrepreneurs should be working on. [78:35] And that's what we where the government [78:36] can help. That's where Trump's uniquely [78:38] qualified. He is the regulatory breaker. [78:41] He got nuclear back on the agenda. No [78:43] other president had gotten nuclear back [78:45] on the agenda for America. He can get [78:47] housing back on the agenda. We have to [78:49] break those things and not start foreign [78:51] wars, to my earlier point, and start [78:53] creating houses for Americans. Another [78:55] amazing episode of the All-In podcast. [78:57] Gentlemen, [78:58] Chamath, you always give yourself the [79:00] last word, J Cal. Well, I go last. If [79:02] you would like, you can moderate and [79:04] I'll go first. But I always go last. [79:06] But you can you can I'm you can give me [79:09] a round of applause or you can throw a [79:10] tomato. Go ahead, Sax. It's all good. [79:12] It's all good. Go for it. All right, [79:13] everybody. Another amazing episode of [79:17] the All-In podcast. Thank you, Bestie [79:18] Brad, for joining us. Thank you, Bestie [79:20] Brad. Love you, boys. Bye-bye. [79:24] Let your winners ride. [79:26] Rain Man, David Sacks. [79:31] >> [music] [79:31] >> And I said, we open-sourced it to the [79:32] fans and they've just gone crazy with [79:34] it. Love you, Bestie. Queen of Kinwah. [79:37] >> [music] [79:40] >> Let your winners [79:41] ride. [79:44] Besties [music] are back. [79:46] And is my dog taking a dump in his [79:48] driveway, Sax? [79:51] Oh, man. [79:52] And if that guy should meet me at the [79:54] party, we should all just get a room and [79:55] just have a one big huge orgy because [79:57] they're all just useless. It's like this [79:58] like sexual [music] tension that they [80:00] just need to release somehow. [80:02] What your big feet What your What your [80:05] big feet [80:06] What [80:07] We need to get merch. Besties are back. [80:22] Hey.