Anthropic Targets $2 Trillion Valuation in October IPO as Revenue Soars

营收飙升，Anthropic拟于10月IPO寻求2万亿美元估值 华尔街融资与庞大的企业支出正推动AI实现创纪录增长，但在模型取代人类智能之前，严峻的物理能源限制最终将制约软件收入。 主持团队与嘉宾Gavin Baker分析了快速演变的AI与科技格局，探讨了Anthropic在面临物理能源与算力限制的同时，能否支撑传闻中2万亿美元的IPO估值。他们将Mark Zuckerberg的开源AI愿景与有效利他主义者（Effective Altruists）高度监管的中心化方案进行对比，认为开放获取对人类能动性和国家安全至关重要。此外，与会者还探讨了Nvidia规模达5000亿美元的GPU融资架构、Amazon就配送分包商用工问题展开的法律纠纷、xAI凭借Grok 4.6确立的竞争地位，以及Silver Lake拟以430亿美元收购Workday的提案。 据报道，Anthropic正筹划在10月进行IPO，基于其达到1000亿至1200亿美元的年化营收运行率，目标估值达到创纪录的2万亿美元。 (2:36) 物理层面的制约（尤其是工业燃气轮机和电力供应限制）已取代芯片供应，成为构建AI超大型数据中心的主要瓶颈。 (12:44) 媒体关于数据中心破坏当地供水和电网的说法大多言过其实；现代设施采用闭环水系统运行，并为当地电网升级提供资金支持。 (20:00) 低成本、高质量开源AI模型采用率的提升将推动企业级应用，同时阻止专有前沿实验室维系长期的商业垄断。 (20:00) AI领域的意识形态分歧体现在：担忧技术广泛扩散的有效利他主义安全拥护者，与担忧权力高度集中的Mark Zuckerberg和Elon Musk等市场支持者之间的对立。 (25:00)

The All-In hosts and guest Gavin Baker analyze the rapid financialization and geopolitical stakes of artificial intelligence, debating whether unprecedented growth forecasts from proprietary labs can survive physical power bottlenecks and open-source competition. Anthropic's IPO Rumors and Revenue Projections (2:36) According to reports cited by the panel, Anthropic is exploring an October public offering targeting a $2 trillion valuation, trading at an estimated 16 to 20 times sales on projected year-end annualized revenues of $100 billion to $120 billion. The company is also reportedly in talks to purchase chip-efficiency startup Descartes for $6 billion. While David Sacks notes that Anthropic's triple-year 10x trajectory would theoretically push it toward $1 trillion in Annual Recurring Revenue, Gavin Baker warns that underwriting leaks often inflate expectations, predicting 2027 exit ARR will land closer to $400 billion to $500 billion due to physical limits. Jason Calacanis expresses deeper skepticism, projecting a steep deceleration toward $300 million to $400 million as open-source alternatives like GLM and Meta's Llama reach performance parity at a 90% discount. Nevertheless, the panel agrees that Anthropic's forthcoming S-1 filing will function as the technology sector's macroeconomic benchmark, providing the first audited GAAP validation for enterprise AI demand across the hardware supply chain. Physical Constraints and Nvidia's $500B Financing Plan The panel underscores that physical infrastructure, rather than capital or silicon alone, has become AI's binding bottleneck. Modern clusters face acute shortages of industrial gas turbines and grid capacity, prompting operators to repurpose retired jet engines for on-site power generation. Baker counters public criticisms of data centers, arguing their water use relies on closed-loop systems and that private microgrids insulate local municipalities while boosting municipal tax revenue. To counter liquidity constraints on data center expansion, Nvidia introduced a $500 billion financing initiative in collaboration with institutions including Blackstone, BlackRock, Goldman Sachs, and KKR. Under this model, third-party lenders fund GPU deployments secured by Nvidia's residual value guarantees on secondary market rentals, with Nvidia taking top-line royalties. Sacks characterizes this mechanism as Nvidia acting as the 'Federal Reserve of AI,' securitizing compute infrastructure into yield-generating assets comparable to commercial aircraft leasing. Open-Source Models and the Ideological AI Divide (25:00) Examining Mark Zuckerberg’s manifesto, 'The Future is for Everyone,' the hosts contrast Meta's decentralized vision with the safety frameworks pushed by Effective Altruists (EAs) and Anthropic. Sacks and Baker reject apocalyptic rhetoric comparing AI to nuclear warheads, contending that regulatory licensing regimes risk creating corporate cartels that stall innovation. Baker advocates for an individual 'right to bear AI' to prevent centralized ideological control, warning that restricting domestic open-source models would surrender technological dominance to China. However, Calacanis highlights contradictions in Meta's posture, observing that while Zuckerberg champions the right of models to distill observable internet data, Meta has historically litigated aggressively against third parties scraping the Facebook and Instagram social graphs. The group also debates labor externalities, contrasting Amazon's delivery subcontractor model with corporate responsibility, as Calacanis argues externalizing healthcare costs fuels public support for regulatory intervention. xAI's Grok 4.6 and Enterprise Software Repricing (87:32) The hosts review xAI's release of Grok 4.6, a 1.5-trillion-parameter model that benchmarks show entering the top tier alongside Claude at lower operating costs. Sacks describes Elon Musk’s dual compute hedge: utilizing massive GPU capacity like the Colossus cluster to train proprietary frontier models while retaining the flexibility—via 90-day cancellation clauses—to rent excess compute to outside labs if training margins compress. The conversation concludes with Silver Lake’s reported $43 billion buyout bid for Workday, driving a 17% stock surge. Baker argues that private equity buyers view mature SaaS multiples as oversold after an 18-month drawdown. Rather than facing total extinction by frontier labs, established software vendors can adopt open-source models internally to trim operational headcounts and expand cash-flow margins. The discussion illustrates that the AI sector is shifting from pure algorithmic exploration to high-stakes capital structuring and physical execution. Whether these multi-trillion-dollar valuations hold depends heavily on upcoming audited disclosures in Anthropic's S-1, the resilience of Nvidia's asset-backed debt model, and whether proprietary labs can outrun rapidly advancing, low-cost open-source models.

本期节目聚焦人工智能产业的资本化重构与产业格局分化。嘉宾 Gavin Baker 与主持团队深入探讨了 Anthropic 传闻中的 2 万亿美元上市预期、Nvidia 联手华尔街建立的 5000 亿美元算力资产证券化模式、开源与专有模型的路线之争，以及底层能源对 AI 扩张的硬性物理制约。 Anthropic 估值争议与算力物理瓶颈 (2:36) 据市场传闻及《金融时报》报道，Anthropic 计划最早于 10 月推进首次公开募股（IPO），目标估值高达 2 万亿美元，其年化营收预计在年底达到 1000 亿至 1200 亿美元。David Sacks 指出，若保持每年 10 倍的增速，该公司理论上将在次年实现 1 万亿美元 ARR。但 Gavin Baker 与 Jason Calacanis 对此持怀疑态度，指出投资银行往往故意泄露过高数字以提高承销门槛，且 Jason 预测开源替代方案将导致其营收放缓至 3 亿至 4 亿美元区间。 算力扩张还面临严酷的实体供应链短缺。从硅片转向发电能力的瓶颈日益凸显，工业燃气轮机和专用叶片在全球仅有两家工厂全天候生产，迫使部分运营商通过拆解退役私人飞机引擎改建地面固定发电机。Gavin 指出，模拟人类大脑计算能力所需的数据中心能耗相当于 100 万户家庭持续运行 6 至 9 个月，执行单次顶级推理查询即可消耗相当于 1000 户家庭的电量，能源短缺已成为制约模型指数级扩张的核心物理屏障。 开源生态突围与中心化监管博弈 (25:00) 针对 Mark Zuckerberg 发布的 6500 字 AI 宣言《未来属于每个人》，讨论深入剖析了有效利他主义（EA）阵营与开源生态之间的意识形态分歧。Sacks 批评 EA 与 Anthropic 试图在华盛顿游说建立类似联邦航空管理局（FAA）或原子能委员会式的牌照监管，本质是将 AI 描述为核武器以便实现行业垄断。若新模型审批像飞机适航认证一样耗时 5 至 8 年，美国将迅速丧失对海外竞争对手的先发技术优势。 与专有模型的封闭监管相对，Meta、Elon Musk 与 Gavin 提倡“去中心化”与“拥有 AI 的权利”。当前开源模型（如 GLM 与 Llama 系列）已能以专有模型 10% 的成本提供接近 Claude Opus 的性能。Gavin 提出顶级闭源前沿模型将扮演“奥本海默”式的任务总协调人，下达子任务给廉价开源模型执行；预计开源将承担 80% 的查询量，但前沿模型仍可攫取 65% 至 85% 的核心经济价值。 Nvidia 算力证券化与宏观需求辩护 Nvidia 联合高盛、黑石、贝莱德及 KKR 等机构推出了 5000 亿美元的计算资产融资框架，将 GPU 转化为类似商用地产或飞机租赁的生息资产抵押证券。在该架构下，Nvidia 提供资产残值担保以确保机构承销商资金安全，承担约 25% 的风险，同时通过上行租金分成转化为高利润率的轻资产计算撮合方，被 Sacks 戏称为“AI 业的美联储”。 针对部分宏观投资者提出的 AI 循环融资泡沫论，Gavin 进行了强力反驳，澄清行业内绝大多数 Token 生成在单位经济层面已实现现金盈利。根据测算，美国拥有约 1.5 亿至 1.6 亿就业人口与 10 万亿至 12 万亿美元总薪资盘，企业若将 5% 至 10% 的劳动力薪酬支出分配至生产力 AI 工具，仅美国本土就将释放 5000 亿至 1 万亿美元的真实企业软件预算。Anthropic 未来的 S-1 招股文件将成为验证整个行业实际财务成色的关键“领航车”。 Grok 突围与企业软件资产价值重估 (87:32) 在多模态与推理竞争中，xAI 凭借 1.5 万亿参数的 Grok 4.6 迅速打破 Anthropic 与 OpenAI 的双寡头格局。在 CursorBench 等基准测试中，Grok 4.6 位于帕累托最优边界的“高智能、低成本”象限。Sacks 强调 Musk 拥有算力双向对冲策略：自建的大规模集群既是训练模型的“看涨期权”，亦可通过预留 90 天取消条款将多余计算资源以现货溢价转租给竞争对手作为“看跌期权”。 与此同时，私募股权巨头 Silver Lake 拟以 430 亿美元收购人力资源软件巨头 Workday 的消息引发关注。在经历了长达 18 个月、股价自 300 美元跌至 100 美元区间的 SaaS 抛售潮后，该交易标志着传统软件资产估值的触底反弹。开源 AI 的普及并未彻底颠覆存量企业软件，反而让私募机构能够利用 AI 大幅缩减 80% 的传统运营冗余，将核心企业数据和留存现金流进一步货币化。 劳工外包争议与电商末端履约成本 (75:00) 针对纽约和新泽西州总检察长起诉亚马逊“配送服务合作伙伴”（DSP）模式以规避法定雇主责任的诉讼，主持团队展开了激烈辩论。Jason Calacanis 批评亚马逊将末端履约风险转移给脆弱的独立分包商，导致社会福利成本被外部化至纳税人，主张每单增加 25 美分成本将司机时薪提至 21 至 22 美元并提供医疗福利。 David Sacks 则依据测算驳斥该提议，指出强行将 DSP 转化为全职雇员将使每件包裹的配送成本上升约 5.20 美元，折合每个美国家庭每年新增 664 美元开支，并危及约 5000 个配送岗位。Sacks 警告过度监管会以保护劳工为名损害供应链效率与消费者物价体验，市场应警惕将私营合同全面公有化或政治化的规制倾向。 本期节目的核心分歧集中在技术发展的物理可扩展性与商业利益分配上。尽管资本市场展现出极高的资产证券化热情，但能源基础设施的刚性短缺、开源对专有模型溢价的挤压，以及围绕反垄断与劳工福利的政治监管，仍将持续对前沿实验室的估值兑现构成实质考验。

## Background 本视频由科技商业播客频道 All-In Podcast 于 2026 年 8 月 14 日发布在 YouTube 平台上。根据页面简介，本期节目邀请了特别嘉宾 Gavin Baker，与主持人共同深入探讨 Anthropic 拟以 2 万亿美元估值冲刺 IPO、扎克伯格发表的 AI 宣言、英伟达高达 5000 亿美元的 AI 融资计划、Amazon 配送服务伙伴（DSP）模式引发的法律争议，以及 xAI 发布的 Grok 4.6 和 Silver Lake 拟私有化 Workday 等重磅科技财经议题。 ## TL;DR 本期节目重点剖析了 Anthropic 以 2 万亿美元估值筹备上市的内幕及其内部超高速增长与隐患，并深入讨论了马克·扎克伯格倡导开源与分布式的 6500 字 AI 宣言。嘉宾们还对英伟达将 GPU 算力打造为可融资资产类别的金融创新进行了拆解，探讨了计算资源过剩与算力链条金融化的潜在风险。此外，节目还就 Amazon 配送外包模式背后的资本主义与社会主义思潮展开了激烈论战，并评估了 Grok 4.6 发布对大模型市场格局及传统 SaaS 软件估值底座的深远影响。 ## Atmosphere & style 整期节目氛围专业且充满高浓度的干货输出，主持人群体与嘉宾之间的互动默契而犀利，既有对前沿技术趋势和硬核金融架构的深邃洞察，也有对商业伦理与社会议题的火爆辩论。讨论语言风格生动而形象（例如将数据中心建设比作硬核美剧《土地机械师》），从对 AI 未来狂妄幻想的警示，到对企业社会责任的激情对峙，展现出硅谷顶级投资圈与思想者们在理性客观分析与强烈观点碰撞交织下的思想盛宴。 ## Key takeaways - [02:36](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=156s) Anthropic 正在筹备公开上市（IPO），目标估值高达 2 万亿美元，计划于 10 月登场，预计年底年化运行收入（ARR）将达到 1000 亿至 1200 亿美元。 - [05:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=300s) 马斯克与 Anthropic 创始人 Dario Amodei 关系和解，并将 Colossus 超级计算集群的算力出租给 Anthropic，体现出算力合作的灵活性。 - [08:13](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=493s) AI 编程赛道已成为头部大模型竞争的核心战场，Anthropic 通过抓住 Cursor 的爆发确立优势，而 OpenAI 也通过优化编程能力实现了显著的月增幅。 - [15:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=900s) AI 基础设施扩张面临物理世界供电和设备的硬瓶颈，燃气轮机和电力供应成为关键制约要素，促使企业探索改装退役飞机引擎等创新供电方案。 - [23:45](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1425s) 开源模型凭借近 90% 的成本优势正在加速企业端普及，这在保障企业数据自主权的同时，也对闭源巨头的高价格模型造成冲击。 - [27:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1652s) 扎克伯格发表 AI 宣言，主张免费开源与分布式 AI 路线，反对将 AI 权力高度集中在少数闭源巨头与政府机构手中。 - [50:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=3000s) 未来的模型生态将呈现开源与前沿模型的共生格局：开源模型处理高频基础任务，而高智商的前沿模型则作为“调度大脑”获取大部分经济价值。 - [55:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=3300s) 英伟达与华尔街顶级机构合作推出 5000 亿美元算力基金，通过残值担保和收入分成机制，将 GPU 算力打造成一种全新的金融资产类别。 - [01:07:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=4052s) 全美劳动力若将 5% 至 10% 的预算用于配置 AI 工具以提升生产力，仅美国市场即可支撑数千亿至上万亿美元的 AI 支出，驳斥了“AI 纯属泡影”的论调。 - [01:14:29](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=4469s) 亚马逊的配送服务伙伴（DSP）外包模式引发了严厉的法律监管与政治争论，凸显出企业极致追求效率与一线员工薪资福利保障之间的深层冲突。 - [01:27:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=5252s) xAI 推出的 Grok 4.6 在基准测试的帕累托前沿上表现优异，马斯克通过高效执行力和跨公司调集核心人才，成功杀入前沿大模型第一梯队。 - [01:35:21](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=5721s) 私募股权巨头 Silver Lake 拟以 430 亿美元私有化 Workday，标志着经历了断崖式下跌的传统 SaaS 软件股正在形成估值底座，展现出 AI 时代应用层重构的潜力。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=0s) 特别嘉宾 Gavin Baker 登场与精简版 Besties 阵容 主持人 Jason Calacanis 宣布新一期节目开播，本期邀请到了 Gavin Baker 作为特别嘉宾，与 David Sacks 共同坐镇。尽管本期录制团队人数较少，但讨论议题极为丰富。Jason 首先向 Gavin 表达了祝贺，提及他此前在 SpaceX 项目上的巨大成功，Gavin 谦虚地回应称 SpaceX 是一个奇迹般的时刻，但他借用名教练 Bill Belichick 的名言“现在是向下一站（Cincinnati）迈进的时候了”，并强调 SpaceX 的故事远未结束，投资人总是需要用最新的业绩证明自己。 交流随后转向了近期热议的政治话题。Jason 提及他与前芝加哥市长、前白宫幕僚长 Rahm Emanuel 进行了一场内容火爆的专访。Jason 总结认为，Emanuel 代表了克林顿与奥巴马时代的民主党温和派立场，极力摒弃极端激进的民主社会主义者（Democratic Socialists）。Sacks 也表达了对党内温和派崛起的期许，认为全美政坛若能出现更多中立、理性的候选人将是莫大的幸事。 --- > "To quote Bill Belichick, you know, it's on to Cincinnati." --- ### [02:36](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=156s) Anthropic 拟以 2 万亿美元估值冲刺 10 月 IPO 节目进入核心重磅新闻讨论：英国《金融时报》（FT）突发报道指出，人工智能巨头 Anthropic 正在筹备公开上市，估值目标高达 2 万亿美元，计划于 10 月正式登场。若此消息属实，该规模将一举打破此前 SpaceX 创下的 1.75 万亿美元 IPO 纪录。预测市场 Polymarket 上的交易员给出 80% 的概率认为 Anthropic 会在年内完成 IPO；同时，Fifth Bestie Brad Gerstner 也曾暗示这一进程正在加速推进。 在业绩方面，Anthropic 的年化运行收入（ARR）预计在今年底达到 1,000 亿至 1,200 亿美元之间。这一营收飙升速度在硅谷历史上前所未有（过去三年每年实现近 10 倍增长）。按照 2 万亿美元估值计算，其市销率（P/S）仅为 16 至 20 倍，远低于 SpaceX 或 Palantir 上市时的估值倍数。此外，市场也传出 Anthropic 正洽购 AI 初创公司 Descartes（交易额约 60 亿美元），旨在通过芯片优化技术显著降低 AI 训练与推理成本。 为了解内部细节，Jason 询问 Gavin 是否持有 Anthropic 股份。Gavin Clarify 自己并非 Anthropic 股东，并打趣道 Anthropic 与 SpaceX 之间的解冻来得稍晚了些，自己通常不会投资埃隆·马斯克（Elon Musk）的直接竞争对手。 - **估值与时间表**：拟定 2 万亿美元估值，计划 2026 年 10 月上市（距录制仅剩 6-8 周）。 - **营收增速**：预计年底 ARR 达 1,000 亿–1,200 亿美元，连续三年同比翻 10 倍。 - **市场份额分布**：Polymarket 显示 Anthropic 在年底保有“最强模型”称号的概率为 67%，但 Grok 与 OpenAI 正迅速缩小差距。 --- ### [05:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=300s) 马斯克与 Dario 的冰释前瞻及 Anthropic 的业务护城河 Gavin 详细补充了马斯克与 Anthropic 创始人 Dario Amodei 关系和解的内幕。马斯克在拥有 Colossus 超级计算集群的庞大算力储备后，决定将其出租给 Dario 和 Anthropic。更重要的是，马斯克对 Dario 的道德品质建立了信任，正如马斯克公开所言：“他们没有触发我的邪恶警报（evil detector）”——要知道马斯克过去曾凭此敏锐直觉迅速斩断了与 Sam Bankman-Fried（SBF）的会面。 探讨 Anthropic 的业务增长时，Gavin 指出即便 Anthropic 在绝对百分比上可能在向 OpenAI、开源模型及 Grok 输掉部分市场份额，但整个 AI 饼图（TAM）正在以几何级数扩张（包括未被统计的开源“暗黑 Token”），因此其绝对收入依然呈爆发式增长。Gavin 引用了好友 Eric Vistria 两年前的思想实验：即便 OpenAI 和 Anthropic 在最底层的模型竞争中失利，凭借其卓越的产品封装、工作流套件（harness）以及沉淀的用户习惯，依然具备极高的商业价值。 关于 2 万亿美元的 IPO 估值，Gavin 保持了一定谨慎。他指出投行泄露给媒体的高估值往往是未争取到主承销商（Lead Left）位置的银行所为，意在抬高门槛让主承销商难堪。如果公司以 2 万亿美元定价上市，很可能是为了给员工与后续解禁期留出缓冲空间，避免高开低走引发剧烈波动。 --- ### [08:13](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=493s) 编程赛道突破、OpenAI 的反扑与万亿营收假想 David Sacks 深入剖析了 Anthropic 的成功路径。Anthropic 此前将战略重注押在 AI 编程（Coding）领域，事后证明极具远见。Gavin 补充道，Anthropic 很大程度上是因为观察到 Cursor 等新兴工具在其平台上构建并爆火，随后果断决定亲自入局主导这一市场。 然而，随着行业巨头察觉到这一趋势，竞争正在急剧加剧： OpenAI 在推出 GPT-5.6 后全面转向编程优化，据内部消息其月增长率已冲上 20% 以上（换算成年化增长超过 10 倍），增长引擎重新被点燃。Sacks 提出，虽然 Anthropic 目前势头强劲，但如果按照每年 10 倍的增速外推，其 ARR 将在明年底达到 1 万亿美元。这就带来了一系列物理与商业极限的追问：全球 TAM 是否足够大？电力供应与数据中心算力能否跟上？ 针对这一极其激进的增长假想，Gavin 透露了 Anthropic 内部近乎狂热的自信：内部高层甚至声称“Anthropic 最终可能会成为全球唯一的私营公司”，届时世界上将只剩下政府和 Anthropic。Gavin 警示这种言论带有强烈的自大与狂妄（Hubris），直言这开始有点陷入 SBF 式的虚妄幻想，极易导致“骄兵必败”（Pride cometh before the fall）。 - **编程战略**：Anthropic 观察到 Cursor 的爆发后直接下场统治 AI 编程生态。 - **OpenAI 追赶**：GPT-5.6 转向编程后，月增速突破 20%，重新展现出 10 倍年化增长势头。 - **内部高层豪言**：内部曾传出在 600 亿美元 ARR 时期规划一年内冲击 6,000 亿美元，甚至宣称要成为“世上唯一的私营企业”。 --- ### [15:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=900s) 供给侧瓶颈：电力、燃气轮机与“Billy Bob Thornton”式硬核建设 在讨论 Anthropic 能否将收入从千亿继续提升至万亿美元时，Gavin 将问题拆解为需求（Demand）与供给（Supply）两个维度： 在**需求端**，全球知识工作者（Knowledge Work）的市场规模在 25 万亿至 65 万亿美元之间。目前的 AI 并非简单的劳动力替代，而是大幅加速了生产力与需求增长（例如目前软件工程岗位的招聘需求比一年前更高）。只要这种加速效应持续，未来 18 个月内就能在国家经济账本中看到明显的 GDP 拉动。 在**供给端**，真正的考验在于物理世界的重工业建造（Atoms, not bits）。数据中心建设需要极度硬核的现场调度，Gavin 形象地将其比作美剧《土地机械师》（Landman）里的物理世界——在二叠纪盆地（Permian Basin）摄氏 44 度（110 华氏度）的高温下，需要像 Billy Bob Thornton 扮演的角色那样，在偏远荒漠中组织数千名工人与毒枭威胁抗衡。 目前全球数据中心面临的核心供应链瓶颈之一是冷切叶片燃气轮机（Turbine）。全美及欧洲仅有两个大型工厂能够生产此类 40 吨重压机设备。然而，资本主义的市场机制正在快速自我修复： 1. 燃机工厂正在开展 24 小时三班倒生产，西门子能源（Siemens Energy）等巨头全力扩产。 2. 鉴于私人飞机二手残值飙升，大量企业开始直接拆卸退役喷气式飞机的引擎，将其改装为数据中心发电的燃气轮机。 3. 埃隆·马斯克据报道已亲自收购了一家燃气轮机公司，以求用最快速度推进数据中心电力自给。 针对近期政治与监管动态，Gavin 提到德克萨斯州州长 Greg Abbott 宣布对全州数据中心开展能源审计。只要项目方采取自带能源（BYOE, Bring Your Own Energy）方案，就不会面临阻碍；该举措主要是为了防止缺乏独立供电方案的数据中心拖垮德州脆弱的电网。针对舆论对数据中心耗水耗电的指责，Gavin 澄清许多主流报道数据存在数万倍的夸大误差。实际上，数据中心落地 Ellendale 等衰落小镇后，不仅使当地税收翻了 10 倍，更重新激活了整个区域经济。 ### [20:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1200s) 数据中心恐慌与媒体虚假宣传 围绕 AI 数据中心的各种恐慌和负面报道，实际上是媒体制造虚假新闻与社会焦虑的典型例证。关于数据中心耗尽社区水资源的论调完全是谎言——数据中心的水资源消耗完全可控且低于标准高尔夫球场，况且水资源在系统中循环使用。此外，若数据中心建立自给自足的发电设施，不仅能将剩余电量反哺电网、承担电网升级费用，还能降低当地电价。蓝州（民主党主导州）电价上涨的真正原因，在于激进的去碳化政策与繁重的绿色环保法规，而非数据中心。 数据中心引发的噪声污染确实存在，但解决方案非常简单且符合常理——只需将数据中心选址在远离居民区的广袤土地上即可。全面禁止数据中心建设的做法愚蠢至极；相反，积极拥抱数据中心建设的社区正通过增加的税收收入，大幅改善当地的学校与公共服务。 尽管短期内新建数据中心仍需依赖天然气作为最快速、可扩展的能源，但 AI 带来的庞大能源需求正在加速对太阳能、电池与核能等清洁能源的投资与规模化应用。这不仅能降低清洁能源的价格，对全人类也是一项长期利利好。美中不足的是，主流媒体抹黑数据中心造福小镇及低收入群体的行为令人遗憾，亟需有人站出来揭露真相。 - 相比高尔夫球场每年消耗数亿加仑的水资源，数据中心的用水完全在可控范围内。 - 埃隆·马斯克（Elon Musk）正在德克萨斯州投资 100 亿美元新建太阳能制造工厂。 - 天然气作为最清洁的碳基燃料，将是近一两年内支撑数据中心规模化扩展的核心过渡能源。 --- ### [23:45](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1425s) 开源模型对 Anthropic 的市场冲击 虽然 Anthropic 等头部 AI 公司的增速惊人，但其明年的收入增长预计将大幅放缓。由于企业和创业公司开始广泛采用开源模型（如 GLM 5.2），Anthropic 等闭源巨头明年极难实现数万亿美元的营收目标，其收入更可能停留在 3 亿至 4 亿美金区间。开源模型的调用成本较 Claude Opus 等顶级闭源模型便宜了近 90%，性价比优势极为突出。 正如过去二十年的 IT 发展史所证明的那样，企业界绝不希望被单一科技巨头锁死或勒索，因此开源始终是企业更安全、更可控的选择。这种趋势正在企业界和创业圈加速蔓延。 尽管 Anthropic 的团队秉持良好初衷，通过其“宪法 AI”（Constitutional AI）制定他们认为最有利于人类的规则，但充满混沌与不可预知的现实世界往往伴随着意想不到的后果。相比之下，开源生态带来了多元与丰富的 AI 体系，完美避免了少数单一主导模型掌控世界的风险。 - 相比 Claude Opus，新兴开源模型的调用成本降幅高达 90%。 - 企业界在过去 20 年选择开源系统的核心逻辑，是为了防止被大型科技巨头“抽地毯”（Rug Pull）或技术勒索。 - 埃隆·马斯克创立 Grok 并坚持开源，旨在追求客观中立的真相与技术的全民共享。 --- ### [27:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1652s) 扎克伯格 AI 宣言：分布式对决集中化 马克·扎克伯格（Mark Zuckerberg）发表了一篇长达 6,500 字的 AI 宣言《未来属于每个人：通往积极 AI 未来的道路》（*The Future is for Everyone: The Path to a Positive AI Future*），接过了由 OpenAI 放弃的“通用人工智能福利”大旗。扎克伯格在文章中勾勒了一个极其乐观的 AI 世界观：通过免费发布开源模型，为全球每个人配备专属 AI 智能体与导师，从而创造无尽的物质丰裕。 扎克伯格在宣言中直击有效利他主义者（EA）与末日论者（Doomers，如 Anthropic 创始人 Dario Amodei）的逻辑矛盾：如果他们真认为 AI 未来如此反乌托邦、会导致大量人员失业甚至毁灭人类，为什么还要如此狂热地去创造它？这种矛盾背后的本质，是一种“受选者的远见”（Vision of the Anointed）——知识分子自以为只要掌握绝对权力，就能将社会导向善良，然而这种思维往往是极权计划与国家权力滥用的开端。 AI 发展路线的真正分歧，不仅在于开源与闭源，更在于**集中化（Centralized）与分布式（Decentralized）**的路线之争。以华盛顿政客和 EA 机构为代表的集中化派系，试图将 AI 塑造成极度危险的权力垄断产物，通过限制芯片和模型出口，将权力集中在少数两三家与政府深度绑定的“垄断企业”（Cartel）手中；而分布式派系则主张赋予每个个体充分的 AI 掌控权与数据主权。 > “AI 如此危险以至于唯一安全路径就是极度集中权力的概念，本身就存在根本性问题。纵观历史，寄希望于绝对权力在足够开明后善待人类，从未带来过更安全或积极的结果。” —— 马克·扎克伯格 - 扎克伯格的 AI 宣言长达 6,500 字，勾勒了开源、免费且分布式的 AI 愿景。 - 宣言指出“超智能是发明而非自动化”，AI 的安全应建立在权的平衡而非单一集中控制之上。 - 拜登政府前官员及智库曾试图将 AI 监管套用“原子能委员会”模式，通过限制出口和企业与政府联姻实现集中控制。 --- ### [33:30](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=2010s) 技术安全观与个人 AI 权利 Anthropic 及有效利他主义（EA）运动的核心信念是“AI 技术过于危险，因此不能散布（Too dangerous to distribute）”；而马克·扎克伯格、埃隆·马斯克与黄仁勋（Jensen Huang）等人的信念则是“AI 技术过于危险，因此绝不能集中（Too dangerous to centralize）”。人类历史已经多次证明，当面对这种抉择时，去中心化与民主化永远是更好的道路。 EA 派系误以为通过竞争会导致“向底线竞争”（Race to the bottom）并丧失安全性，但健康的自由竞争实际上能激发出技术最优秀的一面。将 AI 比作“核武器”并试图建立全球监管机构的论调完全是荒谬的脑补与幻觉——没有普通消费者或企业需要核武器，但每一个普通人和企业都需要 AI。AI 本质上是一种普惠的消费级技术，而非纯粹的军用灭世武器。 正如美国宪法赋予公民持枪权以防御暴政一样，在未来的数字时代，每一个个体都应当拥有属于自己的 AI 权利。公民有权拥有一款价值观与自身保持一致、保护自身利益的 AI，而不是由某个自诩为“AI 救世主”的大公司 CEO 来决定全人类的利益所在。 - 有效利他主义者（EA）深信自己能赚尽天下财富、打造出单一的“印钞机公司”来拯救全人类，这种救世主心态是典型的夸大妄想（Delusions of Grandeur）。 - 主张禁绝开源与集中控制，不仅会导致美国在 AI 与地缘政治竞争中败给中国，还将彻底剥夺个体的技术自主权。 - 应对未来潜藏威胁的最佳策略并非向少数巨头让渡权力，而是让每个人都拥有能够代表并保护自身利益的分布式 AI 工具。 ### [040:00] 扎克伯格 AI 宣言背后的争议与监管风险 节目的对话围绕马克·扎克伯格（Mark Zuckerberg）最新发布的 AI 宣言展开。嘉宾们深入剖析了这份宣言中的两大要点：第一，扎克伯格对自动化与就业替代的看法十分客观，他指出虽然人们担心自动化超越个人能力增长，但近期数据表明后者的提升完全可以匹配甚至超越前者。这正如英伟达 CEO 黄仁勋所言：“你不会被 AI 淘汰，但会被掌握 AI 的人淘汰。”第二，扎克伯格强调所有 AI 模型都源于人类知识，并主张保护“从可观察到的任何事物中学习”的蒸馏权利。然而，主持人指责这存在明显的双重标准与伪善： Meta（Facebook/Instagram）多年来起诉了无数试图对其社交图谱进行索引或蒸馏的初创公司，将自身数据严密封锁，如今却呼吁全网数据的免费蒸馏。 随后，话题转向 Anthropic 等前沿模型公司面临的商业模式与监管危机。大卫·萨克斯（David Sacks）指出，如果 Anthropic CEO 达里奥·阿莫代伊（Dario Amodei）之前在华盛顿推行的强监管体系（如类似 FAA 或 FDA 动辄数年的审批制度）得以实施，其公司高达两万亿美元估值的 IPO 计划将彻底破灭。因为前沿模型之所以能够收取高昂的 Token 溢价，完全建立在领先开源模型 6 个月的时间差之上；一旦引入漫长的监管审批，这 6 个月的领先优势将瞬间荡然无存，产品沦为通用商品，甚至被中国竞争对手赶超。 * **扎克伯格宣言的核心观点**：提出了“个人能力增长可赶上自动化步伐”的认知框架，打破了传统的就业恐慌论。 * **双重标准质疑**：Meta 一方面捍卫 AI 模型“蒸馏人类知识”的权利，另一方面却针对索引其 Instagram 社交图谱的初创公司发起了数百起诉讼。 * **监管对商业模式的毁灭性打击**：航空（FAA）与药物（FDA）式的 AI 审批机制将直接蚕食 Anthropic 6 个月的技术领先期，剥夺其定价权。 --- ### [045:00] 科技 CEO 的“Go Direct”趋势与扎克伯格 2.0 随着网络舆论场的演变，科技巨头的顶级 CEO 们正在掀起一场“去中介化（Go Direct）”的沟通风潮。扎克伯格在社交平台 X 上亲自撰写并发布长文，经 AI 检测证实为 100% 人类纯手工打造。包括 Palo Alto Networks 的 CEO 尼凯什·阿罗拉（Nikesh Arora）、英伟达的黄仁勋在内的多位高管，也纷纷激活或亲自运营 X 账号直接发声。讨论指出，如果公共人物或企业 CEO 仅依赖公关部门（PR）作为代理人，在面对国会问责或舆论危机时将极为脆弱；在个人账号上建立真实的声音不仅是表达真相的渠道，更是抵御外部风险的“铠甲”与“超级能力”。 与此同时，嘉宾们打趣探讨了扎克伯格近年来的形象蜕变。从电影《社交网络》中杰西·艾森伯格（Jesse Eisenberg）扮演的生硬形象，演变为《继承之战》中杰里米·斯特朗（Jeremy Strong）式的硬派风格。他在夏威夷和太浩湖冲浪、练习 MMA 的视频展现出了独特的个人气场（Aura）。大家调侃希望看到他手持美国国旗在考艾岛（Kauai）的公牛鲨水域冲浪，并鼓励更多 CEO 展现真实的自己。 * **个人品牌的防御价值**：CEO 亲自发声（Go Direct）被定义为企业公关之外的“武装与印记”，能够防范未来的监管与舆论审查。 * **形象转变（Zuck 2.0）**：从过去的硬塞公关话术，转变为亲力亲为、展现极客与硬汉特质的真实人格。 --- ### [050:00] 开源与前沿模型的共生关系及 Anthropic 营收预测 对于开源模型与商业闭源模型的关系，加文·贝克（Gavin Baker）提出了全新的解构角度：开源模型非但不会杀死前沿模型，反而会让顶级 Token 更加昂贵且有价值。他以曼哈顿计划（Manhattan Project）做类比：即便有众多诺贝尔奖级别的顶尖物理学家，仍需要奥本海默这样的核心领袖进行统筹调度。同理，一个 IQ 250 的前沿模型（如 Anthropic、OpenAI 或 Grok）可以完美编排管理无数个 IQ 150 的开源模型，将其作为分工工具。预测未来的市场格局将是：开源模型占据 80% 的使用总量（Volume），而前沿模型占据 65% 至 85% 的经济价值（Economic Value）。 在谈及 Anthropic 的财务表现时，嘉宾们展开了激烈的 Over/Under 下注猜想。尽管市场预测 Anthropic 到 2026 年底将达到 1000 亿美元的 Run Rate，但在预测 2027 年的退出年化营收（Exit ARR）时，大家普遍认为 4000 亿至 5000 亿美元是一个现实可期的区间（相当于从 100 亿到 4000 亿实现 40 倍暴涨）。此外，加文还提到了人类大脑的极致能效比：为了在数据中心中模拟人类大脑的智力水平，需要消耗数十万至上百万户家庭的电能运行数月；相比之下，人脑在能源利用效率上具备物理极限层面的巨大优势，这也注定了人类智慧在极高计算成本时代依然不可或缺。 * **前沿与开源模型的二元格局**：开源承担高频基础工作，前沿模型扮演“调度大脑”，拿走大部分市场经济价值。 * **营收暴涨预测**：预测 Anthropic 退出 ARR 将在下一年达到 4000 亿至 5000 亿美元。 * **人脑能效比优势**：相较于数据中心动辄数万户家庭用电量的推理消耗，人类大脑是地球上计算效率最高、最省能源的智能形态。 --- ### [055:00] 英伟达 5000 亿美元 AI 计算融资计划与资产证券化 节目插播了 2026 年 All-In Summit 峰会的最新阵容与福利，随后重磅聚焦于英伟达的金融创新。英伟达宣布与高盛（Goldman Sachs）、黑石（Blackstone）、贝莱德（BlackRock）等华尔街巨头合作，计划筹集 5000 亿美元的 AI 计算基金。黄仁勋在 X 上撰文，将“GPU 计算”定义为一种全新的“可投资资产类别（Investable Asset Class）”。在这种新型融资机制下，企业无需一次性支付数十亿美元购买芯片，而是通过借贷购买英伟达系统，随后通过出租计算资源（Inference/Compute Rental）产生的现金流偿还贷款，其逻辑类似于传统的汽车融资（如福特、通用金融）或住房抵押贷款支持证券（MBS）。 嘉宾加文·贝克分析认为，华尔街顶级资产管理公司（收费标准高达 2/20）愿意为英伟达背书并出资，标志着顶级金融资本正式认可了 GPU 作为可计算现金流折现的优质资产。虽然市场上存在对“循环融资（Round Tripping）”和计算资源过剩的担忧，但这种独立的第三方融资平台成功将硬件采购转化为可分期的资产运营成本。与此同时，市场上亦有超大规模计算厂商（Hyperscalers）开始将上千亿的底层架构从纯闭源前沿转向开源生态，推动了整个 AI 算力链条的金融化与普及化。 * **资产类别的重构**：英伟达将 GPU 算力打造为可产生稳定现金流、可独立融资的金融资产类别。 * **华尔街资本入局**：黑石、高盛等顶级机构设立独立融资平台，为企业租赁算力提供杠杆，降低前瞻性 AI 基础设施的门槛。 * **商业模式转型**：从一次性资本支出（CapEx）转向基于出租收益偿还债务的运营流模式。 根据你提供的逐字稿部分，以下是详细的对话记录与内容复盘： ### [01:00:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=3600s) 英伟达的商业奇招：提供残值担保与算力资产化融资 英伟达正在扮演一个“牵线搭桥”的角色。因为英伟达 GPU 的底层架构极其灵活，其使用寿命足够长，因此能够以比其他专用算力更低的利率进行融资。目前三大（乃至四大）中国开源大模型（如 Qwen、Kimi、DeepSeek 以及 GLM）的架构演进路线各不相同，而这种架构的多样性恰恰对英伟达和 GPU 极为有利，因为市场需要兼具灵活性和长寿命的计算资源。 为了帮助扩大并培育这个市场，英伟达正在实施一项非常聪明的举措：他们联络了 Blackstone、Goldman Sachs、KKR、Apollo、BlackRock 等六家顶级私人股权和华尔街金融机构，提出了残值担保（Residual Value Guarantee）方案。如果这些机构参与交易，英伟达将担保在 3 年或 4 年后，这些 GPU 仍能以特定速率出租。这项担保大大降低了融资成本，并被直接纳入承销模型的风险评估中。英伟达凭借对算力供需的顶尖监控能力（Telemetry），只需承担约 25% 的残值风险，就把极其庞大的融资资本撬动了起来。 当被问及“这是否类似于次级房贷（MBS），究竟什么环节可能出问题”时，讨论转向了硬件的实际寿命。CoreWeave 表示，其 2020 年推出的 Ampere 架构 GPU 到了 2029 年依然能够维持盈利性的出租，这意味着该批硬件拥有长达 9 年的生命周期，就像古董车被转卖到海外继续发光发热一样。 但在理论上，如果强行寻找风险点，最大隐患不在于需求不足，而在于算力过剩（Glut of compute）导致的过度建设。类似于互联网泡沫破裂后的“暗光纤”（Dark Fiber），如果出现大量闲置的“暗 GPU”，对于那些按每瓦 30 至 50 美元现货价格预期建设计算基础设施的公司来说将是灾难。不过，政治阻力（如建设数据中心面临的舆论恐慌与各种阻碍）反而形成了一种反向保护，变相遏制了算力供给过剩的风险。 此外，英伟达此举最绝的一招在于：作为提供低风险担保的交换，英伟达还能获得保底收益之上的收入分成（Revenue Share）。摩根士丹利（Morgan Stanley）近期在一份报告中指出，这些收入分成实际上就是特许权使用费（Royalties），英伟达借此可以极其迅速地演变成一家高利润、轻资产的巨型云服务商。黄仁勋（Jensen Huang）之所以这么做，是因为 AI 的总可寻址市场（TAM）太庞大，已经受到了资本融资能力的限制；他通过解除融资枷锁，让市场的实际规模得以彻底释放。 例如，埃隆·马斯克（Elon Musk）计划明年新增 6 到 8 吉瓦（GW）的算力，资本支出高昂。虽然公司刚完成了 1000 亿美元的股权与债权融资，但仍需进一步寻求资金支持。英伟达利用这些大银行和私募股权机构构建的信用额度，本质上是在用 GPU 未来产生的预期现金流做资产证券化融资（类似于航空公司的飞机融资），让下游买家全面受益。英伟达正在借此成为“AI 领域的中央银行”。 - 英伟达合作的六大顶级金融机构包括：Blackstone、Goldman Sachs、KKR、Apollo、BlackRock 等。 - CoreWeave 确认：2020 年发布的 Ampere 架构 GPU 在 2029 年仍能实现盈利出租，生命周期达 9 年。 - 摩根士丹利评估：英伟达通过收入分成机制，正在转化为轻资产型的巨型云服务商业形态。 ### [01:07:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=4052s) AI 泡沫辩论：熔断机制与 Anthropic 上市的“领航车”效应 在算力建设狂潮中，物理世界的建设周期（电力、数据中心建设）是一个天然的“调节阀”（Throttle），而主要客户则构成了“死人开关”（Dead Man's Switch，即列车驾驶员若发生意外松开手，列车就会自动刹车）。目前最大的两家算力买家是 Anthropic 和 OpenAI。如果需求不及预期、算法效率大幅提升，抑或是开源模型带来阻力，这些巨头就会松开购买按钮，整个链条便会迅速降温，从而避免引发系统性的次贷危机。 正因如此，Anthropic 的 IPO 计划对整个市场至关重要。一旦 Anthropic 成为上市公司，其按 GAAP（通用会计准则）合规披露的季度财报将成为全行业最重要的“领航车”（Pace Car）和需求信号灯。 目前市场上许多宏观投资人和价值投资人断言“AI 是个巨大泡沫、缺乏 ROI（投资回报率）、Token 价格全靠补贴”。但这种观点完全脱离了实际。事实是：Anthropic 已经实现盈利并产生现金流，开源模型的 Token 销售能够盈利，OpenAI 和 SpaceX 的算力业务也极具盈利能力。整个产业链（从 Anthropic 到 SpaceX，再到英伟达，以及上游的台积电、美光、SK 海力士）中的大部分 Token 交易都是能够带来正向现金流的。 如果将视角拉高到全美宏观经济来看：美国约有 1.5 亿至 1.6 亿雇员，薪酬总额达 10 万亿至 12 万亿美元。企业为每位年薪 8 万美元的员工支付 10%（即 8,000 美元）或至少 5%（4,000 美元）的 AI Token 费用来提升其生产力，逻辑上完全通顺——这就像当年企业为员工配置 SaaS 软件、电脑和手机一样。光是在美国市场，AI 支出就能轻松达到 5,000 亿至 1 万亿美元的规模。因此，所谓“AI 融资将陷入自我循环并崩盘”的末日论调并不成立。 - 算力产业链利益传导：Anthropic 每吉瓦算力可创造巨大收入，向 SpaceX 支付现货算力费用，SpaceX 再向英伟达采购芯片，英伟达随后向台积电、美光（Micron）和 SK 海力士（SK Hynix）下单。 - 宏观经济逻辑：全美 1.5 亿以上劳动力，若将薪资的 5%~10% 用于配置 AI 增强工具，仅美国本土的 AI 市场规模即可达到 5,000 亿至 1 万亿美元。 ### [01:14:29](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=4469s) 亚马逊 DSP 外包模式引发争议：新泽西州起诉与纽约政治风暴 话题转向了亚马逊（Amazon）在物流配送领域的法律与政治争议。新泽西州总检察长起诉了亚马逊，指控其通过“配送服务合作伙伴”（Delivery Service Partners, 简称 DSP）这一第三方外包模式来规避责任、剥削司机。与此同时，纽约市议员、民主社会主义者 柔然·曼达尼（Zohran Mamdani）也加入声讨，发布视频要求通过市政法规强制亚马逊将外包司机直接录用为正式员工。 批评者指出，DSP 模式使亚马逊能够在发生交通事故死亡等惨剧时切断法律责任，同时避开了工会谈判和雇主劳动义务。然而，从市场与商业效率的角度来看，DSP 模式是市场自然演进的结果： 1. **控制消费成本**：研究表明，如果强制取消 DSP 独立承包商模式，普通家庭每年可能需要额外支付数百美元的物流费用，配送速度也会大幅变慢。 2. **弹性与就业创造**：DSP 模式允许物流系统在假日高峰期快速扩容，并在纽约等高密度城市维持低成本运营，同时也为数以千计的小企业主创造了创业机会和就业岗位。 针对政客的干预，针对性的回应是：应该坚持合同自由与自愿交易原则。亚马逊的物流效率远远超越了任何政府公共服务（下单后当天或次日即可送达），如果政府强行插手干预并加重运营负担，最合理的做法就是直接在账单上列出一项“柔然·曼达尼附加费”（Zohran Mamdani Fee），让消费者清楚地看到是哪项政策导致了物价上涨。 其实，这些 DSP 司机的时薪（通常在 18 至 22 美元之间）在当地并不算低，该模式在员工、外包商和亚马逊三方之间本属于互利安排，盲目通过行政手段将其取缔只会损害市场的灵活性与整体效率。 - 争议核心：DSP（Delivery Service Partners）外包配送模式。 - 政治与法律动向：新泽西州总检察长提起反垄断与劳动诉讼；纽约政治家要求立法强制直雇。 - 薪资水平：DSP 司机在相关城市地区的时薪约为 18 至 22 美元。 ### [01:20:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=4800s) Amazon配送模式引发的争议与资本主义的反思 Jason Calacanis 继续针对 Amazon 的配送服务伙伴（DSP）模式发起猛烈批评，指出这种架构被精心设计出来的唯一目的，就是帮 Amazon 规避雇员责任，并将这些底层员工的失业、医疗与福利负担转嫁给美国纳税人。纽约州和新泽西州对其提起诉讼完全是理应当之。Jason 表示，作为 Amazon 的资深长期客户与 Prime 会员，他认为将这些一线配送员纳入 Amazon 官方正式员工体系的成本极低——在纽约每单仅需增加大约 0.25 美元，就能将他们的时薪从 18-19 美元提高至 21-22 美元并提供正规福利。他痛批科技与商业领域存在严重的盲点：过分追求极致效率，以至于演变成对社会最弱势群体的剥削，这违背了资本主义的应有之义。如果 Amazon 不主动自我纠偏，民主社会主义者（他称为“社会主义共产党人”）将在未来的选举中利用这一极具说服力的议题大做文章。 David Sacks 则质疑 Jason 的逻辑，指出如果在 Uber 的案例中利用独立承包商被抨击，那么 Amazon 通过 DSP 模式至少还保留了应对节日高峰的弹性。Sacks 追问 Jason 究竟该如何划清哪些人留任 DSP、哪些人转为 Amazon 直雇。Jason 回应称，这些 DSP 公司规模极小且高度分散，一旦发生严重交通事故就会瞬间破产以帮 Amazon 隔绝法律责任，而配送员们其实极度渴望获得 Amazon 的 RSU 股权与健康保险，他们干的是每天 10 到 12 小时的全职高强度工作，并不像 Sacks 所说的那样只是享受兼职弹性。Jason 强调，当年霍华德·舒尔茨（Howard Schultz）主动为星巴克咖啡师提供生活工资与福利，是资本主义历史上最明智的举措之一；资本家们必须反思自己是如何一步步推动年轻一代走向社会主义的。 - **配送升级成本推算**：在纽约市每送达一个包裹仅需增加大约 0.25 美元的成本，即可实现薪资与福利的全面提升。 - **配送员工作现状**：配送员普遍承担每天 8 至 12 小时（多为 10 至 12 小时）的高强度全职班次，且多次尝试组织工会却遭到抑制。 - **资本主义的自我修正**：Jason 强调自己作为 Amazon 长期大股东，主张企业应主动承担社会责任，避免将人力成本转嫁给政府救济（如食品券、失业金与 Medicaid）。 ### [01:25:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=5100s) 消费者成本与社会主义思潮的正面交锋 针对 Jason 充满情怀的演讲，David Sacks 给出了残酷的经济数据回应。他指出，理想化的诉求固然动听，但如果按照诉求全面改制，每个包裹的真实配送成本将暴涨约 5.20 美元，导致每个纽约家庭每年多支出 664 美元，并可能危及约 5,000 个工作岗位。Sacks 强调，当民众作为选民时，或许会被社会主义的宏大叙事所打动；但当他们切换回消费者身份时，核心诉求只有“便宜”。一旦包裹配送费飙升，消费者会立刻对这种社会实验产生强烈反感。 在 Jason 因网络连接问题暂时掉线后，主持人将话题转向人工智能，邀请 Gavin Baker 评价 xAI 刚刚发布的 Grok 4.6 以及配套的 GrokBot 产品。Sacks 询问 Gavin，此前普遍认为前沿 AI 大模型领域已经缩减为 Anthropic 与 OpenAI 形成的“双头垄断（Duopoly）”，Grok 4.6 的横空出世是否打破了这一格局。 - **消费者真实经济负担**：若按诉求强制转型，每单包裹成本增加 5.20 美元，单个家庭年均支出增加 664 美元。 - **就业与市场冲击**：强制改制预计将直接威胁约 5,000 个配送岗位。 ### [01:27:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=5252s) Grok 4.6 的技术颠覆与 Elon Musk 的突破 Gavin Baker 证实，Grok 4.6 确实展现出了极高的颠覆性，特别是在基准测试和帕累托前沿（Pareto frontier）图表上的表现非常惊人。图表清晰地展示了在横轴（成本，向右更便宜）与纵轴（智能与能力，向上更强）上，Grok 4.6 成功杀入了右上角的理想区域，实现了高性价比与强性能的兼得。即使在 Databricks（估值达 1900 亿美元的权威机构）等第三方基准测试中，Grok 4.6 的表现也超越了此前被公认为“黄金标准”的模型（如 Fable 5）。Ruby on Rails 的创始人 David Heinemeier Hansson (DHH) 以及 37signals、Mercor 等团队均对其在真实开发环境中的效果给予了高度评价。 <Image alt="Grok 4.6 在基准测试帕累托前沿上的位置图表" caption="Grok 4.6 帕累托前沿性能图" src="image_agent_tag_2850901992980053873"/> --- Gavin 分析了埃隆·马斯克（Elon Musk）能在短短六个月内带领 xAI 实现惊人追赶的两大核心原因：第一，建立了执行力极强、高度聚焦的团队；第二，马斯克对 xAI 的管理层进行了大刀阔斧的重组，直接从 SpaceX 和 Cursor 调集了最顶尖的“杀手级”核心人才。马斯克虽然在时间节点上经常迟到，但最终总能超额交付。此外，本次发布的 Grok 4.6 仅是一个 1.5 万亿参数（1.5T）的中型模型，而能力更强的大尺寸模型 Grok 4.7 已蓄势待发，将在几周内正式发布。 - **Grok 4.6 参数规模**：当前发布的模型大小为 1.5 万亿参数（1.5T）。 - **后续模型规划**：性能更强的大模型 Grok 4.7 预计在数周内推向市场。 - **应用生态匹配**：GrokBot 的推出被比作 AI 领域的“OpenClaw 时刻”，大幅降低了普通人定制与使用 AI 智能体的门槛。 ### [01:30:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=5400s) SpaceX 的算力对冲与开放生态博弈 节目深入探讨了 xAI 与 SpaceX 之间独特的算力布局与商业闭环。马斯克在组建超大规模算力集群时，赋予了自己极具优势的期权组合：如果成功训练出顶尖前沿模型，xAI 就能一举跻身第一梯队；即使模型未达预期，也能将庞大的算力作为现货（Spot compute）租售给 Anthropic 等前沿实验室。据报道，马斯克在与 Anthropic 的算力协议中加入了 90 天取消条款，这使得他不仅能获取稳定收益，还能在算力紧张时随时撤回算力用于自身的模型训练。 Gavin 补充指出，许多二级市场投资者在评估 SpaceX 时，往往局限于星链（Starlink）的直连手机业务（Direct to Cell）或轨道算力，却忽略了 Grok 的巨大潜能。考虑到 Anthropic 的估值在极短时间内完成了暴涨，Grok 作为具备极高能力且完全掌控算力底座的 AI 力量，其价值应当被重新注入到 SpaceX 的整体估值模型中。 - **算力租约结构**：协议包含 90 天双向取消条款，使得算力定价基本锁定在“现货价格减 90 天溢价”的区间。 - **算力控制权**：马斯克保留了在算力紧缺时期随时从客户处收回算力、优先供给自身模型的权利。 ### [01:35:21](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=5721s) Silver Lake 拟收购 Workday 与软件股的估值底座 在节目接近尾声时，突发重大商业新闻：私募股权巨头 Silver Lake 正洽谈以约 430 亿美元（股价受消息刺激瞬间暴涨 17%）私有化企业级软件巨头 Workday。这一事件被全场视为公开软件市场的重要拐点。过去 12 到 18 个月里，传统 SaaS 软件股经历了断崖式暴跌（Workday 股价曾从 300 美元一路下挫至 100 美元），市场一度担忧 AI 模型层会彻底吞噬软件应用层的价值。 Gavin Baker 与 David Sacks 分析指出，私募股权基金（PE）重新重仓入场，标志着公开市场可能过度低估了传统软件资产。随着开源 AI 模型的迅速崛起与竞争加剧，应用层企业不再被单一的大模型巨头垄断捆绑。像 Bending Spoons 这样的新锐收购者已经证明，利用 AI 改造传统软件公司——裁减 80% 的冗余人员，保留 20%“AI 优先”的核心员工——能够创造出极高的现金流与盈利能力。Workday 的收购案证明，软件行业的估值底座已经形成。 - **Workday 交易体量**：Silver Lake 洽谈收购总估值约为 430 亿美元，消息公布后股价直线上涨 17%。 - **历史股价轨迹**：Workday 股价在此前软件股熊市中从 300 美元高点直接重挫至 100 美元区间。 - **PE 运营新范式**：以 Bending Spoons 为代表的私募运作模式，通过优化 80% 的人力结构、转向 AI 驱动，大幅提升了软件企业的现金流产出。 <FollowUp label="想深入了解 Grok 4.6 与 Anthropic 模型在大模型性能图表上的具体对比吗？" query="请详细对比分析 Grok 4.6 与 Anthropic 各代模型在基准测试、算力成本及商业化定位上的差异。"/> ## Notable quotes > [00:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=0s) To quote Bill Belichick, you know, it's on to Cincinnati. — Gavin Baker > [05:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=300s) They didn't trigger my evil detector. — Elon Musk > [08:13](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=493s) Anthropic 最终可能会成为全球唯一的私营公司 — Anthropic 内部高层 > [08:13](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=493s) Pride cometh before the fall. — Gavin Baker > [27:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1652s) The Future is for Everyone: The Path to a Positive AI Future — Mark Zuckerberg > [27:32](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=1652s) AI 如此危险以至于唯一安全路径就是极度集中权力的概念，本身就存在根本性问题。纵观历史，寄希望于绝对权力在足够开明后善待人类，从未带来过更安全或积极的结果。 — Mark Zuckerberg > [33:30](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=2010s) AI 技术过于危险，因此不能散布 — 有效利他主义者 (EA) > [33:30](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=2010s) AI 技术过于危险，因此绝不能集中 — Mark Zuckerberg, Elon Musk, Jensen Huang > [40:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=2400s) 你不会被 AI 淘汰，但会被掌握 AI 的人淘汰。 — Jensen Huang > [40:00](https://www.youtube.com/watch?v=kVzYGVJ8zUk&t=2400s) 从可观察到的任何事物中学习 — Mark Zuckerberg ## Mentioned Gavin Baker, Jason Calacanis, David Sacks, Rahm Emanuel, Bill Belichick, Anthropic, Dario Amodei, SpaceX, Elon Musk, Polymarket, Brad Gerstner, Descartes, Sam Bankman-Fried, OpenAI, Grok, Eric Vistria, Cursor, Siemens Energy, Greg Abbott, Mark Zuckerberg, Meta, Jensen Huang, Nvidia, Nikesh Arora, Palo Alto Networks, Goldman Sachs, Blackstone, BlackRock, KKR, Apollo, CoreWeave, Morgan Stanley, TSMC, Micron, SK Hynix, Amazon, Zohran Mamdani, Howard Schultz, Starbucks, Uber, xAI, GrokBot, Qwen, Kimi, DeepSeek, GLM, Databricks, David Heinemeier Hansson, 37signals, Mercor, Silver Lake, Workday, Bending Spoons [Source](https://www.youtube.com/watch?v=kVzYGVJ8zUk)

[0:00] Gavin Baker: To quote Bill Belichick, you know, it's on to Cincinnati. [5:00] Elon Musk: They didn't trigger my evil detector. [8:13] Anthropic 内部高层: Anthropic 最终可能会成为全球唯一的私营公司 [8:13] Gavin Baker: Pride cometh before the fall. [27:32] Mark Zuckerberg: The Future is for Everyone: The Path to a Positive AI Future [27:32] Mark Zuckerberg: AI 如此危险以至于唯一安全路径就是极度集中权力的概念，本身就存在根本性问题。纵观历史，寄希望于绝对权力在足够开明后善待人类，从未带来过更安全或积极的结果。 [33:30] 有效利他主义者 (EA): AI 技术过于危险，因此不能散布 [33:30] Mark Zuckerberg, Elon Musk, Jensen Huang: AI 技术过于危险，因此绝不能集中 [40:00] Jensen Huang: 你不会被 AI 淘汰，但会被掌握 AI 的人淘汰。 [40:00] Mark Zuckerberg: 从可观察到的任何事物中学习

[4:47] Gavin Baker: I don't generally invest in Elon's competitors; I don't think it's a wise course of action. [20:00] David Sacks: The data center panic is a testament to the power of the media to create hoaxes and hysteria over literally nothing. Every single aspect of the data center story is wrong. [25:00] David Sacks: Anthropic and people in the Effective Altruism movement believe that this technology is too dangerous to distribute. And what Mark Zuckerberg, Elon, and Jensen believe is that this technology is too dangerous to centralize. [35:00] David Sacks: If we ban open source, if we let this control be centralized, we will lose AI, we will lose geopolitics, we will lose everything to China. It's just that simple. [40:00] Jason Calacanis: You’re not going to lose your job to AI; you’re going to lose it to somebody with AI. [45:00] Jason Calacanis: If you are a public figure, if you are the CEO of a company, and you do not have your own voice on X, you are at risk. It’s power. It’s armor. It’s a sigil. [65:00] David Sacks: In some ways, NVIDIA is kind of becoming the central bank of AI... the Federal Reserve of AI. [70:00] Gavin Baker: The overwhelming majority of tokens are profitable for everyone in the chain [75:00] David Sacks: If I was Amazon... I would just put a surcharge: 'Hey, this is the Zohran Mamdani fee that is being imposed on you.'

[0:00] All right, everybody, welcome back. The all in podcast the number one podcast in the world. David Sachs, and... Gavin Baker are with us this week. We got a short crew. but a long docket and we are gonna rock it. How are you doing, Gavin? Nice to see you. Fantastic. You're still in that SpaceX afterglow. This is the biggest win of your career, correct? You know, SpaceX was a magical moment. There's only one SpaceX, but yeah. To quote Bill Belichick, you know, it's on to Cincinnati. Got it. And there's a lot I think I had for SpaceX. So the SpaceX story isn't over. Yeah. And you're only as good as your last investment in our business, I guess. Everybody's like, what now? Sachs, you're back. How are you doing, brother? How's the summer wrapping up for you? Good. I'm looking forward to talking to Gavin. Oh, fantastic. Well, I'll try to stay half the way and not step on you. You missed a really great discussion we had on the All In Interview show. With your guy, Rahm Emanuel. Rahm Emanuel. How'd that go? I heard actually it got kind of spicy, you know? It got super spicy. ROM came in like full guns blazing, but it's a really great interview. And we'll take anybody who wants to do the interview show who's running in 2028. Did you catch it, Gavin, by chance? Did you catch it? I haven't watched it yet. Okay, got it in the queue. Beautiful. Of course. And basically the verdict on Rahm is that he's kind of a throwback to the Clinton-Obama DNC. I think if you were thinking of a Clinton moderate... I'm Donnie DSA, Democrat Party. This is the great irony. And it's a very similar moment, I think, to what happened in your Republican Party where you had this sort of civil war internally. And I'm hoping the mods, the moderates have a chance here. But, you know, who knows what's going to happen with these lunatic democratic socialists. We don't have too much of that. Did he have a take on that? Yeah, I mean, he... I'd say his... First of all, he's... um got incredible amount of experience having worked for obama Clinton, the mayor of Chicago, and ambassador to Japan. So he was awesome on foreign relations, etc., Yeah. And to your point, he thinks hard no on democratic socialists and the socialist policies are a road to nowhere. So he just went totally hard at it. As hard as you would, Sachs. In fact, I hope he has. Nothing would be greater than to have a moderate or more moderate candidates. I think we can all agree on that. So we'll see. We'll see. It's worth it's worth checking out. All right, let's get to the docket breaking news. At the time of this recording, for those of you who don't know, we record on Thursdays, published on Fridays and Propik targeting a $2 trillion IPO. According to the Financial Times, this would obviously break the record setting $1.75 trillion IPO by SpaceX. FT reported Wednesday night. that investors expect the company to go public in October, which is like six to eight weeks away. This lines up with the Polymarket, the Sharps at Polymarket saying 80% chance of anthropic IPO this year and Brad Gerstner, a fifth bestie, he also was signaling that this would happen sooner rather than later. Annualized run rate. We talked about this sacks a bunch. will end the year between 100 and 120 years. billion dollars this is an extraordinary ramp up of revenue that we have never seen in silicon valley At a $2 trillion valuation, obviously, you all can do the math, but it's 16 to 20 times sales, which is a fraction of what SpaceX and Palantir and some of these highly valued stocks went out at. So pretty interesting revenue growth over the last year. year 10x. Anthropic still has AI's top model, but Grok and we're going to get into it today is catching up quickly and opening. I not too shabby as well. Polymarket says 67% chance. that at the end of the year, Anthropic will have the top model breaking on Thursday. Anthropics talks to buy an AI startup called Descartes. For six, Billy. This software lowers the cost of AI training and inference by making chips more efficient. I'll stop here, Gavin. If you could comment, and I don't know if you're a shareholder anthropic or not, so it'd be great to get that out of the way. But, What's your take on just straight up the revenue ramp we're seeing here? and then what to expect from an IPO. So a few things, I'm not currently an anthropic shareholder. I'd say the anthropic SpaceX detente probably came a little too late. I don't... generally invest in Elon's competitors. I don't think it's a wise course of action. That tracks and the detente was obviously Elon decides [5:00] head when he had so much extra compute with Colossus, that he would... Rent it. to Dario and Athrapik. Absolutely. And I think he almost probably more importantly got comfortable with with Dario's morals and that Dario is a good person and Like all humans, to be human is to err, so maybe he's made some mistakes in the way he's handled government relations. But he got comfortable that his morals were in the good place. As Elon said, they don't set off my evil detector. And he has a pretty good evil detector. famously cut his conversation with the SBF short. So a couple of things on Anthropic, the rumored numbers are exceptional. We've really never seen anything like this. I still think those moments in April and May were some of the most extraordinary moments in the history of capitalism. And what's wild to me is that probably on the margin, Anthropic is losing share to OpenAI, to open source, And to Grok. and they're still growing so fast. The numbers are still exceptional. So I think, you know, that speaks to the strength. And why is that? This is like an important thing for people to understand. The pie is getting ginormous. So even if on a percentage basis, anthropic is losing on a percentage basis, the real number, and we talked about open source being dark tokens that aren't tracked anywhere, you know, this is a major, major pie growing moment. That's what you mean there, Gavin? That's the point. I think AI broadly accelerated in the month of July. which was kind of crazy to contrast with the public stock market action. Now we have a higher resolution view of what was happening. But a few things on the Anthropik IPO. Listen, it's an exceptional business. They've executed really well and they haven't just executed on the model. Their products are amazing. I always think about Eric Vistria. I don't think you've ever had on the pod, but I think he'd be a great guest. He's a good friend. He said, as a thought experiment nearly two years ago, that he thought OpenAI would still, and Anthropic, would still have a lot of value. even if they lost at the model layer. Because the product, the harness, the user familiarity, it's all... it's all so important. And I think there's an element of truth to that. But as far as the IPO, listen, the first thing I would just say is the $2 trillion. I'm a little skeptical. We saw this with SpaceX. If everybody's jockeying to be lead left, for anthropic. And if you lose that, First thing you want to do is make whoever's lead left look bad. So you link to the press and you say, hey, we're in the room. You know, the bar is a $2 trillion IPO. You said on a previous episode. 3 to 4 trillion. Absolutely, but I think that's probably where it might trade. So they're sandbagging now. Yeah, we haven't seen the S1. I do think if you're bringing a company out responsibly, you want to price it in such a way that you can absorb the lockup. Because you don't want to price it really high, create a lot of volatility. That's very disconcerting for employees. You do not want anyone distracted at this moment. So I think that they will price it smartly. Maybe that is $2 trillion. Maybe it is $2 trillion and it trades to $3 trillion. But you just got to remember these leaks are always coming from the bankers who probably lost lead left and they want to make lead left look bad. So if it comes out of $2 trillion, it probably means that the testing the waters and the roadshow just went incredibly well. And people, they do want to own this... We've never, you know, these numbers are unprecedented. So are you saying the roadshow probably occurred already? or quietly occurred? No, definitely not. I'm saying if they price it at two, it means like the testing the waters went well, the roadshow went well, and they've got a lot of confidence about where it's going to trade. Saks in the history of Silicon Valley, you pointed out multiple times here during this podcast in 2026 that we've never seen a revenue ramp like this. Can it continue? And if it does continue? What would that say about the value of AI to corporations and enterprises where Claude is the definitive leader? Well, I think it can continue, although it is a legitimate question because if you're ending the year at, let's say, $100 billion run rate up 10x plus year over year, And by the way, they've grown 10x year over year for the last three years. So it's growing exponentially. If that rate of growth were to continue, that hit a trillion dollars of ARR by the end of next year. And then the question you have to ask is, well, is the TAM big enough for that? But also, is there enough compute? Is there enough energy? I think you start to get into physical constraints. I do think the TAM is big enough. I think the demand is going to be there. There's so many. [10:00] new applications of AI. Agents are just taking off now. You saw that with the launch of GrokBot. All these other companies are going to keep extending the uses of AI into more and more contexts. So the demand for tokens is going to keep growing exponentially. I think the question is whether they can physically meet that demand. And then also Gavin's point about market share, Other companies are seeing what they're doing and they're competing more effectively to take share. Remember, Anthropic made this huge bet on coding. And that ended up being a very prescient bet. I don't know if it was because... of ideology they thought that was the way to get to recursive self-improvement first i don't know if it was because They saw... cursor. That's the reason building on top of them. Yeah, they watch cursors utilization and they said, what we're going to take that business. And that's such an important note. for you to make sense. Yeah, so they've started moving into verticals where they see their own users. building successfully on top of their platform. So that was, you know, I think that definitely was a signal to them to get into coding. Nonetheless, They got into it first. They got into it very successfully. That led to a boom. We hear that open AI now They've basically pivoted into coding. They're doing a better job with that now with GPT 5.6, we've heard their growth rate is accelerating. They used to be growing at three to four X a year. I've heard that Over the past two months, that growth rate is now over 20 percent month over month which if you extrapolate it out over 12 months would be a 10x growth rate so you're now seeing OpenAI, growing as fast as Anthropic, having made that pivot. We can talk about whether we think SpaceX AI may be able to get there as well. But in any event, look, I think that... regardless, of whether Anthropic is able to grow 10x next year. I think they're going to grow very fast. And then the question is just, what does the market share look like relative to their competitors? And I guess, Gavin, it would be worth double clicking on two points there. The first point, is there going to be a market for these tokens and can the growth keep up? In other words, is there a demand? I think all three of us agree. Yeah, sure. Of course, there's demand. But the next two pieces, which Sachs mentioned, were energy. data centers. And that is a question I want you to answer. Can they get to 1 trillion in revenue? Let's slow down here and actually try to you know, do some back of the envelope math, and this is your wheelhouse, what would it take in terms of infrastructure for them to go from 100 billion dollars a year in revenue and 10 exit. And is that humanly possible in the next year? So a few thoughts. I think it's... the right questions. First, internally, Anthropic is very confident. I have been told by multiple people I trust that Dario has said that Anthropic might be the only private company in the world at some point. Think about that. In this vision, an anthropic maximalist vision, There's anthropic. And then there are governments. And that's it. So there's a lot of confidence. I'm sure they have more advanced checkpoints than Fable up their sleeve. They've executed really, really well. I would probably take the under on them being the only private company in the world. Um, I think it's going to be a long time before they land a rocket or their... you know, and deliver your lunch. There might be other businesses in the world. Zip line might want to bring you a burrito Uber might want to take you somewhere away. Mom, I want to drive you home. But sure, Dario, I might interpret that as a negative signal because it's so hubristic, you know? Yeah. This is getting into SBF land a little bit. Yes. I would certainly discourage Dario from saying that ever again to anyone. So you don't, so AI Jesus should not claim that he is AI Jesus. Okay. Pride cometh before the fall. Yes. And that will be a long fall if they, I mean. But no, but look, I have heard from other people at the company. I heard that when they were at about 60 billion of ARR, that they thought they could get to 600 billion in one year. So meaning they thought they were on that 10X, ramp from the middle of this year. Now they're at over 80. Doesn't seem to be slowing down. So I think, look, I think this is the big question is, How fast can they grow next year? Does it go all the way from $100 or $120 to $1 trillion? Look, if they merely got to $400 to $500 billion, that would make them the biggest software company. If they get to 200 billion, that's an amazing outcome. So where does that put them versus Microsoft's Office franchise? Like where or Google's search franchise? Like we're starting to become a peer to those two companies. A hundred. Absolutely. And so just to break the question down, this is a good question, and to demand and supply. [15:00] So demand, depending on how you count it, there's 25 to... You know, 65 trillion in knowledge work. 25 is kind of a number I've used. I was told by, you know, the head of the AI Institute at one of the three largest investment banks that it was way low. So this is the TAM that they're going after today before we have robotics. And then it's a question of, is it labor substitution or accelerating growth? Thus far, it really does look like it's accelerating growth. There are more job openings for software coders today than there were a year ago. There's more demand for coders and that is kind of in the epicenter of the blast zone. There is some evidence that for very young people coming right out of college, that maybe there's been a negative impact. But broadly speaking, it has to come from one of those two. And I just think it's, we, It's better for everyone if it comes from accelerating growth. And these companies are now at a scale where if it's going to come from accelerating growth, we will start to see that in the national accounts over the next 18 months. I am going to... Okay. You want to go to the energy part? Supply. Yeah. And listen, this is... This is where the rubber beats the proverbial road. The people who build these data centers, the best way to understand it is go watch that series Landman. And it's like you need like Billy Bob Thorntons out there. Permian Basin time. The Permian Basin. You're in the patch. You know? you're actually worried about narcos. You know, it is... It's the tip of the spear. It's like... 110 degree heat. You have to orchestrate thousands of people in these remote locations. It's atoms, not bits. It's atoms, not bits. And it's really, really hard. Now, America, we're actually really good once we gear up for something like that, America and capitalism. We're very good at solving those problems. Everybody talks about the ultimate constraint, or these chirp-eyed blades. They're only made in two facilities in America and Europe, and each facility is a giant building, and there's some 40-ton press. But you know what? Now they're running 24-hour shifts. Necessity is, in fact, the mother of innovation, and nothing like capitalism will make that happen. We'll solve that. Boom! supersonic making a new jet for you know supersonic deposit those jets he's like you know what we're going to build a turbine business first because we have so much demand and then elon I understand there was a news report that he bought a turbine company and he did it personally because that speaks to me that he just wanted to go fast. Right. And on top of that, you know, residuals for private jets are like extremely strong these days. And a big part of the reason is people are taking jet engines off old planes and repurposing them as turbines to power data centers. A hundred percent. That's insane. It's crazy. This is like a big business and growing really, really fast. you have Siemens Energy, they're all expanding capacity fast. I think the supply question, the Billy Bob Thorntons here in America, we're going to work with those companies and get these turbines on the ground. It's politics and it's regulation. And the fact that Texas has kind of announced you know, what I think is probably an energy audit is what you're referring to. Governor Abbott announced this week he's going to do an energy audit. Not that he's going to slow things down, not that there's a moratorium, but that everybody has to go through an energy audit if you BYOE. you're going to be fine. But if you don't have a solution for that, and you want to tap the Texas, the great state of Texas is grid, You're going to need to maybe pass some audit because we don't want the whole... grid going down here, which it does every January when things freeze over. Yeah, and it's like right now, the narrative on AI is not good. It's that a data center takes all the water because someone made a mistake in a book three years ago, and they're off by, I think, a factor of 100,000 in the amount of water a data center uses. So it's just wrong. raise your cost of electricity, which is just wrong. They lower your cost of electricity. The Wall Street Journal just read a story about a town called Ellendale, where it was a dying town, a data center got built, it's revitalized the town, tax revenue was 10xed, it's been the best thing that's ever happened. [20:00] But nobody is telling those stories and people need to. And I think, you know, we've discussed on the show many times, the Chinese Communist Party, they do have a very active, anti-AI, anti-data center campaign, because that's what's good for them and bad for America. And someone needs to tell the truth about AI. One of the best phrases in the Bible is, "The truth shall set you free, Yes. Can't set you free if nobody tells the truth. Yes. Yeah. Yeah. I mean, the data center panic is a testament to the power of the media to create hoaxes and hysteria over literally nothing. Every single aspect of the data center story is wrong. It's a lie, like you said, they don't use up a community's water, the water use is quite manageable, less than a golf course. The water recirculates. It brings down electricity costs, assuming that you have the data center stand up its own power generation, because they will sell the excess back to the grid and they pay for grid upgrade costs. The real reason why electricity prices are rising in blue states is because of decarbonization. and all the other green type regulations they have that make power generation more difficult. What are the other accusations? I mean, they create pollution. Yeah. Noise pollution is legit. So there's nobody in their right mind. They don't have to put it in anyone's backyard. Yeah. Nobody in their right mind is doing that today. We have a lot of land in this country. Yeah. People did do that for a bit. If we just use common sense. Yes. Common sense. You just use common sense and there's no problem. The idea that you have to have a complete ban. on data center construction is just silly. And the communities that are embracing them are seeing – They have better schools. They have better services because they're making tax revenue. Well, and then to your point sacks when you know entrepreneurs get on something we're all of a sudden seeing a solar, batteries, nuclear, and more clean energy being tapped to power these. And that demand will lower the prices of those services over time. Solar panels will get cheaper, batteries will get cheaper because they're going to be scaled much faster, which is good for humanity. This is basic, folks. It's very basic. The green stuff can't scale in the short term. It's all gas powered for the new data centers. That's the only way to kind of get the amount of scale that you need in the next year or two. But but in any event, look, you have that that this I think so is more like, you know, it'll be viable. But, you know, it's obviously not as fast as popping up neck ass. Yeah, I largely agree. I will say, you know, Elon's building a $10 billion new solar manufacturing plant in Texas. For a reason. My home state. And, you know, I think solar, the world will eventually run on sunlight. But we're going to run on natural gas for a while. We have a lot of natural gas in America. It's a clean fuel. It's the greenest carbon fuel. And just this media hysteria about data centers is doing such a disservice to low-income Americans, to small-town Americans, often to minority Americans. It's so good in every way, and the media is doing such a disservice, but no one's taking the other side, and someone needs to take the other side and tell the truth. Yeah, well, it's... I don't know who's relying on the media for their facts now, but I think there's a reason why we were number four in the world last week is because, hey, listen, we actually are. part of this build out we know about it and we can explain it to you a little bit more succinctly noise pollution that's an issue Pollution of the air, that is an issue from NACS. Keep these things far away from people, as Saks just said, and you don't have as much of an issue. There is a minor issue. If you put these at scale in a certain location, you can raise the temperature. of that valley or wherever you put it a couple of degrees that's something worth monitoring these are all absolutely manageable issues. And if you really want to get obsessed about water, please, for the love of God, stop eating almonds. Like, and maybe like, you know, the golf course doesn't have to use a billion gallons of water every year, I will say, on the demand side, and on the pricing side, if you pull up this quick note i have here i think anthropics growth rate is going to slow considerably next year and it's been obviously on a tear there's no way they get to a trillion in revenue next year uh i think maybe they could you know triple get to three or four hundred million And the reason is, watching startups and then also watching some of the large companies now embrace GLM 5.2 and Zuckerberg, which we'll get to in a moment, releasing really great open source models. You just look at the pricing here of GLM 2 versus Claude Opus. And it is 90% cheaper to use these products. I'm using them. The founders I invest in are using them and increasingly, corporate America is [25:00] going to embrace these solutions. It's just a little bit harder to set them up. And it's coming folks. And if you think it's not, that corporate America is not going to embrace open source, you have not been paying attention to the last 20 years, when they did exactly that. And 20 years ago, there was a big argument, oh my God, corporate America will never trust open source in corporate America said, we don't want to get rug pulled or have some large tech corporation hold us hostage for their technology, which they're experts at doing. So open source is the safer bet for us. And that's what's happening in corporate America today and inside of startups. This is going to be very real. Gavin, you look like you're jumping out of your seat. Yeah, no, no, this is a larger point. For sure it's good for corporate America. For sure open source is good for the world. For sure it's good for America. But, you know, Dworkesh posted something. Dworkesh, who's a podcaster, a thoughtful guy close to the AI scene, he posted something that really resonated with me. He said at the end of the day, Anthropics AI and their constitution, it is wired to do what Anthropics thinks is best for humanity. And let take Elon judgment as gold and he right His don be evil detector did not go off And they genuinely good people which I actually believe them to be who are doing what they think is good for humanity. I actually believe that. The problem is, there are always unintended consequences, and the world is chaotic and unpredictable. And I just think what's great about open source is it means that we're going to have a rich variety of AIs. We're going to have a diversity of AIs. And that is a good world for humans in America. We do not want one, two, or three dominant models. And that's why, by the way, don't want to always talk about Grok, but it's great. I think history will judge Elon kindly for having Grok dedicated towards the pursuit of objective truth. Objective truth. Talk that book. The whole point, Gavin, is if you made the bet, that means you used your intelligence and your savvy to pick what you think the winner is. That actually is, I think, super accurate. And I think Elon explicitly believes in open source, and he believes that this is a technology that should be shared for the good of it And in fact that was his first foray into this which was backing OpenAI And OpenAI manifesto was rewritten this week Topic 2, by Mark Zuckerberg, quite paradoxically or ironically, in a 6500-word essay. It's titled The Future is for Everyone the Path to a Positive AI Future. He laid out an extremely optimistic AI worldview where he would, as now OpenAI 2.0, Facebook, he would release open source models. He would create an agent for everybody. He would make it free. He would provide everybody in the world with their own tutor and unlimited abundance. He has picked up the abundance crown, which Sam Waltman dropped when he went for the private company. And he says, super intelligence is invention not automation and that ai safety should be based on the balance of power no singular centralized intelligence i'm assuming sacks you read or scanned or got the coverage of zuckerberg's manifesto and you have any thoughts on his position yeah i mean i read it and i found myself nodding in agreement with much of it i really like the point he made pretty early on where he pointed out that a lot of the doomers and I think he was kind of taking a shot at the effect of altruists and anthropics Dario. Dario, he said, look, why are you guys rushing to create a future that you don't believe in, that is so dystopian? And if you think this is such a bad future where everyone's out of work and AI goes rogue and creates all these safety issues, why are you so excited about creating it? So I think he does point out a fundamental contradiction. Now, there is an answer to that question, which is – actually, there's two answers. One is they could be trying to create a future they don't believe in for let's call it mercenary reasons. It could be lucrative, although I don't think that's the real reason. I think most of those guys are missionaries, and what they believe is that that future is dangerous. and only under their enlightened control can humanity be protected. It's right out of the vision of the anointed, which was a book Thomas Sowell wrote 30 years ago, where intellectuals throughout history have thought that if they're maximally empowered, then they can engineer society in a more benevolent direction. And this has always backfired. It's always led to not just empty or broken promises, but also it becomes an excuse [30:00] is for totalitarian people. schemes and state power. And actually Zuckerberg goes on to say something that I think gets at this. So right after he kind of made the point about, you know, why are you rushing to build a future? that You don't want. You don't want. He says... The notion that AI is so dangerous that the only safe path is an extreme danger. concentration of power, seems inherently problematic. Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safer positive outcomes. So I think the point he's getting at here is that a major frame on the whole AI debate, it's not just about open or close is also about centralized versus decentralized. And whether you think a centralized outcome is the right outcome or a decentralized outcome, I'm on the side of decentralized. I mean, how could you not be on the side of decentralized? I mean, it's like the most obvious argument ever. The EA types are not. And by the way, I experience... Yeah, they're delusional. Well, I experienced this when I got to Washington. And, you know, all the former Biden people had just moved over to Anthropoc government affairs. I also dealt with think tank people who are part of this like click. And their frame on AI was that it was inherently dangerous and that we had to centralize control over it. In the government, there should be... you know, just two or three companies, And they should essentially work together. It should be made into a cartel. Remember, you know, Mark Andreessen had a meeting. with those Biden officials where he came out-- - Yeah, they said, "Don't bother. Don't bother competing." - Yeah, now, they dispute that characterization. Maybe they didn't use those words exactly, but I do believe that Mark was correct in his interpretation. of what he heard because i heard similar things and they really believe that the way to control what they essentially thought as of being maligned or at least very dangerous was first you limit it all to the united states so you don't export the chips You don't export the models. It's basically you retain total control over it in the U.S., which frankly is a pipe dream because the rest of the world is going to get AI. Of course. But that was sort of part number one. And then part number two was you form like an atomic energy commission or something like that. to cartelize the industry. And, and then you have a deeply, embedded with the government. You effectively have a merger of of corporate and state power. So it was this extremely centralized vision of the future. I think they still believe in that to some degree. Clearly Zuckerberg listens to the All In podcast and clearly I think we have a seat open next week. Mark Zuckerberg, somebody reach out to him. Come on the pod next week. Thursdays we'll tape at about 10, 11 a.m. Your time, whether you're in Hawaii or... in Tahoe, we will get a Starlink to you and you can do it from your EFOIL, guaranteed. But it would be great to have Mom back to talk about this. Let me get Gavin involved. Just to wrap up, though, I think where Zuckerberg goes with this that I agree is we want maximum empowerment of the individual. That's the way to solve this problem. We want decentralized outcomes. We want open models like you're saying jcal we want data sovereignty look there can also be these proprietary frontier models from Anthropic and OpenAI. But, We want to create a decentralized, empowering outcome for individuals. And I think that's the side that Zuck is on, and I very much agree with that. Gavin, I assume you read it, yeah? Or did you just get the high? You didn't have time for 6,500 words? No, of course I read it. By the way, if he writes nine more... He's got the average nonfiction book, which is right around 60,000, 70,000 words. So Harper Business. Hollis, go call Zuckerberg and get the other eight parts of this manifesto going. I read it. I agreed with it. And I would just... Super agree with everything David said, and I would boil it down to, I think, Anthropic and people in the effect of altruism movement. believe that this technology is too dangerous to distribute. And what Mark Zuckerberg and I think Elon and Jensen believe is... that this technology is too dangerous to centralize. And history is spoken. And when given the choice, it is always better to distribute and decentralize. There is really no counter example that I can think of. And just... Forget as an American, but as a human, I want an AI. that looks out for me Looks out for me. The individual. Be the individual, my conception of what's good for America. I do not want Dario Amadei or Anthropic deciding what's in my best interests. Hmm. I just fundamentally don't want that. You don't trust AI Jesus. I mean, AIG seems like a great guy, although it is funny, somebody said, he was cursed with resting smiling face and it is a little weird to listen to him talk [35:00] of this terrible vision for humanity, while he's smiling happily. He's literally got that grin on while he explains to you the Terminator scenario and that only he can save us. And that even he doesn't know, that like there's a 20% chance that he might not be able to save us all. Gavin, I love that framing about it being too dangerous to centralize. And I think one other closely analogous point is what you think about competition and So I think that the EAs think that competition will create a race to the bottom. and we'll create a lack of safety whereas what i believe is that competition brings out the best And so they think it brings out the worst. Now, look, there are examples of probably both, but fundamentally, they want to centralize control and have less competition. We want to decentralize and have more competition. And I think I want in the Constitution, I have the right to bear arms. And I think that's important. I grew up in Texas. I think it is kind of like when you come home, I have defense against tyranny. I love it. When you're coming home, I'll be back soon. I promise. Okay. But I want the right to have my own AI. I want the right to my own AI. where its values are aligned with me. Not... Not someone as good a person as they might be. The right to bear arms is a good analogy because we do allow you to have a gun even though you might misuse it. Guns are used to kill things, right? Whereas AI has a much more creative use and much more salutary uses. It's beneficial. Fundamentally, it's a consumer technology that everyone's going to want to use. So if we allow you to have guns, I agree. You should be allowed to have AI. Let me correct you there, Sax, now that you're in the great state of Texas with me. Guns also help you feed your family and defend you against evil forces in the world. So those are virtuous uses of a gun. Yeah, I get it. They are. That's self-defense, but I'm saying you'd still use the gun to kill the… Use the gun to kill or hurt the person. in self-defense. I get it. I'm not against guns. I think this is the delusion of Dario. If we think about it with these you know, effective altruists, you can look it up folks, but they think that basically, they're going to make all the money in the world, create the one company, the money printing machine, and then they're going to save all of us. It's massive delusions of grandeur. That, you know, is the technology is going to diffuse anyway, this idea that you can keep the genie in the bottle is absurd. But by the way, I think Saks, the way they view it is, they view this literally as super intelligence and the nuclear bomb. And if you convince yourself that you're building the nuclear bomb. Andrew building the Terminator that you yourself cannot control. And it's super intelligence that will make a billion Terminators and a billion bombs. They literally have diluted themselves while taking P.O. Day or LSD at Burning Man that this is the reality. It's not the reality you could unplug it at any time. They're suffering. from delusions of grandeur period full stop this is not the nuclear bomb well you're you're right about that when i first came into washington that was the dominant analogy that they and their their sort of related think tanks their people in their group had promoted this idea that AI was like a nuclear weapon and we needed an Atomic Energy Commission, a global one, to regulate this. Total pipe dream, never going to happen, but you're right that they were using that analogy, and I think The reason why they're using that analogy is because you do not have a right to basically have a nuclear weapon, to build your own nuclear weapon. You have the right to a gun. but not a nuclear weapon. So they wanted to put AI... in that bucket. But As Jensen has said, no individual, no consumer needs a nuclear weapon. Every consumer, every business needs AI. It's fundamentally different. This is a consumer technology. It's a consumer technology before it's a military technology. And this idea that it's primarily like a nuclear bomb is just kind of absurd. I mean, in fairness, in fairness, Gavin, Cyberdyne Industries, they also thought they were building good technology. So... Science fiction has literally polluted people's brains. If we're going to have a Terminator thing, I want my own Terminators who like me. Yeah, open source the Terminator. Yeah, but just very simply, just bullying everything else. I'm an American, I'm a patriot. If we ban open source, if we let this control be centralized, we will lose AI, we will lose geopolitics, we will lose everything to China. It's just that simple. So if you're an American patriot, you should be for the distribution, democratization, decentralization, individualization, of this technology. Hey, J. Cal, man, what are your thoughts? What are your thoughts on the essay? Well, like you, Sachs, I found myself nodding the whole way because I think he took, he kind of like ran in front of the parade and grabbed the baton and was like, I agree with the best takes here. And the best take is open source, empowering the individual. And this isn't nuclear technology. [40:00] I just want to point out two other points that I think got buried because he did try to cover a lot of ground. The first is I want to give him credit for his view of specifically job displacement because he says, hey, listen, you know, that is possible job displacement. But he says people fear that automation will outpace individuals capability growth leading to job displacement, followed by a difficult period as people learn new jobs. I think we can all agree that that is a concern. But there is no rule that AI must increase automation faster than it increases individuals' capabilities or demand for new skills. Recent statistics suggest it may be more likely that individuals' capability growth could match or outpace automation. I thought this was like a really good framing and is just another way of saying what Jensen said. Hey, you're not going to lose your job to AI. You're going to lose it to somebody with AI. And that's the game on the field right now. every time I use this technology, three new business opportunities open up and I create two more job racks and people are doing five or 10 times as much work if they embrace the technology. And that's a big if there and if people will do it The second thing I thought was pretty interesting was he makes a play on having the right to index anybody content and learn from it the quote he says here is all ai models are derived from human knowledge okay fair enough i'm just pretty pretty plain vanilla there some have tried to frame distillation as harmful but i think it's important to protect the principle that you can learn from anything you can observe in other words he's preemptively saying i want to scan the world's data well i have a message for him which is zuckerberg please explain how this concept is compatible with the fact that you have sued every startup many of which i've invested in that has ever tried to index or use the social graph at instagram or facebook they have hundreds of lawsuits where they have sued people for trying to distill you know who are the top influences on instagram who are who's in my social graph and And they lock that whole thing down. So there's a bit of hypocrisy here that reminded me, Sachs, of our discussion on previous episodes with Claude ripping apart old books and distilling them. We've had the distillation debate many times. Let me just go back to this point about there a huge irony in the fact that Dario did not get his way in Washington for the most part during this administration Because I think if he had gotten his way if he had realized his dream of this highly regulated AI apparatus I don't think that company would be set up for a hugely successful IPO. Why? Because their whole right to charge this massive premium for tokens, which, Jcal, that's the point you were making before, is you're wondering how sustainable that is because of all these open models that are coming up and are offering tokens for much less. Their whole pricing power, That premium they're able to charge is because they're six months ahead of those open models. And that six months will be eaten up in the blink of an eye if they were subject to regulatory approval. Dario himself keeps writing these blog posts where he says, we did an FAA for AI. You know how long it takes the FAA to certify and approve a new model, in this case for airframes? Minimum of five years. I mean, if you want to change a table in your plane, it takes eight years. You know how long it takes the FDA to approve a new medicine? Years, okay? You're gonna apply that regulatory framework to AI model releases? Game over. Forget about it. Your six months will be gone like that. Your model will be commoditized, to Gavin's point. The Chinese competitors will catch up And that bad for the United States It bad for our competitiveness It bad for the AI race I mean obviously we don want China beating us but aside from just the geopolitical dimension it'd be horrible for Anthropik as a company because again, all their pricing power depends on them remaining in the lead by six months. If they ever slip up, hold on. If they ever slip up and lose that six months. What does that do to revenue? Exactly. so now look i actually think that their pricing power may be quite sustainable because i think people are willing to pay not the whole market not the whole market but a decent subset of the market is willing to pay a premium for true frontier intelligence so as long as hey if the top 20 percent sacks the top 20 is willing to pay 10x they're golden right like just like i said before it's apple versus android right most of the market is willing to go with android is cheaper but there's a significant subset of the market that's willing to pay a huge premium for Apple because it's a better experience. Same thing. If you're a company that's in a competitive market and you want the absolute best for competitive or other reasons, you're going to pay a premium for true frontier intelligence. But Anthropic and OpenAI are on that hamster wheel. They have to stay in the lead in order to justify that premium. [45:00] And by the way, I think they can. And they think they can. And they think, in fact, their elite is growing because they're on the way to recursive self-improvement. I just want to make another point about the going direct movement. Zuck has had his Twitter account since the beginning when he tried to buy the company at FinkD, F-I-N-K-D, which was his handle in college, I believe. And in the last... 30 or 60 days, Nikesh from Palo Alto Network's friend of the pod, he started tweeting and doing long form. Jensen opened an X account. He started going long form. Thanks for the follow, Jensen. And then, Zuckerberg, has activated his account and he's going direct and i want to just point out that Zuckerberg previously, I think, was letting other people craft his message and send it to the market. And I think he has matured and decided I am going to write my own missive, this missive. I believe was written by him, probably with some help, but I ran it through an AI detector. It's 100% human crafted, bespoke human words. I think it's fantastic. that he's now getting ahead and saying these are the potential problems in the future And it would have been so great if he had done that with teenage girls and the impact of social media on kids and addiction with Instagram and Facebook earlier on. I give him a lot of credit. for deciding i will go direct i will write my essay i will speak my truth And man, X is getting... exceptional at these dialogues and these missives now, it's fantastic. Yeah, so I probably agree with everything you said, and I want to come back to a couple of things. that Zach said, about how open source actually makes Anthropic more valuable, and computational efficiency of humans is the case for humans. But I think what you said is really important. If you are a public figure, If you were the CEO of a company, and you do not have your own voice on X, you are at risk. It's just that simple. Risk of not unpacking. You, you, you, Ultimately, in the same department as the truth shall set you free. The only person who can tell someone's own individual truth is themselves in their own authentic words. And obviously, Mark Zuckerberg has super voting stock. He's never going to be thrown out of meta. But there's many CEOs who think, "Oh, my corporate PR department's keeping me safe. They're my voice." Well, guess what? Maybe you retire. Maybe the next CEO doesn't like you. And you get called in front of Congress. And I do think at some point, the Democrats have talked about how they're going to have an accountability agenda. And they're going to go after everyone. And I mean, I'm not sure that's like a good message for the Democrats. But be that as it may, if you do not have your own voice and your own brand on X, you're at risk. It's power. It's armor. It's a sigil. Yeah, it is a superpower. And I don't know about you, Sax, but I am loving Zuckerberg's Jeremy Strong era much better than the Jesse Eisenberg. Are you doing this MMA stuff? There is a video of him on like a barge or something. Yes, on a floating barge. A floating barge. What is that? A barge in Lake Tahoe? Yeah, dude is-- I would be careful. I'd be careful about saying bad things about MJ Cal. No, no, no. I am-- listen, I have to have turned over a new leaf. I am-- I am-- I'm into Zuck 2.0. I hope he likes JCal 3.0. Come on the pod, Zach. I want him on the e-foil with the American flag every 4th of July. And I want it live-streamed And if he falls, that would be aura maxing for him. If he gets back up and records again, it doesn't live. Yo, he's got to do that in Kauai with the bull sharks. Let me see some fins chasing him. He is aura farming times a thousand. Way to go, Zuck. And like we need more of this like authenticity. Base, CEO, go direct. 100% man, go direct, protect yourself. Yeah, and be yourself, man. And like he's being himself. Let your freak flag fly. It's all good, folks. two points that I think are really important. Open source makes frontier tokens more valuable. So let's just say, Let's use an example. Let's say that, you know, Anthropic, OpenAI and Grok, they're at like a 250 IQ or 200 IQ. Well, if you have a 200 or 250 IQ intellect that can orchestrate these open source 150 IQ intellects, Like, That makes the 250 IQ intellect more valuable. is because it can farm out past. And just in the same way, if we're going to use the Manhattan Project analogy, everybody talks about Richard Feynman and who's the guy who ran it all. And there's these 10 really smart physicists. Who's the guy who ran it? [50:00] who said, I have become death. Who's the guy who ran the Manhattan Project? Come on. Oppenheimer. You had Oppenheimer, and you had 10, effectively, Nobel laureates, maybe 30 of them. but they needed like the 2,000, you know, merely, you know, top 10,000 in the world businesses. So having more... having more affordable intelligence, it actually makes the frontier even more valuable. And I think a very plausible outcome is one in which frontier tokens are 65 to 85% of the economic value. Open source tokens are like 80% of the volume. And that's good for everyone and everyone wins. Tell me the revenue, Gavin Baker, of Anthropic by the end of 2027. What is their 2027 revenue if 2026 ends at exactly 100 billion? What is their 2027 revenue? I will take the over. Over what? A trillion? A lot of estimates. I'm not going to take the over on a trillion. Okay, I'm going to put $500 billion over or under Gavin Baker. How about this? I'm going to take the over. $500 billion. No, I'm going to take... the over, whatever the Wall Street consensus is for 27, which I don't know, I'll take a 25 to 30 percent over on that, assuming that this regulation campaign doesn't succeed. Going out on a limb here saying that the underwriters are going to sandbag revenue. OK, congratulations. Give me my answer. Over 500 or under? I think four to five billion is within the zone of realistic and achievable. Can I just say one other thing? Hold on, Saks. Over under 500 billion, Saks. I actually was going to say four to 500 billion is exit ARR for next year. Seems very comfortable. So right under. Okay, great. Two unders. Yeah, and the only reason it's not higher is because I do think that somehow you get into physical limitations at that point. But maybe not. Maybe there'll be enough compute. I mean, let's have St. Dear, though, that we all believe... that they're going to hit 400 billion next year. And they're going to 4X... 4x. $100 billion. Like, Just let it sink in. They'll be at 120. So I guess that, you know, implies, you know, 4X would be more like 500. Yeah, 4X. Yeah. That's kind of what we're coming up. I think the over-under really is 500 billion exit ARR. for next year. And I think they think they'll do the over. We were talking about them at 10 billion. 10 months ago. 10 billion to 400 billion is 40x, gentlemen. Are they going to thank me for preventing Dario from shooting himself in the foot? I think they should. He literally was like, he put two guns... on two ankles and he just emptied both clips and you had taken the bullets out you think they should give you one percent i actually don't think the bullets were taken out i think he had them pointed in his temples oh you're saying diddy now he did your hunter and sax pulled him down and like he just shot himself in the foot this is where their hostility to open models comes from i think is that if they're going to slow down they have to slow down the open models as well otherwise you they do get commoditized and their business goes away. However, The United States does not have jurisdiction over what happens in China, and China was not going to slow down. That was always... I think a huge flaw in this whole conception, these EA types thought they're going to institute global compute governance. Somehow they're going to get some global agreement with China. Total pipe dream, never going to happen. If we slow down, China is just going to race ahead. It was always a naive vision. I thought he was going to create the Paris Accords 2.0, and that didn't work. So slow your roll, kid. Can I just say one thing that I think is also important about open source and just like why I think AI is going to be good for humans? It's going to be good for Americans. It's that like if we are short compute, we are talking about compute slowing down anthropics growth because literally we're just, you know, we're running out of energy where we can't bring on. We're running into physics and like... planetary scale limitations, and I think this is a reality. Just think about how energy efficient we are. All of our brains, to approximate our brains, you need to have a data center that consumes the power of a million American homes. running for probably six to nine months, and then you get our brains, and then every time You want to answer a question. You need the power of like a thousand American homes, okay, to operate at the level of our minds, You know, we're smart, but like, and all those figures are going to get better. But we're so energy efficient. there's going to be a use for the most computation [55:00] efficient intelligences. on planet Earth, and those are human intelligences. It's so well said because if you look at what Cerebrus and Grok and NVIDIA are up to, they're making inference ten times, five times every year, and you're the chip master here, Gavin. But they'll be five to ten times more efficient every 12 to 18 months. Then you look at the models. And by the way, congratulations. You got Zuckerberg on your ass. Go ask MySpace. Go ask Snapchat what it's like to have Zuckerberg in the review mirror. That's literally the T-Rex coming at the Frontier models. He is a beast at copying, mimicking, and competing. He's going to be the open source hero from America. He is going to have the best open source model in the next year. I guarantee it here on the audience. I might take the under. There's another American hero, Jensen Wang. Okay, Nemotron. You might see Nvidia at the open source frontier. He doesn't like us talking about that. I got the back channel. He's like, We're not in the open source business. I mean, we do have Nemo Tron and we can give you a full stack. But one of the hyperscalers, by the way, that we know, Gavin, one of the hyperscalers told me they just did a nine figure. double Lindy reverse, where they went from doing a frontier model on 100 billion dollars in infrastructure and they're like Bing, it's going to be open source. not Frontier. So people are looking at this saying it's time to think maybe move over to some open source models. All right, listen, All in sum is coming September 13th to 15th in Los Angeles. We told you about the Giants last week, Jensen, Satya, Jared. Gwen, I mean, this is all first name club. But here we go. Steve Hilton. running for governor of california is going to come spencer pratt i'm going to talk about his failed la run but successful in terms of capturing people's imagination. Abigail Schreier. author of Bad Therapy, a fantastic book, is going to be with us with a very special surprise Guess. And hey. Bestie Gavin. our guy bestie gavin might even make an appearance and my guy blake show CEO of Boom Supersonic, who I mentioned. earlier in the program. And, this year. We're going. to turn the outside of the shrine into a festival-like experience from you for you a frontier tech hub featuring ai and robotics demos and a gifting suite The gifting suite, like you're a celebrity. merch.com has made an incredible all in swag suite for you wellness suite with red light therapy facials and IVs if you uh hey you stay out a little too late saxipoo that sometimes happens don't miss these important conversations and by the way we're going to announce all the djs and parties which is the reason freeberg does this gonna blow your mind the all in summit.com to apply today go ahead and apply Hey, speaking of our guy, Jensen, sixth bestie here on the pod. Nvidia. His partnering with Goldman, BlackRock, and others to raise 500 billion in AI compute. Jensen laid out his vision in an X article. As we mentioned, Jensen is now on x titled nvidia ai factory compute is becoming an investable asset class the firms will establish independent financing platforms raising third-party capital the idea is to treat gpus like a financeable income producing asset like hey maybe mortgage-backed securities but hopefully not with the same outcome here's how it works at a basic level instead of a company paying billions to buy chips up front they borrow money They buy Nvidia-based systems and use them to start generating revenue, i.e. renting compute. And then they can use the cash flows to pay back the loan. Nvidia's role is matchmaker connecting customers who need compute with lenders. Obviously, GM, Ford have their own finance divisions. Gavin, what do you think of this announcement vis-a-vis round tripping and the hand-wringing concerns we talked about on this very podcast a year or two ago. Well, I do think what's important here is what's essentially happened is like, Blackstone, KKR, these firms, Goldman Sachs, they would not have done this. They would have not have done that CNBC episode with Jensen if they did not believe that NVIDIA GPU Compute was a financeable asset. And so what's really smart here is you're getting some of the smartest asset managers in the world who charge $2.20 for many products to essentially value. [60:00] validate this market. And NVIDIA is being a matchmaker. And what they're basically saying is, hey, our compute, because it's so flexible, is going to have a long enough life that you can finance it at lower rates than other kinds of compute. And I think that's smart. That's good for everyone. And it is interesting. If you look at the underlying architectures of the three, what I call big Chinese open source models, maybe even through a few or four, if we have Quinn, if we have Kimmy, if we have DeepSync. and then we have GLM, they're actually evolving in very different ways. And that architectural variation works in NVIDIA's favor and in GPU's favor, because it means that you do need this more flexible compute to be able to finance it, to be able to think that you can have a long life. But what NVIDIA is doing that I think is really smart is they're saying, hey, we are going to help create this market, grow this market. But they're going to provide two very important functions. They're going to say, Hey, Blackstone, Goldman Sachs. KKR, Apollo, BlackRock, I'm leaving someone out and I'm so sorry. There are six of them. Come to us. If you have a deal, you can bring it to us. and we will give a residual value guarantee. And I think the way that that residual, which will further lower the cost of financing these, and that residual value guarantee can then be incorporated into those companies underwriting. And what that basically means is they're saying that after three years, four years, we guarantee that these GPUs - can be rented at a certain rate. And then the risk that they are bearing - is the risk between that value and wherever the market rate is. And right now, everything's going straight up. NVIDIA has better telemetry than almost anyone. and to the supply and demand of compute. So they can put that at a very, very smart place, further make it financeable, you're risking... you know, they're only bearing 25% of the risk, some of the world's smartest underwriters. you know, in terms of reputation. What could go wrong, Gavin? I mentioned mortgage-backed securities, etc. And obviously, The lifespan is the key issue here. The different hyper scalers are saying, hey, four or five, six years, Amazon and the big debate happened. But what we've seen is, to your point with the open source models, is they can be tweaked to run on the oldest hardware. So maybe we get year six, seven, eight out of these. That's the key to this? No, no, we're going to be more than that. CoreWeave said... that they are renting Amperes at economically profitable rates today for 2029. So an Ampere that came out in 2020 is going to have a nine-year life. And I think that's what NVIDIA sees. It's like old American cars, they get sent overseas just... The idea that these GPUs... Where could it go wrong? Here, I'll tell you where it could go wrong in theory. In theory, just so we're steelmanning it. The biggest risk to me is not on the demand side. The biggest risk is that you get a glut of compute and you get an overbuild And the same way that we had dark fiber after the dot-com crash, if you had dark GPUs, that'd be a disaster for everyone, especially if you built out your computer infrastructure expecting a spot price of $30 to $50 a lot. as you know elon said that they were expecting right so if all of a sudden There are too many people racing to provide this compute, and now there's an oversupply, and the market crashes. That'd be the risk factor. In a weird way, All the political headwinds, I think, ensure against that outcome because it is so hard to build data centers for all the reason we said there's a whole moral panic slash panic. hysteria slash hoax going on, that actually it's those political headwinds I think will almost guarantee that there's not an oversupply relative to the exponentially growing demand. So in a weird way, you're protected against that. But Jake, can I shift gears for a second to tell you what's so brilliant about what Jensen did here? There's one other part in return for that. They get a revenue share. above that floor who gets the revenue share above the floor explain yeah So I think the way I think the deals will work is, hey, we're willing to guarantee offtake at you know, nowhere near 30 or $50 a watt, we're willing to guarantee offtake at a very low risk level, that with our really high resolution view, of productive capacity and future model trends we're comfortable with. [65:00] In return, we get some revenue share above probably a higher level. And Morgan Stanley wrote a great note, I think two days ago, Nvidia could become, and these are effectively royalties, these revenue shares, they could very quickly become a very large cloud with a capital light business. Let me tell you why I think what Jensen did was so brilliant. It's because the numbers are getting so big. that the TAM is getting constrained by the ability to finance this build out. And what he's doing is alleviating that finance constraint so that he can grow as big as the TAM actually is, right? It's removing that constraint. So just to take one example, Elon wants to add somewhere around six to eight gigawatts next year. we know that that would cost $300 to $400 billion of CapEx. The company just raised $100 billion in its equity and debt offerings. So obviously, they would have to go out and finance that somehow. And as we talked about in our previous episode, the simplest way to finance it would be to get seller financing from NVIDIA, especially given that the payback period could be as quick as one year. So now Jensen is creating the, you could say the line of credit, using these big banks, using these big private equity shops, and he's making that available And that's going to now benefit all of these downstream purchasers. I don't think it's circular. This is not... a circular situation like girly's worried about. This is, I think Wall Street banks and private equity firms providing financing, based on the expected cash flows that will be delivered from these GPUs. In some ways, NVIDIA is kind of becoming the central bank of AI. the Federal Reserve of AI. That is like a way to kind of conceptualize what they are doing. And in the same way, the Federal Reserve, like a lot of monetary policy, it works through private banks. Here you're going to have, and probably more banks will opt into this, you have private markets. who are charging fees, underwriting these deals, and NVIDIA is helping to facilitate that. So I don't think it's really circular financing. Yeah, I watched that CNBC special as well. One point that really stuck with me was when the private equity guys said that this was a little bit like what they do for plane financing, where when an airline buys airplanes, they're able to finance it, not just based on the credit worthiness of the airline, but rather because the airplanes have value. And that makes it much easier to finance because you know that even if the airline gets in trouble, you're protected, right? You have asset assets. back financing and that's what they're basically doing here with gpus now in order for that to happen Nvidia has to standardize, they have to create essentially reference designs and so forth, and so on, so that it can all be standardized enough, for these Wall Street guys to now package it up and securitize it, turn it into securities. And that's basically what they're doing like they did with mortgages. Yeah, I'll say there's two There's a reason why I'm not worried about this build out. There's two So there's a governor, like a throttle, Gavin, if you will, there's a governor on this, which is the physics of the real world that we talked about in building data centers. There's also a dead man switch on this, which is, you know, the switch that you have to hold down to run the train. But if you have a heart attack, your hand comes off the switch. which is anthropic, and open ai these are the two biggest customers one of them has scaled back open ai their ambitions in this regard of their build out from like 1.4 trillion to like but 600 billion and then there's We just said, okay, they're going to go to 400 billion in revenue next year or something like that, or they'll end the year at that run rate. If they don't need the compute or compute gets so efficient or open source models create a headwind, whatever it is, they're going to take their hand off the buy button. And that dead man switch will then slow everything down so we don't get a mortgage-backed security. Well, that could lead to a car crash. But actually, this is why I think the Anthropic IPO is really important for the market, is getting those quarterly earnings and be able to see their numbers every quarter, I think it will become probably the most important signal that the entire industry has. Why? Because you've got all these people out there saying that AI is a bubble, this capex is not justified, there won't be an ROI. and Everyone's going to be looking to... Anthropics quarterly earnings as the pace car, not just them, it'll be them and OpenAI and SpaceX, but you'd have to say that Anthropic right now is the pace car. And they're gonna be looking to them to see if the demand signal is there. And if there's a hiccup or a wobble, - It'll show up there first, yes. - If Anthropic slams on the brakes, there'll be a pileup of companies behind it because Anthropic, they're making 100 billion per gigawatt of compute. [70:00] That's why they're able to pay SpaceX say 50 billion a gigawatt, you know, spot price for compute, That's why SpaceX is then able to pay $30 billion to NVIDIA for chips for the build-out. That's why NVIDIA is able to pay TSMC and... Micron and SK Hynix and all the way down. It's the entire food chain So if there is like a wobble, If somebody slams on the brakes, if the pace car slows down, the horse is going to feel it. But it's still better. For those numbers to come out on a quarterly basis? It's such... in a vetted, GAP-compliant way. So Gavin, as we're sort of saying here, there's a kill switch, there's a break, whatever you want to call it, Is there enough space? Is there enough distance between the cars... to not have a pile up. In other words, if Dario does slam on the brakes, if there are headwinds, does everybody go slamming into them? Or is there enough for the entire industry to say, okay, yeah, we're going to go to 20% slow, we're going to go 30% slower. It's very simple. If he's slimming on the brakes because there's not demand, there will be a pileup. OK, there's there will be a pop. There's not enough room between the cars. Yeah, if it's because of demand. Now, if Dario is getting passed by SpaceX or OpenAI or an American Open Source champion, there won't be a pileup. But I do think the Anthropoc S1 is going to be really important. And it's going to break a lot of people's brains. And the reason is, is I think there are a lot of macro and value investors. who are very confidently making these prognostications about AI with the assumption that tokens are subsidized. Anthropic is generating cash. And like they kind of this fact is, I would say, kind of known in the world and I've actually never really spoken to Anthropic. But it just feels like there are enough signals And then if you're to read the tea leaves of what has been said in public, they're generating cash, they're profitable. Open source tokens are profitable. Anthropic is profitable. OpenAI, if they're not generating cash, they will be imminently. SpaceX, you know, same thing. And this... When I hear these macro and value investors who are ignorant, they're very smart, but they're ignorant of the facts, and they're making an assumption that tokens are subsidized and this is all going to collapse in some circular financing bonfire, they're just wrong. The overwhelming majority... of tokens are profitable for everyone in the chain. Everyone. And I think the Anthropic S1 is kind of going to make that clear. But when I listen to these macro and value investors talk about AI, it's a little bit about me saying, you know, I'm very bearish on the world economy because oil is at $500 a barrel. And you know what? If oil wasn't $500 a barrel, That'd be a good reason to be bearish. It's just not like they're just wrong. Like the simple fact is wrong. They're paid in some ways to be doom scrollers and merchants of doom. I can tell you, you know, one framing of this, that I've been working on, which is the number of people employed in the United States, like 150 million 160 million, what their collective salaries are 10 trillion, 12 trillion, whatever it winds up being. There is no world in which I don't see. corporations and people in the economy spending five or ten percent of the salaries of their employees on the equivalent in tokens in other words somebody makes eighty thousand dollars a year on average they're gonna you you you will spend eight thousand dollars ten percent of their salary or a minimum of four thousand dollars to make them more efficient we have an analogy for this it's sas software and compute computers on their desktop mobile phones etc just their general use of technology if you put those numbers together you're looking at a trillion dollars in ai spend just in the united states it's obviously And if I'm wrong and it's half and it's 500 billion, well, that's kind of where these companies are trending. And that's not even talking about the rest of the planet. I think we're going to be fine, folks. All right, let's wrap up here. I got two more stories I want to get to. We'll go lightning round here. Very interesting story coming out of New Jersey and New York around Amazon. New Jersey's AG has sued Amazon last week. They're arguing the company is cheating basically by hiring drivers through subcontractors. This has been well known. It's called DSP. This concept of delivery service partners is a very clever, perhaps too clever of a way for our Amazon to shield themselves according to these lawsuits that are starting to pile up because if an Amazon [75:00] driver crashes there's liability and it gets you out of some you know, things you might have to pay for if they were Amazon employees on Monday. New York City mayor in front of the pods, the Zoran Mondami jumped into the fray he is throwing support and he did one of his big grin socialist democratic socialist videos that he's now becoming famous for telling uh and demanding through municipal laws that will require amazon to hire the drivers themselves quote Corporations like Amazon, build billion dollar business models by insulating themselves from accountability through a system of exploitative subcontracting. This has been an issue for a long time. There have been, tragically, a number of deaths because the subcontractors are put under massive pressure. And people get killed tragically when they run into people because of them operating in the real world and Because the drivers are technically employed by the DSP, Amazon, claims you know hey we're not the employer so we're not on the hook and we don't have to get into union bargaining or labor obligations so sacks i'm curious your take on this in terms of the overall surge. in people believing in socialism now and believing that corporations are cheating on the margins is amazon cheating here is this too clever and maybe they need to roll this back i'm curious your opinion because it's getting momentum and it's going to be one of the big issues in the midterms and obviously in 2028. Well, I mean, this is a practice that's widespread. It's not focused on New York. I think Amazon's been using these networks of independent contractors as drivers for a while. And what they say is that if you deny them the ability to use these independent contractors, it will... raised delivery costs. And there was a study that showed that the average household could pay hundreds of dollars extra. If you change this, it would slow down service. It would put thousands of jobs at risk at small contractor businesses. And Amazon says that it could lead them to shift operations outside the city. Also, One thing that this DSP model does is it enables rapid scaling for peaks, so holiday surges, dense urban routing, lower costs. It's a much more flexible arrangement than if everyone was just an Amazon employee. So this is how the market has evolved is I think maybe when they started, Amazon probably was doing everything. And they started to decentralize it. to these networks of small businesses, these contractors, it actually ends up creating more jobs, it creates more entrepreneurship, I think these DSPs, employ thousands of people in New York City. Those jobs to be threatened. If you basically made them illegal, So I think this is a case where Where do you stand, Sachs? Where do you stand on it? I would just let freedom of contract and volunteer exchange. prevail here. And look, this market structure evolved for a reason. The workers liked it. The DSP owners liked it. Amazon liked it. These are mutually beneficial arrangements. I don't see the reason why Mamdani has to insert the city in the middle of this. What I would do if I was Amazon, and I'm sure they will not do this, I would just put a surcharge. Hey, this is the Zoran Mamdani fee. that is being imposed on you. Loads of people use Amazon, not just wealthy people in New York. Lots of people use Amazon. And this is the Zoran Mom Donny Fee. as a result of this ill-advised policy, or have it as a line item. in every bill. Because that's what it's going to result in. When you compare Amazon service to anything government provides, It's insane how efficient Amazon is. You order your thing and it can be there the same day. or at worst it gets delivered to you the next day. It's unbelievable how amazing Amazon service is. It's optimized for low prices for consumers. And we're going to basically get in the middle of that and have the government get in the middle of Amazon's driver network. What city service does Does Mamdani provide that remotely is as efficient as what Amazon does, that he now is going to tell them how to run their business. It's insane. If you look at what Starbucks did, when Schultz was running it, he gave people a job worth having. And he took pride in taking the lowest paid people in the organization and giving them a living wage. And that living wage is somewhere around 20 bucks, 22 bucks in these cities. These DSP drivers are actually paid quite well, 18 bucks, 19 bucks, 20 bucks. But this was not created in the free market as you're sort of saying, Saks, and it was just what everybody wanted. [80:00] No, this was created explicitly. to benefit Amazon and to take the burden of these employees and put it onto the American taxpayer. And they are getting called out on it, rightfully so, by New Jersey and New York. And I believe you know, and I'm a long-term Amazon customer, I've been, you know, a Prime subscriber since the beginning, it would be a de minimis cost for them to make these employees, full-time employees and you'd still have some dsps to scale up and down sure but this is an opportunity for them to make those frontline workers part of the amazon family it would cost but 25 cents per delivery in new york to move them from this 18 19 an hour to give them a slightly higher pay raise maybe 21 22 bucks and give them benefits this is an easy thing to do And what our industry is? what our industry has a crazy blind spot for. is that they look for efficiencies and amazon is the most efficient company in the world to the point of it being abusive to the most, vulnerable members of our society and for being against the esprit de corps of capitalism And this is a level of capitalism that needs to be self-corrected for. And I think the government's going to have very valid arguments here. I think it would be amazing for Amazon to get ahead of this. Because this will be... what the democratic socialists also known as socialist communists in my mind this is a very valid argument that will sell with Americans who believe that this group of people is being taken advantage of, and that we can easily afford to pay 25 cents more per package delivery yeah sax you're right it might add a hundred dollars why don't we have uber do that too then jcal absolutely uber has because uber was basically castigated for the exact same thing using independent contractors instead of making them full-fledged employees it's actually make a great point about uber and this would be the equivalent of if uber had three or four thousand people representing their drivers massive lawsuits that they have to contend with for what those drivers do. So it's distinctly different. And they've gone through this. I'm not a proponent of socialism. But I do think that there's an opportunity for Amazon, which is going to be automating along the way to just take a little bit of a haircut on their amazing business, but then be on the right side of history and not unemployment, food stamps, and health insurance, and give that to the public to do. And these socialists are going to point this out in the next two election cycles to the point of absurdity. And it's going to play. Socialism is playing. Socialism is making inroads, and it's infecting people's brains. And it's up to capitalists to say, you know what, we don't need to be as cutthroat and don't need to be as clever. So that's my message as a long term shareholder in Amazon. I have a. Very big position in Amazon, and I keep increasing it every year. They should do the right thing here and get ahead of this so that they can say, hey, we are an incredible employer as opposed to, hey, we really are trying to hide and move these expenses around. That's my only point on it. Not a socialist, not a cuck, not interested in – messing with capitalism, I think they got a little too clever here. That's my only point. And I think they should do the right thing here and treat those the the most vulnerable the docket what's that now i know why you put this on the docket so you can get on your soapbox well no no i put it on the docket because it's one of the top news stories and i put it on the docket because we talk about socialism every week and we talk about all the virtues of capitalism and there are downsides to capitalism where we can get too clever and then let me ask you a question i think we spent too much time on this but let me ask you a question you said - Go ahead. - That you want Amazon to employ all these drivers. By the way, they don't-- - Not all of them. I think they should employ a bunch of them. - I know, you said you wanted a bunch to still remain with the DSPs. - I think the DSPs can happen-- - So how do you decide which is which? I think they could proactively do that. It's a great question. Andy Jassy could proactively decide just like Schultz did. And he wants to get rid of the DSPs. He wants to make them all. And I think there's a compromise here, which is there are DSPs to scale up when there's demand during the holidays. Seems like a great idea, but I think he should get ahead of it. As soon as you acknowledge that the DSPs are better. because there are more flexible working arrangements that people want. Why are you meddling in this arrangement to begin with? That's actually not true. The people who work in the DSPs, the DSPs themselves, the people who own those fragmented, very small companies that go out of business the second they get in an accident to shield Amazon, those employees would very much like to have Amazon equity. They would love to get RSUs and they would love to get health care and benefits. I don't know, I don't think you can speak for them. Some of those drivers, they work for Uber a lot of the time. They're trying to, I don't know. [85:00] I actually can. They toggle back and forth between being a driver for Amazon, being a driver for Uber. They... Like the flexible working arrangements. You're actually misinformed here. You're incorrect. These are people who are working full shifts. They're working eight to 12 hour shifts. I think it's actually 10 to 12 hour shifts. They're exactly. I don't think you can speak for all these workers. You can speak for them if you they can speak for themselves and they are organizing and Amazon has been trying to break the organization efforts when they try to unionize. And this is a tool for breaking that by forcing them to be part of the DSP. That's the game on the field, Sachs. Well, you already acknowledged that you think the DSPs should stick around. I think a portion of them should do that, and I think Andy Jassy and the team at Amazon should get ahead of this because it's a bad look. It's not in the best interest of capitalism to be this cutthroat. I think he should do it proactively. I don't think he should be forced to do it. That's my personal belief. The same way Schultz said there's an opportunity here to make baristas have a living wage and have benefits. And it was one of the best things Starbucks ever did. And it's really what capitalism should start thinking about. What is our role? What is our role in driving the move to socialism amongst young people? And this is an example of one of the reasons people are starting to believe in socialism is because they believe that companies like Amazon are cheating a little bit on the margins and maybe trying to take the expenses that they should incur and put them on the American taxpayer. And you know what? Mondami and New Jersey are going to win these lawsuits. I predict it now. Well, look, J. Cal, that all sounds really nice, and I want to save capitalism as much as you do. And if Amazon doing that would really save capitalism, then... I guess I'm game for it, but I'm not sure it will. What we do know it will do is raise the cost of packages by about $5.20, per package and it will cost families $664.00. more per year and it puts something like 5 000 jobs at risk so there are real costs in terms of doing things the way that you're saying. When people are wearing their hat as voters, yeah, they fall for the socialism stuff, but then when they're consumers, when they're wearing their consumer hat, they just want to pay less for packages. They will hate the fact that packages cost $5 more. And like Gavin said, they should put the Mamdani surcharge on. on their orders and it's not 25 cents it's five dollars $5.20 per package, $664 per household per year in New York. That's crazy. That's a real cost. Talk about cost of living adjustment. That's a huge cost of living adjustment. JCal has had internet issues and we're just moving on. Gavin. Last issue. What did you think of GrokBot? What did you think of the releases that XAI made this week? I guess it was GrokBot and then also it was... Grok 4.6, right? And people were saying that it is close to the frontier or at the frontier, but a lot cheaper. So I've been maintaining, as of even last week, that we were down to a frontier lab duopoly. with Anthropic and OpenAI. Was I wrong about that? Is this number three, Is it more flexible and fluid than I thought? Yeah, I think it is. I mean, I think it's can it can we show up, show that graph of the Pareto frontier? This is pretty wild. Now on the left, what this shows is kind of like quality or intelligence on the Y axis. and cost on the x-axis. So you want to be in the upper right quadrant. And you can see here you just want to be on the outside of this and you can just see like how disruptive Grok 4.6 is pricing. Right. So as you move right on the x-axis, it's getting cheaper. It's getting cheaper. And the y-axis is the capability and it's getting more and more capable as you move up. Exactly. They invert the x-axis so that as you move to the top right it's better right exactly and i mean this is now and this is on cursor bench and XAI, SpaceX is almost certain they have a right to acquire cursor, they're working with cursor, so maybe this is creating your own homework. Well, right here on the right... This is from Databricks, which just raised money at $190 billion. This is a very real company that is very sophisticated. And you can see, like even on this, You're well ahead of Fable 5. which I think most people would kind of agree is the... you know, the gold standard. Like this is your higher quality, that's a slightly lower price. So I think Grox 4.6 is, you know, the numbers are the numbers. on X, I think we put it down. We do have to wait and see the vibes, but I think it's pretty interesting. DHH, who I think created Ruby on Rails. David Hanmeyer-Hanson. Yeah, he posted something really positive. [90:00] Base camp, base camp. the DHAs, David Hanmere. 37 signals, great. 37 signals, Basecamp, yes. He posted something really positive. You saw the Databricks evaluation. You saw our Mercor evaluation. And I think the early vibes are really good in these benchmark scores are bearing out in the real world. Elon has caught up, in other words. And if you were to attribute why he's caught up, team, they're a dialed in team. Number two, he reshuffled the management of XAI and brought in his killers, From SpaceX. Yeah, Gavin, those are the two reasons. Absolutely. He brought in the SpaceX Aces along with the cursor aces and some new people. And he did that in six short months, Gavin. Never count Elon out. He's frequently late. Never wrong. And he tends to over deliver when it does come. And that's to his own admission. He's like, I'm always late. But we get it right eventually. Yeah. Absolutely, and this is still a relatively small model. It's 1.5 trillion parameter models. A larger model, Grok 4.7, that should be significantly more capable, is coming out in a few short weeks. GrokBot also seems... just as important. It feels kind of like another open claw moment for AI, for this kind of personalization of AI, of making it democratizing AI. The same thing that Zuckerberg said in his manifesto. He wants to create bots for everybody. And that's what Claude Cowork and Claude Tag both do exceptionally well, as well as Hermes, Perplexia Computer. I'm just going to admit, for me, some of those are, and I try to be very technical, but I'm Like, Rock bot feels... Not too hard to use, but it feels much easier to use for... No, no, rock bot is much easier to use. Rock bot is much easier to use. Open claw and Hermes, you need to be... Hermes? Hermes. They're just like the freedom. If you've ever had that channel, it's like that. Facts. Like you, our wives order some channel once in a while and they break with us. It is pretty amazing that in six months he caught up. But Saks, this creates a tension. This does create a little tension. He's hosting Anthropic. He's catching up to Anthropic. What do you think about that tension here that he's there, Amazon Web Services, Elon Web Services at the same time is competing with them increasingly? What I love about it is that he's creating these huge compute clusters, right? And that gives him the ability to train a big model, a big potentially frontier model. So you can think of that as like the call option is to catch up and become a true frontier lab. But if he fails at that, he's got the put option of just selling the compute. Yes. Spot at pretty attractive prices. to the frontier labs and he has the right sacks to take that compute back from anthropic he put that clause according to reports into the anthropic deal so he could take back compute if he wants to right both sides have a 90-day cancellation which means that if the spot price of compute goes up he benefits if it goes down by the way it could hurt him because Either side could renegotiate within 90 days, right? So – Basically, Elon is able to charge spot minus 90. Mm. So that's the risk. But it all it all goes back to, you know, we were talking about before, which is, do you think that compute will be supply constrained going into next year and the year after? And if so, that's a very advantageous dynamic for Elon, because, again, he's very close to charging true spot for his compute. but he can also essentially reroute it to his own service if he's able to take the lead. Can I say what's interesting? Sorry to interrupt. I talk to a lot of investors about SpaceX, okay? And what's just amazing to me is everyone will have a nuanced view on Starlink. Maybe they're a skeptic on Starlink Direct to Sell. I think the plan to use kind of these distributed Starlink, effectively almost, we call them small cells, femto cells, is interesting. Everybody will have a nuanced view on orbital compute. Whether they're bullish or bearish, they will have a view. They'll have a view on terrestrial compute. Very few of these investors talk about Grok at all. And it is Pareto dominant on a lot of measures, and we have an existence proof that Anthropic went from whatever it is, a billion to 50 billion, really fast. And if Anthropic is worth, you know, with two trillion per the FT, you know, three or four trillion, you know, according to yourself on a previous episode, settle out. I said I thought that was a market clearing price. I didn't say that's where I thought it was worth. you know, maybe people should start to consider rock. [95:00] when they think about SpaceX. SpaceX has so many ways to win, and I think that's the higher end of it here. We're obviously all talking our book here. The folks on this podcast are amongst the earliest investors in SpaceX. I was part of leading SpaceX. a small part of leading the second round to 10 billion. Hey, there's one thing that broke during the show that we should talk about, and that's that Silver Lake may be buying Workday. Workday's up 17%. And if this is true, this is, I think, a significant moment for the stock market. Software has been in a death spiral that really kind of has started, that probably troughed maybe three, four months ago, but it was in a tough spot for 12, 15, 18 months. And the return of a private software really changes the investing landscape. And, I don't know what the potential buyers of Workday are thinking, but I do know that the rise of open source and the increasing competitiveness of open source is a godsend for the American software industry. It is an absolute godsend. Yeah, if you want to throw up a charter Workday. you know, like a five year chart, you can just see it. But a world where you have one, two, or three dominant frontier models... I mean, you could just see the stock went from, yeah, we'll call it 300 straight down to 100. really short period of time. And even with this bid, you've only recovered a little bit. but a world where Anthropik, OpenAI, and maybe even SpaceX are kind of the only model providers that dominate the model layer. That is a hard world for software. If I were to guess, why the private equity bid for software might be coming back, it's because they see what's happening with open source. Pretty amazing. And that is the backstop, isn't it, Sax? We've talked about this. If... Nobody in the public markets wants to own these companies and private equity sees away. We talked about bending spoons, I think, last week, a whole bunch. There is now, I think, I don't want to call them bottom feeders, but what would you call them, Gavin? Like, you know, sharp acquirers of equipment? money printing businesses that want to attract even more revenue from it. Yes, Axe? This is a good turn of events that there is a buyer for these companies. Absolutely. I mean, what Silver Lake is saying, they've been oversold. They've been oversold, yeah. And so here they go. There must be some value in that. And if you use AI to run these, I think that was the thing Bending Spoons is doing really well. They get rid of 80% of the employees and then put the 20% who are left as AI first employees and they figure out how to. get all of this incredible revenue. There's your breaking news, Gavin Baker. Amazing. Tied right now. Just so you know, I looked at the metrics before we got on today. You're now tied. with Brad Gerstner in terms of ratings. You and Brad Gerstner are neck and neck and tied for fifth bestie, which means one of you will drop to sixth bestie if this continues. And we'll see. Brad Gerstner may be on next week. So we'll see which bestie takes the fifth slot. David Sachs, great to see you. Hey, maybe we'll get a year-round later. Maybe we can get a steak. You want to get a steak later? I got Thursday night off here. I got a hall pass tonight. Maybe we can go get a steak. August 14th. See you next time. Bye-bye. Oh, man. you