Micron Sales Quadruple on AI Demand, Triggering Tech Price Hikes

AI需求推动美光销售额翻四倍，引发科技产品涨价 全球高带宽内存的严重短缺正挤压硬件制造商的零部件成本，从而将科技基础设施成本直接转嫁给消费者。 本期节目重点聚焦重塑科技格局的政治与经济变革，首先讨论了美国民主社会主义者在纽约市赢得关键初选。随后，主持人和嘉宾分析了中国如何通过模型蒸馏与华为本土硬件在AI领域快速追赶，并探讨了美光凭借在全球高带宽内存领域的垄断地位所取得的亮眼收益。最后，讨论转向前沿基础设施经济学，评估了天基轨道计算中心相较于地面数据中心的可行性，并剖析了Cerebras和Anthropic近期的公开市场表现。 在纽约市长 Zohran Mamdani 的背书支持下，民主党社会主义派系在纽约市国会初选中实现三战全胜的横扫战绩。 (1:21) Chamath Palihapitiya 认为 AI 是当今时代最强大的经济平衡器与平等工具，但硅谷在向公众传达其价值方面却遭遇了严重失败。 (2:49) David Sacks 指出，美国政治的未来正在演变为一场明确的意识形态抉择：民主党的社会主义路线与共和党的民族主义路线之间的对决。 (6:17) 中国人工智能公司 Z.ai 推出拥有 7440 亿参数的开源模型 GLM 5.2，不仅在 SWE 编程基准测试中超越 GPT-5，且 API 成本大幅降低。 (45:00) 中国正通过大规模“iPhone 农场”从云端 API 提取推理轨迹进行模型蒸馏，将与美国前沿模型的差距大幅缩小至约 6 到 9 个月。 (45:00)

The All-In Podcast panel argues that socialist political gains in New York City reflect shifting class demographics and Silicon Valley's narrative failures, while artificial intelligence development faces acute hardware bottlenecks and rising geopolitical competition. The hosts and guests analyze how domestic regulatory hesitations collide with rapid open-source advances in China alongside an unprecedented memory supply crunch. Socialist Primary Victories and Tech's Political Narrative (1:21) Following endorsements by New York City Mayor Zohran Mamdani, Democratic Socialists of America (DSA) candidates achieved a three-race primary sweep in New York, an outcome Polymarket had given just a 26% chance before election day. Brad Lander unseated two-term incumbent Dan Goldman in the 10th District, 32-year-old Chevalier defeated a five-term incumbent backed by House Speaker Hakeem Jeffries in the 13th District, and Claire Valdez won an open seat in the 7th District. David Sacks described the DSA as an ideological insurgency seeking to dismantle traditional constitutional institutions, driven by coalitions of downwardly mobile, educated progressives and recent immigrant voting blocs. Chamath Palihapitiya argued that Silicon Valley enabled this populist surge through self-inflicted branding crises and internal litigation, which distracted from promoting artificial intelligence as an accessible economic leveler capable of democratizing expertise. Chinese Open-Source Capabilities and U.S. Regulatory Friction (45:00) The panel reviewed the release of GLM 5.2 by China's Z.ai, a 744-billion-parameter open-weight model with a 1-million-token context window distributed under an MIT license. Benchmark metrics show the model outperforming GPT-5 on the SWE coding benchmark while trailing Claude Opus 4 by under one percentage point, offered at an API cost 85% lower than GPT-5.5. Gavin Baker attributed this rapid catch-up to distillation through masked API queries to capture reasoning traces, noting that GLM 5.2 now executes its own reinforcement learning. Sacks reported that Z.ai's GLM 5 family and DeepSeek V4 were trained on clusters of indigenous Huawei Ascend 910B chips, warning that while China trails the U.S. by roughly six months on frontier models and 24 months on silicon, domestic regulatory rollbacks like the withdrawal of Anthropic's Fable risk leaving global markets open to state-backed Chinese 'AI in a box' hardware-software bundles. High-Bandwidth Memory Shortages and Consumer 'AI-flation' (60:00) Micron reported quarterly revenue surging fourfold year-over-year from $9 billion to $42 billion, raising fourth-quarter revenue guidance to $50 billion against consensus expectations of $43 billion. As one of only three global producers of High Bandwidth Memory (HBM) alongside Samsung and SK Hynix, Micron has fully committed its supply through 2026, securing long-term supply chain agreements with floor pricing across four primary customers representing half of its revenue. Baker identified DRAM and HBM stacking—moving from 8-die configurations toward 12- and 16-die vertical stacks—as the primary structural bottleneck across AI infrastructure, with memory projected to consume 30% to 40% of hyperscaler capital expenditures next year. The panel observed that component shortages are directly causing consumer hardware price hikes, including a 14% increase on Apple's entry-level MacBook Neo (from $699 to $799) and a 25% increase on the Mac Studio, alongside price increases across video game consoles. Terrestrial Data Center Bottlenecks and Distributed Architecture (69:00) Terrestrial computing deployment faces severe environmental and power restrictions, with approximately 40% of planned data center projects contested since 2021. The panel estimated that constructing a one-gigawatt terrestrial facility currently requires $35 billion in silicon and $25 billion in power, cooling, and labor infrastructure, driving total facility costs toward $70 billion. In contrast, Baker noted that reusable Starship launches could theoretically place a gigawatt of compute into orbit for roughly $5 billion in transport costs, altering data center cost structures. On the ground, operators are mitigating constraints by shifting toward modular 'Megapod' hardware deployed at existing power locations, such as vehicle charging stations, and disaggregating inference into capacity-bound pre-fill stages and bandwidth-bound decode stages using decode-specialized chips like Cerebras or Grok to extend the operational life of legacy GPU racks. The discussion leaves unresolved whether U.S. capital markets and electrical infrastructure can sustainably accommodate the projected $4 trillion pipeline of impending AI public offerings, and whether Western regulatory frameworks will adapt quickly enough to prevent foreign open-source architectures from dominating enterprise standards.

本期节目探讨了美国政坛左翼社会主义势力崛起与科技行业的公关失焦，同时分析了中国在开源人工智能领域的突破、全球AI内存瓶颈对算力基础设施的重塑，以及公开市场消化万亿级科技IPO所面临的现实挑战。 纽约初选左翼大胜与硅谷形象危机 (1:21) 在纽约市国会民主党初选中，得到纽约市长Zohran Mamdani支持的美国民主社会主义者（DSA）阵营在三个目标选区中大获全胜，而Polymarket在选举日前预测全胜的概率仅有26%。在富裕的第10选区，Brad Lander击败了两届现任议员Dan Goldman；在哈莱姆等贫困选区集中的第13选区，32岁的DSA候选人Chevalier击败了众议院民主党领袖Hakeem Jeffries支持的五届现任议员；Claire Valdez则在第7选区获胜。David Sacks指出，DSA纲领主张废除参议院与选举人团、解散移民及海关执法局（ICE）及主要企业公有化，这标志着激进左翼正加速渗透民主党建制派。 Chamath Palihapitiya认为，人工智能本应是人类历史上最强大的经济平衡工具与平等推手，能够像为每个人配备顶级创业导师一样打破阶层壁垒。然而，硅谷严重缺乏合格的品牌代言人，将公关舆论场拱手让给“末日论”、失业恐慌以及水资源消耗等虚假焦虑。科技界内部的纷争让公众对资本主义失去信心，进而转向社会主义叙事，导致犹他州和纽约等地的选战实际上演变为由Anthropic与OpenAI资助团体之间的AI代理人战争。 中国开源模型突破与美国监管自缚 (45:00) 中国Z.ai发布了开源模型GLM 5.2，拥有7440亿参数和100万token的上下文窗口，采用宽松的MIT许可协议。该模型在SWE编码基准上超越了GPT 5，与Anthropic的Claude Opus 4差距不足一个百分点，而API使用成本比GPT 5.5低85%。Gavin Baker分析指出，中国团队利用大量设备通过云API抓取前沿推理轨迹（distillation）进行强化学习，大幅缩短了技术差距。此外，DeepSeek V4与GLM 5系列均被指在华为昇腾910B集群上完成训练与推理优化，显示出中国软硬件自主化的快速进展。 David Sacks对美国的监管现状提出尖锐批评，认为中国在模型上仅落后美国约6个月、芯片制造落后24个月，但正以极快速度追赶。他指责Anthropic的Dario Amodei等行业高管游说推动监管审查，导致Anthropic的Fable和OpenAI 5.6等美国顶级模型因越狱事件而陷入审批停滞。Sacks强调，试图通过限制本土模型来防范网络安全风险是完全错误的，因为中国技术很快就会赶上；美国必须保持技术出口与迭代，否则中国将以极具价格竞争力的全套硬件加模型方案迅速占领全球中立市场。 HBM内存短缺加剧与算力架构重构 (60:00) 美光科技（Micron）公布了极为亮眼的财报，季度营收同比激增4倍，从90亿美元跳升至42亿美元，并给出500亿美元的强劲第四季度指引，远超华尔街预期的43亿美元。作为全球仅有的三家高带宽内存（HBM）制造商之一，美光已将2026年的产能全数售罄，仅四家大客户就锁定了其约50%的收入。高昂的组件成本正传导至消费电子市场，苹果被迫将入门款MacBook Neo提价14%至799美元，Mac Studio提价25%，微软Xbox以及其他主机也面临涨价或提价预期。 由于地面建设面临巨额成本与审批延宕，建造1吉瓦地面数据中心通常需要350亿美元芯片和250亿美元电力冷却设施。严苛的电网审批促使行业探索新架构：一方面是特斯拉与SpaceX通过商标注册披露的“Megapod”模块化集装箱式数据中心，可将建设周期从数年缩短至90天；另一方面，若采用完全可重复使用的星舰（Starship），将1吉瓦算力发射至轨道的发射成本仅约50亿美元，轨道计算的总成本（400亿美元）将显著低于地面的700亿美元。在推理端，业内正利用Grok与Cerebras等专精解码（decode）的芯片配合旧款英伟达GPU，形成分布式微型数据中心以延长硬件寿命。 万亿级AI上市潮与公开市场定价机制 (87:54) 公开市场正面临前所未有的科技资产供给，预计储备管道中的IPO规模高达4万亿美元，其中Anthropic等巨头上市后的估值预期甚至达到3万亿美元。与会者认为，市场具有足够的流动性承接这些资产，资金将从私募市场转向更庞大的公开市场。关于SpaceX解禁后的抛压担忧被认为过度夸大，因为其员工和早期投资者在过去十年中每六个月就享有一次流动性窗口，且近半数员工在IPO时按新估值认购了股票。 针对Cerebras上市首季后跌破发行价的现象，Gavin Baker深入剖析了公开市场的心理机制：大量公募基金经理将跌破发行价视为买入逻辑破坏而机械性抛售，从而给空头创造了砸盘获利的空间。Cerebras的问题在于未向公开市场充分沟通硬件交付周期——其与OpenAI签订的200亿至250亿美元订单需经台积电4个月晶圆制造、2个月服务器组装及电力部署，实际收入需到下半年才会体现。与会者呼吁更多拟上市公司在IPO时采用荷兰式拍卖（Dutch auction），以更透明的市场博弈取代投行主导的定价漏洞。 本期讨论表明，地缘政治层面的意识形态对立与技术出口管制，正与算力基础设施的物理极限及资本流动交织在一起。面对中国在开源模型与国产算力上的快速逼近，以及本土内存供给瓶颈带来的全产业链通胀，美国科技生态系统能否在摆脱监管自缚的同时维持公开市场信心，仍是悬而未决的关键命题。

## Background 本期内容由知名播客频道《All-In Podcast》于 2026 年 6 月 27 日发布在 YouTube 平台上。根据简介，本期特邀嘉宾 Gavin Baker 与前 Uber 联合创始人 Travis Kalanick 共同做客，核心讨论了曼达尼背书的民主社会主义者横扫纽约国会初选、民主党未来路线与中东议题、中国开源大模型的迅速追赶、美光爆炸性财报折射出的内存挤压、太空轨道算力与分布式数据中心的经济账，以及 Anthropic、SpaceX 和 Cerebras 等超级科技 IPO 的最新市场风向。 ## TL;DR 节目首先深入剖析了纽约国会初选极左翼社会主义势力（DSA）大胜对民主党建制派与全美政治格局的深远震荡，反思了硅谷在叙事真空下的信任流失。在硬科技领域，主持群与嘉宾聚焦智谱 GLM-5.2 等中国开源大模型的崛起、模型蒸馏内幕及软硬一体化趋势，并结合美光创纪录的单季业绩拆解了 AI 算力潮向消费电子蔓延的“内存危机”。最后，嘉宾们从第一性原理核算了地面超算与 SpaceX 轨道算力的成本差异，拆解了特斯拉 Megapod 等模块化机房与分布式推理的架构革新，并针对 Anthropic 3 万亿美元估值前景及 Cerebras 破发后的资本市场机制展开了深刻推演。 ## Atmosphere & style 全场兼具硅谷一线宏大叙事的极度自信与面对政治极化撕裂时的严肃忧患，兼顾了顶级对冲基金经理的数据严谨与老友重聚间的风趣互损。从对极左政治纲领和官僚监管的辛辣抨击，到探讨芯片堆叠与太空算力时的硬核兴奋，主持与嘉宾间的思维碰撞火花四溅、节奏极为紧凑。整期播客既充斥着对百亿乃至万亿美元估值时代的纵横捭阖，又穿插着“烤土豆当暖手宝”式的草根忆苦思甜，听感既宏大深邃又充满鲜活的江湖人情味。 ## Bulletin - [01:05](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=65s) 由纽约市长佐兰·曼达尼背书的民主社会主义者（DSA）在纽约国会初选中全胜横扫，包括拉下资深建制派众议员，显示极左翼正通过低投票率动员与移民/高知选民联盟系统性颠覆传统民主党控制权。 - [45:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2700s) 智谱 AI 推出拥有 7,440 亿参数及 100 万上下文的 GLM-5.2 开源模型，在 SWE-bench 评测中击败 GPT-5 且紧咬 Claude 4 Opus，并在华为昇腾 910B 集群上完成训练与推理优化，推动中国“AI in a box”低成本全栈出海。 - [01:01:46](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3706s) 美光季度营收同比暴增 4 倍达 420 亿美元且 HBM 产能全线售罄，内存瓶颈推动 AI 通胀席卷消费电子硬件，苹果 MacBook 与 Mac Studio 售价普遍上涨 14% 至 25%。 - [01:11:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4305s) 算力基础设施呈现分化：地面 1GW 数据中心建造成本受审批与变压器短缺影响逼近 600-700 亿美元，而依托 SpaceX 星舰复用的轨道算力成本可压低至 400 亿美元，同时推理端正通过 Pre-fill 与 Decode 的解耦将 GPU 使用寿命延长至 7-12 年。 - [01:27:22](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5242s) 超级 IPO 潮引发资本博弈，加文·贝克预测 Anthropic 凭借超千亿美元年化收入及 85% 推理毛利率市值可达 3 万亿美元；而 Cerebras 财报后因机构跌破发行价机械止损与供应链放量滞后惨遭做空破发。 ## Key takeaways - [01:05](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=65s) 纽约国会民主党初选中，DSA 支持的候选人以全胜战绩击败丹·高德曼等传统建制派，暴露出高学历富裕知识分子与未同化移民组成的极左联盟正在架空民主党传统工薪阶层基础。 - [03:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=180s) AI 被查马斯视为人类史上最伟大的经济平等推进器，能让普通人即刻拥有顶级联合创始人级别的智慧与行动力，但科技界的末日论与政治内讧制造了叙事真空，反噬了资本主义的公信力。 - [06:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=360s) DSA 纲领代表着对美国传统宪政的颠覆（废除参议院、选举人团与监狱体制），其高层明确表态仅将民主党作为获取选票准入的工具载体，进入体制后推动反宪政的激进议程。 - [14:48](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=888s) 特拉维斯·卡兰尼克提出“真相与正义是社会的免疫系统”，人皆有渴望不劳而获的共产主义惰性，一旦体制免除怠惰代价并包庇谎言，社会免疫机制便会迅速崩溃。 - [20:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1200s) 公共职能大量外包非政府组织引发了典型的“柯利效应”制度化腐败，加州与纽约翻倍投入流向年薪数十万美元的 NGO 官僚，街头无家可归等社会危机却持续加剧。 - [24:10](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1450s) 五眼联盟国家推进 16 岁以下社交媒体禁令引发激烈交锋：支持者认为能斩断极左短视频民粹动员，反对者则直指这是以保护儿童为幌子剥夺匿名权、推行全网实名审查的特洛伊木马。 - [33:20](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2000s) 巴以问题成为民主党内部不可弥合的致命代际断层，纽约第十选区中兰德借批评加沙行动击败坚定挺以的高德曼，50 岁以下选民对以色列的传统支持基础已经彻底瓦解。 - [47:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2820s) 下游开源模型正通过大规模设备集群伪装并发抓取前沿 API 的思考轨迹完成“模型蒸馏”，但 GLM-5.2 已跨过自主强化学习的临界点，未来企业将走向由闭源模型担任把关指挥家的“LLM 委员会”组合架构。 - [51:40](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3100s) OpenAI 联手博通自研芯片打破了与英伟达的默契，而中国在面对制裁下通过智谱与华为昇腾 910B 软硬一体化闭环，正在全球以极致性价比输出“AI in a box”解决方案。 - [01:05:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3900s) 高带宽内存（HBM）垂直堆叠制造壁垒极高，AI 服务器的狂热需求挤占了全球晶圆产能，导致明年超大规模云厂商 30%-40% 的资本支出将流向 DRAM，迫使消费电子全线承受“AI 通胀”。 - [01:11:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4305s) 地面数据中心因电网短缺、变压器匮乏及审批抗议导致建造成本逆向飙升至 600-700 亿美元，而依托星舰发射的轨道虚拟数据中心在电力和散热上具备极强通缩优势，倒逼算力上天。 - [01:14:48](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4488s) 特斯拉“Megapod”商标揭示了数据中心模块化集装箱化趋势，各大云厂商为解算力枯竭被迫妥协放宽设计准则（BOD），将机房建设周期从数年大幅压缩至 90 天。 - [01:23:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5025s) 大模型推理在底层发生架构解耦：预填充（Pre-fill）依赖显存容量，解码（Decode）依赖显存带宽；将 Groq 或 Cerebras 解码芯片与老款 GPU 结合，可将服务器使用寿命大幅延长至 7 到 12 年。 - [01:30:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5400s) Cerebras 财报后破发折射出公募基金“破发即机械抛售”的严苛交易纪律，且其与 OpenAI 签订的百亿美元晶圆订单需要经历台积电制造、整机集成和通电并网的漫长供应链周期，营收放量存在滞后性。 - [01:35:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5700s) 超大规模独立算力企业的真正核心瓶颈在于搞定电力并网的速度；而面对独角兽海量股权融资，投资人推崇以“荷兰式拍卖”对抗投行的人为定价失真。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=0s) 特邀嘉宾入场与纽约初选“社会主义横扫” 全美顶级播客《All-In Podcast》迎来第 278 期。本期固定主持之一的 David Friedberg 在经历纽约市民主党初选由社会主义派大获全胜的政治震荡后，“请了一天心理健康假（mental health day）”。节目组为此邀请了两位重磅嘉宾：Adams 创始人、前 Uber 联合创始人 Travis Kalanick，以及刚结束在得州星舰基地（Starbase）辉煌一周、沉浸在 SpaceX 成功上市喜悦中的 Atreides Management 创始人 Gavin Baker。Gavin 坦言，SpaceX 的 IPO 是数十载艰辛付出的终极成果，但也借用名帅比尔·贝利奇克（Bill Belichick）的名言“奔向辛辛那提（on to Cincinnati）”以及纽约尼克斯队系列赛“从零比零重新开始”的心态，强调胜利后必须立刻专注下一个目标。 随后，主持人 Jason Calacanis 将焦点引向周二刚刚尘埃落定的纽约州国会民主党初选：由纽约市长佐兰·曼达尼（Zohran Mamdani）背书的三位民主社会主义者（DSA）候选人以“三投三中”的全胜战绩横扫对手，彻底震动了华盛顿和华尔街： - **纽约第 10 选区（NY-10）**：市主计长布拉德·兰德（Brad Lander）击败了两届现任众议员、知名建制派丹·高德曼（Dan Goldman）。该选区汇聚了纽约最富庶的街区，包括西村（West Village）豪华排屋区、华尔街金融区，以及布鲁克林的丹波（Dumbo）、科布尔山（Cobble Hill）、卡罗尔花园（Carroll Gardens）和公园坡（Park Slope）。 - **纽约第 13 选区（NY-13）**：DSA 候选人雪瓦利埃（Chevalier）爆冷掀翻了由众议院少数党领袖哈基姆·杰弗里斯（Hakeem Jeffries）全力支持的五届资深众议员阿德里亚诺·埃斯佩拉特（Adriano Espaillat）。该区覆盖哈莱姆区（Harlem）与西布朗克斯（West Bronx），是纽约最贫困的选区之一。 - **纽约第 7 选区（NY-7）**：克莱尔·瓦尔迪兹（Claire Valdez）击败了现任议员钦定的接班人，拿下了布什维克（Bushwick）、威廉斯堡（Williamsburg）、长岛市（Long Island City）及绿点（Greenpoint）等被称为“红化走廊（commie corridor）”、充斥着文青与背带裤咖啡师的 DSA 大本营。 Polymarket 数据显示，在选前曼达尼阵营三席全胜的预测概率仅为 26%，这一结果堪称超级冷门。Jason 指出，这群胜选者和曼达尼一样，在年轻、高学历、高收入群体中极受欢迎——“这些人有资本成为社会主义者（folks who can afford to be socialists）”。这正是典型的“富家穷人（rich poor）”现象：腰缠万贯却假装无产，从上世纪 30 年代、50 年代的红色恐慌一路沿袭至今。如果聪明人投身商界就会成为资本家；若聪明人钻进象牙塔学术界，久而久之便拥有了相信“社会主义极具吸引力”的奢侈特权。 ### [03:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=180s) 硅谷的叙事真空与 AI 作为终极平衡器 Chamath Palihapitiya 从更宏观的技术与社会心理维度展开剖析。他直言，资本主义阵营正在“丢掉剧本（losing the script）”，而硅谷本身就是罪魁祸首。科技界不仅没能向公众展现资本主义的希望，反而将行业内部的私怨、内讧、不安全感和公开撕裂暴露无遗，导致科技精英在普通民众心中威信扫地。在科技界制造的叙事真空中，左翼民粹得以趁虚而入，轻易向大众兜售“只要不是当下的资本主义，任何体制都比现状好”的说辞。 Chamath 进一步揭示，在犹他州和纽约等关键国会选区中，选举本质上已演变为一场围绕“AI 发展路线”的公民投票：由 Anthropic 等势力资助的“反 AI / 监管阵营”与 OpenAI 等支持的“挺 AI 阵营”激烈交锋，所幸挺 AI 力量最终险胜保住了阵线。 Chamath 将人工智能定义为“人类有生之年所能遇见的最大的经济平等推进器（the greatest economic leveler）”，并详细拆解了其底层逻辑： - **第一次经济生产力大解锁**：25 年前互联网与谷歌将全人类的所有知识与信息收集整理，通过搜索引擎实现普惠。然而，当时的技术只能提供原始信息，缺失了将海量信息转化为即时“专业技能与智慧”的能力。 - **AI 的根本突破**：AI 不仅吸收了世界知识，更能将其转化为直接行动力。这意味着，地球上的每一个男女老少，都能瞬间拥有一个等同于 Travis Kalanick 级别的“超级联合创始人”时刻陪伴左右——这位伙伴能帮你思考难题、以工程化思维解决瓶颈、智胜商业对手，且没有任何传统阶级守门人能够阻拦这种能力的获取。 然而，科技界却深陷于“AI 毁灭人类、夺走工作、耗尽淡水”的末日论（doomerism）泥潭。Chamath 愤怒地斥责这些全是由少数 AI 竞赛中别有用心的特定势力资助编造的谎言与假信息。正是硅谷这些极不合格的“品牌代言人”，让公众彻底倒向了更具煽动性的社会主义美好蓝图。 ### [06:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=360s) 民主党全面极左化与激进的 DSA 政治纲领 David Sacks 随后接管话语权，指出美国未来的政治格局正在演变成民主党内部的“共产主义/社会主义”对抗共和党的“民粹民族主义”。他强调，决不能把 DSA 当成普通的进步派，他们的政治主张代表着对美国宪政体系的彻底颠覆： ```mermaid flowchart TD accTitle: DSA 激进政治纲领架构图 accDescr: 展现 DSA 在宪政重构、司法与执法体制、经济外交等维度的激进主张 subgraph Constitutional ["宪政体制彻底颠覆"] A["废除联邦参议院 (Abolish Senate)"] B["废除选举人团制度 (Abolish Electoral College)"] C["重构行政与最高法院<br/>完全隶属于众议院"] D["推行多党制与比例代表制<br/>全面落地排序复选制 (RCV)"] end subgraph Justice ["废除司法与边境执法"] E["废除监狱与警察体制<br/>(Abolish the Carceral State)"] F["废除 ICE / 边境大赦<br/>坚决反对一切驱逐出境"] end subgraph EconForeign ["经济与外交转向"] G["大型核心企业全面公有制"] H["大幅削减国防部预算 (Defund Dept of War)"] I["激进挺巴勒斯坦立场 (Free Palestine)"] end ``` Sacks 警告，民主党建制派如今已陷入极度恐慌，因为他们彻底丧失了对党内权力的控制，党权正在落入佐兰·曼达尼及其盟友手中。以纽约第 13 选区的惨败为例，被击溃的现任议员阿德里亚诺·埃斯佩拉特是国会西裔党团主席、资深重臣；而战胜他的竟然是一位 32 岁、在校读博 10 年却从未上过一天班的待业博士生雪瓦利埃。Sacks 列举了其骇人听闻的过往激进言行： - 公然宣称要“终结西方文明（eradicate Western civilization）”； - 曾在社交媒体宣称自己用美国国旗当餐巾纸擦手； - 在 2023 年 10 月 7 日哈马斯袭击以色列平民的惨剧发生仅仅一天后，便走上街头参加集会公开庆贺针对平民的屠杀； - 公开辱骂白人女性为“丑陋的殖民者（ugly colonizers）”； - 坚决主张废除边境、废除全部监狱，甚至多次公开痛骂警察为“猪（pigs）”。 面对这种甚至远超传统 DSA 极端底线的激进分子堂而皇之赢得国会初选，主持人们惊叹如今的政治极化已进入完全失控的新阶段。 ### [10:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=600s) 民主社会主义者（DSA）的借壳夺权与“后美国”意识形态 在谈及纽约州民主党初选中的极左派候选人时，节目讨论了其社交媒体上的激进言论——不仅曾公开将美军士兵斥为“战争罪犯”，宣称美国是一个“耻辱的国家”，还公开发文赞美共产主义并呼吁剥夺生产资料。这种思潮正逐渐演变为民主党内不可忽视的新现实。即便民主党在即将到来的 11 月中期选举中夺回众议院，哈基姆·杰弗里斯（Hakeem Jeffries）顺利出任众议院议长，如何管理这一批新当选的美国民主社会主义者（DSA）成员也将成为他的巨大梦魇。DSA 绝未将民主党建制派视为同盟，而是将其看作前行路上的绊脚石。正如 DSA 联席主席乔希·布洛赫（Josh Bloch）所言：“我们只是把民主党当作获取选票准入资格的工具载体，并非因为我们认同其目标。我们建立自己的组织，借用民主党的名义参选，在有利可图时与民主党组成党团，并在其内部强推我们自己的议程。在我们眼中，民主党建制派不是归宿，而是障碍。” 面对这种被蚕食的现状，嘉宾直言建制派民主党人完全是“玩愚蠢的游戏，得愚蠢的报应（play stupid games, win stupid prizes）”。正是建制派推行的大规模非法移民与宽松边境政策，为 DSA 输送了庞大的选民基本盘。以佐兰·马姆达尼（Zohran Mamdani）在纽约市长选举中的胜利为例，如果仅由纽约本土出生的本地选民投票，他绝无胜选可能，正是大量涌入的移民选民彻底逆转了选举走势。目前 DSA 的核心联盟由两股力量拼凑而成：一边是庞大的外来移民群体，另一边则是“过度受教育”且面临向下流动的白人激进进步派。这些人顶着高学历光环，却大量流向学术界和非营利组织（NGO），充当极左意识形态的急先锋。这两者的结合直接掏空了传统民主党的控制权。 更深层的危机在于美国社会同化机制的崩塌。过去数十年来维系国家凝聚力的“大熔炉”机制，已被多元文化主义和觉醒文化（wokeism）彻底打碎，国家不再要求移民主动融入美国体制。其结果就是催生了像舍瓦利耶（Chevalier）这类公开发表激进言论的政客。他们不仅是反美的，更是“后美国的（post-American）”，对美国的宪政秩序、自由企业制度毫无敬意，试图照搬移民原籍国的政治制度来重构美国。如果一个国家持续引入海量移民、拒绝进行文化制度同化，并任由马克思主义意识形态领袖进行政治动员，最终美国的社会体制将不可避免地沦落为这些移民所逃离的原籍国模样。 ```mermaid flowchart TD accTitle: DSA 权力崛起与选民联盟模型 accDescr: 展示 DSA 如何利用移民与受过高等教育的精英结合，通过民主党选票机制夺取政治话语权。 subgraph Base["极左翼新生代基本盘"] A["大规模外来移民<br/>（缺乏制度同化）"] B["向下流动的名校精英<br/>（流向高校与 NGO 机器）"] end subgraph Strategy["政治渗透路径"] C["借用民主党竞选外壳<br/>（获取选票准入机制）"] D["内部路线分歧<br/>（视民主党建制派为阻碍）"] end subgraph Consequence["体制性冲击"] E["后美国意识形态崛起<br/>（解构宪政与自由企业制度）"] F["传统工薪选民基础解体"] end A --> C B --> C C --> D D --> E D --> F ``` - **乔希·布洛赫核心表态**：明确 DSA 只是将民主党作为“获取选票准入的载体（ballot access vehicle）”，进入体制后推行激进议程，视建制派为障碍而非同路人。 - **大熔炉机制破裂**：多元文化主义消解了传统的国家同化政策，使得部分候选人走向反宪政、反自由市场与“后美国”叙事。 --- ### [12:55](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=775s) 机构解构之辩：建制拆除、USAID 与拉美右倾潮 针对激进派议员试图拆除大量既有政府架构的动向，现场展开了一场关于左翼与右翼“拆除机构” playbook 的对比。主持人指出，左翼试图全面裁撤各类职能部门，在形式上看似与唐纳德·特朗普（Donald Trump）扬言废除国税局（IRS）、美国国际开发署（USAID）、教育部（Department of Education）以及环境保护署（EPA）的做法类似。但嘉宾随即反驳并强调两者的本质区别：特朗普绝无意废除美国宪法，也无意拆解西方文明的根本基石，两者存在根本性的界限。 话题随即聚焦到被特朗普列入废除目标的 USAID（美国国际开发署）上。嘉宾指出，USAID 长期以来充当着中情局（CIA）推进各类涉嫌干涉主权国家内政与左翼政权更迭行动的前沿阵地。近期南美洲接连有七场大选由右翼阵营获胜，一个不容忽视的地缘政治背景就在于 USAID 资金断供与政治黑钱（payola）的隐退。当外部非政府组织的干预渠道和政治酬庸激励机制被切断后，拉美各国的政治生态便迅速脱离了人为操控的左倾轨道，自发向右摆动。 --- ### [14:48](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=888s) 特拉维斯·卡兰尼克：真相与正义是免疫系统，人皆有“共产主义惰性” 话题转向嘉宾特拉维斯·卡兰尼克（Travis Kalanick）。主持人调侃特拉维斯曾将阿兰·兰德（Ayn Rand）代表个人主义与卓越精神的名作《源泉》（*The Fountainhead*）作为推特头像长达一二十年。特拉维斯澄清这纯属外界夸大其词，自己只是当年每读完一本书就轮换头像，该头像实际只挂了约六个月，后续还换过《汉密尔顿》（*Hamilton*）和《安德的游戏》（*Ender's Game*）。但谈及如今部分地区集体主义思潮泛滥对个人主义的蚕食，特拉维斯用两个精辟的警句给出了剖析： 首先，特拉维斯提出：“真相与正义是社会的免疫系统。当免疫系统被抑制时，所有社会顽疾都会爆发（Truth and justice is the immune system for society. When the immune system is suppressed, all the social ills flare up.）。”无论是主流媒体还是社交媒体对事实的遮蔽，都是社会恶化的早期警报；同样，如果在司法层面人们违法犯罪却毫无后果，社会机能便会走向崩塌。要判断一个文明未来的走向，只要观察“当下的真相是在赢还是在输，正义是在赢还是在输”。现场以新冠疫情期间安东尼·福奇（Anthony Fauci）引发的争议为例，指出公众苦苦等待了近五年，才有一小部分真相被逐步揭露，正义依然遥遥无期。 随后，特拉维斯提出了另一个让许多人感到意外的洞察：“共产主义潜藏在我们每个人的骨子里，存在于我们人类的基因中。”面对众人的疑惑，特拉维斯反问：“你一生中是否曾偷过懒？你一生中是否曾想过不劳而获（something for nothing）？”每个人内心深处都有怠惰和坐享其成的欲望。健康社会与病态生态的区别，在于绝大多数人不会将其作为终极生存法则。然而，一旦外部生态系统开始奖励不劳而获、免除怠惰的代价，这种思维模式就会迅速聚集起临界质量，最终吞噬整个制度。 - **社会免疫指标**：以“真相”和“正义”作为评估文明健康度的两大原子指标；一旦犯罪缺乏代价、谎言不受惩罚，社会免疫即宣告崩溃。 - **人性惰性与制度临界点**：认为对“不劳而获”的渴望是普世的人性弱点，当制度开始免除怠惰的后果时，共产主义式思维便会形成不可逆的扩张生态。 --- ### [17:10](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1030s) 加文·贝克：向下流动的精英阶层与生产力背离的 NGO 机器 加文·贝克（Gavin Baker）从宏观经济与阶层动态的视角切入讨论。他首先指出，人工智能（AI）无疑将成为即将到来的中期选举以及下一届总统大选的核心决定性政治议题。回顾自己的政治谱系，加文提到自己大半辈子都是民主党人，过去的传统民主党曾代表工薪阶层，致力于为底层创造向上流动的经济机会，消除不平等，为黑人与拉美裔争取权益。尽管人们可能对过去民主党的方法论存有分歧，但其初心依然具备崇高性。 然而，今天的 DSA 已经彻底背离了这些目标。从选民人口统计学来看，DSA 的选民基础是一群“相对富裕却面临向下流动（downwardly mobile）的白人自由派”。数据清晰地表明：DSA 正在系统性地失去工薪阶层选民、失去底层贫困选民、失去黑人选民以及拉美裔选民。民主党长期以来赖以立足并试图代表的传统草根群众，在 DSA 的政治光谱中早已无足轻重。加文痛心疾首地指出，选出那些极端反美、甚至曾公开呼吁通过暴力摧毁美国制度的候选人，是当代政治的巨大悲剧。 加文深入剖析了这一激进阶层的成因：这是一整代毕业于名牌高校的精英子弟，他们成长于优渥富足的家庭环境，毕业后却没有进入创造财富与价值的工业实体，而是成群结队涌入庞大的 NGO、非营利组织和官僚机构中。这些人的产出不仅极度匮乏，其生活轨迹也与那些脚踏实地创造社会价值的人彻底脱节。加文总结道：“埃隆·马斯克（Elon Musk）为地球脱碳事业所做的实际贡献，比全世界所有环保活动家加起来还要多上十倍。”正是这种生产力的巨大反差，揭示了当下激进政治背后的荒谬与虚伪。 - **阶层倒错的政治联盟**：DSA 的实际狂热支持者集中在面临资产缩水、向下流动的名校白人知识分子阶层，而非传统产业工人和少数族裔。 - **生产性实体与寄生性 NGO 的对立**：精英阶层脱离实体生产、寄生于庞大 NGO 机器，造成激进意识形态与现实生产力创造严重脱节。 ### [20:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1200s) 非政府组织寻租与“柯利效应”：数千亿资金与恶化的公共治理 嘉宾们深入探讨了政府将大量公共职能和资金外包给非政府组织（NGO）所带来的深层腐败。这种运作模式完美契合了政治学上的“柯利效应”（Curley effect）——源自历史上波士顿市长的执政策略：推行明知会对选区整体经济造成灾难性后果的政策，借此将政治对手和纳税中产驱逐出境，随后再将年薪高达 60 万美元、毫无实际产出的 NGO 主管职位分肥给自己的政治盟友与亲信。这种操作本质上是规模极其庞大的“制度化腐败”。相比于特朗普政府接受卡塔尔私人飞机等放在台面上引发争议的事件，全美范围内流向各类 NGO 的资金高达到数百亿甚至数千亿美元，却几乎处于监管盲区。 在加利福尼亚州和纽约州，这一问题表现得尤为触目惊心。过去几年中，加州和纽约在解决无家可归问题上的人均财政支出增加了一倍乃至翻了两番，几乎全部由 NGO 承包运作，然而街头的实际状况和治理成果却以肉眼可见的速度在持续恶化。嘉宾们指出，美国民主社会主义者（DSA）等极左翼势力的扩张，核心根源并非在于其理念有多么科学可行。其主张在传统上仅能吸引极其狭窄的群体——那些自视受过良好教育却在经济上向下流动的富裕白人群体，他们身处某种自视清高的虚假贫困中，心怀善意却脱离实际生产。 - 治理恶化机制：加州在无家可归问题上的人均支出翻了四倍，资金悉数流向 NGO，街头危机却愈演愈烈。 - 官僚分肥体系：以解决社会问题为名设立的 NGO 高管年薪动辄达 60 万美元，形成依附于极左政治体制的食利阶层。 --- ### [21:40](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1300s) 左翼力量的崛起：佐兰·马姆达尼的政治魅力与“零度可乐版共产主义” DSA 之所以能在政坛迅速崛起，关键因素在于推选出了极具政治天赋的核心人物——佐兰·马姆达尼（Zohran Mamdani）。嘉宾们评价他是“这辈子见过的最具天赋的政治家之一”。马姆达尼不仅擅长发表极具煽动性的公开演讲，在面对媒体专访时对答如流，更像一只极具适应力的变色龙，能够精准捕捉大众情绪。相比之下，曾经被视为民主党新生代招牌的 AOC（亚历山大·奥卡西奥-科尔特斯）在马姆达尼面前甚至显得“只是一场热身赛”。民主党建制派目前在全美范围内几乎找不出任何一位在个人魅力和传播能力上能与马姆达尼抗衡的候选人。 ```mermaid flowchart TD accTitle: 青年激进化与社会主义思潮上升路径 accDescr: 展示现实经济困境如何与魅力型政客结合，通过社交媒体将青年推向极左思潮 subgraph Reality["现实经济困境"] A["高房价与租金飞涨"] B["沉重的高校助学贷款"] C["医疗危机与全面通胀"] end subgraph Discontent["代际绝望情绪"] D["两代年轻人深感被剥夺<br/>认定经济体制被操控"] end subgraph Catalyst["极左政治催化"] E["佐兰·马姆达尼 / AOC 等魅力型政客"] F["社交媒体算法扩音器"] end subgraph Ideology["极左思潮包装"] G["'民主社会主义'<br/>（零度可乐版共产主义）"] H["没收企业股权与私人财富"] end Reality --> Discontent Discontent --> Catalyst Catalyst --> Ideology ``` 这种政治魅力恰好击中了美国正在形成的“两代失落的年轻人”。他们确信自己的经济状况将远逊于父辈：面对飞涨的住房成本、难以摆脱的学生贷款、昂贵的医疗体系和失控的通胀，青年一代认定现有的资本主义体制已经被既得利益者彻底操控，自己根本无法从中分得一杯羹。由于他们对苏联历史或冷战时期的共产主义灾难毫无概念，DSA 巧妙地包装出“民主社会主义”的概念。嘉宾讽刺道，这就像是“社会主义界的零度可乐（Coke Zero）”——听起来毫无负担，剥开包装其核心依然是传统的共产主义，主张直接剥夺他人资产、没收一半以上的企业股权与财富。而建制派在过去几年里不仅未能解决住房供给、教育医疗和地缘无休止冲突（包括对以色列的争议性支持），反而让现实矛盾更加激化。 - 代际剥夺感：年轻一代面临高房价、天价助学贷款与通货膨胀，对传统体制的参与感与信任度彻底破裂。 - 意识形态包装：“民主社会主义”被视为激进共产主义的温和伪装，其核心依然是通过国家权力强行剥夺生产者的合法财富。 --- ### [24:10](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1450s) 社交媒体禁令之辩：查马斯论“青年去激进化”与五眼联盟的社会实验 查马斯分享了他此前在梅根·凯利（Megyn Kelly）节目中阐述的宏观观察。他指出，在意识形态滑向社会主义方面，加拿大、英国和澳大利亚这三个国家实际上远远走在了美国前面。剥离掉冠冕堂皇的政治修辞，从任何客观衡量治理成果的指标来看，这三国的现状都堪称“一场彻头彻尾的灾难”。长年累月在社会议题、开放边境移民政策以及激进气候变化议题上的“道德绑架式表态（virtue signaling）”，导致这三个国家深陷社会撕裂、严重内耗与政治动荡的火药桶状态。 然而，这三个国家近期出现了一个共同的重要转折点：全面出台法令禁止 16 岁及以下青少年使用社交媒体。查马斯认为，马姆达尼、AOC 以及刚刚胜选的极左翼政客舍瓦利耶（Chevalier）等人，无一例外外表出众、极富魅力、紧扣时代情绪，并且精通社交媒体算法传播，实质上是在网上“精心培育一支青年狂热军队”。切断青少年接触社交媒体的渠道，就如同防止儿童过早接触成瘾性毒品。剥离了算法投喂的极端信息流，青年群体被迫重回均衡的信息消费环境，步入成年投票年龄时的激进化倾向将大幅减弱。查马斯指出，佛罗里达州已经通过了类似立法，全美各州都应当严肃考虑对 16 岁以下群体实施社交媒体禁令，以恢复长期的政治稳定。 - 制度先例：加拿大、英国与澳大利亚相继推进 16 岁以下社交媒体禁令，试图从信息源头遏制青少年群体的激进化。 - 政治传播截流：极左翼政治力量高度依赖社交媒体短视频与算法裂变，禁令将直接切断其最重要的政治动员渠道。 --- ### [26:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1605s) 特拉维斯与加文的反驳：实名制审查的特洛伊木马与言论自由的底线 特拉维斯·卡兰尼克（Travis Kalanick）对查马斯的观点提出了坚决反对。他强调，虽然所有人都承认过度沉迷社交媒体会导致成人和儿童的“大脑腐蚀（brain rot）”，危害堪比甚至超过香烟，但以保护儿童为由出台禁令，其实是当权者设下的特洛伊木马。这种基于年龄的限制，核心目的根本不是保护儿童，而是为了强制成年人进行数字实名认证（de-anonymize），从而在全社会建立起滴水不漏的全面审查体制——英国政府目前的一系列立法构想正是这一方向。特拉维斯尖锐指出：“审查从来不是为了过滤有害内容，而是为了封杀让当权者不爽、反对当权者的言论。在统治者眼里，只要支持他们就不是有害内容，反对他们就必须定罪。” ```mermaid flowchart LR accTitle: 社交媒体禁令背后的审查机制风险 accDescr: 特拉维斯指出年龄限制背后的真正目的是强制实名制与审查 A["表面理由：<br/>保护16岁以下儿童远离毒害"] --> B["技术推论：<br/>全网用户强制年龄核验"] --> C["实质后果：<br/>剥夺成年人匿名权（强推实名认证）"] --> D["终极风险：<br/>建立全面审查制度，将异见定罪化"] ``` 加文·贝克尔（Gavin Baker）对特拉维斯的警示表示高度赞同。加文强调，欧盟和西方当权者最迫切希望铲除的，正是 X 平台上那些让官僚阶层头疼的匿名批评者；失去匿名性和言论自由，整个西方世界会滑向更为黑暗的深渊。加文指出，人类历史的悲剧在于人们总无法从过往经验中吸取教训，尽管共产主义在不同文化、不同机制下的每一次试验都以绝对的失败、死亡和苦难收场，但每一代人似乎都想重新试验一次。唯有依托真正的言论自由，公众才能看清极左翼与 DSA 政策在现实中的荒唐后果： - 治安与人命：权威研究数据明确证实，一旦城市选出共和党籍地方检察官（DA），该地区年轻黑人男性的全因死亡率立竿见影下降 7%。 - 环保反讽：极左进步派设立了极其繁杂的官僚审批和环保法规，导致清洁光伏项目根本无法获批落地，阻碍了能源向阳光转型的步伐。 - 教育倒退：极左派打着平权旗号取消公立精英学校选拔机制，甚至在中学课程中砍掉代数与微积分，彻底剥夺了底层贫困家庭子弟依靠个人努力实现阶级跃升的机会。 ### [30:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1800s) 极少数惯犯推高犯罪率与 DSA 的基层选举渗透 讨论由城市治安的残酷现实切入：治安恶化对普通大众造成了极其恶劣的伤害，而统计数据清晰地揭示了暴力犯罪的根源。绝大多数严重暴力事件并非随机分布，约 75% 的暴力犯罪实际上是由拥有 10 次以上前科的极少数群体制造的，甚至有统计指出“0.1% 的核心人群犯下了 70% 的犯罪行为”。如果司法系统能够针对这 0.1% 的惯犯施加切实管束，城市治安本可以大幅改观。然而，现实中的放任正在引发深刻的历史忧虑：美国建国以来凭借地缘和经济体制维系了极高的稳定性，二战后甚至拥有独步全球的支配力，但如果激进派别彻底掌控核心城市、将最具创造力的生产力人群全部挤走，这些城市将面临无法逆转的衰退深渊。 这种激进治理的蔓延并非孤例，而是美国民主社会主义者（DSA）全国性组织运作的战果。在洛杉矶，DSA 已经渗透并掌握了市议会近半数席位；在纽约，由于该州属于深蓝州，初选选民投票率仅在 17% 左右。DSA 恰恰善于利用这种极低投票率的生态：依靠狂热的意识形态凝聚力、高强度的线下动员组织，以及对“选票收集”（ballot harvesting）和排名选择投票（Ranked-Choice Voting）等制度规则的熟练钻营，DSA 屡屡以小博大。佐兰·马姆达尼（Zohran Mamdani）俨然已成为这一运动的精神旗帜。在如今这批激进年轻派眼中，甚至连 AOC（Alexandria Ocasio-Cortez）或伯尼·桑德斯（Bernie Sanders）都已被打上了“妥协叛徒”（sellout）的标签，他们的极端程度远超老一代左翼。 更有趣的是，DSA 吸收了特朗普式的“大帐篷”（big tent）策略。伯尼·桑德斯、佐兰·马姆达尼以及罗·康纳（Ro Khanna）等人虽然政见存在细微差异，却选择在同一个大帐篷下抱团取暖。建制派领袖为了汲取这股基层动能，不得不主动屈膝迎合。在刚刚结束的国会初选中，马姆达尼支持的候选人一举拿下三场关键胜利，其中两场直接将老牌重量级现任议员拉下马。这种巨大震荡向所有深蓝选区的民主党议员释放了严厉警讯：如果不主动在议案表决、政治表态和演说措辞上向 DSA 靠拢，任何建制派政客都可能在下一次初选中面临灭顶之灾。 - **犯罪集中度**：全社会约 75% 的暴力犯罪由有 10 次以上前科者犯下；另有数据表明仅 0.1% 的极少数人群对 70% 的犯罪负有直接责任。 - **低投票率杠杆**：纽约初选投票率仅约 17%，极低参选率成为 DSA 凭借高组织度翻盘的温床。 - **初选威慑效应**：DSA 候选人连续击倒两位实力雄厚的现任国会议员，倒逼民主党温和派全盘左倾。 --- ### [33:20](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2000s) 纽约第十选区震荡：以色列议题成为民主党致命分水岭 在纽约第十国会选区（NY-10）的初选中，民主党内部最深刻的裂痕暴露无遗——这便是关于以色列与巴勒斯坦的态度立场。长期以来，两届连任的现任国会议员丹·高德曼（Dan Goldman）地位稳固，他曾担任特朗普首次弹劾案的民主党首席法律顾问，在各大主流有线新闻台拥有极高出镜率，几乎完美符合传统进步派的各项考核指标。然而，高德曼却在本次初选中遭遇滑铁卢，以惨败告终，击败他的正是纽约市审计长布拉德·兰德（Brad Lander）。 这场对决的真正分野纯粹聚焦在外交立场上。高德曼与兰德均为白人犹太裔从政者，背景极其相似，但在以色列政策上截然对立：高德曼坚定支持以色列，为其过去数年的军事行动全力辩护；而兰德则直接走进清真寺，公开发表演说谴责以色列在加沙犯下“种族灭绝”（genocide）。这使得该选区的对决演变成一场关于以色列政策的纯粹公投，最终兰德以明显优势轻松胜出。这标志着 DSA 纲领中“解放巴勒斯坦”的诉求，已经从大学校园抗议外溢为能够直接摧毁民主党建制派老将的决定性选举力量。 ```mermaid flowchart TD accTitle: 纽约第十选区初选选情分化图 accDescr: 展示丹·高德曼与布拉德·兰德因以色列立场差异而导致的选举结果走向 subgraph Candidates ["NY-10 候选人背景（均为犹太裔进步派）"] DG["丹·高德曼 (Dan Goldman)<br/>两任国会议员 / 特朗普弹劾案总法律顾问"] BL["布拉德·兰德 (Brad Lander)<br/>纽约市审计长 / DSA 核心盟友"] end subgraph Stances ["核心议题分歧：以色列与巴勒斯坦政策"] ProIsr["坚定挺以立场<br/>全面辩护军事反恐行动"] AntiIsr["激进批以立场<br/>赴清真寺谴责加沙行动为种族灭绝"] end subgraph Outcome ["党内初选决选结果"] Defeat["高德曼意外惨败<br/>传统建制派防线失守"] Victory["兰德大获全胜<br/>DSA 动能主导深蓝选区"] end DG --> ProIsr --> Defeat BL --> AntiIsr --> Victory ``` --- ### [35:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2100s) 以色列议题的代际断层：从两党共识到全美民意撕裂 民调数据的剧烈逆转印证了这一政治现实：目前高达 80% 的民主党受访者明确表示不认同以色列的行为。在过去数十年里，支持以色列是华盛顿两党高度统一的政治禁忌与不言自明的共识，如今这一共识彻底崩塌。大学校园内爆发的持续抗议浪潮展现了年轻一代民主党人对该议题的极度执念，巴以问题已成为调动年轻选民参与初选投票的核心推力。 与此同时，撕裂也在共和党内部激烈上演。塔克·卡尔森（Tucker Carlson）因此议题与传统共和党决裂，梅根·凯利（Megyn Kelly）情绪激烈，白宫和保守阵营内部同样暗流涌动。这一现象的本质并非单纯的左右党争，而是深刻的**代际断层（Age Gap）**： - **50 岁以上选民**：无论是建制派共和党人还是福克斯新闻（Fox News）的传统受众，依然对以色列保持着极高的传统支持度与好感。 - **50 岁以下选民**：新生代早已抛弃传统电视媒体转向播客与社交网络，其对以色列的不满情绪呈现压倒性态势；在 50 岁以下的保守派受众中，对以色列的反对比例已经飙升至 57%。 这种代际更迭意味着，以色列在美国政治中“神圣不可侵犯”的共识基础正在加速瓦解，未来两党的对以政策必将面临系统性重构。 --- ### [36:30](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2190s) 9/11 创伤类比：内塔尼亚胡、以色列国与犹太群体的逻辑混淆 面对日益极化的社会舆论，查马斯（Chamath Palihapitiya）重申了其在接受梅根·凯利专访时强调的核心观点：公众必须在逻辑上对四个维度进行清晰的解耦与剥离——即**犹太人个体**、**以色列国**、**生活在以色列的普通民众**，以及**比比·内塔尼亚胡（Bibi Netanyahu）政权**。当前舆论场最大的悲剧在于将这四者机械地捆绑在一起，将对个别领导人的愤怒无差别宣泄到整个族群和宗教之上。 查马斯借用美国现代史展开了深度类比：正如 2001 年的“9·11 事件”一样，2023 年“10·7 袭击”中对本土的突袭与平民屠杀，给整个国家带来了无以复加的情感摧残、心理创伤与现实剧痛。在国家遭遇灭顶之灾的至暗时刻，民众通常会将无限权力让渡给当权领袖以求迅速纠偏复仇。若非经历 9/11 的巨大震撼，《爱国者法案》（Patriot Act）这种在常规环境下绝无可能面世的严苛法案绝不可能顺畅通过。领导人在历史危机时刻被推上风口浪尖行使复仇授权，这种动能完全可以被理解。 然而，战事演变至今，内塔尼亚胡已经陷入“彻底失控”的普遍共识之中。对比比个人的强硬政策提出尖锐批评、呼吁其适时止步、乃至敦促以色列内部完成领导层更迭以重塑其国际声誉，都是完全合情合理的诉求。然而，世界舆论却陷入了一种病态的“全称归咎”： > “以色列人民是令人赞叹的，犹太人是令人敬佩的。但把所有概念强行折叠在一起，因为一个人（内塔尼亚胡）的行为，就将其全盘迁怒到整个国家和整个宗教身上，这完全是不可理喻的疯狂。” --- ### [38:40](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2320s) 加文·贝克：以色列在公关战场上的全面溃败 加文·贝克（Gavin Baker）随即直指核心，指出以色列正遭遇前所未有的公关（PR）危机。在现代信息战与社交媒体的博弈中，以色列政府的表现堪称彻底失能。他认为以色列亟需在关键盟友国部署全新的公关代表团队：他们必须是 35 岁以下、兼具美以双重背景、英语母语级流畅，并且深谙社交媒体传播语境与算法算法逻辑的年轻人；在法国、德国及各主要大国，同样需要这样年轻化、本土化的传播枢纽。 贝克提到一个令人扼腕的现实细节：在 10·7 事件爆发初期，曾有一位极为出色的年轻美以双籍发言人（影射埃隆·莱维 Eylon Levy）在欧美舆论场上表现惊艳，其敏锐犀利的辩才与现代传播技巧为以色列赢得了大量同情与理解。然而，据传仅仅因为内塔尼亚胡的妻子萨拉对其心生不满，这位公关干将就遭到草率解雇，且此后再未找到同等水平的替代者。 如今，以色列在国际舆论场上“每分每秒都在承受重击，却毫无招架还手之力”。以色列官方的惯常做法极其脱节——他们往往推选出一位在以色列本土备受尊崇、年过六旬的退伍战争英雄走向前台。尽管这些老将功勋卓著、英语在字面语法上精准无误，但他们根本不懂当代社交网络的情绪脉搏与受众心理，在快节奏的数字舆论战中被降维打击，导致以色列在西方年轻一代中的话语权全面崩盘。 ### [40:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2400s) 美国教育的激进化、爱国叙事的缺失与“大熔炉”文化 讨论从美国教育系统的问题切入。嘉宾指出，许多在新冠疫情（COVID-19）期间转投特朗普（Donald Trump）的民主党选民，根本原因在于居家隔离期间第一次听到了学校到底在教孩子什么，那种赤裸裸的反美叙事让他们感到极度震惊并被彻底“激进化”。全美教育体系中必须停止灌输反美情绪；虽然美国绝非完美无缺、犯过许多错误，但它依然是人类文明史上“做得最好的国家之一”，应该在每个年级客观且坚定地讲述这一事实。例如奴隶制在东亚某些国家之外曾是世界性的普遍现象，并非美国独有的原罪。正因为现在的孩子从小被灌输“做美国人是可耻的”、“自己的身份认同是可耻的”以及“美国是邪恶的”，他们才会在社交媒体的宣传与政治洗脑面前毫无抵抗力。 美国是人类历史上唯一一个取得显著成功的多元文化社会，也就是传统的“大熔炉”（melting pot）理念。然而在如今的极左文化下，谁要是公开赞美“大熔炉”，就会因为所谓“抹杀他人物质文化”而遭到取消（canceled）。真正的理念应该是保留各自的族群文化，同时融入共同的美国文化。 现场嘉宾随即打趣调侃，提到自己的文化不再是穿 Loro Piana，而是“赢的文化”（culture of winning）——追求学习、进步与冒险，甚至引用电影《塔拉迪加之夜》中瑞奇·鲍比（Ricky Bobby）的经典台词，开玩笑称自己“每天尿出来的都是卓越”（piss excellence）。 --- ### [42:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2520s) 民主社会主义者接管民主党：马姆达尼的“特朗普式”演讲与蛊惑力 对话转向纽约政坛的震荡。民主社会主义者（DSA）正在像寄生宿主一样全面渗透并接管那些无所作为的民主党建制派。嘉宾引用了纽约市 DSA 联合主席古斯塔沃·戈迪略（Gustavo Gordillo）的公开言论：DSA 虽然身处民主党党团，但完全不认同其官僚机制，正试图通过自发组织的方式建立独立性。他们尖锐指出，民主党一边拿着亿万富翁金主的政治献金，一边声称代表工人阶级，而现实是政治必须在亿万富翁阶层与工人阶级之间二选一。 这种基层夺权策略与特朗普当年颠覆共和党如出一辙。建制派当年千方百计想把特朗普踢出局，但特朗普凭借极具穿透力的沟通风格和顶级脱口秀演员般的感染力，彻底俘获了底层选民。如今，极左派代表人物佐兰·马姆达尼（Zohran Mamdani）完全复制了这套打法。马姆达尼展现出类似“奥巴马与特朗普结合体”的超强魅力，正在从内部摧毁民主党。 主持现场还爆发了一场幽默的争论：一位身为纽约尼克斯队（New York Knicks）铁杆死忠的嘉宾坦言，尽管自己对社会主义深恶痛绝，但听到马姆达尼此前发表的一段关于尼克斯队的演讲时，依然被震撼得目瞪口呆，“简直像被催眠了一样”。虽然大家一眼就能看透马姆达尼挂着虚伪的“鳄鱼式微笑”，骨子里推行的是导向痛苦与贫困的共产主义，但他极具煽动性的演说才华依然让人不得不防，甚至其他嘉宾开玩笑说“必须马上对你进行去脑电波洗脑”。 --- ### [45:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2700s) 智谱 AI 推出 GLM-5.2：中国开源大模型全面逼近前沿梯队 ```mermaid flowchart TD accTitle: 模型蒸馏与微调工作流 accDescr: 展示下游开源模型如何通过大规模伪装调用前沿 API 捕获推理轨迹并反哺训练 subgraph FrontierLab ["前沿闭源实验室"] API["Claude / OpenAI API 接口"] end subgraph DistillationFarm ["自动化蒸馏系统 (设备集群)"] Farm["数以万计的手机 / iPad / PC 集群<br/>(伪装多账号高频并发提问)"] Traces["捕获完整 Token 与推理链<br/>(Reasoning Traces)"] end subgraph TargetModel ["开源模型训练 (如 GLM-5.2)"] PreTrain["预训练阶段注入 (Pre-training)"] RL["强化学习对齐 (RL 阶段)"] SelfRL["具备自主高质量强化学习能力<br/>(摆脱单向依赖)"] end API -->|"输出完整推理 Token"| Farm Farm --> Traces Traces --> PreTrain Traces --> RL RL --> SelfRL ``` 话题随后转向人工智能领域的最新重磅事件：中国开源模型正在迅速抹平与美国前沿闭源模型的差距。由智谱 AI（Z.ai）发布的开源模型 **GLM-5.2** 展现出了前所未有的强悍实力： - **基础规格**：拥有 7,440 亿参数（744B），支持高达 100 万 Token 的上下文窗口（1M context window）。 - **完全开源许可**：采用极其宽松的 MIT License，没有任何区域限制、API 锁或使用壁垒，允许全球开发者自由分叉（fork）、自建托管以及完全免费用于商业化搭建公司。 - **评测跑分**：在 Artificial Analysis 智能指数上斩获 51 分，创下全球开源权重（open-weight）模型最高纪录；在权威的软件工程基准测试 SWE-bench 上击败了 GPT-5，落后 Claude Opus 4 不到 1 个百分点。 - **调用成本极低**：在同等能力表现下，API 成本比 GPT-5.5 便宜足足 85%。 - **未来预测**：智谱创始人向埃隆·马斯克（Elon Musk）透露，具备可比前沿能力的下一代开源模型将在 2027 年第一季度甚至更早问世。 这一突破引发了关于“出口管制与自我审查是否已经形同虚设”的讨论。领先 6 个月在长远的人工智能竞赛中是否还具备决定性壁垒，成为核心议题。 --- ### [47:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2820s) 揭秘“模型蒸馏”内幕与未来的“LLM 委员会”多模型架构 针对开源模型为何能如此迅速跟进，加文·贝克（Gavin Baker）援引了一位 TPU 工程师的观点，明确指出这一过程中必然发生了极其庞大规模的“模型蒸馏”（Distillation）。所谓模型蒸馏，本质上类似规模化的作弊小抄（cheat sheet）： 1. 组织者搭建数以万计的 iPhone、iPad 和电脑集群； 2. 利用被掩饰的伪装账号，日夜不停地向 Anthropic 的 Claude 等顶级 API 接口发起海量高复杂度查询； 3. 由于开发者在 API 接口中可以逐字抓取每一个 Token，他们借此完整提取出前沿模型的“思考过程”或“推理轨迹”（reasoning traces）； 4. 这些高质量的推理轨迹随后被作为黄金数据集，重新喂入自研模型的强化学习（RL）乃至预训练（pre-training）流程中，从而以极低成本逼近前沿水平。这就像把谷歌所有的搜索结果全部扒下来再组装一个自己的搜索引擎一样。 但贝克强调，GLM-5.2 已经跨过了一个关键临界点：它自身的综合能力已经足够优秀，完全可以基于自身输出开展独立的强化学习，不必再单纯依赖外部投喂，“猫已经彻底从袋子里放出来了”。 对于未来格局，企业端绝不会押注于某一家单一模型，而是会走向安德烈·卡帕西（Andrej Karpathy）所预言的**“LLM 委员会”（Council of LLMs）组合架构**： - **组合化模式**：企业通常会挑选至少两家前沿模型（例如 Grok、Anthropic、OpenAI 或 Google，其中 Grok 凭借对事实真理的执着、敢于向企业主直言政治不正确但业务关键的真相，应作为常备选项）； - **混合分发**：企业会结合利用自身专有业务数据做过强化学习的开源权重模型。约 50% 到 85% 的常规查询将直接由低成本的开源模型处理，只有最困难、最复杂的极端边界问题，才会上报或由前沿闭源模型做校验把关，从而实现真正的帕累托最优（Pareto dominant outcomes）。 - **商业影响**：开源模型的繁荣并不会摧毁 AI 产业，反而对底层基础设施（算力、芯片、能源等基础设施提供商）是巨大的红利，它只是将前沿闭源实验室的高昂利润率，重新分配回了整个 AI 基础设施生态。 ### [50:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3000s) 可组合模型与企业级路由架构：前沿模型退居“指挥家” 讨论首先聚焦于模型生态的利益分配与未来架构形态。虽然开源模型凭借极高的使用频率已经处理了超过 80% 的 Token，但截至目前，前沿闭源模型依然牢牢摄取了整个行业 90% 的商业经济价值。针对这种结构，嘉宾们提出了“可组合模型”（Composable Models）才是未来企业落地终极形态的论断。为了具体阐述这一概念，现场将话题引向了特拉维斯·卡兰尼克（Travis Kalanick）正在低调推进的新项目。众人打趣他这家“绝密初创公司”，特拉维斯则笑着纠正道：“它不是超级绝密（super secret），而是超级牛逼（super awesome）。”据说不少投资人近期都在热烈打探其进展。 所谓的“可组合模型”在实际业务中的运转，就像一套精密配合的交响乐系统。企业内部通过一个中枢智能路由器（Router），将接收到的用户查询或具体任务精准分发至内部经过强化学习（RL）针对性微调的垂直开源模型中处理。只有在特定工作流的校验、复杂逻辑补充或高难度决策环节，昂贵的前沿模型才会作为最终把关者介入。这种机制将前沿大模型降维为“交响乐团的指挥家”，而不是在每个简单任务上都全盘包揽，极大地平衡了经济成本与推理性能。 ```mermaid flowchart TD accTitle: Composable Model Architecture accDescr: Enterprise query routing to specialized open-source models with frontier models acting as conductors. UserReq["用户请求 / 企业任务"] --> Router["企业中枢智能路由器 (Router)"] subgraph Execution["专门任务执行层"] Router --> RL1["强化学习微调模型 A"] Router --> RL2["开源轻量模型 B"] Router --> RL3["垂直场景专有模型 C"] end Execution --> Judge{"是否需要<br/>高阶验证 / 复杂合成?"} Judge -- 否 --> Output["最终输出结果"] Judge -- 是 --> Frontier["前沿闭源模型 (指挥家/仲裁者)"] Frontier --> Output ``` ### [51:40](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3100s) 芯片与模型的竞合暗战：OpenAI 涉足自研与英伟达的反击可能 话题随后切入大模型公司与硬件巨头之间的微妙角力。针对主持人关于美国开源大模型领头羊何在的提问，现场给出了一个出人意料的答案：英伟达（NVIDIA）才是美国目前真正的开源冠军。凭借其强大的算力储备与工程实力，英伟达完全有能力在任何时候发布媲美甚至超越智谱 GLM 5.2 的顶尖模型。然而，英伟达至今按兵不动，完全是出于经典的渠道冲突（Channel Conflict）考量——黄仁勋极力避免与自己的核心大客户正面夺食，破坏硬件销售的大盘平衡。 然而，大模型巨头对硬件垄断的挑战正在动摇这一默契。随着 OpenAI 联合博通（Broadcom）正式推出代号为“墨西哥辣椒”（Jalapeno）的自研 ASIC 芯片，OpenAI 实际上已经在供应链层面给英伟达亮出了明枪。在埃隆·马斯克自研芯片并发展 xAI 的同时，OpenAI 若将自研算力推向外部客户，将彻底打破平衡。现场嘉宾大胆预测：“如果 OpenAI 执意要自研芯片并对黄仁勋说‘滚蛋’，千万不要惊讶英伟达是否会顺势反扑，亲自下场做一家 OpenAI 的正面对手——这是你在《All-In》节目里第一次听到的预言。” - 前沿闭源模型目前攫取了市场 90% 的经济回报，而开源模型消化了超过 80% 的推理调用量。 - OpenAI 发布自研 ASIC 芯片 Jalapeno，由博通（Broadcom）代工制造，被视为直接摆脱英伟达束缚的激进举动。 - 英伟达受制于渠道冲突暂未全力释放其开源软件模型能力，但若硬件客户持续越界，英伟达拥有直接入局模型竞赛的杀手锏。 ### [53:30](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3210s) 智谱 GLM 5.2 的震撼赶超：监管泥潭与 Dario 的“引火烧身” 现任总统科技顾问委员会（PCAST）负责人的大卫·萨克斯（David Sacks）深入分析了中国在开源模型层面的最新突围。智谱 AI 最新推出的 GLM 5.2 在编码、软件工程以及长上下文智能体（Agent）任务上展现出统治级的表现。在权威的 SWE-bench 基准评测中，GLM 5.2 的得分仅略逊于 Claude 4 Opus，几乎与 GPT-5 平起平坐。这意味着中国开源权重模型已经完全追平了美国上一代前沿闭源模型的水准。 更致命的是美国头部模型公司当下的自我停滞状态。由于 Anthropic 的 Fable 模型因越狱安全隐患被政府紧急叫停，OpenAI 的 GPT-5.6 同样陷入极其繁琐的政府合规审批泥潭，导致市场上可实际调用的最强模型反倒与中国开源模型拉不开差距。萨克斯严词批评了 Anthropic 首席执行官达里奥·阿莫代伊（Dario Amodei）游说政府设立类似“AI 领域的 FAA”的做法，认为其试图通过构筑监管护城河（Regulatory Moat）搞监管俘获，结果反被自己的规则反噬，陷入了典型的自食其果（“Hoisted on his own petard” / FAFO）尴尬境地。 - 智谱 GLM 5.2 在 SWE-bench 等关键编码和软件工程跑分中逼近 Claude 4 Opus，紧咬 GPT-5。 - 萨克斯指出，美国本应全力以赴赢得 AI 竞赛，却因 Anthropic 制造的安全恐慌导致 Fable 被勒令回滚、GPT-5.6 审批受阻，白白给竞争对手赠送宝贵时间。 - 对于网络安全漏洞的真正解法，萨克斯强调必须将最强模型迅速交给白帽黑客，在全行业推动漏洞挖掘与安全补丁升级；单纯依靠冻结模型并不能阻挡对手的技术扩散。 ### [57:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3420s) 华为昇腾与软硬一体化：中国“AI in a Box”出海与算力自主化浪潮 中国大模型在面临严苛制裁的背景下，正在全力加速本土全栈供应链的闭环。中国政府正不遗余力地推动顶尖 AI 实验室在国产硬件上完成开发与训练。尽管外界对部分宣传持保留态度，但此前 DeepSeek-V4 据称已在华为芯片上完成训练，而智谱 AI 更是明确表示其 GLM-5 系列大模型完全依托华为昇腾 910B（Ascend 910B）计算集群训练完成，且 GLM 5.2 的推理阶段已针对华为硬件做了专门的底层适配。 萨克斯指出，这标志着中国正在执行极其强硬的“本土化”（Indigenization）战略，全力将华为打造成国家级芯片冠军。尽管中国在大模型算法上落后美国 6 到 9 个月，先进制程硬件制造上落后约 24 个月，但通过将国产芯片与专门优化的开源模型高度绑定，中国正在打造极具杀伤力的“软硬一体机”（AI in a box）。届时中国厂商将再次祭出经典的“更好、更便宜、更快”打法，以极低的价格在全球范围内输出算力与模型整体解决方案。萨克斯警告，美国政府若继续实施阻碍出口并限制自身企业研发步伐的策略，将在全球市场竞争中彻底丧失主动权。 - 智谱声称其 GLM-5 全系模型完全依托华为昇腾 910B 集群训练，并深度优化了昇腾芯片上的推理性能。 - 智谱创始人透露，中国前沿能力预计将在第一季度前追平 Fable 水准，在特定关键领域（如网络安全）通过集中算力实现快速追赶。 - 中国正将华为芯片与国产开源模型软硬一体打包为“AI in a box”解决方案，以极低成本主打全球出海，对抗高价的西方软硬件方案。 ### [01:00:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3600s) 中国开源模型的审查机制与开源“修复” 在围绕中美大模型竞争的讨论中，主持人 Jason Calacanis 分享了自己试用中国最新开源大模型（部署在 z.ai 平台上的智谱 GLM 模型）的亲身体验。当他尝试向模型提问关于“台湾”或“天安门广场”的敏感历史政治议题时，模型表现得非常不悦并拒绝作答。然而，当转而让其规划巴黎旅游攻略时，模型的表现极为出色；唯独在要求将这些行程整理成信息图表并制定三日详细日程时，系统提示算力不足、处理超时，并建议切换到其他模型，因为当前模型已过载。 针对模型内置的审查机制，David Sacks 指出，中国模型在开箱即用状态下带有政治审查和意识形态倾向是毋庸置疑的事实，但对于开源生态而言，这并不是致命缺陷。美国科技公司完全可以通过 Fork（分叉）开源代码并在私有版本中进行后处理和微调来绕过这些限制。例如 Perplexity 早期在引入中国开源模型时，就成功修补并恢复了关于天安门事件等敏感内容的输出能力。因此，尽管在官方托管的网页端无法查到敏感历史信息，开源社区和海外企业依然能够根据自身需求对模型进行本地定制与重构。 ```mermaid flowchart TD accTitle: 开源模型分叉与审查绕过机制 accDescr: 展示海外公司如何引入中国开源模型并修复其内置的内容审查限制 subgraph Source["中国开源基座"] A["中国开源大模型<br/>(如 GLM / z.ai)"] --> B["出厂预设审查<br/>(敏感历史/政治话题拒绝回答)"] end subgraph Fork["海外机构二次开发"] B --> C["企业拉取并 Fork 开源权重"] C --> D["本地微调与后处理<br/>(如 Perplexity 早期方案)"] end subgraph Result["实际应用落地"] D --> E["解除审查的内容服务"] D --> F["通用推理与行业任务<br/>(行程规划/代码生成等)"] end ``` --- ### [01:01:46](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3706s) 美光业绩大爆发与“内存危机”席卷消费电子 话题转向芯片领域最轰动的事件：美光科技（Micron）公布了极为炸裂的季度财报。作为全球仅有的三家能够量产高带宽内存（HBM）的半导体巨头之一（另外两家为韩国的 SK海力士与三星），美光直接与英伟达（NVIDIA）GPU 协同封装的高性能芯片产能已全线售罄，其 2026 年的全年产能早在很久前就已被各大云厂商预订一空。美光本季度营收同比飙升 4 倍，从 90 亿美元暴增至 420 亿美元，超出市场预期 16%；第四季度指引更是从市场预期的 430 亿美元大幅上调至 500 亿美元。其股价较周期低点已飙升 10 倍以上。Jason 特别提到，早在 2025 年的年度预测节目中，Gavin Baker 就曾预言以美光为首的 HBM 制造商将是这轮 AI 周期中表现最佳的资产类别，而美光至今涨幅已高达 14 倍。 与此同时，AI 带来的内存挤压已经迅速蔓延至普通消费级硬件。此前苹果（Apple）凭借统一内存架构成为本地大模型运行热潮的最大受益者，大量专业用户争相抢购 128GB 和 256GB 的 MacBook Pro 与 Mac Studio。但如今苹果再也无法独自消化急剧上涨的内存成本，不得不将成本转嫁给消费者。对标 Chromebook 的入门级轻薄本 MacBook Neo 起售价从 699 美元跳涨 14% 至 799 美元，而 Mac Studio 的售价更是暴涨了 25%，宣告“通胀正式降临桌面端”。 Gavin Baker 对此展开了深刻剖析。他嘲讽 X 平台上那些整天用 Claude 搜索冷门日本供应商、四处寻找虚假瓶颈的“瓶颈狂热粉”（bottleneck bros）完全找错了重点——真正的核心瓶颈从来不是激光器、电容器、电源管理芯片、NAND 闪存或机械硬盘，而是 DRAM 和 HBM DRAM。正因内存容量与带宽直接决定了所有 AI 模型的性能上限，埃隆·马斯克才将超级晶圆厂“TerraFab”的核心研发重心全部押注在内存上。美光本季度的最核心亮点在于其与前四大客户锁定了包含“价格保底与上限”的长期供应链协议（SCAs），这覆盖了其近 50% 的营收；更为恐怖的是，这些合同设定的最低保底利润率，甚至超越了上一个半导体周期的巅峰水平。 - **美光财报核心指标**：单季营收由 90 亿美元同比增至 420 亿美元（增长 4 倍），超预期 16%；下季度指引上调至 500 亿美元（预期 430 亿美元）。 - **行业竞争格局**：全球具备 AI 级 HBM 量产能力的厂商仅三家（美光、SK海力士、三星），2026 年全部产能已售罄。 - **消费硬件提价潮**：苹果入门款 MacBook Neo 由 699 美元涨至 799 美元（上涨 14%），Mac Studio 提价幅度达 25%。 --- ### [01:05:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3900s) 超算资本支出、HBM堆叠黑科技与“AI通胀”蔓延 深入探讨技术本质时，Gavin 指出制造 HBM、LPDDR 以及英伟达所称的 SOCAM 与生产普通消费级 DRAM 截然不同，其物理难度“是科学能达到的、最接近魔法的领域”。中国长鑫存储（CXMT）即将推进 IPO，未来虽然可能通过向市场倾销廉价的消费级 DRAM 来解救苹果等消费电子厂商，但对 AI 服务器所依赖的尖端堆叠内存毫无替代作用。据测算，明年各大超大规模云服务商（Hyperscalers）数千亿美元的资本支出（CapEx）中，将有整整 30% 到 40% 直接流向 DRAM。从社会博弈的角度看，飞涨的内存成本将单吉瓦数据中心的造价推高到令人发指的程度，这反而可能是一件好事——它如同一道经济刹车，迫使陷入囚徒困境的科技巨头们不得不重新权衡投资回报率，从而为整个人类社会赢得调整和重建“社会契约”的宝贵缓冲期。 从硬件构造来看，高带宽内存之所以存在极高的技术壁垒，是因为它是通过先进封装技术将多枚 DRAM 裸片进行立体垂直堆叠（3D Stacking）。HBM3 已经实现了 8 层裸片堆叠，而下一代技术正在向 12 层乃至 16 层迈进。AI 数据中心对 HBM 爆发式的吞吐需求犹如一台巨型吸尘器，在短时间内抽空了全球所有的先进制造产能。而半导体晶圆厂的扩产周期通常长达数年，导致短期内供需严重失衡，直接引发了席卷全行业的“AI 通胀”（AI-flation）。由于 AI 数据中心对价格极不敏感、愿意溢价抢购，而消费电子用户对价格高度敏感，结果直接造成了消费电子领域的“需求破坏”：微软上调了 Xbox 主机售价，任天堂 Switch 和索尼 PlayStation 也面临迫在眉睫的涨价压力。 最后，嘉宾们聊到了美国本土制造芯片的现实困境。虽然美光是三大内存巨头中唯一的美国本土厂商，但其在纽约州北部的超级芯片工厂遭遇了典型的行政官僚阻碍——该工厂才刚举行奠基仪式，当天就因离奇严苛的环保法规被迫全面停工。David Sacks 和 Gavin 调侃道，在法规泛滥的深蓝州（Deep Blue State）试图建设高能耗、高用水的先进晶圆厂根本行不通，地方政府许诺的天价补贴在层层监管红线下毫无意义。这就像推进太阳能等绿色能源一样，如果严苛法规导致设备根本无法完成安装审批，一切美好设想都是空谈。美光这座耗资百亿的超级工厂，最终极大概率只能放弃纽约，搬迁至营商与监管环境更加务实的得克萨斯州。 ### [01:10:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4200s) 晶圆厂扩建困境与数据中心建设的逆向通胀 讨论从半导体晶圆厂（Fab）的落地困局展开。以往都是各州政府争先恐后地提供税收优惠来招揽工厂，而现在的局面彻底逆转——变成了晶圆厂拿着大笔资金到处询问：“究竟哪个州能真正把厂子建起来？开个价就行，告诉我们把钱打到哪里，送几个装满现金的信封完全不是问题。”针对埃隆·马斯克（Elon Musk）主导的“TerraFab”项目，加文·贝克（Gavin Baker）被问及需要耗时多久才能落成。加文指出，通常情况下建造一座常规晶圆厂需要耗时两年到三年半，但马斯克在工程提速方面有着惊人的过往记录，且此次还拥有英特尔（Intel）合作的基础优势。加文坦言，尽管没有人能给出确切时间表，但以马斯克的履历，他立起 TerraFab 的速度很可能会刷新行业纪录。即便如此，这依然是一项极度艰巨的挑战，晶圆制造堪称“魔法与科学的交汇点”，其复杂程度常人难以想象，就连英伟达 CEO 黄仁勋（Jensen Huang）也曾用“超人类般的不可能任务”来形容马斯克的工程执行力。 ```mermaid flowchart TD accTitle: 地面与轨道 1GW 数据中心建造成本结构对比 accDescr: 对比地面与太空部署 1GW 算力的资本开销与长期通胀/通缩趋势 subgraph Terrestrial["地面数据中心 (总计 ~600 亿美元)"] T_Chips["英伟达半导体芯片: 350 亿美元"] T_Infra["电力、冷却与变压器设备: 250 亿美元"] T_Labor["安装与人力成本 (高度通胀)"] T_Bottleneck["审批受阻、电网短缺、40% 项目遭抗议"] T_Infra --> T_Labor end subgraph Orbital["太空轨道数据中心 (总计 ~400 亿美元)"] O_Chips["太空级半导体芯片: 350 亿美元"] O_Launch["星舰 Starship 发射成本: 50 亿美元"] O_Tech["激光互联虚拟机架 + 轨道太阳能供电"] O_Trend["完全可复用发射技术 (强通缩属性)"] O_Launch --> O_Trend end Terrestrial -->|"成本预计上升至 700 亿美元"| EndGame["地面存量高功率机房升值为“希望之钻”"] Orbital -->|"Token 边际成本随规模化大幅下降"| SpaceXAdvantage["SpaceX 建立长期不对称算力优势"] ``` 随后，嘉宾们打破了科技行业一个长期存在的默认假设：随着时间推移，建立新数据中心的门槛与成本本应逐渐降低。现实情况恰恰相反——建立数据中心正在变得越来越昂贵、越来越困难。虽然能源本身或许能在部分地区维持低价，但关键硬件的供应链已陷入死锁。内存价格持续暴涨，GPU 未见降价，而大型变压器（transformers）与配电开关设备（switchgear）更是极度短缺。雪上加霜的是，地方土地审批规划（entitlements）和政治阻力与日俱增，全美乃至全球范围内能够容纳全新超大型数据中心的物理地点已所剩无几。 - **常规晶圆厂周期**：业界标准通常需要 2 至 3.5 年，但 TerraFab 借助英特尔合作资源和马斯克的极限工程管理，有望打破此周期。 - **供应链瓶颈**：变压器、配电设备与先进内存的供需矛盾，导致新建机房的周期被无限拉长。 - **建设阻力**：地方政府审批收紧与能源并网限制，使物理数据中心的选址几近饱和。 --- ### [01:11:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4305s) 轨道算力第一性原理：SpaceX 对决地面超算中心 查马斯·帕里哈皮蒂亚（Chamath Palihapitiya）与加文·贝克算了一笔震撼的经济账，用第一性原理剖析了“太空轨道计算”（Orbital Compute）的必然性。在地面建设一座 1 吉瓦（GW）规模的数据中心，硬件采购需要大约 350 亿美元的英伟达芯片，而配套的电力基础设施与冷却设备还需要额外耗资 250 亿美元。这 250 亿美元中很大一部分是极其高昂的现场安装与人工作业开销，具有极强的通胀属性。相比之下，在太空轨道部署 1GW 算力，同样的 350 亿美元芯片被送入轨道；一旦 SpaceX 的星舰（Starship）实现完全可重复使用，将这批算力发射入轨的成本仅需约 50 亿美元。很多人脑海中想象的太空数据中心是笨重的五角大楼式建筑，但现实中它不过是漂浮在太空中、通过高速激光互联紧密通信的计算插卡机架，本质上是一个运行在近地轨道的“分布式虚拟数据中心”。 加文指出，地面建造 1GW 数据中心目前总成本约为 600 亿美元，且受制于人力与基建膨胀，三四年后极可能飙升至 700 亿美元；而轨道的发射成本随着星舰的高频复用呈现高度通缩趋势，总成本可控制在 400 亿美元左右，日常的电力供给与散热每年仅需大约 10 亿美元。自 2021 年以来，全美约有 40% 的大型数据中心项目面临公众抗议或法律诉讼，未来这一比例只会更高。查马斯直言，外界对 AI 能耗需求的预测存在严重失衡——现实是“无限的需求对应极其贫乏的供给”。这种供给断崖将急剧提前人类将算力推向太空的经济奇点。最终，规模化运作的 SpaceX 产出 Token 的边际成本将全方位击溃地面的超大规模云厂商（Hyperscalers）。与此同时，地面上那些真正能稳定运行数百兆瓦至吉瓦级的存量机房，其稀缺性将堪比世界名钻“希望之钻”（Hope Diamond）。 - **1GW 地面算力成本**：350 亿美元半导体 + 250 亿美元电力冷却基建 = 600 亿美元（未来或因通胀达 700 亿美元）。 - **1GW 轨道算力成本**：350 亿美元芯片 + 50 亿美元星舰发射成本 = 400 亿美元（且发射成本具备通缩属性）。 - **抗议率激增**：自 2021 年起，地面多达 40% 的数据中心规划遭遇选址当地的抗议与诉讼阻碍。 --- ### [01:14:48](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4488s) 特斯拉 Megapod 商标曝光与模块化数据中心浪潮 话题转向资本市场热议的特斯拉与 SpaceX 潜在合并传闻。查马斯自信地调侃合并“很快就会发生”，并打趣说作为两家公司的股东，自己将是最大赢家。杰森·卡拉卡尼斯（Jason Calacanis）顺势开起带颜色的玩笑，调侃合并时“到底谁在上面、谁在下面”，查马斯则幽默地接梗说“我就像平时周四一样自我取悦”、“就像普通的周六之夜”。玩笑过后，杰森展示了一份特斯拉于 2026 年 6 月 18 日最新提交的商标申请——“Megapod”。商标文件将其定义为：用于人工智能计算的模块化数据中心硬件，整合了服务器网络、AI 处理硬件、配电单元及一体化冷却系统，以整套独立模块对外销售。 ``` ┌────────────────────────────────────────────────────────┐ │ Megapod 模块化算力单元 (特斯拉设计) │ │ │ │ ┌───────────────┐ ┌───────────────┐ ┌──────────────┐ │ │ │ AI 加速芯片 │ │ 模块化配电 │ │ 一体化冷却 │ │ │ │ GPU / 算力机架│ │ (PDU / 电池) │ │ (风冷 / 液冷)│ │ │ └───────┬───────┘ └───────┬───────┘ └──────┬───────┘ │ └──────────┼─────────────────┼────────────────┼──────────┘ ▼ ▼ ▼ ┌──────────────────────────────────────────────────────────────────────────────┐ │ 标准集装箱预制化交付 │ │ · 预制工厂组装 (芯片预装 + 模块整合) ──> 卡车运输 ──> 吊车入场 ──> 快速通电 │ │ · 建造周期从以“年”计算压缩至惊人的 90 天交付上线 │ └──────────────────────────────────────────────────────────────────────────────┘ ``` 坊间传言 Megapod 本质上是将高密度 GPU 机架与特斯拉巨型商用电池（Megapack）深度集成，以便直接铺设在特斯拉已获电力批文的超级充电站网络中。查马斯从数据中心开发者的第一视角对此进行了深度技术拆解：在地面搞算力，拿地相对容易，土地规划许可（zoning）较难，而获取充沛电力则是难如登天。目前主流 AI 机构如 Anthropic、OpenAI、CoreWeave、AWS 和 GCP 都会向机房开发商提供一份长达 50 到 500 页的“设计准则”（Basis of Design，简称 BOD）。但现在大模型端频繁报错让用户“稍后再试”（Come back later），证明算力已濒临枯竭，“动态激增定价”（Surge pricing）呼之欲出。为了抢夺时间，各大云厂商开始被迫妥协，放宽过去严苛的 BOD 规范：从英伟达和谷歌力推的极致液体冷却超算机架，退一步拥抱兼容传统的风冷系统。这直接催生了类似 Dell 和 Vertiv 推动的模块化集装箱方案——在工厂预制完成、平板车拉到现场、吊车一放、通上电就能立刻运行。查马斯激动地表示，如果特斯拉对外出售 Megapod，他自己的数据中心项目将成为超大买家，因为这种模式能将长达数年的机房交付周期缩短至不可思议的 90 天。 - **Megapod 商标核准范畴**：涵盖 AI 处理硬件、服务器集群、电力分配单元和冷却系统的自包含一体化模块。 - **BOD 规范妥协**：为缓解算力挤兑，头部 AI 公司放宽长达数百页的技术设计准则，从单一依赖高规格液冷向兼容预制风冷集装箱退让。 - **90 天极速交付**：模块化硬件可在库房提前预装，通过公路运输后吊装并网，大幅缩短施工周期。 --- ### [01:18:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4680s) 超充站算力争议与 Travis 的“算力厨房”实验 将 Megapod 铺设在特斯拉超级充电站的设想虽然美好，但在落地可行性上引发了激烈争论。超充站虽然拥有现成的充沛电力配额和富余土地，但其运行环境与尖端企业级数据中心存在严重冲突。加文等人指出，高端 AI 算力中心有着极高的物理信息安全要求（InfoSec）与严苛的高效液冷维护条件，而超充站完全是对公众开放的露天场所，普通车主与闲杂人员来回穿梭，极易引发安全隐患。杰森随即开玩笑打趣道：“那就在每个充电桩旁边派一个擎天柱（Optimus）人形机器人当全职保安好了。” 面对这一难题，优步（Uber）联合创始人特拉维斯·卡兰尼克（Travis Kalanick）提出了极具野心且另辟蹊径的解决方案。特拉维斯透露，他旗下的云厨房帝国在全美拥有超过 500 处实体商业地产。这些物业本身就是电网重度用户，早已接通了高负荷工业电力、强劲的商业机械冷却系统，并且管道直接接通了天然气（Natural gas）。特拉维斯表示，为了支撑自己目前正在推进的机器人技术与“物理 AI”（Physical AI）研发，他们团队正严肃考虑把算力机房直接塞进自家的餐饮厨房里，打造前所未有的“算力厨房”（Compute Kitchens）。特拉维斯直言：“这种模式下，你买到所谓‘模块化数据中心’放在哪里根本不重要。我直接去采购英伟达 GPU，然后把它们全部部署在厨房后厨，插上电就能跑，就像当年人们在车库和地下室挖比特币一样简单直接！” - **超充站部署痛点**：缺乏物理安全隔离屏障，人流杂乱难以满足大型机构的 InfoSec 审查，且复杂液冷维护难度大。 - **分布式后厨算力**：特拉维斯拥有 500 多处兼具大功率电力、机械制冷与天然气直供的云厨房物业，天然具备承载边缘算力的基础设施。 - **物理 AI 协同**：将离散 GPU 模块直接接入分布式商业物业，形成了低摩擦、去中心化的全新本地算力供给形态。 ### [01:20:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4800s) 分布式计算的现实壁垒：合规、安全“防尾随隔离门”与物理延迟的残酷定律 围绕非传统地产搭建算力集群的设想，查马斯（Chamath Palihapitiya）率先泼下一盆冷水，强调构建云算力池绝非将硬件简单机架堆叠（rack and stack）那么轻松。如果试图建立一个合成算力池并接入新兴云服务商（neoscaler），就必须签署严格的违约赔偿金（liquidated damages）和服务等级协议（SLA），承担巨大的运营责任。特拉维斯·卡兰尼克（Travis Kalanick）随即澄清，自己设想的核心是内部自用而非转售给超大规模企业（hyperscaler）。然而，除了液冷系统、大功率机械风冷与电力负荷等物理工程外，实体数据中心的“物理安全架构”构成了最大门槛——正规数据中心普遍配备被称为“曼陀门/防尾随门”（man traps）的严格物理隔离舱，必须在指纹验证通过后单向闭锁，才能开启下一道门进入机房。杰森（Jason Calacanis）对此打趣调侃自己也想装一个来“网罗人才（trap some men）”，引发现场一阵哄笑。 回到技术本质，讨论转向了分布式训练（Distributed Training）与分布式推理（Distributed Inference）之间不可逾越的物理鸿沟。查马斯与加文·贝克（Gavin Baker）指出，若将算力贡献给社区化的分布式网络（如 BitTensor、Venice 或 Pluralis），在推理层面确实大有可为，但分布式训练在物理距离上却寸步难行。GPU 节点之间若无法紧密排布在同一物理机房，即便仅相隔两公里并通过自建专有光纤互联，跨节点的任务传递延迟（lag）也会彻底摧毁整体集群效率，带来数个数量级的性能断崖。简而言之，任何在物理定律上宣判分布式训练死刑的延迟限制，反过来恰恰构成了分布式推理生根发芽的土壤。 ### [01:23:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5025s) 推理任务的架构解耦：Pre-fill与Decode分离，以及二手GPU的“长寿时代” 加文·贝克深入拆解了分布式推理在边缘侧落地的创新模式。目前已有初创公司探索在居民住宅外墙安装配备储能电池的 4 卡 GPU 密封模组，利用峰谷电价折扣换取家庭空间，并在社区本地就近处理低延迟推理请求；与此同时，Crusoe 等公司则利用类似 18 轮大型集装箱卡车的移动模块化架构快速组装数据中心。更为底层的变革来自于大模型推理在计算架构上的彻底解耦（Disaggregation）——将推理完整切分为预填充（Pre-fill）与解码（Decode）两个独立阶段： ```mermaid flowchart LR accTitle: 推理任务解耦架构 accDescr: 展示大模型推理中 Pre-fill 与 Decode 的计算差异及异构硬件搭配 subgraph Phase1 ["预填充阶段 (Pre-fill)"] A["用户 Prompt 输入"] --> B["理解语义与上下文"] B --> C["瓶颈：显存容量 (Memory Capacity)<br/>承载超长上下文窗口"] C --> D["硬件适配：传统大显存 GPU<br/>(如 NVIDIA A100 / H100)"] end subgraph Phase2 ["解码阶段 (Decode)"] D --> E["逐 Token 自回归生成"] E --> F["瓶颈：显存带宽 (Memory Bandwidth)<br/>决定生成输出速率"] F --> G["硬件适配：解码专用芯片<br/>(如 Groq LPU / Cerebras WSE)"] end G --> H["高性能高性价比混合推理流"] ``` 加文解释道，预填充阶段负责吃透提示词与历史上下文，本质是“显存容量受限型”（Memory Capacity Bound），即记忆力越强、记住的词越多越好；而解码阶段则是逐字生成下一个 Token 的过程，本质是“显存带宽受限型”（Memory Bandwidth Bound），讲究吐字速度越快越好。这种特性催生出一种极具性价比的异构搭档：将查马斯早期投资并已被行业推崇的 Groq，或是 Cerebras 这类专为解码优化的芯片，置于传统旧款 NVIDIA GPU（如 A100、H100）前端。运营商完全可以把老机房里退役的 GPU 搬进集装箱，配合前端解码芯片焕发新生。这一架构革命使得 GPU 的服役寿命由过去的 3 年大幅拉长至 7 到 12 年，从而大幅降低了硬件的资产折旧与融资成本，让席卷全球的 AI 基础设施具备了长期的金融可行性。 - **显存容量与预填充**：处理大型 Prompt 需要足够的显存空间来容纳上下文，旧款 GPU 的大容量显存依然具备不可替代的缓冲价值。 - **显存带宽与解码生成**：以 Groq 的 SRAM 架构和 Cerebras 的晶圆级引擎（WSE）为代表的硬件，在超高内存带宽支持下能将 Token 生成吞吐推向极致。 - **开源生态与防审查价值**：查马斯与加文一致认为，无论最终是何种商业分成模式，去中心化推理是构建美国本土开源活力、对抗外部竞争以及捍卫言论自由不可或缺的安全防线。 ### [01:25:35](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5135s) 闲置算力网络与特斯拉生态畅想：BitTensor、Powerwall与星链结合体 讨论进一步延伸至去中心化算力交易平台。在无许可网络中，闲置算力（surplus compute）正在被大规模重新整合。例如在 Targon 平台上，用户能以每小时 3 到 4 美元的极低价格按需租用 H200 算力。这种去信任化协作的核心依赖于 BitTensor 等协议的验证者节点（validators），它们持续对接入节点的实际硬件型号与可用算力进行动态验证，杜绝虚标作弊。 杰森基于此脑洞大开，设想特斯拉完全可以借助自身硬件生态建立全球最大的分布式家庭算力底座：特斯拉未来可以强制或优惠捆绑销售 Powerwall 家用储能电池与自研 AI 算力芯片，要求用户“买电池即配算力”，再在屋顶架设一组星链（Starlink）终端。在家庭电力充足且算力空闲时，特斯拉不仅通过电网回购向业主支付电池费用，还能将其组成覆盖全美的去中心化边缘推理网络，相当于用户免费获得了备用电源系统，而马斯克则锁定了长达 20 年的排他性分布式算力基础设施。席间还顺带提及了市场中关于马斯克有意收购 T-Mobile 以补全地面通信链路的业界传闻。 ### [01:27:22](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5242s) 超级IPO巨浪来袭：Anthropic剑指3万亿美元、SpaceX破发与流动性神话 话题最后聚焦于资本市场的历史级 IPO 洪流。Cerebras 在盛大上市后跌破了发行价（deal price），而 SpaceX 在经历暴涨后也从每股 200 美元的非理性高点回落。杰森提问，面对 OpenAI、Anthropic（估值均达数百亿至千亿美元级）以及 SpaceX 合计高达 4 万亿美元规模的潜在新股供给，公开市场究竟是否有足够的深度来承接？资金是会从比特币等另类资产回流，还是被散户流动性耗尽击垮？ 加文·贝克给出了一个令全场震惊的激进预测：他认为 Anthropic 若作为上市公司交易，其实际价值完全足以达到 3 万亿美元。加文披露，Anthropic 预计在今年年底时的年化营收表现将远超 1000 亿美元大关；更为关键的是，大规模商业化阶段的模型收入将被推理工作流绝对主导，而业内报告显示其在推理侧的毛利率高达惊人的 85%，高盈利能力足以支撑其在未来以 10 倍左右的远期市销率（P/S）进行估值。 ``` 3万亿美元市值逻辑： 年化营收目标突破 1000 亿美元级别 × 85% 推理超高毛利支撑 × 10倍左右公开发行市销率估值区间 ``` 针对“市场能否消化如此庞大体量”的担忧，加文坚定地反驳了流动性枯竭论。他指出，这仅仅是将早已在巨额一级私募市场中沉淀的资本搬迁到容量更宽广的二级公开市场而已。对于华尔街部分对冲基金发布的“SpaceX 解禁期（lockup）必崩盘”的做空报告，加文同样嗤之以鼻：SpaceX 在过去十年间每六个月就向员工和早期投资人（如 Founders Fund）提供一次合规的一级市场转让套现通道，并不存在积压已久的“抛售堰塞湖”；恰恰相反，在 SpaceX 的 IPO 中，近半数公司内部员工不仅没有离场，反而自掏腰包认购了公司股票，市场基本面展现出强烈的长期持有信念。 ### [01:30:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5400s) SpaceX 的流动性博弈与 Cerebras 破发背后的机构抛售风暴 嘉宾们就私人巨头二级市场流动性与新近上市的芯片公司 Cerebras 展开深入探讨。针对 SpaceX 的流动性问题，有嘉宾指出自己早在 2018 年就持有 SpaceX，过去每年估值以 20% 到 30% 稳步递增，而上一年估值达到约 3500 亿美元，一年暴涨 8 倍（8x）可能会促使不少早期投资人套现。但 Gavin 认为情况并不完全如此：首先，大量员工依然在以新估值买入股份；其次，大量投资人是通过特殊目的载体（SPV）间接持股，这些份额短期内根本无法解锁或分配股票；更重要的是，Elon Musk 绝不可能卖股票，大量核心多头投资人也完全没有减持意愿。事实上，许多重仓 SpaceX 的机构在 IPO 时也是积极的买方。 话题随后转向刚刚公布作为上市公司首份季度财报的芯片公司 Cerebras。Cerebras 财报发布后经历了极其惨烈的两天。Gavin 剖析指出，这一现象背后折射出华尔街投资经理（PM）生态中的一个残酷铁律：现在有一批人主张在 IPO 定价时“榨干石头里的最后一滴血”，把估值拉到极致；但硬币的另一面是，公募基金中有相当一部分组合经理遵循极其严苛的纪律——一旦新股跌破发行价（Deal Price），不论基本面如何，立刻无条件抛售（"sell it no matter what"），因为在他们眼中这代表着承销商和管理层打破了契约与承诺。 这种机械式、对价格不敏感的抛售（Price-insensitive selling）直接引发了空头的围猎：当股价逼近发行价时，空头就会全力做空砸盘，一旦击穿发行价触发机构的止损踩踏，空头便能轻松斩获 10% 到 20% 的利润。Gavin 借此告诫所有准备上市的创业者：务必勒令投行合理定价，确保上市后的前 9 个月内绝对不要破发。此外，Cerebras 面对公开市场也缺乏沟通经验，未能妥善传达其长线故事。虽然该公司 3 月份季度增长迅速，但对比当前火爆的 AI 板块整体增速显得偏弱。 Gavin 认为，Cerebras 本应以更清晰的供应链交付逻辑向华尔街解释：公司在 2025 年 12 月与 OpenAI 签署了一笔 200 亿至 250 亿美元的超级大单，随后立即向台积电（TSMC）追加晶圆订单。台积电接单并排产需要约 4 个月完成芯片制造；之后还需要大约 2 个月将芯片组装成整机服务器；若顺利解决电力接入，又需要大约 1 个月为算力集群通电并开始生成 Token。因此，OpenAI 订单带来的业绩增量最早要到当年的劳动节（9 月初）前后才会显现，第三季度才见端倪，随后才会真正放量。 ```mermaid flowchart TD accTitle: Cerebras 算力交付与业绩放量时间线 accDescr: 展示从与 OpenAI 签约到产生 Token 收入的各阶段供应链与通电周期 subgraph Phase1["签约与晶圆生产"] A["2025年12月<br/>与 OpenAI 签署 200-250 亿美元大单"] --> B["向台积电 (TSMC) 下单晶圆<br/>芯片制造周期：约 4 个月"] end subgraph Phase2["系统制造与电力部署"] B --> C["服务器整机制造集成<br/>装配周期：约 2 个月"] C --> D["获取电力配额并通电点亮<br/>通电耗时：约 1 个月"] end subgraph Phase3["商业化变现放量"] D --> E["劳动节前后（9月初）<br/>算力集群开始产出 Token"] E --> F["Q3 初见营收<br/>2027年有望实现年化数十亿美元算力收入"] end ``` - **单吉瓦算力经济账**：按 NVIDIA 生态的参考数据，部署 1 吉瓦（GW）算力约需资本支出 350 亿美元，建成后每年可产生约 150 亿美元的 Token 及云算力营收。 - **算力并网扩张测算**：电话会议中提及每月新增 50 兆瓦（MW）的目标，若 Cerebras 能在 2027 年实现月增 50 兆瓦，到年底退出的年化云计算运行率（Run-rate）将达到约 90 亿美元，而目前公司市值已跌破 400 亿美元。 --- ### [01:35:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5700s) 电力瓶颈、荷兰式拍卖与通胀年代的资本记忆 Gavin 强调，对于投资 Cerebras 这样的算力企业，投资人最核心关注的指标既不是行业竞争格局，也不是新增需求规模，而是“能以多快的速度搞定电力并网”。在超大规模云厂商（Hyperscalers）之外，业内真正交付并点亮超过 1 吉瓦电力基础设施的独立算力企业极少，业内仅 CoreWeave、Crusoe 以及点亮了约 600 兆瓦的 xAI 等少数玩家具备这种工程能力。因此，何时通电、通电多少兆瓦才是决定其变现速度的生死线。 嘉宾们随后回顾了 IPO 破发的先例，例如 Rivian 以及早年 Facebook 在上市第一年破发。破发并不意味着这是一家坏公司，更多是因为市场过度炒作或在定价上过于追求极致。为了避免投行人为平衡利益导致的定价扭曲，几位投资人一致推崇“荷兰式拍卖”（Dutch Auction）机制。前 Uber CEO Travis Kalanick 分享了 Uber 当年融资时的经典案例：当初所有投资方被要求填报一张表格，明确写出在 100 亿、110 亿、120 亿直至 200 亿美元估值档位各愿意出资多少，随后公司通过荷兰式拍卖敲定 15 亿美元的融资额度。当发现部分机构落选时，公司会通知他们有最后一次调整 Excel 表格报价的机会，从而推高估值。该轮融资从 90 亿或 100 亿美元起步，最终以 170 亿美元估值成交。Gavin 也回忆起当年自己在富达投资（Fidelity）主导给 Uber 定价时引发轰动：当时富达以约 140 亿至 170 亿美元估值投资 Uber，结果在媒体上被口诛笔伐，被斥责为“根本不知道自己在干什么”，然而仅仅 6 个月后 Uber 估值就飙升至 420 亿美元。 随后话题转到 Anthropic 寻求以 3 万亿美元估值进行大额筹资的市场传闻。面对如此惊人的数字，大家讨论全球资本市场是否吃得下这种体量的股权供给。嘉宾们认为，即便整体估值极高，公开发售的也只是一小部分流通份额（Float），比如 5% 对应约 1500 亿美元，或者 500 亿美元规模的配售，全球资本完全有能力消化。 - **估值认知的时代变迁**：十年前 140 亿至 170 亿美元的非上市独角兽估值堪称离经叛道，而如今市场动辄讨论数万亿美元估值与百亿美元级别的单轮融资。 - **资本消化的机制**：超大估值企业在上市或私募配售时仅释放极低比例的流通盘，使整体公开市场资金并不需要一次性拿出天量现金即可完成承接。 --- ### [01:40:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=6000s) 昔日“土豆当暖手宝”的艰辛怀旧与节目收尾 面对如今 AI 领域动辄几十亿美元、上百亿美元如流水般的融资盛况，全场开启了充满笑声的怀旧与“忆苦思甜”模式。嘉宾们感慨：“还记得当年融 50 亿美元有多难吗？”、“何止 50 亿，还记得当年融 150 万美元有多艰难吗？” 现场笑称大家现在听起来就像坐在摇椅上的老爷爷在唠叨“当年上学来回都要走上坡路”。一位嘉宾讲起了著名的家庭往事：父亲那一辈上学时家里极度贫困，买不起防寒手套，祖母便在烤箱里烤四个土豆，用锡纸包好，左右口袋各塞一个当“暖手宝”御寒；走到学校后，早饭吃一个土豆，午饭吃一个土豆，每天与早逝的姑姑就靠这几个土豆裹腹上学；而现在的下一代，则在攀比抢购最新的 iPhone 16 还是 iPhone 17。 伴随着关于 Gavin 背景里阿瑟顿（Atherton）短租房里奇怪艺术品的打趣，主持人逐一向嘉宾 Chamath Palihapitiya、David Sacks、Travis Kalanick（TK）和 Gavin Baker（GB）致谢道别，并以标志性的粉丝金句串烧与欢笑声结束了整期节目录制。 ## Notable quotes > [00:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=0s) 这些人有资本成为社会主义者（folks who can afford to be socialists） — Jason Calacanis > [03:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=180s) 人类有生之年所能遇见的最大的经济平等推进器（the greatest economic leveler） — Chamath Palihapitiya > [10:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=600s) 我们只是把民主党当作获取选票准入资格的工具载体，并非因为我们认同其目标。我们建立自己的组织，借用民主党的名义参选，在有利可图时与民主党组成党团，并在其内部强推我们自己的议程。在我们眼中，民主党建制派不是归宿，而是障碍。 — Josh Bloch > [14:48](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=888s) 真相与正义是社会的免疫系统。当免疫系统被抑制时，所有社会顽疾都会爆发（Truth and justice is the immune system for society. When the immune system is suppressed, all the social ills flare up.） — Travis Kalanick > [14:48](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=888s) 共产主义潜藏在我们每个人的骨子里，存在于我们人类的基因中。 — Travis Kalanick > [17:10](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1030s) 埃隆·马斯克（Elon Musk）为地球脱碳事业所做的实际贡献，比全世界所有环保活动家加起来还要多上十倍。 — Gavin Baker > [26:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=1605s) 审查从来不是为了过滤有害内容，而是为了封杀让当权者不爽、反对当权者的言论。在统治者眼里，只要支持他们就不是有害内容，反对他们就必须定罪。 — Travis Kalanick > [36:30](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=2190s) 以色列人民是令人赞叹的，犹太人是令人敬佩的。但把所有概念强行折叠在一起，因为一个人（内塔尼亚胡）的行为，就将其全盘迁怒到整个国家和整个宗教身上，这完全是不可理喻的疯狂。 — Chamath Palihapitiya > [51:40](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=3100s) 如果 OpenAI 执意要自研芯片并对黄仁勋说‘滚蛋’，千万不要惊讶英伟达是否会顺势反扑，亲自下场做一家 OpenAI 的正面对手——这是你在《All-In》节目里第一次听到的预言。 > [01:11:45](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4305s) 无限的需求对应极其贫乏的供给 — Chamath Palihapitiya > [01:18:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=4680s) 这种模式下，你买到所谓‘模块化数据中心’放在哪里根本不重要。我直接去采购英伟达 GPU，然后把它们全部部署在厨房后厨，插上电就能跑，就像当年人们在车库和地下室挖比特币一样简单直接！ — Travis Kalanick > [01:30:00](https://www.youtube.com/watch?v=w8ah_tA0yfg&t=5400s) 一旦新股跌破发行价（Deal Price），不论基本面如何，立刻无条件抛售（sell it no matter what） — Gavin Baker ## Mentioned All-In Podcast, David Friedberg, Travis Kalanick, Gavin Baker, Jason Calacanis, Chamath Palihapitiya, David Sacks, SpaceX, Starbase, Atreides Management, Zohran Mamdani, DSA (Democratic Socialists of America), Brad Lander, Dan Goldman, Chevalier, Hakeem Jeffries, Adriano Espaillat, Claire Valdez, Polymarket, Anthropic, OpenAI, Josh Bloch, Donald Trump, USAID, Anthony Fauci, Elon Musk, AOC (Alexandria Ocasio-Cortez), Megyn Kelly, Tucker Carlson, Benjamin Netanyahu, Eylon Levy, GLM-5.2, Zhipu AI (智谱 AI), Claude Opus 4, GPT-5, GPT-5.5, GPT-5.6, Andrej Karpathy, Grok, NVIDIA, Jensen Huang, Broadcom, Jalapeno ASIC, xAI, PCAST, Dario Amodei, Fable, Huawei (华为), Ascend 910B (昇腾 910B), DeepSeek-V4, Micron Technology (美光科技), SK Hynix, Samsung, Apple, MacBook Neo, Mac Studio, CXMT (长鑫存储), TerraFab, Starship, Megapod, Megapack, Tesla, CoreWeave, AWS, GCP, Dell, Vertiv, Crusoe, BitTensor, Venice, Pluralis, Groq, Cerebras, Powerwall, Starlink, T-Mobile, TSMC (台积电), Rivian, Facebook, Uber, Fidelity ## Video description (0:00) Gavin Baker and Travis Kalanick join the show! (1:05) Mamdani-endorsed socialists sweep congressional primaries in NYC (22:51) Future of the Democratic Party, the Israel issue, social media bans (45:12) China's open-source AI catch up, distillation, OpenAI's new chip (1:01:46) Micron smashes earnings, AI's memory crunch hitting Apple and consumer hardware (1:10:17) The math behind distributed compute and datacenters in space (1:27:22) IPO update: Anthropic at $3T, SpaceX float, Cerebras drops after breaking deal price Follow Gavin: https://x.com/GavinSBaker Follow Travis: https://x.com/travisk Apply for Summit 2026: https://allin.com/events 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: #allin #tech #news [Source](https://www.youtube.com/watch?v=w8ah_tA0yfg)

[0:00] Jason Calacanis: 这些人有资本成为社会主义者（folks who can afford to be socialists） [3:00] Chamath Palihapitiya: 人类有生之年所能遇见的最大的经济平等推进器（the greatest economic leveler） [10:00] Josh Bloch: 我们只是把民主党当作获取选票准入资格的工具载体，并非因为我们认同其目标。我们建立自己的组织，借用民主党的名义参选，在有利可图时与民主党组成党团，并在其内部强推我们自己的议程。在我们眼中，民主党建制派不是归宿，而是障碍。 [14:48] Travis Kalanick: 真相与正义是社会的免疫系统。当免疫系统被抑制时，所有社会顽疾都会爆发（Truth and justice is the immune system for society. When the immune system is suppressed, all the social ills flare up.） [14:48] Travis Kalanick: 共产主义潜藏在我们每个人的骨子里，存在于我们人类的基因中。 [17:10] Gavin Baker: 埃隆·马斯克（Elon Musk）为地球脱碳事业所做的实际贡献，比全世界所有环保活动家加起来还要多上十倍。 [26:45] Travis Kalanick: 审查从来不是为了过滤有害内容，而是为了封杀让当权者不爽、反对当权者的言论。在统治者眼里，只要支持他们就不是有害内容，反对他们就必须定罪。 [36:30] Chamath Palihapitiya: 以色列人民是令人赞叹的，犹太人是令人敬佩的。但把所有概念强行折叠在一起，因为一个人（内塔尼亚胡）的行为，就将其全盘迁怒到整个国家和整个宗教身上，这完全是不可理喻的疯狂。 [51:40] 如果 OpenAI 执意要自研芯片并对黄仁勋说‘滚蛋’，千万不要惊讶英伟达是否会顺势反扑，亲自下场做一家 OpenAI 的正面对手——这是你在《All-In》节目里第一次听到的预言。 [71:45] Chamath Palihapitiya: 无限的需求对应极其贫乏的供给 [78:00] Travis Kalanick: 这种模式下，你买到所谓‘模块化数据中心’放在哪里根本不重要。我直接去采购英伟达 GPU，然后把它们全部部署在厨房后厨，插上电就能跑，就像当年人们在车库和地下室挖比特币一样简单直接！ [90:00] Gavin Baker: 一旦新股跌破发行价（Deal Price），不论基本面如何，立刻无条件抛售（sell it no matter what）

[0:00] Gavin Baker: on to Cincinnati [12:49] Travis Kalanick: truth and justice is the immune system for society [12:49] Travis Kalanick: communism is in all of us [14:55] Gavin Baker: Elon has done more to decarbonize the planet than every activist combined [40:00] Panelist: America is awesome [55:00] David Sacks: AI in a box [63:00] Gavin Baker: bottleneck bros [66:00] Brad Gerstner: social contract [55:00] David Sacks: Dario got hoisted on his own petard here [73:00] Panelist: Megapod [73:00] Panelist: compute kitchens [80:00] Panelist: man traps [83:05] Panelist: Pre-fill [83:05] Panelist: Decode [95:01] Panelist: auctions are the way [100:42] Panelist: get a room and just have one big huge orgy

[0:00] All right, everybody, welcome back. Welcome back to the number one podcast in the world. The All In Podcast, episode 278. And Freeberg, he took a mental health day today after the socialist sweep in New York City. So we invited two guests. They both said yes. Travis Kalanick is here from Adams. How you doing, brother? I'm good. I'm good. Good to see you. Good to see you. Good to see you. And after a triumphant week at Starbase, the one the only everybody's favorite gavin baker of atreides management how are you doing gavin great man thanks for having me you're still floating on cloud nine after the spacex ipo yeah that was a very special moment and it was a culmination of you know some cases decades of hard work but man to quote uh bill belichick on to cincinnati yeah yeah exactly right yeah zero zero is uh the how the knicks talked about the next game in every series they're like it's zero zero we come into this as if it's like the first game of the series even if we're up two games or three games the socialists have swept new york in the congressional democratic primaries on tuesday new york city mayor mandami went three for three in the candidates which he endorsed and they all won their primaries 10th District Leader Brad Lander defeated two-term incumbent Dan Goldman. 10th is one of New York's richest districts, includes the West Village, all those townhouses, Wall Street, Dumbo, Cobble Hill, Carroll Gardens, Park Slope. That's some weird geography put together there. In New York's 13, Chevalier beat a five-term incumbent who was backed by House Speaker Hakeem Jeffries. And apparently the socialists are coming for him. 13 is one of the poorest districts, Harlem and the West Bronx, the boogie-down Bronx. She's a 32-year-old Democrat socialist with a history of spicy remarks. New York's 7th district, Claire Valdez, won the open seat over the handpicked successor, the incumbent. 7th is a DSA stronghold, Bushwick, Williamsburg, Long Island City, Greenpoint, known as the commie corridor. A lot of hipsters and baristas with suspenders in that neighborhood. According to our partners at Polymarket, this was a pretty big upset. The Mondami sweep, chances were just 26 before election day yeah that would be like the trifecta there if you were gambling These candidates just like Mondami did a lot better with younger college educated and high income folks folks who can afford to be socialists And these are all safe democratic seats. The DSA will very likely win. So the DSA caucus. That's the best way you just said. People who can afford to be socialists. It's always the rich poor. You know, they're rich, but they pretend to be poor. It's perfect. This is the history of our country from the 30s, the 50s, Red Scare, Black Panthers. I mean, we'll get into it, Chamath. But basically, if you're intelligent and you get into business, you become a capitalist. If you're super intelligent and you go into academia, eventually you have the luxury of belief that socialism is awesome. What's your take here, Chamath? And then I'll go to you, Saxipo, because I know I can see you're chomping on something over there. You're just ready to go. Honestly, I think that we are losing the script. and part of it is because we've been our own worst enemy. I'll just keep saying this, that I think AI is a very good prism into this problem. I think AI is the greatest economic leveler we'll ever find in our lifetime. I think it's the thing that can create the greatest amount of equality I think that it can even the starting line for every single person on earth But we done such a poor job in representing it in bringing it to market in talking about it We've let all of our own personal trials and tribulations and insecurities and fights spill out into the open. As a result, Silicon Valley has lost even more credibility with the people at large. And in that vacuum, what other people can paint is a picture of how anything other than what capitalism looks like today is a better version of what they see. And this is why you're seeing, I think, a lot of these people get a lot of momentum. I think if you look at some of the key congressional races, they were a referendum on AI. And the good news is we were able to hold the line in some of these key places, but just barely. In Utah and in New York, there was a couple of very important races where it was essentially anthropic-funded. anti AI groups, which is again insane against in some ways open AI funded pro AI groups and the pro AI groups one You just explained that it's a great leveler. Can you give a couple of examples and just expand on that briefly? How is it a great leveler? I mean, I know but for the audience's sake, you've said this a couple of weeks now. So I think unpack it as well. [5:00] The best way to explain it is I think that the first real major unlock of economic productivity was when the internet and specifically Google went and harvested and collected all the world's knowledge and all the world's information and they made it available via search. What we figure out though 25 years later, despite the fact that they built a great business, What was missing was then being able to take that knowledge and information and transform it into expertise and intelligence. And that's effectively what AI does. It takes that world's knowledge and it allows you to act upon it so that every single man, woman, and child has an equivalent Travis Kalanick as his co-founder, a super founder, this brilliant person that can think through all your problems, can out-engineer people, can out-think people. And they sit beside you and you have that. And there is no gatekeeping that can prevent you from having that. And so now you're only as good as your ability to direct that energy into something productive that you value. That is an incredibly powerful thing. And instead we gotten caught up in doomerism and jobs being lost and water being consumed all of which are lies all of which are complete fabrications and misinformation And these have been created in order to specifically help one small set of actors inside the AI race. And it's been fed and funded by those folks. So in this vacuum, Jason, we've allowed all these other people to paint the other version. And right now, the other version looks way more compelling than the current version because the current version has very poor brand ambassadors. Saks, what's your take on this? I guess some people are looking at this as the left's version of the populist takeover that Trump did over the last 10 plus years. What's your take on what we're seeing here with socialism and its amazing appeal and its winning at the election box? I think there is some truth to that. I mean, I think the choices of the future are going to be communism, or if you want to call it socialism of the Democrat Party, or nationalism of the Republican Party. I mean, that is where we're headed. Those are the two populist directions. But let look at what these DSA candidates stand for So let look at what their platform is They actually say they want to abolish the Senate They want to abolish the carceral state That means basically police forces and prisons. They want to abolish ICE and grant amnesty for all. They do not support any deportations whatsoever. They want to replace the president and Supreme Court with an executive and judiciary that is chosen by and subordinate to Congress, which basically now I guess this means this House. And with respect to House elections, they want to abolish the Electoral College. They want to replace the two-party system with a multi-party democracy. And they want to expand the House of Representatives, implement proportionate representation, and rank choice voting in all elections. So this would be a total makeover of our constitutional system. They want to free Palestine. They want public ownership of major corporations. They want to defund the Department of War. This is a very radical organization. And you would laugh at a lot of these types of proposals, but you can't really laugh at it anymore because these guys are taking over the Democratic Party. And you can see the Democratic establishment is in complete panic right now because they have lost control of the party to Zoram Mamdani and his allies So Jason like you said I mean let take this one race here New York 13 You got this ally of Hakeem Jeffries, longtime incumbent congressman, Espelot, I guess is his name. He is the chair of the Congressional Hispanic Caucus, and he was defeated by an unemployed 32-year-old PhD candidate. it. She's never had a job. She's been in college for 10 years, I guess, writing this PhD thesis. And I think even by DSA standards, she might be kind of a lunatic. So she has declared that she wants to end Western civilization. She wants to eradicate Western civilization. Wait a second. What? We're soaking in it. Yeah. She actually said she used the American flag as a napkin to clean her hands. Oh, she attended a rally one day after October 7, celebrating the slaughter of Israeli civilians. I mean, she's very pro-Palestine, but even to the point of celebrating Israeli civilian deaths, she calls white women ugly colonizers. She's called for the complete defunding of the police and abolishing all prisons and borders, doesn't want a single deportation, hates the police, openly calls them pigs or has [10:00] on social media before, calls U.S. service members war criminals and says the U.S. is a disgrace of a country. She's written favorably about communism and seizing the means of production and on and on and on. So this is basically the new Democratic Party. It's going to be even if the Democrats do take the House in November and Hakeem Jeffries becomes speaker, this is going to prove to be a huge headache for him managing all these new DSA members because they do not actually see the traditional establishment wing of the Democratic Party as an ally. They see it as an obstacle. This is the DSA co-chair, Josh Bloch, said, we're using the Democratic Party as a ballot access vehicle. Not because we share its goals. We build our own organization, get elected under the Democratic label, caucus with Democrats when it's useful, and push our own agenda from the inside. We see the Democratic establishment as an obstacle, not a home. So the DSA is coming from the Democratic Party. It controls the base now. It's where all the energy is. I think this takeover will continue. I think the DSA will gradually take over more and more of the Democratic Party And all I can say to these establishment Democrats is play stupid games win stupid prizes You supported this open border policy that brought in this wave of mass migration That is a huge part of the base of this new DSA wing. Mamdani would not have won the mayoral election in New York if it had just been native-born New Yorkers voting. It was the mass migrant vote in New York that swung it to Mamdani. It's not exclusively the DSA base. It's the migrants plus these overeducated white progressives who I say overeducated because they're more downwardly mobile. They end up going to work in academia or NGOs, that kind of thing. They're hard left wing. They're kind of the vanguard. And it's this combination of these recent college graduates who are kind of the organizers and this migrant movement who are really taking over the Democratic Party. But again this all goes back to Democratic Party policies They may not have intended for this to happen They may not have intended for them to lose control but it was their open border policy It was also the fact that they have cracked the melting pot and the policy we had for many years in America of assimilating migrants right immigrants Yep That all got cracked by multiculturalism and wokeism We don really do that anymore So now you got these candidates like Chevalier openly declaring they not just anti-American, they're post-American. They don't have any respect for the American system, our constitutional order, our free enterprise system. They want to introduce something in much different. And it's going to look a lot like the countries where these migrants are coming from. yeah so again you know if you import massive numbers of migrants don't assimilate them into our system and then you have these marxist leaders you're going to end up with the american system coming to resemble the countries for which it's really interesting as you mentioned like how she wants to get rid of all of these you know delete all of these uh units it's very similar to trump's you know he wanted to delete a separate set of things like the irs and usa id department of education, EPA, they each use that playbook as well. Like, hey, we're going to delete some of this stuff. Travis, you famously had the fountainhead. Trump didn't want to delete the US Constitution. No, definitely not. Travis, you're used to- Slight difference. Slight difference, I agree. He didn't want to delete Western culture. I don think he wants to delete Western culture No So go out there You might want to delete USAID because it was a festering It was basically a front for all these CIA and CIA Yeah By the way you may you may have noticed that in South America we had seven elections where the countries have swung to the right And I think a big part of the reason why it might be you don't have the USAID there conducting all these left wing regime change operations. payola's payola disappeared yeah disappeared and the incentives disappeared with it travis you uh famously had the fountainhead an rand's uh novel as your twitter avatar for a decade or two uh no they've been writing about it for a decade or two this is like a six-month thing where i would read a book and then rotate it my avatar yeah yeah hamilton for a while i had Ender's game for a while, but yeah, it became a thing. But I mean, you, having known you for a while, I think believe in the individual and their exceptionalism and trying to have a little ruggedness there. This is obviously the exact opposite. What's your take on what's happening in these pockets? Because it's not national yet, but it's definitely notable. All right. I've got sort of two aphorisms for it. [15:00] Okay. First is truth and justice is the immune system for society. When the immune system is suppressed, all the social ills flare up. okay so if you're seeing us losing truth like social media mainstream media whatever you call it like that's an early indicator for bad things happening in society okay and it's not just like social media mainstream media it's just everything around us um and the same on the justice side too if people commit crimes and there's no consequences it's a nice early indicator so you can kind of watch these things and say, is truth winning today or is it losing? Is justice winning today or is it losing? And what is the trajectory? We'll tell you, are we going to get worse or are we going to get better? Right. COVID makes, in the Fauci case, now we're seeing a lot more information come out. We have a really hard time getting the truth and maybe we'll get justice eventually. It seems like we're getting a little more truth five years But it like a perfect example of what you discussing Yeah There a lot packed into that but you can unpack it If we had a lot of time we could unpack it But those become the atomic things that you look at And then the other thing I'd say, and some people get a little surprised when I say this, which is communism is in all of us. Communism is in our blood as humans. And people go, what the hell are you talking about? You're crazy. What do you mean? Well, I'm like, well, have you ever in your life been lazy? And everybody's like, yes, I've been lazy in my life before. I said, have you ever in your life wanted something for nothing? Yeah. The difference is, do you make that a way of life? And when you have ecosystems that essentially allow you to do both of those things without consequence, those ecosystems get a critical mass and start taking hold. And I think that's what we're seeing. Well said. Yeah, I think it's well said. Gavin, I don't know you to talk too much about politics, but talk about markets. But these two things do relate, I think. So give us the economic perspective here on why this is happening Because obviously different generations have had different economic experiences So I don want to lead the witness here but I have had conversations with you about this before Yeah well so I do think obviously AI is, I think, going to be the defining political issue of the midterms and for sure the next presidential election. But I feel like what is going on with the DSA is really a fusion of two things. So the Democratic Party, I was a Democrat for most of my life, and it was the party of the working class people that was trying to create opportunities, you know, maybe level out equality. You know, help black Americans, help Hispanic Americans. And, you know, you can agree or disagree. With their methods, but I think those are all noble goals. And to a large extent, none of those are really present in the DSA. None of them. If you look at who voted for which candidate, the voting base of the DSA are relatively wealthy white liberals who are downwardly mobile. They losing votes with working class people They losing votes with poor people They losing votes with black Americans They losing votes with Hispanic Americans And those, you know, the people that the Democratic Party, I think, for a long time tried, and whether they failed or succeeded, is up to try to represent and give a voice to. They are not of interest to the DSA. And I think the DSA is dangerous. it's tragic that we are electing profoundly anti-american candidates who in some cases have called for violence to eradicate america and i just think there's there's a whole class of people who went to an elite school they grew up in really nice circumstances and instead of going into industry, they went into this kind of giant NGO nonprofit machine and their outcomes have been very different from people who did productive things for the world. You know, I've said on the show many times, Elon has done more to decarbonize the planet than every activist combined. Yeah, time sent. [20:00] Times 10. And I think these NGOs and the fact that the government is increasingly outsourcing a lot of funding to them is a really big part of this problem. And they are pursuing policies. You know, I think we talked about it last time, but the Curley effect, a mayor of Boston, you pursue policies that you know are going to be disastrous for your constituents, but they drive out your rivals. And then you can give jobs, $600,000 a year jobs, run in an NGO that does nothing productive to your friends and allies. And it's really organized corruption happening at a massive scale. You know, the one thing, whatever you want to say about the Trump administration, a lot of people are up in arms about this new plane he's getting from Qataris. Well, one, sounds like we needed a new plane, but two, it's right there for you to see. I think there are tens of billions, maybe hundreds of billions of these payments flowing to these NGOs. And if you look in California and New York I think the per capita spending on homelessness has more than doubled And it all gone to NGOs and outcomes have gotten worse In California, it's like quadrupled and outcomes have gotten worse. So I think this is this is dangerous. And I hope that the mainstream Democrats can find a compelling candidate. because I think the reason that the DSA is increasingly ascendant is not because of their ideas. I think their ideas appeal to this very narrow subset, downwardly mobile, rich white people who increasingly maybe live in quasi-genteeal poverty because they're not doing productive things. They may be well-intentioned, but they're not doing productive things and can find a compelling candidate because I think the reason they're ascendant is zohram mandani i think he is one of the most talented politicians i've ever seen in my lifetime you know he can give a great speech he's good in an interview he can tap into all of this he kind of a chameleon who can shift but i think he is a singularly talented politician and he is the reason that the dsa is ascendant not their ideas or not dissatisfaction with ai but it him and there is no one else in the democratic party i used to think aoc was by far the most talented democratic politician that was the warm-up yeah she was talented she's yes clearly like zero from a plan is clearly talented and if you combine his charisma and ability to communicate with generations like we're now like two of these lost generations that feel like they're going to do worse than their parents and if you look at housing and college debt health care they don't believe that they can participate in the system they feel the system's rigged and if somebody comes along who speaks to them and they have no conception of socialism and what happened you know in germany or during the red scare they have no idea what socialism or communism is they have this incredible rapper they put around this which is you know they're democratic socialists it's like it's it's the coke light it's it's a coke zero of socialism it's actually just communism they want to literally seize people's assets they want to seize half of you know these company stock they want to seize their wealth they they just want to take from the people who have made stuff as we talked about last week and we done nothing to change their mind we haven made more houses we haven't made college more affordable health care is a disaster inflation's a disaster and starting pointless wars obviously the support of israel is somewhere in here as part of all this something yeah sure of course yeah i was on megan kelly yesterday and i had this theory that I've been kind of working through, and I'll just share it with you guys because I'd love your reaction. If you look at the scourge of socialism, if you take the rhetoric away and you actually look at the outcomes, there are three countries that I think have veered far towards socialism well before the United States. Canada, the UK, and Australia. And I think when you look at any sort of reasonable measure of their progress, it's been an unmitigated disaster. So all of the virtue signaling on social issues, all of the virtue signaling on immigration and open borders, all of the virtue signaling on climate change has left each of those three countries in some state of disrepair with enormous amounts of infighting, tremendous political instability. They are all [25:00] sort of powder kegs and there's an interesting thing that happened in each of those three countries that i think has the potential to turn the tide and it speaks to what gavin said each of those three countries now have banned social media when you're 16 and under and i think it's sort of like making sure the kid doesn't get addicted to the drug too early and when you look at somebody like zoran mamdani i think gavin is right i'll go even further he, AOC, the lady that just won, Chevalier, let's just break it down. They're all good looking. They're all charismatic. They have their pulse on the gestalt of the moment. They know how to use social media and they are essentially curating an army. Now, if you cut the legs off by saying young people should actually age into social media, I suspect that the most important channel of information consumption, being taken away from them, actually starts to give them the opportunity and the rest of us quite honestly because we have to deal with kids who are just really stupid constantly asking for this stuff the chance to actually show you how a balanced diet actually allows you to be much healthier in your later life And I think as it relates to information consumption I think you are going to see I suspect a far less radicalized youth aging into the voting rolls in those three countries and I suspect if you start to see political stability and predictability in Canada the UK and Australia you can put your finger right on this ban as the reason why and Florida has already done it in the United States and I think we need to deeply consider it across the rest of the United States. Social media ban is now starting at 16 and Canada has it it may come to the US. Go ahead Travis. I'm a full counter to Chamath on this one. I think we all agree social media is bad for the kids and for the adults. Too much of that stuff is very bad. It's brain rot for real. And it's going to be worse than cigarettes. It's all the things. But the real point of banning under 16 is so that you can force adults to identify themselves and de themselves So you can set up a full censorship regime which they contemplating in the UK And what censorship is really about is not about harmful content It about content that the people in power don want you to see that disagrees with them If it agreed with them, it's not harmful. It's the stuff that doesn't agree with them. And so what they're doing is they're criminalizing disagreeing with them. And so there's a derivative of this age-based stuff. Age-gating. which is really about the adults, not the kids. That is the downside to it. Gavin, you wanted to add? No, I super agree that I think that a social media ban under 16 would be great, but I think it comes at a really high cost outlined by Travis. And that's what's really going on is they want to restrict anonymous accounts on X who say things that particularly the powers that be in the EU do not like. And if there's a way to do that while preserving anonymity and free speech, I'm all for it. But I think if it were not for free speech and X, like I think we'd be living in a very different world today that would be a lot worse. And just to riff on a few things that all of you have said you know on communism it is communism and the great tragedy of human experiences we can learn from the experiences of others And it may not matter that communism has failed utterly and ended in death and misery wherever it has been tried in a variety of different cultures with a variety of different mechanisms. And every generation may need to experiment with it. I think the saving grace and the importance of preserving free speech is that the DSA's policies, and I would say in particular super progressive democratic policies, are measurably bad. They lead to bad outcomes. If you care about black lives, there's a study that is uncontested that when you elect a Republican DA, all-cause mortality for young black men in that city drops by 7%. That's all it takes. You care about the environment? Well, progressives, there's so many regulations that you can't build solar, which is really the only thing that matters and the world is going to run on sunlight. You care about education? Their approach to education, of getting rid of these elite schools that allows low-income people to have a real chance, getting rid of math and calculus, [30:00] it is profoundly disadvantageous to people it leads to terrible outcomes and we all know what happens with crime it turns out that there's a certain percentage of people who you know are i think it's like 60 over 60 of the violence or 75 of the violence comes from people who have like 10 or more convictions yeah i think the specifics that is like 0.1 commits 70 of the crimes and if And if you just dealt with the 0.1%, you'd have effectively no crime. And so I just think the scary thing is, and I do worry for the first time, like as a student of history, the United States has been like a very stable political economic entity, geographically stable for a long time. And that's to its credit, because we could have literally taken over the world after World War II. but man if they get total control of some of these cities and drive out everyone who is productive i don't know how you come back and so that's that is a little worrisome to me in terms of the future of of the united states well and by the way i i don think it just new york I mean that race for mayor in L where was it Rahman somehow beat Spencer Pratt thanks to ballot harvesting after voting day Or I should say votes that were found and counted after election day. I mean, that will be a test of the DSA because they are highly organized and they have learned how to take advantage of all these rules, these ballot harvesting rules and all these types of things. the DSA, I think they've got something like half the city council seats now in LA and they're growing. So especially in these low turnout elections, and look, this was a democratic primary in New York, which is a strongly blue state. So I think maybe 17% turned out. So it's very low, but this is where the DSA really thrives and excels because they care passionately and they're highly organized and they know how to take advantage. This is why they want all these like ranked choice voting and all these types of things. They know how to manipulate and take advantage of those kinds of systems. So I think that you're going to see this in lots of other jurisdictions, wherever they're organized. I think LA will be a really interesting test. So I don't think we can just chalk this up to Zoran's popularity. You know this is a national movement and we going to see it in a lot of places but there no question that Mamdani is now kind of the spiritual leader I mean all of these people don they think AOC is a sellout you know or Bernie is a sellout You know they are way more radical than even those types. To add to that, Saks, they did learn something from Trump, which has built a big tent. So you're seeing Bernie, Mandami, Ro Khanna, they're all kind of like, yeah, we're different, but there's enough room in this tent. Well, what's happening is that these sort of more established progressive leaders. They want to tap into this energy. And even the establishment wing of the Democratic Party is now bending the knee. And so what you're going to see is regardless of how many DSA candidates actually get elected, the rest of the party is now responding to this and they're going to bend, they're going to blow in this direction because they don't want to get challenged in a primary. I mean, think about this. You had three major congressional races where the Mamdani candidate won, two of them unseated, you know, really strong incumbents. These were big upsets. So you got to think now that every congressional race in a pretty blue district, those members are now going to have to take into account that they could get primaried and they going to have to tilt their voting and their views and their rhetoric in this DSA direction because they don want to have happen to them what just happened to Dan Goldman in the New York 10th District And just to make one last point on that, so, J. Cal, you mentioned the Israel issue, and I actually think that is a hugely important and salient issue now in the Democratic Party, in Democratic primaries. Obviously, you saw that as part of the DSA platform, one of the legs is Free Palestine. this defeat of dan goldman two-time congressman he led the impeachment effort against trump he had all the right progressive credentials he checked the box and all the left-wing policies he was on the right cable news channels all the time no one expected him to lose he lost to brand lander really just over this issue of israel dan goldman is is very pro-israel he basically He defended Israel's actions over the past few years, whereas Lander, who like Oldman, is also Jewish. So this was again, white Jewish congressman against white Jewish long time New York politician. So on paper, they're very similar. It was just in this issue of Israel where they disagreed. And Lander went to a mosque in order to denounce what he called the genocide in Gaza. So this was really as close as you can get to a straight up vote. [35:00] on that one issue in this primary and Lander won pretty handily. Now, the reason for this is if you look at polling, 80% of Democrats now say they disapprove of Israel. So, you know, the approval disapproval rating, Israel used to have high approval ratings pretty much across the board. It was sort of a consensus, both Democrats and Republicans. Now, 80% of Democrats say they disapprove. You really can't underestimate how much of a motivator this is for young Democrats. They believe that they were all over campuses. We see it. I mean, it draws people out. And it has been quite polarizing just to call balls and strikes here inside the Republican Party as well. You have Tucker leaving the party over this. You have Megyn Kelly going mental over this. And you have, you know, a lot of civil war inside this administration, according to the report. So this is. Well, let me let me look. I mean, I'm not taking a side in this. I'm just trying to describe what's what's going on. Absolutely. Same as me. I think the Republican Party is a little bit more mixed on this issue, and it really comes down to age. So if you part of the older more establishment Republicans let say you a Fox News viewer you still have high approval ratings for Israel But if you under 50 which means you probably not watching Fox that much anymore you probably onto the podcast I think the disapproval rating for Israel is now 57 Interesting. So it really comes down to age. Young people across the board have serious problems with what Israel is doing. And then as you get into older age groups, that's where you see a big difference between the Democrats and the Republicans. I think Chamath, you did a wonderful job of explaining this on Megyn Kelly. There's Jewish people there's the state of israel there's israelis who live in the state of israel who are also jewish and then there's bb netanyahu and i think there's a lot of concern i mean look i want to say that who is just absolutely out of control is the consensus i think amongst many people across many religions and and political parties not to reash this whole thing but there are a lot of parallels from what happened in 10 7 and what happened in 9 11 in the following ways when you invade a country and you slaughter their people, it creates an enormous injury, an enormous emotional, physical, psychological injury. And what that country typically does is respond by giving the authority to the leader at that time to set the table right And you have to remember there a lot of things that happened post 9 that under any other circumstance would never have happened at the top of the list would have been the Patriot Act which in any other world would never have gotten past and would have seen the light of day, but not for 9-11. So there are moments where leaders are put in a position, and they essentially act on behalf of their country and their people to right or wrong. I understand that. I think everybody understands that. What's happened now, though, is that people cannot logically disambiguate Jews, Israelis, and Bibi. And I think that's very unfortunate, because we're at a point in time where everything gets conflated, and this thing has become this third rail issue, and I find it absolutely shocking. It's completely reasonable for people to have a point of view on Bibi and say, hey, you know what? It's enough, or it's gotten too far, or whatever it is. And I think that there is a very reasonable claim to make that it's time to find different leadership, new leadership inside of Israel and have an opportunity to reset their standing on the global stage. They deserve that. The Israeli people are incredible. Jews are incredible. But the idea that you fold it all together and you look at one person and then you apply it to an entire country and an entire religion is insane May I add one thing May I add one thing Please Gavin I think there's two things, really, and I'll be quick. One, Israel has a giant PR problem. They need a young, sub-35-year-old American-Israeli who's super fluent in English, conversant in social media, and is their kind of spokesperson to America. And they need that in France and Germany and every country. And they do not have that. There was a young, I think his name, I forget his name, but there was a young Israeli who was doing a really good job, Israeli-American, representing Israel after October 7th. And evidently, Bibi Netanyahu's wife didn't like him, so they canned him. And they've never really found a replacement. And I do think this is an urgent issue for Israel because it's like they are taking body blows every second and they're not even responding. And what I think they do, what they tend to do is they'll sometimes roll out someone who's a great man or woman in Israel and is in their 60s and a hero of a war. But isn't that fluent in social media? maybe the command of English is technically precise [40:00] But there's an accident. And those are great people, but they're not great spokespersons. And this is an urgent issue. As far as the antidote, going back to Tomas' comment, I do think a big reason Trump got elected, many of the Democrats I know who voted for Trump, a big part of it was that during COVID, they heard what their children were being taught. For the first time. And it was radicalizing for them. Yeah. And I think it's just really important that, like, as part of the American educational system, just there not be overtly anti-American things. Listen, we've made mistakes as a country. We're not perfect. We're about as good as it gets. And we need to tell that story in every grade consistently. And we can have a debate. We've done so many things wrong. Let's learn from them. But just, you know, slavery was endemic to the world apart from some countries in East Asia. Like it's not a uniquely American problem. And we need to tell those stories because I think a lot of kids the reason they so susceptible to this social media propaganda is they brought up being ashamed to be American ashamed of various things in their identity and being told that America is evil And we need to unwind that because America is awesome. And we're the only vaguely successful multicultural society on planet Earth. Yeah, the melting pot. And if you say that, Gavin, if you say the melting pot is a beautiful thing, you're going to get canceled. Because oh my god, you're getting rid of people's culture. It's like, no no keep your culture and then join this culture and we can all have this great smorgasbord yeah you know what my culture is you know my culture is i don't know it's no longer laura piano i know that i know you're off the train winning winning the culture of winning i love it learning progress adventure do you piss excellence to quote quote ricky bobby oftentimes it's excellent just sometimes it burns but oftentimes um just to warn just to warn the dopey democrats who have been unable to get anything done the socialists are using your party and uh this guy went viral they're just a host like they literally want to infect the oh yeah good image uh they want to infect the democratic party and then like just literally get the voting base And this guy Gustavo Gordillo I don know if you guys saw this he a DSA co in New York City And he just said it outright I was going to read you the quote We part of the Democratic Party caucus, but we don't agree with the way the Democratic Party runs its apparatus. So we're trying to build our own independence by focusing on volunteer-led movement. We think everyone should be able to be trained and become someone who can participate in the political process. And we don't think the democratic party is run that way in terms of the agenda there's a problem the democratic party they are funded by billionaire donors and at the same time they're trying to represent the working class in our opinion you have to choose between the billionaire class and the working class they are just taking over the party and again back to the playbook this worked really well for trump he took the republican party over and he owned them and they tried to get him out they tried to get him out but his message and his communication style was just too on point he knew exactly what people wanted to hear and he knew exactly how to deliver it he is a all-time comedic performer like of non- of non-comedians he's the number one comedian in the world and of actual comedians he's in the top 20 so he communicates perfectly and that's exactly what mandami done he taken the trump playbook and he has applied it here he is taking over with this communication you guys know i a diehard nick fan my whole life cried and was at game five when they won the finals And then Mondami gave this speech that somebody wrote for him about the Knicks And I was just absolutely flabbergasted and upset that it was so great. He's got that Obama-Trump charisma. And he is going to destroy your party from the inside out. Socialism is communism and is the road to suffering and pain, as Gavin said. It's no good will come out of it. Isn't it amazing? okay you know you know ma'am donnie is a communist and what he represents is basically evil and yet because he just gave a speech about the nicks you love him now no i love this speech that's all it takes we're screwed that's all it takes it's true though by the way i mean this guy he's a total phony i mean he doesn't stop smiling he's got this like crocodile smile all the time He's talking about eating you, and it's totally fake. I mean, come on. He had me hypnotized. He had me hypnotized. He was making Knicks references, and it was the most get out of your seat, standing ovation, cheer speech I've ever heard about the Knicks, and I was infuriated. This f***ing guy is such a good orator. You got programmed. I hate him. We're going to have to deprogram you now. No, no, no. It was like getting hypnotized. [45:00] ties and I just pull myself out. I got like pulled in for a second. I have a... Next topic. Next topic, for sure. Oh my God, great, good, strong first topic. I didn't know you guys were going to go all in, so to speak. Topic two, Chinese open source models appear to be catching up with the US frontier models. Let's start with a GLM 5.2 released by China Z.ai. This is a frontier class, open source free to download anywhere model 744 billion parameters, 1 million token context window and it's under the mit license if you don't know what that is open source uh licenses have very super open source it's super open source thank you the most open source the most open source of all open source if you open it up you can use it however you like uh you can fork it you can build your own company based on that no regional restrictions no api fully self-hostable no uh no daria it's just yours it's yours it's yours yours and you just gotta you just gotta reference the you just gotta reference the license that's it yeah you shout out the license and you're good scored 51 points on the artificial analysis intelligence index that the highest score of any open weight model ever stacks up nicely next to the frontier models beat gpt 5 on the frontier swe coding benchmark that a software one trails clawed opus 4 by less than one percentage point api usage cost obviously much cheaper 85 cheaper in fact than gpt 5.5 for comparable performance z.ai founder told elon musk open weight fable capabilities will be here sooner than q1 2027 gavin in other words all this hand-wringing all of these legal restrictions self-imposed restrictions are now completely or close to completely moot if they're going to have a model in q1 does this six months even matter to six months in the grand arc of ai matter or not and what does this mean for frontier models i do think how good glm 5.2 is has challenged some of my beliefs. And there was a great post from a TPU engineer that for sure distillation has happened. There's been an immense amount of distillation, no question. Please explain to the audience what that means. Distillation is when you have, you know, a like, you know, we all have seen videos of these Chinese iPhone farms Just picture a farm like that Tens of thousands of phones iPads and computers that are asking the Cloud API through masked accounts very specific questions And then these what's called reasoning traces are being harvested because if you're on the API, you get to see every token. And those reasoning traces are then fed back into the model during the reinforcement learning process and probably during the pre-training process. And that is a way that you can get really, really close to the frontier at a fraction of the cost. And this is for sure going on and has gone on for a long time. It's a cheat sheet. It's a cheat sheet for other models to catch. It would be like asking Google every question you could ever imagine, every search imaginable, getting all the results and then putting your own search engine together to make it very simple. Exactly. But now that this model is so good, it is good enough to do its own RL. And, you know, the kind of cat may be out of the bag. Now, I don't think we really know how good Mythos is. We don't really know how good the next open AI model is. We haven't seen the next SpaceX model. So maybe that gap opens up again But either way I profoundly believe the future is composable models And you going to every enterprise you going to have what Andre Karpathy called the Council of LLMs, you're going to have, you know, you're going to have Grok, you're going to have Anthropic, you're going to have OpenAI, Google, you're going to have at least two of those. I would argue Grok should always be one of the two because of its dedication to the truth. And it will tell you as a business owner a politically inconvenient truth that you need to know for your data. but you're also going to have your own open weights model that you rled on your data and you're going to put those two together the frontier models in your own model and you are going to get you know real pareto dominant outcomes and you know half the queries are going to be go to the open source model maybe 85 and only the hardest ones are then maybe they all go to open source and only the hardest ones are then checked by the frontier models. So I think this is the future. It's coming. And a misconception that a lot of people have is that open source models are, you know, somehow bad for AI. They're awesome for the AI infrastructure providers. They just shift economic value from the margins of the frontier labs to the infrastructure. And that's not bad for AI. [50:00] Great for them. It's great for them. It's great for them. But I do think there's still a role for these frontier models. And it may be true. To date, frontier tokens are capturing 90% of the economic value. And open source tokens are probably 80% plus of tokens processed. And those ratios may be here to stay. But I just think composable models are the future. What does that mean, composable model? A composable model where you have, if you're a corporation, if you're um dude is there is there a new name for your super secret um startup travis what's the new name the name is called adams it's not super secret it's super awesome super awesome super awesome um for your super awesome startup adams which man loads of people have been calling me to say how like they've been in the you know they've spoken to you and yeah investor the uh Your dog is hunting with investors, Travis. All right, fair enough. Fair enough. Okay. You're going to have a router, and every query that somebody comes in, every task that needs to be done at your company, that router is going to send it to your RL version of the or Demetron Yes Then at some point in the workflow a frontier model may or may not come in to kind of check it, add to it. And that's what I mean by a composable model when you have kind of, you know, kind of a symphony of models working together with kind of the frontier models being, you know, maybe the conductors. But that's what I mean when I say composable. Yeah, understood. Thank you. Saks, formerly AI czar, and now running PCAST. What are your thoughts on China's ascension in open source? And we're still looking for our open source champion, obviously, here in the U.S., but it feels like... No, no, hold me. Can I just say one thing? Sure. NVIDIA is the American open source champion. Thank you. They can release GLM 5.2 or better whenever they want. Okay. But why haven't they done that? they don't want to screw their customers too much of a conflict to be like pushing it out there too often? All of the above. Who knows? We'll see. Okay. Yeah. And you got to be delicate there. I see. Okay. So I'll say it, not you. Jensen does it. Yeah. I mean, it's a, it's a classic channel conflict, right? You you don want to compete with your customers and he is competing with Elon on self now and Elon making his own chips So I think these frontier model companies should think very very carefully about ASICs and the incentive that creates for NVIDIA. On the model perspective, OpenAI, which launched their jalapeno chip this week and announced it being built by, I believe, Broadcom. And they are saying, hey, F you to Jensen and NVIDIA. and they already were full contact with him, don't be surprised if NVIDIA says, you know what, we kind of like the area you're operating in now that you're going to make chips and maybe OpenAI sells those chips to other people. Don't be surprised if NVIDIA starts an OpenAI competitor. You heard it here first on All In. Sax, did you want to jump in there? Sure. I mean, look, I think that China's been good at open source for a while here. There's nothing new about that. But there are a few things that are significant about GLM 5.2. So the first one is it is now, like you said, the best open weight model for coding, software engineering, and long context agent work. And you gave a couple of the SWE bench scores. I mean it was just a tick below Opus 4 and it was right up there with GPT 5 So if you compare this to again the state the frontier models for Anthropic and OpenAI it is right there with the previous model But you have to remember now that the current model, Fable for Anthropic and 5.6 for OpenAI is now in a little bit of a purgatory because of all the reasons we covered last week. Now, look, like I said last week, I ultimately blame Dario and the way he communicated and the way that he primed officials to be on a hair trigger with respect to these models. And when the government got a credible report about a jailbreak from, you know, one of Anthropics' almost trusted partners, you're going to say roll that back. But that is the situation we're in right now is that Fable has been rolled back and GPT 5.6 is trying to navigate these new approval hoops. So we now have a Chinese open weight model that is as good as the currently available models from OpenAI and Anthropic. And look, this is a point I've been making really since I joined the administration is that we are in a very competitive situation with China. I've been saying this from [55:00] beginning our whole ai strategy from the get-go was about winning this ai race defining it as a race as being globally competitive and we cannot afford to do things unnecessarily that slow our companies down i hope that by the day do you think dario got exactly what he wanted it seems to me there's some chance this has been a very calculated strategy to provoke the u.s government into doing what they just did and this is what he wants he has a regulatory moat now He can keep his future models behind this, give it out to Glasswing, use it to distill it for themselves. Do you think this is what they wanted? I think that on a certain level it is what they wanted because they've been advocating to have a federal regulator, basically a new agency. In fact, Dario posted a blog just a few weeks ago saying he wants an FAA for AI. They wanted government approval, a government approval process for AI models. And so in a sense, they've gotten exactly what they wanted. Now, that being said, I don't think they're happy about the fact that Fable has been rolled back. So in a sense you could say that Dario got hoisted on his own petard here or it could be a FAFO situation But look my view on it is we should not reward Dario by giving him exactly what he always craved which is some sort of labyrinthine government approval process that does reward regulatory capture. So I hope that very soon now, I do think that as long as Anthropic has resolved the jailbreak issue, then I do think they should be allowed to come back to market and similarly for open ai i don't think we should be delaying them unnecessarily we do not have months to give away in this race and let me just say one other thing which again is something i've been saying for months which is with respect to risks like cyber it is undoubtedly a risk but what is the response to that the only thing you can do is go out and find all the vulnerabilities first yourself the white hats have the white hats find all the vulnerabilities and do a big upgrade cycle, roll out the patches before they can be exploited. The reality is that if you just clamp down in a way that doesn't even allow these models to be used, the Chinese are going to have these capabilities imminently anyway you know they already at Opus 4 level And the founder of Z he said that before Q1 they have Fable level capability I believe them because look the Chinese have been I say nine months behind our models plus or minus three months, depending on capability. But it's when they know there's been a breakthrough around something like cyber, they can deploy more resources against that particular problem and catch up faster. So like I said, we're on a shot clock here. I've been saying for months that we're on a shot clock. We have to do smart things. We can't just slow everything down because that will not slow down the Chinese. They're not under our jurisdiction. We have to basically get these tools in the hands of our cybersecurity industry. They're the force multiplier. They're the enabler. We have to basically go out and do this big upgrade cycle quickly because we only have a few months left. They're six months behind on the model and they're 24 months behind on the silicon, yet they're only a few months behind in total. So what game are we playing? This is insane. We are gonna lose if we keep doing this stuff to ourselves. So let me make a point about that. So on the silicon, there's been a huge push in China by the government to push their AI labs to develop and train on Huawei chips And look you can take these claims with a grain of salt Maybe they not true but it was claimed that deep seek v4 was trained on huawei chips and now z they are saying that the glm5 family was trained entirely on clusters of huawei ascend 910b chips so now look maybe they're lying maybe they smuggled in some nvidia chips but the claim is that this was all done on indigenous chips. And what I believe is that China is engaged in a strong indigenization push right now. They want to prop up Huawei as the national champion. They want all their companies using Huawei chips. They still need to scale some of the manufacturing, but they're going to do that pretty quickly. And then what they're going to do is they're going to take these Huawei chips, they're going to take these Huawei optimized models. Remember that GLM 5.2, the inference is optimized for the Huawei chips. OK, we know that. And they are basically going to package these things up. They call it AI in a box. They're going to sell it at a fraction of the cost globally. Which is what they do with every technology, right? Better, cheaper, faster. Or almost as good. And that's another thing is I've been saying since the beginning of this administration, we have to be pro export because China is going to be there within a [60:00] One or two years, as I said, we're going to be kicking ourselves because we could have the whole global market to ourselves. We invented reasons not to sell abroad to our friends and partners. And now China is going to be there imminently. Yeah. And with that lower price tag, when you play with this new GL, you get some really interesting responses. I asked it about the country of Taiwan. Was not pleased and didn't give me an answer. I asked it about Tiananmen Square. No answer as well. I'm using the hosted version at z.ai. But when I asked places to visit in Paris, it did an exceptional job, except when I said make these ideas into an infographic and make me like a three-day agenda. It was like, hey, we don't have enough time to do that. And it said, use the other model because this model is too busy. But you can go play with it at z.ai. Just on a point about the censorship. So there's no question that these Chinese models have, you could say censorship and you know there's political bias and they're out of the box but american companies have taken chinese models and then essentially worked around and basically fixed the censorship inside their own forked version so for example perplexity did this very early on with chinese models they showed that you could sort of put back the content on Tiananmen Square and things like that So I think J Cal you absolutely right about the censorship but it not a fatal problem It's something that American companies can fix when they take an open source model and fork it and customize it. Yeah, in the hosted version, you're not going to get a great answer of what agenda you should use for tourism in the great country of Taiwan or your visit to Tiananmen Square. All right, let's keep moving here on the docket. Micron smashed their earnings. If you don't know Micron, they're one of only three companies on Earth that make high bandwidth memory. These are specialized chips. They sit on top of the NVIDIA GPU. and their entire 2026 supply is sold out and has been for some time sk hynix and samsung also make hbm micron smashed earnings revenue up 4x 4x year over year 9 billion to 42 billion beat expectations by 16 big jump in guidance for q4 50 billion versus 43 billion their stock is up 10x shout out to Gavin in our 2025 prediction show He gave a call on HBM makers like Micron as the best performing asset since that time Micron up 14x I not crying in my suit You're not crying in your suit. I got a ton of information here. I think this will just end on the Apple price increases. Everybody knows Apple has been really been a beneficiary of the run local models movement that I'm part of. and people are buying 128 gig, 256 gig, MacBook Pros, Mac Studios. But the gig is up apparently because now Apple, which had not passed on those costs to customers, is having to pass those increases on. So everything from the new MacBook Neo, which is their $699 laptop, kind of competing with Chromebooks, is now $799, up 14%. and Mac Studio up 25%. The costs are just going to be very significant. Inflation has come to the desktop. Your thoughts, Gavin, on Micron and the impact on the industry? And is this a temporary bottleneck, or does this mean everybody has to get into this business quickly? No. Well, one, DRAM is the most important bottleneck. There a whole segment of people on X who are very focused on bottleneck Bottlenecks I call them the bottleneck bros You know they do some work with Claude find some esoteric Japanese company The bottleneck that matters is DRAM And DRAM and HBM DRAM, this is the most important bottleneck simply because memory capacity and bandwidth are foundational to the performance of every AI model. So this is the most important bottleneck. Elon is focusing the TerraFab on memory because he sees it as the most important bottleneck. You know, not lasers, not capacitors, not power supply semiconductors, not NAND flash, not HDDs, DRAM. And I think this bottleneck is going to be with us for a while. And it is kind of astonishing. So I think so a few thoughts like what was important about the quarter. They announced that they have these SEAs, these supply chain agreements that have a floor and a ceiling for prices with an increasingly large group of large customers. And this covers essentially 50% of their revenue, I think, with just four customers. And the floor pricing in these new contracts is ahead of prior cycle peaks from a gross margin perspective. And so this is really, I think, pretty, maybe end up being very transferable. [65:00] for the industry. Most other parts of the semiconductor supply chain have re-rated. You know, LAM Research, you know, the way for fab equipment suppliers, you know, they all trade at huge premiums to DRAM relative to prior cycles and their business models have improved. But, you know, so has the industry structure and business models of DRAM because HBM DRAM is increasingly a customized chip. But as far as other people being able to do this, look, CXMT is going public in China. They may be the cure for Apple's ills. They will flood the market with, to some degree, cheap consumer-grade DRAM. But for the DRAM you need in these AI servers, there are three companies that can make it. It's really hard to do. This is as close to magic as science can get. And, you know, I think TeraFab, TeraFab, you know, is going to be an important part of this solution. But, you know, these these stocks still trade or cross-sectionally cheap relative to the rest of AI. Something I been thinking about memory is DRAM is probably going to be 30 to 40 percent of all hyperscaler CapEx next year Every hundreds of billions of dollars that are spent you know going straight to DRAM It wild But this may actually be very valuable for society because it is probably, you know, going to, you know, inflate the costs of building a gigawatt data center to the point where, like, you know, even for the hyperscalers, economics matter, who are caught in this prisoner's dilemma. and this may give us as a society time to adapt to adapt you know what our friend brad gershner calls the social contract so the high iphone prices you know one cxmt is coming for consumer grade dram but two this may be good for ai it may be good for us as a society and making it is just really pure silicon, right? Like making memory is just incredibly refined silicon. And that might be the pre-bottleneck? Yeah, making that HBM DRAM, making what NVIDIA calls SOCAM, making LPDDR. These are the types of DRAM that are really hard to make, not consumer-grade DRAM and they are increasingly what you need in these AI data centers Semiconductor Yeah so my understanding of HBM stands for high bandwidth memory Again this is part of the you know the GPUs that go in the data center to run AI is that you take the DRAM wafer or die and you actually stack them. And so I think HBM three is like eight, it's stacked eight dies high, but now they're increasing to 12 and even 16. and to basically stack them and then package them all together is that's an advanced technology in and of itself so you're seeing now like gavin's saying there's only three companies that can do it but also this is creating significant price pressure for all the consumer electronics businesses apple had huge news today where they announced massive price increases and again it's because dram now is less available because it's just being hoovered up by all the data centers and And if you're a data center and you need to buy GPUs, again, those chips, they're using immense amounts of DRAM because, again, one HBM chip is using multiple stacks of DRAM. So it's just getting slurped up. And then it takes a couple of years to ramp up new capacity. So these companies are going to do that but that could take a while We saw that in New York remember that was that Micron plant that had just broken ground and then got shut down the same day because of some crazy environmental issue So it not easy to ramp the stuff in the US although Micron is the one provider that's in the US. SK Hynix is in South Korea. Samsung's in South Korea too. But anyway, we're going to see again more of this AI flation, they're calling it. It's just another reason I hate AI is it is in this narrow area of consumer electronics where there's competition for DRAM. It is leading to price inflation now. Microsoft raised the price of the Xbox. You know, it's coming for the Switch. It's coming for the PlayStation. You know, there's demand destruction because the price is in consumer, whereas AI demand is relatively price insensitive. David, I would modify one statement. It's hard to build a micron. It's hard to build a new fab in a deep blue state. You can build that. Why were they trying in New York? It's kind of crazy. You know, New York gave them all these incentives, but it doesn't know those incentives matter. It's a little bit like solar power. You can be as pro-environmental as you want, but if you can't build and install solar because of regulations, it doesn't matter. So, you know, maybe that micron plant ends up getting, you know, built in my home state of Texas. The incentive game has kind of flip-flopped [70:00] Like it used to be the states were courting the factories and the fabs. Now it's the fabs are like, which state can actually build this? We'll pay you whatever you want. Just tell us where to send the envelope. We'll drop a couple of envelopes off. It's not a problem. Gavin, can you say how long it's going to take to stand up the fab at TerraFab? Well, I mean, if it was a normal fab, it would be a two, three, three and a half year process. But, you know, we've seen what Elon has done to other construction processes. And, you know, he's starting with some advantages with the Intel partnership. So I don't think anyone knows. But based on past history, he's probably going to stand up TerraFab faster than other Fabs have stood up. But this is really hard. It's really hard. It's the intersection of magic and science. You can't believe how complicated this is. So it's going to be hard. but you know he has a he has a track record of doing you know what jensen called impossible superhuman and so we see we see how long it takes you know one other point here that i guess it might be relevant to spacex ai although it don have to limit it to this is I think there an assumption that over time it would get cheaper and easier to stand up new data centers, right? But what you're saying is actually it might be getting harder. It might be getting more expensive, right? Because there's competition for these components. The memory's getting more expensive. I'm not sure that the GPUs are getting any cheaper. I guess some of these- The transformers, the switchgear, it's impossible. And the energy might be getting cheaper. And then the entitlements are getting harder. And the political situation is getting harder. There's very few places you can even stand up new data centers. So is it the case that actually it's going to get more and more expensive to? 100%. So to stand up a one gigawatt data center, it's $35 billion in semiconductors, NVIDIA semiconductors. And there's $25 billion of power and cooling equipment. And that is clearly inflationary because a lot of that $25 billion is the human labor required to install it so the calculation that needs to be done for orbital compute is it's 35 billion of silicon in each space and you know in literally outer space in orbit and on ground but if you can get the cost of launch significantly below that 25 billion dollars then the math starts to really mass and when starship is reusable it going to cost five billion billion to put a gigawatt of compute into space And something that drives me crazy is people picture these Pentagon data centers No, it's racks in space linked with lasers. It's kind of a virtual data center in space. Wait, $5 billion? Is that $5 billion of launch costs? $5 billion of launch costs. Now you're $40 billion to put the gig into space. You're at $60 billion terrestrially. and the 25 billion that is powering compute is clearly inflationary. And so it may be that in three or four years, it's 70 billion versus 40 billion, and that five, as Starship becomes rapidly reusable, is likely deflationary. So this is the economics that underpin orbital compute from first principles. And then on an ongoing basis, you're maybe paying a billion dollars a year for the power to run those chips and cool it. If I had to make a guess, I think what's going to happen is that since 2021, about 40% of all data centers get contested, right? I think that number is going to go up. So Saks, I suspect that whatever forecasted energy consumption that we are looking at in AI is grossly imbalanced There is very very meager supply And there effectively infinite demand So that probably pulls forward the economic equation to want to go to space But then again, that's going to prefer SpaceX and their compute stack and their compute decisions over the hyperscalers and over anybody else. And so you're going to have a cost of an output token, think terrestrially particularly from the hyperscalers be a little economically lopsided versus spacex once they get it to scale now that's the key statement whatever is left on the ground though will be incredibly incredibly valuable it'll be a diamond these are diamonds and the thing is you have to find reasonable size right you can't have a 10 kilowatt diamond that's like a little pebble nobody cares but if you're in the reasonable hundreds of megawatts to gigawatts man Those are like hope diamonds. Those are just lock them down, which I own. Travis, there was a interesting trademark filed this week. So more in the investigator, investor, investigator of the Tesla plus SpaceX marriage that everyone seems to believe in. [75:00] It's going to happen shortly. You can say it. You can give me credit for it. Of course, yes. As Chamath has architected in his high perch. But this trademark came out for- By the way, how cool would that be? How cool would it be for those two to come together? It's going to happen. It'd be amazing. And it'll be incredible. And if you're lucky enough to be an owner of both, oh, who's the top and who's the bottom? Chamath is the top, who's the bottom? Everybody can decide to be an owner of both. Jason, I'll be making love to myself when this happens. So no different than any other Thursday. So just another day. On the top and the bottom. Maybe the top and the bottom, just like any other Saturday night. Here's the trademark for Megapod that came out. This is a filing date of 6-18-2026, so a very recent, June 18th. Modular data center hardware for artificial intelligence computing comprised of network of computer servers, computer hardware for artificial intelligence processing, computer network hardware, electric power distribution units, and cooling systems sold as a unit, self-contained modular computing, hardware systems for artificial intelligence workloads, yada yada yada Essentially Well just to explain what this is Well yeah let me just By the way that description makes I don know what it is I still it i still don know essentially what people are saying is this is going to be a giant battery pack with gpus at the supercharger stations which have already been approved that the back channel chamath go ahead you have a couple of issues right now to turn on compute terrestrially so assume you have land that's relatively straightforward assuming you can get it zoned less straightforward assuming you can get power very difficult then you have a very critical design decision so all of these folks publish these things called the basis of design and your your bods essentially tell you here's the anthropic spec here's the open ai spec here's the core weave spec here's the aws spec here's gcp and you get these 50 and up to upwards of 500 page documents of all these technical details the issue that we have is i don't know if you guys have used open ai or anthropic recently where you get the whole thing of like come back later, right? That come back later is completely unacceptable. It just means that they have no compute. So in trying to find... Surge pricing is coming. Yeah, let's go. So in trying to find this solution, what is happening is these basis of design, the restrictions the specificity is being relaxed And one of the key constraints is we moved to an architectural model from Google and NVIDIA that has said look we have to liquid cool these racks these are very complicated big girth supercomputer racks essentially and now we're also going back and saying you know what maybe some of these older stuff that's a little bit less proficient and a little bit less useful but we can air cool them is useful and everybody's like yeah you know what we should try to use everything that's available In that second class, what some very smart people are doing are like, wow, well, here's a shipping container that you can just drop on a concrete pad somewhere, plug in the power and let it rip. And so, Jason, what you're seeing is that level of investment that's happening. So there are companies, you know, like Dell makes these racks, companies like Vertiv makes these modules. And I think if Tesla can make them available, folks like us for my data center project, we would be enormous buyers of these things. Because we would just literally prefab them in a warehouse, right? Get the chips, prefab them, truck them to the place, crane them in, turn it on, off to the races you go. You have like a 90-day build cycle, which is unheard of. So that where these Megapods are coming from I hope it not entirely consumed by Tesla and SpaceX internally Well that the rumor is that he going to put them at the supercharging network where he has a lot of land and he's got power there already. And those are lightly used in some cases. There are other requirements. The problem with some of these things is that you need to have certain levels of access. There are certain InfoSec requirements or certain liquid cooling requirements that make many of the workload applications unfeasible at a place like a supercharger center where random bumble are like traipsing around security guard put an optimist so guys i've got i've got 500 properties with lots of energy lots of mechanical cooling systems and lots of uh lng access or sorry not lng natural gas sorry already piped in. So on all the stuff I'm doing on robotics, AI, physical AI, we're literally looking at putting some of our compute into our kitchens. Compute kitchens. You're going to be able to buy small, modular, let's put data center in a quote. But it doesn't matter. I just go buy the GPUs and I just set it up in a kitchen. right? It's like Bitcoin. [80:00] Hold on. It's not that easy. No, no, no. Hold on. Hold on. It's not that easy. If you guys want to build a cloud and contribute it to a pool, you're going to have to sign up for liquidated damages. And you're not going to sign up to these SLAs. I'm not a hyperscaler. I'm just using it for myself. Rack and stack? Oh, okay. Sure. You can use it for yourself. I'm saying the more interesting opportunity is when... Whatever Travis says, let's talk about the interesting stuff. No, no. I'm just saying what I thought you were going with this, Travis, This is like you could build a synthetic pool and contribute it to a neoscaler. And all I'm saying is to do that is a leap of things that you'd have to do that you probably don't want to do because it veers you away from what you're good at. I think the security thing is the main issue around non-data center type real estate, to be honest. It's the big one. Look, you've got to do liquid cooling. You've got to do mechanical systems for air cooling. You have energy. There's a lot of things you've got to do. But assuming you even had all that, the architecture of a data center is set up like they have man traps, right? Where you go into a room, the door closes behind you that you got into with your fingerprint. Then you go into the next part of it against your fingerprint only when that door closes like Well just out of genuine curiosity how much does one of those man traps cost Why can you put some of those in your Maybe you could Are you really interested in putting in a man trap I'm just trying to help trap some men. I'll try to interpret where Chamath's going a little bit. It's like a megapod that's sitting at a chart, like a public charging situation because of the physical access of it. Like there's probably some stuff you've got to do if you're going to resell it. But if you're going to use it for yourself, you can use it for yourself. The real value, I think, in the Travis example, which I find super exciting, is if you can contribute it to a distributed training pool or a distributed inference pool that has a lower SLA, which is more of how this community-based oriented work. Jason, I think you've talked about BitTensor. I think Venice is a project. I think Pluralis is another project. Travis, that's where I think it's super exciting. But the one thing I'll just throw out there, guys, the distributed training stuff. If these things are far away from each other, meaning not right next to each other, they're so less efficient. The efficiency drops dramatically. You want these things to be right next to each other physically. You get multiple orders of magnitude at least an order of magnitude type efficiency maybe plus plus plus It a big deal So you could have two We have two of our facilities that are like two kilometers Passing the jobs from one GPU to the other is super important And if it's on a peer-to-peer network, it's going to have lag. It's going to take more time. Honestly, even if it's on your own fiber connected like two kilometers away, you're screwed. It's not a thing. This is 100% true, but everything that cuts against training works for inference. respect well latency is super important and i do think you know a distributed inference distributed inference clouds are coming i get that that's a riff and to riff on like what tamas said and all of this like one i mean there is actually a startup that is trying to put four gpu units with kind of a battery on people's houses and give them a discount on their power and then you can do inference for that neighborhood from those four GPUs and it's like lock sealed so nobody can get in. But there's another dynamic that I think we should talk about with all of this. It can play into the Megapods and other people are kind of working on data centers. Crusoe is working on modularly assembling data centers kind of like a data center and think of it as like an 18 wheeler What do you call those things The 18 wheeler You know shipping container whatever it is But that is the disaggregation of inference into pre-fill and decode. When you when a model is answering your question, it's doing two things. The pre-fill part is understanding the question and it's answered thus far. And think of think of that has the more you can remember, the bigger your, you know, your memory capacity. literally the more words you can remember, the better. Decode is the process of generating the next token. And that is a memory bandwidth bound problem. And think of it as the faster you can speak, the better. And these two types of inference are increasingly being disaggregated. And Chamath was an investor in Grok, which NVIDIA bought, and they're going to use this. Cerebrus is the other solution today that is available. And you can put Grok or Cerebrus, decode optimized chips. They would both say you can do more than decode on them, and that's true, in front of old NVIDIA GPUs, like H100s. So you can lift H100s, A100s out of some old data center, put them in one of these megapods, a rack in a shipping container, [85:00] put a Grok or a Seribus in front of it, and you can get a very competitive solution. And so I do think the disaggregation of inference, we're going to be using GPUs for seven years, 10 years, 12 years, and that's great because it lowers the cost to finance them, which makes this AI revolution more financeable. Chamath, you want to riff on Grok? Should we? I really agree with everything. You're incredibly well-steeped in this space. It's so exciting. I don't have any investments or anything in the distributed compute space, but at the intersection of competing against China, having a vibrant American open source community, having a bunch of distributed models for purposes of free speech and otherwise that Travis mentioned earlier. I do think this idea of distributed inference has a real place in the American ecosystem. I don't exactly know where and how and how homeowners would get paid, but whoever figures that out as a pan-American idea, I think is really on to something. It'd be awesome. and yes travis obviously slower but people are contributing compute to this that's like kind of surplus computer or computer that's available recycling unused compute recycling unused compute and then there targon which is just straight up people are putting and you can rent h200s by the hour for three bucks four bucks and it permissionless Anybody can contribute provided and this is kind of the magic of BitTensor is there are validators that make sure you're putting in what you're say you're putting into the network. And so if somebody just had the hardware and Tesla said from now on every Powerwall and I don't put it past him to do this, you buy a Powerwall, we give you a discount on it, every Powerwall has our gpus in it part of the offering is you can't buy a power wall without putting gpus in it and when you're not using it we will pay you for your battery and you put a starlink on your roof and now you've got a distributed system where you get a couple of these um power walls with their own silicon in it and a starlink and now elon's created an infinite number of home battery backup systems people get their battery system for free he gets the exclusive use for 20 years you guys see that there's a rumor that elon's going to buy t-mobile by the way just because cerebrus has blown up and gone through deal price we should talk about cerebrus a little so uh cerebrus had an incredible ipo give us an update on where these ipos are happening obviously spacex had an incredible ipo but has retreated from this you know otherworldly 200 share price so to the extent you can talk about these two as well as the two ipos to come four trillion dollars in backlog we've obviously got open ai and claude i'm sorry anthropic which makes claude both of those would be worth a billy plus you put all this together four trillion of new offerings plus cerebrus in there in the mix how does the market manage this much new inventory being put on the market. Obviously, the flow to SpaceX is notably small, but over time, people like yourselves and other insiders, founders fund, etc., will be unlocked and have the ability to distribute to their LPs. So this is going to be a moving target, I think, on terms of share price. And can the market absorb this? Where does the money come from? Retail? Or does it come from people selling their Bitcoin and moving it over to something more exciting? What's the dynamic here in the market? I know we're in uncharted territory, Gavin. Well, for sure, there's no precedent for any of this, but a few things I would say, like just in no particular order, I get the office worth $3 trillion today. And it's very important. I'm sorry. Did you say anthropic is worth $3 trillion? Yeah, I think that is roughly where it would probably trade as a public company. And wow. Yeah Oh I mean look they going to do they going to end this year Oh oh man they gonna end this year well over 100 billion holy what the 28 number three trillion number is it 200 is it 300 billion it's probably not going to trade it 10 times that number and it will be very profitable at that scale because it'll be inference dominated and people reporting they have 85 gross margins on inference but in terms of the market absorbing this like the markets already absorbed it you know it's it's just shifting from private to public and so in the scale of global capital markets these seem like really big numbers you're just moving from the private markets to the public markets which are even bigger as far as spacex specifically i think one of the more important things um is everybody who's a spacex investor or employee has had a chance to sell every six months for the last 10 years. So they may not be the wall of liquidity that some people are thinking about. I read this New York hedge fund short report that you could just short SpaceX on the lockup because so many people are gonna sell. Really? Well, everybody who's on the cap table, they had an opportunity to sell. And almost half the employees at SpaceX bought on the IPO. Now, I do think, [90:00] But Gavin, one thing though, I got to say, so I've had SpaceX since 2018. What, what? But their little liquidity thing every year, like last year was like 350, 350 billion last year. So you got an 8x in one year, you could have a lot of people selling, right? They were doing little 20% up, 30% up clips for many years. and then an 8Xer could create that liquidity. Maybe, but we'll see. That's possible. But the employees are buying at the new price and they're probably at some level one of the biggest pools of ownership that's going to unlock. And then I do think a lot of people probably own SpaceX through SPVs and those probably don't unlock or get distributed anytime soon. So I would be careful with assuming think elon's a seller no he's definitely not obviously yeah so and a lot of us and a lot of us aren't sellers you know and that's great yeah yeah i mean a lot of people who had large space expositions were large buyers on the ipo on cerebris so cerebris has had a tough two days since they reported their first quarter as a public company And I think there are two things that are worth discussing here. One is there is a whole generation of portfolio managers. There's a lot of people who are advocating for kind of squeezing the blood out of the stone on IPO prices. And the flip side of that is that there are a lot of portfolio managers who, if a stock breaks deal price, they sell it no matter what. They consider it a promise that was broken. And so this is what has happened with Cerebris to some degree over the last two days. And, you know, this may seem irrational, but there are people who run giant funds who I know personally, where if a stock breaks deal price, they sell no matter what. And so if a stock breaks deal price, it can sometimes, you know, go to places you wouldn't think it would go. And this means that shorts, if a stock gets close to deal price, they short it because they want to break deal price. And then, you know, they make a quick 10 or 20 percent. So it becomes a pile on. Because you have this price insensitive selling that can be triggered And this is what has happened to Cerebrus You know people talk about hate sale hate selling but they broke deal price and so just if you going public and you're listening to this tell your bankers price this in such a way that we're not going to break deal price in our first nine months as a public company and that's what i always advise everyone to do and i think it's important but i also think you know it takes companies a while to learn how to tell their story and communicate to public markets. It's a very different audience than VCs. And the way I would have respectfully told the Cerebris story, because what happened to Cerebris is they reported a quarter and they're growing fast, but relative to the rest of AI, they're not growing that fast in the March quarter. So what I would have said is we signed this transformational 20, $25 billion, I don't know the exact number, contract with OpenAI in December, December of 2025. We immediately ordered more wafers from Taiwan Semi. It takes four months from when we make that order Taiwan Semi says yes they start producing It takes four months to make the chip then we then it takes us two months plus or minus to turn that chip into a server And then if we lucky and we can find the power it takes us a month to energize that chip and start making tokens with it. So the first time you're going to see the impact of this OpenAI deal, at the earliest is probably around Labor Day. So you'll see a little bit of it in the third quarter. But then it really, it starts to build. And just like really simple math. So like, let's just use some rough numbers. Let's say, let's take some NVIDIA numbers. It takes some 35 billion to bring on a gigawatt and 15 billion that you can generate 15 billion in token revenue and cloud revenue out of that gigawatt. And somebody on the call talked about adding 50 megawatts a month. If they could add 50 megawatts a month in 2027, forget 26. that means they exit the year at roughly a nine billion dollar cloud computing run rate and you know we're at we're at less than 40 billion of market cap now that is going to be really hard to do and they've never done anything like that before and but i would if i would i would focus [95:00] as an investor i think what matters here is not where they sit competitively not what new demand they can bring on but just how quickly can they bring on power and listen like outside of the hyperscalers the only companies that have ever brought on more than a gigawatt i think are core weave um crusoe um and and spacex ai to bring it on 600 megawatts it's really hard and that is what But like I'm focused on, as an investor, how many megawatts can they bring up? Because we know what they're going to monetize at. And that is the question. And we'll see. Yeah. I think that's well said. Yeah. The other companies that have had this happen of late, I guess Rivian and famously Chamath Facebook traded below IPO price in the first year, I think. And so this isn't necessarily mean a bad company. It just means a lot of hype or maybe going for it in the IPO and pricing it to perfection. And it can go one of two ways. You can either grossly underprice, grossly overprice. And it's really hard to get right. And this is why doing auctions or, you know, auctions are the way auctions are the way. And so it not hard to price if you just do it the way it supposed to be done It hard to price when you want it to be a certain number Yeah And you have many mouths to feed and that influences the price. I don't think that happened here. I think there was a good faith effort to price this thoughtfully. But just, you know, there was such an in-demand IPO. I think it was a hard IPO to price. All right, let's end on... Gavin, you just auction it. I think that's different than underwriting it. You're underwriting. But the bankers, I think they just need to get more in the mode of just doing the auction. Yeah. Final topic was this $3 trillion. Or actually, now it's your $3 trillion call on Anthropic would make it like $6 trillion in offerings. What do we think broadly in the market? in offerings because it's they're going to offer a small slice of that sure and that small slice is very easy to absorb in the context i believe we have no precedent we'll see i might be wrong but it's not hard for global capital markets to absorb 50 billion of an offering it's not like it's not like somebody has to come up yeah it a five percent dollar design 150 billion there enough yeah guys remember remember when 15 billion used to be a whole lot of money How weird is what going on How weird is this crazy? Where is all this money coming? Oh, yeah. Where was it all this time? Can we just reminisce, man? We got my, when I was at Fidelity, my colleagues and I, we got pilloried. I think we priced Uber 14 billion and we got pilloried in the press for not knowing what we're doing and it was seen as, you know, then, of course, like six months later, I think you guys did around at what, like 42 billion or something. But it's like, yeah, 14 billion. That was groundbreaking for a private company back. Yeah, for sure. It was 17, but, you know, 17, 17. You're right. I wanted 14 and we had like a big negotiation. And we ran an auction. No, we just ran an auction. Yeah, you wanted 20 and maybe we landed at 17. What we did was every person who want to be involved had to fill out a sheet of how much money they put it at 10, 11, 12, 13, 14, all the way up to 20. And then we just did the Dutch auction. We said, we want to clear one and a half bill. We just did the auction and cleared it. Went back to people and said hey you not going to get it You have another shot They update their Excel sheet and it just moves the number up a little bit and then you close it down. But yeah, it started, that round started at, I think it was nine or 10 or something like that and ended up at 17. Thank you for your service. Yes, it used to be that that was a lot of money and that was unheard of at the time. And that was, well, look, it was 10 years ago. when we used to walk to school uphill both ways. Yes, we used to have to. Yeah, we used to eat potatoes with no butter, just hot potatoes. We were lucky that they were cooked. I love that. Sometimes they were wrong. No, my dad would tell me that story. It's a famous John Lebeard story. He'd say his mom would put like four potatoes in the oven. They'd wrap them in tinfoil, one in each pocket because they couldn't afford gloves. So you put your hands in your pocket with the hot potatoes. You get to school, you eat one for breakfast, one for lunch. and his sister, Johanna, God rest for soul, my aunt who died too young, and they would just go eat these potatoes. That was their life, walking to school. Yeah, and our kids are trying to- I'll go both ways. Yeah, they're trying to get the new iPhone 16 or 17, I don't know. Remember when it used to be hard to raise $5 billion? [100:00] crazy. Remember when it was hard to raise 1.5 million? Dude, I remember I yeah, I remember when it was hard to raise. Anyways, we don't have to do this. It's okay. We already sound like grandpa's. It's fine. We sound like it was so hard to raise that first 1.5. It's just we don't need to go there. It's so funny, though. I hate this art piece, Gavin behind you. Is that like an art piece? Or you made it? Yeah, no, this is I'm staying at a rented house in Atherton. Oh, okay. There you go. It's just an Airbnb. All right, everybody. For the dictator, Jamal Pai Hapitiya, and for David Sachs, TK, GB, we'll see you next time on the All In Podcast. Bye-bye. Great job, everybody. We'll let your winners ride. Rain Man David Sachs. We open-sourced it to the fans, and they've just gone crazy with it. Love you, Wesley. I'm the queen of kinwaf. I'm going all in. Besties are gone. That is my dog taking an ocean in your driveway. Win it a bitch. Oh, man. My habanazesher will meet me at what? We should all just get a room and just have one big huge orgy because they're all just useless. It's like this sexual tension, but they just need to release them out. What? You're the beat. What? You're the beat. You're a fee. We need to get merch. These are back. I'm doing all in. I'm doing all in.