Big Tech Earnings Quadkill Tests AI Infrastructure ROI as CapEx Surges

资本支出激增之际，四大科技巨头同日放榜考验AI基建ROI 随着Google、Microsoft、Amazon和Meta公布季度财报，华尔街要求其证明在AI数据中心和芯片领域投入的数千亿美元能够转化为高利润率收入。 本期内容聚焦备受瞩目的“科技财报四杀日”：在主要央行公布利率决议后的同一天，超大市值巨头Alphabet、Amazon、Meta和Microsoft集中公布季度财报。主持人前瞻了各科技巨头的财务指标及AI资本支出策略，深入分析巨额基础设施投入究竟带来了切实的投资回报率，还是正在压低利润率。他们还讨论了市场对硬件供应链瓶颈的反应（如Intel股价暴涨），并探讨了社交媒体上走红的一个关于公众信任与集体生存的博弈论思想实验。 央行利率决议叠加史无前例的“科技财报四杀日”，让这一天成为宏观经济与科技市场具有历史意义的一天。 (0:02) Alphabet、Amazon、Meta和Microsoft同时发布财报，这四家公司占S&P 500指数总市值的近20%。 (0:02) 科技巨头正大举推高AI资本支出，但其面临的软件经济效益特征却更接近资本密集型基础设施，而非高利润率软件。 (1:29) Google的搜索单店/单位经济效益正接受严格审视，投资者在评估AI Overviews究竟是能扩大搜索使用量，还是会压缩利润率。 (3:10) Microsoft展现出强劲的企业级AI增长，Azure增速加快至39%，剩余履约义务（RPO）达到6250亿美元。 (4:27)

On a day when the Federal Reserve held US interest rates steady at 3.50% to 3.75%, TBPN previews an unprecedented simultaneous earnings slate from Alphabet, Amazon, Meta, and Microsoft—which together represent nearly 20% of the S&P 500—before analyzing a viral game-theory dilemma on collective human survival. From High-Margin Software to Capital-Intensive AI (1:29) Although core legacy businesses across the tech giants remain healthy, Wall Street is focused on whether massive AI capital expenditures (CapEx) will convert into durable, high-margin revenue. Every Magnificent Seven company is spending at least $100 billion on CapEx, led by Amazon's $200 billion baseline. While a note from SemiAnalysis highlights that accelerating hyperscaler cloud revenues and positive returns on investment should drive upward CapEx revisions that beat expectations, the hosts emphasize a fundamental shift in software economics. Unlike the previous era of building websites with 80% gross margins, AI infrastructure resembles capital-intensive industries like railroads or oil refining. Upfront cash is flowing out to Nvidia, power utilities, and data center builders, creating a "matching problem" as heavy depreciation mounts ahead of bottom-line cash recovery. Divergent Paths to Hyperscaler AI Monetization (4:27) The four reporting giants illustrate distinct models for recouping AI spending through enterprise cloud contracts and advertising. Microsoft, operating at a ~$150 billion CapEx run rate, grew Azure 39% in its prior quarter and reported a 110% surge in Remaining Performance Obligations to $625 billion—roughly 45% of which is tied to OpenAI. Alongside GitHub Copilot surpassing a $500 million annual run rate, investors are watching Microsoft 365 licenses to see if AI agents multiply or consolidate traditional software seats. Similarly, Amazon is backing its $200 billion 2026 CapEx guidance with 24% AWS growth in Q4 (with investors hoping Q1 reaches 30%+ against a $36.7 billion consensus) and a high-margin quarterly advertising engine of $21.3 billion (up 23%). In consumer and ad-driven ecosystems, monetization hinges on unit economics and deployment speed. Google benefits from a vertically integrated stack spanning custom TPUs, DeepMind, and Android, but faces questions over whether AI Overviews and Gemini can monetize search queries quickly enough to preserve traditional search ad margins. Meta, by contrast, bypasses enterprise sales friction entirely. After generating $30.8 billion in Q4 Family of Apps operating income across 3.58 billion daily active users—easily absorbing a $6 billion Reality Labs loss—Meta can immediately A/B test new AI models to optimize Instagram and Facebook ad placements, helping justify a CapEx increase from $72.2 billion last year to $115–$135 billion this year. Supply Chain Bottlenecks and Ecosystem Jitters (14:15) Despite underlying operational strength, investor patience regarding AI spending remains thin. Negative headlines surrounding OpenAI's financial commitments triggered sharp volatility in infrastructure partners like Oracle, SoftBank, and CoreWeave, which fell 4% in after-hours trading before rebounding 8.4% the following day; OpenAI publicly countered that its business is "firing on all cylinders." Synovus Trust analyst Dan Morgan noted that market anxiety is not universal, pointing to solid earnings from IBM, Texas Instruments, and Intel. Following an endorsement from Jim Cramer, Intel shares surged 10% in a single day and 40% over the week, a rally the hosts attribute to market recognition of widespread CPU shortages and AI hardware bottlenecks. Self-Interest vs. Trust in the Button Dilemma (17:33) Shifting from markets to human behavior, the hosts examine a viral thought experiment popularized by Tim Urban and MrBeast: every person on Earth votes Red or Blue; if over 50% choose Blue, all humanity survives, but if 50% or fewer choose Blue, Blue voters perish while Red voters survive. Analyzing the choice in isolation, commentator Mike Solana and roughly 67% of TBPN's live chat favored Red to guarantee personal survival. However, the hosts argue that Blue maximizes expected collective survival because it requires only a 51% majority rather than impossible 100% coordination. Large-scale data supports this prosocial outcome: Blue won 58% in Urban's poll (24 million views), 55.7% in MrBeast's poll (10 million views), and a 3-to-1 margin in researcher David Shor's survey of 14,000 Americans, where choosing Blue strongly correlated with psychometric measures of honesty. As Alphabet, Amazon, Meta, and Microsoft report their quarterly results—followed by Apple on Thursday—the market's central unresolved question remains how quickly massive upfront AI infrastructure spending will translate into structural profit growth before hardware depreciation erodes margins.

TBPN指出，随着美联储按兵不动以及四大科技巨头同日公布财报，市场正聚焦超大规模AI资本开支能否转化为实质性高利润回报；同时，节目通过“红蓝按钮”思想实验，探讨了个体自私理性与群体信任协作之间的博弈。 宏观决议与科技巨头财报交汇 (0:02) 宏观经济与科技行业迎来重要节点。在加拿大央行发布利率决议后，美联储宣布将美国基准利率维持在3.50%至3.75%区间，符合市场主流预期，美联储主席Jerome Powell随后举行新闻发布会。宏观政策落定为盘后的重磅财报拉开序幕：Alphabet、亚马逊、Meta和微软四家超大规模科技巨头于同一天盘后同时披露季度业绩。 这四家科技公司合计占标普500指数总市值近20%，其表现被视为整个宏观经济的重要风向标。SemiAnalysis发布的研报认为，云服务商的收入增长正在加速并展现出正向投资回报（ROI），促使分析师普遍预计各巨头的资本支出预期将被上调并超出华尔街预期。 AI开支激增与软件经济学重构 (1:29) 本轮财报季的核心议题是超大规模AI资本支出（CapEx）的可持续性。此前，亚马逊给出了2026年达到2000亿美元资本开支基准的指引，而美股七巨头成员各自的年度支出均在1000亿美元以上。尽管搜索、电商和办公软件等传统核心业务依然稳健，但投资者对AI基础设施的财务转化机制保持高度审视。 这一商业模式演变带来了严峻的“期限匹配问题”：在数据中心和芯片上的巨额前期现金支出伴随着沉重的资产折旧负担。与传统软件开发拥有80%高毛利、坐收用户流量的轻资产模式不同，当前的AI底层建设更接近铁路或炼油等重工业。大量现金流正流向Nvidia、电力公用事业公司及数据中心建造商，促使华尔街持续评估其转化为底层净利润的周期长度。 四大巨头的商业化路径分歧 (3:10) 四大超大规模企业在AI变现能力上呈现出不同逻辑。Google拥有涵盖TPU芯片、DeepMind、Android、云服务及搜索的全栈整合能力，但面临核心搜索单位经济学的考验：AI Overviews和Gemini能否在搜索查询成本上升或传统广告收入承压的背景下维持高利润率。 微软提供了最直接的企业端落地样本，其Azure业务增速达到39%，未履约合同（RPO）同比激增110%至6250亿美元（其中约45%归属于OpenAI），GitHub Copilot年化经常性收入（ARR）突破5亿美元，支撑着约1500亿美元的资本支出年化规模；亚马逊第四季度AWS增长24%，市场期待其突破30%以消化巨额开支，同时其年化接近1000亿美元规模的高利润广告业务成为其采购GPU和建设算力的核心造血工具；Meta则展现了极高的广告变现转化效率，35.8亿日活用户使得其AI模型优化能直接转化为广告匹配精度的提升，无需经历漫长的企业级软件采购周期，从而为其1150亿至1350亿美元的全年资本支出指引提供了财务支撑。 供应链瓶颈与生态依赖性担忧 (14:15) 尽管基本面稳固，但围绕AI生态依赖性的焦虑依然引发了短期市场波动。由于对OpenAI财务承诺的担忧，Oracle、CoreWeave和软银等AI基础设施概念股出现大幅震荡，其中CoreWeave在盘后下跌4%后次日反弹8.4%，而OpenAI官方对此回应称其业务运转良好、算力获取规划明确。Synovus Trust分析师Dan Morgan指出，IBM、德州仪器和英特尔等硬件与企业服务企业的稳健业绩表明市场担忧并非系统性蔓延。 与此同时，硬件底层供应格局持续分化。在市场意识到全行业CPU短缺及AI硬件供应链制约后，英特尔股票在单日上涨10%的基础上实现了单周40%的涨幅。这种行情重现了ChatGPT推出后市场迅速认识到英伟达芯片瓶颈、从而驱动其估值翻倍的历史逻辑。 红蓝按钮博弈与集体生存机制 (17:33) 节目探讨了由Tim Urban提出并经MrBeast推广的“红蓝按钮”思想实验：全人类参与私密投票，若超过50%的人选择蓝键，则100%人类幸存；若选择蓝键人数在50%或以下，所有按蓝键者死亡，按红键者存活。从孤立的个人自私理性来看，按红键是保证自身免于死亡的绝对优势策略；但从社会整体福利最大化来看，蓝键只需51%的协调比例即可保全所有人，而全员选红键若要实现零死亡则要求严苛的100%协作。 心理学框架深刻影响着人们的选择，实验可被具象为“死亡赌博”（选蓝承担个人死亡风险）或“谋杀赌博”（选红推高蓝方死亡门槛）。在TBPN直播投票中，观众意外偏向红键（蓝键仅占33%）；但真实大规模网络投票中，Tim Urban与MrBeast的统计分别有58%和55.7%的人选择蓝键。研究员David Shor对1.4万名美国人的测度调查进一步显示，蓝键以3比1胜出，且该选择与高度重视诚实和集体福祉的心理指标显著正相关，表明群体在生死抉择中更倾向于通过互信来实现整体生存率最大化。 无论是科技巨头为争夺AI算力而承担巨额折旧风险的长期资本博弈，还是公众在红蓝按钮测试中权衡自保与互信的社会博弈，两者的核心都在于如何评估不确定性中的协同价值。随着财报数据落地，市场各方将继续检验巨额基础设施投入向可持续盈利转化的真实节奏。

## Background 本期内容由科技商业访谈频道 TBPN 于 2026 年 5 月 22 日发布在 YouTube 平台。节目属于旗下 30 分钟精简栏目《Diet TBPN》，由 John Coogan 和 Jordi Hays 主持，聚焦当天科技圈核心动态与重磅讨论。 ## TL;DR 节目上半场聚焦美联储维持基准利率不变之际，谷歌、亚马逊、Meta 与微软四巨头同日披露财报的“科技四杀（Quad Kill）”，深入拆解了高达千亿美元级的 AI 资本支出（CapEx）与重资产折旧周期难题，并横向对比了各家在云业务再加速与广告变现闭环上的真实回报率。下半场则戏剧性切入在社交网络上引发热议的“红蓝按钮”博弈论思想实验，结合直播间观众民调、MrBeast 统计以及数据科学家 David Shor 的全国抽样，探讨了自保理性、群体协调风险与电车难题式的人性框架。整期节目从硬核的科技巨头财报展望自然过渡到了引人深思的社会心理学博弈分析。 ## Atmosphere & style 节目的整体基调兼具快节奏的紧迫感与硅谷极客式的幽默调侃，两位主持人的互动张弛有度、配合默契。上半场剖析四大科技巨头万亿级军备竞赛时，洋溢着一种紧盯美股风暴眼与超级碗决战般的紧绷与敏锐；下半场切换到红蓝按钮思想实验与直播间实时民调时，气氛骤然变得生动活泼，充斥着针对博弈论框架和人性自私与崇高的激烈交锋与风趣吐槽。 ## Bulletin - [00:02](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=2s) 美联储决议将基准利率维持在 3.5%–3.75% 区间不变，同日美股迎来谷歌、亚马逊、Meta 与微软同台发布财报的“科技四杀（Quad Kill）”，四巨头市值合计占标普 500 指数近 20%。 - [01:29](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=89s) 科技巨头 AI 资本支出（CapEx）全面迈入千亿美元量级，亚马逊年化开支预期高达 2000 亿美元，微软剩余履约义务（RPO）同比暴增 110% 至 6250 亿美元，标志着行业商业模式正由高毛利轻软件向铁路式的重资产高折旧模式转型。 - [09:48](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=588s) Meta 展现出独特的无损变现优势，前沿模型突破无需面对漫长的企业端采购周期，通过 A/B 测试即可直接无感部署于覆盖 35.8 亿日活用户的广告推荐闭环，带来数十亿美元的即时利润转化。 - [26:55](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1615s) 社交网络热议的“红蓝按钮”博弈论思想实验中，数据科学家 David Shor 针对 14,000 名全美代表性样本的严谨民调显示，蓝按钮以接近 3:1 的压倒性优势胜出，其核心预测特质为“我讲真话”。 ## Key takeaways - [00:02](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=2s) 美联储符合预期地维持基准利率在 3.5%–3.75% 不变，谷歌、亚马逊、Meta、微软集中公布业绩，其合计两成的标普 500 权重直接主导美股大盘走向。 - [01:29](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=89s) AI 产业正经历根本性商业模式演变，从过去坐享 80% 高毛利的软件轻资产模式转向重资产的基建开发，巨额资本开支与实际现金流回流之间存在严峻的周期匹配问题。 - [03:10](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=190s) 谷歌拥有涵盖自研芯片 TPU、前沿模型到全球终端入口的最完整 AI 全栈布局，但市场正高度审视 AI 搜索概览是否会侵蚀其传统搜索广告的单位经济模型与高利润率。 - [04:26](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=266s) 微软核心 Azure 业务增速达到 39%，未履约合同订单（RPO）同比暴增 110% 至 6250 亿美元（其中 OpenAI 贡献约 45%），向市场证明企业端对算力的真实刚需依旧极其庞大。 - [07:08](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=428s) 随着 AI Agent 展现出单体替代多名员工的潜力，传统的按席位收费（seat-based）SaaS 商业模式面临冲击；微软则凭借深厚的分发渠道复制当年 Teams 击败 Slack 的策略，持续巩固 GitHub Copilot 的企服壁垒。 - [09:03](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=543s) 亚马逊年化近千亿美元且增速达 23% 的高利润广告印钞机，为其 AWS 业务 2000 亿美元的创纪录算力大基建豪赌提供了关键的自由现金流垫底。 - [09:48](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=588s) Meta 在 AI 变现上规避了企业软件漫长的技术扩散难题，算法优化可直接实时注入 35.8 亿日活用户的推荐系统，极微小的转化率提升即可拉动数十亿美元的纯利润。 - [14:15](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=855s) 市场在巨额折旧引发的“AI 焦虑”与硬件需求爆发之间反复拉锯，通用 CPU 短缺的预期驱动英特尔股价实现单周 40% 的惊人涨幅。 - [17:33](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1053s) Tim Urban 提出的“红蓝按钮”生存思想实验引发人性博弈讨论：个人选红可获得绝对的自保生存权，但全员选蓝过半即可实现全人类无伤幸存的最优解。 - [19:53](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1193s) 从群体协调机制分析，全员选红达成零死伤需要 100% 绝对步调一致且容错率为零，而选蓝仅需大于 50% 多数即可挽救所有人，协调门槛反而更低。 - [20:26](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1226s) 虽然直播间小范围投票中自保的红按钮占据多数，但 MrBeast 等平台数百万级的大众真实投票中，蓝按钮得票率均稳定在 55% 以上并实现群体过半幸存。 - [22:17](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1337s) 红蓝博弈呈现出强烈的概念框架分歧：红按钮既能被表述为理性的自保机制，也能被视作将他人推入深渊的“终极谋杀赌局”。 - [26:55](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1615s) David Shor 涵盖 14,000 名样本的严谨学术级民调证实，蓝按钮以接近 3 比 1 获胜，且强烈认同“我讲真话”人格特质的受访者展现出了压倒性的选蓝利他偏好。 ## Detailed walkthrough ### [00:02](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=2s) 超级财报日降临：美联储决议与科技四巨头齐聚 金融和科技市场迎来了一个充满重磅消息的超级大日子。播客以 Jerb Gavin 前一天发布的日程总结开场：美东时间上午 10:00 加拿大央行公布利率决议，下午 2:00 美联储公布最新利率决议。美联储的决定刚刚出炉，将基准利率维持在 3.5% 至 3.75% 的区间不变，虽然有些人此前寄望于降息，但这一按兵不动的决定完全符合市场预期，算是一个“不痛不痒的无事发生（nothing burger）”，随后美联储主席杰罗姆·鲍威尔（Jerome Powell）将在下午 2:30 发表讲话。 然而，真正的风暴眼在下午 4:00：谷歌（Google）、亚马逊（Amazon）、Meta 以及微软（Microsoft）四家科技巨头将在同时间集中发布财报。主持人惊呼这是一场科技财报的“四杀（quad kill）”。这四家公司合计占据标普 500 指数近 20% 的总市值，任何风吹草动都会直接撼动整个美股大盘。 - 美联储在周三会议上将联邦基金利率维持在 3.5%–3.75% 区间。 - 谷歌、亚马逊、Meta、微软四巨头同时在美东时间下午 4:00 公布业绩，合计占标普 500 指数近 20% 市值。 - 半导体与云服务分析机构 SemiAnalysis 当天早间发布研报（由 YC Yield Chad 引述），预计超大规模云厂商（hyperscalers）的资本支出（CapEx）将进一步上调并击败华尔街预期，原因是云业务营收正在加速增长，且前期云投资已展现出正向投资回报率（ROI）。 ```mermaid flowchart TD accTitle: Super Wednesday Financial and Tech Schedule accDescr: Visual breakdown of events on Super Wednesday showing rate decisions followed by earnings reports. subgraph Macro["宏观经济决策"] CAN["10:00 AM 加拿大央行利率决议"] FED["02:00 PM 美联储利率决议: 3.5%-3.75% 维持不变"] POW["02:30 PM 鲍威尔讲话"] end subgraph TechQuad["04:00 PM 科技四巨头财报 (Quad Kill)"] MSFT["微软 Microsoft"] GOOG["谷歌 Google"] AMZN["亚马逊 Amazon"] META["Meta"] end CAN --> FED --> POW --> TechQuad ``` --- ### [01:29](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=89s) AI 军备竞赛的根本难题：铁路式重资产 vs 80% 毛利的软件幻梦 市场上最大的核心问题在于：整场 AI 大基建到底进展如何？科技巨头们的资本支出已经攀升到了不可思议的量级，上季度亚马逊甚至给出了高达 2000 亿美元的年化资本支出预期。如今在美股“科技七巨头（Magnificent Seven）”中，每家每年的资本开支起步都是千亿美元级别（11 位数）。尽管各家的核心成熟业务表现依然稳健——谷歌搜索在增长、亚马逊核心电商在增长、微软的 Microsoft 365 企业席位也在稳步扩张——但投资者真正焦虑的是与 AI 基础设施挂钩的“持久性收入（durable revenue）”。 这里存在一个严峻的“周期匹配问题（matching problem）”：科技巨头们投入了数以千亿计的资金，每年产生庞大的资产折旧，现金流何时能真正回流？这种新收入的护城河到底有多深？主持人指出，软件行业过去那种“建一个网站，用户涌进来，坐享 80% 高毛利”的美好幻梦已经终结；当前的 AI 产业正在演变出一种更接近“铁路或石油开采”的重资产商业模式。虽然现金源源不断地从超大规模云厂商流向英伟达（Nvidia）、电力公司以及数据中心建设承包商，但全市场都在追问：从投入到更高的收入和净利润，究竟有怎样的确切转化周期？如果只是无休止地消耗现金，盈利能力最终会被拖垮。 --- ### [03:10](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=190s) 谷歌：最完整的全栈生态与搜索商业化大考 在四大巨头中，谷歌拥有公认整合度最高的 AI 全栈实力：从 Google DeepMind 的尖端基础模型训练，到底层自研 TPU 定制芯片，再到庞大的消费端产品分发渠道。谷歌可以将 AI 功能塞入整个产品矩阵——Google Workspace、搜索、YouTube、Android 和 Google Cloud。按理说，这套飞轮应该能够飞速运转。 然而，投资者最为关心的核心命题是：AI 是否会过快地破坏搜索的单位经济模型（unit economics）？大语言模型、Gemini 以及 AI 搜索概览（AI Overviews）带来的商业化变现速度，能否足以弥补传统搜索广告可能出现的下滑？市场正紧盯这些新形态的 AI 搜索究竟是在扩大整体搜索频次、提升广告 ROI，还是在侵蚀谷歌原有的高利润商业模型。 --- ### [04:26](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=266s) 微软：从 6250 亿 RPO 到企服 AI 变现的“晴雨表” 微软同样是交出亮眼成绩单的选手。微软总营收同比增幅达 17%，包含 Azure、M365 和部分 LinkedIn 业务的广泛“微软云”部门增速为 26%，而单独拎出来的核心 Azure 云业务增速更高达 39%。Azure 正成为推动巨额 CapEx 的最强引擎，目前其资本支出年化运行率约在 1500 亿美元。 微软上季度财报中最令人瞠目的数字是“剩余履约义务（RPO, Remaining Performance Obligations）”达到了创纪录的 6250 亿美元，同比暴增 110%。其中约 45% 来自 OpenAI 的算力承诺，其余则来自签署了长期算力大合同的众多企业合作伙伴。这笔庞大的待履约订单直接向市场证明：巨额资本开支并非微软的盲目自嗨，而是整个行业对算力有着确凿的高增长需求。此外，硬件折旧问题似乎也未如预期那样可怕，上一代主力显卡 H100 目前依然在顺利产生利润。 - 微软整体云业务增长 26%，其中 Azure 独立增速达 39%，CapEx 年化运行率约为 1500 亿美元。 - 上季度 RPO 达到 6250 亿美元（同比增 110%），其中 OpenAI 贡献了约 45%。 - 微软是观察 AI 渗透进真实商业社会的最清晰窗口，核心跟踪指标包括：Azure 增速、云业务毛利率（是否面临降价压缩）、Copilot 企业付费采纳率、M365 席位增长以及 GitHub Copilot 的持续势头。 ```mermaid flowchart LR accTitle: Microsoft Enterprise AI Monetization Funnel accDescr: Pipeline showing how CapEx flows into RPO commitments and downstream enterprise metrics. subgraph Inflow["资本投入与承诺"] CapEx["CapEx 约 1500 亿美元"] RPO["RPO 6250 亿美元<br>(OpenAI 占 45%)"] end subgraph CloudCore["云基础设施"] Azure["Azure 增速 39%"] CloudMargin["云业务毛利率监控"] end subgraph EnterpriseApps["企业应用与席位"] M365["M365 席位数增长"] Copilots["Copilot 增量付费与 GitHub Copilot"] end CapEx --> Azure RPO --> Azure Azure --> CloudMargin CloudMargin --> M365 M365 --> Copilots ``` --- ### [07:08](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=428s) 席位制 SaaS 的生死存亡与 GitHub Copilot 的“Teams 式”渠道碾压 讨论深入到了企业级软件的根本商业模式。目前各大科技公司出现了一定规模的裁员，而 M365 的席位增速将透露宏观经济与用工需求的真相。更长远的问题是：未来的“AI Agent（智能体）”究竟是需要单独购买一个账号席位，从而让总席位数激增；还是一个 Agent 将直接替代 20 名员工，导致 20 个人合用一个席位？如果席位数断崖式下滑，以席位收费（seat-based）为核心的传统 SaaS 模式将面临巨大冲击。 在代码助手赛道，GitHub Copilot 虽然最早入局并迅速冲上 5 亿美元 ARR，但在舆论声量上，近年来 Windsurf、Cognition、Cursor 以及 Anthropic 推出的 Claude Code 等初创产品往往更受极客追捧。然而，这恰如当年 Slack 与微软 Teams 的竞争：Slack 占据了所有的热度和极客口碑，但微软凭借极其强大的企业分发网络（GTM）与销售渠道，将 Teams 强行推入所有企业客户。即便 GitHub Copilot 不是当下“最酷炫、最性感”的产品，微软依然有能力依靠深厚的企业渠道壁垒继续扩大其市场份额。 --- ### [09:03](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=543s) 亚马逊：AWS 的 2000 亿豪赌与每年千亿的隐形广告印钞机 对于亚马逊而言，全市场的目光都聚焦在 AWS 的加速增长上，以此来验证其领跑科技七巨头的资本支出计划。亚马逊给出的指引是 2026 年 CapEx 将达到创纪录的 2000 亿美元，因此 AWS 的营收增速和利润率成为财报的最大看点。 第四季度亚马逊整体表现健康：净销售额增长 14%，AWS 云计算业务增速为 24%。与此同时，亚马逊内部还有一个极其庞大、却经常被外界低估的“隐形巨无霸”——广告业务。该季度广告业务同比增长 23%，单季体量达 213 亿美元。这意味着亚马逊几乎仅靠广告业务就能实现每年近 1000 亿美元的营收。这项高利润业务源源不断产生的大笔自由现金流，正是支撑亚马逊在数据中心和 AI 大基建上展开 2000 亿豪赌的最大底气（该季度总体营业利润达 250 亿美元）。 ### [09:48](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=588s) 亚马逊与 Meta 财报前瞻：算力扩张与变现闭环 关于亚马逊的现金流，其自由现金流虽然仍高达 112 亿美元，但由于在人工智能领域的资本支出（CapEx）大幅增加，该数字已出现明显收窄。然而，如果 AWS 业务能够重新加速增长，那么目前市场上看似激进的资本开支，就会被重新解读为“在需求彻底爆发前抢购稀缺算力储备”。在算力与 GPU 为王的时代，拥有充裕算力储备的巨头将占据绝对先机，届时外界会视其为高瞻远瞩的明智之举。目前华尔街对 AWS 收入的一致预期约为 367 亿美元，同比增速在 20% 左右的中位区间；但市场真正的期盼是增速能够以“3”字开头（即达到 30% 左右）。一旦 AWS 展现出强劲的再加速势头，势必将引爆投资者的乐观情绪。 ```mermaid flowchart TD accTitle: Meta AI 驱动广告闭环流程图 accDescr: 展示 Meta 如何将前沿 AI 模型快速 A/B 测试并部署于全系 App 广告投放系统，从而直接转化为数十亿美元利润且无扩散损耗。 subgraph MetaAI["Meta 内部研发与敏捷分发"] A["前沿模型突破 / 算力基础设施<br/>(推理/代理能力)"] --> B["广告匹配与推荐系统 A/B 测试"] B --> C["全系应用矩阵直接部署<br/>(Instagram / Facebook / Reels)"] end subgraph Revenue["实时变现转化"] C --> D["数十亿用户无感体验<br/>(精准推荐日常商品)"] D --> E["广告转化率与客单价提升<br/>(+200 亿美元级别潜在增量)"] end ``` 转到 Meta，其处于一个非常微妙且令人兴奋的节点。Meta 拥有极其扎实的第四季度表现，即便市场围绕顶级人才争夺战、实验室调整、偶有瑕疵的模型发布（例如 Behemoth 等）以及 Meta Vibes 短视频应用采用率等问题充斥着疑虑（FUD），但有一点毋庸置疑：AI 已经在切实为 Meta 创造巨额财富。 - **Reels 与用户规模**：Meta 全系应用（Family of Apps）每日活跃用户数（DAU）已达到惊人的 35.8 亿。 - **财务与广告指标**：上季度总营收同比增长 24%，达到近 600 亿美元；广告展示量（Impressions）增长 18%，单次广告均价上涨 6%。全系应用营业利润高达 308 亿美元，这让 Reality Labs 每季度约 60 亿美元的亏损显得微不足道——完全属于巨头有能力承担的未来平台押注。 - **CapEx 激增预期**：Meta 去年的资本支出为 722 亿美元，今年的指引区间在 1150 亿至 1350 亿美元之间（几乎翻倍）。两位主播甚至调侃扎克伯格可能会在电话会议上彻底放飞自我，把 CapEx 指引直接轰到 2500 亿美元，毕竟他“连太空白昼反射镜（mirrors in space）都想搞”，势要成为终极算力玩家。 Meta 相比企业级软件服务商拥有巨大的天然优势：企业软件公司在推出前沿模型后，需要面临极其漫长的销售周期、组织变革管理、安全与合规审查，才能逐步推动收入；而 Meta 只要在底层模型上取得突破，就能立即通过 A/B 测试把算法应用于广告竞价与推荐分发中。消费者甚至不会察觉算法升级，只觉得“刚刚想买车就刷到了汽车广告，想买衣服就看到了完美的衬衫”。这种点击率和转化率的微小提升，直接就能转化为高达 200 亿美元的额外利润。Meta 完全不存在所谓“技术扩散难题（diffusion problem）”，不需要费尽心思说服用户使用 AI，用户每一次滑动屏幕都在直接为 AI 变现。 --- ### [14:15](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=855s) 科技超级碗：财报季的 AI 焦虑与芯片热潮 当前整个科技巨头阵营都在试图回答同一个核心命题：能否在庞大的折旧成本赶上来之前，将高额的 AI 资本支出转化为独占的分发渠道、更高的用户留存率以及可量化的收入增长？这无疑是属于巨头们的“超级碗”，而《华尔街日报》等媒体也敏锐指出了重燃的“AI 焦虑（AI jitters）”。 此前受 OpenAI 相关融资与算力传闻影响，Oracle、CoreWeave 及软银等持股或深度绑定公司的股价曾出现盘后大幅震荡。虽然 CoreWeave 在隔夜盘后暴跌 4% 后，当天日内强势反弹 8.4%，但这依然暴露了投资者的脆弱神经——市场极度担忧巨额基础设施投资无法转化为预期的爆发式利润。Synovus Trust 投资组合分析师 Dan Morgan 形象地指出：“冰层非常薄，缰绳勒得极紧。”任何对 OpenAI 或 Anthropic 发展势头的质疑都会迅速引发抛售。然而，从 IBM、德州仪器以及英特尔等芯片与硬件企业稳健的财报来看，需求基本面并未受损，OpenAI 官方也回应称其业务“全速运转，内部士气极为高昂”。散户社区与投资人同样紧绷神经，甚至有推特网友调侃“今晚要是财报不及预期，孩子们可能要挨饿了”，而 Robinhood 首席执行官 Vlad Tenev 也即将做客节目。 在芯片领域，CNBC 知名主持人 Jim Cramer 惊呼“英特尔简直是一匹势不可挡的骏马（Intel is such a horse）”，并直言自己“完全找不到看空的理由”。英特尔股价当天再次大涨 10%，单周涨幅已达到惊人的 40%。这一现象源于市场正在全面消化全行业潜在的“通用 CPU 短缺”逻辑。主播回顾道，正如当年 ChatGPT 问世后 Nat Friedman 与 Daniel Gross 在 TechCrunch 上率先指出“英伟达可能极其重要”随后英伟达市值翻了两番一样，华尔街目前正在重新审视并重估 AI 供应链中被忽视的基础组件。 --- ### [17:33](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1053s) 互动思想实验：红按钮与蓝按钮的生存博弈 节目的后半段切入了一个在社交网络上掀起巨浪的博弈论思想实验。这一话题最初由著名博客 *Wait But Why* 的作者 Tim Urban 在 4 月 24 日发布推文（获得了超过 2400 万次浏览），随后超级网红 MrBeast 在四天后跟进，同样吸引了上千万次观看。主播在直播聊天室中发起了实时投票，邀请观众共同参与这一残酷的人性测试。 | 选项 / 情形 | 结果判定条件 | 最终结局 | |:--- |:--- |:--- | | **蓝色按钮占绝对多数** | 超过 50% 的人按下蓝按钮 | **全员存活**（所有人安然无恙） | | **蓝色按钮未能过半** | 蓝按钮得票率 $\le$ 50% | **仅红按钮者存活**（按蓝按钮的人全部死亡） | Tim Urban 提出的思想实验规则极为严苛：全世界每个人都必须在不公开的前提下进行一次秘密投票，选择按下红按钮或蓝按钮。如果按蓝按钮的人数超过 50%，则全人类存活；但如果按蓝按钮的人数不足或刚好等于 50%，那么所有按下蓝按钮的人将全部被处死，只有按下红按钮的人能够活下来。 主播在分析中剖析了两种截然不同的人性逻辑： 1. **直觉型与利他型（Normie Answer）**：直觉会倾向于选蓝按钮，因为只要大家互相信任并抱有善意，就可以实现全人类平安脱险的最优解。 2. **超级理性博弈（Hyper-rational Choice）**：如果站在个体生存的角度极限推演，按下红按钮才是绝对的安全解。因为一旦你按下红按钮，无论大众选蓝过半还是不过半，你个人都必定存活——如果人人都陷入这种“冷酷的绝对理性”，全员选择按红按钮，虽然蓝按钮得票率为零，但按红按钮的全体人类同样能幸存。然而这种选择存在着极高的信任崩塌风险与道德挣扎，主播坦言自己一开始也是坚定支持按蓝按钮，但在博弈机制面前不由陷入了深思。 ### [19:53](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1193s) 红蓝按钮博弈：自保理性与群体博弈的权衡 节目围绕社交媒体上热议的“红蓝按钮思想实验”展开了激烈的理性与博弈论探讨。讨论者剖析了从常人直觉到博弈思考的心路历程：如果从纯粹的自保理性出发选择红按钮，自己就能获得 100% 的生存保障；然而，一旦有人选择了蓝按钮，而红方没有达到全员一致的特定门槛，那些选蓝的人就必死无疑。因此，选择红按钮在某种意义上相当于“对选蓝的人投了反对票（死刑票）”。 从拯救全员的角度来看，二者存在完全不同的协调门槛：如果希望所有人都活下来且选红，就必须确保 100% 的人全部选红，不允许出现任何偏差；但如果走向选蓝的路径，只需要达到 51% 的多数门槛，全员就能安然无恙。这意味着选蓝只需要说服少得多的人，对群体协作的容错率反而更高。 ```mermaid flowchart TD accTitle: Red vs Blue Button Coordination accDescr: Decision flow comparing the coordination requirements of choosing Red Button versus Blue Button. subgraph RedChoice ["选择红按钮路径"] R1["个人生存率 100%"] --> R2{"全员存活条件"} R2 -->|"需要 100% 全体选红"| R3["极低容错率：一人出错即有死伤"] end subgraph BlueChoice ["选择蓝按钮路径"] B1["个人承担死亡风险"] --> B2{"全员存活条件"} B2 -->|"仅需 >50% 多数选蓝"| B3["较高容错率：全员生还门槛更低"] end ``` ### [20:26](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1226s) 直播间民调惨淡与海量真实样本的对比 主持人迅速在直播间发起实时投票，让观众输入“1”投红、“2”投蓝。然而，TBPN 的直播弹幕呈现出压倒性偏向红按钮的态势。这一结果与网络上著名的几项大型民调形成了鲜明反差： - 在 Tim Urban 的民调中，有 58% 的人选择了蓝按钮，蓝方胜出，无人死亡。 - 在顶流博主 MrBeast 发起的数百万人规模投票中，55.7% 的人投给了蓝按钮，蓝方同样过半胜出，成功免于灾难。 面对直播间仅有约 33% 选蓝、即将“全员覆灭（cooked）”的残酷局面，嘉宾们开玩笑称可能有敌对国家行为体（nation-state）在操纵投票。随后，主持人援引了知名科技评论人 Mike Solana 的公开表态。Solana 坦言自己最初出于本能选了蓝，但细想之后立刻转向了红，直言从理性角度根本找不到选蓝的逻辑，虽然能理解选蓝者的善良愿景，但认定他们“彻底算错了账”。演播室里的泰勒（Tyler）则坚定捍卫蓝方立场，他认为现实世界永远存在方差（variance），总会有人因误操作或理解偏差投错；如果人人都是 100% 绝对理性，确实大家选红最稳妥，但只要理性率稍有下滑（例如 90%），选蓝反而是唯一能救下所有人的方案。 ### [22:17](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1337s) 框架效应的博弈：“终极死亡赌局”对决“终极谋杀赌局” 讨论进一步聚焦于人们如何通过不同的概念框架（Framing）来解构这两个按钮。推特用户 Kermit 提出了直观的视觉对比： 1. **红按钮**：标签为“什么都不会发生（Nothing happens）”。 2. **蓝按钮**：标签为“终极死亡赌局（Ultimate death gamble）”——按下即代表你与所有按蓝者都将丧命，除非超过 50% 的人类同样按下它。在此视角下，按红自保是显而易见的常识。 随后，另一位网友 Prerat 则给出了完全针锋相对的翻转版本： 1. **蓝按钮**：若人人按蓝，什么都不会发生，人人平安。 2. **红按钮**：实为“终极谋杀赌局（Ultimate murder gamble）”——一旦你选择按红自保，除非红方占到多数，否则你将亲手引爆选蓝者的死亡判决。 这种心理博弈演变成了两难质问：你宁愿冒着生命危险参与死亡赌局，还是愿意双手沾满鲜血、把他人推向深渊？甚至有人调侃，选择按蓝可能单纯是因为某些人寻求刺激（thrill-seeking）。 ### [23:55](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1435s) 道德审判、电车难题与社媒投票的真实有效性 网络舆论随之升级为针对人性的道德审判。网友 Simone 讽刺红按钮支持者属于“在灾难中会扔下妻儿独自逃生的人，只因孩子跑得不够快”；而反方则反唇相讥，指责蓝按钮支持者是将家人的生死无谓地押在不可靠的群体协调上。甚至有激进言论将选择红按钮直接定性为参与“群体灭绝（mass genocide）”。 网友 Self Maxer 还绘制了一幅改良版“电车难题”插画：红队成员安稳地站在铁轨旁看热闹，而蓝队成员则站在铁轨上拼命拉动操纵杆试图拯救全员。主持人笑称 TBPN 的观众显然是冷眼旁观的红队，而泰勒这种选蓝者就像明知有危险还往铁轨上站的憨人。 针对 MrBeast 等社媒大规模调查的有效性，现场展开了辩论： - **正方观点**：在成百上千万人参与的海量样本下，每一次蓝方都稳定超过 50%，充分说明在统计学意义上大众的群体协调能够成功，蓝方就是最优解。 - **反方观点**：社媒信息流完全没有真实利益纠葛（no real stakes），用户刷到后凭直觉几秒钟就点完了；如果是真正关乎生死的抉择，深度思考后的社会大众很可能会大举倒向红按钮。 但主持人也指出，选蓝本质上是一种“重建对人性信仰”的英雄主义叙事，如同现实世界中的国家牺牲、社区互助，或是电影《绝世天劫》（*Armageddon*）中宇航员飞往小行星引爆炸弹的悲壮奉献，是人类为了更大善意承担牺牲的缩影。 ### [26:55](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1615s) David Shor 权威全国民调出炉与尾声 节目临近尾声时，主持人引用了数据科学家 David Shor 针对该问题的严谨全国代表性民调数据，为红蓝按钮的争议给出了实证结论： - 在 14,000 名具有全美代表性的样本中，蓝按钮以接近 **3 比 1 的压倒性优势**获胜。 - 该研究交叉对比了 204 个常用心理测量学（psychometrics）问题。 - 预测按钮选择的最关键人格特质是**“我讲真话（I tell the truth）”**：强烈认同自己讲真话的人压倒性地倾向于按下蓝按钮；而不讲真话的人虽然整体仍偏向选蓝，但选择红按钮的比例明显上升。 节目最后，主持人感叹这是一场极为发人深省的社会心理学实验，并在道别前呼吁听众在 Apple Podcasts 与 Spotify 上给予五星好评，同时订阅 TBPN.com 官方通讯。 ## Notable quotes > [00:02](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=2s) 四杀（quad kill） > [01:29](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=89s) 建一个网站，用户涌进来，坐享 80% 高毛利 > [09:48](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=588s) 在需求彻底爆发前抢购稀缺算力储备 > [09:48](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=588s) 连太空白昼反射镜（mirrors in space）都想搞 > [09:48](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=588s) 刚刚想买车就刷到了汽车广告，想买衣服就看到了完美的衬衫 > [14:15](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=855s) 冰层非常薄，缰绳勒得极紧。 — Dan Morgan > [14:15](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=855s) 全速运转，内部士气极为高昂 — OpenAI > [14:15](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=855s) 今晚要是财报不及预期，孩子们可能要挨饿了 > [14:15](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=855s) 英特尔简直是一匹势不可挡的骏马（Intel is such a horse） — Jim Cramer > [22:17](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1337s) 终极死亡赌局（Ultimate death gamble） — Kermit > [22:17](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1337s) 终极谋杀赌局（Ultimate murder gamble） — Prerat > [23:55](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1435s) 在灾难中会扔下妻儿独自逃生的人，只因孩子跑得不够快 — Simone > [26:55](https://www.youtube.com/watch?v=IJCQ5YU8_UM&t=1615s) 我讲真话（I tell the truth） — David Shor ## Mentioned TBPN, John Coogan, Jordi Hays, Jerb Gavin, 加拿大央行, 美联储, Jerome Powell, Google, Amazon, Meta, Microsoft, SemiAnalysis, YC Yield Chad, Nvidia, Google DeepMind, Gemini, AI Overviews, OpenAI, Slack, Microsoft Teams, GitHub Copilot, Windsurf, Cognition, Cursor, Anthropic, Claude Code, AWS, Mark Zuckerberg, Reality Labs, Oracle, CoreWeave, 软银, Dan Morgan, Synovus Trust, IBM, Texas Instruments, Intel, Jim Cramer, Nat Friedman, Daniel Gross, Robinhood, Vlad Tenev, Tim Urban, Wait But Why, MrBeast, Mike Solana, David Shor ## 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=IJCQ5YU8_UM)

[0:02] 四杀（quad kill） [1:29] 建一个网站，用户涌进来，坐享 80% 高毛利 [9:48] 在需求彻底爆发前抢购稀缺算力储备 [9:48] 连太空白昼反射镜（mirrors in space）都想搞 [9:48] 刚刚想买车就刷到了汽车广告，想买衣服就看到了完美的衬衫 [14:15] Dan Morgan: 冰层非常薄，缰绳勒得极紧。 [14:15] OpenAI: 全速运转，内部士气极为高昂 [14:15] 今晚要是财报不及预期，孩子们可能要挨饿了 [14:15] Jim Cramer: 英特尔简直是一匹势不可挡的骏马（Intel is such a horse） [22:17] Kermit: 终极死亡赌局（Ultimate death gamble） [22:17] Prerat: 终极谋杀赌局（Ultimate murder gamble） [23:55] Simone: 在灾难中会扔下妻儿独自逃生的人，只因孩子跑得不够快 [26:55] David Shor: 我讲真话（I tell the truth）

[1:29] We're going into an entirely different era... as opposed to something like you build a website, people just show up, and it's 80% margin, which was the dream of the previous software era. [10:39] If there's a breakthrough and they have a model that is placing ads more effectively, they can quickly A/B test that, run that across the entire family of apps, and you don't even notice it as a consumer... and just a little bit better means like another $20 billion dollars in profit. [14:15] The ice is thin. The leash is very tight... Any evidence that would come out that would add doubt about OpenAI, Anthropic, or any of these companies is obviously going to create a sell-off.

[0:02] What a day in financial markets and [0:04] technology markets. There is a ton of [0:07] news today. Gerb Gavin summed it up [0:11] well. [0:12] A huge day tomorrow. He's posting this [0:14] yesterday. 10:00 a.m. Canada interest [0:16] rate decision. 2:00 p.m. This is Eastern [0:18] time. USA interest rate decision. That [0:21] is in Fed held rates constant. So, sort [0:26] of a nothing burger, I guess. Not the [0:29] best news. I think some people were [0:30] hoping for a cut, but everyone sort of [0:32] expected this. It met It met [0:34] expectations, but the news is that just [0:36] 5 minutes ago, the Federal Reserve has [0:38] kept interest rates on hold at a range [0:40] of 3.5 [0:42] to 3.75% [0:44] at Wednesday's meeting. And Jerome [0:46] Powell is giving a speech in just 30 [0:47] minutes at 2:30 Eastern. Then, at 4:00 [0:51] p.m. Google earnings, Amazon earnings, [0:53] Meta earnings, and Microsoft earnings. [0:56] It is a massive day. [0:57] >> deal. No big deal. They only represent [0:59] just under 20% of the total market cap [1:03] of the S&P 500. [1:05] >> Let's go. [1:05] >> All reporting within [1:07] >> But people are optimistic. Semi analysis [1:08] put out a note this morning according to [1:10] YC Yield Chad, expecting hyperscaler [1:13] capex to be revised upwards and beat [1:15] street expectations as hyperscaler cloud [1:18] revenue is accelerating. And they are [1:19] seeing positive ROI on cloud [1:22] investments. I can sort of go through my [1:25] earnings preview. It's a tech earnings [1:27] quad kill today. [1:29] And question is just how is the AI build [1:32] out going? Obviously, the capex numbers [1:35] were huge. We saw the first time ever [1:37] we've seen a $200 billion number from [1:39] Amazon last quarter. But everyone's up [1:43] in the Is there even a word for [1:46] triple? It's like what? 11 digits or [1:49] something that 100 billion. Everyone's [1:51] Everyone's spending at least 100 billion [1:52] these days. At least if you're in the [1:54] mag seven. But financial performance has [1:56] actually been strong even in sort of [1:58] legacy areas such as search Google [2:02] search is growing e-commerce sales [2:04] Amazon core business is growing [2:07] enterprise software seats Microsoft 365 [2:10] we're going to have more news on all of [2:11] that even though all of those businesses [2:13] they're working they're chugging along [2:15] the big question is around durable [2:17] revenue tied to AI infrastructure [2:19] because you have this matching problem [2:21] you spend a bunch of money there's [2:22] depreciation when do you actually get [2:24] the cash flow back? How durable is this [2:26] revenue? What are the moats around this [2:27] revenue? What do the growth rates and [2:28] margins look like in this new era of [2:31] something that looks maybe a little bit [2:32] more like a railroad business and oil [2:35] business as opposed to something [2:38] like you build a website people just [2:40] show up and it's 80% margin which was [2:43] the dream of the previous software era [2:45] we're going into an entirely different [2:47] era but it is very exciting and we're [2:49] getting a lot of data today. So everyone [2:51] has seen cash flow from these [2:54] hyperscalers to Nvidia to power [2:57] companies to data center builders [2:59] but everyone's wondering what's the [3:00] exact conversion cycle to higher [3:02] revenues and higher profits. What's the [3:04] what's the pathway there because you [3:05] don't want to just be drawing down on [3:06] cash endlessly eventually you stop [3:08] making money entirely. So [3:10] let's start with Google. [3:11] Google has the most fully integrated AI [3:15] stack arguably they have consumer [3:16] distribution they have model training [3:19] with DeepMind they have custom chips [3:21] with the TPU and a bunch of product [3:23] services where they can stuff AI [3:25] features. [3:26] >> Google workspace [3:27] >> Google workspace [3:28] >> underrated [3:28] >> search YouTube Android cloud they can [3:31] deploy solutions all over the place and [3:33] so the flywheel should be spinning very [3:35] very quickly. The key question that [3:38] investors are asking is does AI change [3:41] the unit economics of search too [3:43] quickly? Does AI do LLMs do [3:47] AI search overviews Gemini generally are [3:50] they able to monetize those results fast [3:53] enough to offset any potential declines [3:56] in search ad revenue? Uh and so, people [3:59] will be looking for how is search [4:01] monetizing? How is How are the new AI, [4:04] you know, disruptors uh monetizing? And [4:07] so, uh but there are tons of places to [4:09] pick up growth even if growth does slow [4:11] down in the core search business. Uh [4:14] tons of opportunity in cloud. Uh but the [4:16] question again is uh AI overviews and [4:19] Gemini, are they expanding search usage? [4:21] Are they increasing ad ROI? Or are they [4:24] compressing the model, the the financial [4:26] model? Microsoft is also coming off a [4:27] strong quarter. Honestly, everyone's [4:29] coming off strong quarters. Everyone's [4:30] doing very well. Revenue's up 17% at [4:32] Microsoft with cloud, which includes [4:34] Azure, M365, some LinkedIn stuff. [4:37] There's a It's a big bucket. Cloud [4:39] Microsoft cloud's growing at 26%, but if [4:41] you dive in and you double click on [4:43] Azure, Azure is growing at 39%. [4:46] And of course, Azure is a bigger lever [4:49] on CapEx, uh which is run rating around [4:51] 150 billion, not bad. Uh the biggest [4:54] number in the Microsoft earnings is RPO, [4:57] remaining performance obligations. Last [5:00] quarter it was listed at 625 billion, up [5:03] 110%. That was the eye-popping number of [5:06] the last earnings. About 45% of that is [5:08] coming from OpenAI, but they have lots [5:09] of other partners that have signed on [5:11] for really long compute contracts. And [5:14] that is starting to show up in the [5:15] financials saying, "Hey, we're spending [5:17] all this CapEx, but we have this RPO, [5:19] and we have these deals signed where [5:21] companies are the It's not just us that [5:24] we're forecasting some really high [5:25] growth here. The entire industry is [5:27] forecasting high growth." And so, we [5:29] have done deals to justify the CapEx [5:31] that we're spending right now even with [5:32] depreciation, which seems like a less of [5:34] an issue than people thought it was [5:36] since H100 still seem to be monetizing [5:38] just fine. But we can go into that. So, [5:40] Microsoft has the cleanest read on [5:42] enterprise AI monetization, which I [5:45] think is something people have been [5:46] really looking for, looking for numbers. [5:48] They know that you know, subscription [5:50] LLM chat apps monetize at a decent rate [5:53] that the margins that a lot of these [5:54] companies are are okay. The token APIs [5:56] are working, but what does it actually [5:58] mean to deploy AI into the American [6:01] economy, into everyday businesses, into [6:03] large-scale businesses? All of those [6:04] businesses are on Microsoft [6:06] overwhelmingly. And so, there are a ton [6:08] of data points that can help you [6:10] understand how AI is flowing through the [6:14] global economy, but also the American [6:15] economy. So, what are we talking about [6:17] specifically? Azure growth, how much [6:19] cloud hosting is going on, gross margins [6:21] for Microsoft cloud, that's very [6:23] important. Are you seeing cloud [6:24] compression? Because if you go to your [6:26] Microsoft cloud provider and you say, [6:27] "Hey, I'm going to this Teams thing. I'm [6:29] paying a lot of money for it. I actually [6:30] vibe coded something. You got to give me [6:32] a discount." That would show up in cloud [6:34] margins. Will it show up? I don't think [6:37] it will, but we'll see. I think it's [6:39] going to be fine. [6:40] Copilot adoption. [6:41] And our poop, how much are they actually [6:44] rolling these out? Are they actually [6:45] getting incremental spend? Are companies [6:47] willing to send more of their [6:49] hard-earned dollars to Microsoft for [6:52] better AI services, better features, [6:54] Copilots? Then M365 seat growth, that's [6:57] a really important one. Are people [6:59] adding more seats? Are they hiring? Like [7:01] we've seen some layoffs in big tech, but [7:03] how is the overall economy doing? How [7:05] many more seats are are being rolled [7:07] out? [7:08] Again, with the question of like, does [7:09] your AI agent need a seat? And so, you [7:11] have more seats or does your AI agent [7:13] replace 20 seats? And then you only have [7:14] one seat. And you have a 20 agents, [7:16] they're all logging in through the same [7:18] M365 seat. These are like more long-term [7:20] questions, but we're getting an early [7:22] read today. GitHub Copilot momentum, [7:24] this will all paint a picture of what is [7:26] happening with AI adoption in [7:28] enterprises broadly. Interestingly, [7:30] Copilots at Microsoft, it hasn't been [7:32] the most hyped product. GitHub Copilot [7:35] early, very, very early to the party. [7:38] Huge run rate. [7:39] You know, rocketed up to 500 million ARR [7:42] very very quickly. But the the horse [7:44] race has always been you know, Windsurf [7:47] Cognition and Cursor and and Codex and [7:50] Claude Code and like the battle has been [7:53] the the the the startups for the most [7:56] part. Gemini's been in there. GitHub [7:58] Copilot has felt like it hasn't been [8:00] dominating the narrative, but we're [8:02] going to find out is still growing [8:03] because the market is really really big. [8:05] It's possible that everything's growing [8:07] even if there's you know, a horse race [8:08] back and forth between the leading labs. [8:10] There will be a bunch of other questions [8:12] answered around, you know, [8:14] how nuanced is the diffusion adoption [8:17] question. So, [8:19] Microsoft has an incredible go-to-market [8:21] team, an incredible go-to-market motion, [8:22] enterprise sales motion. And so, [8:25] is there an advantage that they have [8:27] that they can press GitHub Copilot even [8:29] if it's not the sexiest product next to [8:32] whatever's hot this week, can they get [8:34] that into people's [8:35] >> Teams, right? [8:36] >> Same thing. Yeah. Yeah, Slack was [8:38] definitely like the hot one, the hyped [8:40] one, and Teams wound up doing very well. [8:43] And so, we'll get us we'll get a [8:44] stronger read on what's going on there. [8:46] And then of course that M365 seat growth [8:49] will tell us a lot about what does the [8:51] future look like for seat-based [8:53] enterprise software, the SaaS model [8:55] broadly, because if all of a sudden [8:57] that's falling off a cliff, well, it [8:58] probably doesn't look good for other [9:00] seat-based SaaS companies. For Amazon, [9:03] everyone wants to see strong AWS [9:05] acceleration to justify the Mag 7 [9:08] topping CapEx numbers. They guided at [9:10] 200 billion in CapEx in 2026. So, expect [9:14] a lot of focus on AWS revenue growth and [9:16] margins. Q4 was healthy though. Net [9:18] sales up 14% for Amazon. AWS growing at [9:21] 24%. [9:23] The sneakily huge ads business over at [9:26] Amazon grew 23% to $21.3 [9:30] billion. [9:32] They're almost making a hundred billion [9:35] dollars a year just on ads. That's [9:37] remarkable. And that's a lot of cash [9:39] flow to fund CapEx and other, you know, [9:41] AI initiatives. And so, operating income [9:44] overall was 25 billion for last quarter, [9:47] and they generated free free cash flow [9:48] of 11.2 billion. So, still [9:51] huge cash flows. They're obviously [9:52] drawing down on those and that 11.2 [9:54] billion number was down [9:57] because of increased AI spending. So, if [10:00] AWS accelerates, all the CapEx looks [10:03] like buying scarce capacity ahead of [10:05] demand. Like, they will be GPU rich or [10:08] just compute rich generally at a time [10:11] when it's good to be GPU rich and [10:13] compute rich, and they will look like [10:14] geniuses. So, people are hoping for [10:17] strong AWS acceleration. Consensus for [10:20] AWS revenue is around 36.7 billion with [10:24] growth in the mid-20% range. But the [10:27] market's really hoping that it starts [10:28] with a three. Everyone's hoping for 30% [10:32] something like that at least this year [10:34] at some point. You know, re-acceleration [10:36] would be a treat for the market. [10:39] Lastly, Meta. Meta's in an interesting [10:40] spot. Super clean Q4. Nothing really to [10:44] prove today despite all the flood, all [10:47] the fear, uncertainty, and doubt around [10:49] the talent wars, new team members, the [10:52] new lab, the sometimes clunky model [10:55] releases. get Behemoth out. What what, [10:58] you know, different adoption of Meta [10:59] vibes, the video app. Like, there [11:01] there's been all these questions about [11:02] the AI strategy. All of that needs to be [11:04] put to the side in the face of just AI [11:07] is already working for Meta. [11:09] >> Let's get an update on Reels. [11:11] >> Uh yeah, they're it I mean, they're [11:13] they're they're they're absolutely [11:14] crushing it. So, 3.58 billion DAUs, [11:17] daily active users, and they grew the ad [11:19] business a ton. And so, [11:22] the revenue overall grew 24% to almost [11:25] 60 billion last quarter. Ad impressions [11:28] rose 18%, average price per ad rose 6%, [11:31] family of apps operating income was 30.8 [11:35] billion which makes losing 6 billion at [11:38] Reality Labs like quaint. It's just like [11:40] who who cares? It's totally worth taking [11:42] a side bet on the future of devices and [11:45] maybe you get a platform out of that. [11:47] Makes a ton of sense. And so capex was [11:49] 72.2 billion last year. The guide this [11:52] year is somewhere between 115 135. I [11:56] think we might see that tighten up today [11:58] because it's a little bit wider than [11:59] some of the other hyperscalers that are [12:01] targeting, but clearly [12:03] a near doubling almost doubling of [12:05] capex. What are you laughing about? [12:06] >> There's one possibility where he goes, [12:09] you know, he blows it out. [12:10] >> 250. [12:11] >> We can't We can't We can't count that [12:13] out. We can't count that out. They came [12:15] out with a They came out with a solid [12:17] They They came out with a solid model. [12:19] >> Yeah. [12:19] >> He's putting mirrors in space. He wants [12:21] to be a player. [12:22] >> Yeah. Yeah. Yeah. I mean a lot of focus [12:24] is on the the the the new meta models, [12:27] but all of that it just financially at [12:30] least in the quarterly earnings. Like it [12:31] will just take a backseat to [12:33] what's going on with the AI in the ad [12:38] placement ad monetization funnel because [12:40] that's where Meta makes so much money. [12:42] And so the question is how much more [12:44] juice will AI bring to the ad business? [12:48] Expectations imply revenue growth of [12:50] around 31%. And the question's always [12:53] the same. Is there enough incremental [12:55] growth to justify the capex? An AI [12:58] advancement delivers better ad [13:00] performance basically immediately and [13:02] this is the good news for Meta. So [13:04] improvements in AI move the needle at [13:06] Meta incredibly quickly. Like a new [13:08] model can come out and there can be some [13:10] breakthrough in like thinking, [13:11] reasoning, coding agents. And if you're [13:13] in the enterprise software world, it can [13:15] take time to do a deal, roll it out to [13:18] different developers, different [13:19] organizations, change management, [13:21] understand the the the the guardrails, [13:24] the security, how does this actually [13:27] flow through an enterprise to drive [13:29] incremental demand? Not so at meta. If [13:32] there's a breakthrough and they have a [13:33] model that is placing ads more [13:35] effectively, they can quickly AB test [13:37] that, run that across the entire family [13:40] of apps and you don't even notice it as [13:42] a consumer. You're just like, oh, like I [13:44] was shopping for a car and I saw an ad [13:46] for a car. I was shopping for a shirt [13:47] and I saw an ad for the perfect shirt. [13:49] And the ads just got a little bit better [13:51] and just a little bit better means like [13:53] another 20 billion dollars in profit. [13:55] So, [13:56] it can be [13:58] So, they just don't have a diffusion [14:00] question. They don't they they don't [14:01] have a diffusion problem. There's no, [14:04] oh, we have this amazing model. If only [14:06] we could get people that open Instagram [14:08] to use it. Like there's nothing there [14:10] because they use it when they scroll and [14:12] they see an ad. And so that's an [14:13] incredibly fortunate place to be in. So, [14:15] every company is trying to answer the [14:17] same question. [14:18] Can you turn AI capex into proprietary [14:21] distribution, higher customer retention, [14:24] measurable revenue growth before the [14:26] depreciation catches up. It's the Super [14:28] Bowl for big tech, so get your popcorn [14:30] ready. Everyone's Everyone's excited. [14:33] There are There are AI worries, AI [14:35] jitters in the Wall Street Journal. AI [14:38] worries have returned to Wall Street now [14:40] come earnings. And so everyone is going [14:43] back and forth. On the open AI news, [14:45] shares of Oracle, Coreweave and SoftBank [14:48] slumped, but Coreweave is up majorly [14:51] today, 8.4%. [14:53] They They dropped at least 4% after [14:56] hours yesterday, but then they're back [14:58] up. That revived investors' worries that [15:00] technology giants' massive investments [15:02] in artificial intelligence won't produce [15:04] the blockbuster profits many expect. The [15:07] report was particularly jarring coming [15:09] ahead of key earnings for major players, [15:11] Alphabet, Amazon, [15:13] Microsoft, Meta, all scheduled to report [15:15] Wednesday with Apple following on [15:17] Thursday. The ice is thin. The leash is [15:20] very tight, said Dan Morgan, portfolio [15:23] analyst at Synovus Trust. Any evidence [15:27] that would come out that would add doubt [15:30] about Open AI, Anthropic, or any of [15:31] these companies is obviously going to [15:33] create a sell-off. Morgan said he hadn't [15:35] adjusted his positions on Tuesday and [15:38] didn't think investor concern was [15:40] broad-based because companies including [15:42] IBM, Texas Instruments, and Intel had [15:45] reported strong earnings in recent days. [15:47] Instead, Tuesday's losses centered on [15:50] companies with big stakes in Open AI's [15:52] business. Open AI had forged close ties [15:53] with these companies to secure funding [15:55] and gain access to computing resources, [15:56] which are essential for training AI [15:58] systems and providing answers to queries [16:01] and executing user requests. And so, [16:04] Oracle and CoreWeave we talked about [16:06] fell. [16:07] Open AI has defended its financial [16:09] footing and said its leaders are aligned [16:11] on securing computing resources. The [16:13] business is firing on all cylinders and [16:14] the mood internally is incredibly [16:16] positive, the company said in a response [16:19] to the journal's article on Tuesday. [16:21] Colin Tadar says, "My children might not [16:24] eat tonight if this doesn't go well [16:26] because he is excited for all four of [16:29] these earnings on Robinhood." And we [16:30] have Vlad from Robinhood coming on the [16:31] show later today. [16:34] >> Cramer says, "Intel is such a horse." [16:38] >> I have no bear case. [16:39] >> I have no bear case. [16:41] >> And Intel is up another 10% today. [16:43] [laughter] [16:43] >> Yeah, up 10% today. [16:44] >> Wow. [16:46] I mean, we [16:47] >> on the week so far. [16:48] >> 40% on the week. [16:51] >> Is that good? [16:52] >> Wow. [16:53] >> Based on everything you've seen so far [16:54] in your 6-month career, Tyler, is that [16:57] good? [16:57] >> That seems incredible. [16:58] >> Looks pretty good to me so far. [16:59] >> Yeah, I mean, there is a [17:01] There's like the idea of a of a CPU [17:05] shortage generally is just being [17:08] digested by the market right now. I [17:10] think [17:11] it took [17:12] I mean, I remember Nat Friedman and [17:14] Daniel Gross going on TechCrunch and [17:15] talking about [17:17] after ChatGPT saying like yeah Nvidia [17:20] seems like it could be important. [17:23] Like quadrupled in value. [17:25] And this this tends to happen as people [17:28] digest the new reality of demand around [17:31] certain pieces of the AI supply chain. [17:33] >> And we have an interactive game set up. [17:35] Is that [17:36] >> We do. So there has been a viral [17:41] uh I don't know. What is it? A brain [17:43] teaser? [17:44] >> It's a poll. [17:45] >> It's a poll but it's more it's a thought [17:47] experiment. It's a thought experiment. [17:48] It's not it's not purely a poll. It's a [17:51] It's a it's a brain teaser. It's a [17:53] thought experiment. And did this [17:56] has this existed before? Did this start [17:59] with It did this didn't start with Mr. [18:01] Beast, right? [18:02] >> No, so It started with Tim Urban. Yeah, [18:03] Tim Urban [18:05] I I don't think Tim Urban like created [18:06] this either but I think his post was the [18:08] one that first went like really big. But [18:09] I I don't think this is new. I feel like [18:11] I've heard about this before at some [18:12] point. But Tim Urban was the first one [18:14] to make it like go viral this [18:16] >> Yeah. [18:17] >> cycle. [18:17] >> So he got 24 million views on April [18:20] 24th. And then Mr. Beast [18:23] came from behind four days later with 10 [18:26] million views on a very similar [18:28] question. Let's start with the original [18:31] Tim Urban question. And we will ask [18:34] everyone in the chat to participate and [18:37] vote for what you would pick in this [18:41] thought experiment. So here is the [18:43] question posed by Tim Urban of Wait But [18:45] Why. Everyone in the world has to take a [18:49] private vote [18:51] by pressing a red button or a blue [18:53] button. [18:54] If more than 50% press the blue button, [18:58] everyone survives. If more than 50% of [19:01] people press the blue button, everyone [19:02] survives. If less than 50% of people [19:05] press the blue button, only the people [19:07] who press the red button survived. Which [19:10] button would you press? What do you What [19:12] do you pressing? [19:13] >> So So the sort of like normie answer is [19:16] you immediately just go blue because it [19:19] feels like a lot of people would go [19:20] blue, but like the safe answer is you [19:22] just if you if you're hyper-rational [19:24] you'd be like obviously you hit red cuz [19:26] if everyone's just super rational, [19:28] everyone hits red and then everyone's [19:30] all good. You don't hit over 50%, right? [19:33] >> Yes. [19:34] >> Yeah, because like no matter what it So [19:35] if I press red, no matter what anyone [19:37] else picks, like I survive, right? [19:39] >> Yes. But there is a risk. [19:42] >> But But like you know, if you want [19:44] everyone to survive, then you need 100% [19:46] of people to pick red, [19:48] right? [19:48] >> Yes. Yeah. [19:49] >> But Okay. So Okay, so originally I was [19:51] blue. I think I was like complete [19:53] normie, right? And then I went to [19:55] Jordi's take. I was like, wait, no, that [19:57] doesn't make sense. Like even though I [19:59] mean, we'll see what happened in the [20:00] poll, right? But But you know, I was [20:02] like, okay, if I'm thinking about it [20:03] super rationally, I should just pick [20:04] red, right? Then I save myself 100% of [20:06] the time. [20:06] >> Yeah. [20:07] >> But then like if there's some people [20:08] that pick blue, then like they're going [20:10] to die for sure. [20:11] >> Yeah. [20:11] >> Like I'm basically voting against them. [20:13] >> Yeah, you're you're sort of [20:14] >> Like you know, if you want everyone to [20:16] live, if you want everyone to live and [20:18] you're and you're going for red, you [20:19] need 100% of people. But if you're going [20:21] for blue and you want everyone to live, [20:22] you only need 51% of people, right? [20:24] >> Yes. [20:24] >> So you need to convince way less people. [20:26] >> Quickly, let's [20:27] Let's tell the chat how to vote. [20:29] >> Yeah, so one is for red. [20:31] >> One is for red. [20:32] >> here and two is for blue. [20:33] >> Two is for blue. So type a a single [20:36] character, just a one if you want to [20:39] vote for red and just a two if you want [20:42] to vote for blue. Oh, do we have this [20:43] up? [20:44] Ooh, we are tilted towards red. This is [20:48] crazy. This is not what happened with [20:51] Tim Urban's and with [20:54] uh Mr. Beast because Tim Urban 58% [20:58] said blue. Blue won, no one died. And [21:01] with Mr. Beast, [21:04] 55.7% [21:06] went blue, no one died. It's not looking [21:08] good for the blue team here. [21:11] >> [laughter] [21:11] >> 33% of the chat is is going to get [21:14] cooked right now. [21:15] >> It's possible a nation-state is voting. [21:18] >> Yes, that would be very, very high risk. [21:20] Mike Solana says, "For what it's worth, [21:23] and this is embarrassing, but I'm going [21:24] to admit it. My instinct was blue, and I [21:26] pressed blue. Then I thought about it [21:29] for a moment, and the answer was clearly [21:31] red. I can't make a rational case for [21:33] blue, but understand where blues are [21:36] coming from." Parentheses, "They are [21:38] wrong." So, your final answer, you you [21:41] you flipped. You're you're blue. [21:42] >> blue, yeah. [21:43] >> You're still going blue. [21:44] >> Because I I think I you have to assume [21:45] that there's going to be some variance. [21:47] But some There's some people will just [21:48] pick the wrong one. [21:50] >> Yes. [21:50] >> Okay. [21:51] >> And so [21:51] >> The the the correct one is red. If [21:53] everyone is, [21:54] you know, 100% logical, everyone should [21:55] be picking red. [21:56] >> Mm uh [21:58] yes. If if 100% of people are rational, [22:01] they will all pick red, and no one will [22:02] die. [22:02] >> Correct. [22:03] >> But if 90% of people [22:05] >> If 90% of people are doing it. [22:06] >> Yes. If 90% of people are rational [22:08] >> Yes. [22:08] >> then everyone should just be picking [22:09] blue because you want everyone to live. [22:11] >> Yes. Yes. And then and then the people [22:13] that make the mistake, they go for the [22:15] red, they are saved anyway. [22:17] >> Right. [22:17] >> But uh there are there are some folks [22:19] who are who are sharing different [22:22] different illustrations. Uh Kermit has [22:25] one here. He says, "If this was how the [22:27] buttons looked, what portion of humanity [22:30] would would press blue?" And he has two [22:33] images pulled up. One is the red button, [22:35] nothing happens. [22:37] And [laughter] the other one is the [22:38] ultimate death gamble. Press blue, you [22:41] enter the ultimate death gamble. You and [22:43] everyone who presses this button dies, [22:46] unless more than 50% of people choose to [22:49] press this button. [22:51] And so that makes it pretty clear. I [22:52] think this is the Mike Solana position [22:54] as well. Uh you got to press red in this [22:56] case because nothing happens. Um [23:00] but uh but that's not but that's not [23:02] entirely true. And so people have sort [23:03] of have sort of flipped this around [23:05] because uh Prerat here has the opposite, [23:08] which is instead of the ultimate death [23:10] gamble, it's the ultimate murder gamble. [23:11] So, blue, if everyone presses blue, [23:14] nothing happens. Red is the ultimate [23:16] murder gamble because you initiate the [23:18] ultimate murder gamble. Everyone who [23:20] pressed blue will die unless 50% of [23:23] people choose red. And so, it goes back [23:25] and forth, and you can see both of [23:27] those. Do you want the blood on your [23:28] hands from playing the ultimate murder [23:30] gamble, or do you want the risk of the [23:33] ultimate death gamble? I I think it's [23:35] underrated to just be uh a little [23:37] thrill-seeking and want to press blue [23:39] because, you know, you're you're bored, [23:40] and you want [23:41] >> can tell blue is the correct answer just [23:42] because of the massive pools where over [23:44] 50% of people picked blue. [23:46] >> That's true. That's true. [23:47] >> blue because you don't want to kill, [23:48] yeah, you know. [23:49] >> Once you're at Mr. Beast's scale, this [23:51] is like a real-world simulation. [23:53] >> It is. It is. Well, we're still we're [23:55] still red. [23:57] >> Uh [23:58] Cook. [23:58] >> Uh Simone says, "Red button pushers will [24:01] leave their wives and children to fend [24:03] for themselves in disastrous situations [24:06] because their kids can't run fast [24:08] enough." Mixalana says, "Blue button [24:09] pushers are literally leaving their [24:11] wives and children behind to fend for [24:13] themselves." Um are that they aren't [24:16] guaranteed to push red. Yes. Uh people [24:19] are going back and forth. Red cells [24:21] won't believe this, but some people make [24:23] a distinction between nothing and mass [24:25] genocide. So, they see pushing the red [24:28] button as effectively committing mass [24:31] genocide because you are potentially uh [24:33] condemning the blue button pushers to [24:36] death in this scenario. Self Maxer [24:38] shares a version of the uh [24:42] of the trolley problem where the red [24:45] team is not on the rails, but the blue [24:47] team is pushing the lever to save [24:50] themselves. Is this an accurate [24:52] description? Is this Is this a fair [24:55] Is this a fair uh contextualization [24:58] around this? It's obviously charitable [25:00] to the red team, which of course the [25:02] TVPN audience seems to be represented by [25:04] standing next to the trolley lines and [25:07] not standing on the on the rails. [25:10] But you would be there. You Tyler, you [25:12] would be a blue [25:13] button pusher pushing on the on the [25:15] rail. [25:16] >> that makes It's like I'm an idiot. [25:17] [laughter] [25:17] >> Seems a little risky in this scenario. [25:20] It does. When you have the option just [25:21] to stand there and watch the trolley go [25:23] by, you know, it's it's all about frame [25:25] of mind here. [25:26] >> No, but like you can tell that from the [25:28] polls, from the big polls, you can just [25:29] tell that blue is the correct option. [25:31] >> Potentially. Do you know that [25:32] beforehand? [25:33] >> No, you don't. But those people who [25:34] voted in it did not know beforehand. [25:37] >> Yeah. [25:37] >> And so if if 10 million or how many [25:39] hundreds of thousands people voted [25:41] >> Yeah. [25:41] >> I think you can just assume that that's [25:42] >> The question is like if this actually [25:44] was a real vote. [25:45] >> Yes. [25:46] >> Would people spend a little bit more [25:48] time thinking about the problem and [25:49] would that influence their [25:51] >> Yes. Yes. Yes. It is It is a very [25:53] >> Because the social media poll people are [25:55] scrolling, they see Mr. Beast ask a [25:57] question, they're just like curious. [25:59] There's no real stakes. And so they just [26:01] answer quickly. [26:02] >> But the blue is very much a It's a faith [26:05] in humanity restored moment, right? If [26:07] everyone works together, nobody dies at [26:09] all in this hypothetical. David Shore [26:11] says, "We asked this to a large [26:13] >> hits the death gamble button together [26:17] >> Yeah. [26:18] But there [26:18] >> Humanity wins. [26:19] >> There are there are analogies to that in [26:21] real life. [26:22] You know, national sacrifice, community [26:25] sacrifice, everyone pitching in for a [26:28] greater good to save those who didn't [26:32] even chime in to serve or save the [26:34] world. [26:36] The you know, you think about [26:37] Armageddon, the sacrifice of going to [26:39] the the [26:41] the the asteroid to destroy it. These [26:43] are heroic These are heroic stories. [26:46] What What is What's the chat saying? [26:49] >> They're not uh I don't think they're [26:51] they really [26:52] are big fans of my position. [26:53] >> No. [26:54] >> [laughter] [26:55] >> Well, [26:55] uh you know who is? Uh a large sample of [26:58] nationally representative Americans. [27:00] Blue won by a large margin, three to one [27:03] margin. [27:04] Um [27:04] and uh everyone voted [27:08] uh blue. Here, they asked a whole bunch [27:10] of people uh to close out the blue red [27:12] button discourse. They pulled 14,000 [27:15] people cross-tabbed survey responses by [27:17] 204 commonly used psychometric [27:20] questions. Uh the top four personality [27:24] questions most predictive of button [27:26] choice were displayed below. I tell the [27:28] truth. The blue button pushers strongly [27:30] agree. I tell the truth. People who [27:32] don't tell the truth were still more [27:33] likely to select blue, but little bit [27:35] more likely to select red on average. [27:38] Uh [27:38] a very very very interesting social [27:40] experiment. Have a great day, everyone. [27:42] Leave us five stars on Apple Podcasts [27:44] and Spotify. Sign up for our newsletter, [27:46] tvpn.com. [27:48] We'll see you tomorrow. Goodbye.