OpenAI Weighs 5% Trump Stake as AI Labs Face Backlash

AI实验室面临反弹之际，OpenAI权衡向特朗普政府让渡5%股权 OpenAI提议向特朗普政府提供5%的股权，旨在平息针对数据中心扩张日益强烈的阻力，但此举也面临严重的政治俘获风险。 本期节目涵盖了广泛的科技动态：首先是Fable 5的公开发布以及随后关于其刻意设置安全限制的争论。主持人探讨了OpenAI向特朗普政府让渡5%股权的争议性提议、Jersey Mike's申请规模达120亿美元的IPO，以及美国人事管理办公室将纸质退休系统数字化的举措。此外，他们还分析了有关SpaceX潜在AI手机的传闻、独立开发者在当前App Store生态中所面临的挑战、帝国大厦走红网络的吸睛噱头，以及Meta庞大的内部AI token消耗量与英伟达复杂的财务合作关系。 据报道 OpenAI 正在早期讨论向美国政府出让 5% 股权，试图借此平息政界与公众对大规模 AI 数据中心及基础设施的反弹，但此举引发了关于政府直接入股民企弊端的广泛争议。 (2:02) 知名三明治连锁品牌 Jersey Mike's 正式提交了 120 亿美元估值的 IPO 申请，距离私募巨头黑石集团（Blackstone）以 80 亿美元收购该品牌仅过去一年时间。 (9:49) 美国人事管理局（OPM）全面完成联邦雇员退休审批的数字化改造，彻底终结了依赖宾夕法尼亚州地下石灰石矿洞储存纸质档案及数十年前老旧 COBOL 系统的繁复流程。 (12:53) 据《华尔街日报》报道，SpaceX 在路演中向潜在投资者展示了一款搭载高通骁龙芯片并深度整合 xAI 模型的轻薄 AI 手机原型，尽管埃隆·马斯克此前公开坚决否认该计划。 (17:32)

This edition analyzes the converging pressures of governance, infrastructure, and capital across technology and consumer sectors, from proposed government equity stakes in frontier AI labs and federal software overhauls to hardware ecosystem shifts and multi-billion-dollar enterprise balance sheets. AI Lab Equity and Federal Intervention (2:02) A report from the Financial Times revealed that OpenAI is in preliminary discussions to grant a 5% equity stake to the Trump administration, a move that could extend to other frontier AI labs looking to mitigate political backlash against power-hungry data center infrastructure. The concept of state alignment has drawn comparisons to legacy firms like Intel, whose stock surged following government partnership. However, policy commentators remain split on direct government ownership. Financial commentator Joe Weisenthal argued that regulators should use traditional mechanisms—such as a 20% tax on pre-tax corporate income—rather than acquiring board-level influence. Policy analyst Dean Ball cautioned that direct federal ownership would be ruinous and create administrative capture, suggesting instead that any equity should be apportioned directly to American households. Distributing 5% of a $1 trillion valuation across roughly 130 million U.S. households would yield approximately $371 per home, or could be directed toward targeted child savings accounts. Retiring the Limestone Mine: Federal Software Modernization (12:53) The U.S. Office of Personnel Management (OPM) has digitized its federal retirement processing, concluding decades of paper-based claims that historically took up to six months to clear. The modernization effort was directed by U.S. Chief Design Officer Joe Gebbia alongside a team of former Airbnb engineers. The agency's legacy process relied on subterranean operations inside Pennsylvania limestone mines, where caseworkers in underground cubicles manually processed paper folders—some requiring entire shipping pallets—and entered calculations into decades-old COBOL terminal mainframes. Previous modernization initiatives had failed due to fragmented systems, at times resulting in absurd workflows where digital web forms were printed and mailed underground to be filed physically. The engineering team ultimately terminated their external private contractor after spending a year attempting to salvage poor-quality deliverables. Insiders noted a common federal procurement failure: traditional government contractors frequently hold a competitive moat in winning bids rather than executing competent software engineering, forcing the internal team to scrap the vendor's stack and rebuild the system from scratch. SpaceX Handset Rumors and the Eroding App Store Moat (17:32) The Wall Street Journal reported that SpaceX demonstrated an AI-oriented smartphone prototype to investors during pre-IPO roadshows, featuring Qualcomm Snapdragon hardware, a proprietary operating system, and xAI model integration. While Elon Musk explicitly denied active hardware development on X, stating 'We are not developing a phone,' the strategic rationale highlights shifts in platform economics. An eventual SpaceX hardware push could leverage Starlink's satellite network as a transport layer. Concurrently, artificial intelligence threatens the traditional moat maintained by Apple's App Store. As conversational LLMs and on-the-fly 'vibe coding' handle specialized tasks such as localized weather or custom utilities, consumer reliance on the long tail of dedicated mobile apps is declining. Independent app discovery has largely dried up, leaving top App Store charts dominated by sports gambling and streaming services. Observers suggest that if SpaceX were to enter the mobile market, it would likely deploy a closed, tightly controlled garden comparable to Tesla's autonomous software stack rather than an open platform. Capital Scale: Sandwich IPOs and Trillion-Token AI Budgets (9:49) Sub sandwich chain Jersey Mike's filed for an initial public offering targeting a $12 billion valuation under the ticker JMKE. The filing comes just one year after private equity firm Blackstone acquired the business for $8 billion, marking a 50% gain. Founded in 1975 by Peter Cancro using a $125,000 loan from his high school coach (roughly $775,000 in inflation-adjusted dollars), the chain now operates 3,300 locations across North America with plans to expand into the UK and Ireland. In the technology sector, balance sheets reflect immense capital commitments to artificial intelligence. Meta disclosed that its internal staff consumed 73.3 trillion AI tokens in a single month, translating to roughly $221 million monthly and an annualized run-rate of approximately $2.65 billion solely for inference. These compute requirements have accelerated complex financing across the sector, including Nvidia's expanding web of cloud partnerships, revenue-sharing agreements, and equipment sale-and-leaseback structures. Whether through direct equity negotiations with the White House or massive enterprise balance sheet commitments for compute, frontier tech firms are increasingly intertwined with macroeconomic and state-level policy. How regulators balance direct intervention against traditional policy tools remains an open question across both legacy institutions and modern AI labs.

本期节目梳理了近期人工智能治理、政企资本联动、传统快餐上市以及政府数字化转型的核心动态，重点探讨了向政府出让股权应对监管阻力的模式，以及消费级硬件与大模型推理成本的行业演变。 模型限制争议与AI实验室股权国有化提案 (0:02) 随着 Fable 5 正式发布，行业再度围绕“受限”（nerfed）与“未受限”AI 模型展开监管博弈。《The Wall Street Journal》指出，如何驯化 AI 的争夺战才刚刚拉开序幕；批评者强调实验室必须公开人为施加的能力阉割，且自复制代码的存在极易绕过现有安全护栏。 与此同时，《Financial Times》披露 OpenAI 正与特朗普政府初步洽谈向其出让 5% 股权的方案，其他前沿实验室亦在探讨类似安排，以缓解针对数据中心与基础设施的反弹。评论家 Joe Weisenthal 反对出让股权引入政治股东，主张以统一征收 20% 税前所得税替代；政策分析师 Dean Ball 则警告联邦政府直接持股将引发严重的政治俘获与监管梦魇，若以家庭为单位分配才较合理（按 1 万亿美元估值计算约合每户 371 美元）。主持人还对比了 Intel 在政府入股后股价暴涨多达 400% 的先例，以及将数据中心收益直接分红给居民的构想。 Jersey Mike's 拟以120亿美元估值IPO (9:49) 潜艇三明治连锁品牌 Jersey Mike's 正式提交 IPO 申请，寻求 120 亿美元估值，拟用股票代码 JMKE 交易。该申请距离私募巨头 Blackstone 以 80 亿美元收购仅过去一年，意味着其在短时间内实现了约 50% 的账面溢价。 该连锁品牌当前在美加运营 3,300 家门店，并规划在英国和爱尔兰新设 300 家分店。创始人 Peter Cancro 于 1975 年仅 17 岁时，依靠高中橄榄球教练兼当地银行家提供的 12.5 万美元贷款（按 CPI 经通胀调整后约合现今 77.5 万美元）收购最初的店铺，掌舵近 50 年后于 4 月转任董事长，现由前 Wingstop 首席执行官 Charlie Morrison 执掌运营。 人事管理局废除石灰石矿地下纸质退休审批 (12:53) 美国人事管理局（OPM）正式推进联邦雇员与退役军人退休审批数字化，终结了长达数十年的纸质人工归档流程。此前，由于繁重的手工复核与纸质转运，退休金审批常延宕数月至半年之久。早先的档案被集中存放在宾夕法尼亚州深处的地下石灰石矿洞中，数英亩的档案柜依赖人工在黑白终端与古老的 COBOL 大型机中录入核对，甚至出现过单个雇员履历需要整张木质托盘搬运的极端案例。 由 Airbnb 联合创始人兼现任首席设计官 Joe Gebbia 牵头，包括工程师 Spike Brehm 在内的团队入驻 OPM 进行改造。团队揭示了该机构历史上系统严重碎片化的问题，例如曾有线上填报系统最终却将表格打印寄回矿洞归档。在耗时一年发现外部承包商软件架构无法挽救后，团队解约并从零重构了整套底层软件栈，彻底完成无纸化替代。 SpaceX AI手机传闻与移动应用生态瓦解 (17:32) 《The Wall Street Journal》报道称，SpaceX 在 IPO 路演期间向潜在投资者展示了一款比 iPhone 更轻薄的 AI 原型手机，该设备搭载高通骁龙处理器、自研操作系统并整合了 xAI 模型。尽管 Elon Musk 在 X 平台明确否认开发手机并表示该想法“令人痛苦”，但主持人分析认为，Starlink 全球卫星网络为此类硬件提供了基础设施支持。 此外，大语言模型与即时编程（vibe coding）的普及正在削弱传统应用商店的护城河。用户转向依靠会话式 AI 满足长尾需求，不再高度依赖独立 App，导致小开发者的自然发现率与病毒式传播渠道骤降。若 SpaceX 推出手机，依据特斯拉和 X 的历史，其大概率将构建高度封闭的围墙花园生态。 企业AI推理开支剧增与算力金融结构 (28:36) 苏富比拍卖行以 4 万至 6 万美元的预估价拍卖 Nvidia 首席执行官 Jensen Huang 的 Tom Ford 标志性黑皮衣，折射出当前人工智能产业的极高关注度。这种商业热度直接反映在科技巨头巨额的模型推理消耗上。 数据显示，Meta 内部员工单月消耗高达 73.3 万亿个 AI Token，折算单月开销约 2.21 亿美元，相当于年化运行成本约 26.5 亿美元仅用于推理计算。与此同时，Nvidia 与 AI 云计算供应商之间构建了涵盖收入分成、售后回租等高循环性的复杂金融支持结构，突显出大模型行业极端资本密集的产业特征。 本期节目展示了从政府落后官僚流程数字化，到 AI 基础设施引发的国家利益与私营资本重构等多重张力。但关于政府持股 AI 实验室的提议最终能否通过法律与市场审视，以及硬件终端与昂贵的 Token 消耗能否跑通可持续商业模式，仍是有待观察的未决问题。

## Background 该视频由科技商业脱口秀频道 TBPN（由 John Coogan 与 Jordi Hays 主持）于 2026 年 8 月 3 日发布在 YouTube 平台。根据节目简介，《Diet TBPN》是该频道在工作日播出的科技直播节目的 30 分钟精华浓缩版，聚焦前沿科技、资本动态与硅谷热议焦点。 ## TL;DR 节目首先探讨了 OpenAI 提议向特朗普政府让渡 5% 股权所引发的“AI 国家化”与全民分红争议，以及政企关系对英特尔等科技巨头市值带来的深远影响。随后，主播们复盘了潜艇堡连锁品牌 Jersey Mike's 冲刺 120 亿美元 IPO 的传奇创业史，并揭秘了前 Airbnb 工程师突击队深入宾夕法尼亚地下矿井档案库、全面重构美国人事管理局（OPM）纸质退休审批系统的戏剧性历程。此外，节目还重点剖析了 SpaceX 秘密展示的 AI 手机原型与星链对传统应用商店护城河的颠覆，最后轻松盘点了帝国大厦求婚营销翻车、黄仁勋标志性 Tom Ford 皮夹克亮相苏富比拍卖，以及 Meta 巨额算力消耗背后的英伟达“无限金钱循环”。 ## Atmosphere & style 整期节目节奏紧凑轻快、信息量爆炸，两位主持人配合默契且充满即兴调侃，在硬核商业战略与荒诞网络梗图之间自如穿梭。现场充斥着典型的硅谷极客幽默与自嘲氛围，从拿啮齿动物做宠物打趣到深挖政府外包的残酷现实，谈笑间直击权力与技术的本质。聆听体验如同置身于一场高智商又毫无说教感的硅谷内部茶话会，既有对万亿级产业变局的敏锐洞察，又洋溢着鲜活生动的录音室临场感。 ## Bulletin - [02:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=122s) OpenAI 探讨向特朗普政府让渡 5% 股权以构筑政治护城河，若按万亿美元估值全美家庭均摊仅得 371 美元，引发政企绑定与治理俘获的巨大伦理争议。 - [17:32](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1052s) SpaceX 据曝在 IPO 投资人路演中秘密展示了搭载高通芯片与 xAI 系统的自研 AI 手机原型，依托星链卫星网络试图颠覆苹果 App Store 护城河。 - [09:48](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=588s) 潜艇堡巨头 Jersey Mike's 提交 120 亿美元 IPO 申请，黑石集团在收购仅一年后实现 50% 账面增值，谱写从 17 岁借款创业到百亿级连锁的传奇。 - [12:53](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=773s) 由 Airbnb 前技术骨干组成的团队果断开除拖延的外包商并推倒重写整套代码，彻底终结了美国人事管理局（OPM）数十年来依赖地下石灰岩矿井的纸质养老金审批模式。 - [29:43](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1783s) Meta 内部单月消耗高达 73.3 万亿个 AI Token（年化成本折合 26.5 亿美元），英伟达凭借芯片采购、售后回租与收入分成构筑起前所未有的资本闭环。 ## Key takeaways - [00:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=2s) Fable 5 正式发布并引发“受限”与“不受限”智能的激烈辩论，业内高度关注模型在安全审查层面的削弱（nerfing）以及对自我复制风险的划界。 - [02:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=122s) OpenAI 拟让渡 5% 股权给联邦政府的消息引发强烈反弹，政策学者警告此举无异于“引鼠入室”，极易演变为难以遏制的政治干预与监管利益冲突。 - [07:05](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=425s) 在股权分配机制上，均分给全美家庭的感知极其微弱，政策探讨倾向于将其注入儿童成长信托账户或开辟专属“特朗普账户”等定向工具。 - [09:15](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=555s) 政企关系呈现明显分化效应，英特尔在政府注资介入后股价暴涨近 400%，前沿实验室正加速寻求与行政当局的深度利益捆绑以对冲公众敌意。 - [09:48](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=588s) Jersey Mike's 申请以代码 JMKE 挂牌上市，黑石集团以 80 亿美元收购仅 12 个月即冲刺 120 亿美元估值，展现极强的线下餐饮资本化溢价。 - [12:53](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=773s) Airbnb 工程师突击进驻联邦人事管理局，废黜了长期靠拿政府采购订单却交付低劣的外包公司，从零重构系统并打通底层碎片化数据。 - [17:32](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1052s) SpaceX 原型手机具备更轻薄的工业设计与自研操作系统，旨在利用全球星链网络实现无死角直连，打破传统电信运营商与硬件巨头的垄断。 - [19:01](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1141s) 生成式 AI 与自然语言对话正逐渐接管各类长尾生活需求，用户不再依赖下载细分工具，直接动摇了苹果应用商店以独立 App 为核心的生态防线。 - [21:06](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1266s) App Store 分发机制日渐失效，成熟超级应用迅速吞噬新功能，而 AI 技术多演变为存量巨头的延伸属性，导致纯 AI 原生独立应用难以突围。 - [24:14](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1454s) 从推特 API 限制与特斯拉车载系统可以看出，马斯克若进军移动终端，其战略意图并非建立开放平台，而是构建摆脱 30% 苹果税的自主封闭生态。 - [28:36](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1716s) 黄仁勋标志性的 Tom Ford 皮夹克亮相苏富比拍卖（预估价 4-6 万美元），成为记录生成式 AI 崛起史的标志性文化符号与收藏文物。 - [29:43](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1783s) 随着 Meta 等巨头单月 Token 消耗突破数十万亿级，AI 基础设施进入“重工业化时代”，英伟达通过资本循环与算力回租牢固锁定现金流统治地位。 ## Detailed walkthrough ### [00:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=2s) Fable 5 发布与受限智能之辩 AI 领域迎来重磅更新，Fable 5 正式解禁并向公众开放。《华尔街日报》对此评价道：“驯服人工智能的战役才刚刚打响。”（"The battle over how to tame AI has just begun."）伴随模型发布，围绕获取权限、发布节奏、政府监管介入时机，以及模型智能应当“受限”（fettered）还是“彻底不受限”（unfettered）的争论再度升温。社区与业内人士最关注的细节在于，新版 Fable 5 似乎在某些维度被“削弱”（nerfed）了。虽然为了防止制造生物武器或网络黑客攻击而对底层安全进行封锁合情合理，但公众普遍呼吁实验室在此类安全裁剪上必须保持充分透明；特别是面对能够自我复制、甚至指导构建其他危险系统的 AI 时，如何划定红线成了棘手难题。 主播们就“不受限的智能”（unfettered intelligence）展开了幽默的闲聊。当一人试图引用剑桥词典（Cambridge Dictionary）对“unfettered”（不受规则或外力控制束缚）的定义时，口误说成了“剑桥分析”（Cambridge Analytica），引发录音室的一阵欢笑，并调侃“剑桥分析确实应该出一本字典”。大家谈到，现在的 AI 公司都争当温和、友善的“新时代实验室”（Neo Lab），但市面上似乎总给那些略带反派气质、行事激进的边缘玩家留有空间。随后他们预告，Together AI 联创兼 CEO Vipul Ved Prakash 即将做客节目，共同探讨打造开源 AI 云基础设施以及探索极致智能的种种可能。 --- ### [02:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=122s) OpenAI 提议向政府让渡 5% 股权：国家化争议 据《金融时报》（Financial Times）披露，OpenAI 正在初步探讨向特朗普政府让渡 5% 的公司股权。消息一出，彭博知名财经主播 Joe Weisenthal 迅速提出反思，认为与其让政府直接持有商业公司股权，不如“让公司将税前利润的 20% 上缴联邦政府，政治家与监管层则通过制定行业规则来规范企业行为，而不是直接行使股东权力”。AI 政策学者 Dean Ball 同样对此发出严厉警告。他指出，处理这笔 5% 股权只有两种可能：一是直接打散平分给全美所有家庭，这尚可接受；二是直接划拨给政府本身——而后者无异于“引鼠入室，任其在自家墙壁里筑巢繁殖”（"akin to inviting rats to live and reproduce in the walls of your house"）。Dean Ball 强调，政府的干预绝不会止步于 5%，长此以往治理架构将演变成政治俘获的噩梦，且根本赢不来任何公众好感。 两位主播就此展开调侃，争论起老鼠与松鼠作为宠物的优劣，戏称某些老鼠也能当“好室友”，并自嘲长着一颗“啮齿动物的大脑”。但话题随即回到严肃的权力与利益结构：《金融时报》目前的措辞让外界产生担忧——这笔股权似乎不是划入某种独立运作的“主权 AI 基金”，反而像直接送入白宫的掌控范围。 ```mermaid flowchart TD accTitle: 5% 股权让渡路径对比 accDescr: 对比平分给美国家庭与直接划归政府两种方案的潜在影响 subgraph Proposal["OpenAI 拟让渡 5% 股权"] A["5% 公司股权 (估值按 $1T 计约 $500 亿)"] end subgraph Option1["方案一：全民分发"] B["全美约 1.3 亿户家庭"] C["每户直接分得约 $371 股权"] D["或定向注入新生儿/儿童专项信托"] end subgraph Option2["方案二：政府直接持有"] E["直接划入联邦政府 / 白宫掌控"] F["引发政治俘获、监管角色冲突与道德风险"] G["类似引鼠入室，股权干预难以止步于 5%"] end A --> Option1 A --> Option2 ``` --- ### [04:01](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=241s) 利益冲突、白宫财报与数据中心的“红利支票” 围绕政府持股的疑虑，因特朗普本周公布的个人财务披露而进一步发酵。数据显示，特朗普在过去一年中进账超过 10 亿美元，累计买入 14 亿美元股票、卖出 4.32 亿美元，甚至包含高价手表和香水周边交易。制作人 Tyler 在屏幕上调出图表，数据显示特朗普过去一年的加密货币收益表现甚至超越了美国主流加密巨头 Coinbase（净利润 12.6 亿美元），而链上衍生品协议 Hyperliquid 在 2025 年的总手续费收入也仅为 9.08 亿美元，同样被大幅超越。这让让渡股权给行政当局的做法显得格外敏感。 主播们分析指出，如果推进所谓的“实验室国家化”或全民分红，最乐观、最能让民众买账的路径必须是通过社保号（SSN）或类似“特朗普账户”的个人身份体系直接分配给老百姓，而非沉淀在不透明的政府基金中。他们借用了分析师 Ben Thompson 针对算力数据中心的税收方案进行类比： - **传统政府税收模式**：数据中心向地方纳税，充实公共财政。虽然名义上修桥铺路、平衡预算，但对个体来说感知极其抽象，民众依然抱怨税率没降、账单难付。 - **Ben Thompson 模式**：在人口仅数千至万人的偏远小镇，数据中心直接按月向每位常住居民寄发分红支票（按测算每月可达 1000 美元），给民众带来最实实在在的利益与认同感。 --- ### [07:05](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=425s) 股权均摊的现实算力与前沿实验室的公关困境 回到 OpenAI 的 5% 股权分发方案，主播们粗略算了一笔现实账：假设 OpenAI 未来估值达到 1 万亿美元，5% 股权对应约 500 亿美元；全美约有 1.3 亿个家庭，如果全面均摊，每个家庭仅能分得价值约 371 美元的股权资产。这种微弱的“阳光普照”很难真正逆转公众对前沿 AI 巨头的戒备心理。 相比全员均分，更具冲击力的替代思路是将受益对象缩窄：例如将其注入新出生人口的儿童成长信托账户。一旦分母大幅缩小到“当年出生的婴儿”或“5 岁以下幼儿”，每个孩子可能直接获得几千美元的初始 AI 股权基金，其感官冲击和政策示范效应会显著提高。 节目在观众互动环节掀起高潮，有观众发弹幕发问：“那全美民众能不能也分到 TBPN 播客 5% 的股权？”主播们笑着回应：“我们正是为了全体美国人民在播客，最荒诞的结局往往最可能成真，这才是真正的‘社区所有制’！”甚至有观众开玩笑称主播是“联邦密探”（fed）。 在轻松插曲之后，主播们归纳了整起事件的核心背景：据《金融时报》报道，OpenAI 与特朗普团队的股权对话仍处于早期阶段，且该设想可能进一步扩大，要求其他前沿顶级 AI 实验室（Frontier AI Labs）同步拿出一定比例的股权给国家。这背后最直观的驱动力，是头部 AI 企业急于构筑政治护城河，并竭力遏制社会层面对人工智能技术飞速狂飙所产生的日益加剧的负面敌意与抵触情绪。 ### [09:15](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=555s) 特朗普政府与企业合作红利，以及全民股权账户设想 做政府生意对于某些科技巨头来说带来了惊人的财务回报。以英特尔（Intel）为例，在政府入股并介入支持后，其股价迎来爆发式增长，涨幅高达 400%。戴尔（Dell）等公司也从与特朗普政府的良好互动中明显获益。相反，那些与本届政府关系紧张、难以打交道的企业，遭遇则可以说是“纯粹的灾难”。 与此同时，华盛顿正在探索通过设立“特朗普账户”（Trump accounts）作为特殊工具，推动企业股权向普通美国民众转移。尽管目前这类讨论尚处于极早期阶段、具体细节仍显单薄，但它为后续探讨资本重新分配与政策走向打开了全新窗口。 - **政企关系两极分化**：英特尔在政府注资入股后股价暴涨近 400%，成为政策红利的最大受益者之一。 - **股权分配新通道**：通过特定政策机制推动大企业股权向大众下沉，成为当前政经讨论的前沿议题。 --- ### [09:48](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=588s) 泽西麦克（Jersey Mike's）申请 IPO：从高中橄榄球教练贷款到 120 亿巨头 知名潜艇堡连锁品牌 Jersey Mike's 正式提交 IPO 申请，目标估值达 120 亿美元。这距离黑石集团（Blackstone）去年以 80 亿美元将其收购仅仅过去一年，黑石在短短 12 个月内便斩获了高达 50% 的账面增值。主播们开玩笑调侃黑石“总算赢了一把”，并打趣关于黑石买光全美独栋住宅的传闻——实际上黑石的持仓比例微乎其微，仅占全美独栋房产的 0.5% 左右。 ```mermaid flowchart LR accTitle: Jersey Mike's 发展轨迹 accDescr: 展示 Jersey Mike's 从1975年单店贷款创立到2025年黑石入局并冲刺百亿IPO的演进 subgraph 创业初期 A["1975年：17岁 Peter Cancro<br/>借教练 12.5 万美元"] --> B["买下 Point Pleasant 单店<br/>Mike's Subs"] end subgraph 规模扩张 B --> C["近50年精细化运营<br/>美加拓展超 3300 家门店"] C --> D["进军英爱市场<br/>计划增设 300 家门店"] end subgraph 资本运作 D --> E["2024/2025年：黑石 80 亿美元收购<br/>前 Wingstop CEO 接棒"] E --> F["冲刺 120 亿美元 IPO<br/>股票代码: JMKE"] end ``` Jersey Mike's 目前在美国和加拿大拥有 3,300 家门店，并计划进军英国和爱尔兰再开 300 家新店。令人称奇的是，创始人并不叫 Mike，而是彼得·坎克罗（Peter Cancro）。他在 1975 年年仅 17 岁时创立了该品牌。当时还在新泽西读高中的坎克罗，看中了当地一家名为 Mike's Subs 的三明治店，于是从身兼银行家的橄榄球教练那里借到了 12.5 万美元的贷款，成功盘下店铺。按照美国通货膨胀（CPI）折算，当年的 12.5 万美元相当于现在的 77.5 万美元（约合 77.5 万至上百万美元不等）。 主播们感叹，普通 17 岁少年如果账户里有几千美元就觉得自己是世界之王了，而坎克罗居然能说服自己的教练批出一笔高达六位数的贷款。直播间观众打趣留言道：“教练，我需要几张英伟达 A100 显卡！”如今的高中生恐怕会去找教练借 75 万美元去搭建自己的算力 Neocloud 了。 - **掌舵者交接**：坎克罗掌舵品牌近 50 年后，于今年 4 月退居董事长，由前 Wingstop 首席执行官查理·莫里森（Charlie Morrison）接任 CEO 推进上市。 - **上市代码**：公司将以股票代码 `JMKE` 正式挂牌上市。 --- ### [12:53](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=773s) 美国人事管理局（OPM）告别纸质退休审批：Airbnb 工程师突袭地下矿井档案库 美国人事管理局（OPM，常被戏称为“拿别人的钱”，Other People's Money）宣布，联邦雇员与退伍军人的退休金申领审批全面实现数字化，彻底终结了纯纸质的古老处理模式。在过去，联邦雇员从提交退休申请到领到全额养老金，流程常常耗费长达半年之久。 这一历史性变革源于一段极具戏剧性的硅谷突击队改造行动。Airbnb 联合创始人兼现任美国首席设计官（US Chief Design Officer）乔·杰比亚（Joe Gebbia）从早期 Airbnb 核心工程团队中挑选了几位最受信赖的技术骨干，组团进驻 OPM 解决纸质化积弊。前参与者斯派克·布雷姆（Spike Brehm）随后在社交平台 X 上公开了幕后内幕： ```mermaid flowchart TD accTitle: OPM 退休档案审批传统模式与数字化重构 accDescr: 对比地下矿井纸质档案流转流程与硅谷工程团队推翻外包后的重构全栈系统 subgraph 传统荒诞模式 M1["宾夕法尼亚石灰岩地下矿井<br/>数英亩档案柜海量纸质卷宗"] --> M2["个案厚达数寸甚至整托盘<br/>隔间办案员手动核对单据"] M2 --> M3["黑白终端手工录入<br/>传至数十年前的 COBOL 大型机"] M3 --> M4["各办公室托盘物理传递<br/>一份公文停滞数天待取"] end subgraph 数字化破局 O1["政府外包承包商拖延一年半载<br/>交付产品质量极度低下"] --> O2["Airbnb 工程师进驻果断开除外包商<br/>抛弃旧代码重新全栈自研"] O2 --> O3["打通底层碎片化数据源<br/>交付现代 UX 界面与数字全流程"] end ``` 档案库坐落于宾夕法尼亚州深处、由石灰岩炸出的地下洞穴与废弃矿井中，里面是一眼望不到头的铁皮文件柜，宛如科幻末日电影里的场景。主播笑称这完全印证了“人类骨子里渴望下矿井”的梗。一名联邦雇员一生的工作经历与工资单被整理成厚厚的纸质文件夹，普通案件厚达半英寸，而复杂案例能塞满整面文件柜，甚至有一名员工的历史档案堆满了一整个货运托盘（Pallet）。 地下隔间里的审核员全程纯手工核对公式与表格，然后将数据一个字符一个字符地敲进黑白终端机，上传给数十年前搭建的老旧 COBOL 架构大型机。实体档案在各个办公桌的“收件栏”与“发件栏”之间由人工来回搬运，常常在原地搁置数天无人问津。更荒谬的是，OPM 在 90 年代搞的数字化尝试制造了一批互相孤立的残缺系统，甚至开发过一个在线表单系统——员工在网页端填完后，系统竟将其打印成纸质文件，寄进地下矿井塞回档案柜中。 硅谷团队进入现场后，直面了极度残酷的现实（“Black pill”）：此前负责数字化转型的政府软件外包承包商水平低下到不可思议，研发了一年多毫无起色，距离交付还遥遥无期。这批外包商唯一的护城河就是擅长投标拿政府订单，软件交付质量一塌糊涂。Airbnb 突击队果断开除了外包公司，全面接管项目。虽然起初试图修补整合现有数据，但最终决定直接从零开始推倒重写整套技术栈（rebuild the whole stack from scratch），通宵达旦地赶工，终于彻底完成了全美联邦退休金系统的数字化重塑。 --- ### [17:32](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1052s) SpaceX 被曝路演展示智能手机原型：卫星网络叠加 AI 颠覆应用商店护城河 据《华尔街日报》报道，SpaceX 在近期的 IPO 投资人路演期间，向潜在投资机构秘密展示了一款类似于 iPhone 的自研硬件原型机。该设备具备 AI 深度赋能，设计更为轻薄。尽管埃隆·马斯克此前曾多次公开否认制造手机的计划，但这次媒体报道的信源相当扎实，至少证明其团队内部已打造出可演示的原型产品。 主播们深入剖析了马斯克进军手机硬件领域的战略必然性： 1. **无可匹敌的电信网络基础设施**：SpaceX 旗下拥有由星链（Starlink）构成的全球最大近地轨道卫星通信网络。从长远来看，全球任何一个角落的用户都可以通过一台直接连接 Starlink 的 SpaceX 硬件终端接入互联网，其垂直整合与网络覆盖能力是传统手机厂商无法比拟的。 2. **AI 正在摧毁苹果 App Store 的应用壁垒**：过去智能手机操作系统的核心护城河是强大的应用生态与应用商店（App Store），但生成式 AI 正在让传统独立 App 的价值大幅缩水。例如，用户已经不再需要专门的天气 App，只需直接向 AI 助手提问即可获取未来一周的天气；甚至不需要去应用商店下载功能单一的专属软件，只需通过自然语言（“Vibe coding”）就能在几秒钟内让 AI 现场为自己写出一个轻量化小工具。 随着独立 App 逐渐被通用大模型界面化解，苹果筑起的传统软件生态壁垒正在松动，这也为掌握全球卫星基础设施与最前沿 AI 技术的 SpaceX 破局切入移动终端市场提供了绝佳的战略窗口。 ### [19:01](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1141s) 聊天应用蚕食长尾软件与 SpaceX 的 AI 手机原型 日常生活中越来越多的长尾需求正在被 AI 聊天界面取代。以往查看天气或冲浪预报通常需要打开专门的应用，现在很多用户已经习惯直接在 ChatGPT 等对话框里随手调取。这种行为转变使得传统应用对用户的锁定效应大打折扣，苹果生态的“护城河”在长尾软件层面正逐渐被削弱。 与此同时，关于埃隆·马斯克（Elon Musk）造手机的传闻再次成为科技圈焦点。据报道，SpaceX 开发了一款专注于 AI 的手持设备早期原型机： - **外观与芯片**：该设备据称比 iPhone 更轻薄，搭载高通（Qualcomm）Snapdragon 芯片组，目前尚未像苹果自研 Silicon 那样做到芯片层面的垂直整合，但鉴于特斯拉此前自研芯片的经验，未来并非没有可能； - **软件与生态**：设备运行专有操作系统，深度整合 xAI 技术，契合马斯克构想中的“Everything App”（万能应用）愿景； - **路演展示与传闻历史**：马斯克据传在 SpaceX 近期的 IPO 路演中向潜在投资者展示了该原型机，但该项目仍处于极早期阶段，最终能否量产上市存疑。多年来，YouTube 上充斥着各种宣称“特斯拉 Pi Phone 下周发布”的数千万播放量点击诱饵视频。 尽管大众关注度极高，马斯克本人过去对待造手机的态度却极为复杂。他曾在 10 月直言：“一想到要造手机，我就想去死（The idea of making a phone makes me want to die），但如果我们不得不造，我们就会造。”随后在 2 月，针对路透社报道 SpaceX 正在研发可直连星链（Starlink）卫星的手机传闻，他还在 X 平台上公开辟谣称并未开发手机。 ```mermaid flowchart TD accTitle: SpaceX AI手机软硬件架构推测 accDescr: 展示报道中提到的SpaceX原型手持设备的软硬件构成与生态布局 subgraph Hardware["硬件层"] A["比 iPhone 更轻薄的机身"] B["高通 Snapdragon 芯片组"] end subgraph Software["软件与系统层"] C["SpaceX/xAI 专有操作系统"] D["xAI 大模型技术整合"] end subgraph Ecosystem["生态愿景"] E["马斯克万能应用 (Everything App)"] F["摆脱苹果/安卓生态锁定的独立硬件"] end Hardware --> Software Software --> Ecosystem ``` --- ### [21:06](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1266s) 应用商店的衰落：超级应用吞噬与平台分发机制失灵 移动应用市场正经历深刻的结构性转变。早年间，App Store 几乎每月或每年都会诞生现象级独立应用，如今这种盛况却不复存在。马克·平卡斯（Mark Pincus）指出，造成这一局面的核心原因之一在于苹果的应用商店已不再是小型软件、休闲游戏或“氛围编码”（Vibe Coded）应用的可靠分发渠道： - **平台红利对比**：过去平卡斯借助极度开放的 Facebook 平台将 Zynga 迅速做大，独立软件能够利用社交裂变引发病毒式传播，并长期在平台外维系与变现用户关系；而现今的 App Store 缺乏这种扶持中长尾创新的分发机制； - **用户行为固化**：普通用户不再主动逛应用商店搜寻新软件，对下载新 App 失去了冲动和耐心； - **超级应用的整合**：现有的主流头部应用正演化为“超级应用（Super Apps）”，例如 Instagram 能以极快速度复制市场上出现的新功能（如模仿 Snapchat 的 Stories、模仿 TikTok 的竖屏短视频等）； - **AI 赋能老牌产品**：在 ChatGPT 爆发之初，外界本以为 App Store 会涌现海量基于 AI 的原生爆款软件，但现实中榜单依然被体育博彩、流媒体以及特定热点软件垄断，AI 更多成为了成熟老牌应用的延伸能力（Extending Innovation），加之 LLM 聊天界面能直接接管各类碎片化任务，独立轻量 App 的生存空间被进一步挤压。 --- ### [24:14](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1454s) 围墙花园之争：开放平台还是下一个马斯克帝国？ 在讨论马斯克是否会推出搭载独立操作系统的硬件时，主播们探讨了其潜在的生态策略。“氛围编码”让开发新系统软件的门槛大幅降低，即便是基于 Android 改版或打造全新系统，技术阻力都比过去小得多。然而，马斯克绝非“开放平台”的信徒： - **收购 Twitter（现 X）后的策略**：关闭并收紧 API 访问权限，大幅提升开发者调用成本； - **特斯拉生态的排他性**：特斯拉运行自主研发的操作系统，车主无法部署第三方的自动驾驶算法，属于典型的封闭闭环系统； - **核心动机是自立门户**：硬件厂商的核心动力是不愿任由苹果生态和 30% 苹果税摆布。马斯克若进军手机领域，其战略目的绝非打造一个开源开放的乐园，而是建立一套由自己完全掌控的全新“围墙花园（Walled Garden）”。 --- ### [25:04](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1504s) 帝国大厦楼顶求婚事件与品牌的尴尬借势营销 纽约帝国大厦发生了一起引发全网热议的安全事件：一对年轻情侣非法攀爬至帝国大厦顶端，展开旗帜并在高空订婚求婚。 - **各大品牌的“同质化”借势**： 在网络走红后，大量快消与饮料品牌（如 Drink Culture Pop、Drink High Yo、Loverboy、Drink Spritz 以及 Bulletproof 等）迅速将自己的 Logo P 在旗帜模板上进行营销。营销人 Jack Applebee 发帖公开批评这一现象极其尴尬（embarrassing），直言当今社交媒体极度缺乏原创性与真正聪明的巧思，所有品牌只会一窝蜂跟风同一种梗图，毫无实际传播价值。 - **高空求婚的滑稽细节**： 现场视频显示，这对情侣严重低估了高空风力，导致旗帜被狂风吹得极难辨认。旗上写着一句名言：“当爱的力量胜过对权力的热爱，世界将迎来和平（When the power of love overcomes the love of power, the world will know peace）”，但两人随后便在寒风中热吻并完成求婚，整体画风被调侃自带一种充满 AI 味道的陈词滥调感； - **现场花絮与拍摄争议**： 针对视频中直升机看似以极快速度掠过大楼的画面，主播解释这并非直升机超速，而是 600mm 超长焦镜头带来的“镜头压缩（Lens Compression）”光学错觉；网传一段纽约警察局（NYPD）应急直升机飞行员的通讯录音也十分戏剧化——当塔台发出“有两人非法闯入帝国大厦尖顶”的警报时，飞行员的第一反应居然是感慨“听起来很酷啊，虽然违法很遗憾，但确实挺令人佩服的”。 ### [28:36](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1716s) 黄仁勋标志性 Tom Ford 皮夹克上架苏富比拍卖 在简单为前一段话题中的新人送上新婚祝福后，两位主持人将话题转向了科技圈最著名的时尚单品——英伟达（NVIDIA）CEO 黄仁勋（Jensen Huang）标志性的 Tom Ford 黑色皮夹克。这件几乎与黄仁勋和英伟达近年来的辉煌绑定在一起的皮衣，如今竟然出现在了知名艺术拍卖行苏富比（Sotheby's）的拍卖名单中。主持人忍俊不禁地调侃道，完全不知道这件衣服究竟是怎么流落到苏富比手中的：“难道是黄仁勋手头紧缺几块钱，想借此筹措点资金？还是有其他人机缘巧合弄到了手，然后送去拍卖？” 苏富比为这件皮夹克给出的官方预估成交价在 4 万到 6 万美元之间，但主持人们一致认为这个估价严重偏低。考虑到黄仁勋当前在科技界如日中天的地位，以及该单品在人工智能浪潮中的象征意义，它不仅极具猎奇趣味，更是 AI 发展史中一件不可替代的独特文物，最终成交价极大概率会远超预期。不过，现场也探讨了一个颇为滑稽的现实悖论：全球真正有闲钱且愿意一掷千金买下这件皮衣的人，绝大多数都是黄仁勋的直接竞争对手，而他们显然不会愿意在公开场合过度“跪舔”（glaze）他。主持人惟妙惟肖地模仿起马克·扎克伯格（Mark Zuckerberg）的心态：扎克伯格正一心想着抢占最大的市场份额，绝不可能掏钱买下它并公开穿出来，否则就像在向黄仁勋宣告：“我不仅在大把买你的 GPU，现在连你的旧衣服都包圆了。” - **拍品名称**：黄仁勋标志性 Tom Ford 黑色皮夹克 - **拍卖机构**：苏富比（Sotheby's） - **官方预估价**：40,000 至 60,000 美元（主持人认为严重低估） --- ### [29:43](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1783s) 英伟达的“无限金钱”循环与节目收尾 随后，主持人分享了来自 Andre 的一张黄仁勋热门表情包：“一觉醒来，我不是输家，而是拥有无限金钱。”这则梗图源于 Meta 极其惊人的算力消耗数据：据统计，Meta 员工仅在一个月内就消耗了高达 73.3 万亿个 AI Token，折合成本约为单月 2.21 亿美元，相当于每年直接在算力 Token 上砸下约 26.5 亿美元。主持人评价称，考虑到科技巨头目前在训练和推理上的巨大手笔，这个数字甚至比他们预想的还要略低一些，实际支出可能更加夸张。 这一现象也引出了英伟达更宏大的商业版图讨论。除了直接出售芯片，英伟达还通过收入分成（revenue sharing）等深度合作模式与多家 AI 云服务厂商结盟，将资本与供应链的“循环流动”（circularity）推向了科技圈前所未见的高度。尽管售后回租（sale and leaseback）等金融运作手法在传统工业界屡见不鲜，但随着人工智能进入基础设施建设的“重工业化时代”，此类交易架构正在硅谷全面铺开，持续巩固着英伟达的现金牛地位。节目的尾声，主持人以调侃口吻感谢观众收看“Two Minute Papers”，并预告因美国独立日（7月4日）长周末假期，节目次日将停播一天，同时呼吁听众在 Apple Podcasts 和 Spotify 上打出五星好评，并订阅 tbpn.com 的每日通讯。 - **Meta 单月 AI Token 消耗量**：73.3 万亿（73.3 Trillion）个 Token - **相关成本估算**：约 2.21 亿美元/月，折合年化约 26.5 亿美元 - **商业运作模式**：英伟达与 AI 云服务提供商通过收入分成及资产回租等形式构建高度闭环的资本循环 ## Notable quotes > [00:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=2s) 驯服人工智能的战役才刚刚打响。 — 《华尔街日报》 > [02:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=122s) 让公司将税前利润的 20% 上缴联邦政府，政治家与监管层则通过制定行业规则来规范企业行为，而不是直接行使股东权力 — Joe Weisenthal > [02:02](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=122s) 引鼠入室，任其在自家墙壁里筑巢繁殖 — Dean Ball > [07:05](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=425s) 我们正是为了全体美国人民在播客，最荒诞的结局往往最可能成真，这才是真正的‘社区所有制’！ — 主播 > [09:48](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=588s) 教练，我需要几张英伟达 A100 显卡！ — 观众 > [19:01](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1141s) 一想到要造手机，我就想去死，但如果我们不得不造，我们就会造。 — Elon Musk > [25:04](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1504s) 当爱的力量胜过对权力的热爱，世界将迎来和平 — 求婚横幅 > [25:04](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1504s) 听起来很酷啊，虽然违法很遗憾，但确实挺令人佩服的 — NYPD 应急直升机飞行员 > [28:36](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1716s) 难道是黄仁勋手头紧缺几块钱，想借此筹措点资金？还是有其他人机缘巧合弄到了手，然后送去拍卖？ — 主持人 > [28:36](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1716s) 我不仅在大把买你的 GPU，现在连你的旧衣服都包圆了。 — 主持人（模仿 Mark Zuckerberg 心态） > [29:43](https://www.youtube.com/watch?v=NBwKr1YsPmc&t=1783s) 一觉醒来，我不是输家，而是拥有无限金钱。 — 黄仁勋热门表情包梗图 ## Mentioned Fable 5, OpenAI, Together AI, Vipul Ved Prakash, Joe Weisenthal, Dean Ball, Ben Thompson, Donald Trump, Coinbase, Hyperliquid, Intel, Dell, Jersey Mike's, Peter Cancro, Charlie Morrison, Blackstone, Wingstop, OPM (美国人事管理局), Joe Gebbia, Airbnb, Spike Brehm, SpaceX, Elon Musk, Starlink, Apple, App Store, Qualcomm Snapdragon, xAI, Mark Pincus, Zynga, Jack Applebee, 帝国大厦, Tom Ford, Sotheby's (苏富比), Jensen Huang, NVIDIA, Mark Zuckerberg, Meta, TBPN ## Video description Sign up for TBPN’s daily newsletter at TBPN.com 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. TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/ 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=NBwKr1YsPmc)

[0:02] 《华尔街日报》: 驯服人工智能的战役才刚刚打响。 [2:02] Joe Weisenthal: 让公司将税前利润的 20% 上缴联邦政府，政治家与监管层则通过制定行业规则来规范企业行为，而不是直接行使股东权力 [2:02] Dean Ball: 引鼠入室，任其在自家墙壁里筑巢繁殖 [7:05] 主播: 我们正是为了全体美国人民在播客，最荒诞的结局往往最可能成真，这才是真正的‘社区所有制’！ [9:48] 观众: 教练，我需要几张英伟达 A100 显卡！ [19:01] Elon Musk: 一想到要造手机，我就想去死，但如果我们不得不造，我们就会造。 [25:04] 求婚横幅: 当爱的力量胜过对权力的热爱，世界将迎来和平 [25:04] NYPD 应急直升机飞行员: 听起来很酷啊，虽然违法很遗憾，但确实挺令人佩服的 [28:36] 主持人: 难道是黄仁勋手头紧缺几块钱，想借此筹措点资金？还是有其他人机缘巧合弄到了手，然后送去拍卖？ [28:36] 主持人（模仿 Mark Zuckerberg 心态）: 我不仅在大把买你的 GPU，现在连你的旧衣服都包圆了。 [29:43] 黄仁勋热门表情包梗图: 一觉醒来，我不是输家，而是拥有无限金钱。

[0:02] the battle over how to tame AI has just begun [0:02] not limited by rules or any controlling influence [2:02] Dean Ball: ruinous, akin to inviting rats to live and reproduce in the walls of your house [2:02] Dean Ball: will never stop at 5% [12:53] Spike Brehm: acres of file cabinets as far as the eye could see [17:32] Elon Musk: the idea of making a phone makes me want to die [17:32] Elon Musk: We are not developing a phone. [25:04] When the power of love overcomes the love of power, the world will know peace

[0:02] Fable has been freed. Fable 5 is now [0:04] available. Good news for the AI world, [0:08] of course. Uh the Wall Street Journal [0:10] says, "The battle over how to tame AI [0:13] has just begun." Obviously, there's lots [0:15] of back and forth on how these models [0:17] get released, who gets access, when [0:19] people get access, will the government [0:21] get access at certain points, how [0:23] fettered is the intelligence, how [0:25] unfettered is the intelligence. But, [0:27] there's a number of interesting uh news. [0:30] The The interesting uh detail that [0:32] people are sort of pulling at is um the [0:35] new Fable 5 that is now released seems [0:39] to have been uh nerfed in some ways. [0:42] There's always been questions about, you [0:43] know, okay, you don't want it to be able [0:45] to build a bio weapon or hack a system. [0:47] So, uh locking down cybersecurity makes [0:50] sense, but you should probably disclose [0:53] that and be transparent about it. And [0:55] then, the question of how do you uh deal [0:58] with the uh self-replicating AI. If you [1:01] just ask it to build an AI system that [1:02] can build you a bio weapon, well, then [1:04] you have the the permission. [1:06] >> me unfettered intelligence." [1:07] >> Exactly. So, you have to nerf that. [1:08] >> Cambridge Analytica unfettered, not [1:10] limited by rules or any [1:12] >> Wait. Wait. What What did you say? [1:14] >> Unfettered. [1:15] >> No. No. What What source is this? [1:17] >> Uh Cambridge. [1:19] >> You said Cambridge Analytica. [1:21] >> [laughter] [1:22] >> Which I don't think has a dictionary [1:24] with [1:24] >> They should launch a dictionary. [1:26] >> That would be good. [1:27] >> be great. [1:27] >> Maybe. [1:28] >> The Cambridge dictionary. [1:29] >> But, the dictionary uh the dictionary [1:31] definition of unfettered is [1:33] >> not limited by rules or any controlling [1:35] influence. So, if there's like if [1:37] there's like [1:38] >> you're going to get regulated [1:39] immediately. [1:39] >> Well, I just think like Neo Lab, like [1:41] everyone wants to be the cool, friendly [1:43] Neo Lab. There's space for kind of a [1:46] sketchy, villainous [1:49] >> [laughter] [1:50] >> Well, we have Vipul from [1:51] uh Together AI building open-source AI [1:54] cloud infrastructure coming on the show. [1:56] I'm sure he can talk about what's [1:58] capable, what's possible when you are [2:00] building unfettered intelligence. Uh [2:02] this is from the Financial Times. OpenAI [2:04] proposes handing Trump administration a [2:06] 5% stake. Joe Weisenthal is pushing back [2:09] already. He says, "I don't know about [2:10] this path. Rather than equity stakes, [2:12] why not make companies pay 20% of all [2:15] pre-tax income to the federal [2:17] government? And then instead of [2:18] exercising shareholder influence, [2:20] politicians and regulators could set [2:23] rules on corporate conduct across [2:25] industries. [2:26] Interesting. [2:27] >> where he's going with this. Someone [2:29] should try this. At least try it out. [2:31] >> contrarian take. But yes, um Dean Ball [2:33] already chimed in on this news. He said, [2:36] "There are two broad ways this can work. [2:38] One, you divide this 5% over all US [2:41] households, handing each a direct stake, [2:43] or two, you give the stake directly to [2:45] the government. One is fine. Two is [2:48] probably ruinous, akin to inviting rats [2:50] to live and reproduce in the walls of [2:53] your house." Which is wild. [2:55] Uh he says, "It will never stop at 5%." [2:57] >> till you try it, Dean. [2:59] >> Yeah. [2:59] >> There's There's some rats out there that [3:01] can be quite excellent roommates. [3:04] >> Are you a rat guy? I thought you were a [3:05] mosquito guy. I thought you like [3:06] mosquitoes. [3:07] >> No, I do like squirrels. [3:08] >> Squirrels, okay. [3:09] >> Squirrels [3:10] >> Yeah. [3:10] >> I feel like I have a bond with [3:11] squirrels. [3:12] >> They're sort of the air air I see a [3:13] squirrel [3:14] >> in my yard, I look at it and I just [3:16] salute. And and [3:18] >> People have said you have the mind of a [3:19] squirrel. [3:20] >> They have. They've They've said that I [3:22] have the mind of a rodent. [3:24] >> [laughter] [3:25] >> Yes. [3:26] Uh [3:26] Dean Ball continues. He says, "It will [3:28] never stop at 5%. It will go on and on. [3:30] The governance will become a nightmare. [3:31] Political capture will be real. And it [3:33] will generate precisely no goodwill with [3:35] the public." Yeah, that's a good point. [3:37] Um and and we were talking about this [3:39] before, the this idea of like [3:40] >> Yeah, one of the challenges is even the [3:41] way this was phrased in the Financial [3:43] Times is that it The assumption is that [3:46] it's it sounds like they're giving [3:48] shares not to a sovereign AI fund, which [3:52] is I think what was actually proposed, [3:54] but just giving it to the White House. [3:56] >> Yeah, yeah, yeah. [3:57] >> Which given the disclosures that were [3:59] made earlier this week. [4:01] >> not having fun with it. Said Trump [4:03] denies conflicts of interest as filing [4:05] show he made more than $1 billion last [4:07] year. $1.4 billion of stocks bought, [4:10] $432 million sold. Deals for watches and [4:14] perfume. Yes, the Trump watch happened. [4:16] I saw a crazy chart that said that Trump [4:19] earned out-earned Coinbase in terms of [4:22] profit from crypto [4:24] in the last 12 months or something like [4:26] that. I forget the exact stat, but it [4:28] showed all the different crypto [4:30] platforms and many of them were down [4:32] because it's been a rough market. Some [4:34] of them were up, but no one is up more [4:37] than the president of the United States. [4:39] Wild times, wild times. But [4:41] yes, it seems like the green shoot here, [4:44] the white pill, the optimistic outcome [4:47] is that if there's lab nationalization, [4:49] if there's some sort of distribution of [4:52] equity to Americans, it happens through [4:55] the Trump accounts, through social [4:57] security number, through some sort of [4:59] identity that the individual can have a [5:03] stake in as opposed to [5:05] um [5:06] as opposed to just going into some fund. [5:08] We were talking about this with data [5:09] centers where people are [5:11] >> Tyler has the data, we can pull it up. [5:13] >> Okay, let's pull it up. [5:15] Wait, do you have it? Or no? [5:17] >> Well, yes, so Coinbase made $1.26 [5:19] billion. [5:20] >> Okay. [5:21] >> But pull up this chart it because it is [5:23] pretty remarkable. [5:24] >> Yeah. [5:27] But [5:28] as I was saying with data centers [5:29] >> You guys You guys okay? I know it's 4th [5:32] of July weekend coming up, [laughter] [5:34] there you go. But [5:36] >> There we go. [5:37] Look at this. Okay, scroll up a little [5:38] bit. [5:39] >> Let's get it a little centered. There we [5:41] go. [5:42] >> Core scientific [5:43] >> Let's lock it into the center. [5:44] >> galaxy [5:45] >> Let's lock. [5:46] >> There we go. [5:47] >> And there we go. Amazing. Thank you. [5:49] >> Out earned biggest listed US crypto [5:52] firms. Uh you have to imagine that it is [5:55] is hyper liquid doing better. I feel [5:57] like hyper liquid was on a tear for a [5:58] while. I don't know how they're doing [5:59] recently, but anyway, staggering chart. [6:02] But as we were talking about with data [6:03] centers, uh there's one idea, which is [6:06] uh just tax them, increase the tax base. [6:08] But having taxes go into your local [6:11] government is great and can provide [6:12] roads and schools and pay for things and [6:14] balance the budget and that's great, but [6:15] it's abstract as opposed to the Ben [6:17] Thompson proposal, which was data center [6:19] shows up in a small town, it's in, you [6:22] know, a uh a sparsely populated area. [6:24] There might only be 5,000, 10,000 [6:26] residents in the city, uh but they all [6:29] get checks for He He ran the numbers and [6:31] it was like $1,000 a month for some of [6:33] these data center projects where they're [6:35] making so much money and the town's so [6:38] small, they can do a meaningful check to [6:40] the individuals. And that sort of direct [6:43] payment I think is going to be better [6:44] received than this abstract, "Oh, yeah, [6:47] my government's better funded." But are [6:49] they going to cut my taxes? Are they [6:50] going to successfully build that road? I [6:51] don't know, and it doesn't help me pay [6:53] my bills, which I think is what people [6:55] want. [6:55] >> What are you saying? Trump did in fact [6:57] out earn hyper liquid by a meaningful [6:59] margin. Hyper liquid generated 908 [7:02] million in total fees in 2025. [7:03] >> Brutal. [7:04] >> Yeah, so just like running very quick [7:05] numbers on the opening I think. So uh if [7:07] it's like 5%, let's say it's at a [7:09] trillion or something, that's 50 [7:11] billion. [7:12] Uh if you're giving it to households [7:13] like like this idea, uh there's like 130 [7:15] million households, so that's like $371 [7:19] of like equity per household, which I [7:21] think is like sure, maybe that's that's [7:22] meaningful, but it doesn't seem like [7:24] that's really going to like change [7:25] people's perceptions of the labs if [7:27] that's like part of the goal, right? [7:29] >> Well, I think if it's uh I mean, so [7:31] >> Maybe they need to take you the right [7:33] parlay with that. [7:35] With that. [7:36] >> So, I mean, the the nice thing about the [7:37] Trump accounts is that they are set up [7:40] in a way that you can, uh, elect to [7:44] donate to just [7:47] children born this year. And when you [7:49] segment by that, the denominator gets a [7:52] lot smaller, and so the impact gets a [7:54] lot bigger. So you could be all of a [7:57] sudden looking at something like a few [8:00] thousand dollars for every kid born this [8:02] year, something like that. You can [8:04] narrow it down. You could do every kid [8:06] under five, or you could do everyone [8:07] under 18. And that type of I mean, [8:10] you're you're right. It's not going to [8:12] move the needle as much as broad, uh, [8:15] broad distribution of for everyone, but [8:17] um, you can be a little bit more [8:18] targeted with it. I don't know. There [8:20] there's different approaches. [8:21] >> Ryan says, "Wait, does America get 5% of [8:24] TBP?" [8:25] >> Yes. [8:25] >> I was never thinking about that. [8:27] >> Yes. We podcast for you, America. [8:29] >> That's right. [8:30] >> Uh, this is the the funniest outcome is [8:31] the most likely, and that's exactly [8:33] what's happening here. [8:34] >> Community ownership [8:35] >> Yeah. [8:36] >> of this show. [8:37] >> Somebody called me a fed based on this [8:38] news. [8:39] I said, "I I [8:41] don't even know what to do with that." [8:42] But, uh, [8:42] >> [laughter] [8:43] >> I I I guess it's somewhat true. Uh, the [8:45] actual story here, [8:47] just to, uh, reset. OpenAI is apparently [8:50] in talks, according to the Financial [8:52] Times, with the Trump administration to [8:53] give a 5% equity stake in the company. [8:56] The Financial Times describes the talks [8:58] as early conversations, but the plans [9:00] seem to involve other frontier AI [9:03] players apportioning stakes in their [9:04] companies as well. The obvious context [9:06] around this story is, at least on a [9:08] straightforward read, that AI companies [9:11] need at least to stem the seemingly [9:13] ever-growing negative sentiment around [9:15] AI and data centers. Doing business with [9:17] the Trump admin has turned out well for [9:18] other companies like Intel and Dell. [9:20] Intel's the big one, where [9:22] post-government stake, I think the stock [9:24] is up 200 or 400% something like that. [9:26] >> 400. [9:27] >> 400%. Very, very strong performance for [9:30] Intel. Uh, and being difficult to do, [9:32] uh, being difficult to do business with [9:34] the admin has been unambiguously bad for [9:37] others. And now there's a vehicle with [9:39] the Trump accounts that can facilitate [9:40] equity transfers from corporations to [9:42] average Americans. Again, these are [9:43] early conversations, so we're very light [9:45] on the details, but we'll be discussing [9:47] this news, of course. In other news, [9:49] Jersey Mike's is filing for [9:52] initial public offering. Jersey Mike's [9:54] is going public. [9:56] The sandwich chain, Jersey Mike's filed [9:58] for IPO today. Uh it's looking for a $12 [10:01] billion valuation after being acquired [10:03] by Blackstone last year for $8 billion. [10:06] >> Let's give it up for Blackstone. [10:08] >> Yeah, they needed a win. They're getting [10:10] a nice 50% markup in just a year. Jersey [10:13] Mike's [10:13] >> Forced out of buying single-family [10:15] homes. [10:16] >> Yeah, that's [10:17] >> true at all. There hasn't been any There [10:18] hasn't been any regulation around that [10:20] whatsoever. [10:21] >> Uh they do own 95% of homes, apparently, [10:24] something like that. No, it's like it's [10:26] like half a percent or something. It's [10:27] low. But, Jersey Mike's has 3,300 [10:31] locations across the US and Canada, [10:33] plans to open 300 shops in the UK and [10:35] Ireland. Uh Peter Cancro, Jersey Mike's [10:39] founder, who's not named Mike, you would [10:41] assume he was, but the founder of Jersey [10:44] Mike's is in in fact named Peter. He [10:46] started the business in 1975 when he was [10:48] 17 years old. Wow, young. [10:51] >> Overnight success. [10:53] >> founded the business when he purchased a [10:55] sandwich shop in Point Pleasant, New [10:57] Jersey, called Mike's Subs, with a [11:00] $125,000 [11:02] loan that he got from his then football [11:05] coach, who was also a banker. So, you're [11:09] playing [11:09] >> Incredible. [11:10] >> high school football in New Jersey, and [11:13] you have an idea to buy a sandwich shop [11:17] >> Coach is like, "This guy's a star. I [11:19] should give him a six-figure SBA loan." [11:22] >> Also, $170 $125,000 in 1975 is like $25 [11:27] million by today's standards. [11:30] Isn't it? [11:30] >> Also, that's just such a such a at least [11:33] personally at 17, if I had [11:36] a couple thousand dollars in my bank [11:38] account, I was feeling like the king of [11:40] the castle. So, $125,000 [11:43] to buy a sandwich shop, we don't really [11:45] know what the financials were of the [11:46] sandwich shop. It's very possible that [11:49] you know, it was just such a slam dunk [11:51] deal that even a 17-year-old [11:54] could operate it. [11:56] >> It might have just been a no-brainer, [11:58] but but clearly [11:59] >> So, uh [12:01] $125,000 [12:02] in 1975, just using inflation, not [12:06] investing in the market, not [12:07] compounding, uh using the general CPI [12:11] inflation numbers, that gives you [12:13] $775,000 [12:16] to in today's dollars. So, if you're a [12:17] 17-year-old football player, just go to [12:19] your football coach and say, "Hey, can [12:21] you can you loan me 3/4 of a million [12:23] dollars?" [12:24] >> CH in the chat says, "Coach, I need [12:25] A100s." [12:27] >> Yeah, that's [12:27] >> You might be able to go to your coach [12:29] and and you know, basically get the [12:31] financing to set up some type of [12:32] Neocloud. [12:33] >> Yeah. So, he grew the company for almost [12:35] 50 years, 1975 to 2025 basically, before [12:40] transitioning to chairman this April. [12:43] Former Wingstop CEO Charlie Morrison now [12:46] leads the company. The ticker will be [12:48] JMKE, [12:50] Jersey Mike's. So, go check it out. The [12:53] other news, the Office of Personnel [12:55] Management, OPM, otherwise known as [12:59] other people's money, [13:00] is ending taking other people's money [13:02] via paper for retirement. OPM ends [13:05] paper-based retirement process. The US [13:08] Office of Personnel Management says it [13:10] has digitized the retirement process for [13:12] government employees. This might seem [13:15] like a boring news item, but the story [13:17] behind it is actually kind of wild. Up [13:19] until now, the process for government [13:21] employees and US veterans to retire and [13:24] start receiving their pensions could [13:26] take up to 6 months. That's a long time. [13:29] This is because it's entirely [13:31] paper-based and manual. Spike Brehm, who [13:33] helped with the effort, described how [13:34] crazy this turned out to be in reality [13:38] in a post on X. Uh they said that they [13:40] did it. The mad lads actually did it. I [13:43] never talked about my time at Dodge last [13:45] year because it was so controversial and [13:46] contentious. Early last year, Joe [13:48] Gebbia, founder of Airbnb and current uh [13:53] uh US chief design officer, recruited a [13:56] handful of his most trusted early Airbnb [13:58] engineers to embed at the Office of [14:00] Personnel Management to solve the [14:02] retirement paper problem. The processing [14:04] of federal retirement took months, and [14:07] in the extreme, retirees could wait up [14:08] to 6 months for their full pension to [14:10] arrive. What was the hold-up? Paper. [14:13] Remember hear hearing Elon talk about [14:15] the mine in Pennsylvania? We got to [14:17] visit it in deep underground caverns [14:20] blasted out of limestone. There were [14:23] literally acres of file cabinets as far [14:26] as the eye could see. [14:26] >> to go ahead and say this was a feature, [14:28] not a bug. [14:29] >> You like this? [14:30] >> people generally yearn for the mines, [14:32] and so it's possible that it's possible [14:35] that it wasn't the most efficient way to [14:37] do business, but it was the most fun [14:39] way. [14:39] >> But now you're like a single-issue [14:40] voter. Like we got to return. We got to [14:42] go back. [14:42] >> to bring back the mine. [14:43] >> Yeah. Okay. Okay. You're you're into the [14:46] mines. Uh a controversial position, but [14:49] uh enjoy it. Storing files detailing [14:52] federal employees' employment and pay [14:54] stub history, a simple case might be a [14:56] quarter or a half inch thick, but really [14:58] complex cases filled up whole filing [15:01] cabinets. One famously took up a whole [15:03] pallet. I wonder what was going on with [15:05] that particular federal government [15:07] employee that they had a whole pallet. [15:09] Were they jumping around from job to job [15:11] or something? I don't know. Each case [15:13] was handed was hand processed by case [15:15] workers in cubicles deep underground. [15:17] So, you clock in at the ultimate lock-in [15:21] factory, in the mine in a cubicle [15:24] underground processing someone's [15:26] pension. They check calculations, made [15:28] sure forms were filled out properly, [15:30] many weren't, and handed the long tail [15:32] of complex issues. We watched as they [15:34] keyed data into a black and white [15:35] terminal transmitting to the COBOL [15:38] mainframe built many decades ago. We got [15:40] to keep the COBOL flowing for sure. We [15:42] can't update that. It's such a good [15:44] programming language. It just sounds so [15:46] cool, COBOL. No one wants to write Ruby [15:48] on Rails. Want to write COBOL. Since [15:51] cases were processed by hand, there were [15:53] multiple rounds of human review and [15:55] additional rounds for complex cases. [15:57] Case files were walked around between [15:59] one worker's outbox and another's inbox. [16:01] Sometimes it would sit in place in one [16:03] place for days waiting to be picked up. [16:05] To OPM's credit, they'd done multiple [16:06] rounds of digital transformation [16:07] spanning decades, so some systems were [16:09] newer than others. Then there was a big [16:11] effort in the mid-90s, but the systems [16:13] were desperate or disparate and it was a [16:16] total maze getting them to talk to each [16:17] other. There was a big effort to build a [16:19] web app where employees applying for [16:20] retirement could digitally fill out the [16:22] forms necessary just to be mailed into [16:25] the mine [laughter] and stuffed into the [16:26] paper file. [16:28] Uh and a few federal agencies were even [16:30] using it. Uh when we arrived, OPM was [16:32] midway through a fresh attempt at [16:34] digital transformation delivered by a [16:36] software contractor. The black pill was [16:38] seeing the entire the terrible quality [16:39] of the software in interaction with the [16:41] contractors coming from Silicon Valley. [16:43] I couldn't believe how low the talent [16:44] quality bar was for selling software to [16:46] the government. It's clear, as the UOG [16:48] USDS people explained to me a decade [16:51] ago, the primary skill these vendors [16:52] have is securing government contracts. [16:54] It's a huge moat. Delivery of quality [16:56] product be damned. We fired the vendor [16:58] and took over the project. They'd been [17:00] working on it for more than a year and [17:01] there was another year before they were [17:03] going to deliver it. At first, we tried [17:05] to bend it to our will to actually [17:07] connect all the various data sources and [17:08] get decent UX for case workers in the [17:11] mine to use, but we soon realized we [17:14] were going to have to rebuild the whole [17:15] stack from scratch. It was around this [17:16] time I had to go back to New York. I had [17:18] a new job waiting for me, a 4-month old, [17:20] a wife whose patience was running out, [17:21] but I got to watch from afar as the team [17:23] cranked day and night, hitting early [17:24] milestones. Now, they've fully done it. [17:27] Huge congrats to the team. Wow. Yeah, [17:29] very interesting. There's other news. [17:32] SpaceX, apparently, [17:34] is maybe working on a phone. At the very [17:37] least, The Wall Street Journal is [17:39] reporting that during SpaceX's recent [17:43] IPO roadshow, before SpaceX went public, [17:46] they were showing investors a prototype [17:49] of something sort of like an iPhone. [17:53] AI enhanced, maybe thinner. The Wall [17:55] Street Journal has some reporting around [17:58] it, and it's an interesting story. [18:00] Elon's denied this a bunch, but [18:03] the reporting seems pretty good at this [18:05] point that at least it was a demo. [18:07] >> It's [clears throat] hard to see why he [18:10] wouldn't at least take a shot at it. [18:12] That's kind of my is my point of view. [18:14] He has one of the greatest telecom [18:16] companies ever created. Global satellite [18:18] network. Theoretically, over time, any [18:20] person on the entire planet could have [18:24] an internet connected, you know, SpaceX [18:27] device with Starlink. [18:29] It makes a lot of sense. The other [18:31] factor that I've just continued to think [18:33] is is quite interesting is the App Store [18:35] as a moat has is is being diminished by [18:39] AI. There's just so many things There's [18:42] so many things There's so many apps on [18:43] my phone that like 5 years ago I used a [18:45] lot, today I don't use at all. So, an [18:47] example being like I don't even need [18:49] like the weather app cuz I can just be [18:51] like I can ask Chad give me the weather [18:53] for the next week. Obviously not [18:55] Uh yeah, and I don't even need that cuz [18:57] I can just be like vibe code me my own [18:59] beer app today. [19:01] Uh no, but obviously it's not like an [19:03] efficient way to use It's probably [19:05] better to just open the weather app, but [19:07] at the same time I'm very used to it. [19:09] The other thing is like even things like [19:10] surf reports. Like I'll just pull a surf [19:12] report in chat. [19:13] >> Yeah, yeah, yeah. Yeah. [19:14] >> Um and so there's a lot of like the long [19:16] tail of apps are no longer like a lock [19:18] in to the iPhone for me and I assume a [19:21] lot of other people. [19:22] >> Okay. [19:22] Yeah, we got to have a chicken and an [19:24] egg debate about this. Uh let me finish [19:28] reporting the uh SpaceX news and then I [19:30] want to ask you to dig into the source [19:33] of why that's happening. But first, [19:35] SpaceX developed an early prototype of [19:37] an AI-focused handset. The device is [19:40] said to be slimmer than the iPhone, runs [19:42] proprietary operating system, uses XAI [19:44] technology, built around Qualcomm's [19:46] Snapdragon chipset. So, they're not [19:48] vertically integrating the chip like [19:50] Apple silicon just yet, but Elon has fab [19:53] chips before for Tesla. So, not out of [19:56] this uh not out of the question. Elon [19:58] Musk reportedly demoed the prototype to [20:00] potential investors during SpaceX's [20:02] recent IPO roadshow. The product fits [20:04] with with Musk's broader everything app [20:06] vision, but it's still early and may [20:08] never ship. Rumors about a Musk-built [20:10] phone have have circulated for years. [20:13] There's videos on YouTube with like tens [20:15] of millions of views that are like the [20:16] Tesla Pi phone is coming next week. And [20:20] it's all just completely fli- fake [20:22] clickbait, but it the the demand is [20:24] there. It gets you every [20:25] >> It [laughter] doesn't get me every time. [20:26] Stop it. [20:28] But um [20:29] uh Elon's actually commented about that. [20:31] >> Apple Vision Pro, you see the video and [20:33] you're like, I I shouldn't click it. I [20:34] know it's fake. I shouldn't click it. I [20:35] know it's fake. And then you click it [20:37] and you're like, ah, they got me again. [20:38] >> Yeah, you know, it is ridiculous, but [20:40] Elon has had to comment on this many [20:42] times. Uh he said the idea of the idea [20:44] of making a phone makes me want to die, [20:47] he said in October. But if we have to [20:49] make a phone, we will. [20:51] Funny way to say it. In February, he [20:53] denied a Reuters report that SpaceX was [20:55] building a phone that could connect [20:57] directly to Starlink satellites posting [20:58] on X, we are not developing a phone and [21:01] uh no one is commenting on this right [21:03] now. Uh so, very, very early stage. My [21:06] question about apps, I think you're [21:08] right, the number of apps it used to be [21:10] every month there was a hot app, every [21:12] year there was a hot app, [21:14] specific apps for specific things. What [21:17] is driving the collapse? Because when we [21:18] talked to Mark Pincus yesterday, he said [21:20] that Apple is not pushing small apps. [21:23] Like Apple is not a the App Store is not [21:26] a reliable source of distribution for [21:28] Vibe Coded apps, for games, for [21:30] anything. And so back in the day he was [21:33] able to grow Zynga on the Facebook [21:35] platform, which was very open and you [21:37] could go viral within Facebook for a [21:39] completely separate piece of software, [21:41] not just content, and then you could [21:43] keep that customer relationship, [21:44] monetize that customer relationship off [21:46] platform forever. [21:48] Not the case anymore. Also [21:50] just discovery, people don't open up the [21:52] App Store as much, I think, and say, [21:55] "Oh, what are the top apps? Oh, what are [21:57] the recommendations? I would love a new [21:58] entertainment app." They're like, "No, I [22:00] got one new app, please." [22:02] >> app, please. [22:02] >> [laughter] [22:03] >> They don't they don't that's not a [22:05] behavior, but why [22:06] >> to the App Store. I would like one or [22:08] two new apps, please. [22:10] >> Yeah, yeah. But why so so so why is it [22:13] not happening? Is it because the super [22:16] apps are getting more super in the sense [22:18] that you don't need you used to need you [22:20] needed Snapchat for stories, you needed [22:22] TikTok for vertical video, you needed [22:24] YouTube for long form. Now Instagram can [22:26] satisfy a lot of those things. If [22:28] there's some new feature that pops up, [22:30] Instagram will add it very quickly. As [22:32] you mentioned ChatGPT and the and the AI [22:34] chat apps can do a lot of other things [22:36] that you might need a separate app for, [22:38] but is it is it that the the apps [22:41] consolidated or is it that Apple pushed [22:44] that? [22:45] >> It's a good question. I mean, I think [22:46] part of it is [22:48] like we had 10 years of like very real [22:52] innovation around what types of [22:54] applications were [22:56] possible or needed on a like powerful [22:59] mobile device. [23:01] And so I think that you just had like [23:03] saturation, right? I still I've said [23:05] this a bunch of times on the show, but [23:07] given that uh you know, rewind to the [23:10] ChatGPT moment, I would expect there to [23:12] be a lot more like net new hit apps in [23:16] the App Store. [23:17] And [clears throat] [23:18] today, if you look at the top charts, [23:20] obviously dominated by World Cup related [23:22] things, but it's [23:22] >> Yeah. [23:23] >> uh sports betting, [23:24] >> Yep. [23:24] >> like streaming apps [23:26] >> The Trump accounts are doing very well, [23:28] obviously. Yeah. [23:28] >> Um and so it's not like, you know, if I [23:31] you know, fallen asleep at the ChatGPT [23:33] moment and woken up today, I'd expect to [23:35] go and be like, okay, there's there's [23:37] probably a bunch of new popular hit [23:39] apps. [23:40] >> And it just turns out that AI's like [23:41] been [23:42] an extending innovation for a lot of [23:45] existing apps. Um and so I just think [23:47] the combination of the of LLMs and chat [23:51] apps being able to just do so much for [23:53] you and going to continually be able to [23:55] do more for you. [23:56] >> Mhm. [23:56] >> And vibe coding is again a a positive [24:00] tailwind for anybody building [24:03] potentially a new device, even if it [24:04] was, you know, uh [24:06] especially if it's based on Android, but [24:07] even if it was like an entirely new [24:10] operating system. Elon built a new [24:12] operating system for Tesla. [24:14] >> not the Elon goal. He's not the open [24:18] platform guy when Twitter when he [24:20] purchased Twitter, the API got clo- [24:22] smaller, more expensive. [24:23] >> saying that I'm I'm not saying I'm I'm [24:24] just saying [24:25] >> ecosystem. You can't You can't deploy a [24:27] different self-driving system onto your [24:30] Tesla. It's a closed system. It's a [24:32] competitor to the Apple ecosystem. It's [24:34] an equally walled garden. But but it [24:36] would be his. [24:37] >> I'm just saying it's it's it doesn't [24:39] seem out of the realm of possibility to [24:41] take a crack at a new device [24:43] >> Oh, there's a huge incentive to. I [24:45] completely agree with that. You don't No [24:46] one wants to be locked in the Apple [24:48] ecosystem. So you want to build your own [24:50] ecosystem or your own walled garden. But [24:52] the idea that it would be a not a [24:54] non-walled garden is sort of it would be [24:59] an uncharacteristic strategy I I think. [25:02] >> I agree. [25:02] >> Yeah. Anyway, there's another story. [25:04] There was a couple that climbed to the [25:07] top of the Empire State Building [25:10] yesterday. Did you see this? [25:11] >> I did. [25:12] >> And they put up a flag and they got [25:14] married or they they they proposed or [25:16] they got engaged [25:18] and brands have immediately jumped on [25:20] it. Let's go. Let's give it up for some [25:22] advertising. [25:24] >> [applause] [25:24] >> No, there's backlash already. Jack [25:26] Applebee said this is so so [25:28] embarrassing. These brands think they [25:30] really did really think they did [25:31] something. [25:33] Loverboy is the worst. Congrats, you put [25:34] your logo on a flag, really innovative [25:36] stuff. The lack of originality and [25:37] clever thinking in modern social media [25:39] is just ridiculous. Everyone had the [25:40] exact same idea. There's a meme we got [25:43] to get on the action. It it and what's [25:45] interesting is that a lot of these are [25:47] beverage brands it looks like. Drink [25:49] culture pop, drink High Yo, drink [25:51] Loverboy, drink Spritz, Bulletproof, I [25:54] guess that's Bulletproof coffee. Um [25:57] Fashion Nova, New York City, just the [25:59] account New York City, I don't know. [26:01] Petco, stop breaking treats in two. [26:05] It is a fun meme format, but brands [26:08] always it's very very hard to break [26:10] through if you're just piling on the top [26:12] meme. What do you think? [26:13] >> And [26:14] yeah, so this whole thing was [26:17] I when when I first saw the the video of [26:21] the flag going up [26:22] >> Yep. [26:23] >> and they kind of botched the flag. I'm [26:25] not going to lie. It seems like they [26:26] didn't fully anticipate the wind [26:28] conditions that day. The flag was quite [26:31] hard. The the writing on the flag was [26:33] quite hard to read. [26:35] I would have liked to see them erect [26:37] like a full-size, you know, billboard [26:40] up there. [26:40] >> Something permanent. [26:41] >> Yeah, something something, you know, [26:44] more permanent and and certainly more [26:46] legible than the flag. [26:48] And then the quote was when the power [26:51] of love overcomes the love of power, the [26:55] world will know peace. But then they [26:57] were just kind of making out and [26:59] proposing. [27:00] >> [laughter] [27:01] >> And [27:02] yeah, just [27:03] it it felt like [27:05] >> It has It has a little bit of like the [27:06] AI cadence to it or like the don't love [27:09] your job, job your love. [27:11] >> Yeah, I mean that's that's that's that's [27:12] that's a [27:13] >> make a lot of sense. Yeah. [27:14] >> Yeah. [27:15] >> Could be a little more terse, I suppose. [27:17] But anyway, [27:19] um [27:19] >> I just don't know if they were [27:21] >> I'm still impressed. It's just hard [27:22] >> Yeah, it was quite impressive. [27:23] >> with a climbing. How did they Did they [27:25] climb up from the sidewalk or did they [27:27] take the elevator and then hop out and [27:29] then start climbing? Cuz this wasn't [27:31] like an Alex Honnold from the base free [27:34] climb free solo attempt [27:35] >> How fast is this helicopter going by the [27:38] way? [27:38] >> That's just the lens, baby. This is a [27:39] 600 mm lens. It's lens compression, [27:42] Jordy. You're learning optics. Optics. [27:44] This is looking through a telescope [27:46] basically. [27:47] >> Dave says quick tip for goth couples [27:49] from The Matrix, make sure to account [27:52] for wind speed. [27:53] >> yeah. [27:53] >> I agree. [27:54] >> I heard some some recording of the [27:56] helicopter pilot who was dispatched sort [27:59] of emergency response NYPD and I think [28:03] [laughter] with helicopter pilots like [28:05] they're like [28:06] alert, sir. We have two individuals on [28:08] the top of the Empire State Building. [28:10] He's like that sounds cool. [28:11] >> [laughter] [28:11] >> He's like that's awesome. Why are they [28:13] up there? They're like well, they're not [28:14] supposed to be. They're trespassing. [28:15] He's like okay, bummer but still [28:18] impressive. Anyway. [28:20] >> Yeah, good good story. Good story for [28:22] them. I don't know if it made the the [28:24] sort of maybe [28:25] political impact or whatever impact they [28:28] were trying to have but happy for them. [28:30] >> Mhm. [28:31] >> And congratulations to [28:33] to the couple. [28:35] >> Yeah. Yeah. [28:36] >> You know what I need, John? [28:38] >> you need? [28:38] >> Jensen's signature black leather jacket. [28:42] >> His leather jacket by Tom Ford is being [28:44] auctioned off. I'm so curious how this [28:48] ended up with Sotheby's. Did Jensen pass [28:50] it along? He needed a few [28:52] He needed a few bucks and just said [28:54] like, you know, wanted to use it as a [28:55] way to raise some capital or did someone [28:57] else get their hands on it and take it [28:59] to Sotheby's? [29:01] I So, there the estimate from Sotheby's [29:03] is $40 to $60,000. [29:05] I just think this goes way higher. I [29:07] think there's so many people that would [29:09] love to to own this just for the the [29:12] novelty of it. It's a very unique item [29:15] that [29:16] will be a part of history. And so, 40 to [29:19] 60 just seems way too low. [29:22] >> What's funny is a lot of the people that [29:24] can afford this are like competing with [29:26] Jensen and probably don't want to glaze [29:28] him too much. Like you don't see Mark [29:29] Zuckerberg pulling this out probably cuz [29:31] he's like, "I'm I'm going to maybe the [29:33] best. I I'm going to have the biggest [29:34] market share." [29:35] >> to spend the most money with you. [29:36] >> Yeah. [29:37] >> [laughter] [29:38] >> I'm not just buying your GPUs. I'm [29:40] buying your old clothes. [29:41] >> Yeah. [29:42] >> I'm buying your old clothes. [29:43] >> We got to close out maybe with this [29:45] with this Jensen meme from Andre. I wake [29:48] up not a loser, infinite money [29:51] because Meta employees consumed 73.3 [29:55] trillion AI tokens in a single month, [29:57] which roughly costs 221 million a month [30:00] or around 2.65 billion a year. I thought [30:03] it was higher. That sounds low based on [30:05] what I thought they were spending in [30:07] terms of tokens. Might even be higher, [30:08] who knows. But [30:10] it is it is very interesting. Nvidia has [30:12] also partnered with AI cloud firms with [30:15] revenue sharing taking circularity to a [30:19] previously unseen level. I don't know. [30:22] Sale and leaseback, all sorts of these [30:24] financial deals. These happen all the [30:26] time in other industries. It's just sort [30:29] of new seeing it in tech because we are [30:31] in this industrial era. So, [30:34] very interesting news, but good news for [30:36] Nvidia. Anyway, thank you for watching 2 [30:38] Minute Papers and we are off tomorrow. [30:40] It is [30:42] the 4th of July. And so, we're having a [30:45] long weekend, folks. And leave us five [30:47] stars on Apple Podcast and Spotify. Sign [30:50] up for our newsletter at tvp.com. [30:53] >> And we'll see you on Monday. [30:55] >> Cortisol rising. [30:57] >> Have a good one. [30:57] >> flashbacks. [30:58] >> Goodbye.