Intel Jumps on Apple Chip Manufacturing Deal as Trump Brokers White House Pivot

特朗普促成白宫转向，英特尔获苹果代工协议股价大涨 在台积电晶圆产能瓶颈迫使Mac硬件延期数月后，苹果和Nvidia正将关键芯片生产转向英特尔的14A制程节点。 本期节目深入解析了英特尔近期的股价反弹。在部分源于美国政府干预的推动下，英特尔与苹果和Nvidia达成了初步芯片供应协议。主持人将10大科技巨头主导市值的蓬勃发展的AI经济，与喜忧参半的就业数据、举步维艰的家电制造商以及主题公园等体验式消费保持稳定等实体经济指标进行了对比。他们还从奥克兰关键的OpenAI联邦庭审现场发回报道，分析了被改编成网络热门音乐的被泄高管短信，并解读了包括DeepSeek巨额70亿美元融资以及SpaceX/xAI向Anthropic提供算力等行业重磅突发新闻。 英特尔就使用其即将推出的14A制程节点为苹果代工芯片达成初步协议，推动英特尔股价大涨近20%并创历史新高。 (0:00) 特朗普政府在将约90亿美元联邦补贴以每股21美元的价格转为10%股权后，积极为英特尔代工业务牵线科技合作交易。 (0:00) Nvidia和SpaceX也对英特尔做出了重大承诺，其中Nvidia以每股23美元投资50亿美元，联合开发定制数据中心CPU。 (0:00) “AI十大巨头”在美股总市值中的比重已达到约40%，反映出资金在AI硬件和云基础设施建设上的高度集中。 (8:14) 宏观经济呈现显著分化：专门的AI行业GDP增长了31%，而非AI经济的增长仅为0.1%。 (8:14)

TBPN highlights an intensifying divergence between capital-heavy artificial intelligence ecosystems and the broader traditional economy, marked by aggressive public-private interventions, extreme market concentration, and massive compute deals. State-Backed Intel Turnaround and Big Tech Alliances (0:00) Intel shares surged around 20% to reach $118–$125 following reports of a preliminary foundry deal to manufacture chips for Apple, aiming to alleviate Apple's supply bottlenecks at TSMC. The turnaround follows aggressive intervention from the Trump administration, which previously converted roughly $9 billion in grants into a 10% equity stake at $21 per share. Under CEO Lip-Bu Tan, who replaced Pat Gelsinger in March 2025, Intel has targeted its upcoming 14A process node to secure manufacturing commitments from Apple, SpaceX, and Nvidia, the latter having previously invested $5 billion at approximately $23 per share. The AI Economic Divergence and Capital Concentration (8:14) Public equity markets reflect unprecedented concentration, with the top ten AI-driven firms—the 'AI Big 10'—accounting for roughly 40% of total U.S. stock market capitalization, comparable to historical railroad peaks. Citing analysis by Greg Ip in The Wall Street Journal, the hosts note that dedicated AI sector GDP expanded by 31% alongside surging capital expenditures in tech hardware (+43%) and data centers (+22%), while the non-AI economy grew by just 0.1% with personal consumption rising 1.6%. Labor Market Stasis and Consumer Spending Shifts (10:24) Despite broader economic unease, the U.S. labor market added 115,000 jobs in April, beating economist consensus estimates of 55,000. Job additions were concentrated in non-AI disrupted sectors such as healthcare, warehousing, retail, and transportation, leading KPMG Chief Economist Diane Swonk to characterize the environment as a 'high anxiety job market' where employees hesitate to leave roles and hiring remains frozen. Concurrently, consumer behavior is shifting toward an economic 'barbell': capital equipment and housing-adjacent durables like Whirlpool suffered dividend cuts and an 80% five-year stock drop, while experiential spending like theme park operator Six Flags showed attendance and revenue growth. OpenAI Trial Revelations and Compute Infrastructure Deals (24:51) Federal court proceedings in Oakland involving OpenAI highlighted internal governance crises, featuring video depositions from Mira Murati and Helen Toner alongside live testimony from former board member Shivon Zilis regarding accelerationism and leadership tensions surrounding Sam Altman's temporary ouster in November 2023. Meanwhile, global compute demand continues to drive unprecedented transactions. China's DeepSeek is raising a $7 billion round at a $50 billion valuation, with its founder directly committing $3 billion to secure infrastructure for frontier models, and SpaceX/xAI has partnered to supply Anthropic with compute despite previous friction between Elon Musk and the Anthropic team. The hosts observe that market valuations are increasingly torn between an extreme AI bull case of hyper-productivity and systemic bubble or displacement risks, though pragmatic demand for high-end compute continues to override both corporate rivalries and traditional macroeconomic drag.

本期节目聚焦人工智能引发的宏观与产业两极分化：在白宫斡旋下，Intel凭借14A制程与重组迎来估值修复并与Apple等巨头接洽；与此同时，资本与就业市场呈现出AI基建高歌猛进与传统实体经济疲态并存的“杠铃”结构。 白宫斡旋与Intel代工重组 (0:00) Intel与Apple就代工芯片达成初步意向，促使其股价当日暴涨近20%，突破每股118至125美元。此项潜在合作是双方秘密谈判一年的结果。此前Apple完全依赖TSMC，但近期Mac mini与Mac Studio因TSMC先进制程晶圆短缺而遭遇数月发货延迟甚至被迫涨价，迫使Apple寻求供应链多元化。白宫在此期间发挥了关键斡旋作用，特朗普政府曾将近90亿美元的联邦补助金转为每股21美元、占股10%的Intel股权，目前账面收益显著。特朗普与商务部长Howard Lutnick积极向Tim Cook等科技领袖推介Intel，旨在增强美国本土制造韧性。 在管理层层面，陈立武（Lip-Bu Tan）于2025年3月接替Pat Gelsinger出任Intel CEO，并引入前TSMC高管罗唯仁负责先进封装与定制芯片，尽管此举引发了TSMC的诉讼。目前Intel正全力押注14A制程节点，除Apple外，Nvidia此前也以每股约23美元向其投资50亿美元以联合开发定制数据中心CPU（账面回报达5倍），Elon Musk的SpaceX亦围绕“TeraFab”规划与其保持接触。 AI十巨头与双轨经济割裂 (8:14) 公开市场的资本集中度已达到历史极端水平，被称为“AI十巨头”（在原先七巨头基础上加入AMD、Broadcom和Micron）的企业合计占据美股总市值的40%左右。节目引述历史对比指出，这一集中度仅次于1835至1910年美国铁路热潮时期（当时铁路股曾占据全市场63%的权重）。 宏观数据显示出AI基建与非AI经济的严重脱钩。引自《华尔街日报》Greg Ip的测算显示，专用AI领域的GDP增速高达31%，而非AI经济的增速仅为微弱的0.1%。作为经济支柱的总个人消费仅增长1.6%。传统商业投资在住宅、写字楼、普通工厂和运输设备上普遍下滑，而科技硬件资本支出猛增43%，企业软件投资增长23%，数据中心建设投资增长22%。 劳动力冰冻与消费杠铃分化 (10:24) 美国4月份新增非农就业115,000人，虽较3月向下修正后的185,000人有所放缓，但大幅超出华尔街预期的55,000人。毕马威首席经济学家Diane Swonk将其形容为“高焦虑就业市场”：在职者因担忧前景而不敢离职，求职者则面临普遍的招聘冻结；新增就业主要集中在医疗、零售、仓储与运输等不易受AI直接冲击的领域，科技大厂裁员并未动摇超1亿人总就业盘的绝对数据。 消费端则呈现耐用品与体验型支出的“杠铃分化”。拥有百年历史的家电巨头Whirlpool因成屋销售下滑3%、高额债务以及来自LG和Samsung的激烈竞争，自1950年代以来首次削减股息，股价5年暴跌80%；与之相对，主营纯可选消费的游乐园运营商Six Flags（FUN）第一季度入园人数、游客消费与营收全面增长。分析认为，消费者会推迟可维修的大家电置换，却不愿牺牲子女成长期的线下游乐体验。这一现象延伸出“杠铃投资假说”：资本一边流向以Oracle、OpenAI为代表的高科技AI基础设施，另一边流向主题公园、职业体育等无法被AI自动化复制的真实物理体验（即所谓的“抗粗制滥造/anti-slop”资产）。 全球算力角逐与非常规同盟 (30:06) 法庭现场直击披露了OpenAI早期治理裂痕，前董事Shivon Zilis到庭就历史摩擦作证，前董事Helen Toner与前CTO Mira Murati的录像证词亦曝光了2023年11月管理层内讧期间的通讯细节。在诉讼纷扰之外，全球算力争夺正在重塑资本与商业版图。中国AI企业DeepSeek正在以500亿美元估值筹集70亿美元资金，创下中国AI领域融资纪录；其创始人以对冲基金背景个人出资30亿美元（占本轮40%）并保持90%持股，意图全力扫购算力，押注在数月内重返前沿大模型行列。 与此同时，算力短缺的现实甚至压倒了昔日的商业与个人恩怨。尽管Elon Musk此前多次公开抨击Anthropic团队，但SpaceX及xAI已与Anthropic达成算力供应协议。凭借Colossus等超大规模数据中心集群的快速电力交付能力，xAI实际上已化身为算力云供应商，向Anthropic输出关键算力支持，印证了商业计算需求在现实利益面前的务实妥协。 本期讨论表明，AI基础设施竞赛正吸纳巨量资本并驱动半导体先进制程回流美国本土，但这种繁荣与疲软的传统实体经济形成鲜明对立；市场在生产力无限释放的极度乐观与技术泡沫破裂的深度焦虑之间摇摆，算力瓶颈与商业同盟的演变仍是左右行业走向的核心变量。

## Background 本期内容由科技脱口秀频道 TBPN 于 2026 年 5 月 22 日发布在 YouTube 平台。节目属于 30 分钟精简版栏目 Diet TBPN，由 John Coogan 与 Jordi Hays 共同主持，主要围绕苹果与 Intel 潜在芯片代工交易、美国新增 11.5 万非农就业数据、DeepSeek 筹备以 500 亿美元估值融资，以及科技巨头近期的重磅动态展开讨论。 ## TL;DR 节目首先聚焦 Intel 在白宫牵线与苹果初步代工协议推动下的股价暴涨，指出其与 Nvidia、马斯克联手构建的代工生态正折射出“AI 十巨头”主导美股的极度集中度。随后，主持人们对比了 AI 投资的高歌猛进与传统实体经济的萎靡，借助惠而浦与六旗乐园的财报分化阐释了拥抱实体稀缺体验的“杠铃投资策略”。最后，制作人 Tyler 亲述了马斯克诉 OpenAI 庭审首日的现场见闻，并结合 DeepSeek 创始人自掏 30 亿美元巨额融资以及 SpaceX/xAI 向 Anthropic 转售算力等突发新闻，展现出当下 AI 赛道兼具疯狂狂热与深层焦虑的现实图景。 ## Atmosphere & style 整场节目充斥着硅谷极客播客特有的快节奏与冷幽默，主持人在犀利的商业洞察与无厘头的烂梗调侃之间自如切换，甚至将法庭短信证据魔改成 Emo 流行乐播放。现场氛围活跃而接地气，既有亲历奥克兰法院排队抢座的生动见闻，也有对半导体估值过热“如同无伞跳伞”的形象比喻，让观众在轻松密集的笑料中深切感受到当前 AI 资本热潮下混合着兴奋与末日隐忧的集体焦虑。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=EgugwJdzeOc&t=0s) 传苹果与英特尔（Intel）达成初步芯片代工协议，白宫牵线并持有英特尔 10% 股权，推动英特尔股价单日飙升近 20%。 - [10:25](https://www.youtube.com/watch?v=EgugwJdzeOc&t=625s) 美国 4 月新增非农就业 11.5 万人，远超预期的 5.5 万人，显示传统基石产业就业大盘依然稳健，打破经济熄火论调。 - [31:33](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1893s) DeepSeek 拟以 500 亿美元估值融资 70 亿美元，创始人梁文锋个人出资 40%（约 30 亿美元），资金将全力用于储备算力以追赶前沿模型。 - [32:21](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1941s) 马斯克旗下 xAI/SpaceX 与 Anthropic 达成算力合作，标志着其基础设施正向新型云服务商（Neo-Cloud）迅速转型。 ## Key takeaways - [01:30](https://www.youtube.com/watch?v=EgugwJdzeOc&t=90s) 苹果正在寻求台积电之外的双代工渠道，在白宫穿梭协调与产业政策注资的推动下，初步与英特尔达成晶圆代工协议。 - [04:13](https://www.youtube.com/watch?v=EgugwJdzeOc&t=253s) 英特尔新任 CEO 陈立武顶住政治质疑推行重组，押注 14A 尖端制程，成功绑定英伟达、特斯拉/SpaceX 与苹果三大核心大厂。 - [06:24](https://www.youtube.com/watch?v=EgugwJdzeOc&t=384s) 台积电先进制程产能被 AI 芯片买家瓜分，导致苹果面临 iPhone 与 Mac 等硬件交付延期的困境，议价能力与优先级受到挤压。 - [06:24](https://www.youtube.com/watch?v=EgugwJdzeOc&t=384s) 包括 AMD、博通与美光在内的“AI 十巨头”（AI Big 10）占标普总市值近 40%，市场集中度逼近 19 世纪铁路繁荣时期的极端水平。 - [09:02](https://www.youtube.com/watch?v=EgugwJdzeOc&t=542s) 美国经济呈现严峻的内部脱钩：AI 相关经济增速达 31%，而非 AI 经济增长仅 0.1%，商业建筑与传统消费动能相对疲软。 - [10:25](https://www.youtube.com/watch?v=EgugwJdzeOc&t=625s) 4 月非农就业新增 11.5 万人，零售、仓储与医疗保健表现强劲，证明科技行业裁员噪音并未撼动拥有上亿人口的宏观就业大盘。 - [13:34](https://www.youtube.com/watch?v=EgugwJdzeOc&t=814s) 惠而浦削减股息与六旗乐园逆势增长体现了“AI 杠铃策略”：一端极度聚焦 AI 基建，另一端拥抱具备稀缺性、不可被虚拟化的线下真实体验。 - [23:41](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1421s) 半导体板块追高情绪过热，估值处于历史高位，分析师警示当前买入无异于高空无伞跳伞，需高度依赖极端完美的后续利好承接。 - [24:49](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1489s) 马斯克起诉 OpenAI 案在奥克兰正式开审，首日披露了米拉·穆拉蒂宣誓录像与内部短信，聚焦 2018 年控制权争夺与 2023 年罢免风波。 - [31:33](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1893s) DeepSeek 计划以 500 亿美元估值完成 70 亿美元融资，幻方量化掌舵人自掏 30 亿美元，押注巨额算力储备以突破下一代 Frontier 模型的性能瓶颈。 - [32:21](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1941s) Anthropic 采购 xAI/SpaceX 算力印证了新型云服务商（Neo-Cloud）的崛起，强大的快速并网与基建落地能力正压倒 AI 阵营间的文化与商业敌意。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=EgugwJdzeOc&t=0s) 经济过山车与“K形之中的K形”：Intel暴涨与美股分化 周五的开场依旧充斥着熟悉的调侃与喧嚣。主持人打趣道，大家必须坐稳扶好，因为“美国经济过山车”的剧本又在不间断地上演。美股市场正在疯狂拉升，甚至让人觉得主持人都该穿上全白西装来应景庆祝。在这一片狂欢中，最亮眼的明星莫过于 Intel——其股价日内暴涨近 20%。但在这股狂热之下，更深层的宏观现实是“AI 经济”与美国“实体经济”之间日益扩大的鸿沟。不仅两大经济体正在脱钩，甚至在看似统一的走势之下，实体经济内部也层层嵌套着“K 形之中的 K 形（K shapes within the K shapes）”，不同企业间的命运正在剧烈撕裂。 引爆 Intel 股价的最直接催化剂，是《华尔街日报》（The Wall Street Journal）披露的一则重磅独家：Intel 已与苹果公司（Apple）达成初步的芯片代工协议。尽管这笔交易尚未彻底敲定，但双方已经展开了长达一年多的密集谈判。对长久以来呼吁重振 Intel 的科技分析师（如 Ben Thompson）和市场观察者而言，这场长达数月的拉锯战终于迎来了实质性突破。 - **市场异动**：Intel 股价在消息传出当天暴涨近 20%，触及高位。 - **核心传闻**：苹果与 Intel 达成初步代工协议，双方已进行了长达一年以上的深度谈判。 - **宏观格局**：资本市场呈现极度分化，AI 概念股与实体产业之间、实体产业各公司之间均出现严重的 K 形分化走势。 --- ### [01:30](https://www.youtube.com/watch?v=EgugwJdzeOc&t=90s) 苹果重回Intel代工：白宫牵线与三大巨头的“接盘” 苹果与 Intel 的渊源深厚，整个 1990 年代至 2000 年代双方都是最紧密的盟友，直到苹果在 2020 年正式启动 Apple Silicon 计划，将桌面芯片全面转向台积电（TSMC）代工。如今 Intel 试图挽回这位老客户，尽管目前尚不明确 Intel 将具体为苹果的哪些设备生产芯片。Intel 长期以来兼具芯片设计与晶圆代工（Foundry）双重业务，但在过去几年里，两块业务的双双疲软让前任 CEO 帕特·基辛格（Pat Gelsinger）推行的复兴计划步履维艰。局面的真正转折点来自美国政府的强力介入：特朗普政府此前促成了一项重大协议，将近 90 亿美元的联邦芯片补贴直接转换为 Intel 的股票，使美国政府获得了 Intel 约 10% 的股权。 这笔国家级注资入场时的成本约为每股 21 至 23 美元，而如今 Intel 股价已经飙升至 125 美元左右，实现了约 5 倍的丰厚账面浮盈，白宫团队“显然感觉好极了”。与此同时，Nvidia 首席执行官黄仁勋（Jensen Huang）当时也跟投了 50 亿美元，如今单日浮盈就高达一二十亿美元，主持人开玩笑称这笔投资收益“正好够老黄给马斯克的 Grok / xAI 结账付款”。《华尔街日报》指出，特朗普政府与商务部长霍华德·卢特尼克（Howard Lutnick）在促成这笔交易中扮演了关键推手，多次召集苹果 CEO 蒂姆·库克（Tim Cook）、SpaceX/特斯拉的埃隆·马斯克（Elon Musk）以及英伟达的黄仁勋进行游说施压，强调必须为了美国制造业的供应链韧性、摆脱对台湾单一外源供应链的过度依赖，给 Intel 下一代晶圆厂一张兜底的底牌。随着苹果的初步入局，Intel 已经成功将苹果、Nvidia、马斯克（特斯拉/SpaceX）这三大顶级客户全部拉上了自己的战车。 ```mermaid flowchart TD accTitle: Intel Foundry Revival Ecosystem accDescr: Diagram showing the US government, Nvidia, Musk, and Apple backing Intel's foundry business. subgraph CapitalAndPolicy["政策与资本兜底"] Gov["美国政府 / 商务部<br/>90亿美元补贴转10%股权"] Trump["白宫穿梭外交<br/>(施压与游说三大巨头)"] Gov --- Trump end subgraph BigTechCustomers["三大战略客户"] NV["Nvidia (黄仁勋)<br/>投资50亿 / 定制数据中心CPU"] Elon["马斯克 (Tesla/SpaceX)<br/>TeraFab 长期晶圆合作"] Apple["苹果 (Tim Cook)<br/>初步代工协议 / 分担产能压力"] end subgraph IntelHub["Intel 制造核心"] Node["14A 先进制程节点<br/>(Intel Foundry)"] end Trump --> Apple Trump --> NV Trump --> Elon Gov --> Node Apple --> Node NV --> Node Elon --> Node ``` - **白宫股权注资**：美国政府将近 90 亿美元芯片法案拨款转换为 Intel 10% 的股权，入股价约 21 美元/股。 - **Nvidia 巨额浮盈**：Nvidia 以每股 23 美元价格注资 50 亿美元，随股价冲上 125 美元，已斩获约 5 倍回报。 - **三大战略同盟**：Intel 目前已先后与 Nvidia（定制数据中心 CPU 合作）、埃隆·马斯克（TeraFab 项目及芯片采购）以及苹果公司（代工制造协议）达成不同层面的深度捆绑。 --- ### [04:13](https://www.youtube.com/watch?v=EgugwJdzeOc&t=253s) 陈立武的重组铁腕：从撤职危机到押注14A先进制程 回顾过去十年，Intel 因一系列技术失误、频繁的高层人事更迭以及失败的整合尝试，在先进制程上被台积电和三星电子（Samsung Electronics）远远甩在身后，导致外部代工客户纷纷削减或撤销订单。2025 年 3 月，董事会罢免了帕特·基辛格，任命陈立武（Lip-Bu Tan）为新任 CEO。换帅初期曾上演了一幕极其惊险的政治戏剧：特朗普总统因对陈立武早期在亚洲的商业背景存有顾虑，曾一度公开质疑并要求撤换他。然而，陈立武凭借极具说服力的“魅力攻势（charm offensive）”迅速赢得了特朗普的信任与认可，随后美国政府正式官宣了 10% 的股权入股计划。 获得白宫支持后，陈立武迅速以铁腕手段重构 Intel 的核心管理层。他从台积电挖来了资深高管罗唯仁（Wei-Jen Lo），甚至因此招致了台积电的法律诉讼，标志着双方台面上的彻底决裂。此外，陈立武撤换了原有的产品线负责人，为数据中心处理器部门、客户计算部门以及新设立的定制芯片（Custom Silicon）业务物色了全新掌门人。在技术路线上，他将所有筹码集中押注在代号为“14A”的下一代尖端工艺制程上，期望以“筑巢引凤”之势吸引顶级大厂回流。特朗普在白宫与库克面谈时甚至亲自替 Intel 当起了说客。特朗普后来公开宣称：“只要我们（政府）一进场，苹果进来了，Nvidia 进来了，很多聪明绝顶的人全进来了，政府已经在这笔交易里赚到了几百亿美元。” - **换帅风波与反转**：2025 年 3 月陈立武接替基辛格，经历了从遭遇白宫政治审查到用魅力攻势全面赢得特朗普支持的过山车反转。 - **台积电诉讼**：Intel 挖角台积电前核心高管罗唯仁，引发台积电正式提起侵权诉讼。 - **业务组织重构**：全面设立定制硅片业务线，技术演进路线孤注一掷聚焦于 14A 先进制程。 --- ### [06:24](https://www.youtube.com/watch?v=EgugwJdzeOc&t=384s) 苹果的产能困境与“AI Big 10”主导的市场集中度 苹果选择与 Intel 接触，背后有着严峻的供应链现实。长期以来，苹果的 iPhone、iPad、Mac 等核心产品线几乎完全仰仗台积电代工。然而，在最近两次的财报电话会议上，库克直言不讳地指出，先进制程芯片的短缺是导致苹果无法满足 iPhone 旺盛市场需求的关键原因。这种产能瓶颈目前已波及 Mac mini 和 Mac Studio 等桌面产品线，平衡供需甚至需要数月之久，逼得苹果直接上调了 Mac mini 的起售价格。过去苹果是台积电说一不二的头号客户，但随着生成式 AI 爆发，Nvidia 等 AI 芯片巨头对台积电先进制程产能的争夺已达白热化，导致苹果在台积电那里的议价能力和排产优先级受到严重挤压。面对地缘政治与实际缺货的双重夹击，苹果必须尽快找到第二代工来源（Dual-source）。 这一轮半导体产业链的合纵连横，折射出美股市场前所未有的头部高度集中现象。市场的所有超额回报几乎全部被少数几家万亿美元市值的超级科技巨头包揽，如今 Intel 也凭借重组和概念加入了这场狂欢。彭博社分析师甚至将这一阵营从“美股七巨头（Mag 7）”扩展定义为涵盖了 AMD、博通（Broadcom）和美光（Micron）的“AI 十巨头（AI Big 10）”。这十家核心科技公司的市值权重已经占据了整个标普市场的 40%。这一极端数字在美股历史上罕见——主持人引用数据指出，唯有在 1835 年至 1910 年的铁路大基建时代，铁路股票曾一度占据市场 63% 的极端份额；而在日经泡沫、Nifty 50 时期以及 2000 年互联网泡沫中，都不曾见过如此恐怖的单一行情集中度。这正是《华尔街日报》资深经济专栏作家格雷格·伊普（Greg Ip）所剖析的核心命题：当几家大厂绑架了整个指数，繁荣与泡沫的界限已变得无比模糊。 - **苹果供应短缺**：先进制程产能受限已直接影响 iPhone 交付，Mac mini 与 Mac Studio 面临长达数月的供需失衡，导致终端售价被动上调。 - **议价权易手**：Nvidia 等 AI 芯片设计商对台积电先进制程的爆买，削弱了苹果作为台积电最大单一客户的产能锁定优势。 - **“AI Big 10”市场集中度**：Mag 7 加上 AMD、博通、美光构成的“AI 十巨头”已占美股总市值的 40%，仅次于 19 世纪中后期铁路股曾创下的 63% 历史峰值。 ### [09:02](https://www.youtube.com/watch?v=EgugwJdzeOc&t=542s) AI 经济与传统经济的脱钩：冰火两重天的增长数据 《华尔街日报》（The Wall Street Journal）近期试图厘清美国整体经济与 AI 经济之间的关联，探寻当前的增长动力究竟源自何处。粗略估算显示，AI 经济的增速高达 31%，而与之形成鲜明对比的是，非 AI 经济的增长率仅有孱弱的 0.1%。从宏观指标来看，占美国 GDP 最大比重的个人消费支出仅录得 1.6% 的微弱增长；对住房、商业建筑、工厂以及卡车、飞机等运输设备的投资全线下滑。主持人们感叹，如今社会在数据中心上的建设支出甚至已经超过了写字楼和商业住宅。 与此同时，科技硬件设备的投资飙升了 43%，这主要由算力芯片和 GPU 的疯狂采购所拉动；软件投资增长了 23%，数据中心投资也激增了 22%。这种加速反弹不仅体现在软件公司的营收端，也反映在科技行业韧性的展现上。像 DoorDash 这样依靠强网络效应而非纯软件护城河的企业，在经历市场波动的冲击后，大多在一个季度到半年内证明了自己的抗风险能力。尽管仍有部分软件企业股价受挫，但科技产业链内部的高额资本开支与传统实物资本投入的低迷构成了极其割裂的经济图景。 - **AI 经济估算增速**：31%，而非 AI 经济同期增速仅为 0.1%。 - **固定资产投资分化**：科技设备投资增长 43%，软件投资增长 23%，数据中心投资增长 22%；住房、工厂、写字楼及运输设备投资均陷入萎缩。 - **个人消费支出增长**：1.6%，显示宏观消费动力相对疲软。 --- ### [10:25](https://www.youtube.com/watch?v=EgugwJdzeOc&t=625s) 4月非农爆表与 Texas Roadhouse 暴涨插曲 正当市场认为除 AI 之外的整体经济已陷入停滞时，《华尔街日报》公布的最新就业数据却给出了完全相反的信号：美国 4 月新增非农就业岗位 11.5 万个，远超分析师此前预期的 5.5 万个。尽管较 3 月修正前的 18.5 万有所回落，但这一超预期翻倍的表现直接打破了“经济即将熄火”的悲观叙事。新增就业主要集中在零售、运输、仓储和医疗保健等基础行业，而这些领域基本没有受到 AI 技术的直接冲击。 直播间里，观众 Gabe 突然插播突发消息称美式连锁牛排餐厅 Texas Roadhouse 股价单日暴涨 15% 至 20%，弹幕戏称“你对 Texas Roadhouse 还是不够看好”，主持人们借此打趣最近全网病毒式传播的“桐生一马滤镜”（Kiryu filter）迷因。回到严肃的经济话题，主播们指出当前宏观叙事存在巨大错配：许多人以为英伟达的财报和 AI 正在一肩扛起全球经济和美股大盘，但实际上拥有超过 1 亿就业人口的庞大真实经济体仍然具备极强的抗跌韧性。科技大厂裁员几千人固然在社交媒体上引发轩然大波，但在宏观就业大盘中几乎微不足道。正如毕马威（KPMG）首席经济学家 Diane Swonk 所评价的：“这依然是一个高度焦虑的就业市场——在职者牢牢抓紧工作，而求职者则深感被拒之门外。” ```mermaid flowchart TD accTitle: 劳动力市场与 AI 叙事的割裂 accDescr: 对比传统大盘就业市场的韧性与科技行业小规模裁员引发的舆论焦虑 subgraph Macro ["宏观就业大盘（超1亿就业人口）"] A["4月新增就业 11.5万人<br/>预期仅 5.5万人"] --> B["增长核心支柱"] B --> C1["医疗保健"] B --> C2["零售与仓储"] B --> C3["交通运输"] end subgraph Tech ["科技与 AI 叙事圈"] D["科技大厂裁员数千人"] --> E["社交媒体广泛焦虑"] F["英伟达财报高歌猛进"] --> G["被误以为独木支撑全美经济"] end Macro -.->|"体量悬殊，韧性托底"| Tech ``` --- ### [13:34](https://www.youtube.com/watch?v=EgugwJdzeOc&t=814s) 惠而浦对决六旗乐园：杠铃策略与“不可替代的线下体验” 为了探寻非 AI 领域真实经济的运行逻辑，节目深入对比了两家反差强烈的传统企业财报：家电巨头惠而浦（Whirlpool）与主题乐园六旗（Six Flags）。惠而浦拥有上百年历史，自 1950 年代以来历经互联网泡沫破裂和大萧条，从未中断过股息发放，如今却因沉重债务和激烈竞争宣布下调股息，导致其股价重挫，过去 5 年跌幅高达 80%。相反，本应属于极端可选消费、在经济困难时期最先被砍掉的六旗主题乐园（股票代码：FUN），其第一季度营收、入园游客人次和客单消费却全线增长，股价在经历橄榄球星特拉维斯·凯尔西（Travis Kelce）财团注资 2 亿美元后的回踩中逐步走出底部。 这两家公司的迥异走势揭示了消费行为的新分化：惠而浦生产的冰箱和洗衣机虽然是生活必需品，但属于“可延期消费的大件商品”。在成屋销售环比下跌 3% 以及 LG、三星等国际对手围剿下，消费者完全可以选择推迟更换或修补旧机器；但孩子们处于适合坐过山车的黄金年龄是不可逆且无法延期的。这种分化精准契合了 AI 时代正逐渐盛行的“杠铃投资策略”（Barbell Thesis）：一端极度拥抱 AI 基础设施，另一端则重注具有唯一性的线下真实体验或稀缺 IP。例如埃里森家族（Ellison family）一方面通过甲骨文（Oracle）重仓 AI 算力底座，另一方面通过华纳兄弟控股蝙蝠侠、超人等经典文化资产；风险投资家乔希·库什纳（Josh Kushner）则左手重仓 OpenAI，右手买入旧金山巨人队（SF Giants）。正如节目中所说，过山车是典型的“反 AI 垃圾内容”（Anti-slop）资产——“因为你不可能通过 Vibe Coding 凭空敲出一座飞越太空山（Space Mountain）。” | 对比维度 | 惠而浦 (Whirlpool) | 六旗乐园 (Six Flags) | |:--- |:--- |:--- | | **行业属性** | 耐用家用电器制造（冰箱、洗衣机） | 线下主题公园与娱乐（过山车体验） | | **消费属性** | 昂贵的生活必需品，但消费“可无限期延后” | 纯非必需可选消费，但具备成长年龄的“不可逆性” | | **近期财务动向** | 70年来首次削减股息，债务承压，过去5年股价跌超80% | 一季度营收增长，入园人数及客单消费双双提升 | | **外部制约/催化** | 成屋销售下滑 3%，面临三星/LG 激烈竞争 | 特拉维斯·凯尔西财团注资 2 亿美元，线下客流复苏 | | **AI 杠铃定位** | 容易被宏观周期与国际供应链替代的夹心层 | 无法被虚拟化或 AI 生成替代的“Anti-slop”实体资产 | --- ### [17:41](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1061s) 英国《金融时报》的人工智能终局图景：人类灭绝还是无限富足？ 节目随后讨论了 Flexport 创始人瑞安·彼得森（Ryan Petersen）在社交平台上分享的一张来自《金融时报》（Financial Times）的未来推演图表。该图表严肃而具象地探讨了人类面临的几种截然不同的 AI 终局：在最极端的超级乐观情景（即终结匮乏的技术奇点）下，全球人均实际 GDP 将突破 100 万美元大关；而在人类灭绝的情景下，一切产出与经济数据归零。 在灭绝与天国这两个极端之间，还存在着一条 AI 驱动稳步增长的现实基准路径。主持人们认为，无论是追求稳步攀升的增长斜率，还是奔向消除稀缺性的终极繁荣，这才是全人类与科技界应该全力争取的目标。然而，正是因为未来的不确定性在“百万人均 GDP”与“文明归零”的两极之间剧烈摇摆，当前市场才蔓延着一种挥之不去的深层焦虑感。 ### [18:30](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1110s) AI 泡沫的双重焦虑与估值辩证 关于当前人工智能浪潮究竟是真正的科技飞跃还是资本泡沫，市场上存在一种奇特的“双重焦虑”：一方面，如果 AI 进展过于迅猛，人们会陷入对大规模失业的极度恐慌；但另一方面，如果 AI 被证实只是过度炒作的泡沫并迅速破裂，整体经济便会滑入衰退，引发同样灾难性的失业潮。无论站在技术末日论还是经济过热论的立场上，人们所面临的经济不确定性风险在本质上极为相似。 不过，两位主持人指出，尽管目前市场上各大 AI 公司的估值已经高得令人咋舌，但如果仔细审视它们的营收增长曲线，会发现业务增速同样惊人，因此现在断言泡沫破裂还为时过早。 ### [19:19](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1159s) “餐巾纸计算”对决“信封背面计算” 在进入严肃数据分析前，主持人之间展开了一场充满极客幽默的争论：做商业构想时，究竟属于“餐巾纸计算”（Napkin Math）派还是“信封背面计算”（Back of the Envelope Math）派？一人随手拿出了此前参加 WonderCo 活动带回来的餐巾纸留念，但随即吐槽在薄薄的餐巾纸上写字其实极其困难，稍微用力就会划破，相比之下信封不仅纸质平整、面积充足（甚至能达到标准的 8.5×11 英寸规格），还能列出多列单元格画出一整张小型电子表格，更具“严肃商业工具”的体面感。 另一方则坚定捍卫“餐巾纸哲学”，认为餐巾纸代表了一种商业直觉：“如果一个商业机会的逻辑无法在一张小小的餐巾纸上讲清楚，那它就根本不值得让人兴奋。”信封承载的内容太多太复杂，反而失去了直击本质的魅力。聊天室观众也插话提醒，餐巾纸在生活中的可获取性远超信封，许多传奇的商业合作并非诞生于正式会议室，而是在餐馆酒席间随手抓起餐巾纸勾勒出来的；两人还顺带调侃了在高级餐厅布质餐巾上乱画的失礼行为，以及家长是否该制止孩子拿蜡笔在桌布上涂鸦的日常趣事。 ### [20:48](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1248s) 实体化“每日邮件箱”：现代通讯焦虑与反向创业畅想 主持人分享了近期被海量即时通讯工具压得喘不过气的心力交瘁感——每天睁开眼就是处理不完的 Slack 消息、各类工作邮件，以及 Instagram、X（Twitter）、LinkedIn 等各大平台的私信。为此他脑洞大开地提出了一个“反向创新”服务：用户把所有收件箱接入该服务，平台每天集中将所有新消息打印成纸质文件，并在固定时间（比如两小时的时间窗口内）投递到家门外的实体小盒子里。 用户只需每天一次走到屋外取出信件翻阅，决定哪些需要回复。这种一天仅处理一次的物理仪式感，能强迫人们更加审慎、深思熟虑地撰写回信，摆脱全天候即时通知的神经衰弱。主持人打趣这与虚拟邮箱服务 Earth Class Mail 恰好相反，聊天室观众则纷纷无情吐槽：“这个革命性服务不如命名为‘美国邮政署’（USPS）吧！”或者“干脆把这个箱子数字化，改叫它 Email 算了”。他们还提到唐纳德·特朗普（Donald Trump）其实一直践行着类似的极简习惯：助理负责把邮件打印出来，他用粗马克笔在纸上批注回复，再由助理扫描发回。 ```mermaid flowchart LR accTitle: Physical Inbox Service Concept accDescr: A satirical workflow turning digital notifications back into physical mail. subgraph Digital["数字通讯碎片"] D1["Slack / 邮件"] D2["X / IG / LinkedIn 私信"] end subgraph Aggregator["每日打印与递送"] P1["集中汇总并打印纸质文件"] P2["每天固定时间投递"] end subgraph Physical["实体仪式感"] B["门外实体收件箱"] R["用户每日翻阅 / 审慎批复"] end Digital --> Aggregator Aggregator --> Physical ``` ### [23:41](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1421s) 半导体市场的极度过热与“无降落伞跳伞”隐喻 话题转向炙手可热的半导体板块。分析师 Justin Spittler 刚刚发帖提醒投资者在当前点位买入半导体股必须极度谨慎，市场情绪已经过热到了极点。节目画面随后切到极限运动传奇特拉维斯·帕斯特拉纳（Travis Pastrana）在 Nitro Circus 中的经典震撼视频：帕斯特拉纳完全不背降落伞，手里甚至还悠闲地拿着一罐红牛饮料，直接从万米高空的飞机上一跃而下。 他在空中一边做特技动作一边喝饮料，任由下坠的紧张感不断累积，直到最后时刻才与空中接应的队友紧紧相拥、扣紧安全背带并开伞落地。主持人感叹，在当前估值处于历史天花板的背景下追高半导体股票，心理体验简直就像这趟不带降落伞的盲目高空跳伞——全凭空中接应的运气，稍有闪失便是粉身碎骨。 ### [24:49](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1489s) Tyler 探访奥克兰法庭：OpenAI 诉讼案首日与 Mira Murati 证词回放 节目的制作人 Tyler 讲述了自己过去两天亲赴加州奥克兰联邦法院，现场旁听埃隆·马斯克起诉 OpenAI 及山姆·奥特曼（Sam Altman）首日庭审的经历。为了确保能拿到向公众开放的入场名额，他在周二节目结束后直奔机场，周三凌晨 5:30 就顶着寒风赶到法院大门外排队。 - **严苛的旁听席位**：该案属于联邦法庭审理，法律规定必须向公众开放席位，但现场旁听配额仅有 20 到 30 个左右，其中还预留了相当一部分给持证媒体，普通公众名额极为紧俏。 - **与主流媒体同台竞争**：Tyler 抵达时已有三四人在排队；知名科技记者迈克·艾萨克（Mike Isaac）因为代表《纽约时报》持有一张与凯德·梅茨（Cade Metz）轮流共用的专用媒体通行证，直到清晨 7:30 至 8:00 才现身。 - **首日庭审节奏与焦点**：庭审从早上 8:30 一直持续到下午 2:00 左右，法官开场致辞、陪审团入场，中间穿插了两次 20 分钟的休庭。 - **Mira Murati 录像作证**：庭审核心环节播放了 OpenAI 前首席技术官米拉·穆拉蒂（Mira Murati）的宣誓录像证词，重点还原了 2023 年底管理层与董事会“罢免奥特曼风波”的具体时间线，并首次在法庭上公开展示了核心当事人之间的大量内幕短信记录。 ### [27:01](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1621s) 庭审证据魔改：将萨姆·奥特曼短信改编为 Emo 歌曲与 Gen Z 烂梗 节目开场展示了一段在网络上走红的 AI 创意作品：有人把萨姆·奥特曼（Sam Altman）发给米拉·穆拉蒂（Mira Murati）的短信记录，改编成了一首 2011 年风格的青少年 Emo 伤感流行乐（teeny teenage heartthrob anthem）。伴随着极为契合的动效图形，歌曲唱出了两人的原话：“你能指示一下方向是好还是坏吗？”（Can you indicate directionally good or bad?）、“萨蒂亚和其他人都很焦虑，方向上非常糟糕”（Satya and others anxious, directionally very bad）。主持人打趣称这段音频应该拿去放给 AI 怀疑论者听，看看他们面对如此成果还怎么忍得住不去动手创造。 制作该作品的作者透露了其具体工作流程：先利用 OCR 工具识别法庭公开的短信截图转化为纯文本，剔除对话中的人名后输入到 Suno 中，前后迭代生成了 20 到 30 次，最终选出了这段洗脑的旋律。随后主持人甚至脑洞大开，读起了一段由 Gen Z “大脑过载烂梗”（brain rot slang）重写的版本：“Yo fam，帮我 vibe check 一下，我是不是真要凉了（am I cooked or nah for real for real）？萨蒂亚这帮人 low-key 焦虑到飞起……”。 ```mermaid flowchart LR accTitle: 短信转歌曲生成工作流 accDescr: 展示将法庭披露的短信记录制作成 Suno 歌曲的三个步骤阶段。 subgraph S1["数据提取"] A["短信截图证据"] --> B["OCR 文字识别"] end subgraph S2["文本清洗"] B --> C["去除对话发言人姓名"] end subgraph S3["AI 音乐生成"] C --> D["输入 Suno 生成音乐"] D --> E["迭代生成 20-30 次选优"] end ``` ### [28:53](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1733s) OpenAI 庭审现场见闻：沉闷的录像放映与证人出庭 随后话题切回当天的法庭现场。整场听证会的前一个小时几乎全在播放米拉·穆拉蒂此前录制的作证录像（deposition video），形式极其枯燥，宛如中学课堂上“老师不想讲课直接给学生放电影（movie day）”。法官中途甚至忍不住发火，担心陪审团会无聊到睡着，甚至当场张罗着给陪审团倒咖啡提神。在看完全部短信证据后，前 OpenAI 董事会成员希冯·齐里斯（Shivon Zilis）亲自出庭作证。希冯在 2020 年至 2023 年担任董事，直到埃隆·马斯克（Elon Musk）创立 xAI 之前的 2023 年 2 月才退出董事会，正好亲历了诸多核心矛盾。 现场旁听的主持人回忆，听证会的焦点反复在两个时间线之间跳跃：一段是 2018 年前后萨姆·奥特曼、格雷格·布罗克曼（Greg Brockman）、伊利亚·苏茨克维尔（Ilya Sutskever）与埃隆·马斯克之间的权力博弈；另一段则是 2023 年 11 月引发轰动的奥特曼被罢免事件。希冯作证后，法庭又播放了长达一小时的海伦·托纳（Helen Toner）录像，内容全围绕罢免风波展开。 - 旁听席上座无虚席，普通观众席基本全由没拿到正规媒体证的科技记者填满，整个区域回荡着疯狂敲击笔记本键盘记录证词的声音。 - 庭审使用的行话并没有想象中那样高深硬核（不如 Dwarkesh Patel 播客的技术密度），律师们提问尤为浅白，而希冯作证时则聊到自己为何进入 AI 行业并提及了“加速主义”（acceleration）等圈内术语。 - 现场还发生了一些趣事：同行记者迈克·艾萨克（Mike Isaac）连日作战带上了专属臀部坐垫（butt pillow），而主持人则因嫌准备零食麻烦，全程硬撑断食看完了庭审。 ### [31:33](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1893s) 幻方量化掌舵人自掏腰包：DeepSeek 估值 500 亿美元融资 70 亿 话题转入 AI 行业的重磅资本动态。深度求索（DeepSeek）正筹备以 500 亿美元的估值募集一笔高达 70 亿美元的巨额融资，这将成为中国人工智能领域有史以来规模最大的一笔融资。最让外界感到震撼的是其创始人的出资比例：创始人（梁文锋）个人将直接出资该轮融资的 40%——即掏出整整 30 亿美元的个人资金注入公司。 在此次融资前，创始人个人持有 DeepSeek 约 90% 的股份，如此高比例的个人持股与巨额自投在当前百亿美元体量的科技独角兽中绝无仅有。DeepSeek 诞生于创始人旗下极为成功的量化对冲基金（幻方量化），其自掏腰包的核心动力是为了全力扫货算力。尽管图表显示近期中国开源模型在性能增长曲线上稍有落后，但他押注通过庞大算力储备在数月内重新回到最前沿，打造出新一代 Frontier 模型。 ### [32:21](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1941s) SpaceX/xAI 与 Anthropic 算力交易：宿敌合作与 Neo-Cloud 崛起 节目最后讨论了当周突发的重量级交易：扎克·布拉克斯（Zack Brax）在社交媒体上调侃称“祝贺 Anthropic 在市场上击败 Grok，并尽情享用手下败将的算力遗存”。外界普遍关注马斯克旗下的基础设施力量（SpaceX / xAI）与 Anthropic 达成的算力合作。数月前马斯克还曾在公开场合对 Anthropic 团队出言不逊，此前多数观察家虽认为两家合作在商业理性上行得通，但都判定强烈的文化差异会阻碍合作落地，然而“对算力的极致渴求打破了一切壁垒”。 主持人分析，这坐实了马斯克旗下实体转型为“新型云服务商”（Neo-Cloud）的趋势。凭借极其惊人的基建落地速度和快速通电能力，为外部大模型团队转售算力是极为顺理成章的变现路径。此前外界曾认为 Cursor 的采购协议已经吃满了这些基础设施资源，但随着马斯克多个 Colossus 超级数据中心集群的持续投产与扩张，其庞大的算力底座完全有余力同时支撑多种超级客户的需求。 ## Notable quotes > [04:13](https://www.youtube.com/watch?v=EgugwJdzeOc&t=253s) 只要我们（政府）一进场，苹果进来了，Nvidia 进来了，很多聪明绝顶的人全进来了，政府已经在这笔交易里赚到了几百亿美元。 — 唐纳德·特朗普 > [10:25](https://www.youtube.com/watch?v=EgugwJdzeOc&t=625s) 这依然是一个高度焦虑的就业市场——在职者牢牢抓紧工作，而求职者则深感被拒之门外。 — Diane Swonk > [13:34](https://www.youtube.com/watch?v=EgugwJdzeOc&t=814s) 因为你不可能通过 Vibe Coding 凭空敲出一座飞越太空山（Space Mountain）。 — 主持人 > [19:19](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1159s) 如果一个商业机会的逻辑无法在一张小小的餐巾纸上讲清楚，那它就根本不值得让人兴奋。 — 主持人 > [27:01](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1621s) 你能指示一下方向是好还是坏吗？ — 萨姆·奥特曼 > [27:01](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1621s) 萨蒂亚和其他人都很焦虑，方向上非常糟糕 — 米拉·穆拉蒂 > [27:01](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1621s) Yo fam，帮我 vibe check 一下，我是不是真要凉了（am I cooked or nah for real for real）？萨蒂亚这帮人 low-key 焦虑到飞起…… — 主持人 > [32:21](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1941s) 祝贺 Anthropic 在市场上击败 Grok，并尽情享用手下败将的算力遗存 — 扎克·布拉克斯 > [32:21](https://www.youtube.com/watch?v=EgugwJdzeOc&t=1941s) 对算力的极致渴求打破了一切壁垒 — 主持人 ## Mentioned 苹果 (Apple), 英特尔 (Intel), 台积电 (TSMC), 英伟达 (Nvidia), 特斯拉 (Tesla), SpaceX, xAI, OpenAI, Anthropic, 深度求索 (DeepSeek), 幻方量化 (High-Flyer), 微软 (Microsoft), 甲骨文 (Oracle), 惠而浦 (Whirlpool), 六旗乐园 (Six Flags), 德州公路牛肉餐馆 (Texas Roadhouse), DoorDash, 博通 (Broadcom), 美光 (Micron), AMD, 三星电子 (Samsung Electronics), Suno, Cursor, Flexport, 华尔街日报 (The Wall Street Journal), 金融时报 (Financial Times), 纽约时报 (The New York Times), 约翰·库根 (John Coogan), 霍尔迪·海斯 (Jordi Hays), 蒂姆·库克 (Tim Cook), 黄仁勋 (Jensen Huang), 埃隆·马斯克 (Elon Musk), 唐纳德·特朗普 (Donald Trump), 霍华德·卢特尼克 (Howard Lutnick), 陈立武 (Lip-Bu Tan), 帕特·基辛格 (Pat Gelsinger), 罗唯仁 (Wei-Jen Lo), 本·汤普森 (Ben Thompson), 格雷格·伊普 (Greg Ip), 黛安·斯旺克 (Diane Swonk), 萨姆·奥特曼 (Sam Altman), 米拉·穆拉蒂 (Mira Murati), 萨蒂亚·纳德拉 (Satya Nadella), 希冯·齐里斯 (Shivon Zilis), 格雷格·布罗克曼 (Greg Brockman), 伊利亚·苏茨克维尔 (Ilya Sutskever), 海伦·托纳 (Helen Toner), 梁文锋, 特拉维斯·凯尔西 (Travis Kelce), 特拉维斯·帕斯特拉纳 (Travis Pastrana), 乔希·库什纳 (Josh Kushner), 瑞安·彼得森 (Ryan Petersen), 泰勒 (Tyler), 迈克·艾萨克 (Mike Isaac), 凯德·梅茨 (Cade Metz), 扎克·布拉克斯 (Zack Brax), 14A 先进制程, Colossus 超级计算机集群, Grok, iPhone, Mac mini, Mac Studio ## Video description Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive [Source](https://www.youtube.com/watch?v=EgugwJdzeOc)

[4:13] 唐纳德·特朗普: 只要我们（政府）一进场，苹果进来了，Nvidia 进来了，很多聪明绝顶的人全进来了，政府已经在这笔交易里赚到了几百亿美元。 [10:25] Diane Swonk: 这依然是一个高度焦虑的就业市场——在职者牢牢抓紧工作，而求职者则深感被拒之门外。 [13:34] 主持人: 因为你不可能通过 Vibe Coding 凭空敲出一座飞越太空山（Space Mountain）。 [19:19] 主持人: 如果一个商业机会的逻辑无法在一张小小的餐巾纸上讲清楚，那它就根本不值得让人兴奋。 [27:01] 萨姆·奥特曼: 你能指示一下方向是好还是坏吗？ [27:01] 米拉·穆拉蒂: 萨蒂亚和其他人都很焦虑，方向上非常糟糕 [27:01] 主持人: Yo fam，帮我 vibe check 一下，我是不是真要凉了（am I cooked or nah for real for real）？萨蒂亚这帮人 low-key 焦虑到飞起…… [32:21] 扎克·布拉克斯: 祝贺 Anthropic 在市场上击败 Grok，并尽情享用手下败将的算力遗存 [32:21] 主持人: 对算力的极致渴求打破了一切壁垒

[17:41] In the bull case, real GDP per capita goes north of a million dollars. And in the human extinction scenario, it goes to zero, of course. But the AI boosted growth path is a steady trend up... [27:17] Yo fam, can you give me a vibe check on am I cooked or nah for real for real? Satya and the gang low-key stressing. Mira Murati: yeah, you're skibidi. This isn't a rare L... Little bro, who ratioed me?

[0:00] The great lock-in continues. We will go [0:02] into a review of the last few days. It's [0:05] Friday. [0:05] >> Hey, we're having fun. We're going to be [0:07] hanging out here on the stream telling [0:09] you about the American economic roller [0:12] coaster. You've heard this story [0:15] >> before. [0:17] >> [laughter] [0:18] >> It's just non-stop. [0:19] Okay, so the American economic roller [0:21] >> back. We miss you guys. [0:22] >> It's good to be back. Yeah. [0:25] We got a bunch of stuff going on. A [0:26] whole bunch of economic news over the [0:28] last couple days. Stock market is [0:29] absolutely ripping [0:30] >> We should be in white suits. Intel's up [0:31] 20% today. So almost. And there's this [0:34] split between the AI economy and the [0:36] real economy, the American economy. And [0:38] there's a whole bunch of different [0:39] things that like you need to puzzle [0:40] together to get a picture of what's [0:42] going on. And then I'm discovering that [0:44] there are K shapes within the K shapes. [0:47] Even in the real economy, there's divide [0:49] between different companies. And so I [0:51] wanted to walk through that. So let's [0:52] start with the Intel news. It's up [0:53] almost 20% today on the news that they [0:55] will be making chips for Apple. There's [0:58] a report in the Wall Street Journal [0:59] about this. They've really [1:00] >> reached a preliminary chip making [1:02] agreement. It's not completely locked [1:04] in, but there's a whole bunch of extra [1:06] context here that we've been tracking [1:08] for the last 6 months, 12 months as some [1:10] of this was expected. The revitalization [1:13] of Intel was something that a lot of [1:14] folks were clamoring for. Ben Thompson, [1:17] myself, a lot of folks. We're all hoping [1:18] for something and and plans are starting [1:21] to come together. So the talks have been [1:22] described as intensive, intensive talks [1:25] between the two companies have been [1:26] going on [1:27] for more than a year. Apple's been [1:28] talking to Intel. [1:29] >> Interesting. Of course, Intel and Apple [1:31] have been long-term partners for the [1:34] what? All the 90s, 2000s up until the [1:37] Apple Silicon program where they started [1:40] going Apple started going direct to [1:41] TSMC. Can they get them back? It seems [1:43] like potentially in some form factors, [1:45] but we're going to dig into it. So they [1:47] have hammered out a formal deal in [1:49] recent months. And it's still unclear [1:51] which exact Apple products Intel would [1:54] make chips for and manufacture them both [1:56] its own designs. So, Intel has two main [1:58] businesses. It's both a design shop and [2:00] a manufacturer. They have a fab and a [2:03] and a design unit. In its foundry unit, [2:06] both businesses have been [2:07] underperforming for years before Pat [2:09] Gelsinger came in as the new CEO and was [2:13] vowing to revitalize them. And so, last [2:15] summer, the Trump administration struck [2:16] a deal to convert nearly $9 billion in [2:19] federal grants into Intel stock, giving [2:21] the US a 10% stake. [2:24] >> 2021 a share. It's now at $125 a share. [2:28] So, [2:28] >> White House has got to be feeling pretty [2:30] good. [2:30] >> Yeah, huge. I mean, I'm pretty sure [2:32] Jensen has made over a billion dollars [2:35] on the $5 billion investment. I mean, [2:36] just today, if it's up 20% and he must [2:39] have seen a big growth in that position [2:41] since then. So, [2:43] just today probably a one or two [2:45] billion-dollar print. And so, [2:47] the reporting here in the journal says [2:49] that the Trump administration was [2:50] actually key in bringing Apple to the [2:52] table, putting pressure on sort of both [2:54] sides to say, "Hey, let's think about [2:56] the future of American manufacturing [2:58] resiliency, reducing Taiwanese [3:01] dependence on a on a foreign supply [3:03] chain, and giving Intel a real shot at [3:05] underwriting the next big fab that they [3:07] want to [3:07] >> Nvidia got in at $23 a share with their [3:10] $5 billion investments. So, [3:13] uh uh [3:14] around a 5x. [3:15] >> Not bad. That's not bad at all. [3:17] >> really needed it. [3:18] >> He paid for Grok with that. That's [3:20] pretty sick. [3:21] >> Oh, yeah. [3:21] >> He just paid He paid for Grok [3:23] >> for perfectly. [3:24] >> That's why he wired [3:25] >> before they actually closed. [3:25] >> like, "I'm up. I'm I I I I I I I I I I I [3:27] I I I I I I I I I I I I I I I I I I I I [3:28] I I I I I I I I I I I I I I I I I I I I [3:28] I I I I I I I I I I I I I I I I I I I I [3:29] I I I I I I I I I I I I I I I I I I I I [3:30] I I I I I I I I I I I I I I I I I I I I [3:31] I I I I I I I I I has met repeatedly [3:33] over the last year with high-ranking [3:35] Apple officials, including CEO Tim Cook, [3:37] as well as SpaceX chief Elon Musk, [3:39] Nvidia chief Jensen Huang [3:41] to try and convince them to get into [3:42] business with Intel. There was a [3:43] discussion about should Nvidia dual [3:45] source the Grace CPUs from Intel. [3:49] SpaceX and Elon's been with Samsung for [3:51] a lot of Tesla chips, but there's been [3:54] uh discussion there. There was that [3:55] rumor about Global Foundries, and Elon's [3:57] always been in the in and around the fab [3:59] world, uh and he's obviously going a lot [4:02] deeper with the long-term TeraFab [4:03] project. So, Commerce Secretary Howard [4:05] Lutnick was trying to convince them to [4:06] get into business with Intel, some of [4:07] the people familiar with the matter [4:08] said. With With the Apple deal, Intel is [4:10] now signed partnerships with all three. [4:13] So, over the last decade, Intel fell [4:15] badly behind rivals, this is from the [4:16] Wall Street Journal, such as TSMC and [4:18] Samsung Electronics after a series of [4:20] technical mishaps, leadership changes, [4:22] and failed attempts at consolidation led [4:24] outside foundry customers to pull or [4:26] curb their businesses. When [4:29] Intel hired Tan in March of 2025, wow, [4:31] just a year ago, to replace ousted Chief [4:34] Executive Pat Gelsinger, President Trump [4:36] raised concerns that Tan's close ties [4:38] with China would compromise him and [4:40] called for his ouster. It was a very, [4:41] very dramatic moment, but Tan won Trump [4:45] over with a charm offensive. I like [4:47] that. And the government announced its [4:49] 10% investment in Intel shortly after. [4:51] Following the investment, Intel shares [4:53] uh price rose sharply. On Friday [4:56] morning, it rose 7 and 1/2% to an [4:57] all-time high of nearly $118 per share. [5:00] It's up more now. Tan has been reshaping [5:02] Intel's top leadership ranks in recent [5:04] months, uh including hiring former TSMC [5:07] executive Wei-Jen Low, a move that [5:09] prompted a lawsuit from TSMC. So, [5:11] they're not friends anymore. Intel's CEO [5:13] has also ousted his head of product and [5:15] hired new executives to lead the [5:17] company's data center processor and [5:19] client computing units, as well as newly [5:21] formed custom silicon business. He's [5:23] also invested heavily in Intel's most [5:25] advanced manufacturing process known as [5:26] 14A, and that is the node that uh Intel [5:29] is hoping that all of the uh potential [5:32] customers will jump together and say, [5:34] "Hey, we're going to buy from this if [5:36] you make it." If you build it, we will [5:37] come. President Trump personally [5:38] advocated for Intel to cook in a [5:40] meeting, which is funny because it [5:42] sounds like he's advocating for Intel [5:44] just to cook generally, but he's No, [5:47] he's he's advocating for Intel uh to Tim [5:50] Cook in a meeting at the White House [5:52] according to people familiar with the [5:53] matter. [5:54] Trump said, "I like Intel in January." [5:57] He said that the government [5:58] made tens of billions of dollars from [6:00] Intel deal and that the government's [6:02] backing of the company had attracted [6:03] important partners to Intel. "As soon as [6:05] we went in, Apple went in, Nvidia went [6:07] in, a lot of smart people went in." [6:09] Trump said. Nvidia, the world's largest [6:11] chip firm, invested 5 billion in Intel [6:13] in September and the two companies [6:14] announced partnership under which Intel [6:16] would build custom data center CPUs, the [6:18] processing brains for most computer [6:20] systems, for Nvidia. And last month, [6:22] Elon, of course, announced the Terafab [6:24] project. [6:25] And so, Apple relies on TSMC to make [6:27] chips for iPhones, iPads, Macs, and [6:30] other devices and is under pressure to [6:31] find additional chip suppliers. There's [6:33] also chip shortages, so it could be good [6:36] to dual source in independent of the [6:38] geopolitical discussion. [6:40] On Apple's last two earnings conference [6:42] calls, Cook blamed a lack of [6:45] availability of advanced chips for [6:47] Apple's inability to meet customer [6:48] demands for iPhones. The constraints are [6:50] expected to continue in the current [6:52] quarter affecting several Mac models, [6:53] Cook said. Quote, "We think looking [6:56] forward that the Mac mini and the Mac [6:58] Studio may take several months to reach [7:00] supply-demand balance." [7:02] Um And after the earnings call, Apple [7:05] raised the Mac mini starting price. And [7:07] so, TSMC's manufacturing capacities [7:09] capabilities far surpass those of [7:11] Samsung and Intel, makers of other kinds [7:13] of chips for memory and storage and [7:15] storage, for example, are more [7:16] competitive with one another giving [7:18] Apple multiple sources of supply. [7:20] Although, of course, they are memory [7:21] constrained. Apple's long been TSMC's [7:23] top customer, but skyrocketing demand [7:26] for its manufacturing capacity from [7:27] Nvidia and other designers of AI chips [7:29] means Apple's no longer has as much [7:33] leverage to secure the supplies that it [7:34] needs. Starting in 2006, Apple used [7:37] Intel designed CPUs as its main [7:39] processor of its personal computers, but [7:40] switched to its own custom CPUs [7:43] based on an arm design in 2020. That's [7:46] the dawn of Apple silicon. [7:48] And so there's been an incredible [7:51] economic and financial performance [7:54] concentrated in just a handful of [7:56] trillion-dollar tech companies. Intel's [7:58] starting to join and starting to perform [8:00] like that. There's a few memory stocks. [8:02] I saw one report that referred to it as [8:04] the mag 10 and they'd added a few other [8:07] AI names to that group, but there's just [8:10] a few companies that are driving the [8:11] vast majority of returns in the stock [8:13] market. [8:14] >> We had three sort of beautiful weeks [8:17] where pretty much everyone started [8:19] saying there's no way it's a bubble. [8:20] >> Yeah. [8:21] >> Look at the revenue growth. There's no [8:22] way it's a bubble. But now people Steve [8:25] over at Bloomberg, we can pull this [8:26] chart up. The concentration [8:28] >> Sure. [8:29] I've seen this. [8:31] >> the market you can see [8:33] >> The AI big 10. So mag 7 plus AMD, [8:38] Broadcom, and one other. Who is that? [8:41] Micron. [8:41] >> Yeah. [8:42] >> Now make up [clears throat] 40% of the [8:44] market. Railroads for reference during [8:47] 1835 to 1910 were 63% of the market. And [8:52] he comps it to Japan, the Nifty 50, tech [8:55] and telecom throughout various periods [8:57] of time. And so I was reading this post [8:59] by or this piece by Greg Ipp in the Wall [9:02] Street Journal and he was trying to [9:04] disentangle what's happening in the [9:07] overall American economy from the AI [9:09] economy. What's where is the growth [9:12] coming from? He sort of back the [9:14] envelope did. He said the AI economy [9:16] grew 31% while the non-AI economy just [9:19] 0.1% [9:21] and he cites a few economic statistics [9:23] that are recent. Personal consumption, [9:26] the biggest component of GDP grew a [9:28] relatively muted 1.6%. Investment fell [9:31] in housing, business structures. I think [9:33] we're building we're spending more in [9:34] data centers than housing now or office [9:37] buildings I guess. I I think it eclipsed [9:38] office buildings Uh and factories and [9:40] transportation equipment like trucks and [9:42] aircrafts. Meanwhile, investment soared [9:44] 43% in tech equipment. Obviously, that's [9:47] chips and GPUs. 23% in software, [9:50] surprising, [9:51] and 22% in data centers. [9:53] >> meaningful bounce back. Certainly on the [9:55] revenue side. [9:55] >> Absolutely. [9:57] >> there's been some [9:59] acceleration. [10:00] >> It really was extremely widespread where [10:02] you had, I mean, you had like DoorDash [10:05] and and and companies with strong [10:07] network effects where the software was [10:09] not their moat, they still had to go and [10:11] answer to the market and some of them [10:13] got in and out of that in a week or a [10:16] month and some of them it took a quarter [10:17] or two to show that there's resiliency. [10:19] Some of them are still beaten up beaten [10:20] down, but overall, [10:22] you know, software is still doing very [10:23] well. On the flip side, there's another [10:25] headline that hit the journal yesterday. [10:28] US adds 115,000 jobs in April with solid [10:32] hiring across sectors. [10:34] >> Retail, transportation, warehousing, and [10:37] healthcare [10:38] >> All showed All showed strong growth and [10:41] led to expectation-beating jobs growth. [10:45] And so, [10:46] this is [10:47] the weird dynamic that you know, you're [10:50] seeing all of this growth in the in the [10:52] AI economy and yet [10:54] the overall employment is still chugging [10:57] along. We can pull up the chart from the [10:59] Wall Street Journal, but the US job [11:00] market blew past expectations in April [11:03] buoyed by gains in in industries [11:05] including retail, transportation, [11:07] warehousing, healthcare. Not completely [11:09] unexpected given that those are not [11:11] particularly AI targeted [11:13] areas, whereas you might see, you know, [11:15] the layoffs in the tech community, but [11:17] that doesn't it's just not enough [11:20] because the the US employs over 100 [11:23] million people and so a layoff of 4,000 [11:26] people at some tech company just doesn't [11:28] really move the needle when you're [11:29] talking about it. [11:29] >> Breaking news from Gabe in the chat. [11:31] Texas Roadhouse is up. [11:33] >> What? Why? He says 20%. I'm seeing 15%. [11:36] >> Why is it up so high? [11:37] >> Uh, he says you're not bullish enough on [11:40] Texas Roadhouse. I have never been. [11:42] >> Maybe they beat uh [11:43] >> Yeah, yeah. [11:44] >> Everyone says this is a Kiryu [11:46] filter. It really it really is. [11:48] >> [laughter] [11:49] >> It's crazy how vi- how viral that [11:51] >> Incredibly viral. [11:51] >> that meme format has gone. I thought [11:53] that was just like a thing that I I sent [11:55] to you because I'd be dying laughing [11:57] about it. [11:58] >> No, no, no. It's quite pop- quite [11:59] popular. [12:00] >> broken containment fully. The American [12:01] economy added 115,000 jobs in April. [12:04] Uh, this was down from a net gain of [12:06] 185,000 [12:08] jobs in March. Uh, but it was much [12:10] better than what expectations were for [12:13] April. Uh, analysts pulled by the Wall [12:15] Street Journal were expecting 55,000 [12:18] jobs. And the real number came in more [12:20] than twice that, which is like it just [12:23] breaks the narrative a lot. There is [12:24] this disconnect between like the AI [12:26] economy and the real economy. I was [12:28] talking to Sagar about this. It's like [12:29] >> in some ways it's like AI is Nvidia [12:33] earnings is holding up the global [12:34] economy and it's holding up the global [12:35] stock market. It's not actually holding [12:38] up the the the economy if you view the [12:39] economy as all of the different jobs and [12:43] and and activities that happen. Even if [12:45] there isn't incredible growth there, [12:47] there's actually a surprising amount of [12:49] strength and resiliency. So, the [12:51] unemployment rate has stayed [12:52] >> Personally, I'm I'm waiting to hit the [12:54] gong until we see what the revisions [12:56] come at in at. [12:57] >> We haven't seen any revisions from March [12:59] this time. [13:00] >> Yeah, yeah. It comes like it comes like [13:02] it comes like over six months later. [13:03] >> Uh, the April report coming in after [13:05] strong job gains in March shows how the [13:08] labor market is holding up better this [13:10] year than last. While health care is [13:11] still leading the way in job gains, [13:12] other sectors now appear to be picking [13:14] up. Businesses are seeing conditions [13:16] stabilizing and they have weathered the [13:18] tariffs, so many are hiring. It's [13:20] looking somewhat better than it did last [13:22] year. Diane Swonk, chief economist at [13:24] KPMG, says it's still a high anxiety job [13:26] market. Those who have a job are clearly [13:28] clinging on, while those who are looking [13:31] for a job are feeling frozen out." I was [13:34] I was digging into, you know, where [13:37] where is there strength within the real [13:38] economy? Once you go outside of AI, what [13:41] is going on? And [13:43] there were two recent earnings reports [13:45] that were sort of [13:47] disconnected, but [13:48] showed a little bit of the picture of [13:50] what's going on. So, it was Whirlpool [13:52] and Six Flags. Both had very different [13:55] reports. So, Whirlpool you probably know [13:57] from refrigeration and [13:59] washing machines, dryers, that type of [14:01] appliance. They've been making [14:03] appliances for for over 100 years, and [14:05] they've also been paying a dividend [14:07] since the 1950s consistently. They've [14:09] never They've never suspended their [14:10] dividend even through all the Great [14:13] Recession, the dot-com crash. Every year [14:15] they've been able to pay that dividend. [14:17] They just cut their dividend, uh which [14:19] is a really, really big deal for this [14:21] company since it's a dividend stock. The [14:23] stock traded down on the news, and the [14:25] stock is down 80% of the past 5 years. [14:28] Um and they're in some financial [14:29] trouble. Like they have a lot of debt, [14:31] and they have a lot of competition. But [14:34] you could play that as like, "Okay, the [14:37] real economy is like completely chucking [14:39] to a halt." At the same time, you had [14:41] Six Flags, which is should be the thing [14:44] that is the most discretionary. Like do [14:46] you go to the roller coaster theme park [14:48] or not? And Six Flags just reported [14:51] higher first quarter revenue. They're [14:53] growing, and they're growing attendance [14:54] and customer spending. And so, uh [14:57] now Six Flags is not exactly a [14:58] juggernaut of business. It's only worth [15:01] 2.3 billion. Uh and the stock is down [15:03] over the past 2 years by 50%. Uh but the [15:05] business is growing, and you wouldn't [15:07] expect that during a time of like deep [15:09] economic weakness. And so, there's [15:11] something odd going on there where as [15:14] you think of refrigeration as extremely [15:15] necessary, roller coasters as the [15:17] ultimate like you don't you you can [15:19] definitely skip it if you are [15:21] cash-strapped. but the actual dynamics [15:24] of the market are very different. So, [15:25] Whirlpool sells big-ticket necessities, [15:29] these these refrigerators, [15:31] but they are deferrable. So, you can put [15:34] off getting a new refrigerator, you can [15:35] repair an old refrigerator if money is [15:37] tight, but your kids are only really [15:39] roller coaster age for a limited time. [15:42] And Whirlpool also faces brutal global [15:45] competition, and existing home sales are [15:47] down 3% month over month. And so, all of [15:50] that drives fewer appliance upgrades, [15:52] and they are not necessarily a [15:53] beneficiary of all the international [15:55] competition that they face from LG and [15:56] Samsung and other international players. [15:59] And so, [16:00] I I've sort of come [16:01] >> correction, Six Flags or or new [16:03] information Six Flags has has been [16:06] up since [16:08] since around around November. It sort of [16:11] bottomed, Travis. I remember I was [16:12] thinking of this cuz one, they have the [16:14] ticker FUN, which is fantastic. The [16:16] stock has not been having fun, but [16:18] Travis Kelce [16:19] >> and the group put in 200 million. [16:21] >> That's right, we talked about it. [16:22] >> traded down pretty substantially [16:25] >> Yeah, I remember that. [16:26] >> after the investment, but it's basically [16:28] recovered to where it was. And [16:31] >> Well, it's still down over the past like [16:33] 5 years. [16:34] >> Yeah, yeah. [16:34] >> 2 years ago, it was it was like a $5 [16:36] billion company. [16:37] >> Yeah, down down big. [16:38] >> But but but but but what's interesting [16:40] is that at this time of like economic [16:42] uncertainty and all these questions [16:43] about hiring, questions about economic [16:46] resilience, they are increasing [16:48] revenues, increasing attendance, [16:49] increasing customer spend. And that's [16:51] what's driving the stock today. And so, [16:53] I was looking at these two companies and [16:55] I was like, they they sort of fit into [16:56] this [16:57] this barbell thesis of the AI future, [17:00] which I've been seeing pop up more and [17:01] more. And the two examples that I always [17:03] give are one, the Ellison family is both [17:05] like long slop and long anti-slop. They [17:08] have a ton of infrastructure investments [17:10] in AI through Oracle, and then they own [17:13] legacy media, like Batman, Superman [17:15] through Warner Brothers. And then Josh [17:17] Kushner's doing a similar thing with [17:18] OpenAI and the San Francisco Giants. [17:20] Like two completely opposite ends of the [17:22] spectrum. So, you can think of roller [17:24] coasters as potentially like anti-slop [17:26] because you can't vibe code Space [17:28] Mountain. [17:29] But but but it's interesting to sort of [17:32] like dig into the weeds of what's going [17:34] on in the global economy and the [17:36] American economy and seeing like where [17:39] the unsuspecting winners are. Ryan [17:41] Peterson shared a post from the chart [17:44] from the Financial Times that perfectly [17:47] illustrates our possible futures. This [17:49] is so funny that this is in in the [17:52] Financial Times, but they they fully [17:54] embraced what happens in the human [17:57] extinction tech singularity and the end [17:59] of scarcity tech singularity. [18:01] In the bull case, [18:03] real GDP per capita goes north of a [18:06] million dollars. And in the human [18:08] extinction scenario, it goes to zero, of [18:11] course. But the AI boosted growth path [18:13] is a steady trend up. words and that is [18:16] that is the goal that I think everyone [18:17] should be working for. Potentially the [18:19] end of scarcity outcome as well. And I [18:21] and I think this this chart that Ryan [18:23] shares is sort of sort of I think why [18:25] there's anxiety in the market because [18:28] there's this you know that the whole [18:30] like you're not prepared if it's not a [18:32] bubble concept, but there's there's like [18:35] a dual anxiety where like if if AI gets [18:38] too good at there's mass unemployment, [18:40] everyone's worried about that. But if AI [18:42] is a bubble and it collapses, you go [18:44] into recession. Everyone loses their job [18:46] and like you're in the similar scenario [18:48] of like economic anxiety for both the [18:51] not necessarily the true doomers. I [18:52] mean, obviously they they they face a [18:54] similar [18:55] you know opinion. But even if you're [18:57] just like in the it's overheated camp, [19:00] there's real risk to the stability. And [19:04] I think that's where a lot of the the [19:06] the bubble concerns come from. Although, [19:08] we're certainly not seeing very many [19:10] signs of of the bubble. I mean, [19:12] obviously, the valuations are high, but [19:14] uh so are the revenue growth charts. So, [19:17] we can go through um [19:18] some of the folks [19:19] >> Before we do that, [19:20] are you more of a back of the envelope [19:22] guy or a napkin math guy? [19:24] >> Oh, that's interesting. I have a napkin [19:25] here from WonderCo from the uh event. I [19:29] brought it [19:29] >> Oh, you want to look at that? A [19:30] souvenir. [19:31] >> I like uh I like a napkin. I like a [19:34] napkin. Actually, now that I'm thinking [19:35] about it, napkin math is insane. It's [19:38] extremely hard to write on a napkin. [19:40] >> Totally. [19:40] >> It's way easier to write on the back of [19:42] an envelope. [19:42] >> Yeah, and you got a lot more space. [19:44] >> envelope. You can get an envelope that's [19:46] 8 and 1/2 by It's a It's effectively [19:48] >> An envelope is sophisticated. An [19:49] envelope is is is kind of a core [19:53] business utility. Napkins can be used [19:55] for anything. [19:56] >> You can You can You can You can You can [19:57] You can You can You can You can You [19:58] >> I will say I'm much more of a napkin [19:59] math guy. I can't I mean, just just just [20:03] um the phrase that you use spiritually, [20:05] spiritually, and the phrase that I [20:06] pulled, but if I can't get excited about [20:09] an opportunity based on what can fit on [20:11] a napkin, [20:13] then I'm I'm never going to get excited [20:14] about it. [20:15] >> So, you think a napkin is is [20:17] definitionally smaller? It's a less [20:19] amount of math? [20:20] >> not It's not It's a It's a lesser [20:23] vehicle than [20:25] an envelope. [20:26] >> Yeah, cuz a big envelope, you could do a [20:28] full spreadsheet on. You can You could [20:29] You could write out columns [20:31] and have multiple cells. [20:33] >> Um No, so I'm I'm I'm in in practice, [20:36] I'm a napkin math guy, but if something [20:38] was really serious, I would probably [20:40] trade [20:41] >> envelope. [laughter] [20:44] Yeah, we've been doing too much napkin [20:45] math around here. It's time to pull out [20:46] the big guns. Let's get out the [20:48] envelope. [20:48] >> Speaking of envelopes, so this morning, [20:51] I was telling Tyler, um I'm I'm I'm [20:54] incredibly overwhelmed with Slacks, [20:56] emails, you know, DMs on every platform, [21:00] LinkedIn DMs, Instagram DMs, X DMs, all [21:03] the different messaging apps, and I I [21:05] thinking, how cool would it be if there [21:07] was a service where you connected all [21:09] your inboxes to and then every day they [21:12] would print out all of the messages and [21:14] then bring them to you and you could put [21:18] like a box. [21:20] >> Yeah. [21:20] >> You could put a box outside of your [21:22] house and they would just put them in [21:23] there and then at a specific time, you [21:25] know, maybe in the roughly like 2-hour [21:27] window, you could go out and grab all of [21:29] the [21:29] >> Yeah. Back to back to business. [21:32] >> sort of like leaf through them, decide [21:33] what you need to respond to. You would [21:35] potentially, because you're only getting [21:37] messages once a day, you would probably [21:39] be a lot more intentional about what you [21:41] wrote back and forth, right? [21:43] And I think that could be [21:45] >> Sort of the opposite of Earth Class [21:46] Mail, those virtual mailboxes. Cuz [21:48] they'll they'll they'll take your [21:50] physical mail service. [21:50] >> I want everything. I want everything to [21:52] be in this box. I want to put a box [21:54] outside of my house and people put [21:57] and I think there's something there. [21:59] >> Yeah. Trump I mean Trump literally does [22:00] this. [22:01] >> He does that, right? [22:02] >> He gets emails printed out [22:03] and then he'll he'll write with a big [22:05] marker his response and then his [22:06] assistant will scan it and then email it [22:08] back. [22:08] >> Yeah. [22:09] >> More more on C Fry says a napkin is more [22:12] available than an envelope. Very true. [22:15] Often the best Often the best business [22:17] meetings [22:18] >> Mhm. [22:19] >> were not scheduled as a business [22:21] meeting. [22:21] >> So there aren't envelopes around. [22:23] There's napkins. [22:23] >> around but there's plenty of napkins. [22:25] >> Also you're I mean you're saying napkins [22:26] are smaller but if you're at a [22:27] restaurant and you have like a fabric [22:29] napkin and you unfold that, those can be [22:30] pretty big. [22:31] >> So you're you're but I feel like it's [22:33] bad form to be at a restaurant. Like [22:35] you're not supposed to ruin the napkins. [22:37] Or do you carry a pen that has washable [22:40] ink? [22:40] >> You have the crayons for the kids, [22:41] right? People are [22:43] >> The chat is saying [22:44] >> ask you to [22:45] >> is saying I could call my service the [22:47] United States Postal Service. [22:48] >> Thank you, Alex. That's a great name. [22:50] >> That's sort of like the the the San [22:51] Francisco artificial intelligence [22:53] company or something like that or the [22:56] the browser company of New York. It it [22:58] it would be inspired by that, yeah. [23:00] Uh, [23:01] I I I think that even with the crayons [23:03] and the kids [23:04] >> He [23:04] closed-fisted grip, right on the table. [23:07] >> On the tablecloth, okay. [23:08] I Yeah, I I feel like [23:11] if if you give if you're given crayons, [23:14] you are expected to maintain that the [23:17] child uses the crayons only on the [23:19] children's menu, which is typically made [23:21] out of paper, very disposable. If you [23:23] see the child using the crayons on a [23:26] cloth napkin, you are expected as a [23:28] patron to intervene. [23:29] >> Jackson says, "Digitize this service and [23:32] call it email." Here we go. We got to We [23:34] got to get We got to establish the [23:36] mailbox first. [23:37] >> We got an idea. [23:39] >> But then we can go there. But I like [23:40] where you're going. [23:41] >> Well, there's some other There's some [23:43] other posts about the the the markets. [23:45] Justin Spittler says then, [snorts] "Be [23:48] careful buying semis here. Obviously, [23:50] the market's very very overheated." [23:53] >> Travis [23:54] >> this video. Oh, this is Travis Pastrana. [23:56] >> Yes. [23:57] >> That's amazing. Nitro Circus, we [23:58] mentioned many times on the show. [24:01] Is he drinking a Red Bull? [24:03] >> Yeah, no parachute. [24:05] >> No Oh, no parachute. [24:06] >> Yeah, so he jumps out of the plane no [24:08] parachute and he's drinking a Red Bull. [24:10] >> That's actually [24:10] >> he connects. He's got to do some tricks [24:13] first. [24:14] >> taking his time here. The tension is [24:16] building, building, building. [24:18] And so he connects with someone who [24:19] wraps him up. And then do they pull the [24:22] parachute? Or do they [24:24] >> They they they strap him to them. [24:27] >> Oh Oh, okay. There is a strap. [24:28] >> Strap themselves to him. [24:29] >> seems sort of crazy. Just a bear hug and [24:31] and hope for the best. [24:32] >> [laughter] [24:32] >> Your bear hug [24:33] >> I don't know. It's Red Bull, anything [24:35] could happen. [24:35] >> Um [24:36] >> I would I would I would expect him to be [24:38] delivered a full parachute, his own [24:41] parachute that he then, you know, dons [24:43] and and But he made it. Wow. Well, [24:46] that's what it's like investing in [24:48] semiconductors right now, I guess. [24:49] Anyway, how was your last 2 days, Tyler? [24:51] What What did you get up to? [24:52] >> It was sick. I was at the I went to the [24:54] trial. [24:54] >> to to trial? [24:55] >> Opening uh trial. [24:57] >> Okay, walk me through it. You left the [24:59] the studio on Tuesday. You leave the [25:01] office at uh 2:00 after we wrap the show [25:04] on Tuesday. You went straight to the [25:05] airport? [25:06] >> I did. [25:06] >> Okay. [25:06] >> Yep. [25:07] >> And then, what time did you go to bed? [25:08] Cuz you woke up really early, right? [25:10] >> Yeah, so I I I got to the trial, so it's [25:12] in Oakland. I got to the courthouse at I [25:14] think like something around like 5:30. [25:16] >> 5:30? [25:17] >> Yeah, cuz you got to get in line. So, [25:18] basically, you know, it's a public [25:20] cuz it cuz it's like a federal case. [25:21] >> Yeah, yeah. [25:22] >> Um there like has to be some room for [25:24] the public. Uh but there's only like I I [25:26] think the number is somewhere like 20 or [25:28] 30. Cuz I I think it's closer to 20 [25:30] because I think part of the 30 is [25:32] reserved just for media. [25:33] >> Okay. And when you got there at 5:30, [25:35] were there already people standing in [25:36] line? [25:36] >> Yeah, so so there were I think there [25:38] were like three or four people there [25:39] already. [25:39] >> Okay. [25:40] >> Um in line. [25:41] >> bird getting the worm. [25:42] >> But yeah, well, I cuz I didn't want to [25:43] be late, obviously. Cuz you know, I cuz [25:45] >> Did you talk to the other people? [25:46] >> Uh a few of them. I so so [25:48] >> Were you there before Mike Isaac? [25:50] >> I was there before Mike Isaac. [25:51] >> But he has a media pass. [25:52] >> So so yeah, because last time when Mike [25:54] Isaac came on the show, he said that [25:56] because each like news [25:58] >> Yeah, he uh New York Times got one pass, [26:00] and he was splitting it with Cade Metz. [26:02] And so they were sort of going back and [26:04] forth, and I guess he had a media pass [26:05] that day. [26:06] >> Yeah, so that day he had a media pass. [26:07] He He got there pretty late. He got [26:08] there like 7:30 or 8:00. [26:10] >> Okay. [26:11] >> Um [26:11] >> So, you're there from 5:30, he gets [26:13] there at 8:00. What time does the trial [26:14] actually start? [26:15] >> Around 8:30. [26:16] >> 8:30, okay. [26:17] >> Yeah. So, basically, I'm just posting up [26:18] for like 3 hours. [26:19] >> Okay. And then, this is Wednesday, so [26:21] walk me through like what what what are [26:23] you seeing? There's these opening [26:24] remarks from the judge. They sort of [26:25] welcome everyone in. [26:26] >> Yeah, jury comes in. [26:27] >> Okay. [26:28] >> Um so, there are basically three main [26:30] like segments. So, basically, I'll say [26:32] it starts 8:30, ends at I think 2:00 [26:34] p.m. There's uh two breaks, two [26:37] 20-minute breaks. So, starts off, and we [26:40] we watch like the deposition of Mira [26:41] Murati. So, this is all live-streamed, [26:43] right? The [26:44] The audio is [26:45] >> The audio is live. [26:45] >> So, you're seeing a video. We're [26:47] watching the video of her deposition. [26:48] >> Okay. [26:49] >> It's mostly covering the the ousting, [26:51] the timeline of the ousting. [26:52] >> And and did This is where This is where [26:54] all the the texts came out. [26:55] >> Yeah. Yeah. Yeah. And And And [26:57] were they showing you the actual text? [26:59] Were they playing a video? Or were they [27:01] playing the AI [27:03] reenactment of the text? Cuz we have [27:06] this Someone turned Sam Altman's text to [27:09] Mira Murati into a 2011 style emo teeny [27:12] teenage heartthrob anthem. Did this make [27:15] it into the courtroom or no? Let's play [27:17] this. [27:18] >> You have the audio? [27:19] >> Can you indicate directionally good or [27:22] bad? [27:24] Satya and others anxious directionally [27:28] very bad. [27:29] >> Good motion graphics, too. [27:31] >> [music] [27:32] >> Whoever made this is talented. [27:35] >> And this is [27:38] very bad. [music] [27:41] Can I call [27:42] >> What It Is this Suno? Can you put in [27:45] text into Suno and get a full song? [27:47] >> this for an AI skeptic. [27:48] >> Yeah. [27:49] >> And And see what their reaction is [27:51] because [27:51] >> How do you see this and not want to [27:52] build? I I've I've seen these before and [27:54] I've always really enjoyed them, but [27:56] with most of the AI music tools, I've [27:59] never found a great way to put in a full [28:02] script that I've written to get the [28:04] results, but does Suno have that? [28:06] >> So the author said his steps to make [28:08] this. So first, he OCR'd the images and [28:10] turned them into plain text. So images [28:12] of the actual text messages. And then [28:14] remove the names from the dialogue, [28:15] paste it into Suno, and iterated like 20 [28:17] to 30 times. [28:18] >> Ooh. [28:19] >> And then finally thought this one was [28:20] catchy. [28:20] >> Okay. [28:21] >> Um yeah. [28:22] >> Put up a version of this with the with [28:24] the Gen Z brain rot slang. Yo, fam, can [28:26] you give me a vibe check on am I cooked [28:28] or nah for real for real? Satya and the [28:31] gang low-key stressing. [28:33] Uh Mira Murati, yeah, you're skibidi. [28:35] This isn't a rare L. This isn't [28:37] Ohio-level [28:39] generational oral loss. Little bro, who [28:42] ratioed me? Sam Altman says. Mira [28:44] Murati, some rando zesty NPC Twitch [28:47] looksmaxing Sigma. I'm not even going to [28:49] go further. What else happened? Did [28:51] someone actually get on the stand or was [28:52] it all video? [28:53] >> Yes, [laughter] so the first like hour [28:54] >> It's It's straight up teacher played a [28:56] movie. It's movie day. [28:57] >> No, the [28:58] the judge actually did get angry because [29:00] like she was like, "Oh, the jury's going [29:02] to get bored." And like she was like [29:03] offering them coffee and stuff. [29:04] >> Okay. So basically, yeah, the first [29:05] maybe hour was Mira deposition video and [29:08] then we go through some of the actual [29:09] like documents again of the text [29:11] messages. And then Shivon Zilis is [29:13] testifying, so she's physically there. [29:15] >> Okay. [29:15] >> She was a Yeah, OpenAI board member 2020 [29:17] to 2023. [29:18] >> Okay. [29:19] >> And she steps down when Elon started [29:21] xAI. [29:22] >> Yep. [29:22] >> I have no idea how long that was. That [29:23] was the majority of the [29:25] >> It was during the whole conflict, so [29:27] perfect person to uh [29:29] >> to testify. [29:29] >> Well, no, so so she actually left the [29:31] board before the ousting. So she left in [29:33] February 2023. Ousting is November. [29:36] >> Oh, okay. Interesting. I didn't I I [29:38] The timeline always gets so jumbled up [29:40] because like there's like 2018 battles [29:42] between Sam and Greg and Ilya and Elon. [29:45] And then there's the board [29:47] >> all throughout this hearing there's like [29:49] Yeah, there's there's basically [29:50] >> of two main moments, right? Yeah, and [29:52] we're sort of clicking back and forth [29:54] between them. [29:55] >> Yeah. [29:55] >> Uh chat wants to know what kind of [29:57] wambas were being thrown around. Was [29:59] Lorraine being used? People saying, "Can [30:01] you please Lorraine this email?" [30:04] >> Uh Shivon was actually dropping some [30:06] good lingo. I I I don't know if you [30:07] remember [30:08] >> Yeah, there's something [30:09] like Mike Isaac always [30:11] talks about like they keep referencing [30:13] Dota and they keep Everyone on the stand [30:15] is using a ton of jargon. How jargony [30:17] was it? Was it like Dwarkesh Patel [30:19] podcast level jargon or like actual like [30:22] researchers like talking to each other [30:24] jargon? [30:25] >> I would say it was not very [30:26] sophisticated, especially among the [30:28] lawyers, right? Okay. Shivon, she did [30:31] drop some good lingo. I I Yeah, it was [30:33] some referencing some like why she got [30:35] into AI, but you know, acceleration and [30:37] all this. [30:37] >> Okay. Okay. [30:38] >> And then after that it we watched [30:40] another like hour-long video of Helen [30:42] Toner's deposition. That was just about [30:43] the the house thing. [30:44] >> Okay. [30:45] >> Yeah, it was very fun. I enjoyed a lot. [30:46] Like basically almost everyone like [30:48] sitting in my section, which was [30:49] basically the public and the and the [30:51] media. Like everyone that came as the [30:53] public were basically also media. They [30:55] just didn't get the media pass. [30:56] >> Sure. Sure. [30:57] >> So everyone surrounding me is is on [30:58] their laptop like typing basically. [31:00] Yeah, just like taking down what's [31:02] what's being said. And then I'm just [31:03] sitting there enjoying it cuz it's like, [31:04] you know, [31:05] it's good stuff. [31:06] >> How would you rate Mike Isaac's like [31:08] snack and overall food supplies for the [31:11] day? Did he bring Did he [31:12] >> It looks pretty good. I I believe he did [31:13] bring the butt pillow. [31:14] >> It seems like he's not making progress [31:16] on the food front. I would expect by day [31:20] seven of this he would have a [31:21] smorgasbord in front of him. [31:22] >> I I fasted the whole time. I was kind of [31:24] I just forgot about [31:24] >> didn't you figure out We we have talked [31:26] about the food situation so many times. [31:28] How did you not think ahead there? [31:29] >> I just wanted to make it more [31:30] challenging. [31:31] >> Okay. [31:31] >> Yeah, I was like this is going to be too [31:32] easy. [31:33] >> Deep Seek is raising a monster $7 [31:35] billion round at $50 valuation making it [31:39] China's largest ever AI raise. But what [31:43] shocks A.J. Ross here Crypto Punk the [31:45] most is the founder. He's personally [31:47] contributing 40% of the round himself. [31:49] Wow, $3 billion coming from the founder [31:51] directly. Owns 90% of the company. [31:54] Unheard of at this valuation. Deep Seek [31:56] was founded inside [clears throat] of [31:57] his hedge fund, one of China's most [31:58] successful funds. What a beast. Uh he's [32:01] got to acquire as much compute to push [32:03] out new Deep Seek models. You know, we [32:04] saw that chart that showed that uh all [32:07] Chinese open source models were sort of [32:08] falling behind a little bit on a [32:10] different growth curve in terms of [32:12] performance, but uh you know, he's [32:14] certainly betting on getting back in [32:16] line, having a uh you know, frontier [32:19] model within a within a couple months. [32:21] >> Zack Brax says, "Congrats to Anthropic [32:22] for defeating Grok in the market and [32:24] feasting upon the compute of their [32:26] fallen enemy." [32:27] >> Yeah, basically every time we take every [32:29] time we take a day off the show, like [32:31] something big happens uh on Wednesday. [32:34] It was the was the xAI's or SpaceX [32:37] Anthropic deal. A lot of people have [32:39] been predicting that. I I was not simply [32:42] because I thought there it it was like [32:45] the rational decision for the parties, [32:47] but I thought that the tension between, [32:49] you know, Elon who had only a couple [32:51] months ago been, you know, hurling [32:53] insults at the Anthropic team. I didn't [32:55] think they would be able to uncover that [32:57] those cultural differences. Uh but [33:00] demand for compute finds a way. [33:02] >> Yeah, I think that you and others had [33:04] identified the possibility of becoming a [33:06] neo cloud selling the compute where the [33:08] >> Yeah, last last year last year I was [33:10] talking about that a lot all the time. I [33:12] was like, "Look, they're incredible at [33:13] building infrastructure really really [33:15] fast bringing power online. This feels [33:17] like a very strong a strong uh use for [33:21] Elon Inc. [33:23] >> Yes. [33:23] >> Uh and uh but I but as things evolved I [33:26] just didn't see I didn't see this coming [33:28] together. [33:28] >> I didn't see this coming together [33:29] specifically because of the cursor deal. [33:31] I thought the cursor deal was the [33:32] long-term solution for all that compute [33:34] and then the compute sort of got sold [33:36] twice maybe, but of course uh there are [33:38] multiple clusters, multiple Colossus [33:40] data centers, and plenty of work to be [33:43] done as uh [33:45] SpaceX continues to uh grow their [33:47] ambitions.