Benioff Rejects SaaS Demise as Salesforce Bets Millions on Anthropic

贝尼奥夫驳斥SaaS消亡论，Salesforce斥资数百万美元押注Anthropic Salesforce首席执行官Marc Benioff驳斥了人工智能将毁灭企业软件的担忧，认为成熟平台仍将是不可或缺的分发中枢，同时该公司每年投入3亿美元采购Anthropic模型以实现内部开发自动化。 Salesforce首席执行官Marc Benioff与主持人们深入探讨了重大的宏观经济与技术变革，话题始于特朗普与习近平峰会的外交利害关系及中美关系。对话探讨了“SaaSpocalypse”（SaaS末日论）以及企业软件如何适应AI智能体，同时分析了Apple、OpenAI和Anthropic等公司的战略转变。节目最后展望了即将到来的创纪录厄尔尼诺现象所带来的严峻气候前景，深入剖析了私募市场中未经授权的“暗箱SPV”（Dark SPVs），并深情悼念了YouTube前首席执行官Susan Wojcicki。 特朗普与习近平举行高层峰会，双方在中东安全问题上达成初步共识，中方承诺大规模采购美国大豆、原油、液化天然气及200架波音客机，意图通过深化经贸交织规避“修昔底德陷阱”。 (1:14) 围绕向中国出口先进半导体芯片的辩论，Salesforce 首席执行官 Marc Benioff 指出，中国 AI 实验室已学会在缺乏顶级芯片的情况下训练并优化出极具竞争力的模型，严格的出口管制可能已失去实际效果。 (20:09) 针对企业软件板块市值缩水超1800亿美元的“SaaSpocalypse”恐慌，Benioff 称这只是市场被 AI 叙事“催眠”，企业软件正迎来两极分化：低端轻量级工具可能被 AI 替代，而掌握底层核心数据和客户黏性的水平平台将成为 AI 模型落地的关键分发渠道。 (31:48) 有报道称 OpenAI 正考虑就两年前宣布的 ChatGPT 整合合作起诉苹果，矛盾集中在苹果严格限制访问入口、强制要求用户口播唤醒词导致使用率低迷，以及双方在隐私和 AI 硬件领域的竞争加剧。 (47:24)

During Episode 273 of the All-In Podcast, Salesforce CEO Marc Benioff and the hosts argue that while geopolitical tensions, public market panic over software, and climate disruptions are intensifying, technological integration and institutional adaptation remain powerful counterweights. U.S.-China Summit and Semiconductor Export Policy (1:14) The summit between President Donald Trump and President Xi Jinping opened with tentative bilateral commitments, including Chinese pledges to purchase 200 Boeing aircraft alongside U.S. soybeans, crude oil, and LNG. Polymarket odds cited on the show placed the likelihood of a Chinese invasion of Taiwan at 6% for 2026 and roughly 17% by the end of 2027. Chamath Palihapitiya framed the broader geopolitical reality as a natural trade between U.S. leadership in energy and AI against China's dominant position in rare earths and manufacturing. The panel debated whether restricting advanced semiconductor exports to China remains viable. Marc Benioff argued that export controls are largely ineffective because Chinese AI teams train competitive models using open-source tools and algorithmic efficiency without top-tier hardware like NVIDIA's Blackwell chips. Consequently, Benioff recommended selling chips freely to maximize commercial returns. Palihapitiya added that domestic manufacturing, including TSMC's operational Arizona facility, will reduce Taiwan's strategic bottleneck within 18 months. The SaaS Rerating and Enterprise AI Deployment Enterprise software providers have experienced steep public market drawdowns—Salesforce dropped 37%, ServiceNow 42%, and Workday 45%, wiping out over $180 billion in sector market capitalization—driven by speculation that autonomous AI agents will displace traditional software interfaces. Benioff dismissed the downturn as market overreaction, emphasizing that Salesforce projects over $46 billion in revenue and $16 billion in cash flow, which supported a $50 billion share buyback. Palihapitiya asserted that while low-end software for small businesses is vulnerable, complex enterprise monoliths will become the essential distribution channel for major AI labs seeking returns on trillions in infrastructure spending. To integrate AI reliably, Benioff highlighted Salesforce's $8 billion to $9 billion acquisition of Informatica to establish an enterprise data semantic layer, noting that Salesforce also spends roughly $300 million annually on Anthropic's API services to power internal automated coding workflows. Apple-OpenAI Friction and Evolving AI Architecture Reporting surfaced indicating OpenAI is considering legal action against Apple over their two-year-old Siri partnership. Friction centers on Apple's restrictive user interface, which requires users to explicitly invoke 'ChatGPT,' as well as Apple's privacy concerns and frustration over OpenAI's hardware initiatives involving former Apple designer Jony Ive. Benioff noted that the broader AI industry has aggressively pivoted away from novelty consumer tools and video generators toward autonomous coding agents, following Anthropic's early lead. The co-hosts debated the future of AI hardware paradigms. Jason Calacanis argued that large unified memory hardware will allow users to run frontier models locally to protect private communications. David Friedberg countered that modern workflows require persistent cloud models synced across devices. Evaluating real-time multi-modal startups like Mira Murati's Thinking Machines, Benioff rejected predictions of runaway enterprise token costs, forecasting the emergence of smart middleware that routes simple requests to low-cost edge models and reserves expensive frontier models for complex tasks. Super El Niño and Projected Agricultural Shocks (64:04) David Friedberg warned that an impending El Niño event is tracking toward sea surface temperature anomalies up to 4°C above historical baselines, rivaling the record 1877 event. The world's oceans currently store an excess 11 million terawatt-hours of heat energy—comparable to roughly 500 years of global human energy consumption of 25,000 terawatt-hours annually. Friedberg stated with 99% confidence that this thermal release will make the coming year the hottest in recorded history. The thermodynamic shift threatens widespread disruption: atmospheric rivers across the U.S. West and Gulf Coast, record heatwaves in the American interior, and potential monsoon failure across South Asia. A disrupted monsoon jeopardizes food supplies for 1.5 billion people and the livelihoods of 150 million agricultural workers in India, compounded by Persian Gulf fertilizer supply crunches. Friedberg noted that agricultural commodity markets are already aggressively pricing in these projected supply shocks. Private Venture Structures and Governance Scrutiny (71:40) The panel examined Anthropic's regulatory push to clamp down on unauthorized, multi-tiered Special Purpose Vehicles (SPVs) marketing secondary shares to retail investors. These vehicles frequently layer administrative fees, imposing up to 10% load-in charges, stacked carried interest, and share prices marked up 10% or more above official venture rounds. Palihapitiya and Benioff condemned these intermediary structures, arguing that late-stage private companies such as SpaceX, OpenAI, and Anthropic must pursue public listings to clean up capitalization tables and protect retail buyers. Palihapitiya warned that failure to transition to public transparency will trigger significant litigation once these companies eventually float shares and investors discover the extent of middleman fee dilution. The discussion leaves unresolved whether enterprise SaaS incumbents can capture the commercial gains of autonomous agents before frontier AI labs build independent distribution channels, alongside open questions on how global food markets will absorb the severe climate shocks forecast for the upcoming agricultural cycle.

本期节目聚焦中美高层峰会背后的双向经贸互锁与半导体技术地缘博弈，剖析企业级软件面对AI浪潮时的估值重构与真实护城河，并深入探讨了前沿AI巨头生态摩擦、边缘计算智能路由前景以及创纪录超级厄尔尼诺对全球粮食供应的严峻挑战。 中美峰会经贸互锁与技术地缘博弈 (1:14) 特朗普及多位美国商界领袖与中国国家主席习近平展开第七次面对面会谈，双方在霍尔木兹海峡自由通行与阻止伊朗获得核武器议题上达成一致。经贸方面，中方承诺采购200架波音客机，并扩大进口大豆、原油及液化天然气；Polymarket预测市场数据显示2026年台海冲突概率仅为6%，至2027年底升至约17%。David Friedberg指出，AI与自动化驱动的技术革命有望打破“修昔底德陷阱”下的零和争夺；Chamath Palihapitiya则强调，中美正在以稀土与制造业对比能源与前沿AI展开务实筹码交换。 针对半导体出口管制，Salesforce首席执行官Marc Benioff主张放宽芯片出口，指出中国AI实验室即便缺乏NVIDIA Blackwell等顶尖芯片也已训练出具备竞争力的模型，限制出口反而会削弱美企并为华为让渡市场空间。同时，随着台积电亚利桑那州工厂投产以及中国本土设备研发推进，Chamath预计未来18个月内本土晶圆制造将逼近先进制程，台湾作为全球供应链单一脆弱点的地缘溢价正在逐步减弱。 SaaS估值重估与企业级AI落地 公开市场担忧AI代理将取代传统软件交互界面，引发企业级SaaS板块剧烈回调，Salesforce市值下跌37%（蒸发约900亿美元），ServiceNow与Workday跌幅均逾40%，板块整体缩水超1800亿美元。Benioff将估值倍数压缩归因于市场对AI概念的过度“催眠”，强调Salesforce本财年预期营收超460亿美元并产生逾160亿美元现金流，且已执行500亿美元的股票回购。Chamath分析指出，AI虽将颠覆轻量级SMB软件，但万亿美元级的AI基建投入终将面临ROI拷问，基础模型必须借助拥有深厚工作流沉淀与客户信任的成熟SaaS平台实现变现。 为消除企业级模型“幻觉”并打通异构数据，Salesforce斥资80亿至90亿美元收购数据集成商Informatica以构建底层语义层，并通过Agentforce自主代理推进自动化销售与客服。同时，Salesforce推进Slack无界面化（Headless），接入MCP与AXL流式API标准，使应用逻辑能直接嵌入第三方大模型。Benioff透露，公司今年在Anthropic API上的支出预计达3亿美元，借助代码代理大幅提升了内部研发与交付效能。 巨头竞合摩擦、代码代理与边缘路由 OpenAI与苹果两年前达成的ChatGPT系统集成合作因体验不佳与信任缺失出现裂痕，传出OpenAI考虑采取法律行动。核心分歧在于苹果强制要求用户使用显式语音指令唤醒ChatGPT，且因数据隐私顾虑以及防范OpenAI涉足硬件（特别是拉拢Jony Ive）而未大力推广，导致预期的订阅增量严重落空。在行业路线上，Anthropic专注于自主编程代理的战略大获成功，迫使竞争对手削减视频生成等边缘项目转而全力追赶；Benioff透露软件研发正加速向语音结合脚踏板的对话式开发转型，Salesforce也正在Slack内原生构建AI代码生成能力。 在端侧与云端架构之争上，尽管苹果M5芯片及超大统一内存支持本地模型以保护用户隐私，但Friedberg指出多终端上下文无缝持久同步是现代工作流的刚需。针对Mira Murati创立的Thinking Machines所展示的每200毫秒捕获环境的实时多模态系统，未来企业Token消耗量可能增长千倍。Benioff对此反驳称，成本不会线性暴增，未来将崛起关键的中间件智能路由层，把简单任务分配给廉价端侧模型，仅在复杂推理时调用高价前沿模型。 超级厄尔尼诺预警与全球农业危机 (64:04) David Friedberg在科学环节展示了海洋温度异常预测，指出即将到来的厄尔尼诺周期中，海洋表面温度异常预计将高出历史基准4°C，强度恐刷新1877年的历史极值。海洋作为全球气候系统的热蓄电池，当前累积了约1100万太瓦时（TWh）的过剩能量，相当于人类全球年均能源消耗量（2.5万TWh）的近500倍。Friedberg以99%的置信度预测，随着这股热量释放至大气，未来一年将成为人类有记录以来最炎热的一年。 该气候现象将引发连锁性的全球物理与经济冲击。北美将遭受大气河流洪水与破纪录热穹顶侵袭；南美巴西、阿根廷及澳大利亚等粮食出口大国将面临严重干旱减产，冲击依赖粮食进口的发展中国家。最为危急的是南亚地区，极大概率遭遇的印度季风失效将威胁1.5亿农业工人与15亿人的粮食安全，叠加霍尔木兹海峡局势导致的氮肥供应紧张，将形成双重打击。大宗商品衍生品市场已开始提前定价这一严重的供给冲击。 二级市场SPV治理与企业长期责任 (71:40) 针对资本市场的结构性风险，Anthropic近期严厉打击了未经授权、多层嵌套的“暗池SPV”（Dark SPVs）。这类特殊目的工具专门向非合格投资者兜售未上市公司的二级市场股份，层层叠加高达10%的前端入场费、额外管理费和双重业绩提成，且售价常较官方轮次大幅溢价。Chamath对此表达强烈支持，痛批阻碍企业上市的金融掮客工具，并预言SpaceX、OpenAI等独角兽在最终IPO时将面临因费用盘剥引发的诉讼潮，呼吁晚期企业应当及早上市以澄清股权结构。 在企业社会责任方面，Benioff重申了Salesforce坚持27年的“1-1-1”慈善模式，即将1%的股权、1%的利润及1%的员工时间投入公益。该模式已累计发放超10亿美元资助，提供超1000万小时志愿服务，并免费支持5万多家非营利机构运作。最后，Benioff动情悼念了因癌症逝世的YouTube前首席执行官兼Salesforce前董事Susan Wojcicki，呼吁科技行业在追求商业与估值之外坚守人道关怀与社会担当。 从地缘博弈的互锁平衡到企业软件的真实壁垒，再到边缘路由架构与全球气候危机，商业世界正经历多维度的去伪存真。资本狂热与估值重构不可避免，但长期价值最终取决于底层生产力突破、严谨的商业化ROI，以及直面全球环境系统性风险的韧性与责任担当。

## Background 该视频由知名科技商业播客《All-In Podcast》于 2026 年 5 月 27 日发布在 YouTube 平台上。节目简介显示，本期邀请了 Salesforce 联合创始人兼 CEO Marc Benioff 作为特邀嘉宾，深入探讨特习北京峰会与中美经贸、AI 冲击下的“SaaS 末日”大洗牌、OpenAI 考虑起诉苹果违约、Thinking Machines 发布的实时多模态模型、2026 年历史级厄尔尼诺气候危机，以及 Anthropic 整顿一级市场“暗箱 SPV”等重磅焦点。 ## TL;DR 节目首先拆解了特习会背后的经贸大单与势力范围重构，Benioff 与主持团队一致认为技术封锁已难奏效，双向商业深度捆绑才是维系地缘和平与生产力增长的关键。针对二级市场上演的“SaaS 末日”与软件股抛售狂潮，Benioff 和 Chamath 驳斥了盲目的 AI 颠覆论，指出坐拥深厚企业数据语境、稳定留存率与充沛现金流的核心平台不可动摇，Salesforce 启动的 500 亿美元巨额回购便是明证。随后，嘉宾们探讨了 OpenAI 与苹果合作破裂背后的硬件博弈、实时多感官 AI 带来的交互革命，以及端云协同智能路由对企业 Token 成本的重塑。最后，Friedberg 深度预警了 2026 年打破极值的超级厄尔尼诺对全球粮食供应链的灾难性冲击，全场在痛斥多层套叠 SPV 乱象与 Benioff 对已故前 YouTube CEO Susan Wojcicki 的深情追思中结束。 ## Atmosphere & style 全片充斥着硅谷顶级圈子特有的谈笑风生与思想碰撞，在犀利的互损调侃与极高密度的硬核战略拆解之间自如切换。现场既有大国博弈、万亿美元 AI 资本回报率大考的冷峻现实主义剖析，也穿插着关于恐慌症、踩炭火励志营与流浪狗救助的鲜活插曲。整场对话节奏跌宕起伏，从老牌科技巨头面对估值腰斩时的从容老道，一路升华至尾声 Benioff 眼含热泪缅怀挚友的人性温度，给听众带来极为丰富而真实的临场感。 ## Bulletin - [01:12](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=72s) 特习北京峰会首日就霍尔木兹海峡通航与伊朗去核化达成共识，中方作出增购美豆、原油、LNG 及 200 架波音客机的实质承诺，但在台湾问题上中方严正警告若处理不当两国将迎头相撞。 - [31:41](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1901s) 面对二级市场对传统软件板块的估值血洗与“SaaS 末日”恐慌，Salesforce CEO Marc Benioff 宣布启动史上罕见的 500 亿美元股票回购，并透露今年将采购约 3 亿美元 Anthropic 算力全面重塑研发流水线。 - [47:24](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2844s) OpenAI 因不满苹果限制唤醒交互及侵入硬件领域，正考虑就 ChatGPT 整合协议起诉苹果；同时巨头研发焦点正从伴侣机器人紧急转向由 Anthropic 领跑的 Coding Agent。 - [01:03:01](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3781s) 科学角落预警 2026 年将迎来历史级超级厄尔尼诺，全球海洋过剩蓄热高达 1100 万太瓦时，叠加伊朗危机导致的化肥断供，印度及南亚地区面临严峻的季风失常与粮食赤字威胁。 - [01:11:40](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4300s) Anthropic 公开向缺乏监管的多层“暗箱 SPV”宣战并废除违规转售，投资人警告层层剥皮的前端费用与 Double Carry 将在上市清算时引爆海啸般诉讼。 ## Key takeaways - [02:52](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=172s) 中美领导人避免“修昔底德陷阱”的根本路径在于跳出静态存量争夺，借助 AI、自动化与生物科技做大全球蛋糕，进入科技驱动的充裕时代共享增量。 - [05:45](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=345s) 通往中美和平缓和的坚实纽带是双向的深度经济捆绑，美国应向中国高价值市场倾销重型装备、能源、农产品及核心芯片，以现实经贸交换明确双极势力范围。 - [09:58](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=598s) Salesforce 采取严格合规的“Salesforce on Alibaba”模式在华运营，而特斯拉则凭借独资车企地位成为深度掌控本地制造与 AI 数据的绝无仅有的特例。 - [19:50](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1190s) 对华芯片出口管制已失去实质意义，中国本土大模型正实现高速跟随，全面开放半导体销售不仅能依靠美企垄断挤压本土竞品，更能通过经贸利益绑定规避地缘冲突。 - [24:33](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1473s) 随着美台先进制程差距缩短至 1-2 纳米以及中美各自加快自主晶圆代工建设，台湾在全球经济与国家安全层面的不可替代性将在未来 18 个月内被实质性稀释。 - [31:41](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1901s) 二级市场对 SaaS 的抛售严重脱离财务基本面，低端单功能工具虽面临淘汰，但手握核心业务逻辑、庞大现金流与深厚客户信任的企业级巨头护城河依然稳固。 - [35:50](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2150s) 大模型在顶级企业的落地门槛极高，迫使 OpenAI 耗资 40 亿美元组建实施实体；当 3 万亿美元 AI 资本开支面临 ROI 审问时，大模型厂商最终仍需依赖传统 SaaS 渠道分销 Token。 - [40:04](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2404s) Salesforce 斥资 80 亿至 90 亿美元收购 Informatica，旨在构建统一数据与语义层，消除大模型幻觉，使 Agentforce 能够在全流程交互中无缝接管上下文。 - [41:36](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2496s) 企业级软件架构正在向无头（Headless）演进，Slack 与日常沟通记录沉淀了企业最真实的实时上下文，成为驱动大模型构建“企业世界模型”的核心资产。 - [44:47](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2687s) 企业采购正从昂贵的垂类现成软件转向灵活的底层水平平台，结合生成式 AI 智能体自行构建专属工作流，以极低成本重塑销售拓展与运营链路。 - [48:09](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2889s) AI 行业迎来赛道掉头，巨头紧急砍掉伴侣机器人等旁枝项目转向代码 Agent；工程师工作流演变为语音流输入与脚踏板结合，PM、开发与设计的职能边界彻底瓦解。 - [52:21](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3141s) Google 凭借 Gmail、Drive 等私域数据生态在个人助手战局中占据绝对护城河，苹果若要抗衡则必须在自研模型、收购成熟团队或贴牌第三方之间做出艰难抉择。 - [56:10](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3370s) Thinking Machines 等实时多模态系统将桌面、音频与摄像头以每 200 毫秒高频采样投喂给双协同模型，未来全时感知交互将促使 Token 消耗量激增上千倍。 - [57:54](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3474s) 基于语言文字概率预测的 LLM 并非通往 AGI 的唯一正解，具备多维生物级感知的多感官模型（Multi-sensory Models）将掀起下一波核心技术范式转移。 - [59:55](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3595s) 针对暴增的算力开销，未来架构将依赖智能路由层根据任务复杂度调度流量，通过边缘小模型与云端前沿大模型的无缝协同彻底打破算力垄断与高昂成本。 - [01:03:01](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3781s) 2026 年超级厄尔尼诺释放高达人类 500 年用电量的海洋过剩蓄热，直接引发北美大气河流、全球极热热穹顶及南半球主要粮仓减产，大宗商品期货正激进重估风险。 - [01:11:40](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4300s) 针对延迟上市催生的“SPV 套 SPV”暗箱交易，投资界呼吁必须彻底清除多层管理费与双重分成乱象，促使科技独角兽理顺股权结构并早日回归透明公开市场。 - [01:13:30](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4410s) Salesforce 践行 1-1-1 慈善模式（1% 股权、利润与工时），累计捐赠超 10 亿美元；Benioff 呼吁全行业加入 Pledge 1% 倡议，并动情缅怀挚友兼董事会成员 Susan Wojcicki。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=0s) 欢迎特邀嘉宾 Marc Benioff：政治光谱与白宫待遇 节目开场，主持人 Jason Calacanis 宣布这是《All-In Podcast》的第 273 期。由于常驻主持 David Sacks 缺席，本期迎来了重量级嘉宾——Salesforce 联合创始人兼 CEO Marc Benioff。现场随即开启了标志性的硅谷互损调侃：Jason 笑称，之所以请 Benioff 来，是因为其他所有顶尖科技高管全跟着美国总统特朗普访问中国去了，Benioff 是“唯一留在国内的软件业 CEO”。Benioff 则幽默接梗，称自己正是因为“独守空房”才拿到了上节目的资格。 随后，话题转向 Benioff 的政治立场。Jason 犀利发问：Benioff 过去在公众眼中一直是以支持民主党闻名的自由派科技领袖，如今怎么似乎也融入了特朗普的阵营？Benioff 当场断然否认了这一标签，直言自己“从来都不是什么左翼民主党人”。他强调自己的核心准则是“支持国家、为美国服务”，自己既非民主党人也非共和党人，而是一个“纯粹的美国人（American）”。Jason 顺势调侃起 Benioff 在政商社交圈的超高待遇——从身着燕尾服出席英国温莎城堡晚宴、接待查尔斯王子访美，再到与白宫高层一同出席沙特国宴，Benioff 始终手握全球顶级政治社交场合的核心入场券。 ### [01:12](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=72s) 特朗普与习近平北京峰会首日：经贸大单与台湾局势 在因伊朗局势推迟两个月后，备受全球瞩目的“特习会”（Trump-Xi Summit）终于在北京正式开幕。这是自 2017 年以来美国总统首次访华，也是特朗普与习近平主席的第七次面对面会晤。在首日的会谈中，双方迅速达成了一系列关键地缘共识：中国同意霍尔木兹海峡应保持畅通（中方无需承担军事部署承诺），并一致认为伊朗绝不能拥有核武器。 但在台湾问题上，气氛明显带有火药味。习近平发出明确警告，指出台湾问题如果“处理不好，两国将迎头相撞甚至发生冲突，使中美关系陷入极其危险的境地”。与此同时，预测市场 Polymarket 上的押注数据显示，参与者认为 2026 年台海爆发冲突的概率极低，仅为 6%（成交量高达 2300 万美元）；但到 2027 年底前发生冲突的概率则陡增至约 17%，几乎是当下的三倍。 在经贸合作层面，中方作出了实质性采购承诺，包括增购美国大豆、原油、液化天然气（LNG）以及多达 200 架波音（Boeing）客机。习近平表示此次会谈为两国构建“建设性、战略性、稳定的双边关系”描绘了愿景；而特朗普则一如既往地毫不掩饰对两人私交的赞美，哪怕在发言中特意加上免责声明称“大家总是不喜欢听我夸奖习主席”。 - **霍尔木兹海峡与中东安全**：中美达成一致，确保海峡通航自由，坚决反对伊朗拥有核武。 - **台湾议题红线**：中方表态“若处理不当，两国将迎头相撞甚至冲突”。 - **Polymarket 赔率数据**：2026 年底前中国收复台湾概率仅为 6%；2027 年底前概率攀升至 17%。 - **首波采购大单**：涵盖农产品（大豆）、化石能源（原油、LNG）及 200 架波音客机。 ### [02:52](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=172s) Friedberg 的视角：跳出“修昔底德陷阱”与做大蛋糕 针对峰会的深远意图，David Friedberg 重点分析了中美元首开场白中关于如何避免“修昔底德陷阱”（Thucydides Trap）的讨论。Friedberg 引述哈佛学者格雷厄姆·艾利森（Graham Allison）的理论指出，历史上当一个崛起的大国遭遇相对衰落的守成大国时，零和博弈往往不可避免地将双方拖入冲突泥潭，和平过渡的先例少之又少。 ```mermaid flowchart TD accTitle: 零和博弈与充裕时代对比图 accDescr: 对比传统资源受限下的零和争夺模式与技术突破驱动的做大蛋糕充裕模式 subgraph 传统零和模式["静态世界：零和博弈与冲突"] A1["资源有限 / 经济蛋糕静态"] --> B1["大国争夺存量市场与地缘势力"] B1 --> C1["落入修昔底德陷阱（必然走向冲突）"] end subgraph 充裕技术模式["充裕世界：科技驱动共同做大蛋糕"] A2["AI、自动化、生物科技爆发"] --> B2["全球生产力与物质充裕度飞跃"] B2 --> C2["多极协作共享增量，避免零和冲突"] end ``` Friedberg 提出，关键分水岭在于大国处于“静态资源争夺”还是“资源扩张”的历史阶段。如果全球深陷资源紧缺，各方只能通过掠夺土地、地缘特权来维持增长；但当下人类恰恰身处由人工智能（AI）、全自动化、生物技术所掀起的工业与科技革命临界点，一个真正的“充裕时代（abundance）”正在到来。在这种背景下，中美领导人最理想的战略视野不是争抢存量，而是如何确保未来 30 年内两国都能从加速扩大的全球蛋糕中成比例受益。对于特朗普而言，若能落实一系列兼顾国内就业岗位、提升民众收入并引入产业投资的经贸协议，将是一场战术与战略上的双重胜利。 ### [05:45](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=345s) Chamath 的视角：以双向经济深度绑定实现地缘缓和与“分蛋糕” Chamath Palihapitiya 则从地缘政治现实主义的角度给出了更为直接的剖析。他指出，中美两国底层价值观迥异：中国是自上而下的儒家集体主义治理哲学，而美国则是高度推崇个体主义的自下而上架构。然而，正是这种根本性的体制差异，反而为彼此留出了巨大的合作缓冲区，而连接双方的核心纽带唯有商业。 Chamath 引述一位资深民主党朋友的私下谈话：主流媒体只盯着“特朗普为何带一堆商业巨头访华”做表面文章，却忽略了最深刻的战略常识——通往中美免于冲突的缓和（détente）的最坚实路径，就是深度的“经济相互纠缠（economic entanglement）”。过去数十年这种关系是畸形单向的，美国只是单纯接收中国廉价商品；如今要实现真正的稳态，必须转变为双向绑定，由美国向中国庞大的消费市场倾销高附加值的重型装备、能源、汽车与核心芯片。 针对 Friedberg 偏向理想主义的“做大蛋糕”，Chamath 毫不避讳地指出，两国元首在闭门密室里真正在做的，其实是现实层面的“切分全球势力范围”： - **资源与技术置换**：中国高度渴求美国的能源补给与不可替代的关键底层科技。 - **地缘利益交换（Quid Pro Quo）**：美国以此为筹码，要求中国在关键战略区域降温降级介入程度，首当其冲是中南美洲（美国的后院），其次是中东，其三则是就亚太势力范围达成一种彼此默认的动态平衡。 - **清晰的两极格局**：通过这种地缘与经贸交换，世界被重新划定为一种边界清晰、可预测的双极体系（bipolar construction）。 ### [08:01](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=481s) Benioff 的视角：顶级销售团队出海签单 在回顾中美经贸关系经历的波折时，现场探讨了近年来的激烈摇摆：从去年的严厉关税战，到疫情后美国在医药、防疫物资、芯片及能源供应链上的“去风险与脱钩”呼声，如今两国却再次坐到同一张长桌前修复纽带。对此，Marc Benioff 极为乐观地指出，习近平在发言中明确使用了“希望打开一扇更宽的大门”的表述，这意味着商业合作的开放空间将显著扩大。 在 Benioff 看来，特朗普此行最精彩的一笔，是带去了全美最无可挑剔的“超级销售天团”。除了被称为“最具销售天才的总统”特朗普本人以外，代表团成员个个都是各自领域的顶级“销售冠军”： - **埃隆·马斯克（Elon Musk）**：代表特斯拉与尖端先进制造； - **黄仁勋（Jensen Huang）**：亲自在现场推销英伟达（NVIDIA）的计算芯片； - **凯利·奥特伯格（Kelly Ortberg）**：波音公司新任 CEO，一举斩获 200 架商用客机订单，并积极谋求美国金融服务的大单； - **布赖恩·赛克斯（Brian Sikes）**：全球最大私人粮食巨头嘉吉公司（Cargill）CEO，亲自挂帅向中国出口海量美国大豆。 Benioff 动情地强调，商业从来都是“推动世界改变最伟大的平台”。特朗普与习近平之间原本就具备良好且熟稔的个人私交，两人在峰会上笑容满面，深谙商业交易之道。随着全美顶尖 CEO 们全力以赴，这次访华必将让美国代表团带着满载的签约订单簿凯旋而归。 ### [09:58](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=598s) Salesforce 与外企在中国：阿里云合作与特斯拉的独特性 关于美国软件企业是否被允许在中国销售软件，Salesforce 首席执行官 Marc Benioff 详细解释了他们所采取的极为审慎的策略。由于中国拥有极其严格的数据本地化与数据驻留法律，Salesforce 从未在中国设立任何办事处，也从未直接雇佣员工。Benioff 直言，他们唯一在中国留下过痕迹的情况，就是在收购其他公司后发现对方在中国有业务分支，随后 Salesforce 必须立刻将其剥离。目前，Salesforce 在中国完全依赖与阿里巴巴建立的独家战略合作关系，所有业务与代码全部交由阿里巴巴托管并运行在后者的服务器上，形成了全球绝无仅有的“基于阿里云的 Salesforce（Salesforce on Alibaba）”运作模式。 这种合作主要服务于跨国顶级品牌。Benioff 拿全球奢侈巨头举例，比如路易威登（Louis Vuitton）以及 Chamath 格外偏爱的诺悠翩雅（Loro Piana）。Loro Piana 在中国销售大量的羊绒衫，其全球其他门店均统一使用标准的 Salesforce 架构，唯独在中国的零售与客户管理，必须无缝接入阿里巴巴的本地云端，以完全符合中国所有数据必须留存在境内、可被合规审查的监管要求。 ```mermaid flowchart LR accTitle: Salesforce 在全球与中国的运作架构对比 accDescr: 展示 Salesforce 全球通用云架构与中国阿里云本地合规架构的区别 subgraph Global["全球市场架构"] A["全球客户 (如 LVMH/Loro Piana)"] --> B["Salesforce 全球云端"] B --> C["全球直营办事处及数据中心"] end subgraph China["中国大陆特殊架构"] D["中国门店零售数据"] --> E["阿里云合规基础设施"] E --> F["基于阿里云的 Salesforce"] end ``` 随后话题转向了在中国运作的另一极端案例：特斯拉（Tesla）。Benioff 评价称，埃隆·马斯克（Elon Musk）在中国做成了一件不可思议的事情——特斯拉在中国完全独资，没有设立任何合资企业或本地代理，马斯克与中国最高领导人之间展现出了高度的相互尊重与客气。这些搭载大量摄像头和前沿 AI 技术的全智能电动汽车，在中国的工厂制造并在中国的大街小巷穿行。当被问及马斯克究竟是如何做到这一点时，Benioff 毫不吝啬地夸赞道：“我认为埃隆是全世界最伟大的推销员，这就是原因所在。” 不过，主持群也提出了更现实、甚至更带有一丝冷酷色彩的商业竞争视角（coopetition）：让特斯拉进入中国建厂，很大程度上也是为了让本土供应链学习其创新与工程体系，而如今中国车企在电动车领域已经发展成为特斯拉最强大的全球竞争对手。 - **零本地雇员**：Salesforce 从未在中国直接雇人，遇收购必迅速剥离本地资产。 - **独家通道**：除中国通过阿里云运行“Salesforce on Alibaba”外，全球其他市场均采用自主运营云架构。 - **特斯拉的唯一破例**：作为对比，特斯拉在中国拥有独资车企地位并全面掌控生产与 AI 车辆运营，成为极少数免除本土合资绑定的外企特例。 --- ### [12:07](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=727s) 比亚迪、关税威胁与中美贸易对美国中期选举的影响 当聊到中国电动车的发展速度时，现场气氛相当热烈。有人提起比亚迪最新推出的鸥翼门车型，直呼“太酷了，我也想搞一辆”，但也有人随即泼冷水反驳其安全性存疑。更深层的担忧在于产业冲击：有传言称此类车型可能通过各种贸易途径进入美国，而主持人立刻指出，如果中国制造的 2 万到 3 万美元高性价比电动车涌入美国本土，美国汽车产业几乎将在“一夜之间被彻底掏空并宣告终结”。同样身处中国数据合规旋涡的还有苹果的 Tim Cook，苹果必须把用户数据存放在中国本地数据中心，这一妥协在面对持不同政见者隐私要求时经常招致美国国内的激烈抨击。 话题随之切入地缘政治与美国国内政治的交叉点：在通胀持续高于 3%、选民因伊朗冲突等对外政策产生受挫感的情绪下，美国普通民众并未感到中美关系的宏大叙事能直接改善日常生活。特朗普面临着仅剩六个月的中期选举时间表以及随后的后半个任期，而中方的战略规划往往着眼于数十年乃至百年的长期棋局。 David Friedberg 分析指出，中国所能带来的二阶与三阶效应对选民的实际生活至关重要——最核心的就是“就业安全感、收入增长与生活成本”。选民的投票倾向取决于中美经贸谈判能否产生通缩效应： - **物价降低**：若贸易协议能够降低那些加征关税商品的门槛，让超市货架上的日常消费品价格直降 30% 到 40%，普通家庭就能立刻感受到生活成本下降的实惠。 - **出口拉动**：若能在协议中扩大美国优势产业的大宗出口，例如大豆、原油、核心技术与商业软件，将直接促进国内生产力，并为本土产业带来稳定的收入与就业保障。 Friedberg 总结道，到 11 月选民走进投票站时，这种最直观的钱包感受才是决定他们选择“投蓝还是投红”的真正关键。 --- ### [15:20](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=920s) 中期选举背后的选区重划、a16z 政治献金与中美大国分工 对于中期选举的推演，Chamath Palihapitiya 坦言这超出了自己的“薪资级别”，但他认为决定中期选举胜负的，与其说是这次中美元首峰会，不如说是近期最高法院关于杰利蝾螈（gerrymandering，选区划分不公）的裁决、弗吉尼亚州最高法院的判例，以及接下来各红州和蓝州立法机构对国会选区版图的重构。此外，海量政治资本的流向也是关键推手。《纽约时报》本周披露的一则令人意外的消息显示，硅谷风投巨头 Andreessen Horowitz（a16z）竟然成为了本轮选举周期中最大的捐助方。 Chamath 分析称，a16z 显然是在谋求成为美国金融基石的核心成员。从商业逻辑来看，作为管理费与资产规模驱动（AUM）的机构，在迈过 1000 亿美元大关后，它们的目标是走向万亿美元规模，成为科技与私市投资界的“下一个黑石（Blackstone）”。当一家机构开始涉足规模更大、深度卷入地缘政治的资本周期和关键经济部门时，政治布局就变成了必不可少的常态机制。 ```mermaid flowchart TD accTitle: a16z 资本升级与政治参与机制 accDescr: 说明 a16z 从千亿 VC 向万亿全阶段资管巨头转型必须融入政治的商业路径 A["传统风险投资 ($100B AUM)"] --> B["迈向全资产管理巨头 ($1T 目标 / 成为下个黑石)"] B --> C["重仓布局涉及地缘政治与产业政策的资本高壁垒行业"] C --> D["大规模政治献金与游说建构话语权"] ``` 从中美战略博弈的大局看，Chamath 认为两大强国之间有着非常强烈的“划分势力范围并达成交易”的底层动力： 1. **中方筹码**：掌握着全球绝大部分关键矿产资源与稀土供应链的绝对控制权。 2. **美方优势**：在人工智能算力情报（AI）上拥有几乎无可撼动的统治力，在能源领域亦掌握决定性的话语权。 3. **互换与妥协**：两国无需在所有细枝末节上寸土不让。例如，中国并不非得在智利、委内瑞拉或巴拿马过分深入扩张；美国也未必非得在马六甲海峡问题上维持极其极端的绝对立场。双方完全可以用美国的石油与技术换取中国的稀土与制造业产品，在理性的利益妥协中寻求各自的富足与平衡。 --- ### [17:40](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1060s) 中国的内部挑战、五亿中产阶级与“贸易委员会”构想 在看待中国当下的战略诉求时，与会者指出，外界常带有一种盲目的恐华情绪，但中国国内目前同样面临着极为严峻的系统性挑战，包括人口老龄化与出生率下滑、房地产行业面临的困局以及经济增速放缓。领导层最关心的核心命题，始终是让社会保持运转并维持充分就业。Marc Benioff 补充强调，中国领导层历史上最伟大的执政成就之一，就是将多达 5 亿人口从贫困中带入中产阶层，这与美国政治家的关注焦点大不相同，也是其政策的首要出发点。 正因为这一经济成就如今面临着外部需求萎缩和制造业外迁的挑战，中国对于扩大对外贸易的渴望反而更加强烈。虽然外界普遍认定其最想要的是收复台湾以及确立区域霸权，但现场讨论认为，其更现实、更紧迫的需求在于维护经济引擎的健康运转。 这一点在中美元首峰会的最新动态中体现得格外清晰： - **从和平委员会到贸易委员会**：特朗普不仅提到了推动和平的委员会，还首次公开抛出了设立“贸易委员会（Board of Trade）”的全新构想，这在以往的对话中前所未见。 - **“更宽敞的大门”**：中方领导层在会谈中公开表态希望打开“更宽的合作之门”。 - **跨国金融巨头的身影**：随同出席此次外事活动的还有 Visa 和万事达卡（Mastercard）的首席执行官。中国国内的日常支付与商业生态长期被阿里巴巴等本土巨头及其超级 App 所主导，并未对 Visa 和 Mastercard 完全敞开怀抱；如果能借助此次贸易谈判叩开中国庞大消费市场的大门，对这两家全球金融网络巨头而言将是历史性的巨大红利。 ### [19:50](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1190s) 经贸合作与AI芯片禁令：封锁是否已经失去意义？ 在讨论中美关系时，Marc Benioff 强调经济深度捆绑才是维护和平的最坚固纽带。无论是波音客机、大豆、支付服务，还是英伟达和高通（高通 CEO Cristiano Amon 也在访华团中）的芯片，美国应该尽可能多地向中国出口商品和服务。在他看来，两国之间的经贸合作越紧密，爆发战争的概率就越低。面对 Jason Calacanis 抛出的尖锐问题——“我们是否应该向中国出售最先进的 AI 芯片？”，Benioff 给出了出人意料的干脆回答：封锁在现阶段已经毫无意义。 Benioff 指出，中国本土的大模型已经极具竞争力，中国团队早已学会在缺乏顶级芯片的情况下完成模型训练。美国对拥有最尖端的 Blackwell 架构等芯片津津乐道，某种程度上只是一种“自我心理满足（ego gratification）”。事实上，中国厂商的“快速跟随（fast follow）”能力极强，美国一年前拥有的最顶尖模型能力，他们现在已经完全具备。因此，美国应该顺水推舟，能卖什么就卖什么，通过深度的商业利益绑定来换取地缘稳定。 ```mermaid flowchart TD accTitle: 芯片封锁与经贸和平的逻辑推演 accDescr: 对比全面出口与技术封锁对中美地缘政治及产业竞争的不同影响路径。 subgraph TradePolicy["经贸与出口策略"] A["向中国开放高端芯片与各类商品出口"] B["严格实施半导体出口管制与封锁"] end subgraph DirectImpact["直接产业结果"] A --> C["Nvidia 等美企垄断市场份额<br/>挤压华为等本土竞品生存空间"] B --> D["倒逼中国巨额补贴本土供应链<br/>华为与国产光刻设备加速突围"] end subgraph MacroOutcome["地缘政治与经济终局"] C --> E["全球生产力提升<br/>深度利益捆绑降低冲突概率"] D --> F["双方各自建立独立晶圆厂<br/>台湾地缘战略价值逐步稀释"] end ``` ### [21:01](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1261s) 晶圆制造本土化与台湾地缘价值的重塑 David Friedberg 进一步延展了芯片议题，认为随着中美两国内地半导体制造能力的建立，台湾地区的战略敏感性正在被实质性稀释。美国本土的芯片法案虽然伴随着磕磕绊绊，但台积电在亚利桑那州的工厂已经正式运转投产；与此同时，中国在华为的牵头下正全力推进芯片制造本土化，政府通过数十年的长期产业规划进行投资，不仅涵盖晶圆制造厂，更深入到半导体设备制造，以防范 ASML 等光刻机厂商断供带来的供应链冲击。 Friedberg 提出一个深刻的问题：一旦中美各自建成了自主可控的先进晶圆厂和半导体设备体系，台湾在经济和国家安全层面的不可替代性是否就不复存在了？届时或许并非美国“背叛”或抛弃了台湾，而是各方都不再对其过度神经紧绷。中国可能计划在 2040 年前后实现两岸的全面融合，到那时美国无论从国家安全还是经济利益出发，可能都不会再有强烈的介入动机。 - **全球生产力论**：Friedberg 认为，唯有推动全球生产力指数（Productivity Index）整体上升，人类福利才能最大化。将先进技术人为禁锢在一侧只会激化东西方冲突，技术的扩散才能带来更低的商品成本与更长久的和平。 - **外交筹码交换**：针对 Jason 提出的外交假设——美国停止对台军售以换取中国停止对伊朗出售武器，Friedberg 认为在务实的外交谈判中，这种互换利益的交易框架具有可探讨的现实空间。 ### [24:33](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1473s) 18个月技术倒计时：让黄仁勋赢，还是让华为赢？ Chamath Palihapitiya 给出了一个极具确定性的时间表：再过 18 个月，台湾问题将不再像今天这样成为国际舆论的核心交锋点。他解释称，目前美国在先进制程上距离摆脱对台湾关键战略代工的依赖仅剩 1 到 2 纳米的差距。伴随美国本土先进制程产能的逐步释放，以及跨领域底层工程能力的突破（例如 Neuralink 展示的能够在纳米级尺度下全自动进行脑部电极植入的机械手术设备），高精尖制造的壁垒将被重新定义。 在芯片销售问题上，Chamath 与 Benioff 站到了同一战线，强烈主张“卖掉芯片”。其核心商业逻辑在于抢占生态主导权：必须让英伟达继续大获全胜，而不能通过封锁断供为华为等中国本土竞争对手提供喘息和抢占市场的生存空间（oxygen）。既然模型能力终将被快速追平，政策制定者更应该关注的是建立合理的“了解你的客户”（KYC）审核机制以防止模型被滥用制造生物武器，而不是在硬件销售上自缚手脚。 ### [26:15](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1575s) 封锁假设、神经科技玩笑与 Tony Robbins 的创业启蒙 当 Jason 逼问 Benioff 如果中国决定封锁并收回台湾、美国是否应当出兵防御时，Benioff 直言自己多年来始终不同意历史学家 Niall Ferguson 的悲观论调，认为炒作台海军事摊牌是个“伪命题（nonsense issue）”，海峡两岸最终会走向和解。 随后话题在几位主持人的插科打诨中转向轻松氛围： - **Neuralink 玩笑**：Benioff 拿 Chamath 刚才提到的 Neuralink 机器人打趣，说要帮 Chamath 预约下周二下午 5 点的脑机接口手术，钻开脑袋给他“植入一点同理心”，还调侃自己会亲自在手术台旁拉着他的手。 - **《绿野仙踪》与恐慌症**：几人借着《绿野仙踪》里的铁皮人（缺心脏）和稻草人（缺脑子）互相戏谑。Friedberg 分享了自己带孩子去拉斯维加斯 Sphere 体验《绿野仙踪》的经历，并罕见地袒露自己患有人群恐慌症（panic disorder），在人员密集的密闭空间中容易极度焦虑并胡思乱想逃生通道或突发事件。 - **Tony Robbins 的商业启示**：话题随后拐向了潜能导师 Tony Robbins。Benioff 坦承自己是 Tony Robbins 的头号粉丝，甚至将其请到了当周的 Salesforce 大会上。面对主持人关于是否真的通过付费指导撬开成功大门的追问，Benioff 给出了毫不含糊的定论：“如果当年没有 Tony Robbins，今天就不会有 Salesforce。”正是 Robbins 引导他在创业初期打破思维阻塞、重新审视核心问题，才奠定了后续整个商业帝国的根基。 ### [29:02](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1742s) 托尼·罗宾斯的心力觉醒与第五届 All-In 峰会预热 对话承接上一环节关于个人成长的心得，马克·贝尼奥夫（Marc Benioff）深有感触地引用了著名励志导师托尼·罗宾斯（Tony Robbins）的名言：“你提出问题的质量，决定了你人生的质量。”贝尼奥夫直言，年轻时的自己从没想过这些问题——你真正想要什么？什么对你最重要？你打算如何获取它？又是什么在阻碍你？而当你真正得到它时，你又该如何感知？当这些问题如同醍醐灌顶般砸向他时，他立刻意识到必须将其逐字写下，明确当下的聚焦点。这种前所未有的清晰感，成为他日后源源不断的内驱力。杰森·卡拉卡尼斯（Jason Calacanis）打趣问他是否也参加过托尼·罗宾斯标志性的“赤脚踩炭火”挑战，贝尼奥夫毫不犹豫地笑着承认自己走过好多次，甚至热情相邀“下次我们可以一起去走”。杰森则笑着回忆托尼此前做客节目时，大家本想深挖其心理学底层逻辑，结果全场都被他带着做了一整套深呼吸训练。 随后，杰森插播宣布 All-In 峰会（All-In Summit）即将迎来连续第五届举办，时间定在9月13日至15日。杰森调侃道，正因为贝尼奥夫在往届峰会上的主题演讲评分实在太高，主办方特意让他休整两年以充分拉高观众的期待值，因此贝尼奥夫将在2027年的第六届峰会隆重回归。 --- ### [30:16](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1816s) 弗里德伯格的流浪狗救助呼吁与贝尼奥夫的“云”猫 在切入正题前，戴维·弗里德伯格（David Friedberg）突然插播了一条紧急公益求助（PSA）。弗里德伯格向导播调出链接，神情严肃且动情地替洛杉矶鲍德温公园（Baldwin Park）动物收容所一只名叫“Scooter”的小狗寻找领养人。他解释称自己答应了妻子一定要在节目中发声，因为目前洛杉矶的街头流浪狗泛滥成灾，动物收容所早已不堪重负，大量无辜收容的流浪犬只正面临被安乐死的命运。弗里德伯格感叹，这些小动物并不是自己选择流浪街头，结果却难逃厄运。 面对弗里德伯格的动容，兄弟会成员们开始互相调侃。杰森开玩笑问查马斯·帕里哈皮蒂亚（Chamath Palihapitiya）愿不愿意掏1万美元多给这些狗狗续留一周的命，查马斯则幽默回怼：“里面有纯种的白色金毛巡回犬吗？”弗里德伯格苦笑表示全都是杂种串串。此时贝尼奥夫也分享了一段暖心插曲：他们家此前带孩子去收容所领养宠物，刚迈进大门，一只猫咪就主动朝着孩子们扑过来，他们当场就决定将其领养回家。当被问及这只幸运猫的名字时，贝尼奥夫笑着给出了极具个人烙印的答案：“它叫‘云’（Cloud）。” --- ### [31:41](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1901s) “SaaS 末日”来袭：软件股惨遭重估与贝尼奥夫的淡定回应 轻松的气氛随即转向严肃的商业硬核话题。杰森正式抛出软件行业的重磅议题——AI 对 SaaS（软件即服务）领域造成的惨烈冲击。当前二级市场正笼罩在极度恐慌中： - 赛富时（Salesforce）股价重挫 37%，市值蒸发约 900 亿美元； - ServiceNow 跌去 42%； - Workday 暴跌 45%； - 整个传统企业级软件板块累计损失超 1800 亿美元市值。 市场普遍悲观地认定，未来的企业不再需要 Slack、Salesforce 或 HubSpot，所有人只需对着一块玻璃屏幕对 AI 助手下指令，AI 就能自动解决所有业务流程。面对这种“界面消亡、传统软件归零”的巨大恐慌，贝尼奥夫显得相当从容，幽默地调侃道：“这可不是我经历的第一次 SaaS 末日（SaaSpocalypse）。”他笑称公司上下现在都在喝着菝葜汽水，还养了只“大脚怪”（Sasquatch）吉祥物，让团队变得更加充满斗志与个性（Sassy）。 贝尼奥夫随后严肃地剖析了资本市场的本质。在他看来，企业级软件板块正在经历周期性的整体“估值重估”（rerating）。他在企业软件行业打拼了40年、创办 Salesforce 27载，深知市场每隔一段时间就会经历这种洗牌。以 HubSpot 为例，其估值倍数居然被砸到了2倍市销率（P/S），这在他职业生涯中闻所未闻，即便其刚交出一份亮眼的财报。贝尼奥夫横向对比了排名前十的头部企业软件公司，发现大家季度业绩普遍过硬，但估值倍数无一例外被压制在极低水平。他指出，市场现在陷入了一种盲目的“AI 催眠”，但这种所谓的取代逻辑目前根本没有反映在实际财务数据中，全球范围内企业软件的实际采购需求依然十分庞大。 ```mermaid flowchart TD accTitle: 市场恐慌与 SaaS 现实反差 accDescr: 对比市场对生成式 AI 颠覆 SaaS 的悲观预期与大型企业级软件的实际业务基本面 subgraph Fear["二级市场悲观叙事（AI 催眠）"] A["生成式 AI 接管端到端工作流"] --> B["企业无需购买传统 SaaS 席位"] B --> C["软件股估值暴跌（P/S 遭重挫至 2x）"] end subgraph Reality["企业级软件基本面现状"] D["TOP 10 头部软件巨头业绩依然强劲"] E["每年数百亿美元确定性现金流支撑"] F["复杂企业级业务逻辑与深度客户信任壁垒"] end Fear -.->|"脱离实际数据支撑"| Reality ``` --- ### [34:09](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2049s) 面对估值杀跌的企业定力：锚定客户成功与充沛现金流 面对公募市场无情的估值压制，杰森追问贝尼奥夫如何稳定内部团队军心——尤其是在员工薪酬大量由股权构成、且面临来自 OpenAI、Anthropic 和 SpaceX 等炙手可热的非上市公司严峻挖角压力的情况下。贝尼奥夫坦言，当前环境下非上市公司的确更具抗压优势，因为它们无需承受公募市场残酷的逐日盯市重估。 谈到军心管理，贝尼奥夫给出的心法是“切忌沉溺于股价”（You can't get drunk on the stock price）。他反复告诫员工，如果把个人的情绪波动和价值感完全锚定在当天的股价红绿上，心态必然崩溃，大家必须找到另一个坚实的支点： - **稳固的财务大盘**：Salesforce 本财年营收将突破 460 亿美元； - **充沛的造血能力**：自由现金流将超过 160 亿美元； - **庞大的组织体系**：拥有超过 83,000 名员工； - **长期战略支点**：始终聚焦于客户最终的成功交付（Customer Success）。 贝尼奥夫直言，无论是管理层还是普通员工，谁都无法左右外部股价的波动，唯一能掌控的就是为客户创造实际价值的深度。而在被杰森赞许其过往出色的并购眼光时，贝尼奥夫更展现了老牌巨头的从容：“现在一切标的都变便宜了，这正合我意。” --- ### [35:50](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2150s) 查马斯拆解反向交易：低端 SaaS 覆灭与高壁垒巨头的逆袭 查马斯在随后的分析中明确划分了软件市场的两极分化。他认同低端 SaaS 市场确实已经名存实亡，缺乏护城河的单功能工具注定被 AI 吞噬；但像 Salesforce 这样牢牢占据企业级核心的大型单体系统（monoliths），护城河反而极其稳固。 查马斯援引了本周爆出的重磅交易作为佐证：OpenAI 牵头组建了一家规模达 40 亿美元的部署/实施服务公司，并为外部投资者提供了高达 17.5% 的优先保底收益率。查马斯指出，OpenAI 不惜代价设立这样一家本质上对标安永（EY）、安达信、德勤（Deloitte）、普华永道（PwC）或高知特（Cognizant）的实施集成实体，在传统商业逻辑上是算不过来账的，但这恰恰暴露了 OpenAI 在高端企业级市场碰壁的窘迫——AI 在顶尖企业的落地远非简单的“输入 Prompt、按回车就能自动化”那么轻松，企业级业务逻辑极其复杂沉重。 基于此，查马斯认为公募市场目前对传统企业软件巨头的抛售严重过度，并提出了极具前瞻性的“反向交易”（Opposite Trade）逻辑： ```mermaid flowchart LR accTitle: 查马斯的反向交易演进路径 accDescr: 展示 AI 资本开支从狂热审视到回归现实，最终倒逼大模型厂商依赖传统企业分销渠道的过程 subgraph Phase1["第一阶段：现实敲打"] A["OpenAI 耗资 40 亿做实施外包"] --> B["高端企业落地受阻：落地远非 Prompt 那么简单"] end subgraph Phase2["第二阶段：ROI 大考"] C["资本市场质疑 3 万亿美元 AI 资本开支"] --> D["严厉审问大模型 Token 转化效率"] end subgraph Phase3["第三阶段：反向套利"] E["AI 实验室被迫低头"] --> F["求助贝尼奥夫等传统巨头分销 Token"] G["高留存率、深厚 CXO 信任壁垒的巨头最终通吃"] end Phase1 --> Phase2 Phase2 --> Phase3 ``` 查马斯总结道，决定企业未来胜负的核心资产是高度可预测的净收入留存率（Net Dollar Retention）、极低的客户流失率，以及在大型企业董事会和 CXO 层级中深耕二三十年所建立的极致信任。在过去四年中，科技行业在 AI 基础设施上砸下了 3 万亿美元的巨资；当二级市场的狂热情绪退潮、开始严肃追问“这 3 万亿美元的 Token 投资回报率（ROI）究竟在哪里”时，最终的一幕必然是大模型实验室不得不反过来敲开贝尼奥夫这些传统巨头的门，哀求道：“求求你们帮我卖掉这些 Token 吧！”这才是企业级软件重估潮后真正的主线逻辑。 ### [38:16](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2296s) 史上罕见的500亿美元回购与Anthropic驱动的研发效率 随着软件行业估值倍数从过往的非理性繁荣逐步回落，市场正在寻找新的估值平衡点。主持人指出，Salesforce 启动了一笔堪称史无前例的 500 亿美元巨额股票回购计划。Marc Benioff 对此直言不讳，这不仅是科技史上规模最大的回购之一，更体现了公司全力回购自家股票的决心。在 Benioff 看来，当前的宏观与技术转折点赋予了企业前所未有的新能力，其中代码生成智能体（coding agents）的发展尤为惊人，尤其是 Anthropic 展现出的实力。 Benioff 透露，Salesforce 仅在今年就预计采购并消耗价值约 3 亿美元的 Anthropic 算力与服务用于辅助编程。得益于 AI 编码助手，软件研发成本大幅下降，生产效率呈指数级提升。他表示，自己现在能够做到过去无法想象的事情：一边极速交付软件，一边同步对外销售；人类工程师、AI 智能体与无头（Headless）平台三者无缝协同，彻底打破了以往的研发与组织瓶颈。从技术服务、客户支持到产品分发与营销，Salesforce 内部实现了前所未有的效率跃迁，而 Anthropic 就像一艘势不可挡的火箭，与 Salesforce 平台深度互补，正在重塑交付客户价值的上限。 - **股票回购规模**：启动高达 500 亿美元的股票回购计划，创下历史性纪录。 - **Anthropic 采购开支**：Salesforce 预计今年在 Anthropic 编码与模型调用上投入约 3 亿美元。 - **组织协同新范式**：形成“人类员工 + AI 智能体 + Headless 底座”三合一的高效研发与运营流水线。 --- ### [40:04](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2404s) 斥巨资收购Informatica：数据语境与语义层构建 在危机中寻找机遇，Salesforce 在严格控制人员规模、精简机构与股票回购的同时，完成了近年来规模最大的一笔并购交易——以大约 80 亿至 90 亿美元收购数据集成巨头 Informatica。Benioff 强调，所有的 AI 落地如果脱离了“语境”（Context），都将毫无用处。大语言模型本质上是概率模型，具有很强的不确定性，企业级应用必须将其牢牢锚定在真实、统一的数据源与语义层之上，否则系统就会陷入幻觉与混乱。客户最迫切的需求是对分散在各处的数据进行统一和谐化、联邦化与深度集成，这也是 Salesforce 收购 Informatica 的底层逻辑。 数据统一后，大型模型便能赋能各种前端与后台工作流。Benioff 举例称，如今拨打 Salesforce 经典的“1-800-NO-SOFTWARE”客服热线，27 年来首次接待客户的不再是人工座席，而是 AI 智能体“Agentforce”。系统在快速完成身份验证并摸清意图后，若遇未识别信息，能瞬间无缝转接给人工客服；此时人类座席的屏幕上已同步展示完整的交互上下文。这种电话、网络与人类无缝咬合的“三位一体”服务体验，正是由 AI 与底层数据层共同驱动的成果。 ```mermaid flowchart TD accTitle: Agentforce Context-Driven Architecture accDescr: Informatica provides unified data context and semantic layer to large language models, powering the Agentforce customer interaction and seamless human handoff. subgraph DataLayer["统一数据与语义层"] D1["企业多源业务数据"] --> D2["Informatica 联邦与清洗"] D2 --> D3["单一真实可信数据源 (Single Source of Truth)"] end subgraph AIPlatform["模型与智能体引擎"] D3 --> LLM["大语言模型 / 语义层"] LLM --> AF["Agentforce 智能中枢"] end subgraph Interaction["全渠道交互流水线"] User["客户接入 (1-800-NO-SOFTWARE / Web)"] --> AF AF -- "自动验证与意图识别" --> Resolve["即时自动化解决"] AF -- "遇异常或复杂诉求" --> Human["人工座席 (完整上下文无缝交接)"] end ``` --- ### [41:36](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2496s) Headless Slack与平台API演进：企业的“世界模型” 谈及将 Slack 改造为无头（Headless）架构的战略抉择，Benioff 回应称这在内部其实顺理成章。Salesforce 的底层架构一直引领着 API 的行业演进：从早期的 XML API、SOAP API、REST API，再到如今的 CLI API、MCP（Model Context Protocol）API，平台始终保持全方位开放。应用的逻辑并非写死在前端，而是通过全新推出的 AXL API 实时向外推流，能够随时随地投射并显现到任何大模型、端侧设备或外部界面中。 对话随后引申到 Jack Dorsey 此前的一篇备忘录，后者提出希望通过为企业构建“世界模型”来管理公司。嘉宾们一致认为，要构建企业的世界模型，最核心的语义理解和实时上下文实际上全沉淀在企业日常沟通中——也就是 Slack 和电子邮件里。Benioff 透露，通过 Slackbot 结合大模型，系统正在持续理解公开频道乃至授权私聊中的沟通数据，其洞察企业脉搏的敏锐度甚至超过人类管理层自己。高管只需向 Slackbot 提问，系统便能瞬间回答诸如“目前最大的五个未结单交易是什么”、“员工最不满什么问题”、“本周最重要的三项业务优先级是什么”等深层管理问题。 Jason Calacanis 也分享了自身的实操体验：在为公司升级更高级别的 Slack 方案后，他成功打通了数据导出接口，接入了 OpenAI、Perplexity 与 Claude 等多种工具。他甚至在录制节目前写好了一个全自动 Prompt 流程：系统每两小时扫描一次公司 Slack 中的全部讨论，自动梳理出正在制定的决策、决策者身份，并分别以“幕僚长”、“CEO”与“董事会成员”三个虚拟角色的视角对决策质量进行评分和风险排查。 - **架构进化轨迹**：经历 XML、SOAP、REST，升级至 CLI、MCP 以及全新的 AXL 串流 API。 - **Slack 核心资产地位**：汇聚了企业最真实的实时上下文，OpenAI、Anthropic 等前沿 AI 巨头内部均全盘采用 Slack 与 Salesforce 销售/服务云作为运营基石。 --- ### [44:47](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2687s) 撕掉私有SaaS的遮羞布与水平软件的逆袭 面对大量在 AI 浪潮前创立、曾试图冲刺 IPO 却至今搁浅的非上市 SaaS 企业，Jason 询问 Benioff 会给这些公司的董事会与创始人提出什么建议。Benioff 直白而尖锐地调侃道：“给他们递一盒舒洁（Kleenex）纸巾擦眼泪吧！”他直言许多未上市公司的 CEO 终日抱怨估值缩水、心血得不到回报，但“快长大人吧”。在他看来，公开市场最可贵之处在于其时刻在对一切资产进行理性定价；而在私募市场中，没有真实买家接盘的所谓高估值不过是一场数字幻象。他告诫创业者必须抛弃幻想，将全部精力回归到基本面：专注营收增长、聚焦客户体验、守住自由现金流、追求实际盈利以及持续的产品创新。 David Friedberg 随后从实际采购角度分享了重要趋势观察。在为其农业科技公司 Ohalo 进行大规模 Salesforce 落地实施时，他注意到过去半年内部出现了一个明确转向：他们基本放弃了各类细分垂类软件（vertical software），转而大幅加码对水平平台型工具（horizontal tools）的投入。企业正利用强大的底层水平平台与生成式能力，自行定制符合自身业务特性的专属应用和工作流，而不再采购昂贵的现成通用套件。这一现象暗示着资本市场上垂类与水平软件被一刀切定值的现状可能存在套利空间。 Benioff 对此深表赞同，并以销售端的人工智能变革作为佐证：Salesforce 拥有 1.5 万名销售人员，但内部统计发现在过去的 27 年里，由于人力所限，竟然有高达 2000 万至 3000 万个潜客咨询未被及时回呼。为了解决这一痛点，Salesforce 收购了潜客资格筛查公司 Qualified。就在本周，他们依托 AI 智能体一次性向 5 万名客户进行了自动外呼与商机资格甄别，以极低成本全面重构了传统 BDR（业务拓展代表）与 SDR（销售开发代表）的工作模式。 --- ### [47:24](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2844s) 突发快讯：OpenAI拟就ChatGPT集成违约起诉苹果 讨论接近尾声时，现场切入了一条刚刚由彭博社曝光的突发行业重磅新闻：OpenAI 正在考虑就双方的 ChatGPT 合作协议正式起诉苹果公司（Apple）。回顾两年前，苹果与 OpenAI 达成了一项轰动业界的战略合作协议，宣布将 ChatGPT 深度整合进 Siri、iOS 以及 macOS 操作系统生态中。 然而据彭博社披露，这笔交易在实际推进中对 OpenAI 而言极其不如人意，甚至可能演变为一场指控苹果违约的法庭恶战。OpenAI 的核心不满之一在于苹果对集成交互的苛刻限制：在 Siri 的实际唤醒逻辑中，苹果强制要求用户在发出指令时必须明确说出“ChatGPT”这一特定前缀，系统才会将请求转交给 OpenAI 的模型生成结果。这一极其低效繁琐的交互路径不仅严重削弱了用户的使用体验，也导致实际调用量大打折扣，被嘉宾痛批为将本就体验极差的 Siri 进一步变成了最难用的个人助手产品。 ### [48:09](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2889s) 苹果与 OpenAI 合作破裂：硬件野心与各家 AI 赛道的紧急掉头 Jason 引述彭博社报道指出，苹果与 OpenAI 之间高调宣布的合作并未取得预期成果。OpenAI 原本期望这项整合能带来数十亿美元的订阅分成收入，但苹果几乎完全没有在系统中主动推广 ChatGPT，绝大多数用户依然倾向于使用独立的 ChatGPT 应用程序。报道披露，苹果内部对 OpenAI 的隐私数据处理方式深感担忧，对 OpenAI 管理层缺乏信任；更致命的是，苹果对 OpenAI 进军硬件领域的野心极度不满——OpenAI 不仅秘密研发与苹果竞争的硬件设备，还直接挖走了苹果的前首席设计官 Jony Ive。 针对两家巨头的龃龉，Salesforce 首席执行官 Marc Benioff 呼吁大家把视角拉高来看待整个行业的发展轨迹。他指出，大语言模型兴起初期，各大科技巨头各自走上了截然不同的岔路：Elon Musk 专注于让 Grok 搞伴侣机器人和各种情感交互；OpenAI 则沉迷于 Sora 视频生成以及广告网络等庞杂尝试；Google 忙于 Gemini Nano 等端侧部署；唯独 Anthropic 压注于编程智能体（Coding Agents）。事实证明 Anthropic 踩中了真正的杀手级场景，凭借代码能力一飞冲天。眼看 Anthropic 遥遥领先，整个行业瞬间迎来大洗牌，各大巨头纷纷按下重置按钮，紧急砍掉伴侣机器人和边缘项目，全员重新聚焦代码 Agent。 ### [50:18](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3018s) 编程革命：从键盘时代迈向脚踏板、语音交互与 Slack 的秘密武器 Benioff 强调，当前 AI 行业最令人兴奋的就是这种瞬息万变的动态调整能力。随着 Anthropic 高阶模型的爆发，企业内部人人皆可编程的时代骤然降临，AI 编程从过去微不足道的边际效率提升，变成了革命性的生产力跃迁。Benioff 在节目中透露了一则独家内幕：Salesforce 正在 Slack 内部秘密研发全新的代码协作技术，旨在让所有团队成员更轻松地直接编写代码。他感慨道，自己 15 岁在 Burlingame 高中时，是用双手在键盘上敲 6502 和 68000 汇编代码写电子游戏，而如今产品经理、开发人员与 UI 设计师的角色界限已彻底瓦解，融为一体。 Chamath Palihapitiya 对此深表赞同，并生动描绘了当下一线开发者的颠覆性日常：真正的工程师甚至已经不再用手碰键盘。在 Chamath 投资的公司里，工程师通过脚踏开关触发麦克风，配合 Whisper 与语音流输入工具全口述写代码，“脚踏板加语音输入，键盘时代已经终结”。面对这种狂热的无键盘趋势，Benioff 也幽默打趣称 Salesforce 正在推行自己的“海豚流”（Dolphin Flow），引得全场哄堂大笑。 - **角色融合趋势**：产品经理（PM）、全栈开发者与 UX 设计师之间的职能壁垒正在消失，统一演变为调用 Agent 的“创造者”（Creator）。 - **人机交互变迁**：从早期的底层汇编纯手工输入，演变为语音控制工具（如 Whisper、Flow）配合脚踏开关的连续口述编程。 - **Slack 战略动向**：Salesforce 正在将原生代码构建能力深嵌进 Slack，以降低非技术人员的软件创建门槛。 ### [52:21](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3141s) 个人助理生态战：Google 的数据护城河与苹果的伙伴困境 谈及 OpenAI 屡屡出现波折的外部伙伴关系，David Friedberg 尖锐地指出，OpenAI 似乎始终无法维持稳固的长期商业联盟。Jason 抛出疑问：苹果是否应当彻底放弃 OpenAI，重回老盟友 Google 的怀抱——毕竟双方此前在 Safari 默认搜索引擎的合作中曾让苹果赚取数百亿美元。 Friedberg 分析认为，Google 在打造终极个人 AI 助手方面占据天然优势，因为 Gemini 可以直接深度集成用户的 Google Drive、Gmail、Google Photos 以及 Google 日历等核心私域数据。无论是追踪邮件往来、查询个人日程，还是在企业级 Google Workspace（G Suite）中调用业务流，助理要真正具备实用价值，核心前提是必须拥有对个人全量数据的访问权。苹果虽然坐拥海量存量硬件生态，但若想打造同等级别的体验，要么必须自研底层模型，要么就得选择类似 Anthropic 这样的第三方伙伴贴牌合作（White-label）。但在 Google 庞大的数据飞轮压迫下，苹果将面临一场极其严峻的硬仗。 ### [54:19](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3259s) 本地运行 vs 云端持久化：苹果硬件优势的激辩 Jason 对苹果的前景给出了极其乐观的研判。他认为苹果只需出手收购 Perplexity 或 Mistral 这类头部 AI 实验室，便能依靠其无与伦比的硬件底座跃升为全球排名前三的 AI 巨头。Jason 举例称，自己新购入的配备 48GB 统一内存的 M 系列 MacBook 性能极其强悍，而未来配备近 1TB 统一内存的 Mac Studio 更将成为本地运行大模型的终极神器。在本地端侧运行模型既能彻底索引用户的全部照片与文件，又不必担心隐私泄露给“无人信任的 OpenAI”或 Google，使用高性能端侧硬件的员工生产力能够直接跃升 10 倍。 ```mermaid flowchart TD accTitle: 本地端侧架构与云端持久化架构的分歧 accDescr: 对比 Jason 主张的本地高性能计算与 Chamath 主张的多设备云端同步架构 subgraph LocalView["Jason 的端侧优先主张"] L1["苹果强大硬件<br/>（高统一内存 M 系列芯片）"] --> L2["端侧本地大模型"] L2 --> L3["绝对隐私保障<br/>（无数据上传风险）"] end subgraph CloudView["Chamath 的云端持久化主张"] C1["跨设备场景<br/>（手机 / 网页 / 多台电脑）"] --> C2["云端持久化上下文"] C2 --> C3["全时全场景伴随<br/>（打破单机硬件绑架）"] end ``` 然而 Chamath 坚决反对端侧本地模型的构想。他直言端侧方案存在致命缺陷——缺乏“跨设备的持久性”（Persistence）。在多设备、多身份并行切换的现代数字生活中，用户不可能随身背着一台重达 5 磅的 MacBook Pro 仅为了向本地模型询问医生的健康诊断，一旦脱离单台机器，上下文便发生断裂。尽管 Jason 反驳称苹果可以通过 iCloud 在云端实现无感同步，但 Chamath 坚持认为单纯依赖传统电脑硬件构想未来完全是误判了发展方向。 ### [56:10](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3370s) 下一代硬件形态与 Thinking Machines 的实时多模态系统 Chamath 进一步向 Jason 发难：当前对于苹果优势的设想，本质上都建立在“计算硬件外形规格（Form Factor）长期不变”的前提之上。然而苹果如今的辉煌是过去三四十年针对 PC 和智能手机微调优化的产物，一旦 AI 催生出类似当年 iPhone 问世般的全新硬件形态，庞大的传统组织将极难完成掉头。 Jason 赞同硬件形态即将剧变，并指出这正好与 Mira Murati 创立的新公司 Thinking Machines 最新发布的实时多模态系统不谋而合。结合苹果正在申请的带有摄像头的 AirPods 专利以及各类穿戴设备（如 AI Pin、智能手表），未来的常驻硬件将实时捕获用户的物理现实与桌面环境，促使企业用户的 Token 消耗量暴增上千倍。Jason 揭示了 Mira Murati 团队系统的核心运作机制：该模型同时监控用户的电脑桌面、环境声音与摄像头画面，并且每隔 200 毫秒完成一次高频数据采样与上传，分别投喂给两个协同工作的后台模型——一个负责即时交互的超低延迟模型，以及一个回溯 30 秒历史、负责深度推理的思考模型。 ```mermaid flowchart LR accTitle: Thinking Machines 实时多模态双模型架构 accDescr: 传感器高频数据采样并分发给实时模型与深度思考模型的运作机制 subgraph InputStreams["全时多模态感知输入"] S1["桌面屏幕画面"] S2["多路麦克风音频"] S3["高清摄像头画面"] end subgraph Ingestion["采样与切片"] P1["每 200 毫秒高频上传"] end subgraph DualEngine["双模型协同处理"] M1["实时低延迟模型<br/>（即时人机交互）"] M2["深度思考模型<br/>（回溯分析 30 秒上下文）"] end InputStreams --> P1 P1 --> M1 P1 --> M2 ``` ### [57:54](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3474s) 多感官模型与实时交互：LLM 并非通往 AGI 的唯一解 讨论由 Thinking Machines 等公司推出的实时模型切入。Jason 敏锐地指出，一旦交互模式从传统的“发送提示词等待回复”转变为持续 8 小时的实时双向互动，对 Token 的需求将暴增 1000 倍。这种毫秒级的全时段推理不仅对模型吞吐量提出极端考验，也将直接倒逼普通桌面端硬件的大规模升级。面对 Jason 抛出的“这个美丽新世界该如何应对”的提问，Marc Benioff 给出了冷峻而深刻的技术反思。在他看来，实时多感官模型是“人类迈向通用人工智能（AGI）的一小步，也是巨大的一步”，但他对科技界将大语言模型（LLM）奉为圭臬的狂热持保留态度。 Marc Benioff 直言，行业目前充斥着一种“LLM 能搞定一切、直接通往 AGI”的盲目信仰，但他始终无法理解仅仅基于语言文字的统计概率预测——即“上一个词预测下一个词”的技术范式——如何能够带人类实现《少数派报告》（Minority Report）等科幻电影中描绘的真正智能。Benioff 拿自己打了个生动的比方：“我就是一个运行在生物计算机里的多感官模型。”人类拥有眼睛、耳朵、嘴巴、大脑与心脏，处理着极其复杂且目前尚未被完全解析的多维信号，而绝不仅仅是一个吐字的文本模型。他坚信，多感官模型（Multi-sensory Models）才是 AI 的下一波重大浪潮，尽管它依然不等于真正的 AGI。随着这波新模型的演示震撼面世，所有模型公司都面临残酷的“再次转向（Pivot）”，必须在这条赛道中重新选定阵营。 --- ### [59:55](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3595s) 3亿美元 Token 账单与智能路由：端侧与云端协同降本 针对模型消耗激增的前景，Jason 向 Benioff 发难：Salesforce 每年在 Anthropic 上就要砸下 3 亿美元采购 Token，如果未来每个员工的交互需求扩大 1000 倍，企业的 Token 开销岂不是要从几十万美元飙升至上亿美元？Benioff 针锋相对地驳斥了这一推论，认为这纯粹是被模型厂商“洗脑”后的焦虑误区。 Benioff 拆解了 Salesforce 内部的真实代码编写场景：在海量的编程需求中，绝大多数任务根本不需要调用昂贵的前沿旗舰大模型。他判断，未来必然会出现一层智能中间调度层（Router），由其在用户与 Anthropic、OpenAI 之间进行流量分发，将简单琐碎的日常请求交给廉价的小模型或端侧设备，仅在真正遇到棘手复杂难题时才求助于旗舰云端模型。Benioff 用洗澡打了个极具画面感的比喻：“这就好比我只想冲个澡，你却把整个水库的水一股脑全浇在我头上；我需要的只是一点水，而不是全部的水。”他高度评价了 Jason 与 Chamath 此前提出的观点——分布式边缘智能（Edge Intelligence）与云端能力的无缝融合，将彻底重构成本曲线，当前看似高昂的算力开销只是产业发展初期的暂时阵痛。 ```mermaid flowchart TD accTitle: 智能模型路由与端云协同架构 accDescr: 展示用户请求经过智能路由层根据复杂度分发至端侧小模型或云端前沿大模型的降本流程 User["用户实时多感官交互<br/>（高频连续任务）"] --> Router["智能调度层<br/>（中间路由 Agent）"] subgraph RoutingDecision ["任务复杂度评估与分流"] Router -->|"低复杂度 / 日常任务<br/>（本地响应）"| Edge["边缘端小模型<br/>（低成本/低延迟）"] Router -->|"高复杂度 / 深度推理<br/>（按需调用）"| Cloud["云端前沿大模型<br/>（Anthropic / OpenAI）"] end ``` - **Salesforce Token 年支出**：当前每年向 Anthropic 支付约 3 亿美元用于代码生成与业务支持。 - **调度层商业机遇**：业内将涌现估值可观的新兴初创企业，专精于模型网关与分层路由，打破前沿基础模型公司的算力垄断。 - **核心计算范式**：从单纯依赖中心化云端推理，转向“边缘设备（Edge）+ 云端（Cloud）”协同的分布式智能架构。 --- ### [01:02:05](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3725s) 科学角落前奏：硅谷最强推销员与托尼·罗宾斯玩笑 在进入尾声议程前，Jason 列出两个收尾话题供大家选择：其一是关于 Anthropic 打击二级市场“灰色 SPV”（Dark SPVs）对投融资生态的震荡，其二则是听众翘首以盼的“科学角落”（Science Corner）。几位主持人一致决定先由 Friedberg 带来科学分析，随后再快速进入 SPV 闪电战。 现场气氛随即转向欢快的调侃。面对 Marc Benioff 对科学板块的浓厚兴趣，Jason 与主持人们打趣称 Benioff 是“硅谷有史以来最伟大的销售大师”，甚至开玩笑追问他是如何炼成这种登峰造极的推销艺术的。Benioff 笑着自谦自己骨子里其实是个纯粹的技术极客（Geek），并顺势拿励志演说大师托尼·罗宾斯（Tony Robbins）开涮，戏称大家应该结伴去参加罗宾斯的潜能培训班，一起在炭火上赤脚行走（Firewalking）来历练心性。 --- ### [01:03:01](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3781s) 科学角落：历史级厄尔尼诺来袭与全球粮食、电网危机 玩笑过后，Friedberg 正式开启本期科学角落，展示了一张令人触目惊心的海表温度异常预报图（Sea Surface Temperature Anomalies）。数据显示，全球海洋积聚的热量已达到近代观测史上的前所未有水平，局部海温较常年偏高整整 4°C，各项指标直逼甚至超越了历史上破坏力最强的 1877 年超级厄尔尼诺事件。 Friedberg 将广袤的海洋比作全球天气的“巨型蓄电池”：在一年中的特定周期，海洋疯狂吸收太阳与环境热能，随后在特定季节集中向大气释放，从而彻底颠覆全球信风系统与水汽分布。目前海洋中积蓄的超额能量高达约 1100 万太瓦时（TWh），而全人类一年的总耗电量才不过 2.5 万太瓦时——这意味着海洋里封存了相当于人类文明 500 年用电总和的过剩能量。随着这股庞大能量在未来数月内喷涌进入大气层，气象模型以 99% 的确定性预测，未来一年必将成为人类现代有气象记录以来最炎热的年份。 ```mermaid flowchart LR accTitle: 超级厄尔尼诺连锁反应链 accDescr: 展示海洋超额热量释放如何通过极端天气传导至粮食减产与全球大宗商品危机 A["海洋过剩热能释放<br/>（1100万 TWh）"] --> B["大气环流与信风异变<br/>（极热与大气河流）"] B --> C["极端区域气候灾难<br/>（干旱/热浪/暴雨）"] C --> D["农业减产与粮食禁运<br/>（巴西/澳洲/印度）"] D --> E["次生系统性冲击<br/>（电网瘫痪/大宗商品暴涨）"] ``` 随后，Friedberg 详细列举了厄尔尼诺现象将在全球引发的一系列致命次生灾害： - **北美极端天气分化**：加州、美国西南部及墨西哥湾沿岸将遭遇频繁而剧烈的“大气河流”（Atmospheric Rivers）侵袭，引发毁灭性洪涝；而美国北部至加拿大一线则将面临极度降雪匮乏与超常热浪，极易重演此前席卷加拿大的森林大火。 - **致命热穹顶（Heat Domes）**：美国内陆区域如凤凰城在 5 月份气温就已突破 106°F（约 41.1°C），超级厄尔尼诺将显著拉长热穹顶盘踞的时间，推升出历史罕见的致命高温。 - **南半球农业生命线告急**：阿根廷南部、智利与巴西将直面破纪录热浪。巴西作为全球第一大农产品出口国，一旦大宗粮食作物歉收，全球农产品供应链将遭重创；澳大利亚的小麦减产更将直接冲击高度依赖其粮食进口的印度尼西亚与菲律宾等数亿人口的生存。 - **印度季风瘫痪威胁**：当前模型显示印度季风降雨失常的概率极高，这不仅直接关乎当地 1.5 亿以此谋生的农民生计，更将把全境 15 亿人的口粮供应推向深渊。 - **宏观连锁负面溢出**：极端高温不仅会导致美国西南部用电负荷见顶、电网瘫痪，还将全面引爆全球能源及大宗商品价格，带来严重的抗通胀倒退与地缘动荡。 ### [01:06:50](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4010s) 科学角：超级厄尔尼诺引发的全球粮食危机与大宗商品动荡 大卫·弗里德伯格（David Friedberg）深入剖析了 2026 年这场历史级厄尔尼诺现象对全球地缘政治与农业供应链的连锁冲击。随着气候异常加剧，印度、菲律宾和越南等依赖粮食进口的市场可能面临严峻的供应危机甚至社会动荡。印度的处境尤为棘手：不仅农业极度依赖季风降雨，眼下还因伊朗危机及霍尔木兹海峡局势导致的氮肥断供，遭遇了季风异常与化肥短缺的双重暴击（Double Whammy）。未来一年内，南亚地区极有可能出现卡路里赤字（Calorie Deficit）并引爆严重的经济危机。弗里德伯格强调，最新气象数据已彻底证实，未来数月内将迎来一场“数据爆表（Off the charts）”的超级厄尔尼诺，大宗商品市场正对此展开极其激进的交易押注。 面对杰森·卡拉卡尼斯（Jason Calacanis）对以往厄尔尼诺是否同样引发粮荒的追问，弗里德伯格解释称，虽然巴西和澳大利亚在厄尔尼诺年份减产属于常态，但本次气候异常指数已经远远超出历史极值。作为农产品净出口国，美国本土主要需要担忧加州的风暴、极度高温、电价飙升以及山火季，但至少大西洋飓风活动会因此减弱；相比之下，高负债且高度依赖农业出口的巴西等国，极易被推入主权债务与宏观经济恶性循环的深渊。大卫·萨克斯（David Sacks）提出，全球变暖是否也让加拿大或美国北部等高纬度地区开辟出了新的耕地；弗里德伯格对此表示认同，指出加拿大如今已能大规模种植大豆，但气候变化向大气注入的过量能量极度不稳定，一旦澳大利亚或乌克兰等主要小麦产区再次遭遇极端旱灾，依赖其粮食的印尼和越南等数亿人口根本无法在短期内填补缺口。 在聊到弗里德伯格此前在节目中准确预警俄乌战争引发的粮食危机（并促成了交战方建立黑海小麦与化肥外运安全走廊）后，几位主持人开始幽默打趣。卡拉卡尼斯戏称弗里德伯格为“毁灭博士（Dr. Doom）”，调侃漫威系列 12 月即将上映的重磅电影里不是小罗伯特·唐尼（Robert Downey Jr.）主演毁灭博士，而是由弗里德伯格亲自本色出演。针对软件股暴跌，屏幕上还贴出了一张经典的“第一次绞刑？”（First time?）梗图，以此调侃做客的马克·贝尼奥夫（Marc Benioff）：在全行业哀鸿遍野的“SaaS 末日（SaaSpocalypse）”面前，贝尼奥夫气定神闲地表示自己早已身经百战，正忙着回购自家的股票。 - **南亚粮食危机隐忧**：印度面临季风受阻与霍尔木兹海峡化肥断运的双重挤压，未来一年南亚可能出现大面积卡路里赤字。 - **大宗商品市场异动**：交易员正针对创纪录的厄尔尼诺极端指数重新定价粮食与能源期货。 - **高纬度耕地扩张**：气候变暖与基因育种使加拿大等北部寒冷区域开始大面积种植大豆，但仍难以抵消全球主要产粮国减产造成的整体缺口。 ```mermaid flowchart TD accTitle: 厄尔尼诺气候异常对全球农业与经济的传导链条 accDescr: 展示2026年历史级厄尔尼诺事件如何通过气候与地缘政治双重作用引发全球粮食与金融危机。 subgraph ClimateTriggers["气候与地缘诱因"] A["2026 历史级厄尔尼诺<br/>（指数彻底爆表）"] B["中东地缘冲突<br/>（霍尔木兹海峡受阻）"] end subgraph DirectImpacts["农业与能源直接冲击"] A --> C["南亚季风降雨异常"] A --> D["澳/巴农业受灾减产"] B --> E["全球氮肥供应链中断"] end subgraph MacroCrisis["区域与全球危机传导"] C & E --> F["印度及南亚面临粮食赤字与动荡"] D --> G["巴西农业出口重挫加剧债务危机"] A --> H["美国极端高温与电价飙升"] end ``` --- ### [01:11:40](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4300s) Anthropic 整顿多层“暗箱 SPV”：一级市场乱象与 IPO 呼吁 话题转入一级市场投资的灰色地带。AI 独角兽企业 Anthropic 近期公开向泛滥的“暗箱特殊目的载体（Dark SPVs）”宣战，点名批评了部分转售股权的分销平台，并明确表示将废除这些违规交易。卡拉卡尼斯指出，许多中介机构面向普通散户（如牙医等高净值个人）兜售层层套叠的多层 SPV，不仅层层剥皮收取高达 10% 的认购费（Load-in fees），还叠加了多重业绩分成（Double carry），甚至在上一轮估值的基础上溢价 10% 转售，整个二级中介市场早已演变成缺乏监管的“西部荒野”。 查马斯·帕利哈皮蒂亚（Chamath Palihapitiya）对此表达了坚决支持，认为贝尼奥夫此前关于“成熟企业必须走向公开市场”的观点完全正确。查马斯痛斥这些为了让企业推迟上市而人为搭建的繁琐衍生结构，直言“早该一枪打死这些垃圾机制（need to get a bullet put in its head）”。他警告称，多层套叠的“SPV 套 SPV”结构埋下了巨大雷区。等到未来 SpaceX、Anthropic 或 OpenAI 正式敲钟上市、各方利益结算之时，必将引发海啸般的诉讼纠纷。当年 Uber 上市时就曾有大量投资者发现自己深陷不知情的套层 SPV 中被无情收割，而这种收取高额前端费用与双重 Carry 的暗箱操作，最终注定演变成一场金融灾难。几位投资人一致呼吁，各大科技巨头应效仿 Anthropic 严控此类投机倒把，加速理顺股权结构，早日通过 IPO 回归透明估值。 --- ### [01:13:30](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4410s) 贝尼奥夫谈 Salesforce 1-1-1 慈善模式与缅怀挚友苏珊·沃西基 在节目的尾声，卡拉卡尼斯半开玩笑地念出了一条“来自贝尼奥夫公关部门撰写的粉丝提问”：在贝尼奥夫如今集顶尖创业家与慈善家于一身的生涯阶段，他觉得自己更像史蒂夫·乔布斯（Steve Jobs），还是更像劳伦·鲍威尔·乔布斯（Laurene Powell Jobs）？面对这道幽默的调侃，贝尼奥夫认真地回顾了他在 27 年前创立 Salesforce 时做出的最重要抉择——推行知名的“1-1-1 慈善模式”（Pledge 1%）。该模式将公司 1% 的股权、1% 的产品利润以及员工 1% 的工作时间直接注入慈善基金会。 贝尼奥夫自豪地分享了这项长期承诺结出的硕果：Salesforce 员工至今已累计贡献了超过 1000 万小时的志愿服务，基金会发放的无偿捐赠款项超过 10 亿美元，同时平台还在为全球超过 5 万家非营利组织提供完全免费的软件技术支持。他大力倡导所有初创公司和成熟企业都加入 Pledge 1% 倡议，直言这是其一生中最具远见与价值的决策。 随后，话题转向了贝尼奥夫个人深厚的私人慈善事业。当卡拉卡尼斯提起前 YouTube 首席执行官苏珊·沃西基（Susan Wojcicki）因癌症不幸英年早逝、贝尼奥夫为此投入巨资支持医疗研究时，贝尼奥夫难掩内心的悲伤与触动，眼泛泪光地表示：“你快把我弄哭了……她是我最好的挚友（My best friend, my bestie）。”他动情地向听众缅怀苏珊，盛赞她是这个星球上最卓越、最伟大的人之一——不仅是一位杰出的科技领袖，更是五个孩子的伟大母亲。贝尼奥夫也表达了对沃西基家族的支持与敬意，包括她的妹妹、23andMe 创始人安·沃西基（Anne Wojcicki）以及加州大学旧金山分校（UCSF）的珍妮特·沃西基（Janet Wojcicki）教授。现场在对这位硅谷传奇女性的深切追思中流露出真挚温情。 - **Salesforce 1-1-1 模式成果**：累计超过 1000 万小时员工志愿工时、超 10 亿美元慈善捐助赠款，并免费赋能 5 万余家非营利机构。 - **Pledge 1% 倡议**：贝尼奥夫呼吁全球科技企业均应在创立之初将 1% 的股权、利润与时间回馈社会。 - **缅怀苏珊·沃西基**：贝尼奥夫深情追忆这位挚友在科技界、家庭及癌症抗争中所展现的人格魅力与非凡遗产。 ### [01:15:30](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4530s) 缅怀苏珊·沃西基（Susan Wojcicki）与节目收尾 在节目的尾声，现场流露出真挚而动情的气氛。马克·贝尼奥夫（Marc Benioff）眼含热泪地深情缅怀了前 YouTube 首席执行官苏珊·沃西基（Susan Wojcicki）。苏珊此前不幸罹患了一种极为罕见的癌症，并在几年前过早地离开了人世。贝尼奥夫动情地提到，苏珊曾长期担任 Salesforce 的董事会成员，是科技史上最伟大、最杰出的人物之一。他对苏珊的家人全力成立新基金会来攻坚并攻克这种罕见癌症深表感激，并明确表示自己和公司正全力资助这一事业，同时呼吁所有人都能提供支持。 贝尼奥夫感慨道，世界上每一个人都应该努力成为像苏珊那样的人。为了纪念她，贝尼奥夫在自己的办公室里设立了一个专门的纪念角，摆放着两人的合照，他直言自己“每一天都会深深地思念她”。他的心永远与苏珊的丈夫丹尼斯（Dennis Troper）、他们的孩子们、苏珊的姐妹们以及整个沃西基家族紧紧相连。贝尼奥夫称赞沃西基家族是硅谷不可多得的伟大基石与财富，硅谷能拥有这样的家族实属万幸。 在主持人们齐声赞叹“说得太好了”并向苏珊致以深切哀悼（“Rest in peace, Susan Wojcicki”）之后，全场在温暖的道别与标志性的夏威夷问候语“Aloha”声中结束了本期播客。 - **纪念基金会**：苏珊·沃西基的家人正集中精力成立专门基金会，致力于攻克夺去她生命的罕见癌症，贝尼奥夫已全面注资支持。 - **深厚情谊**：苏珊生前曾担任 Salesforce 董事会成员；贝尼奥夫在办公室常设两人的合影，每日缅怀这位挚友。 - **致敬传奇家族**：贝尼奥夫向苏珊的丈夫丹尼斯·特罗珀（Dennis Troper）、孩子及姐妹们致敬，称沃西基家族为硅谷最重要的基石机构之一。 ## Notable quotes > [01:12](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=72s) 若处理不好，两国将迎头相撞甚至发生冲突，使中美关系陷入极其危险的境地 — 习近平 > [05:45](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=345s) 通往中美免于冲突的缓和的最坚实路径，就是深度的‘经济相互纠缠’ — Chamath Palihapitiya > [09:58](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=598s) 我认为埃隆是全世界最伟大的推销员，这就是原因所在。 — Marc Benioff > [26:15](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1575s) 如果当年没有 Tony Robbins，今天就不会有 Salesforce。 — Marc Benioff > [31:41](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=1901s) 这可不是我经历的第一次 SaaS 末日（SaaSpocalypse）。 — Marc Benioff > [34:09](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2049s) 切忌沉溺于股价（You can't get drunk on the stock price）。 — Marc Benioff > [35:50](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2150s) 求求你们帮我卖掉这些 Token 吧！ — Chamath Palihapitiya > [44:47](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=2687s) 给他们递一盒舒洁（Kleenex）纸巾擦眼泪吧！ — Marc Benioff > [59:55](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=3595s) 这就好比我只想冲个澡，你却把整个水库的水一股脑全浇在我头上；我需要的只是一点水，而不是全部的水。 — Marc Benioff > [01:11:40](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4300s) 早该一枪打死这些垃圾机制（need to get a bullet put in its head）。 — Chamath Palihapitiya > [01:13:30](https://www.youtube.com/watch?v=jJRAvZNGUvI&t=4410s) 你快把我弄哭了……她是我最好的挚友（My best friend, my bestie）。 — Marc Benioff ## Mentioned Marc Benioff, Donald Trump, 习近平, Jason Calacanis, David Sacks, David Friedberg, Chamath Palihapitiya, Elon Musk, Jensen Huang, Kelly Ortberg, Brian Sikes, Tim Cook, Tony Robbins, Graham Allison, Jack Dorsey, Mira Murati, Jony Ive, Susan Wojcicki, Anne Wojcicki, Janet Wojcicki, Dennis Troper, Salesforce, OpenAI, Apple, Anthropic, Thinking Machines, NVIDIA, Tesla, Boeing, Alibaba, Cargill, BYD, Qualcomm, Informatica, Slack, Google, Perplexity, Mistral, SpaceX, HubSpot, ServiceNow, Workday, Visa, Mastercard, Andreessen Horowitz (a16z), Blackstone, ChatGPT, Agentforce, Claude, Gemini, Grok, Sora, Whisper, Polymarket ## Video description (0:00) Salesforce CEO Marc Benioff joins the show! (1:14) Trump-Xi summit, doing business in China as a US company, impact on Americans and the midterms (18:46) Taiwan, chips, AI models, and peace through trade (31:41) AI's impact on software: What SaaS thrives, what SaaS dies? (47:26) OpenAI is considering suing Apple over failed ChatGPT integration (56:54) Thinking Machines releases real-time model, future of consumer AI, multi-sensory models (1:02:24) Science Corner: Impacts of a historically strong El Nino in 2026 (1:11:40) Anthropic goes after "Dark SPVs" Follow Marc Benioff: https://x.com/Benioff Save Scooter the Dog: https://animalcare.lacounty.gov https://www.instagram.com/reels/DYJZFn0R6oY Apply for Summit 2026: https://allin.com/events Follow the besties: https://x.com/chamath https://x.com/Jason https://x.com/DavidSacks https://x.com/friedberg Follow on X: https://x.com/theallinpod Follow on Instagram: https://www.instagram.com/theallinpod Follow on TikTok: https://www.tiktok.com/@theallinpod Follow on LinkedIn: https://www.linkedin.com/company/allinpod Intro Music Credit: https://rb.gy/tppkzl https://x.com/yung_spielburg Intro Video Credit: https://x.com/TheZachEffect Referenced in the show: https://polymarket.com/event/will-china-invade-taiwan-before-2027 https://polymarket.com/event/will-china-invade-taiwan-by-december-31-2027 https://www.nytimes.com/2026/05/14/world/asia/china-xi-trump-taiwan-warning.html https://www.nytimes.com/2026/05/13/technology/andreessen-horowitz-politics.html https://www.instagram.com/reels/DYJZFn0R6oY https://finance.yahoo.com/sectors/technology/articles/openai-launches-4-billion-ai-134916653.html https://www.youtube.com/watch?v=KO53gwuqZUQ https://www.bloomberg.com/news/articles/2026-05-14/openai-apple-partnership-frays-setting-up-possible-legal-fight https://siliconangle.com/2026/05/11/thinking-machines-drops-new-highly-responsive-model-designed-humanlike-interactions-real-time/ https://x.com/thinkymachines/status/2053938892152435174 https://www.coindesk.com/markets/2026/05/12/anthropic-fights-unauthorized-stock-exposure-as-token-markets-imply-trillion-dollar-valuation https://support.claude.com/en/articles/13704655-unauthorized-anthropic-stock-sales-and-investment-scams #allin #tech #news [Source](https://www.youtube.com/watch?v=jJRAvZNGUvI)

[1:12] 习近平: 若处理不好，两国将迎头相撞甚至发生冲突，使中美关系陷入极其危险的境地 [5:45] Chamath Palihapitiya: 通往中美免于冲突的缓和的最坚实路径，就是深度的‘经济相互纠缠’ [9:58] Marc Benioff: 我认为埃隆是全世界最伟大的推销员，这就是原因所在。 [26:15] Marc Benioff: 如果当年没有 Tony Robbins，今天就不会有 Salesforce。 [31:41] Marc Benioff: 这可不是我经历的第一次 SaaS 末日（SaaSpocalypse）。 [34:09] Marc Benioff: 切忌沉溺于股价（You can't get drunk on the stock price）。 [35:50] Chamath Palihapitiya: 求求你们帮我卖掉这些 Token 吧！ [44:47] Marc Benioff: 给他们递一盒舒洁（Kleenex）纸巾擦眼泪吧！ [59:55] Marc Benioff: 这就好比我只想冲个澡，你却把整个水库的水一股脑全浇在我头上；我需要的只是一点水，而不是全部的水。 [71:40] Chamath Palihapitiya: 早该一枪打死这些垃圾机制（need to get a bullet put in its head）。 [73:30] Marc Benioff: 你快把我弄哭了……她是我最好的挚友（My best friend, my bestie）。

[1:24] Xi Jinping: If handled poorly, the two countries will collide or even clash, putting the entire US-China relationship in an extremely dangerous situation. [20:09] Marc Benioff: I think that the Chinese models are as competitive as these U.S. models, and they've learned to make these models without having the highest end chips... We probably should just sell whatever we can at this point. [26:01] Marc Benioff: If it wasn't for Tony Robbins, I don't think there would be a Salesforce. [34:31] Marc Benioff: You can't get drunk on the stock price. If you are focused on today all the time and that is how you're getting your emotional state, it's not going to work for you. You have to find a different anchor. [71:40] Chamath Palihapitiya: All of these tickytacky mechanisms that people can use to stay private longer need to get a bullet put in its head. And these SPVs are the worst—the layered SPVs on SPVs on SPVs... You're getting double carry, you paid 10% on the load-in fee, and you paid a price 10% over what the last round was. It's a recipe for disaster. [73:30] Marc Benioff: She was one of the great people of all time... Everyone should try to be a Susan. She was one of the absolute greats. I have an altar in my office with photos of her and me, and I think of her every single day.

[0:00] All right, everybody. Welcome back to [0:01] the number one podcast in the world, the [0:03] allin podcast. Episode 273. [0:06] It's a huge week. Saxs is our show. We [0:09] got to get Mark Beni off, CEO of [0:11] Salesforce. [0:11] >> I'm the only one left here. You're the [0:13] only one left. That's why that's how I [0:14] made it on. [0:15] >> No software CEOs, right, to China. [0:17] >> There's no software uh CEOs in China. [0:20] >> Interesting. But did you get the invite? [0:22] What's your what's your status with this [0:24] administration because you were [0:25] obviously very famous for, you know, [0:27] being a Democrat on the left? Um, and [0:30] >> I'm not a Democrat on the left. I never [0:32] have been. [0:33] >> What are you talking about? You donated [0:36] to all these folks and now you're kind [0:38] of in Trump's camp. What is it? [0:40] >> Listen, um, uh, the number one thing is, [0:43] hey, I'm here to support the country. [0:45] >> That's what I do. [0:46] >> Okay. [0:46] >> Yeah. [0:47] >> So, you're right in the middle. You have [0:49] uh an allegiance to America. [0:51] >> I hope so. [0:52] >> Yeah. [0:52] >> I'm not a Democrat or Republican. [0:54] >> You're an independent. [0:55] >> I'm an American. [0:56] >> Okay. I like it. Ben off got the invite [0:58] to the two hottest tickets. He got the [1:00] invite to Windsor Castle in Tales and [1:02] then he got the invite uh when Prince [1:04] Charles came here too. I saw he was in [1:06] Tales then too. [1:07] >> And also we were at that Saudi dinner [1:08] too. Weren't you there? [1:10] >> I did not go to that one. All right. [1:12] Well, we have a big docket here. The [1:14] Trumpi summit has begun. That is the [1:18] number one story right now after a [1:19] two-month delay because of the war in [1:21] Iran. This is the first visit to China [1:25] since 2017. [1:27] Seventh face-to-face meeting for Trump [1:29] and President Xi. There are 12 hours, 15 [1:32] hours ahead of us. Today, Thursday was [1:35] officially day one. Here's what [1:37] happened. China agreed that the straight [1:40] of Hormuz should remain open with no [1:44] military commitment and that Iran should [1:46] not have a nuclear weapon. So, we're in [1:48] sync on that. On Taiwan, this is a [1:51] little spicy. She warned that quote if [1:54] handled poorly the two countries will [1:55] collide or even clash putting the entire [1:58] USChina relationship in an extremely [2:00] dangerous situation. Poly market says [2:03] Taiwan is safe for 2026 only 6% chance [2:07] China invades this year on 23 million in [2:11] volume which is a lot of volume 17% [2:13] chance roughly triple [2:16] that something happens by the end of [2:18] 2027. There's been a lot of debate. Uh [2:21] some people say it will happen after [2:23] Trump is out of office. Some people [2:25] think it's going to happen while he's in [2:27] office on trade. She committed to buy [2:30] more soybeans, US oil, [2:33] LNG, and 200 Boeing jets. She said the [2:38] talks laid out a vision for [2:39] constructive, strategic, and stable [2:41] ties. Uh Trump was effusive in his [2:44] praise uh for his friendship, giving the [2:47] disclaimer that people don't like it [2:48] when he praises she. Let's stop here and [2:52] have a bit of a discussion. Freedberg, [2:55] you uh have been a bit obsessed with [2:58] this relationship and the bipolar [3:02] nature of it. What are your thoughts [3:04] here? And what is the goal for Trump? [3:06] What is winning for Trump coming out of [3:07] this? What is winning for she? [3:11] I mean, she made comments in his uh [3:14] opening remarks that it would be ideal [3:19] if the United States and China could [3:21] avoid the Thusidities trap, which as [3:24] you'll recall, we talked about with [3:26] Graeme Allison and we've talked about [3:28] several times over the last few years [3:30] that as a a rising power meets a kind of [3:33] declining power, there's always some [3:36] moment where you end up in this kind of [3:38] state of conflict. And the question is, [3:39] can the US and China avoid conflict and [3:42] find a path to cooperation, which has [3:44] happened a few times in history when we [3:46] found ourselves in this moment, but [3:47] doesn't often happen. And I think the [3:50] way to kind of think about the [3:51] opportunity is if you're in a resource [3:53] expansive world where you are increasing [3:57] productivity and increasing production [3:58] globally at an accelerating pace, [4:01] perhaps there's less of a reason to have [4:03] conflict. Perhaps there's a way that uh [4:06] both societies, both countries can [4:09] increase the quality of life for their [4:11] people without taking it away from the [4:14] other side. And in a world where you're [4:16] more resource constrained or resource [4:18] static, that becomes less possible. You [4:21] have to fight and grab land and grab [4:23] territory and grab resources from the [4:25] other side. [4:26] And it seems like in this moment when we [4:28] are seeing these extraordinary [4:30] technology shifts unlocked by AI and [4:32] automation and biotech and all of these [4:36] kind of moments of which could be true [4:39] abundance ahead of us. It seems like the [4:41] perfect moment to say hey maybe the [4:43] world can be more multipolar. It doesn't [4:45] need to be uniolar. It doesn't need to [4:46] be bipolar but everyone can participate [4:49] in the expansion of the pie. And I think [4:52] that that's kind of the idealistic way [4:53] to look at the opportunity in front of [4:55] the administration and Chinese [4:57] leadership right now. Is there a way to [4:59] kind of look at the next 30 years and [5:02] ask the question, how do we all share [5:05] proportionally in growing the pie and [5:07] not find ourselves in a moment of [5:08] conflict where we assume the pie is [5:10] static? That's I think hopefully the [5:11] message that comes out of this from an [5:13] administration political perspective. It [5:14] just seems like this is going to be a [5:16] major coup for the president if he can [5:17] walk away with a series of trade deals [5:20] with China that increase job security, [5:24] increase prosperity, increase income [5:25] levels, increase investment in America [5:28] and American jobs. It would be a huge [5:30] win. And I think that's the tactical [5:32] stuff that people will be truly looking [5:34] for. And I think the big strategic [5:35] question which I care most about which [5:37] is avoiding conflict. Is there a way we [5:39] can chart an economic and cooperative [5:41] path that doesn't involve eating each [5:43] other and involves sharing in a bigger [5:45] pie? [5:45] >> Chimoth, why is Trump bringing all of [5:48] these CEOs with him and what does [5:50] success look like uh coming out of this [5:53] meeting? [5:53] >> I think that you find a resolution and a [5:56] path forward with China through economic [5:58] cooperation. It's the largest consumer [6:00] economy in the world. [6:03] It's still largely closed off. And so I [6:06] think what this is was about bringing [6:08] some amount of financial firepower with [6:11] them so that they could start to [6:13] penetrate that market. Planes, cars, [6:16] chips, you know, very much hard [6:17] equipment type stuff. I've said this [6:20] before, but I think the Chinese are very [6:22] much a top- down confusion societal [6:24] philosophy, [6:26] whereas Americans are much more bottoms [6:28] up kind of an individualist [6:30] construction. [6:31] The fact that we're so different [6:33] actually gives us a lot of room to find [6:34] cooperation and not find conflict. And I [6:38] think at the center of that is economic [6:40] cooperation. I had a friend call me this [6:42] morning, very prominent Democrat. And [6:45] you know, he was the one that called out [6:47] the media, which I was surprised by, and [6:49] he said, you know, the media gets it all [6:50] wrong because they're talking about why [6:52] is he bringing these CEOs? And instead [6:54] the reality is that the simplest and [6:57] shest path to a no conflict dant with [7:00] China is economic entanglement and it [7:03] has to be birectional because really for [7:05] so many years it's been one way where [7:06] they're sending us the cheap goods that [7:09] they've wanted. [7:11] So I think strategically it's good and I [7:13] think you know I'll just take a slightly [7:14] different take on what Freick said. I [7:16] think behind closed doors they're [7:17] probably just figuring out how to divide [7:18] the pie. [7:20] >> Uh unpack that for a second. divide [7:22] which pie the the globe the economy the [7:25] the western hemisphere versus [7:27] >> look China China still needs a lot of [7:29] energy they also need a lot of critical [7:32] technologies that America provides and [7:34] so I think that there's a trade there [7:35] and the quit proquo is there are certain [7:38] regions of the world where we want them [7:39] to deescalate their participation in [7:43] central South America being probably the [7:45] most important and then the second is [7:46] probably in the Middle East and then [7:47] maybe the third is just to find some [7:49] reasonable view of the Asia pack region [7:51] together and I think there's a trade [7:53] there to be done and in that you kind of [7:55] start to carve up the world into a [7:57] pretty easily identifiable bipolar [8:00] construction. [8:01] They also uh Mark need customers uh to [8:05] buy their goods and we've had a bit of a [8:09] uh mixed signals there. Tariffs [8:11] obviously last year sticking it to China [8:13] in a pretty hard way. the decoupling [8:15] after COVID. Hey, we need to be [8:18] independent energy, PPE, drugs, chips [8:22] from China. Now, we're kind of saying, [8:25] hey, let's come to the table and build [8:28] connective tissue. So, sort that out for [8:30] us on a business level. Should the two [8:33] countries be deeply entwined? And then [8:35] how do we manage having too much [8:37] dependency on China? [8:40] Well, you heard it today from President [8:42] Xi when he said he wants a wider door. [8:45] And that means that he wants the door of [8:48] business to open even wider between the [8:51] two countries. We have our absolute best [8:54] salespeople there. Not just our [8:56] president, who has to be one of the best [8:58] salespeople I've ever met, but also [9:00] we've got Elon, we've got Jensen selling [9:03] chips, [9:04] >> Kelly Ortberg selling planes, [9:06] >> Kelly selling planes. He's already sold [9:08] 200. He wants to sell financial [9:09] services. [9:10] >> Brian is selling soybeans. Brian Sykes, [9:13] the CEO of Cargle. You know, Cargle is [9:15] the world's largest private company [9:17] based in Minneapolis and he is going to [9:20] come away selling a lot of soybeans. So, [9:22] we have a you go through the list of the [9:24] CEOs. Each one is our best salesperson [9:27] in this category and they're going to [9:29] come back with orders and it's going to [9:31] be amazing. And by the way, I think you [9:33] said it very well. you know, this idea [9:35] of business is this kind of greatest [9:37] platform for change. Um, they are [9:40] working together like never before. [9:42] Don't forget, this isn't Trump's and [9:44] she's first meeting. These guys really [9:46] like each other. They are very connected [9:49] together. You can see they're smiling [9:50] with each other. They're very they have [9:52] a sales orientation. So, I expect a lot [9:55] of order books to come back. [9:56] >> I have a question for you. Are you [9:58] allowed to sell software in China? We uh [10:01] well with the right the Chinese have a [10:03] lot of data residency laws. So uh you [10:06] know we we do it is we don't have any [10:08] offices or employees in China. We never [10:10] have. The only time we end up with any [10:12] residue in China is when we buy a [10:15] company and they have a presence over [10:17] there and then we have to devest from [10:19] it. We have an exclusive partnership [10:21] with Alibaba. Everything just goes [10:23] through them and we we don't do anything [10:26] in China. [10:26] >> Is it significant growing? Is it what [10:29] percentage of your revenue is it? [10:30] >> Yeah, it's it's great. You know, we have [10:32] it's a great partnership. Key customers [10:34] of Salesforce like Louis Vuitton, you [10:36] know, or Laura Piana, I know Chimoth [10:38] loves Laura Piana. [10:40] >> Yeah. [10:40] >> And you know, Laura Piana sells a lot of [10:42] sweaters in China and then, you know, in [10:46] all other stores all over the world, [10:47] they use Salesforce, but in China, it's [10:51] Salesforce on Alibaba. [10:52] >> It's still your code, but then it [10:54] resides in their servers for data [10:56] residency stuff. their data residency, [10:59] their their partnership. It's a totally [11:02] unique relationship. It's the only place [11:03] in the world that we do such a thing. [11:05] >> Yeah. And it has to be all the data has [11:07] to be kept in China. All the data has to [11:09] be accessible by the CCP. [11:12] >> That's why it's so extraordinary. By the [11:13] way, what Elon has, you know, Elon has [11:16] Tesla in China, no partnership. He's the [11:19] only one. He has a phenomenal [11:21] relationship with Shei. When you look at [11:24] she and um Elon together, notice how [11:27] differential and respectful they are of [11:29] each other. And here is these AI cars [11:33] with all these cameras, Americanmade, [11:36] you know, um Chinese factories driving [11:39] around China. That is pretty incredible. [11:42] >> How do you think he pulled that off? Is [11:45] that just a one-off exception? [11:46] >> I think Elon's the greatest salesman in [11:48] the world. That's how [11:50] >> I mean the cynical take on a chim would [11:53] be they wanted him there and they wanted [11:54] to study the company and the innovation [11:56] and they have now become the biggest [11:58] competitor uh with their EVs and so it [12:01] is a pretty coopetition I guess it would [12:06] be the cynical way to do [12:07] >> I want one of those new BYD golf you [12:10] know uh uh what do you call those with [12:13] the doors that flip flip up [12:14] >> Gwing [12:15] >> Gwing I want one of those Gwing BYDs [12:18] they have some cool cars have you seen [12:19] those. [12:20] >> They're cool, but they're not safe. [12:21] >> They're incredible. [12:22] >> They're talking in this in this they [12:24] they said they're talking about bringing [12:25] some of those cars to the US. That would [12:28] be incredible. [12:29] >> Uh would hollow out the entire [12:31] automotive industry here if we had [12:32] $20,000, $30,000 cars from China. It [12:36] would be the end of the US automotive [12:37] industry overnight. And uh Tim Cook's [12:40] there. He has to store all his data [12:42] there. That's been a pretty [12:44] controversial issue for him because [12:47] Freedberg, if anybody has private [12:50] information or is a dissident, Tim Cook [12:52] has to give that data over to the [12:55] Chinese Communist Party. How do you [12:58] think this plays in America visav the [13:00] midterms? Freedberg, Americans feel [13:03] ignored, inflation over 3% and they look [13:08] at the Iran war as a betrayal by [13:11] President Trump and they look at China [13:14] as not material to their well-being. So [13:19] when you look at Trump being on a [13:21] six-month timeline for the midterms and [13:24] he's only got two more years in office [13:25] after that, you got she who's working [13:28] off of a hundredyear playbook. How does [13:30] that dynamic play out with Trump and the [13:33] midterms and his core constituency, [13:36] MAGA, which is now evolving into America [13:39] first, America only? And this kind of is [13:41] is rubbing that group of people the [13:43] wrong way. [13:45] >> I think the second and third order [13:47] effects of anything that comes out of [13:48] China is what's going to be most [13:49] consequential to the average voter, [13:51] which is job security and income [13:53] security and income growth. And then [13:55] cost of living impact. The cost of [13:57] living impact would be if there can be [14:00] some deflationary effect from a trade [14:01] deal that makes access to goods lower [14:04] price for Americans. And then anything [14:06] that America is exporting, there's [14:08] increased demand. Like there's a [14:09] follow-on effect to the soybean export [14:12] market. [14:14] There's a follow-on effect obviously to [14:15] the oil export market. There's a [14:18] follow-on effect to for example [14:22] technology and software being exported. [14:24] So that increases economic productivity [14:26] in the US and increases job security [14:29] income. And then the other side of the [14:30] trade deal is can we get stuff that [14:32] Americans really care about and make it [14:34] cheaper. Can people now buy stuff at the [14:36] store that drops in price by 30 40% [14:38] where we've kind of tariffed and reduce [14:40] trade that increases the price of [14:42] things. So if those deals kind of get [14:44] worked out, people will see things get [14:45] cheaper at the store and they'll feel [14:47] good about kind of their income [14:49] security. And I think that could be [14:50] positive. I think that's what's [14:52] ultimately going to make people make the [14:54] blue red decision when they go to the [14:55] polls in November. [14:56] >> That's the schizophrenic nature of this [14:58] chimoth. We have Americans who very much [15:02] want cheap goods from China and then we [15:04] have last year all these tariffs being [15:06] put on and we have this great [15:07] decoupling. So, how do you handicap the [15:11] midterms, Trump's timeline and she's [15:14] timeline and what their goals are and [15:17] then I guess we'll get to Taiwan after [15:19] that. [15:20] >> I don't know. No, it's I think it's a [15:21] little bit above my pay grade. But what [15:22] I will say is that the midterms I think [15:25] are probably going to swing more based [15:27] on these recent jerrymandering rulings [15:29] from the Supreme Court and what happened [15:31] in the Virginia Supreme Court and what's [15:34] going to happen in the state [15:35] legislatures of these red states and [15:38] some of these blue states more than [15:41] what's going to come out of this summit [15:44] with with President Xi. So let me just [15:46] put that over here. And I think that [15:48] money is getting organized. There was an [15:50] article in the New York Times this week [15:52] which really surprised me, but the [15:54] largest donor in the in this election [15:56] cycle is Andre and Horowitz. [15:58] >> Yeah. What is that about? Yeah. [16:00] >> I think it's because they're they're [16:01] trying to establish themselves as part [16:03] of the financial firmament of America, [16:05] which I think strategically as a [16:07] business makes a lot of sense. Like [16:08] they're in the AUM business, right? So [16:11] they can't stop at 100 billion of AUM. [16:13] They're marching towards a trillion. [16:16] they're going to be the next Blackstone, [16:17] which I think is an incredible feat and [16:21] they're building a juggernaut of a [16:22] business. But I think they also realize [16:25] that politics is now a permanent part of [16:27] that if they're going to be investing in [16:30] higher quantum more geopolitically [16:33] influenced parts of the capital cycle [16:34] and parts of the economy. So I think the [16:37] midterms are largely that the money [16:39] that's going to be raised plus the [16:42] gerrymandering that will get done and [16:44] undone over the next few months. Again I [16:46] think I think the the Chinese and the [16:48] and the Americans have a very strong [16:50] incentive to kind of divide up the [16:52] world. Like I think the future at this [16:54] point you have these critical resource [16:56] inputs. China has a strangle hold. You [16:58] have energy and [17:01] intelligence AI where America has a [17:05] strangle hold on one and an effective [17:07] strangle hold on the other which is [17:08] energy and I think there's a trade there [17:11] to be done and as long as you can [17:13] negotiate what a reasonable give and [17:16] take is I think they're just going to [17:18] find common ground like does China [17:20] really need to be in Chile and Venezuela [17:23] and Panama? Probably not. Does America [17:25] have to have an incredibly strong point [17:27] of view about the straight of Malaa? [17:30] Probably not. And so there's probably a [17:32] trade to be done so that we can get [17:34] their rare earths, they can get some [17:36] more oil, everybody can be happy, and we [17:38] can all find abundance. [17:40] >> Yeah. The the key issue here is China is [17:42] and most people are, you know, [17:44] chinophobic, but China has serious [17:47] systematic issues with population, with [17:50] their real estate, with their GDP. All [17:52] of this is in decline. [17:54] And I think she most of all is concerned [17:58] about another Tianaan Square if he [18:00] doesn't get more people working. They've [18:02] had the largest growth of any middle [18:04] class [18:06] you know in history. They have the [18:07] largest middle class in the world. I [18:08] think uh you were sort of referencing [18:10] that Mark with luxury goods and luxury [18:12] goods providers are going there but it's [18:15] tenuous uh [18:16] >> and that's the number one by the way [18:18] that is the number one point of focus of [18:20] President Xi. President Xi's focus has [18:24] been getting 500 million people from [18:26] poverty into the middle class. [18:28] >> It's not exactly how we think about the [18:30] world in the United States. That's the [18:33] main focus of the political leader and [18:34] and he's achieved that. That that has [18:37] been an incredible accomplishment [18:38] >> and and but it's at risk right now [18:40] because their GDP is falling. They need [18:42] their factories to operate. They're [18:44] being undercut by other regions. And [18:46] then of course Taiwan is what they want [18:49] uh most of all and supremacy in you know [18:52] their part of the world. [18:53] >> I'm not convinced of that actually. [18:55] >> I think what they want Oh, I mean I [18:57] think we just kind of said it and to [18:59] summarize they they want to have you [19:02] know economic success. They want to [19:04] continue to grow their middle class. [19:06] They want to continue to have a healthy [19:07] economy. They want more trade. I thought [19:10] it was interesting that they're talking [19:12] about Trump is talking about not only [19:13] having the board of peace now he wants a [19:15] board of trade. That was a discussion [19:17] we've never heard before today. Then uh [19:20] we have she saying I want the wider [19:22] door. Then we have all these folks there [19:25] are selling their products. Two people [19:26] we didn't mention the co Visa and the [19:29] CEO of Mastercard are on the trip. Now [19:31] why is that? Because if you look at you [19:34] know commerce in China it is dominated [19:37] by companies like Alibaba and and these [19:39] organization who use these super apps. [19:41] It's not really all the way open for a [19:44] Visa and a Mastercard. That would be [19:46] incredible for those executives. So in [19:48] the same way that we want to sell [19:50] airplanes, we want to sell chips, [19:52] Nvidia, the Qualcomm coristiano is also [19:54] there, we want to sell soybeans, we want [19:57] to sell payment services, we want to [19:58] sell everything. The more economic [20:01] collaboration and cooperation you can [20:02] get between the two countries, the more [20:04] peace you're going to have. I think 100% [20:08] I agree. [20:09] >> Three, three hard questions, Mark. Uh, [20:10] should we be selling them the latest [20:12] chips? Yes or no? In your mind, [20:16] >> I think it's irrelevant at this point. I [20:17] think that the Chinese models are as [20:19] competitive as these US models and [20:22] they've learned to make these models [20:23] without having the highest end chips. I [20:26] think the highest end chips is kind of [20:27] more of a ego gratification for us. We [20:30] have Blackwell, we have this, we have [20:31] that. But when you look at their models, [20:34] their models are excellent. They fast [20:36] follow us. You know, the best models we [20:38] had ago, they have now. [20:40] >> Should they should we just sell it to [20:42] them since they're figuring it out [20:43] anyway, Mark? [20:45] >> We probably should just sell whatever we [20:46] can at this point. Got it. [20:48] >> And uh I think that as I said, I think [20:51] the more economic cooperation and [20:53] collaboration is better. And if you want [20:54] peace, I think that this [20:57] >> I think Taiwan, it's kind of a circle [21:00] back to that. Yeah. [21:01] >> Freeberg, uh, your take on chips. Should [21:04] we sell them latest chips? And then [21:05] Chimat, I want to get the answer to that [21:07] question from each of you. [21:08] >> I think it's inevitable that they get [21:09] there. I, by the way, I do think this is [21:11] one of the key aspects of this deal. [21:13] I've said this before, but maybe Taiwan [21:16] becomes less relevant to the US and to [21:18] China as both China and the US um, [21:20] mainland FAB. So, as we build out our [21:23] own [21:24] >> manufacturing capacity here in the US, [21:26] and I know we've struggled and there are [21:28] fits and starts and issues with it and [21:29] whatnot, but the the TSMC facility in [21:32] Arizona is running, they're printing, [21:35] and if it works and they figure out how [21:37] to scale it. There'll be other [21:38] facilities in the US and then China's [21:40] mainlanding with Huawei standing up a [21:42] lot of facilities. They've got a multi- [21:44] I think I talked about this on one of [21:45] the shows, a multi-deion [21:47] government investment in ensuring [21:49] competitiveness not just with fabs but [21:52] also with semiconductor manufacturing [21:54] equipment so that they don't get cut off [21:56] on the supply chain with companies like [21:57] ASML blocking sales into mainland. So as [22:01] they build out their own semiconductor [22:03] capital equipment systems and their own [22:06] fabs, does Taiwan really matter? Is [22:08] there really that much of a security [22:10] risk to the United States and to China [22:12] in that sense? And perhaps what [22:13] everyone's focused on, which is this [22:15] pivot point around Taiwan. Maybe it's [22:17] not that the US is selling Taiwan down [22:19] the river, but it's just that no one [22:21] gives a anymore. So, you know, [22:23] there may be this kind of cultural [22:24] moment for China where they want to at [22:26] some point say, "Hey, by 2040," which I [22:28] think is what I've heard, they want to [22:29] be able to kind of, you know, bring [22:31] Taiwan fully into the sphere. maybe the [22:34] US shouldn't care that much at that [22:36] point from a economic and security [22:38] perspective. So I think that's that's [22:40] one of the things that's happening that [22:41] >> so you're in the camp of sell the chips [22:43] to them and it will demotivate the TSMC [22:49] or the capture of those fast. [22:51] >> Put it this way. I think the more the [22:53] global productivity index can climb, the [22:57] better off all humans will be. So the [23:00] question is, do you really want the west [23:02] to see their productivity index climb [23:04] and then you're effectively forcing [23:06] conflict and forcing issues with the [23:08] east? Or if we all grow our productivity [23:11] index, we all make more stuff, everyone [23:13] benefits, everyone has better jobs, [23:15] everyone makes more money, everyone has [23:16] access to more stuff, the cost of goods [23:18] comes down for everyone, doesn't that [23:20] make the world a safer and more secure [23:21] place? [23:22] >> Sure. So I think this idea that like [23:23] technology should be held to one uh side [23:27] of the world and not given to the other [23:28] side of the world is inevitably going to [23:30] lead to conflict or increase the [23:32] probability of conflict. Whereas if you [23:34] let technology diffuse and proliferate, [23:37] everyone benefits and conflict indices [23:39] go down. [23:39] >> It's commoditized, you know, and [23:42] everybody has access to it. So then it [23:44] it becomes [23:44] >> you give everyone a higher standard of [23:46] living and you give everyone's economy a [23:48] boost and you give everyone less reason [23:49] to fight with each other. Since you went [23:50] there with Taiwan, should we take our [23:53] arm sales off of the table, Freedberg? [23:56] And should we ask China to not sell arms [23:58] to Iran? [24:01] >> Yes, we should ask China to not sell [24:02] arms to Iran. [24:04] >> Yeah. [24:04] >> Okay. [24:05] >> I don't know what there is. What is [24:06] there to talk about there? Because I [24:08] think what she wants is for us to not [24:10] sell arms to Taiwan. So, would you make [24:14] that trade, I guess, is what I'm saying. [24:15] >> Oh, yeah. I mean, that's a good [24:17] question. And I think in that context [24:19] that's that's a nuance one. I don't [24:20] >> that's why I bring it up because that [24:22] seems to be [24:23] >> generally you're right. Yeah. I mean [24:24] generally that's probably a a trade that [24:28] you know that should be kind of trade do [24:30] the deal. [24:30] >> So you say make the trade [24:32] >> chimath. I mean I kind of know [24:34] >> we're 18 months from Taiwan not being an [24:38] important moment of conversation the way [24:41] it is today. Why 18 months? because we [24:43] are at a point where we're probably 1 [24:45] to2 nanometers away from being able to [24:48] do what we need Taiwan to strategically [24:51] do for us. And so as we scale up our [24:54] chip fabs, as we get more capacity and [24:57] interestingly there are these orthogonal [24:59] technologies being developed. I don't [25:01] know if you guys saw but neuralink was [25:03] showcasing now a machine that is [25:06] literally operating at the [25:09] almost nanometer scale to do the brain [25:11] operations for the implantation all [25:13] automatically. So when when you when you [25:15] have the dexterity and the capability [25:17] mechanically to make these things, the [25:19] real reason then is a very different one [25:21] than what it is today. Today it's [25:22] economic and if you take that off the [25:24] table, I think we'll have a very [25:25] different attitude to Taiwan. That's [25:26] number one. Number two, sell the chips. [25:28] >> And the reason we should sell the chips [25:30] is we want Nvidia to win. We do not want [25:32] to give enough oxygen for Huawei to then [25:35] all of a sudden emerge and have a [25:37] version of a chip that works. And what [25:38] Mark said is totally right. These models [25:41] are catching up. They're almost at the [25:43] same pace. So the more important thing, [25:45] Jason, is probably we should all agree [25:47] is just like let's have a reasonable KYC [25:50] for how the models get used so that you [25:52] know somebody in a basement doesn't cook [25:54] up some boweapon. Meanwhile, sell the [25:57] chips. Proliferate. Let Jensen win. I'd [26:00] rather he win than Huawei win. [26:01] >> Mark, I'm going to give you the last [26:02] word, but I'm going to force you to [26:03] answer a hard question. China sets up a [26:06] blockade around Taiwan and they decide [26:08] they're taking it. Should the US defend [26:09] it? Yes or no? I I've said this for [26:12] years. I I don't agree with Neil [26:13] Ferguson on that point. I think that [26:15] this is a nonsense issue. I think China [26:17] and Taiwan will reconcile. I think that [26:20] uh Well, first of all, I do want to [26:22] schedule though Chimath for that neural [26:24] link. Um next Tuesday 5:00 we can drill [26:27] your brain. Chimoth, can I get you in? [26:30] >> What are you trying to do? Put some [26:31] empathy into Chimoth. I talked to Elon. [26:33] He said it doesn't. You just drill right [26:35] in. There's a and boom. And then well, [26:38] you can get hooked up. It is the [26:40] craziest thing. [26:41] >> I know. I got you scheduled at 5:00 on [26:44] Tuesday tomorrow. Hold your hand if [26:46] you're too nervous. [26:47] >> And by the way, [26:50] finally had a hard literally like [26:56] for a even better show. [26:57] >> Abs. It would be incredible. Yes. And uh [27:02] I would laugh and I would giggle. We [27:05] could get into show tunes here. Free [27:07] break. I know you're um Yeah. not big on [27:09] on the show too. [27:10] >> If I only had a heart. [27:11] >> If I only had a brain. [27:13] >> A heart. A brain. And what's the third [27:15] one? [27:16] >> If I only had a neural link. [27:19] >> You got to give you courage. REEDBERG. [27:25] >> OKAY, move on. [27:26] >> All right, let's move on here. [27:27] >> Your By the way, I took kids. [27:30] >> Tim stayed home. I took his kids to the [27:32] Wizard of Oz at the sphere. It was [27:34] incredible. If you haven't seen it, [27:37] >> yeah. do something enriching with my [27:40] kids or hit the crafts table. [27:43] >> No, I I'll be honest with you. I get a [27:45] little freaked out in large [27:50] >> like groups of people. [27:51] >> Yeah, I get I get a No, just just like [27:53] when there's a lot of people, I get very [27:54] nervous. [27:56] >> Really? [27:56] >> Is this because of your newfound [27:58] celebrity? Is this is that the issue or [28:00] just always like No, I don't even want [28:01] to go there. I just have these bad [28:03] thoughts of like, oh my god, what if [28:04] something happens? How do we get out and [28:06] like [28:06] >> terrorist attack? [28:08] panic disorder. I don't know. Yeah, [28:10] exactly. You have panic disorder. [28:12] >> Panic disorder. [28:12] >> I don't I don't like these large spaces [28:14] with lots of people. I get a little [28:15] winded up. [28:16] >> Anyway, your kids had a great time. [28:18] Great show. [28:19] >> Tony Robbins lives very near you, [28:20] Shimoth. Don't you think we can get that [28:22] to happen? [28:22] >> Yeah. Let's get Tony Robbins in there to [28:24] clear that blockage. You're a Tony [28:26] Robbins guy, Mark. Why are you so into [28:27] Tony Robbins? [28:28] >> I love Tony Robbins. I had him at my [28:30] conference this week. It was incredible. [28:31] Actually, he did an incredible [28:33] presentation. [28:33] >> Did he unlock for you your success? He [28:36] claims to that he has you on the roster [28:38] of $50,000 a year CEOs? [28:40] >> I am the number one fan of Tony Robbins. [28:42] So I mean [28:44] >> what percentage of your success is [28:45] attributed to Tony's mentorship? [28:47] >> Hey, if it wasn't for Tony Robbins, I [28:49] don't think there would be a sales [28:50] force. So there you go. [28:51] >> Can you actually no jokes aside, can you [28:53] tell us like what like what like what [28:56] were those big kind of blockages and how [28:58] did you how did he help you or how did [28:59] you work through it? [29:00] >> Just focusing on hey the questions you [29:02] know the quality of your questions is [29:04] the quality of your life. It's that [29:05] insight and just are you asking yourself [29:07] the right questions. What do you want? [29:08] What's important to you? How are you [29:10] getting it? What is preventing you from [29:12] having it? How will you know that you [29:14] have it? Just no one had ever said that [29:16] to me. And then it was just like clear [29:17] to me, wait, I need to write that down. [29:19] Like what do I really want right now? [29:21] What is my point of focus? [29:23] >> And that is what gave me motivation as [29:26] soon as I got that clarity. I mean, a [29:28] lot of us do it automatically, but [29:29] that's not where I was when I was a kid. [29:31] That's for sure. [29:32] >> Did you run across the coals? Did you do [29:34] any of that kind of stuff? [29:35] >> Oh, I've many times. [29:36] >> Really? [29:37] >> Yeah, of course. [29:38] >> Okay, there it is, folks. [29:39] >> Hey, we can go together. [29:41] >> Uh, absolutely. We had Tony on the show. [29:43] I tried to get in there and understand [29:45] his psychology, but he just had us do [29:47] breathing exercises. All right. The uh [29:50] All-In Summit is happening again. Fifth [29:53] year in a row. Wow. We've been doing [29:54] this for a while. Allin.com/events [29:57] September 13th, 14th, and 15th. We'll [30:00] have you back, Mark, in year six. But [30:03] when you score so highly on a keynote, [30:05] we give you two years off so that we can [30:07] build up the anticipation. So coming in [30:09] 2027, our guy Mark Benning off. But for [30:12] this year, uh, allin.com/events. [30:15] Okay. Topic two, AI's impact. [30:16] >> By the way, real quick, can can I just [30:18] do a quick PSA? [30:19] >> A public service announcement coming. [30:21] >> Public service announcement. [30:22] >> Yes. [30:23] >> Nick, just pull up this link. This dog [30:25] named Scooter needs a home. [30:27] >> Okay. [30:27] >> Promise my wife I do this. Don't get [30:29] Don't get mad at me. This dog named [30:31] Scooter needs a home. He's in Baldwin [30:32] Park. Dog's about to die. Lives in [30:35] Baldwin Park at the Animal Care and [30:36] Control. They're overwhelmed in LA. [30:38] There are so many homeless street dogs [30:40] that are being collected. They're all [30:41] being put to sleep. [30:42] >> Wow. [30:42] >> Someone give this dog a home. [30:43] >> Gorgeous dog. [30:44] >> Scooter. [30:45] >> Good dog. [30:45] >> Link. Link will be on the YouTube [30:47] channel. Thank you very much. [30:48] >> Are you getting a little emotional, [30:49] Freeberg, about this dog being put down? [30:52] >> I just don't like these, you know, all [30:53] these street dogs. like they they didn't [30:55] choose that situation they were in and [30:57] then they all end up kind of getting [30:58] pulled in by animal care and control and [31:00] then they all get put to sleep. Kind of [31:02] sucks. You're a dog guy. Come on. You [31:04] got to have a heart. [31:05] >> No, I have a big heart. Chimath, it's [31:06] $10,000 to keep these dogs alive for [31:08] another week. Can we count on you for [31:10] that donation? [31:12] >> Any purebred white golden retrievers in [31:14] that bunch? [31:15] >> No, they're just muts. Just muts. [31:17] >> We have a cat that we got at the shelter [31:19] actually. [31:20] >> Kind of a story. walked into the shelter [31:23] looking for a cat. All of a sudden, [31:24] we're in the shelter. This cat's like [31:26] coming at, you know, my kids and all of [31:27] a sudden they're, "Okay, we'll take this [31:29] cat." What is this cat's name? [31:31] >> This cat's name is Cloud. [31:34] >> Cloud. [31:34] >> That's our cat. [31:35] >> Wow. [31:36] >> Absolutely. [31:37] >> Yeah. We have Cloud the cat. [31:40] >> All right. I like it. That's awesome. [31:41] >> All right. AI's impact on software [31:43] rippling through the market. Mark, [31:45] you're on the front lines here. [31:46] >> Thank you for reminding me. [31:48] >> Yeah. Well, I mean in pain and suffering [31:50] are lessons, right? I think the Buddha [31:52] and Tony Robbins both taught us that and [31:54] the only way to the other side is [31:56] through as you know, Mark. So, [31:58] Salesforce down a whopping 37% 90 [32:01] billion in losses for you. service now [32:03] 42% workday 45% [32:06] 180 billion in market cap's been lost [32:08] and the assumption mark is that AI is [32:13] going to make it unnecessary for you to [32:15] use Slack or Salesforce or HubSpot [32:17] whatever it happens to be that you'll [32:18] just ask your AI to solve this for you [32:21] and we'll all look at a piece of glass [32:24] and have no actual software the AI will [32:27] just tell us what to do what's your [32:29] response to the fear the criticism and [32:32] how are you managing this massive change [32:34] internally? [32:35] >> You're right. It's the SAS apocalypse. [32:37] Um, I mean, it's not my first SAS [32:39] apocalypse, honestly, but it's the [32:40] current SAS apocalypse. So, we are all [32:42] now drinking at Salesforce Esprillas. [32:45] And, uh, we actually have a pet [32:47] Sasquatch as well. Yeah, we're all a [32:50] little more sassy. That's what I would [32:51] say. That's number one. [32:53] >> You bring the sass to sass. [32:55] >> We got to bring sass to sassy, you know. [32:57] And then, [32:57] >> well, you've always been a sassy guy. [32:59] >> Yeah. And then number two is look the [33:02] market is rerated. It's not a mystery. [33:04] Everybody knows you know you guys have [33:05] been talking about it for a while. I've [33:07] been living it. So the market software [33:09] market's rerated. It happens every now [33:11] and then. There are cycles. You know [33:12] I've been doing Salesforce for 27 years [33:15] enterprise software for 40 and the [33:18] market's rerated. So you mentioned [33:20] HubSpot. They're trading at two times [33:22] sales. I've never seen that before. They [33:25] just had a great quarter. We saw a lot [33:27] of great quarters. I looked at the top [33:28] 10 major enterprise software companies. [33:30] They all had great quarters and they're [33:33] all trading in two time sales. So why? [33:36] Because of everything you just said. [33:38] There's like, you know, a hypnosis [33:39] around AI and, you know, we haven't seen [33:42] it show up in the numbers yet. If it [33:44] shows up in the numbers, maybe people [33:46] will be right. Right now, all we know is [33:48] there's still a lot of enterprise [33:50] software being sold in the world. [33:51] >> So you've been through this. It's not [33:53] your first time. [33:55] >> This is not my first apocalypse. I mean, [33:57] I remember the cases apocalypse of 2020 [34:00] when COVID happened and all of it, you [34:02] know, everything collapsed. There was [34:05] the great cases apocalypse in 2016. [34:08] What's that? [34:09] >> In all sincerity, what's your strategy [34:11] and how do you deal with, [34:12] >> you know, your internal team? Obviously, [34:14] they are compensated through, you know, [34:16] stock and this is headwind and you have [34:19] competition for employees from OpenAI, [34:21] Anthropic, and SpaceX. So, how do you [34:24] deal with uh rallying the troops and [34:27] then how do you develop a strategy? [34:29] >> Well, you're right. I mean, it's a lot [34:31] easier to be a private company right now [34:33] where you don't have to be rationalized [34:35] by the public markets. If you're in the [34:36] public markets and you get rerated, [34:38] that's the reality of being in the [34:40] public markets. And I think that for [34:42] what I tell my employees, what I just [34:44] told my employees is look, you can't get [34:46] drunk on the stock price. If you are [34:48] like focused on today all the time and [34:51] that is how you're getting, you know, [34:52] your emotional state, it it's not going [34:54] to work for you. You have to find a [34:56] different anchor. So I try not to pay [34:59] really a lot of attention to it. I'm [35:00] focused on my customer success. How is [35:03] our revenue? Look, we'll do over 46 [35:05] billion this year, more than 16 billion [35:07] in cash flow. We have more than 83,000 [35:09] employees. These are the things that I'm [35:11] focused on. What is the level of [35:13] customer success that we're delivering? [35:15] That's a really important I can't [35:17] control the stock price. There's nothing [35:19] I can do. And for our employees, they [35:22] can't control it either. That they need [35:24] to believe in the company and the [35:25] quality of the success they're [35:27] delivering to customers in the long [35:29] term. And then you look at the market, [35:31] you look at these other companies, not [35:32] just us. They're doing great. And I [35:35] mean, I saw some great numbers, but it [35:37] doesn't really matter. The market's [35:38] rerated. [35:40] >> Well, you've got all that free cash [35:41] flow. You've been great at acquisition. [35:43] So, I'm sure you're uh [35:44] >> I've bought some great companies, by the [35:46] way. Everything's a little cheaper. I [35:48] like that. [35:48] >> Yeah. Chimoth, what's your take on this? [35:50] You've been pretty critical of the SAS [35:53] space in previous episodes, pointing [35:56] out, hey, how durable are these [35:58] revenues? So, what's your take? [36:01] >> I think the low end of the market is [36:03] basically finished. I think there there [36:05] is no safe space. I think the high end [36:09] of the market where Mark operates, where [36:12] the large monoliths operate is quite [36:15] safe and the tell was this week. [36:18] The the deployment company deal [36:21] basically shows you what OpenAI is [36:24] running into. So, you know, you have to [36:27] put in $4 billion. You get all of these [36:29] folks. You gave them a 17.5% preferred [36:33] guaranteed return for what? essentially [36:37] to stand up a competitor to Ernston [36:38] Young, Anderson, Deote, PWC, Cognizant, [36:42] etc. If you just look at that, that [36:44] doesn't actually mathematically make [36:46] that much sense if you look at what [36:48] those companies are actually trading at. [36:51] But why is OpenAI doing this? I think [36:54] why is because at the high end of the [36:56] market where all the action is, what [36:59] people are finding is, hey, hold on a [37:01] second. This is a lot harder than we [37:02] thought. It's not like boop boop boop [37:04] put in a put in a prompt and beep boop [37:06] it all works. It's not how it works. [37:09] >> And so I think I think we're [37:12] >> I think we're a little oversold. [37:14] >> And I now I think this consolidation and [37:16] the rerating can happen in the opposite [37:18] direction. So what is the opposite [37:20] trade? [37:21] >> The opposite trade is who has [37:24] constructive net dollar retention, who [37:26] has negative churn that's been really [37:28] predictable, which is a way of saying [37:30] who has the best relationships. meaning [37:32] they're inside the CXOs and the seauite. [37:35] They have relationships with the CEOs [37:37] and they're trusted and they've been [37:39] around for 20 years. Those guys, I [37:42] think, are positioned to crush because [37:44] eventually, Jason, the next trade that [37:46] has to happen is when the public markets [37:48] become a little bit less breathless [37:50] about AI and they ask one simple [37:53] question. Okay, guys, you've spent $3 [37:56] trillion in the last four years. What is [37:58] the ROI of these tokens? [38:01] And what's going to happen is they're [38:04] going to have to go to guys like Mark [38:05] and other people and say, "Please sell [38:08] my tokens." And that's the next shoe to [38:11] drop. If you if you had to ask me. So I [38:13] think you're going to see these revenues [38:15] which are like way out of whack like the [38:16] multiples on revenue, multiples on [38:18] assets, multiples of they're not even [38:19] multiples of those will come way back [38:22] down and I think these go back up and [38:25] you'll find a balance. [38:26] >> Yeah. And you've uh initiated what could [38:29] only be described as an unprecedented [38:31] massive stock buyback mark 50 billion. [38:34] >> I think it's the one of the largest in [38:36] history. Yeah, we we're you know we we [38:38] want to buy back as much as we can and I [38:43] would just say that you know look at a [38:46] high level look there's awesome new [38:49] capabilities like these coding agents [38:51] are awesome. Enthropic is awesome. like [38:53] I am going to probably use $300 million [38:56] of Enthropic this year at Salesforce [38:58] coding. Everything's going to be cheaper [39:00] to make. It's more efficient. I can do [39:03] things that I just could not do before. [39:04] I can go faster than ever before. I can [39:07] implement my software and sell it at the [39:09] same time. I've never been able to do [39:12] that before. I can break through [39:14] obstacles that I've had, you know, just [39:16] by focusing because I have coding agents [39:20] and humans together working together. [39:22] Today I have humans, agents and headless [39:25] platforms all interoperating never [39:27] before. So the opportunity for my own [39:31] company and the efficiency that I have [39:33] in my own company in service and support [39:35] in distribution in marketing across the [39:39] board is unprecedented. What I can do [39:41] for our customers unprecedented and you [39:44] know to that point my gosh have you seen [39:46] Anthropic? I mean it is a rocket ship [39:49] that will not stop because you can use [39:52] this product to do these incredible [39:54] amazing things and then complement it [39:56] with platforms like ours. It's you know [39:59] it's it's it's it's impossible to [40:01] describe what we're going to be able to [40:03] do for customers. It's going to be [40:04] awesome [40:04] >> and never waste a crisis. You've been [40:06] super focused on the efficiency of the [40:09] company, a headcount, share buybacks, [40:11] and I'm guessing you're looking at M&A. [40:15] So, it's not your finished up one of our [40:19] biggest deals ever, Informatica. I think [40:21] it was eight or nine billion dollars. [40:24] It's been in it's been awesome because [40:27] none of this stuff works if you don't [40:28] have context. You know, the AI is very [40:31] probabilistic. that is it can kind of [40:33] kind of figure things out but it needs [40:35] to be grounded in real data and it needs [40:37] to have that semantic layer. That means [40:39] it needs to be locked down into the [40:41] truth into a single source of truth or [40:43] it just cannot work well. And so our [40:46] customers want more harmonized data, [40:48] federated data, integrated data. So we [40:51] bought Informatica. Our customers want [40:53] our apps to use those large language [40:55] models to be able to provide not just [40:58] automation to their employees but also [41:00] to their agents just like we just said. [41:02] So like if you call 1800 no software [41:05] right now for the first time in 27 years [41:07] you talk to agent force hey agent force [41:10] what's happening but then after it [41:13] authenticates you and you get in the [41:14] system and it doesn't know who you are [41:16] then boom it'll autoes escalate you to a [41:19] human being who says oh I can see [41:21] everything you've been doing oh yeah you [41:23] just need to resolve this this and that [41:25] and so that trinity of the phone the web [41:28] and the human being are all together [41:30] it's all made possible buy this AI. We [41:33] just could never do that before. That is [41:35] awesome. [41:36] >> Was it controversial, Mark, to make [41:38] Slack headless? Was that a big decision [41:40] or was that a was everybody kind of [41:42] mostly on board with that? [41:43] >> Well, we were the first to, you know, [41:45] the way that our platform is architected [41:47] and Chimoth, you know it so well cuz [41:48] you're coding your coding agents and the [41:51] company you're building can run right on [41:52] top of it and build awesome stuff also. [41:55] So, you know, number one is we were the [41:58] first with an XML API, a SOAP API, a [42:01] REST API, now a CLA API, MCP API. We [42:06] always wanted to have a platform that [42:08] had every API possible. And our apps are [42:10] not hardcoded where we had to cut the [42:13] top off of them. They've always be [42:14] embedded inside our platform. So now we [42:17] can stream our apps out through a new [42:19] API called AXL and I can manifest it [42:22] into any large language model or device [42:24] or anything and it just works better [42:28] actually. [42:29] >> I think that you know when Jack Dorsey [42:32] wrote that memo Jal and part of what he [42:34] said is [42:35] >> he he's going to try to run his company [42:37] by building a world model. Yes. My [42:39] initial thought was, well, where is the [42:41] context, Mark? What you said before, [42:42] where's the semantic understanding to [42:45] create a world model? And a lot of it [42:47] for a lot of companies is actually in [42:49] Slack. [42:49] >> It's in Slack and email. Yeah, those two [42:51] places. [42:52] >> I showed you that Slackbot, Jason. I [42:54] mean, the thing is all because you run [42:56] your company on Slack, all your DMs, all [42:59] your channels, we're reading that now [43:01] through the AI and we can tell you more [43:03] about your business than you know [43:05] >> because Slackbot is reading stuff that [43:08] >> you know, nobody knew what was happening [43:11] and like I'm using that myself, but I [43:13] can then connect it into other things. [43:15] So, I have the ability to go into [43:16] Salesforce and Google and everything. [43:18] And so, when I'm on Slackbot, I can ask [43:21] it any question about my company. What [43:23] are my top five deals? What am do what [43:26] are my employees upset about? What are [43:28] the top three things I need to focus on? [43:30] And then boom, I get the information [43:32] because it has the data. [43:34] >> I think you should really look at SL [43:36] making Slack more open and cheaper and [43:40] getting more of that context because [43:43] there are limitations in terms of [43:45] getting your data out based on like [43:47] which plan you're on. It's just too [43:49] convoluted. But once I upgraded to a [43:52] higher plan and you know it's whatever 6 [43:54] 12 18k for our small company I was able [43:57] to get more and more context to my [44:00] openclaw plexity computer and claude we [44:03] and we're testing all of those and today [44:06] I wrote literally before I got on the [44:08] show a new prompt that is every two [44:12] hours in our Slack tell me what [44:14] decisions are being made who's making [44:16] those decisions um and what you would [44:19] handicap those decisions. If you were my [44:21] chief of staff, if you were a CEO or if [44:22] you were a board member and I created [44:24] personas to now evaluate decision- [44:28] making going on just [44:29] >> I have to show you the new version of [44:31] Slack. You're going to love it, Jason. [44:32] And you know that open AI and [44:34] enthropping and every AI company is [44:36] standardized on Slack. Yes. [44:38] >> And by the way, and our core sales cloud [44:40] and service clouds as well. [44:42] >> So I got to show you what we're doing. [44:44] Yeah. [44:46] So fun. [44:47] >> What's your advice to let's say you were [44:49] running a software company that was like [44:51] before the AI wave and you're private, [44:53] right? There's a bunch of these [44:54] companies that were supposed to go [44:56] public, didn't go public. What should [44:58] they do? Like what do those boards do? [45:00] What [45:00] >> they here's some Kleenex for them for [45:02] all their tears [45:05] >> because they're I mean I talked to these [45:06] CEOs. They're crying. What my market [45:08] cap? I'm not getting paid for my work. [45:10] Guys, grow up. You know, that's what I [45:12] love about the public markets. They [45:14] rationalize everything all the time. So [45:16] >> great, be in the public markets. You [45:18] want to be in a private market. Your [45:20] valuation is fantasy land until [45:21] somebody's actually going to pay you. So [45:24] I just tell them, hey, you got to focus [45:27] on your revenue, focus on your [45:28] customers, focus on your cash flow, [45:30] focus on your profitability, focus on [45:32] your innovation. How are you going to [45:34] add value to your customers? That's [45:36] what's really truly important. [45:39] >> Freeberg, any thoughts from you? We're [45:40] doing a big Salesforce implementation at [45:42] O'Hollow and I I would say that one of [45:45] my observations over the last six months [45:47] because we've talked a lot about using [45:49] these generative tools is we've kind of [45:52] dropped all the vertical software but [45:55] we're doubling down our investment in [45:57] horizontal tools. So like we're [45:58] investing heavily in these platform [46:00] capabilities that then we're able to [46:03] build apps and workflows on top of them [46:05] that are specific to our business rather [46:07] than try and buy an offtheshelf app or [46:09] workflow tool. So I think that there's [46:11] kind of a really interesting moment [46:12] right now where we talked about this [46:14] last time you could kind of discern [46:15] between founder and non-founder [46:17] enterprise software. I also think that [46:18] there's this kind of vertical horizontal [46:20] shift where they're kind of trading the [46:23] same right now but you could probably [46:24] break them. You could break that trade. [46:26] So, I think there's probably a good [46:27] arbitrage there. [46:28] >> I think everyone who wants to sell those [46:30] potato seeds, you know, is going to need [46:34] >> some tools to sell them. But the thing [46:36] that's kind of cool is, you know, we [46:38] have 15,000 salespeople now. So, they're [46:40] out there selling these products like [46:42] Slack and selling to David and so forth. [46:45] But here's the thing that's interesting. [46:47] In the last 27 years, we calculated [46:50] somewhere between 20 to 30 million [46:52] people, we didn't call back. We did not [46:55] call them back [46:56] >> because we didn't have the people to do [46:58] it. So just this week I called back [47:01] 50,000 people, you know, just through [47:04] agents so I can qualify. I just bought a [47:07] company called qualified so I can [47:09] qualify the agents, call people back the [47:11] BDR, the SDR function. I can go [47:14] outbound. I can do things I could never [47:15] do before. That is made possible by this [47:18] AI automation linking it then to the [47:21] apps. You can go to another level. [47:24] Yeah. Okay. Breaking news. I want to get [47:26] everybody's take on this. Open AAI in a [47:28] breaking news story is considering suing [47:31] Apple over their Chat GPT partnership. [47:33] As we all know, two years ago, Apple and [47:36] Open AAI announced Chat GPT would [47:38] integrate into Siri, iOS, MacOSS. But [47:41] according to Bloomberg, the deal has [47:43] gone so poorly for OpenAI that they [47:45] might sue Apple for breach of contract. [47:48] Here are OpenAI's gripes. Apple chat GPT [47:51] integration within Siri requires users [47:53] to specifically say chat GPT to get [47:56] results. We probably all experienced [47:57] this if we haven't given up on Siri, [47:59] which is the worst personal assistant [48:01] ever created. Uh Apple hasn't it's [48:03] discretionad to the highest level. Just [48:07] discretad with this product. How do you [48:09] get there first and you remain worst? [48:13] Apple hasn't promoted the integration at [48:15] all and users are still overwhelmingly [48:18] going to the standalone chat GPT app or [48:21] others. OpenAI [48:23] expected billions of subscription [48:25] revenue from the deal and it hasn't come [48:27] to fruition. According to Bloomberg, [48:30] Apple's side of the story is, hey, they [48:31] have concerns about Open AI's privacy [48:34] practices. Maybe they don't trust the [48:35] guy in charge over there, which was a [48:37] reoccurring theme in the lawsuit. And [48:40] they're annoyed that OpenAI, and here is [48:43] the palace entry, they're upset that [48:48] OpenAI is building hardware to compete [48:51] with Apple and that they recruited their [48:55] design guru, Johnny IV. [48:58] Mark Beni off, you know, the players. [49:01] What is going on here? Yeah, [49:04] >> you got to trust Sam Alman and Open AI. [49:07] I love him. But let's just upscale this [49:10] conversation just one second. Listen, [49:13] what has happened? What happened is we [49:15] have these LLMs. They starting to come [49:17] out come out. Every company's kind of [49:20] chosen a slightly different path. You [49:22] had Elon, he went out, he had Grock and [49:24] he kind of started building these [49:25] companions and sex bots and all this [49:28] kind of stuff going on. And that was a [49:29] huge focus of his, the sexbot focus. [49:32] Then you kind of had OpenAI and they [49:34] were doing the Sora video thing and [49:36] they're also doing ad networks and crazy [49:38] stuff like that. Then you had Gemini and [49:41] they had the Nano Banana and then [49:43] finally you've got Enthropic and they go [49:45] we don't know about those sex bots and [49:46] we don't know about Nano Banana but [49:48] we're going to do coding agents. And it [49:50] turned out Enthropic was right. And all [49:52] of a sudden the rocket ship took off and [49:54] now everybody's like, "Whoa, where did [49:56] Enthropic go?" Oh, whoa, they're way up [49:58] there. And then they're all like, "Where [50:00] are all they going to do coding agents [50:01] to?" And now they're all resetting and [50:04] kind of hitting their buttons and going, [50:05] "Coding agents. Everybody focus, focus. [50:07] Kill Sarah, kill this. Get rid of that. [50:11] Sex bots off." You know, cursor on. [50:14] Yeah. And that's where we are right now. [50:18] And so everybody, look, this is what's [50:20] cool about right now is that everybody, [50:22] it's such a dynamic moment in our [50:23] industry. It's so exciting that people [50:26] have to pivot and you have to be ready [50:28] to focus and refocus and constantly ask [50:31] the question like, "What do you want [50:32] right now?" And everyone is changing [50:35] what they really want. And if you looked [50:37] at where everybody was a year ago and [50:40] where they are now, they're in a totally [50:42] different place cuz we all know that [50:43] when Enthropic 46 hit, boom, everyone [50:47] could code in their companies. And [50:49] before that, they really couldn't. It [50:50] was a little bit of a productivity [50:52] improvement, but not as much as we [50:54] wanted. Now everybody sees this and [50:56] goes, "Wow, this is unbelievable." We're [50:59] even working on technology inside Slack [51:02] to make it easier for everybody to code. [51:04] So I think it's going to be really [51:07] >> breaking news. Breaking news there. [51:08] >> You're going to see some cool stuff with [51:10] Slack and Code. I'm not ready to talk [51:11] about it yet, but there's no question [51:14] that uh we are in a new moment in [51:16] coding. I mean Chamas got a whole [51:18] company around it. A lot of people have [51:19] companies around it. There's going to be [51:22] before coding was all about humans and [51:24] like that's where I started. I was 15 [51:26] years old, Berlin game high school on [51:29] hands-on keyboard writing video games, [51:31] you know, basic 68,000 6502, [51:35] you know. No, no, now it's like [51:38] everyone. [51:38] >> How's that changed your org? Is it the [51:40] product manager, the developer, or the [51:42] UX designer? [51:45] >> The creator, or did it all blend into [51:46] one blended? And who wins? Who wins? [51:49] It's all blended together. And by the [51:51] way, to to Mark's point, the hands-on [51:53] keyboard thing, you know, most of our [51:55] engineers just speak. They So it's like [51:58] And so, you know, one of our guys just [52:00] has a foot button operated thing and it [52:02] just like [52:04] that, [52:04] >> too. Yeah. Whisper. That's all they do. [52:06] Flow plus. [52:08] >> It's all talking. Pedal plus whisper [52:10] flow and talking. There is no cans on [52:12] keyboard anymore. [52:13] >> All right. So, let's get back to this [52:14] story here. [52:14] >> We're going for dolphin flow at [52:16] Salesforce. [52:18] >> Dolphins on keyboards. Mahalo for you. [52:21] Aalo. Um, Freeberg, two questions. One, [52:25] now that Sam has shut down the sex [52:27] spots, how are you filling that void? [52:28] And then number two, [52:31] thank you, Jama. [52:33] And number two, what's your take on the [52:35] strained relationship here between [52:37] OpenAI and the world and specifically [52:40] the Apple world? [52:42] >> I don't know what to tell you. This that [52:43] there doesn't seem to be a lot of [52:44] long-term partnerships with Open AI. [52:47] That's a very uh generous way to say it. [52:52] What should Apple do? Should they just [52:54] go back to the loving arms of Google [52:56] where they made a hundred billion [52:59] dollars being partners with them on the [53:01] search default your alma [53:03] >> I don't know what to tell I mean look at [53:04] the end of the day I think Google has a [53:07] real opportunity to integrate a Gemini [53:09] assistant into all of your personal [53:10] information and Gemini and calendar [53:13] >> your Google drive your Google photos [53:15] where all of your personal information [53:16] sits it can become your your point of [53:19] calendaring your point of asking it [53:21] questions about your personal [53:23] Hey, when did I email this person, etc. [53:25] And in the enterprise context, I think [53:26] that's also up for grabs with G Suite. [53:30] And so, Google has a real chance to kind [53:32] of own that assistant interface. Maybe [53:34] Salesforce does too, Mark. Apple [53:36] obviously has an embedded install piece. [53:37] How many people are using Apple services [53:41] for doing all of their mail, for [53:43] calendaring, [53:45] for storing their information? I think a [53:47] good number. [53:49] >> Yeah. So that may end up becoming kind [53:51] of the way that the world silos out is [53:52] like you'll have an assistant, but the [53:55] assistant for it to be truly useful to [53:57] you individually, it has to have access [53:59] to all of your information. Whether it's [54:01] in an enterprise context or a personal [54:02] context, Apple is probably going to need [54:04] to either build their own or white label [54:06] with a partner, could be anthropic, [54:08] could be someone else [54:09] >> to get this to really work. But but [54:11] they're going to be up against a [54:12] formidable competitor is my point [54:13] because I think Google Assistant is [54:14] going to end up be being kind of a real [54:16] kick-ass product once they get this [54:18] flywheel going. Apple has the clearest [54:20] path to becoming, you know, a top two or [54:23] three player in AI simply by buying [54:26] something like Perplexity or Mistrol or [54:29] or or some AI lab and then using their [54:33] hardware footprint, which is [54:34] extraordinary. I just got this MacBook [54:36] with 48 gigs of RAM on it with an M5. It [54:39] is unbelievable. And the new ones with [54:41] the the Studio coming out are going to [54:43] have a terabyte of RAM. they are going [54:46] to have a clear path to maintaining your [54:48] privacy running local models and that's [54:51] going to solve people's problems. It's [54:53] going to be very easy to index all your [54:55] images without giving the data to OpenAI [54:58] which nobody trusts or giving it to [55:00] Gemini. And that's I think a big part of [55:03] why they picked the current CEO is [55:05] because he's an engineer and [55:07] hardwarebased. I I think the future of [55:08] this is going to be local models running [55:10] on extraordinary desktop hardware. And [55:13] if you have employees on this level of [55:16] hardware running these models local like [55:18] I have started to do, they become 10 [55:20] times more valuable than the employees [55:22] not running it. I I think Apple's my [55:24] choice for the next year. [55:25] >> I'm not a huge believer in these local [55:27] models. And the the only reason why is I [55:29] think you want persistence. You have [55:31] multiple devices. You have multiple [55:33] persona. You're using a web browser in [55:35] one instance. You're using a different [55:36] computer at home in the other. And the [55:38] idea that you don't have persistence [55:40] that follows you around in 2026, I think [55:42] is a breaking feature. So, I think [55:45] you're not going to lug around a 5B [55:48] MacBook Pro everywhere you go so that [55:50] you can get knowledge. Oh, wait. You're [55:52] asking, "Oh, wait. I got a diagnosis [55:54] from my doctor. Hold on a second." [55:57] >> I don't think that's [55:59] the game on the field. The game on the [56:02] field though is [56:03] >> you will uh be able to use iCloud for [56:05] this as well. So, it's going to be, you [56:08] know, effortlessly. [56:10] >> So, well, look, give me your opinion on [56:11] this. I think the thing that you're [56:14] assuming is that these form factors [56:16] don't need to change. And I just wonder [56:19] like, don't we have an iPhone moment [56:20] that we're just all going to be [56:22] surprised by where somebody shows up and [56:24] they're like this thing and you're like, [56:26] "Oh my god, that's the thing." Do do you [56:28] know what I mean? like yeah that that's [56:31] I [56:32] >> there's enough ingredients here where [56:34] somebody's going to cook something up [56:35] and then you have to think about Jason [56:37] the sunk cost of like you know look [56:40] Apple is incredible [56:42] but there that's the product of 40 years [56:44] 30 years of meticulous process [56:47] optimization [56:48] what do you do if the form factor is [56:50] totally different and the nature of the [56:51] device is different how do you pivot [56:53] that organization [56:55] >> you're hitting on something very [56:56] important and it dovetales with another [56:57] story on the docket mirror Mrage [56:59] released thinking machines new real time [57:04] world [57:04] >> that was impressive [57:05] >> and this is super impressive and Apple [57:07] at the same time has patented putting [57:10] cameras into your AirPods so you're [57:12] right and then I use this plug pin I've [57:14] talked about it before [57:15] >> these persistent hardware devices the [57:17] watch the earpieces are going to know [57:18] your entire reality they're going to [57:20] monitor your entire desktop and this is [57:22] going to lead to a a use of tokens that [57:26] would be a thousand times what user [57:28] business users are currently using [57:30] because the way mirror's and we should [57:32] have her on the pod or at one of our [57:33] events. The way her model works [57:35] Freedberg is it's watching your desktop. [57:38] It's listening to all the voices and [57:40] then it's watching your webcam all at [57:42] the same time and every 200 milliseconds [57:46] it's uploading it to two different [57:47] models. One is deep thinking and looking [57:50] backwards, you know, at a maybe a 30- [57:52] secondond clip and the other one's in [57:54] real time. that will 1,000x the need for [57:58] tokens if people are doing this for 8 [58:00] hours a day cuz it's not the turnbased [58:03] here's my latest prompt here's my [58:05] question I got a response it's in real [58:08] time always querying an LLM which is [58:11] going to require a hardware upgrade to [58:13] the average desktop Mark Beni off what [58:16] are your thoughts on this brave new [58:17] world [58:18] >> it's one small step for man one giant [58:20] leap towards AGI I think that you said [58:25] really well. I mean the we are in the [58:27] we've been so kind of think that LLM is [58:29] the beall end all. We're going to go to [58:31] AGI. I don't really understand how large [58:34] language models which are only about [58:35] language and words and we know how it [58:37] works. It's one word one word therefore [58:39] the next word is this is going to get us [58:41] to where we want to go to which is [58:43] minority report or um [58:45] >> agreed [58:45] >> all the science fiction movies that [58:47] we've seen. But what we saw and and I [58:49] think Jason you'd really articulated [58:51] that super well by the way multiensory [58:54] models and multiensory models. Well, [58:56] here's a good one. Oh yeah, me. I'm a [58:59] multiensory model in a biological [59:01] computer. I'm a multiensory model. I've [59:03] got these eyes, ears, mouth, you know, [59:05] work. Sometimes some brain, heart, [59:07] whatever. And some other things going on [59:09] too that I don't even understand. And [59:11] it's all running in this biological [59:13] computer. I'm not exactly a large [59:15] language model, though I do have one. [59:17] sometimes. So I would say that [59:20] multi-ensory models are the next big [59:23] wave for AI and then but we're still not [59:26] at AGI at that point. But those demos [59:28] like the demo of that model that was [59:31] pretty everyone should see like that was [59:33] pretty awesome. And then I think we can [59:36] kind of say hm where are we now on this [59:39] on the path? But I think every model [59:42] company is kind of go hold on pivot you [59:45] know again pivot like saying pivot [59:47] everybody's pivoting and you're going to [59:49] have to pick your poison and where where [59:51] are you going next you know and you [59:54] can't do everything you can't [59:55] >> you said you were spending 300 million [59:57] on tokens with anthropic. Now imagine [60:00] what this would do to an average, you [60:03] know, employee if they needed a thousand [60:08] times the tokens they have now. So [60:09] instead of spending 150k in tokens, you [60:11] got to spend, you know, a hundred [60:13] million. But you have [60:15] >> I don't think that's true. I I don't [60:17] think that that I think we're getting [60:18] brainwashed on that. [60:19] >> Okay. So explain. [60:21] >> I think that's a mistake to think that [60:22] way. See, I think we are wasting a lot [60:24] of these tokens. So my coding here's my [60:26] I think I convince you this. So here's a [60:29] here we're using $300 million of [60:30] enthropic this year and we're coding [60:32] we're coding we're coding right the vast [60:35] majority of those tokens don't need to [60:36] go to enthropic there needs to be some [60:39] intermediary layer layer that's saying [60:41] oh oh that one has to go to enthropic [60:43] but these ones can handle by smaller [60:45] models [60:47] that can route it to the most affordable [60:49] for the job it's such an early moment [60:52] you know it's an early moment and you're [60:54] and I want to connect back to what you [60:55] said before which was brilliant Jason [60:57] which because the importance of the edge [60:59] because the edge and the cloud are going [61:02] to come together. Okay. And yes, you're [61:05] going to have more distributed [61:06] intelligence. It's going to work [61:08] together. That's going to makes a ton of [61:09] sense what you said. What Chama said was [61:12] also completely right. You were both [61:14] right. Then you combine it to this idea. [61:17] Let's put it together where we're going. [61:18] Yes, we're going to have more [61:19] distributed intelligence. We're going to [61:21] have intelligence on the edge. We're [61:22] going to have multiensory models. And [61:25] yes, you're going to do coding. And it's [61:26] going to be more efficient though. So to [61:28] think that it's going to be always so [61:30] expensive, that is just the moment of [61:33] time that we're at right now. And I [61:35] think there's going to be a hot new [61:36] company that's going to come along and [61:38] say this. I'm going to sit between you [61:41] and Anthropic and Open AAI and I'm going [61:44] to make sure that they you only need [61:46] their tokens when you actually need [61:49] them. [61:50] >> And that is not really where we are [61:51] right now. [61:53] >> Does that make sense what I'm saying? [61:54] Like you're like% [61:56] >> it's like oh I need a shower and Okay, [61:59] great. All the water. No, no, no. I just [62:01] need some water. I don't need all the [62:03] water. [62:04] >> Yeah. [62:04] >> Okay. [62:05] >> All right. Listen, we uh have two final [62:08] topics here. We could talk about [62:09] anthropic unrolling the dark SPVS and [62:12] the impact that has on the market. Or we [62:15] can go to everybody's favorite science [62:17] corner. [62:18] >> Well, do both. Just do a quick uh around [62:20] the around the neighborhood. [62:21] >> Yeah. Okay. Lightning round on uh we'll [62:23] do SPVS last. Freeberg, what do you got [62:26] in Science Corner? Everybody, all the [62:28] all the Freedberg fans are just [62:30] destroyed when we skip it. So, let's [62:31] give [62:32] >> a good one. Mark, do you like Science [62:33] Corner? No, you do. [62:35] >> I do. [62:35] >> Come on. Yeah. [62:36] >> Only the oceans. [62:38] >> Mark is the greatest salesman in Silicon [62:40] Valley. He loves it all and he wants to [62:42] hear more. [62:44] >> Mark, how did you learn sales? How do we [62:47] become half as good as you? [62:48] >> I'm not a good sal. I'm like a I'm a a [62:51] geek. [62:52] >> You are, but you know how to [62:53] >> call it a Tony Robbins seminar. Let's [62:55] go. Walk over the coals together. Let's [62:57] roll. [62:57] >> We're all going to Tony. Put out the [62:59] coals. We're walking across them. [63:01] >> Nick, let's pull up the first chart. [63:03] This science corner is about the dreaded [63:06] El Nino season that is coming up. And so [63:09] this is a forecast for sea surface [63:11] temperature anomalies. And basically [63:14] what this means is that the ocean at [63:16] this point has heated up so much that [63:21] there's now a forecast on ocean [63:23] temperatures that are going to exceed [63:25] anything we have seen in recent history. [63:29] So we're kind of looking at temperatures [63:30] that might be 4° above normal. That [63:32] doesn't sound like a lot, but let's just [63:34] look at this image. That top image is [63:36] the sea surface temperature anomaly. [63:38] That means how different the sea surface [63:40] or the the ocean temperatures are. Why [63:42] does that matter? And this is compared [63:44] to 1877 when we had the biggest El Nino [63:47] year ever. And I'll explain what that [63:49] means in a moment. But why does this all [63:51] matter? The reason is the oceans are the [63:54] battery of weather. What happens at one [63:57] part of the year is that the oceans [63:58] absorb heat energy and absorb absorb [64:01] absorb gets hotter and hotter and hotter [64:03] and then that heat energy is released [64:04] into the atmosphere and then the [64:06] atmosphere drives the weather events [64:07] that happen all over the world and it [64:09] becomes somewhat predictable that you [64:10] can estimate what the weather over the [64:13] next year is going to be like based on [64:15] how the temperatures are sitting in the [64:16] ocean today. And at this point there is [64:19] so much excess energy stored up in the [64:22] oceans. I'll give you guys a statistic. [64:24] It's about 11 million terowatt hours. [64:27] The whole planet Earth uses 25,000 [64:30] terowatt hours in a year. So 11 million [64:33] extra terowatt hours of energy is [64:35] currently stored up. That's 500 years [64:37] worth of human energy in this ocean. And [64:39] over the next few months, that energy is [64:41] going to be released into the [64:42] atmosphere. And that will absolutely 99% [64:46] confidence that will make the upcoming [64:48] year the hottest year on record by far [64:50] that humans have ever experienced or at [64:52] least that we've experienced in recent [64:54] history where we didn't have data going [64:56] going back a long time. It changes the [64:58] trade winds which changes how a lot of [65:01] the weather evolves from a normal [65:02] season. It changes the moisture in the [65:04] atmosphere in certain parts of the [65:06] world. And I'll just give you some [65:08] highlights of what this means. Over the [65:09] coming year, there will be major [65:11] atmospheric river events where you just [65:13] get water dumped on you in the southwest [65:15] in California and the Gulf Coast. You'll [65:18] have very low snowfall and very high [65:20] heat waves in the northern part of the [65:21] US going up to Canada. You guys remember [65:23] all those fires a couple years ago in [65:25] Canada. Interior regions of the US like [65:28] in Phoenix and and so on. They're [65:30] already seeing temperatures at 106 [65:32] degrees in May. A super El Nino like is [65:34] forecasted for this year could extend [65:36] that duration of these heat domes which [65:38] could create temperatures that we've [65:39] never seen in those areas. Southern [65:41] Argentina, Chile, Brazil could see [65:43] record shattering heat waves. And this [65:45] is where things start to get a little [65:46] nasty because when that happens, the [65:48] crops start to fail. And in many parts [65:51] of the world, Brazil, India, Australia, [65:54] these are critical uh local consumption [65:56] for large populations, but also major a [65:59] export markets. So if Brazil's crop [66:02] fails, if Australia's wheat crop fails, [66:05] Australia's wheat crop goes to places [66:07] like Indonesia and the Philippines. [66:08] Hundreds of millions of people depend on [66:10] that that wheat crop. Hundreds of [66:12] millions of people depend on the exports [66:13] out of Brazil. Brazil is the world's [66:15] largest a exporter. And the scariest one [66:17] of all is if the monsoons fail, which is [66:20] now a very high probability event in [66:22] India. 150 million farmers in India that [66:25] depend on their agricultural output or [66:27] they don't make any money, they can't [66:28] survive. and one and a half billion [66:30] people that depend on that food. So the [66:32] importance of this El Nino event goes [66:34] beyond just like an interesting weather [66:36] anecdote. But if you think about the [66:38] second and third effects of this over [66:40] the next year, you could see energy [66:42] prices spiking and electricity spiking [66:44] and the grid failing in parts of the [66:46] southwest, commodity prices spiking all [66:48] over the world. And then you would see [66:50] places like uh India, the Philippines, [66:53] Vietnam starting to face some sort of [66:55] unrest if there isn't enough food supply [66:57] that's coming into those markets. And [66:58] then the question in India is a really [67:00] nasty one because there isn't a really [67:01] good solution. India and markets like it [67:04] that are significantly dependent uh on [67:07] having their monsoon event but also are [67:10] currently facing a shortage of nitrogen [67:13] based fertilizer because of the crisis [67:14] with Iran and the straight of Hormuz [67:16] which we've talked about as a double [67:18] whammy. So over the next year in South [67:20] Asia you could see a calorie deficit and [67:23] a major kind of economic crisis that [67:25] starts to emerge. So, this El Nino [67:26] event, I just wanted to kind of bring it [67:28] up as an interesting science corner [67:29] because the data now seems pretty [67:31] confirmed that over the next few months, [67:33] we're going to have this record-breaking [67:34] El Nino and these are the sorts of [67:36] consequences that will play out over the [67:38] next year. [67:38] >> This is scary. God damn. What do [67:41] you do about the the food problem? [67:44] That's a big problem. [67:45] >> That's a big problem. And this is the [67:47] commodity markets are going wild for [67:49] this right now. [67:49] >> Did this did this happen the last time [67:51] we had an El Nino? [67:52] >> Yes. So when we have El Ninos, there's a [67:54] severity of these sorts of events [67:56] happening. [67:56] >> No, I mean, did we have a food shortage [67:58] problem in the last El Nino? [68:00] >> Oh, yeah. When we have El Ninos, there's [68:02] food shortages that come out of [68:03] Australia, Brazil. You see these events. [68:05] That's a regular part of the the thing. [68:06] The problem this year is the index is so [68:08] off the charts. It is so far beyond [68:10] anything we've seen and it's correlated [68:13] obviously with the severity that it [68:15] triggers a crisis that may be, you know, [68:17] really difficult to manage in a lot of [68:19] places. So the US is a net a exporter. [68:22] So from a US perspective, our big issues [68:24] to worry about are the fire season. By [68:26] the way, this does decrease Atlantic [68:28] hurricanes. You're going to have fewer [68:29] hurricanes this year in the uh in the [68:30] south, but you know, you have these kind [68:32] of major storm events that happen in [68:34] California, the heat waves, electricity [68:36] prices. We're probably more stable than [68:38] other parts of the world. But think [68:40] about the Brazil, the a export market is [68:42] such a significant part of the economy [68:44] and Brazil has a debt problem that this [68:46] could end up becoming an economic [68:47] problem in some parts of the world in [68:50] including Brazil. So the the commodity [68:52] markets are trading this El Nino event [68:54] pretty actively right now. It's become [68:55] kind of a focal area for a lot of folks [68:57] in in commodity markets. Can your super [68:59] seeds help or no? [69:01] >> That's a longer term solution. But I [69:03] mean look, we're like at the end of the [69:04] day having um this actually a good [69:06] point, but having genetics a crop [69:09] genetics that can adapt to a more [69:11] rapidly changing climate is going to be [69:13] critical over the next couple of [69:14] decades. Uh that's a big kind of macro [69:17] for [69:17] >> another part of this Freedberg that as [69:20] certain places heat up they actually [69:23] become viable places to put crops i.e. [69:26] Canada or the northern part of the [69:28] United States. So we're seeing the [69:31] actual places you can grow certain crops [69:34] independent of gen genetic modification [69:36] is expanding. In other words, farmlands [69:38] expanding because of global warming. [69:40] >> Right? So we increasingly grow soybeans [69:42] in Canada. That was not the case. uh two [69:45] decades ago through both plant breeding [69:46] where we've made the genetics adapt but [69:48] also the warmer temperature the shorter [69:49] winters and whatnot they've been able to [69:51] grow that that crop further north but if [69:55] you guys go back a couple of years there [69:56] were major heat waves and droughts in [69:58] Australia [69:59] >> huge fires yeah [70:00] >> huge fires but remember that wheat goes [70:02] to Indonesia it goes to Vietnam it goes [70:04] to these countries with hundreds of [70:05] millions of people dependent on it and [70:06] then they got to go scramble and find it [70:08] elsewhere now on a global basis there's [70:10] always a a bit of a give and a take but [70:13] in a year like this when you have this [70:14] much energy pumped into the atmosphere, [70:16] you may not have the given the take and [70:19] that's where things start to buckle. [70:21] >> All right. So, yeah, it's just like an [70:23] early warning sign. I know I've given [70:24] food warnings, food crisis warnings [70:26] twice before on the show. He gave a [70:28] really good one for Ukraine four years [70:29] ago and it turned out that that warning [70:33] resulted in a negotiation to allow wheat [70:36] to flow out of Ukraine that Russia [70:39] Ukraine and other folks made a specific [70:42] carveout in their war to make sure that [70:44] famine didn't spread. So [70:45] >> and the fertilizer export was negotiated [70:47] at the start of the war as well which [70:48] was critical. [70:49] >> Yeah. [70:49] >> Uh for sure. [70:50] >> I think you played a role in raising [70:52] awareness for that. So I give you credit [70:53] to Freeberg Mark. Uh thanks for coming [70:56] on the program. Sorry that uh at the end [70:58] of the program we all want to commit [71:00] sapuku here but you know we have Dr. [71:02] Doom and he uh he always loves to end [71:04] the show. There he is [71:08] with an upbeat you want to rap with an [71:09] upbeat anthropic [71:11] doomsday starring David Freriededberg [71:14] coming to the Marvel franchise in [71:16] December. Yes. So you thought Robert [71:18] Downey Jr. was playing him. It's [71:19] actually Freedberg. Here's your meme. [71:23] It's not Mark Beni off's first time uh [71:25] here in the SAS apocalypse. [71:27] >> There it is. First time. [71:31] >> He's like, I've been doing this before, [71:33] boys. I'm buying my stock back. [71:35] >> Somebody cut that rope. [71:37] >> Yeah, absolutely. [71:39] >> All right, listen. Uh I'll I'll just [71:41] lightning round this one. Philanthropic [71:43] has had enough of the shenanigans of the [71:45] multi-tiered, multi-layered SPVS being [71:49] sold to dentists for 10% loadin fees. I [71:53] think we all knew this day was coming [71:55] since it's becoming the wild west. They [71:57] called out specific platforms and they [71:59] said we're going to negate them. Chimoth [72:01] Tempest in a teapot or long overdue [72:05] punishment for bad actors. [72:07] >> Love Mark said it right. All these [72:09] companies should go public and get [72:11] evaluation and and focus on the higher [72:14] order bit. And all of these tickytacky [72:17] mechanisms that people can use to stay [72:19] private longer need to get a bullet put [72:21] in its head. And these SPVS are the [72:23] worst. The layered SPVS on SPVS on SPVS. [72:28] >> By the way, I and I will guarantee you [72:29] this. Once SpaceX goes public, once [72:32] Anthropic goes public, once OpenAI goes [72:34] public, you're going to see a litany of [72:37] these lawsuits back and forth between [72:38] the purveyors of these SPVS, they should [72:41] not be allowed. [72:42] >> They should be all negated if possible. [72:46] You can't now you can't unbreak the [72:48] eggs. So, I guess you just have to call [72:50] it what it is. I hope every company [72:51] adopts this and I hope as a result these [72:54] companies go public sooner and they [72:55] rationalize their equity structure [72:58] faster. I think it's the right thing to [73:00] do. So, I like the fact that anthropic [73:01] is [73:01] >> just say it. Descatiad. [73:04] >> Well, it's not discretziad. I just I I [73:06] just think that you are going to have a [73:07] lot. [73:08] >> No. Well, you're going to have a lot of [73:09] people that are going to be very upset [73:10] once this SpaceX thing comes out because [73:12] inevitably, because it happened in Uber, [73:14] somebody gets screwed in a layered SPV [73:16] somewhere that they didn't know and [73:18] they're going to be like, "What [73:19] happened?" [73:19] >> Read the fine print. You're getting [73:21] double carry and you paid 10% on the [73:23] loadin fee and you paid a price 10% over [73:26] what the last round was. So, it's a [73:28] recipe for disaster. The fine print. for [73:30] disaster. [73:30] >> Recipe for disaster. Hey Mark, great [73:33] job. Thanks so much for coming back with [73:35] you guys. Yeah, come back any time. You [73:37] were great. Hey, your [73:40] >> there's a we have a fan question here. [73:42] This is a fan question that came in [73:44] written by your PR department. They want [73:46] to know when you do your charity work uh [73:50] or your innovation as a founder, are you [73:52] more, [73:53] >> you know, at this stage in your career, [73:55] you're known for both of these things. [73:57] Are you more Steve Jobs or more Lorraine [73:59] Pal right now? What your PR department [74:01] worked very very deep question they f [74:04] the best decision I ever made the best [74:07] when I started Salesforce we put 1% of [74:10] our equity 1% of our profit and 1% of [74:13] all of our employees time into a [74:14] foundation. Today we've done more than [74:17] 10 million hours of volunteerism. We've [74:19] given away more than a billion dollars [74:21] in grants [74:22] >> and we run over 50,000 nonprofits for [74:24] free on our platform. That's just [74:27] Salesforce. [74:28] Every I every company should do this. [74:30] It's amazing. Pledge11%.org. [74:33] That's my advertisement for the show. [74:35] >> Everybody should jump on the 111 [74:37] platform. Best decision I ever made 27 [74:39] years ago. Irrespective I've obviously [74:41] done a huge amount of personal [74:43] philanthropy to five hospital. [74:46] >> You did one that was really I think [74:48] personal. Uh Susan Wjeki tragically died [74:51] of cancer and major. [74:52] >> You're going to get me crying. Let's not [74:54] go there. This is not your time. [74:57] My best friend. Bestie. My bestie. [74:59] >> She was a special person, huh? [75:00] >> She's a star. [75:01] >> The best. She was the best. [75:03] >> Just I mean, [75:04] >> I'm not kidding. You're I'm not kidding. [75:06] >> Tell the audience what was so unique [75:08] about Susan. Tell tell them what you're [75:10] It's a good time to do a remembrance. [75:12] >> She This was one of the great people on [75:15] the planet. Incredible woman. Mother of [75:18] five amazing children. Um, and uh, two [75:22] amazing sisters, great family. Uh, [75:25] everybody knows Ann, her sister, her [75:27] sister Janet at UCSF. Um, sorry, my eyes [75:30] are tearing up, but I'll just tell you [75:32] that, you know, she contracted a very [75:34] rare form of cancer, passed away too [75:37] young a couple years ago now. Um, and [75:41] she was on our board. She was one of the [75:43] great people of all time. I'm so [75:45] grateful that her family is focusing on [75:47] building a new foundation to go after [75:50] this rare cancer. Uh we're backing it. I [75:52] think everybody should. Um look, [75:54] everyone should try to be a Susan. She [75:56] was one of the absolute greats and uh I [75:59] have an altar in my office. When you [76:01] come to see me, you'll see with the [76:02] photos of her and me and I think of her [76:05] every single day. And uh my heart just [76:09] stays with her, her husband Dennis, all [76:11] the children, the sisters, the wholeiki [76:14] family. Amazing. And one of the great [76:16] institutions in Silicon Valley. We're [76:18] lucky to have the whole family. [76:19] >> Yeah, she was fantastic. [76:21] >> Well said. [76:21] >> All right. Well said. Rest in peace, [76:23] Susan Waki. [76:24] >> Thank you. [76:24] >> All right. We'll see you all next time [76:26] on the podcast. Love you boys. [76:28] >> See you guys. Aloha. Thanks for Bye-bye.